article stringlengths 1.98k 169k | summary stringlengths 1.01k 4.15k | section_headings listlengths 2 38 | keywords listlengths 0 12 | year stringclasses 11 values | title stringlengths 30 189 |
|---|---|---|---|---|---|
Reversible lysine acetylation is one of the most important protein posttranslational modifications that plays essential roles in both prokaryotes and eukaryotes . However , only a few lysine deacetylases ( KDACs ) have been identified in prokaryotes , perhaps in part due to their limited sequence homology . Herein , we developed a ‘clip-chip’ strategy to enable unbiased , activity-based discovery of novel KDACs in the Escherichia coli proteome . In-depth biochemical characterization confirmed that YcgC is a serine hydrolase involving Ser200 as the catalytic nucleophile for lysine deacetylation and does not use NAD+ or Zn2+ like other established KDACs . Further , in vivo characterization demonstrated that YcgC regulates transcription by catalyzing deacetylation of Lys52 and Lys62 of a transcriptional repressor RutR . Importantly , YcgC targets a distinct set of substrates from the only known E . coli KDAC CobB . Analysis of YcgC’s bacterial homologs confirmed that they also exhibit KDAC activity . YcgC thus represents a novel family of prokaryotic KDACs .
Protein ( de ) acetylation plays critical roles in many key biological processes , for example , transcriptional regulation , aging , and metabolism ( Cohen et al . , 2004; Grunstein , 1997; Lin et al . , 2009; Lu et al . , 2011 ) . Recent mass spectrometry ( MS ) efforts have revealed that many proteins are acetylated in Escherichia coli , although only a single E . coli lysine deacetylase ( KDAC ) , CobB , has been identified so far ( Choudhary et al . , 2009; Henriksen et al . , 2012; Zhang et al . , 2013a ) . The fact that induction of CobB only had a limited impact on reducing the global protein acetylation level suggests that additional KDACs may exist . However , homolog searching has failed to reveal any additional KDACs in E . coli , presumably because these enzymes emerged via convergent evolution . In contrast to bioinformatics methods , biochemical approaches have proven effective for identifying new enzymes resulting from convergent evolution ( Tsukada et al . , 2006; Yamane et al . , 2006 ) , though their laborious , time-consuming nature has limited their applications to high-throughput , proteome-wide screens . Herein , we established a ‘clip-chip’ approach to enable a proteome-wide , activity-based search for novel KDACs in E . coli .
The principle behind the clip-chip approach is the delivery of thousands of purified proteins spotted on a glass slide ( e . g . , a proteome microarray ) to a substrate of interest immobilized on another slide ( i . e . , the substrate slide ) such that thousands of desired biochemical reactions can be carried out in parallel , in order to identify new enzymes of interest ( Figure 1a and Figure 1—figure supplement 1 ) . The substrate slide is created by immobilizing a substrate of interest onto a nitrocellulose-coated slide . After thousands of purified proteins are spotted on a plain glass slide , it is then ‘clipped’ onto the substrate slide in a face-to-face manner , resulting in the delivery of the proteins onto the substrate slide . Owing to the highly porous nature of nitrocellulose and the tiny volume of the protein droplets ( 0 . 3–0 . 5 nL ) , the delivered protein droplets are immediately absorbed and kept locally in the nitrocellulose , preventing cross-contamination . To determine which transferred proteins possess the enzymatic activity in question , the ‘clipped’ substrate slide is then incubated with an appropriate reaction buffer , followed by signal detection . 10 . 7554/eLife . 05322 . 003Figure 1 . Screening the Escherichia coli proteome to discover new KDACs using the ‘clip-chip’ strategy . ( a ) Schematic of the ‘clip-chip’ strategy . ( b , c ) Identification of YcgC as a potential protein deacetylase . E . coli proteome chips were clipped onto three substrate slides separately coated with acetylated RutR , NhoA , and YceC . After incubation in a protein deacetylase buffer , the reactions were terminated by adding wash buffers , followed by a signal detection step with a pan α-AcK antibody coupled with a Cy3-labeled secondary antibody as detection reagent to visualize the loss of signals ( e . g . , black holes in ( b , c ) . To determine the identity of proteins that generated the holes , the substrate slide was subsequently probed with an α-6xHis antibody followed by a Cy5-conjugated secondary antibody . ( d ) Using acetylated RutR proteins purified from E . coli , of the four candidates tested , YcgC showed robust deacetylation activity in vitro . Equal amounts of RutR proteins were used in each reaction and loss of acetylation was detected with the pan α-AcK antibody . DOI: http://dx . doi . org/10 . 7554/eLife . 05322 . 00310 . 7554/eLife . 05322 . 004Figure 1—figure supplement 1 . Design of the ‘clip-chip’ strategy . ( a ) The ‘clip-chip’ strategy uses two slides , a protein slide that contains proteins of interest printed onto a slide with an appropriate surface , and a second substrate slide on which the enzymatic reactions are carried out . The protein slide containing arrayed protein droplets ( b ) is first imprinted onto the slide coated with substrate ( red ) ( c ) to transfer the proteins of interest from the protein slide onto the substrate slide . ( d ) Visible and homogenous watermarks indicate that the protein droplets from the protein slide are effectively and evenly transferred to the substrate slide . ( e ) An appropriate enzymatic reaction buffer is then loaded onto the surface of the imprinted substrate slide and reactions are carried out under appropriate conditions . After a series of stringent washes , the results are recorded with a microarray scanner . DOI: http://dx . doi . org/10 . 7554/eLife . 05322 . 004 To screen for new KDAC candidates in the E . coli proteome , we prepared separate substrate slides for three E . coli proteins , namely NhoA , RutR , and YceC , which were chosen because they have a rather high endogenous acetylation level and because CobB exhibits only modest ability to deacetylate them ( Zhang et al . , 2013b ) . After 4256 individually purified E . coli proteins ( Chen et al . , 2008 ) were spotted on plain glass slides , they were clipped separately onto the three substrate slides , followed by incubation with a standard deacetylase reaction buffer containing NAD+ . The reactions were terminated by adding wash buffers , followed by a signal detection step with a pan α-acetyllysine ( α-AcK ) antibody coupled with Cy3-labeled secondary antibodies as detection reagents . Proteins that efficiently deacetylated the substrates could be readily identified as they left behind pairs of black holes in fluorescence images of the substrate slides ( Figure 1b , c ) . To help determine the identity of the proteins with potential KDAC activity , we subsequently probed the clipped substrate slides with an α-6xHis antibody to visualize the E . coli proteins delivered onto the substrate slides . As a negative control , substrate slides were also processed in parallel without the clipping step . We identified four candidates that showed significant deacetylation activities against at least one of the three substrates tested . To validate the KDAC activity observed above , we purified the four candidate proteins and performed solution phase deacetylation reactions against RutR . CobB was also included for comparison . By evaluating the decrease in acetylation signals using an immunoblot assay with α-AcK , we confirmed that one of the candidates , YcgC , could readily deacetylate RutR in vitro , and that CobB also deacetylated RutR . YjgD did not show any detectable deacetylation activity against RutR , while Gnd and YhbL showed slight activity ( Figure 1d ) . As YcgC also showed KDAC activity against NhoA and YceC ( data not shown ) , we then focused on characterizing the function of YcgC . YcgC is previously known as DhaM , a subunit of dihydroacetone kinase complex and a nonessential gene in E . coli . Because the endogenous level of YcgC is very low , YcgC was overexpressed on the wild-type background in the subsequent experiments . As the M subunit of the dihydroxyacetone kinase complex , the possibility that YcgC has intrinsic enzymatic activity has not been reported previously ( Molin et al . , 2003 ) . Therefore , we first determined whether YcgC also requires NAD+ and/or Zn2+ to deacetylate RutR , as class III deacetylases require NAD+ as a cofactor and other classes are dependent on Zn2+ ( Thiagalingam et al . , 2003 ) . We chose RutR as the substrate for YcgC , because endogenous RutR proteins are highly acetylated and because it is known to regulate genes directly or indirectly involved in the complex pathways of pyrimidine and purine metabolism ( Shimada et al . , 2007; Shimada et al . , 2008 ) . To reduce possible contamination from other proteins , cofactors , or metal ions as much as possible , both 6xHis-tagged YcgC and RutR proteins were affinity purified from E . coli under stringent wash conditions , followed by overnight dialysis . Immunoblotting of the deacetylation reactions clearly showed that YcgC could deacetylate RutR effectively , but this activity did not appear to be dependent on NAD+ or Zn2+ ( Figure 2a , Figure 2—figure supplement 1 ) . High-performance liquid chromatography analysis and inductively coupled plasma-mass spectrometry ( ICP-MS ) , respectively , confirmed that there was no detectable NAD+ or Zn2+ in the reaction ( data not shown ) . Of note , Coomassie staining of the decaetylated RutR protein product band appeared at a slightly lower molecular weight than acetylated RutR and this is explored below . 10 . 7554/eLife . 05322 . 005Figure 2 . In vitro and in vivo characterization of YcgC’s KDAC activity . ( a ) In vitro assays of the KDAC activity of YcgC on RutR demonstrated that its KDAC activity does not require either NAD+ or Zn2+ as cofactors . Incubation with YcgC almost completely abolished the slower migrating acetylated RutR bands ( upper panel ) as evidenced by immonublotting ( lower panel ) . ( b , c ) LC-MS/MS analysis to determine the residues of RutR deacetylated by YcgC . RutR was treated with YcgC first and the untreated RutR used as the control . Both these two samples were resolved on a SDS-PAGE gel side by side . The upper band represents the Kac-containing starting materials and the lower band represents the K-containing product , which were then recovered from the gel and subjected for MS/MS analysis ( inserts ) . Lys52 was identified as an acetylated site in RutR protein ( b ) . After incubating with YcgC , acetylation on K52 was no longer detectable ( c ) . ( d ) RutR is deacetylated by YcgC in Escherichia coli . A 3xFLAG tag was chromosomally inserted at the 3′-end of rutR coding sequence . Acetylation of 3xFLAG-tagged RutR was monitored upon induction of YcgC . While the protein level of RutR was unchanged ( middle panel ) , its acetylation level was dramatically reduced as a function of YcgG induction ( upper panel ) . YcgC’s expression was monitored using a custom-made antibody ( lower panel ) . ( e ) Mutagenesis of RutR confirmed that K52 and K62 are acetylated in vivo . Two single mutants K52Q and K62Q and one double mutant K52/62Q were constructed . These mutants along with WT RutR were produced and purified in E . coli . Equal amounts of purified proteins were Western blotted with the α-AcK antibody , quantitation of acetylation level of these samples were performed . KDAC: Lysine deacetylase; LC-MS/MS: Liquid chromatography–mass spectrometry; IP: Immunoprecipitation . DOI: http://dx . doi . org/10 . 7554/eLife . 05322 . 00510 . 7554/eLife . 05322 . 006Figure 2—figure supplement 1 . Without NAD+ , CobB could not deacetylate RutR . In vitro assays of the KDAC activity of CobB on RutR demonstrated that its KDAC activity is dependent on NAD+ . Incubation with CobB and NAD+ almost completely abolished the slower migrating acetylated RutR bands ( upper panel ) as evidenced by immonublotting ( lower panel ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05322 . 00610 . 7554/eLife . 05322 . 007Figure 2—figure supplement 2 . LC-MS/MS analysis to determine the residues of RutR deacetylated by YcgC . RutR was treated with YcgC first and used the untreated RutR as the control . Both these two samples were resolved on a SDS-PAGE gel side by side . The upper band represents the Kac-containing starting materials and the lower band represents the K-containing product , which was then cut and subjected to MS analysis ( the inlet ) . ( a ) Lys62 was identified as an acetylated site in RutR . ( b ) After incubating with YcgC , acetylation on K62 was no longer detectable . DOI: http://dx . doi . org/10 . 7554/eLife . 05322 . 00710 . 7554/eLife . 05322 . 008Figure 2—figure supplement 3 . Specificity and sensitivity of the custom-made YcgC monoclonal antibody as assessed by Western blotting . Specificity was measured by spiking an E . coli total lysate with affinity purified YcgC , and sensitivity by testing serially diluted YcgC . DOI: http://dx . doi . org/10 . 7554/eLife . 05322 . 00810 . 7554/eLife . 05322 . 009Figure 2—figure supplement 4 . Mutagenesis of RutR confirmed that K52 and K62 are acetylated in vivo . Four single mutants K52Q , K62Q , K52R , and K62R and two double mutants K52/62Q and K52/62R were constructed . These mutants along with WT RutR were produced and purified in E . coli . Equal amounts of purified proteins were Western blotted with the α-AcK antibody . DOI: http://dx . doi . org/10 . 7554/eLife . 05322 . 009 Next , we employed liquid chromatography–mass spectrometry ( LC-MS/MS ) to determine which acetylated lysine residues of RutR were deacetylated by YcgC . We found that Lys52 and Lys62 , present in the peptide sequences LEQIAELAGVSK52TNLLYYFPSK and TNLLYYFPSK62EALYIAVLR , respectively , were acetylated in RutR expressed in wild-type ( WT ) cells ( Figure 2b and Figure 2—figure supplement 2a ) . However , after RutR was incubated with YcgC , acetylation of K52 or K62 was no longer detectable ( Figure 2c and Figure 2—figure supplement 2b ) . Therefore , YcgC effectively deacetylates RutR on residues K52 and K62 in vitro . To determine whether YcgC could deacetylate RutR in cells , we performed immunoprecipitation ( IP ) -coupled immunoblotting to measure changes in the acetylation levels of RutR over the period of YcgC induction . To enable immunoprecipitation of endogenous RutR proteins , a 3xFLAG tag was chromosomally inserted into the 3′-end of the rutR coding sequence . An isopropyl-beta-D-thiogalactopyranoside ( IPTG ) -inducible ycgC construct was then transformed into the rutR:3xFLAG cells and induced for YcgC expression for up to 4 hr . Using IP-coupled immunoblotting analysis , we observed that acetylation levels of RutR proteins were significantly reduced in a YcgC expression level-dependent manner as detected by a custom-made α-YcgC monoclonal antibody ( Figure 2d and Figure 2—figure supplement 3 ) . In contrast , the total amount of RutR was not affected by YcgC induction ( Figure 2d ) . These results confirmed that YcgC effectively deacetylates RutR in vivo without affecting its stability . To examine whether K52 and K62 acetylation sites of RutR were deacetylated by YcgC in vivo , we created two single ( K52Q; K62Q ) - and one double ( K52/62Q ) -mutants of RutR . After transformation of these mutants into E . coli , subsequent IP-coupled immunoblotting demonstrated that , compared with WT RutR , mutation of either K52 or K62 resulted in a substantial loss of acetylation signals in RutR , with the K52/62Q double mutant showing the lowest acetylation signals ( Figure 2e ) . RutR K-to-R mutants ( i . e . , K52R , K62R , and K52/62R ) were also created and tested , and similar results were observed to those with the K-to-Q mutants ( Figure 2—figure supplement 4 ) . These results suggest that both K52 and K62 are major acetylation sites in RutR , and can be effectively deacetylated by YcgC in E . coli . Because KDACs catalyze hydrolytic reactions on lysine residues , we tested a variety of hydrolase inhibitors against YcgC in in vitro deacetylation reactions as described above . We found that Halt Protease Inhibitor Cocktail ( with or without ethylene glycol tetraacetic acid [EGTA]; Thermo Scientific , Rockford , IL ) and Complete Protease Tablet ( Roche , Mannheim , Germany ) could significantly inhibit YcgC’s deacetylase activity ( Figure 3—figure supplement 1a ) . Further analysis revealed that the active component in the Halt Protease Inhibitor Cocktail was a serine hydrolase inhibitor 4- ( 2-aminoethyl ) benzenesulfonyl fluoride ( AEBSF ) , but not the other components . Additional assays demonstrated that YcgC’s deacetylase activity could not be inhibited by well-known deacetylase inhibitors , including trichostatin A , SAHA ( suberoylanilide hydroxamic acid ) , and NAM ( Nicotinamide ) , and that hydrolase inhibitors , phenylmethylsulfonyl ( PMSF ) , leupeptin , ethylenediaminetetracetic acid and EGTA , showed no detectable inhibition of YcgC ( Figure 3—figure supplement 1b , c ) . Thus , it is likely that YcgC belongs to the serine hydrolase family , which has no significant homology to any of the annotated KDACs to date . To identify which Ser residue in YcgC was most critical for its hydrolase ( KDAC ) activity , we examined five Ser residues , namely S7 , S10 , S73 , S77 , and S200 , which are highly conserved among its prokaryotic homologs on the basis of protein sequence alignment . Next , we created a quintuple Ser-to-Ala mutant ( i . e . , 5SA ) in YcgC and tested its ability to deacetylate RutR in vitro . As compared with the WT YcgC , the 5SA mutant appeared devoid of RutR deacetylase activity ( Figure 3a ) . To determine which of the five Ser residues was likely to be the catalytic nucleophile for hydrolase activity , we incubated WT YcgC with hydrolase inhibitor AEBSF and the following MS/MS analysis revealed that Ser200 was the only conserved residue that was covalently labeled with AEBSF ( Figure 3b ) . To confirm its importance , a Ser200-to-Ala ( S200A ) mutant was created and tested in the same in vitro assay . As anticipated , the deacetylase activity was not detectable with S200A YcgC , establishing Ser200 as the likely key catalytic residue . 10 . 7554/eLife . 05322 . 010Figure 3 . S200 is critical for YcgC’s deacetylation activity . ( a ) Two mutants of YcgC , that is , S S8/10/73/77/200A and S200A were constructed through gene synthesis . In vitro assays of the KDAC activity of these two mutants on RutR demonstrated that their KDAC activities were completely abolished . ( b ) S200 on YcgC was identified as an AEBSF binding site by LC-MS/MS analysis . YcgC was incubated with AEBSF , then trypsin digested and subjected to MS/MS analysis . Upon AEBSF-mediated sulfonation , a 183 Da molecular weight increase is predicted . ( c ) In vitro assays of the KDAC activity of YcgC on heat-denatured RutR demonstrated that YcgC was still active ( right panel ) , while the downshift band disappeared ( left panel ) . Similar results were observed when heat-denatured RutR was treated with CobB . KDAC: Lysine deacetylase; LC-MS/MS: Liquid chromatography–mass spectrometry; AEBSF: 4- ( 2-Aminoethyl ) benzenesulfonyl fluoride; WT: Wild type . DOI: http://dx . doi . org/10 . 7554/eLife . 05322 . 01010 . 7554/eLife . 05322 . 011Figure 3—figure supplement 1 . YcgC’s protein deacetylase activity is inhibited by AEBSF . ( a ) YcgC’s deacetylase activity on YcdC was monitored when a variety of hydrolase inhibitors were added individually . The solvents of these inhibitors , that is , DMSO and ethanol , were also included as controls . ( b ) Testing the individual component of the Pierce Halt protease inhibitor cocktail revealed that AEBSF inhibits the protein deacetylase activity of YcgC . ( c ) YcgC’s protein deacetylase activity is not affected by both EDTA and EGTA . DOI: http://dx . doi . org/10 . 7554/eLife . 05322 . 01110 . 7554/eLife . 05322 . 012Figure 3—figure supplement 2 . Lysine 62 is critical for RutR proteolysis . ( a ) Mutant proteins of K52R , K62R , K52/62R , K52Q , K62Q , and K52/62Q with WT , RutR proteins were treated with YcgC and examined with Coomassie stain . ( b ) The ratio of cleavaged RutR to intact RutR was also calculated . DOI: http://dx . doi . org/10 . 7554/eLife . 05322 . 01210 . 7554/eLife . 05322 . 013Figure 3—figure supplement 3 . The Km and Vmax values of YcgC were determined using RutR as a substrate . Because the N-terminal cleavage of RutR is tightly coupled with its deacetylation by YcgC , the downshifted band of RutR in the YcgC deactylation reaction could be conveniently used as a surrogate of YcgC’s activity . YcgC was incubated with serially diluted acetylated RutR . Deacetylation was determined by measuring the intensity of the lower bands on a silver-stained gel . DOI: http://dx . doi . org/10 . 7554/eLife . 05322 . 013 As mentioned , RutR deacetylation by YcgC reproducibly induces faster migration on sodium dodecyl sulfate polyacrylamide gel electrophoresis ( SDS-PAGE ) . This behavior was also observed with deacetylation by the sirtuin family member CobB ( Figure 2a; Figure 3a ) . This phenomenon is not typically associated with protein deacetylation and we considered the possibility that proteolytic degradation of deacetylated RutR was occurring . N-terminal Edman sequencing of RutR after YcgC treatment showed N-terminal truncation with loss of 14 residues ( N-MTQGAVKTTGKRSR ) or 11 residues ( N-MTQGAVKTTGK ) . RutR without YcgC treatment was also sequenced , no N-terminal truncation was observed . We considered the possibility that YcgC might show protease activity in addition to its unambiguous deacetylase activity . However , we believe this is unlikely since treatment of RutR with the sirtuin CobB , mechanistically and structurally unrelated to YcgC , induces a similar N-terminal truncation in RutR , visualized by SDS-PAGE and detected by N-terminal sequencing with loss of 14 residues . We suspect that this N-terminal cleavage of RutR results from autoproteolysis . To explore this possibility , intact RutR proteins were first heat-denatured and then incubated with either YcgC or CobB under the same deacetylation conditions . As shown in Figure 3 , heat-denatured RutR could be partially deacetylated by both YcgC and CobB , but did not show a downshifting on SDS-PAGE . These results are consistent with the possibility that deacetylation of native acetylated RutR sparks autoproteolysis but denaturation inhibits this autoproteolytic activity . To understand whether specific Lys residue ( s ) play a role in facilitating RutR autoproteolysis , K52R , K62R , K52/62R , K52Q , K62Q , and K52/62Q mutant and WT RutR protein were treated with YcgC and analyzed using Coomassie-stained SDS-PAGE ( Figure 3—figure supplement 2 ) . Measurement of the ratios of cleaved/intact RutR revealed that K52Q and K52R RutR behaved similar to WT RutR . In contrast , K62Q , K62R , and the two double mutant RutR proteins showed diminished cleavage . Therefore , we propose that the apparent autoproteolytic activity of RutR is dependent on its deacetylation and that removal of the acetyl group from K62 appears most important for its cleavage . Future studies will be needed to further understand the molecular mechanism for the apparent RutR autocleavage and its biological function . Because the N-terminal cleavage of RutR is tightly coupled with its deacetylation by YcgC , the downshifted band of RutR in the YcgC deactylation reaction can be conveniently used as a surrogate of YcgC’s activity . We thus performed steady-state assays to monitor the enzyme kinetics of YcgC via measuring the production of the downshifted RutR band , and estimated the Km and Vmax of YcgC to be 2 . 13 ± 0 . 65 μM and 0 . 29 ± 0 . 07 μM/min/μM , respectively ( Figure 3—figure supplement 3 ) . Originally identified as a transcriptional repressor of the rutABCDEFG operon in E . coli , RutR ( YcdC ) was later found to bind to 19 additional E . coil chromosomal loci ( Shimada et al . , 2008 ) , including the coding regions of pmrD and gcd ( Umezawa et al . , 2008 ) . However , deletion of rutR alone does not result in a significant elevation of the expression levels of most of its target genes , suggesting that RutR regulates target gene transcription via a different mechanism . To determine whether deacetylation of RutR by YcgC plays a direct role in transcription regulation of RutR’s downstream target genes , we monitored the expression levels of 15 known target genes of RutR in ycgC-induced cells . Interestingly , while induction of ycgC did not change the expression level of rutR , the expression of two rutR targeting genes , pmrD and gcd , was significantly decreased , by as much as fivefold as measured with quantitative polymerase chain reaction ( PCR ) over a 2-hr period of ycgC induction ( Figure 4a ) . On the other hand , induction of cobB in parallel did not affect expression levels of rutR or any of the 15 of RutR’s target genes ( data not shown ) , suggesting the possibility that YcgC regulates a different set of substrates from CobB . 10 . 7554/eLife . 05322 . 014Figure 4 . YcgC and CobB target distinct sets of substrates . ( a ) YcgC regulates gene expression via deacetylating RutR . Expression of gcd and pmrD is significantly reduced upon RutR induction over a period of 2 hr as measured by quantitative real-time PCR . Double asterisks indicate that the observed fold changes are statistically significant , p<0 . 01 . ( b ) Global gene expression analysis of ycgC- and cobB-induced cells . Clustering analysis shows clearly that impact on global transcription of induction of ycgC is distinct from that of cobB . Venn diagram showing that there is no significant overlap between genes down- and up-regulated due to CobB and YcgC induction . ( c ) Overexpression of YcgC affects global protein acetylation levels in E . coli . After ycgC and cobB were separately induced for 1 hr , global acetylation was detected in whole lysates of Escherichia coli using two pan α-AcK antibodies . The WT E . coli strain was also included for comparison . Boxed areas indicate regions that show obviously different staining patterns in ycgC- and cobB-induced cells . PCR: Polymerase chain reaction; WT: Wild type . DOI: http://dx . doi . org/10 . 7554/eLife . 05322 . 014 To test this hypothesis , we examined the impact of induction of ycgC and cobB on global gene expression profiles in E . coli ( Allard et al . , 1999; Yeung et al . , 2004 ) . Using a standard DNA microarray approach , we found that , compared with WT cells , 197 genes were significantly repressed and 93 genes were activated after 4 hr of ycgC induction . In agreement with the above observations , expression levels of pmrD and gcd were significantly reduced ( Figure 4b; Supplementary file 1 ) . A similar analysis in parallel revealed that 4 hr of cobB induction resulted in 195 and 136 up- and down-regulated genes , respectively , compared with WT cells . However , with the exception of xerC , none of the RutR’s targets was affected ( Supplementary file 2 ) . Furthermore , Venn diagram analysis did not reveal any significant overlap between either the up- or down-regulated gene groups in the ycgC and cobB induction experiments ( Figure 4b ) . These results suggest that YcgC profoundly affects global gene expression and it probably functions via distinct biological processes from CobB . This conclusion is further supported by evidence obtained at the protein level . Using immunoblotting with pan α-AcK , we observed that overexpression of ycgC decreases the acetylation levels of many proteins , resulting in a global change in acetylation profiles compared with those of WT cells ( Figure 4c ) . Importantly , changes in the acetylation profile of ycgC-induced cells were different from those in cobB-induced cells . For example , in boxed areas 2 and 3 ( Figure 4c ) , the acetylated bands are almost completely absent in ycgC-overexpressing cells , while they are essentially unchanged in cobB-overexpressing cells . On the other hand , in boxed area 1 , cobB overexpression completely abolished the acetylation signals , whereas only a modest decrease in acetylation signals is observed in ycgC-overexpressing cells ( Figure 4c ) . Taken together , the above results suggest that YcgC and CobB each target a distinct set of substrates . To determine whether YcgC’s KDAC activity is evolutionarily conserved , we searched for its homologs in both eukaryotes and prokaryotes . Although YcgC shows limited homology to components of the phosphotransferase system ( Punta et al . , 2012 ) , no statistically significant homologs were identified in eukaryotes ( Molin et al . , 2003 ) . However , many prokaryotic homologs with high similarity were readily identified . Other than homologs from Escherichia strains , the closet homolog is the DhaM protein from Shigella sp . ( str . 2457T ) , and statistically significant homologs were also identified in more remotely related bacterial species . Therefore , we selected five homologs , representing a wide range of homology , for closer scrutiny ( Figure 5a ) . Sequence alignment of the five representative homologs with YcgC showed that the N-terminal regions ( i . e . , amino acids 1–2501–250 ) of these proteins are more conserved than the C-terminal regions ( Figure 5b ) . To determine whether these YcgC homologs possess protein deacetylase activity , the five selected genes were synthesized , subcloned into the same IPTG-inducible expression vector as that of E . coli YcgC , and transformed into E . coli . After a 4 hr induction of each YcgC homolog , changes in global acetylation profiles were determined with two pan α-AcK antibodies and compared with that for WT cells ( Figure 5c , d ) . Results clearly showed that induction of each of the five YcgC homologs gave rise to a unique protein deacetylation signature that is different from that of the WT and YcgC-induced strains . For example , overexpression of the Klebsiella homolog significantly reduced the acetylation levels of proteins at ~55 kDa ( red box; Figure 5c ) . As another example , overexpression of the Pantoea homolog substantially reduced the acetylation levels of proteins around 60 and 43 kDa compared with the WT strain ( wide and narrow red boxes; Figure 5d ) . Taken together , these results demonstrate that all five YcgC homologs possess readily detectable KDAC activity with different substrate preferences . Because YcgC and its homologs share little similarity with all the known KDACs identified so far , and because its activity does not require NAD+ or Zn2+ , these results strongly suggest that this group represents a novel prokaryotic KDAC family . 10 . 7554/eLife . 05322 . 015Figure 5 . YcgC represents a new family of KDACs . ( a ) Five representative YcgC homologs with protein sequence homology ranging from low to high . ( b ) Amino acid sequence homology analysis between YcgC and five selected YcgC homologs from other bacteria . The consensus strength among the six homologous proteins at each amino acid position of YcgC is indicated with colored bars . Red , orange , green , light blue , dark blue , and blank bars represent 100 , 80 , 60 , 40 , 20 , and 0% consensus strength , respectively . ( c , d ) Changes in global E . coli acetylation profiles upon induction of the five YcgC homologs . The five selected YcgC homologs were cloned , transformed into E . coli , and induced to overexpress . Global acetylation profiles of each induced strain were detected with a pan monoclonal antibody ( Cell Signaling , #9441 ) and a pan polyclonal antibody ( PTM-Biolabs , PTM-105 ) , as shown in c and d , respectively . WT E . coli cells were also processed in parallel as a comparison . An antibody against myelin basic protein was used as a loading control . WT: Wild type . DOI: http://dx . doi . org/10 . 7554/eLife . 05322 . 015
In this study , we have applied a clip-chip approach to identify new KDAC candidates in E . coli . Our in-depth biochemical characterization revealed that the novel KDAC YcgC removes the acetyl groups on K52/62 of its substrates RutR via a previously unknown Ser hydrolyase activity . A surprising observation was that , after deacetylated by either YcgC or CobB , the RutR showed a significant downshift on SDS-PAGE , suggesting possible proteolytic activity . Further biochemical analysis established that this is likely autoproteolysis of RutR that is stimulated by deacetylation of K62 . Our data suggest that acetylation of RutR may enhance its stability . Indeed , endogenous RutR purified from cells grown under standard conditions is heavily acetylated . Although a complete understanding of this phenomenon will require future study; to our knowledge , this is the first example of a protein deacetylation event driving proteolytic activation . Our in vivo functional studies on YcgC revealed that it down-regulates the expression of several RutR target genes by catalyzing the deacetylation of two lysine residues on RutR . It has been puzzling until now how RutR represses target gene transcription by its previously reported binding to coding regions of pmrD and gcd , as the binding sites are located hundreds of base pairs downstream of the start codons , and deletion of rutR does not enhance their expression levels ( Umezawa et al . , 2008 ) . Based on the results of this study , we propose that YcgC deacetylates RutR leading in turn to the recruitment of additional cofactors that enhance silencing of target gene expression . As PmrD has been demonstrated to serve as a connector between multiple two-component signal-transduction systems in Salmonella enterica , E . coli , and other bacteria , our study also suggests the possibility of crosstalk between protein acetylation and phosphorylation in E . coli , a prevalent regulatory mechanism found in eukaryotes ( Eguchi and Utsumi , 2005 ) . This study also highlights several advantages of the clip-chip approach . First , as an activity-based screen , this method can be readily adopted to search for many other types of enzyme activities . Second , the clip-chip approach does not require any prior knowledge of the enzymes of interest as long as a robust biochemical assay is available . Third , it is a proteome-wide , high-throughput screen that does not require further deconvolution ( e . g . , in MS/MS ) of the positive signals because of the use of a protein microarray on which each protein is physically addressable . Finally , the clip-chip approach is capable of functional annotation of enzymes using both gain- and loss-of-signal reactions . We envision that the ‘clip-chip’ strategy will proved to be of wide application for the de novo discovery of enzyme activity in biology .
Unless otherwise stated , all chemicals used in this study were purchased from Sigma-Aldrich ( St Louis , MO ) , and enzymes were purchased from New England Biolabs ( Ipswich , MA ) . E . coli proteome chips were prepared as described previously ( Chen et al . , 2008 ) . In brief , expression plasmid-carrying E . coli cells were cultured , induced , and harvested in 96 deep-well plates . To purify the fusion proteins , cell pellets were treated with lysozyme and incubated with Ni-NTA Superflow ( QIAGEN , Valencia , CA ) in Multiscreen Nylon Mesh filter plates ( Millipore , Billerica , MA ) . After six washes , the proteins were eluted with 250 mM imidazole . To prepare the proteome chip , the purified proteins were re-arrayed from 96-well plates into 384-well plates in a cold room using an Apricot system ( Apricot Designs , Covina , CA ) . The re-arrayed proteins were printed in duplicate onto plain glass slides . A substrate slide was prepared by coating FAST slides with 200 μl acetylated protein at a protein concentration ≥0 . 1 μg/μL . The E . coli proteome chip was imprinted onto the substrate slide . After the removal of the proteome chip , the substrate slide was submerged in protein deacetylation buffer ( 50 mM Tris-HCl , 4 mM MgCl2 , 50 mM NaCl , 50 mM KCl , 1 mM NAD+ , pH 8 . 0 ) . The reaction was carried out at 26°C for 16 hr . The slide was washed three times with 1× Tris-buffered saline and Tween 20 ( TBST ) , 5 min each time , and incubated with an α-AcK antibody ( #9441 of Cell Signaling Technology , Danvers , MA ) . The incubation was carried out with a 1:1000 antibody dilution at room temperature for 1 hr . The slide was washed four times with 1× TBST , 5 min each time , followed by incubation with a Cy3-conjugated secondary antibody from Jackson ImmunoResearch ( West Grove , PA ) . To facilitate the identification of positive spots , the substrate slide was further probed with an α-6xHis antibody followed by a Cy5 conjugated secondary antibody from Jackson ImmunoResearch . A GenePix 4200A microarray scanner was used to record the results . Since this is a loss-of-signal assay , the signal intensity of each protein spot was defined as ‘Background-Foreground’ . The signal intensity of each protein was averaged from the two replicate spots . Signal-to-noise ratio ( SNR ) , that is , signal/standard deviation of background , was set as the final signal of each protein . The cutoff to call protein deacetylase candidates was set as SNR ≥3 . Deacetylase candidates identified by clip-chip were overexpressed and purified in E . coli . In a 20 μL reaction , the three acetylated substrates , that is , 3 μg of RutR and YceC , and 0 . 5 μg of NhoA , were individually incubated with 5 μg of each deacetylase candidate . The reactions were carried out in protein deacetylation buffer at 37°C for 1 hr . These protein samples were then analyzed by both silver staining and Western blotting . Membranes were further probed with an IRDye 800 secondary antibody at room temperature for 1 hr and visualized with an Odyssey Infrared Imaging System from LI-COR Biosciences ( Lincoln , NE ) . Hlat protease inhibitor cocktail was purchased from Thermo Scientific , cOmplete protease tablets were from Roche , AEBSF , aprotinin , bestatin , and pepstatin A were obtained from Sangon Biotech Co . , Ltd ( Shanghai , China ) . The deacetylation assays were performed as described above except for the addition of a variety of hydrolase inhibitors at appropriate concentrations . The solvents of these inhibitors , that is , dimethyl sulfoxide and ethanol , were also tested as controls . Affinity purified YcgC ( 80 µg ) was diluted in 200 µL phosphate-buffered saline ( pH 7 . 4 ) , and denatured at 100°C for 10 min to release any bound NAD+ . After centrifugation at 12 , 000 rpm for 10 min , the supernatant was transferred to a new Eppendorf tube , and acetonitrile ( ACN ) was added to the supernatant at a ratio of 3:1 ( vol/vol ) . The reaction was mixed well , allowed to stand for 20 min at 4°C , then centrifuged at 12 , 000 rpm for 10 min . The resulting supernatant was then subjected to liquid chromatography-coupled high-resolution mass spectrometry ( LC-HRMS ) analysis . An aliquot of pure NAD+ ( 50 µM ) was tested to calibrate the LC-HRMS system and as a positive control . LC-HRMS was performed as described previously ( Vogliardi et al . , 2011 ) on a Waters ACQUITY UPLC system equipped with a binary solvent delivery manager and a sample manager , coupled with a Waters Micromass Q-TOF Premier Mass Spectrometer equipped with an electrospray interface ( Waters Corporation , Milford , MA ) . Briefly , LC was performed on a Syncronis HILIC column ( 50 × 2 . 1 mm , 1 . 7 µm ) ( Thermo Scientific ) . The column was eluted with 200 mM ammonium formate aqueous solution and ACN in gradient mode at a flow rate of 0 . 30 mL/min at 30°C . MS was performed using negative polarity , 2 . 4 KV capillary voltage , 30 V sampling cone , 4 eV collision energy , a source temperature of 110°C , and a desolvation temperature of 350°C . The flow rate for the desolvation gas was set at 600 L/hr . Scan range was set to m/z 50–1000 , scan time to 0 . 3 s , and interscan time to 0 . 02 s . ICP-MS analysis ( Goullé et al . , 2005 ) was performed according to the manufacturer’s instructions . Briefly , 0 . 48 mg of YcgC was prepared in 4 mL concentrated nitric acid and 2 mL deionized water . The solution was then subjected to microwave digestion with a Multiwave 3000 instrument from Anton Paar ShapeTec GmbH ( Wundschuh , Austria ) at 600 W power for 15 min . The digested sample was filtered through a filter paper . The sample was analyzed on an ELAN 9000 ICP-MS instrument from PerkinElmer , Inc . ( Waltham , MA ) . Human hair GBW07601a ( GSH-1a ) from Institute of Geophysical and Geochemical Exploration ( Hebei , China ) was included as a positive control . All experiments were carried out at room temperature in a dust-free area with a relative humidity of 10–85% . YcgC ( 0 . 3 μM ) was incubated with RutR at a series of concentrations , that is , 1 . 5 , 2 . 3 , 3 , 4 . 5 , 6 , 7 . 5 , 9 , 12 , and 15 μM . The reaction was carried out in protein deacetylation buffer without NAD+ at 37°C for 15 min . Protein samples were then resolved by 12% SDS-PAGE followed by silver staining . The gel was scanned with a PowerLook 2100XL from Techville , Inc . ( Dallas , TX ) , converted to an 8-bit grayscale image and analyzed by Image J ( NIH; http://rsb . info . nih . gov/ij/ ) . Acetylated RutR and deacetylated RutR were trypsin-digested and analyzed with a nanoflow LC-MS/MS coupled online with a Q Exactive Plus quadrupole orbitrap mass spectrometer ( Thermo Scientific , San Jose , CA ) equipped with a nanoelectrospray ion source . Briefly , the peptide mixtures were loaded onto a C18 column ( 100 mm inner diameter , 10 cm long , 5 mm resin ) from Michrom Bioresources ( Auburn , CA ) using an autosampler . Peptides were eluted with a 0–35% gradient ( Buffer A , 0 . 1% formic acid , and 5% ACN; Buffer B , 0 . 1% formic acid , and 95% ACN ) over 80 min and detected online with a Q Exactive Plus quadrupole orbitrap mass spectrometer using a data-dependent TOP10 method ( Haas et al . , 2006 ) . E . coli strain ( W3110 ) harboring chromosomal 3xFLAG-tagged RutR was constructed using the Red recombination system ( Poteete , 2001 ) . In short , the DNA cassette for recombination was composed of a 150 bp upstream flanking sequence , the rutR gene , a 3xFLAG tag before the stop codon of rutR , followed by the sequence of the kanamycin resistance gene , and a 150 bp downstream flanking sequence . This cassette was synthesized and cloned into pUC57 by GenScript ( Nanjing , China ) . The cassette was amplified by high fidelity PCR and treated with DpnI . One microgram of the linear DNA fragment was electrotransported into E . coli W3110 cells carrying pKD46 , and these recombinants were selected using kanamycin medium and verified by colony PCR . The plasmid carries ycgC from the E . coli AG1 strain that we used for the construction of the E . coli proteome chip , was extracted and transformed into the E . coli W3110 strain harboring chromosomal 3xFLAG-tagged RutR . The transformed E . coli strain was then cultured in lysogeny broth media to a OD600 of 0 . 6–0 . 8 and induced by 1 mM IPTG at 37°C for 0 . 5 , 1 , 2 , and 4 hr . Cells were harvested and treated with lysis buffer ( 50 mM NaH2PO4 , 300 mM NaCl , 20 mM imidazole , 1× CelLytic B , 50 units/mL of Benzonase proteinase inhibitor cocktail , and 1 mM PMSF , pH 8 . 0 ) at 4°C for 2 hr with vigorous shaking . The 3xFLAG-tagged RutR was then immunoprecipitated using an α-FLAG antibody and protein G conjugated agarose beads . Samples were resolved on a 10% SDS-PAGE gel followed by Western blotting with an α-AcK antibody ( Cell Signaling Technology , Shanghai , China ) and an α-FLAG polyclonal antibody . RutR mutants K52Q , K62Q , and double mutant K52Q/K62Q were synthesized and cloned into pET28a+ with GenScript ( Supplementary file 3 ) . The acetylation level of these RutR mutants was detected with an α-AcK antibody ( Cell Signaling Technology ) and compared with that of the WT E . coli strain . Mouse α-YcgC monoclonal antibody was custom-made by Abmart , Inc . ( Shanghai , China ) . Western blotting was applied to characterize the antibody . The sensitivity of the antibody was tested using serially diluted RutR and its specificity was tested by spiking purified RutR into a whole lysate of E . coli . At a molar ratio of 500:1 , AEBSF and YcgC were incubated at 37°C for 1 hr . After SDS-PAGE and Coomassie staining , YcgC band was cut off and in-gel trypsin digestion was done according to the standard protocol . The digested YcgC was then analyzed using Ultimate 3000 Nano Pump LC system from Thermo Scientific coupled with an electrospray ionization quadrupole time-of-flight mass spectrometer from Bruker Daltonics ( Bremen , Germany ) . The LC setup was coupled online to a Q-TOF using a nano-ESI source from Bruker Daltonics in data-dependent acquisition mode ( m/z 350–1500 ) . Tandem mass spectra were extracted , charge state was deconvoluted and deisotoped by Compass Data Analysis version 4 . 1 from Bruker Daltonics . Mascot version 2 . 4 from Matrix Science ( Boston , MA ) was set up to search the database ( entries ) . Carbamidomethyl on cysteine was specified as fixed modifications , oxidation of methionine was specified as variable modifications . RutR was incubated with YcgC in deacetylation buffer at 37°C for 1 hr . Protein sample was then resolved by 12% SDS-PAGE and transferred to the polyvinyl difluoride membrane . The shift RtuR band was dyed by Ponceau S and cut off , and then N-terminal sequenced by protein sequencer PPSQ-33A from Shimadzu ( Kyoto , Japan ) . The raw data and graphs generated by PPSQ-33A were identified and exported by PPSQ-33A data processing . The N-terminal sequence RtuR was then determined . E . coli cells were cultured in LB medium at 37°C . Before and after induction with 1 mM IPTG for 1 hr , cells were treated with lysis buffer at 4°C for 2 hr with vigorous shaking . Cell debris was removed by centrifugation at 4°C . The protein concentration of the whole lysate was determined using the BCA Protein Assay ( Pierce , Rockford , IL ) ; 100 μg of whole lysate was then resolved on a 10% SDS-PAGE gel followed by Western blotting with an α-AcK antibody ( Cell Signaling Technology ) overnight at 4°C . As a loading control , the protein lysates were Western blotted with an α-myelin basic protein antibody from Abmart , Inc . ( Noinaj et al . , 2013; Spanò et al . , 2011 ) . Results were recorded using an IRDye 800 secondary antibody and the Odyssey Infrared Imaging System ( LI-COR Biosciences ) . Total RNA was extracted using an RNA extraction kit from TIANGEN Biotech Co . , Ltd . ( Beijing , China ) . RNA was then reverse-transcribed to cDNA using a random oligo primer from Promega ( Beijing , China ) , according to the manufacturer’s instructions . Primers were synthesized by Sangon Biotech . ( Shanghai , China ) ( Supplementary file 4 ) and validated by regular PCR and melting curve analysis . Real-time PCRs were carried out using FastStart Universal SYBR Green Master from Roche ( Shanghai , China ) and the ABI 7500 real-time PCR platform ( Life Technologies Corporation , Shanghai , China ) . E . coli DNA chips were purchased from CapitalBio Corp . ( Beijing , China ) . cDNA labeled with a fluorescent dye ( Cy5 and Cy3-dCTP ) was produced by Eberwine’s linear RNA amplification method and subsequent enzymatic reaction ( Guo et al . , 2005; Patterson et al . , 2006 ) . Arrays were hybridized in a CapitalBio BioMixer II Hybridization Station overnight and scanned with a LuxScan scanner and the images obtained were then analyzed using LuxScan 3 . 0 software from CapitalBio Corp . A space- and intensity-dependent normalization based on a LOWESS program was employed ( Yang et al . , 2002 ) . To identify significantly differentially expressed genes , SAM 3 . 02 was used . Unsupervised hierarchical clustering was used to cluster samples or genes . The distance between single samples or genes was based on Pearson’s correlation coefficients . Distances between clusters were calculated using the ‘complete linkage’ method . Venn diagrams were drawn using the R package Vennerable . The size of each circle proportionally reflects the number of unique genes in each group . To identify YcgC’s bacterial homologs , ‘dihydroxyacetone kinase subunit M’ was used as a search term in PubMed under the ‘Protein’ category . This search showed that the top taxonomic groups , that is , Escherichia , Klebsiella , Shigella , Serratia , Citrobacter , Yersinia , Enterobacter , Salmonella , Pantoea , and Providencia , all belonged to the Enterobacteriaceae . The amino acid sequence of YcgC was then Blasted against the most significant taxonomic groups using BlastP . Bacterial strains of the closest homologs from each taxonomic group were determined , that is , Citrobacter koseri ATCC BAA-895 , Enterobacter aerogenes KCTC 2190 , Klebsiella oxytoca KCTC 1686 , Pantoea ananatis LMG 5342 , Providencia stuartii MRSN 2154 , Serratia odorifera 4Rx13 , Shigella flexneri 2a str . 2457T , and Yersinia enterocolitica subsp . enterocolitica WA-314 , and the amino acid sequences of these homologs , along with that for YcgC from E . coli K12 W3110 , were then subjected to phylogenetic tree construction and sequence alignment using LaserGene ( DNAstar Inc . Madison , WI ) . | After proteins have been made , they can be modified in several ways . For example , chemical tags called acetyl groups may be added to ( and later removed from ) the protein to regulate cell activities such as aging and metabolism . Enzymes are proteins that help catalyze the reactions that add or remove the acetyl tags on certain “substrate” proteins . In the bacteria species Escherichia coli , many enzymes that help to add acetyl groups to proteins have been discovered . However , only a single E . coli “deacetylase” enzyme that removes the acetyl group has been identified . Now , Tu , Guo , Chen et al . have devised a technique to identify new deacetylases , called the “clip-chip” approach . In this method , thousands of proteins that are potential deacetylases are arrayed on a glass slide , and substrate proteins are immobilized on another slide . The two slides are then clipped together face-to-face , allowing the potential enzymes to transfer to the substrate slide and interact with the substrates . Using this approach , Tu , Guo , Chen et al . identified a protein called YcgC as a deacetylase in bacteria . Further characterization experiments revealed that YcgC works in a different way to other known deacetylases , and that it targets different substrates to the previously known E . coli deacetylase . Tu , Guo , Chen et al . found that the equivalents of YcgC in other bacteria species are also deacetylases; these enzymes therefore represent a new deacetylase family . In the future , the clip-chip approach could be used to discover new members of other enzyme families . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology"
] | 2015 | YcgC represents a new protein deacetylase family in prokaryotes |
CLC secondary active transporters exchange Cl- for H+ . Crystal structures have suggested that the conformational change from occluded to outward-facing states is unusually simple , involving only the rotation of a conserved glutamate ( Gluex ) upon its protonation . Using 19F NMR , we show that as [H+] is increased to protonate Gluex and enrich the outward-facing state , a residue ~20 Å away from Gluex , near the subunit interface , moves from buried to solvent-exposed . Consistent with functional relevance of this motion , constriction via inter-subunit cross-linking reduces transport . Molecular dynamics simulations indicate that the cross-link dampens extracellular gate-opening motions . In support of this model , mutations that decrease steric contact between Helix N ( part of the extracellular gate ) and Helix P ( at the subunit interface ) remove the inhibitory effect of the cross-link . Together , these results demonstrate the formation of a previously uncharacterized 'outward-facing open' state , and highlight the relevance of global structural changes in CLC function .
CLC transporters catalyze the exchange of Cl- for H+ across cellular membranes ( Dutzler , 2007; Matulef and Maduke , 2007; Zifarelli and Pusch , 2007; Jentsch , 2008; Accardi and Picollo , 2010; Miller , 2015; Accardi , 2015; Jentsch , 2015 ) . In humans , they are critical to a wide variety of physiological processes and constitute therapeutic targets for treating diseases ( Jentsch , 2008; Zhao et al . , 2009; Stauber et al . , 2012; Stolting et al . , 2014; Devuyst and Luciani , 2015; Pusch and Zifarelli , 2015; Zifarelli , 2015 ) . In bacteria and yeast , CLCs are virulence factors and therefore could serve as drug targets to protect against food poisoning and fungal infections ( Iyer et al . , 2002; Zhu and Williamson , 2003; Canero and Roncero , 2008 ) . From bacteria to humans , Cl-/H+ exchange by CLC transporters occurs with a strict stoichiometry of 2 Cl- for every H+ ( Accardi and Miller , 2004; Picollo and Pusch , 2005; Scheel et al . , 2005; Jayaram et al . , 2011; Leisle et al . , 2011 ) . To achieve this stoichiometric exchange , CLCs must follow an alternating access mechanism , in which bound substrate ions access either side of the membrane alternately , i . e . , they cannot access both sides simultaneously ( Patlak , 1957; Jardetzky , 1966; Shilton , 2015 ) . The alternating access mechanism can only be realized by coupling of ion binding , translocation , and unbinding events to conformational changes in the transporter protein . Specifically , movement of ions between solution and the ion-binding sites of the transporter , as well as ion movement between binding sites , needs to be coupled to conformational changes between “outward-facing” ( in which the external , but not internal , solution is accessible to ions ) , “occluded” ( in which neither solution is accessible ) , and “inward-facing” ( in which the internal , but not external , solution is accessible ) states ( Forrest et al . , 2011; Rudnick , 2013 ) . In all other active transporters that have been structurally ( or biophysically ) characterized , the conformational changes governing the interconversion between these major functional states involve significant protein motions , including reorientation of helices or even entire domains ( Shi , 2013; Paulino et al . , 2014 ) . For the CLC transporters , in contrast , it has been proposed that the transport mechanism may be fundamentally different and involve only localized side chain motions ( Feng et al . , 2010; 2012 ) . However , this proposed mechanism is based largely upon the observation that no large-scale CLC conformational change could be detected crystallographically . Given the strong constraining forces in a crystal environment , which often prevent the protein from populating all naturally accessible , functionally relevant conformational states ( Elvington and Maduke , 2008; Gonzalez-Gutierrez et al . , 2012; 2013; Kumar et al . , 2014 ) , alternative approaches for detecting CLC conformational change during its function are strongly motivated . CLC transporters are homodimers in which each subunit independently catalyzes Cl-/H+ antiport ( exchange ) ( Robertson et al . , 2010 ) . There are two key Cl--binding sites within the protein lumen , known as Scen and Sext . The central anion-binding site ( Scen ) is stabilized by a positive electrostatic potential created by the N-termini of Helices F and N as well as by interactions with conserved Ser and Tyr residues , which physically occlude the anion from the intracellular side ( Figure 1A ) . Using cross-linking as an alternative approach to crystallography , Basilio et al . demonstrated that the conserved Tyr contributes to an intracellular “gate” that opens to generate an inward-facing state ( Basilio et al . , 2014 ) . While this inward-facing state has not yet been structurally characterized in detail , the elegantly designed cross-linking studies demonstrated that movement of neighboring Helix O ( Figure 1B ) is required in conjunction with movement of the Tyr-gate residue . 10 . 7554/eLife . 11189 . 003Figure 1 . Structure of CLC transporters . ( A ) Structure of ClC-ec1 ( pdb: 1OTS ) . The bound Cl- ( one in each identical subunit at site Scen ) is coordinated by conserved Ser and Tyr residues ( shown as spacefilled ) . The N-termini of helices F and N ( shown in purple and yellow respectively ) point towards this site and provide a positive electrostatic environment for the anion . The H+-permeation pathways are delineated by two key residues , Gluex and Gluin . Gluex also acts as a “gate” that blocks the Cl--permeation pathway ( green arrows ) from the extracellular solution . ( B ) CLC structure highlighting helices discussed: F ( purple ) , N ( yellow ) , O ( pink ) , P ( blue ) , Q ( brown ) , and R ( aquamarine ) . ( C ) Close-up of the Cl--binding region in WT ( left ) and E148Q ( right ) ClC-ec1 , highlighting intracellular and extracellular gate residues S107 , Y445 , and E148 ( Gluex ) . In the E148Q mutant ( pdb: 1OTU ) , the Gln side chain , mimicking the protonated Gluex , swings away from the Cl--permeation pathway and is replaced at Sext with a Cl- ion . The structure of this mutant is otherwise indistinguishable from the WT structure . ( D ) Cartoon of the Cl--binding region , illustrating the hypothesis that the E148Q structure represents an “outward-facing occluded” rather than an “outward-facing open” conformation . DOI: http://dx . doi . org/10 . 7554/eLife . 11189 . 00310 . 7554/eLife . 11189 . 004Figure 1—figure supplement 1 . Comparison of CLC structures determined at high and low pH . The two homologs shared 80% sequence identity . ( A ) The Salmonella CLC backbone ( PDB 1KPL , orange ) superposes with ClC-ec1 ( PDB 1OTS , blue ) ( RMSD 1 . 5 Å; Cα RMSD 1 . 0 Å ) . ( B ) The side chains of key residues studied here ( Gluex ( E148 ) , F357 , L361 , L411 , M415 , D417 , and Y419 ) are similarly positioned in the two structures . These similarities motivate alternative approaches to crystallography ( this work and others ) , which highlight the fact that a conformation crystallized does not necessarily reflect the ensemble of conformations outside the restraints of crystallization ( Bell et al . , 2006; Elvington and Maduke , 2008; Elvington et al . , 2009; Basilio et al . , 2014; Abraham et al . , 2015 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11189 . 004 At the extracellular side , a highly conserved glutamate residue , “Gluex” , sits above the anion at Scen and blocks it from the extracellular solution ( Figure 1C , left panel ) . Localized side-chain motions of this residue represent the sole differences distinguished in crystallographic studies of CLC transporters ( Dutzler et al . , 2003; Feng et al . , 2010 ) . In the structure of a mutant in which Gln is used as a proxy for the protonated Gluex , the side chain swings upwards and the site previously occupied by the side chain is occupied by an anion ( Figure 1C , right panel ) . Thus , the structure of this mutant has been thought to represent an outward-facing ( OF ) CLC conformational state . However , in this structure the pathway to the extracellular solution is very narrow – too narrow to accommodate Cl- or other permeant ions ( Miloshevsky et al . , 2010; Krivobokova et al . , 2012 ) – suggesting that additional conformational changes are required for the formation of the OF state in order for external anions to access the external anion-binding site ( Sext ) . We therefore hypothesize that the state identified in the E148Q crystal structure is an “outward-facing occluded” state and that a distinct “outward-facing open” state may exist to permit access of external Cl- to the Gluex-vacated Sext site ( Figure 1D ) . Addressing this hypothesis is crucial to understanding the CLC transport mechanism and how it relates to those of canonical transporters . Various experimental approaches have been used to evaluate whether CLC conformational changes beyond Gluex are involved in the transition to an OF open state . Since the pKa of Gluex is ~6 ( Picollo et al . , 2012 ) , a change in pH from 7 . 5 to 5 . 0 will cause Gluex to transition from mostly deprotonated to mostly protonated , and therefore from its position occupying Sext outward towards the extracellular solution . Such pH manipulations can therefore be used to enrich the OF state and probe for changes in protein conformation . Although crystallization at pH 4 . 6 failed to reveal any conformational change ( Figure 1—figure supplement 1 ) ( Dutzler et al . , 2002 ) , spectroscopic approaches have shown that H+-dependent changes do occur outside the restraints of crystallization . Using environmentally sensitive fluorescent labels , Mindell and coworkers showed that Helix R , which lines the intracellular vestibule to the Cl--permeation pathway ( Figure 1B ) , undergoes H+-dependent conformational change during the transport cycle ( Bell et al . , 2006 ) . Using site-specific NMR labeling schemes , our lab has identified H+-dependent structural change at Helix R and also at the linker connecting Helices P and Q ( P/Q linker ) ( Figure 1B ) – a region ~20 Å distant from the Cl--permeation pathway ( Elvington et al . , 2009; Abraham et al . , 2015 ) . Clearly , CLCs undergo H+-dependent conformational changes beyond those revealed by crystallography . The question remains whether and how these conformational changes are involved in regulating ion binding and translocation during Cl-/H+ transport . Here , we study the conformational change in Helix P and the P/Q linker region ( Figure 1B ) in ClC-ec1 , a well-studied prokaryotic CLC , using a combination of 19F NMR , double electron-electron resonance spectroscopy , chemical cross-linking , crystallography , molecular dynamics ( MD ) simulations , and analysis of cross-linking in mutant transporters . Our results show that rearrangement of Helices N and P occurs to widen the extracellular vestibule and generate a previously uncharacterized “outward-facing open” CLC conformational state , thus establishing the involvement of structural changes beyond the rotation of Gluex .
The 19F NMR nucleus is an advantageous reporter of conformational change because of its sensitivity to chemical environment , its small ( non-perturbing ) size , and the lack of endogenous 19F in proteins ( Gerig , 1994; Danielson and Falke , 1996; Kitevski-LeBlanc and Prosser , 2012 ) . Using ClC-ec1 , a prokaryotic CLC homolog that has served as a paradigm for the family ( Chen , 2005; Dutzler , 2007; Matulef and Maduke , 2007; Accardi , 2015 ) , we previously showed that we could replace native Tyr residues with 19F-Tyr and observe conformational changes reported by changes in 19F chemical shift ( Elvington et al . , 2009 ) . Our strategy to enrich the OF conformational state of ClC-ec1 involved lowering the pH of the solution from 7 . 5 to 4 . 5–5 . 0 , as described above . Of the five buried Tyr residues in ClC-ec1 , two reported [H+]-dependent changes in chemical environment . The first , as expected , was at Y445 , which is within 6 Å of Gluex; the second , strikingly , was in a region ~20 Å away , at Y419 near the dimer interface ( Figure 2A ) . 10 . 7554/eLife . 11189 . 005Figure 2 . H+-dependent solvent accessibility of Tyr residues in ClC-ec1 , detected by 19F NMR . ( A ) “BuriedOnly” ClC-ec1 , a mutant in which the five buried Tyr residues ( spacefilled in yellow ) have been labeled with 19F . The seven solvent-exposed Tyr residues have been mutated to Phe . Residues Y445 ( on Helix R , shown in aquamarine ) and Y419 ( linker between Helices P and Q , blue and brown respectively ) were previously identified as undergoing H+-dependent changes in chemical shift ( Elvington et al . , 2009 ) . ( B ) 19F NMR spectra of BuriedOnly ClC-ec1 . Top data panel: low pH was used to enrich the outward-facing conformational state . Changes in chemical shift reflect changes in chemical environment experienced by the 19F nuclei . Middle data panel: spectral changes in response to addition of TEMPOL ( inset ) at pH 7 . 5 . Bottom data panel: spectral changes in response to addition of TEMPOL at pH 5 . 0 . DOI: http://dx . doi . org/10 . 7554/eLife . 11189 . 005 To better understand this conformational change , we performed an accessibility experiment , reasoning that global protein conformational changes often result in solvent exposure of previously buried regions . For this experiment , we exploited the sensitivity of 19F relaxation rates ( and hence spectral line widths ) to the water-soluble paramagnetic probe TEMPOL ( Bernini et al . , 2006; Venditti et al . , 2008 ) . In this experimental setup , movement of a 19F-labeled residue from a buried to a solvent accessible location would be detected by line-broadening and peak attenuation . We first examined whether any of the five buried tyrosine residues in ClC-ec1 exhibits sensitivity to TEMPOL . In “BuriedOnly” ClC-ec1 , a mutant in which all five buried Tyr residues have been labeled with 19F ( Figure 2A ) effects of TEMPOL were observed at both low and high [H+] , with apparently greater sensitivity at high [H+] ( Figure 2B ) ; however , because the 19F resonances are overlapping , we were unable to unambiguously assign the observed changes specifically to effects on chemical shift or line-broadening of a particular resonance . Therefore , to clearly identify the residue ( s ) sensitive to TEMPOL , we generated ClC-ec1 constructs containing only one 19F-Tyr label per subunit ( either Y445 or Y419 , Figure 2A ) , replacing all other Tyr residues with Phe . Although the “Y445only” mutant was unstable and could not be further examined , the “Y419only” mutant ( Figure 3A ) was stable and showed robust , fully coupled Cl-/H+ exchange activity ( Figure 3—figure supplement 1 ) . The functionality of the Y419only mutant may seem startling , given that it involves mutating the highly conserved Cl--coordinating Tyr445 ( Figure 1C ) to Phe , but it is consistent with previous structural and functional studies demonstrating wild-type behavior of the Y445F mutant ( Accardi et al . , 2006; Walden et al . , 2007 ) . 10 . 7554/eLife . 11189 . 006Figure 3 . 19F NMR detects H+-dependent solvent accessibility at Y419 . ( A ) Y419only ClC-ec1 . In this variant , all native Tyr residues except for Y419 have been mutated to Phe , so that only Y419 will carry a 19F label . Y419 is highlighted in the ClC-ec1 structure shown from the point of view of the membrane ( left ) and from the extracellular side ( right ) . The lower panels illustrate that Y419 lies in a buried position ( left: thin slice through the protein at Y419; right , surface representation viewed from the extracellular side . ( B ) 19F NMR spectra of Y419only . The prominent peak centered at -60 ppm shifts upfield ( -61 and -63 ppm ) when the pH is shifted from 7 . 5 to 4 . 5 to enrich the OF state . ( C ) Y419 becomes substantially more exposed to solvent at increased [H+] , as indicated by susceptibility to line-broadening by the water-soluble TEMPOL at pH 4 . 5 ( bottom spectra , green vs black trace ) compared to pH 7 . 5 ( top spectra , orange vs cyan trace ) . ( D ) The change in the Y419 exposure to solvent is reversible , as revealed by return of the signal ( to the expected chemical shift ) when the pH is raised to 7 . 5 ( bottom trace , orange ) . ( E ) Y419 in the channel-like ClC-ec1 background is accessible to TEMPOL at both pH 7 . 5 and 4 . 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 11189 . 00610 . 7554/eLife . 11189 . 007Figure 3—figure supplement 1 . Functional characterization of ClC-ec1 variants . Top: Cl- turnover rates for ClC-ec1 variants examined in this study , with the exception of D417C mutants which are summarized in Table 1 . Bottom: Stoichiometry of transport for WT ClC-ec1 , Y419only ( used in NMR studies ) , D417C ( used in cross-linking studies ) , and Helix-N mutants F357A and L361A ( hypothesized to transmit conformational change from D417C to the Cl--transport pathway ) . Data represent average ± SEM ( n=3–7 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11189 . 00710 . 7554/eLife . 11189 . 008Figure 3—figure supplement 2 . Reproducibility of TEMPOL-NMR experiments . ( A ) Repeat of the experiment demonstrating ( 1 ) TEMPOL causes line-broadening at pH 4 . 5 and ( 2 ) that this line-broadening is reversible by a change to pH 7 . 5 ( cf Figure 3D ) . The signal is enhanced and returns to the expected chemical shift when the pH is raised to 7 . 5 ( bottom trace , orange ) . ( B ) Repeat of the experiment demonstrating that Y419 in the channel-like ClC-ec1 background is accessible to TEMPOL at both pH 7 . 5 and 4 . 5 ( cf Figure 3E ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11189 . 00810 . 7554/eLife . 11189 . 009Figure 3—figure supplement 3 . Overlay of WT and channel-like ClC-ec1 structures . Overlay of WT ClC-ec1 ( grey , 1OTS ) and channel-like variant E148A/Y445A ( purple , 3DET ) , RMSD 0 . 52 Å . The left panel shows a view from within the membrane; the middle and right panels shows views from the extracellular side , with the Y419 side chain depicted in pink ( channel-like ) and grey ( WT ) . The side chain is in an identical ( overlapping ) position in the two structures . DOI: http://dx . doi . org/10 . 7554/eLife . 11189 . 009 Prior to investigating the effect of [H+] on solvent accessibility of Y419 using TEMPOL , we acquired 19F spectra for Y419 only as a function of pH . The 19F NMR spectrum of Y419 only at pH 7 . 5 shows a single 19F peak centered at 60 ppm ( Figure 3B ) . This peak shifts upfield and splits into two peaks when [H+] is increased , indicating that the 19F nucleus has experienced a change in chemical environment . This result is consistent with our previous findings ( Elvington et al . , 2009 ) further supporting the notion that conformational changes occur in the vicinity of Y419 as increasing [H+] promotes occupancy of the OF state in which Gluex is protonated . The appearance of two peaks , at -61 and -63 ppm , indicates that the 19F label on Y419 is experiencing two different environments . This could arise from two conformational states of ClC-ec1 or from a tyrosine ring flip that occurs slowly on the NMR timescale ( <<1000 s-1 ) ( Weininger et al . , 2014 ) . While this information is useful in identifying Y419 as being in a region involved in H+-dependent conformational change , the lack of comprehensive theory for interpreting 19F chemical shifts in terms of structure motivates additional studies to provide more details on the nature of the conformational change . To evaluate whether there might be a change in solvent accessibility of Y419 , we examined the effect of TEMPOL on the 19F spectra of Y419only . Because of the steep distance dependence of nuclear relaxation enhancements mediated by paramagnets like TEMPOL , significant line-broadening requires the paramagnetic center to approach the target nucleus within less than ~10 Å ( Teng and Bryant , 2006 ) . At pH 7 . 5 , there is little sensitivity of the 19F-Y419 signal to 100 mM TEMPOL ( Figure 3C ) which is consistent with the largely buried position of Y419 in the crystallographically captured state of ClC-ec1 , i . e . , 12–13 Å from the protein surface ( Figure 3A ) . In contrast , at pH 4 . 5 , significant line-broadening is observed ( Figure 3C ) , indicating exposure of Y419 to the bulk solution allowing a close approach , or direct contact , of the TEMPOL probe with the fluorine atom ( Esposito et al . , 1992; Niccolai et al . , 2001 ) . This H+-dependent change in accessibility is reversible , as demonstrated by the reappearance of the Y419 signal when pH is returned to 7 . 5 from 4 . 5 in the presence of 100 mM TEMPOL ( Figure 3D ) . The reproducibility of these experiments is shown in Figure 3—figure supplement 2 . The outer- and inner-gate residues of ClC-ec1 ( Gluex and Y445 respectively , Figure 1 ) can be replaced by smaller residues Ala , Ser , or Gly , to yield “channel-like” ClC-ec1 variants ( Jayaram et al . , 2008 ) . This excavation of the gates yields a narrow water-filled conduit through the transmembrane domain , which allows rapid Cl- throughput and abolishes H+ coupling . Thus , it appears that the mechanism of Cl- flux through these variants involves channel-like diffusion that is independent of substrate-dependent conformational change . Consistent with this picture , our previous 19F NMR data showed that E148A/Y445S ClC-ec1 ( exhibiting the highest Cl- permeability among the channel-like variants ) does not undergo the substrate-dependent spectral changes observed in the coupled ClC-ec1 transporters ( Elvington et al . , 2009 ) . In this channel-like background , Y419only exhibits a single NMR peak at ~-61 ppm , and this signal is sensitive to line-broadening by TEMPOL at both pH 4 . 5 and 7 . 5 ( Figure 3E ) . The accessibility at pH 7 . 5 is surprising given that the crystal structure of channel-like ClC-ec1 variant E148A/Y445A superposes closely with WT ( RMSD 0 . 52 Å ) and indicates a buried position for Y419 ( Jayaram et al . , 2008 ) ( Figure 3—figure supplement 3 ) . While in these studies we used variant E148A/Y445S , which has not been crystallized , the two variants are functionally similar ( Jayaram et al . , 2008 ) . In channel-like E148A/Y445S , the accessibility of Y419 to TEMPOL indicates that the channel-like ClC-ec1 variant adopts a conformation in solution different from that observed in the crystal structure and similar to the conformation adopted by WT at low pH . We investigated the functional relevance of the conformational change detected at Y419 by introducing cysteines into this region and examining the effects of inter-subunit cross-linking . We reasoned that this cross-linking would restrict the conformational changes responsible for the increased solvent accessibility of Y419 at low pH , and , if these conformational changes are functionally important , cross-linking should also reduce the efficiency of Cl-/H+ transport . In the X-ray crystal structure , the Cα-Cα distance between the two Y419 residues ( one in each subunit ) is 8 . 8 Å , within striking range for potential disulfide bond formation . We found that Y419C forms spontaneous inter-subunit cross-links and , to our surprise , that these cross-links have no detectable effect on function ( Figure 4—figure supplement 1 ) . Since Y419 lies in the middle of a loop ( the P/Q linker ) , we reasoned that loop flexibility may thwart the intended restriction of motion by the disulfide cross-link . To test this possibility , we examined cross-linking at D417 , the residue immediately following Helix P , which also has an inter-subunit Cα-Cα distance of 8 . 8 Å ( Figure 4A , B ) . Like Y419C , D417C forms spontaneous inter-subunit disulfide cross-links , with ~50% of the protein migrating as a dimer on non-reducing SDS-PAGE ( Figure 4C , top panel ) . To determine the effect of the cross-link on function , we purified D417C transporters under reducing conditions , thereby obtaining a sample in which the majority ( >90% ) of the protein was not cross-linked , and then induced varying amounts of cross-link by titrating with copper-phenanthroline ( CuP ) ( Figure 4C , bottom panel ) . We assessed the functional effect of cross-linking using a Cl- efflux assay ( Walden et al . , 2007 ) . These assays show that cross-linking at 417C correlates directly with a decrease in transport activity ( Figure 4D ) , with Cl- and H+ transport inhibited in parallel ( Figure 4—figure supplement 2 ) . Controls showing the lack of effect of CuP on WT and cysteine-less transporters are shown in Figure 4—figure supplement 3 . Linear extrapolation to 100% cross-linking is summarized for all D417C variants in Table 1 . 10 . 7554/eLife . 11189 . 010Figure 4 . Cross-linking and H+-dependent conformational change at D417C . ( A ) ClC-ec1 with D417 side chain shown spacefilled , viewed from the membrane ( left ) and from the extracellular side ( right ) . ( B ) Close-up view showing D417 and Y419 side chains . ( C ) Detection of inter-subunit disulfide cross-links by non-reducing SDS-PAGE . When the D417C transporters were purified under standard ( non-reducing ) conditions , inter-subunit cross-links formed spontaneously , with ~50% of the protein migrating as a dimer ( top gel , solid arrow ) and ~50% as a monomer ( open arrow ) . By purifying the transporters under reducing conditions , the amount of cross-linking could be reduced to <10% , and then titrated ( up to ~95% ) with addition of increasing amounts of CuP ( bottom panel ) . ( D ) Effect of cross-linking on D417C activity . Left: Representative data traces showing Cl--transport activity of D417C . Right: Summary data showing Cl--transport activity as a function of disulfide cross-linking , which was determined by quantifying the relative intensities in the monomer and dimer bands detected by SDS-PAGE ( as shown in panel C ) . Each data point represents one flux-assay measurement . Error bars ( most are smaller than the symbols ) indicate the uncertainty in curve-fitting to the primary data ( transporter flux and background leak measured in control liposomes ) . Data are from three separate D417C ClC-ec1 preparations , as depicted by three colors ( purple , yellow , and blue ) . ( E ) D417C/channel-like is resistant to CuP-induced cross-linking . The bottom panel shows results from thiol quantification assays before and after treatment with 100 µM CuP . ( F ) Effect of cross-linking on activity of D417C/channel-like ClC-ec1 . Left: Representative data traces . Right: Summary data , as in panel D . Yellow and purple indicate data from two separate D417C/channel-like ClC-ec1 preparations . ( G ) DEER distance distributions reveal a pH-dependent increase in inter-subunit distance at D417C . ( H ) D417C/channel-like does not exhibit the pH-dependent change observed with D417C/WT . ( I ) Comparison of WT and channel-like D417C at pH 7 . 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 11189 . 01010 . 7554/eLife . 11189 . 011Figure 4—figure supplement 1 . Cross-linking of Y419C ClC-ec1 . ( A ) Y419C analyzed on non-reducing SDS-PAGE . Solid and open arrows indicate positions of dimeric ( cross-linked ) and monomeric ClC-ec1 respectively . Y419C purified under non-reducing conditions ( gel at left ) forms spontaneous crosslinks . Y419C purified under reducing conditions ( gel at right ) is largely uncross-linked but becomes cross-linked upon addition of 10 µM CuP . ( B ) Representative raw traces showing Cl- flux through Y419C-reconstituted proteoliposomes . ( C ) Summary data show that the Y419C crosslink has no significant effect on activity ( error bars show SEM for n=4–6 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11189 . 01110 . 7554/eLife . 11189 . 012Figure 4—figure supplement 2 . Cross-linking at D417C inhibits Cl- and H+ transport in parallel . Data represent average ± SEM ( n=4 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11189 . 01210 . 7554/eLife . 11189 . 013Figure 4—figure supplement 3 . Control experiments on WT and cysteine-less ClC-ec1 . The WT ClC-ec1 background was used for Y419C and N-deletion ( crystallization construct ) mutants; the cysteine-less ClC-ec1 background was used with all other constructs in this study . ( A ) Control experiments on WT ClC-ec1 . Left: SDS-PAGE analysis of CuP-treated WT ClC-ec1 . Arrows indicate migration position for the cross-linked dimer ( solid ) or uncross-linked monomer ( open ) . The presence of a prominent monomer band indicates that CuP does not cross-link this template . Middle: Representative traces of Cl--efflux mediated by WT or cysteine-less ClC-ec1 . Right: Relative Cl- transport rates for WT and cysteine-less ClC-ec1 ( mean ± SEM , n=3–4 ) . ( B ) Control experiments on cysteine-less ClC-ec1 . Panels as in ( A ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11189 . 01310 . 7554/eLife . 11189 . 014Figure 4—figure supplement 4 . CuP-treated D417C proteins run as dimers on size exclusion chromatography . Control: CuP-treated D417C proteins run as dimers on size exclusion chromatography ( Superdex 200 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11189 . 01410 . 7554/eLife . 11189 . 015Figure 4—figure supplement 5 . Functional- , CW-EPR , and DEER data analysis for spin-labeled D417C variants . ( A ) MTSSL-labeled D417C retains Cl--transport function , mean ± SEM for n=3–4 . Cl- turnover of labeled samples was measured after the sample was exposed to pH 4 . 5 for 1 hr at room temperature before adjusting back to pH 7 . 5 for reconstitution . ( B ) CW-EPR ( left ) , baseline-corrected DEER signals ( middle ) and fits corresponding to distance distributions ( right ) for D417C at pH 7 . 5 and 4 . 5 . ( C ) MTSSL-labeled D417C/channel-like retains Cl--transport function , mean ± SEM for n=4 . Cl- turnover of labeled samples was measured after the sample was exposed to the pH 4 . 5 condition for 1 hr at room temperature before adjusting back to pH 7 . 5 for reconstitution . ( D ) CW-EPR ( left ) , baseline-corrected DEER signals ( middle ) and fits corresponding to distance distributions ( right ) for D417C/channel-like at pH 7 . 5 and 4 . 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 11189 . 01510 . 7554/eLife . 11189 . 016Table 1 . D417C activity extrapolated to 0 and 100% cross-link . DOI: http://dx . doi . org/10 . 7554/eLife . 11189 . 016D417C variantTurnover at 0% cross-link ( s-1 ) Turnover at 100% cross-link ( s-1 ) WT1440 ± 70280 ± 70E148A/Y445S ( channel-like ) 14400 ± 1300520 ± 5400E148A260 ± 2076 ± 26Y445S850 ± 70-20 ± 105A404L140 ± 2054 ± 17L361A ( Helix N ) 360 ± 30240 ± 30F357A ( Helix N ) 84 ± 9119 ± 8Values for turnover at 0 and 100% D417C cross-link were estimated from extrapolation of fits to data in Figures 4 , 7 and 11 . The uncertainties report the 95%confidence interval in the extrapolated values . Since our NMR results indicated that the Helix P-Q region of channel-like ClC-ec1 adopts a conformation similar to that of WT at low pH ( Figure 3 ) , we hypothesized that the conformational changes underlying the increased solvent accessibility of Y419 may also move the two D417 side chains out of the range for inter-subunit cross-linking . To test this hypothesis , we generated the D417C mutant in the channel-like background and evaluated its sensitivity to cross-linking . Consistent with our hypothesis , D417C in channel-like background does not form spontaneous cross-links and is only minimally cross-linked even when treated with up to 100 µM CuP ( Figure 4E , top panel ) . Because this limited cross-linking of D417C/channel-like ClC-ec1 could be due to non-availability of the cysteines ( due to oxidation ) rather than lack of structural proximity of the two cysteine residues , we used a spectrophotometric assay to quantify the free thiols . Immediately after purification and before CuP treatment , essentially all of the 417C residues are available as free thiols ( Figure 4E , bottom panel ) . Therefore , the deficiency in cross-linking of 417C in the channel-like background compared to 417C in the WT background is not due to unavailability of the free thiols but rather because of a change in proximity of the two cysteines . After treatment with 100 µM CuP , 40% of the cysteines are available as free thiols ( Figure 4E ) . Since only ~25% had been cross-linked , this result indicates that ~35% became oxidized to other ( non-disulfide ) species . Thus , oxidation to non-disulfide species competes with disulfide bond formation and therefore thwarts any attempt to increase the extent of cross-linking beyond ~25% with longer CuP treatments . To rule out the possibility that the crosslinking might be due to inter-dimer ( rather than inter-subunit ) disulfide bond formation , we examined D417C proteins on a Superdex 200 gel filtration column both before and after treatment with CuP and found that they ran as dimers ( and not tetramers , as would occur in the case of inter-dimer cross-linking ) ( Figure 4—figure supplement 4 ) . To the degree it can be cross-linked , D417C in channel-like ClC-ec1 is inhibited similarly to D417C in WT ( Figure 4F ) . Thus , movement of Helix P away from the position observed in the crystal structures is necessary for maximal activity in both transporter and channel-like ClC-ec1 . To evaluate whether this movement of D417C/Helix P is H+-dependent ( as is the movement detected by 19F NMR , Figure 3 ) , we labeled D417C ClC-ec1 with the nitroxide spin label MTSSL ( 1-Oxyl-2 , 2 , 5 , 5 , -tetramethylpyrroline-3-methyl methanethio-sulfonate ) and used double electron-electron resonance ( DEER ) spectroscopy ( Jeschke , 2012 ) to deduce distance changes as a function of pH . At pH 7 . 5 , the distribution is dominated by a single peak at a distance shorter than 20 Å ( Figure 4G ) . Lowering the pH to 4 . 5 induces a shift to a peak at 20 Å ( Figure 4G ) . The channel-like D417C-MTSSL exhibits an altered distance distribution profile compared to the WT background ( Figure 4H ) , suggesting that the protein adopts a different conformation . Notably , the D417C-D417C distance in channel-like is increased relative to that observed in the WT background at pH 7 . 5 ( Figure 4I ) , consistent with the resistance of the channel-like protein to cross-linking ( Figure 4E ) . A decrease in pH does not shift the distance distribution as observed in the WT background , consistent with the loss of pH dependence in 19F NMR experiments on channel-like ClC-ec1 ( Figure 3E , ( Elvington et al . , 2009 ) ) . Together , these results link the conformational changes observed via NMR to those prevented by the cross-link . Crystallization of cross-linked D417C ( WT background ) confirms that the cross-link has trapped the conformation seen in the crystal structures and not some other ( non-native ) conformation . Our crystal structure , determined at 3 . 15 Å resolution , reveals a backbone that superimposes on WT ClC-ec1 with a Cα RMSD of 0 . 57 Å ( Figure 5A , Table 2 ) . Extra density connecting the 417C residues confirms the formation of an inter-subunit disulfide bridge ( Figure 5B ) . The regions around both the Cl- and the H+ permeation pathways are intact and not notably distinguishable from WT ( Figure 5C ) . To confirm the integrity of the Cl--permeation pathway in cross-linked D417C ClC-ec1 , we directly measured Cl--binding affinity using isothermal titration calorimetry ( ITC ) ( Picollo et al . , 2009 ) . Both in the absence and presence of cross-link , D417C binds Cl- robustly , with an affinity somewhat stronger than observed with WT ( Kd ~0 . 1–0 . 2 mM vs 0 . 6 mM ) ( Figure 5D , E ) . 10 . 7554/eLife . 11189 . 017Figure 5 . Structural integrity of cross-linked D417C . ( A ) The cross-linked D417C backbone ( PDB 5HD8 , green ) superposes with WT ( PDB 1OTS , blue ) ( RMSD 0 . 57 Å for 862 Cα atoms ) . ( B ) Extra density between residues 417 on the two subunits was modeled as a disulfide bridge , shown in stereo . ( C ) Close up stereo view of key residues around the Cl- ( upper panel ) and H+ ( lower panel ) permeation pathways . In the upper panel , the residues shown ( E148 , S107 , and Y445 ) are the same as those depicted in Figure 1A . In the lower panel , also shown are E203 , the internal H+-transfer site ( Accardi et al . , 2005 ) and A404 , a residue lining the portal for H+ entry from the intracellular solution ( Han et al . , 2014 ) . Cl- modeled in the central binding site is depicted as green and blue spheres . 2F0-Fc maps are contoured at 1σ . ( D ) ITC experiments show Cl- binding to WT , D417C , and D417C cross-linked with 100 µM CuP . Top panels: heat liberated when 20 mM KCl is titrated into the ITC cell containing 25–50 µM protein ( WT , 25 µM; D417C , 50 µM; D417C+CuP , 30 µM ) . ( E ) Summary data for ITC experiments , ± SEM . WT , n=3 from two separate protein preparations; D417C , n=4 from four separate preparations; cross-linked D417C n=4 from three separate protein preparations . DOI: http://dx . doi . org/10 . 7554/eLife . 11189 . 01710 . 7554/eLife . 11189 . 018Table 2 . Data collection and refinement statisticsa . DOI: http://dx . doi . org/10 . 7554/eLife . 11189 . 018Data collection Space groupC121Unit cell dimensionsa , b , c ( Å ) 231 . 7 , 96 . 1 , 170 . 0α , β , γ ( ° ) 90 , 132 , 90Resolution range ( Å ) 39 . 2–3 . 15 ( 3 . 23-3 . 15 ) Completeness ( % ) 90 . 2 ( 80 . 6 ) Rmerge ( % ) 7 . 7 ( 70 . 9 ) I/σ ( I ) 14 . 8 ( 1 . 7 ) Redundancy3 . 6 ( 2 . 1 ) Refinement statistics Resolution limit ( Å ) 39 . 2–3 . 15No . of reflections41 , 839Rwork/Rfree ( % ) 20 . 5/25 . 7Number of atoms Protein13 , 064Ligand/ion4B-factors Protein69 . 8Ions 118 . 4 r . m . s deviations Bond lengths ( Å ) 0 . 007Bond angles ( ° ) 1 . 139aValues in parentheses are for the highest-resolution shell . Data were collected from a single crystal . The inhibition of channel-like ClC-ec1 activity by the D417C cross-link suggests that inhibition occurs via an effect on the Cl--permeation pathway , given that channel-like ClC-ec1 transports only Cl- and not H+ . But since this conclusion is based on experiments with the atypical channel-like ClC-ec1 – with high transport rate and a continuous water passageway connecting the two sides of the membrane – we sought to strengthen the conclusion by examining uncoupled transporters that display typical transport rates and lack a continuous passageway: ( 1 ) E148A , which lacks the critical Gluex residue that acts both as an extracellular gate for Cl- and as a transfer-point for H+ permeation ( Figure 1 ) , and ( 2 ) Y445S , which is mutated at the intracellular gate ( Basilio et al . , 2014 ) ( Figures 1 , 2 ) . The E148A ( Gluex ) mutant is similar to channel-like ClC-ec1 in that it transports only Cl-; however , it has a much lower turnover rate , comparable to WT ( Accardi and Miller , 2004; Jayaram et al . , 2008 ) . This slow turnover suggests that despite being uncoupled E148A still depends on conformational changes to catalyze transport , a view supported by both 19F NMR and fluorescence-based experiments , which detect H+-dependent conformational change in this mutant ( Bell et al . , 2006; Elvington et al . , 2009 ) . In contrast to D417C/channel-like ClC-ec1 , we found that D417C in the E148A mutant background is readily cross-linked by CuP ( Figure 6A ) . Therefore , uncoupling through E148A alone does not alter the protein conformations sampled in solution as substantially as observed with channel-like ( E148A/Y445S ) . The turnover rate of un-crosslinked D417C/E148A is quite low – as low , in fact , as the extrapolated value for the turnover of fully cross-linked D417C ( Table 1 ) . Nevertheless , cross-linking of D417C/E148A is associated with significant inhibition of activity ( Figure 6B , Table 1 ) . This result is consistent with the conclusion that inhibition occurs via an effect on the Cl--permeation pathway . 10 . 7554/eLife . 11189 . 019Figure 6 . Cross-linking D417C in uncoupled transporter backgrounds . ( A ) D417C/E148A – detection of inter-subunit disulfide cross-links by non-reducing SDS-PAGE . ( B ) Effect of cross-linking on activity of D417C/E148A . Left: Representative data traces showing Cl--transport activity . Right: Summary data showing Cl--transport activity as a function of disulfide cross-linking . Each data point represents one flux-assay measurement , with error bars indicating the uncertainty in curve-fitting to the primary data . Purple , yellow , blue , and dark red each represent data from a separate protein preparation . ( C ) D417C/Y445S – detection of inter-subunit disulfide cross-links . ( D ) Effect of cross-linking on activity of D417C/Y445S , as in panel B . Data are from three separate protein preparations ( indicated in purple , yellow , and blue ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11189 . 01910 . 7554/eLife . 11189 . 020Figure 6—figure supplement 1 . H+ turnover of D417C/Y445S . Left: Representative data traces showing H+-transport activity . Right: Summary data showing H+-transport activity as a function of disulfide cross-linking . Each data point represents one flux-assay measurement , with error bars indicating the uncertainty in curve-fitting to the primary data . Data are from three separate preparations , with data from each preparation shown in a different color . DOI: http://dx . doi . org/10 . 7554/eLife . 11189 . 020 The second uncoupled transporter examined , the inner-gate mutant Y445S , differs from the Gluex mutant in that it is only partially uncoupled , with a Cl-/H+ stoichiometry of ~39:1 instead of the 2:1 stoichiometry observed with WT transporters ( Walden et al . , 2007 ) . The double mutant D417C/Y445S transports Cl- at ~850 s-1 ( Figure 6C ) and H+ at ~20 s-1 ( Figure 6—figure supplement 1 ) yielding a Cl-/H+ stoichiometry of ~43 , similar to that of the Y445S single mutant ( Walden et al . , 2007 ) . Cross-linking of D417C/Y445S proceeds to ~70% and inhibits Cl- transport by ~60% ( Figure 6D ) , with extrapolation to 100% cross-linking yielding a turnover of ~0 ( ± 100 s-1 ) ( Table 1 ) . For H+ turnover , it is difficult to judge whether there is a significant effect of the cross-link ( Figure 6—figure supplement 1 ) . Given the uncertainty in measuring such low H+ fluxes ( ~20 s-1 ) , it may be that H+ is inhibited to the same extent as Cl- , to a lesser degree , or not at all . In the latter cases , the cross-link would in effect “rescue” ClC-ec1 coupling; such rescue could arise from an increase in Cl- occupancy at Scen , which is known to facilitate H+ coupling ( Accardi et al . , 2006; Nguitragool and Miller , 2006; Han et al . , 2014 ) . However , we cannot distinguish these possibilities within the uncertainty of our measurements . The results with channel-like and uncoupled transporters ( Figures 4 , 6 ) support the conclusion that the D417C cross-link inhibits the Cl- branch of the Cl-/H+ transport mechanism but do not rule out an effect on H+ transport . As an approach to examine the effect of the cross-link on H+ transport , we used MD simulations to examine water entry into the hydrophobic region between Gluin and Gluex , which is essential to connect these two major H+-binding sites and thus support H+ transport ( Kuang et al . , 2007; Wang and Voth , 2009; Cheng and Coalson , 2012; Lim et al . , 2012; Han et al . , 2014 ) . Water entry occurs via a narrow portal on the cytoplasmic side of the protein , lined by Gluin together with E202 and A404 ( Lim et al . , 2012; Han et al . , 2014 ) . Previously , we showed that constricting this portal by introducing large side chains at position 404 inhibits water entry detected computationally and H+ transport detected experimentally ( Han et al . , 2014 ) . Since A404 is on the intracellular end of Helix P ( Figure 7A ) , restricting movement of this helix via the D417C cross-link might restrict water entry . To determine whether the D417C cross-link affects water entry , we compared the number of water molecules entering the central hydrophobic region during the simulation of cross-linked D417C compared to WT . In contrast to the A404L mutation , which greatly reduces water permeation through the portal ( Han et al . , 2014 ) , the D417C cross-link has no effect on water entry ( Figure 7B ) . This result suggests that the cross-link reduces ClC-ec1 transport predominantly via an effect on the Cl--permeation pathway rather than on the H+-permeation pathway . 10 . 7554/eLife . 11189 . 021Figure 7 . Computational analysis of water entry through the portal lined by A404 ( Helix P ) . ( A ) ClC-ec1 structure highlighting the location of the A404 “portal” residue at Helix P . ( B ) The D417C cross-link does not affect water entry into the pathway connecting Gluin and Gluex . The aggregate number of water molecules entering the region between the two residues was determined as described previously ( Han et al . , 2014 ) and compared for wild-type ( WT ) and cross-linked mutant ( D417C ) over the same timescales . DOI: http://dx . doi . org/10 . 7554/eLife . 11189 . 021 Our experimental results suggest that there could be functionally important motions of ClC-ec1 that open the Cl--transport pathway and are impeded by the D417C cross-link . To investigate the molecular basis of such motions , extensive MD simulations were conducted either in the absence or in the presence of the D417C cross-link . Note that even the hundreds of nanoseconds of simulations performed here can probe mainly conformational fluctuations of ClC-ec1 near its reference conformation ( in this case , the crystal structure ) , which did not permit direct observation of the opening of the gates . Nevertheless , the sampled dynamics and fluctuations can provide information that can be used to derive collective motions , which are often functionally relevant ( Bahar et al . , 2010 ) . Collective motions are defined as those involving concerted movements of a large number of atoms distributed throughout the protein , and are therefore distinguished from localized conformational changes . A series of collective motions of a protein can be obtained in general by decomposing the fluctuations of a protein sampled through MD simulations , e . g . , through principal component analysis , or by analyzing normal modes of the protein that underlie protein motions ( Bahar et al . , 2010; Gur et al . , 2013 ) . Collective motions can further be used to probe how larger-magnitude conformational change along the identified displacement vectors ( modes ) might involve crucial , functionally-relevant protein motions , such as opening-closing movements of enzymatic active sites , ligand-binding sites on receptors and channel pores ( Tai et al . , 2001; Lou and Cukier , 2006; Shrivastava and Bahar , 2006; Liu et al . , 2008; Jiang et al . , 2011; Isin et al . , 2012; Peters and de Groot , 2012; Fan et al . , 2013; Yao et al . , 2013 ) . For example , collective motions obtained from normal mode analysis ( NMA ) were used to project opening movements of potassium-channel pores ( Shrivastava and Bahar , 2006 ) , and these predicted movements are consistent with those seen in single-molecule and X-ray crystallographic experiments ( Shimizu et al . , 2008; Alam and Jiang , 2009 ) . In this study , we identified collective motions in ClC-ec1 using principal component analysis ( PCA ) of the equilibrium MD simulations ( see Methods ) , which in general identifies similar collective motions to those derived from NMA ( Leo-Macias et al . , 2005; Yang et al . , 2008; Skjaerven et al . , 2011 ) . We then introduced deformations in the reference protein structure along each of the top 20 collective motions identified in our analysis ( ~75% of the motions observed in the equilibrium MD simulation ) . We then examined whether increasing the amplitude of these collective motions ( which overcomes timescale limitations of the simulation ) confer conformational change to the Cl--transport pathway . We specifically examined regions around the extracellular and intracellular gates to the Cl--transport pathway , where motions may lead to opening of either gate ( which are both closed in the reference protein structure ) . The extracellular gate is formed by the juxtaposition of Helix F ( which contains Gluex ) and Helix N ( Figure 1B ) . To scrutinize opening of this gate , we examined Cα distance changes ( Δr ) between several residue pairs on these helices: I356-G149 , F357-E148 , and A358-R147 ( Figure 8A , B ) . A search over the 20 dominant collective motions obtained through the PCA of the entire WT MD simulation revealed that deformations along some of the collective motions increase the distances between these pairs by >1 . 5 Å and thus are coupled to gate opening . To conduct a statistical analysis of these motions , we divided the entire simulation trajectory into six blocks and determined the number of times such collective motions occur in each block ( Figure 8C , blue bars ) . An identical analysis performed on the MD simulation trajectory obtained from the cross-linked D417C mutant revealed that the motions that open the extracellular gate are dampened due to the cross-linking ( P=0 . 002 – 0 . 008 ) ( Figure 8C , orange bars ) . The intracellular gate is formed by two key residues S107 and Y445 ( Walden et al . , 2007; Accardi and Picollo , 2010; Basilio et al . , 2014 ) ( Figure 2A ) . To scrutinize opening of this intracellular gate , we examined distance changes between these two residues as a result of collective motions . As with the extracellular gate , we observed some collective motions that lead to distance changes ( Δr ) of > 1 . 5 Å . Unlike the extracellular gate , however , the cross-link at residue 417 does not significantly dampen the distance changes around the intracellular gate ( P=0 . 338 ) ( Figure 8D ) . 10 . 7554/eLife . 11189 . 022Figure 8 . Coupling of extracellular and intracellular gating motions to collective motions in ClC-ec1 detected computationally . ( A ) Key inter-Cα distances were employed to detect functional motions . The left panel shows the location of the Cl- gates ( dashed box ) and transport pathways ( dashed green line ) in ClC-ec1 . Right panel shows a close-up of the Cl- gates where key inter-Cα distances for both the extracellular and intracellular gates are denoted by dashed double arrows . ( B ) Scheme for determining distance change ( Δr ) caused by a collective motion . Following a collective motion , a native structure ( red helices ) undergoes structural transition ( peach helices ) . As a result , the distance between the helices increases by Δr = r’ – r . ( C ) Opening motions of the extracellular gate . The number ( N ) of collective motions that lead to distance changes ( Δr > 1 . 5 Å ) at each of the extracellular-gate residue pairs was determined from analysis of MD simulations for WT and cross-linked ( “D417C x-link” ) ClC-ec1 , as described in the text . The data are shown in a box-and-whisker plot where the whiskers denote minimum and maximum of the data and the box denotes the range of 25th percentile to 75th percentile of the data when sorted . The horizontal line in the box denotes the median of the data . ( D ) The number ( N ) of collective motions that lead to distance changes ( Δr > 1 . 5 Å ) at the intracellular gate pair 107–445 is not significantly different between WT and cross-linked ClC-ec1 . DOI: http://dx . doi . org/10 . 7554/eLife . 11189 . 022 The comparison of dominant gate-opening motions between WT and cross-linked forms described above suggests that the cross-link at residue 417 likely cripples the opening of the extracellular gate , thereby slowing Cl- transport . However , along this line of reasoning , one must reconcile why the E148A mutants , in which the extracellular gate has ostensibly been removed , are inhibited when the cross-link is introduced . To address this question , we first investigated the bottleneck for Cl- transport in ClC-ec1 based on the crystal structures . The radius profile of the ClC-ec1 Cl- transport tunnel , calculated using the program HOLE ( Smart et al . , 1996 ) , shows an extracellular bottleneck with a minimum radius of ~0 . 2 Å ( Figure 9 ) . Interestingly , the calculated radius profile for both the E148A mutant ( lacking Gluex ) and the channel-like variant E148A/Y445A also reveal extracellular bottlenecks . ( E148A/Y445A was evaluated rather than the E148A/Y445S construct used here because this is the only channel-like variant for which there is a crystal structure . ) With minimum radii of ~0 . 9 Å ( Figure 9 ) these bottlenecks are still too narrow to allow Cl- permeation ( r ( Cl- ) ≈ 1 . 81 Å ) ( Shannon , 1976 ) . Thus , additional opening motions in the gate region are needed for Cl- transport . 10 . 7554/eLife . 11189 . 023Figure 9 . The extracellular gate remains narrow in the Gluex mutant ( E148A ) and in the channel-like variant E148A/Y445A . The pore radius profiles of the ClC-ec1 Cl- transport tunnel for WT ClC-ec1 ( blue ) , E148A ( pink ) and E148A/Y445A ( green ) along the z-axis ( membrane normal ) . Shown are the profiles for subunit 1 . The results for subunit 2 are very similar and thus not shown . The z-position of the central Cl--binding site is chosen as the origin of the z-axis . The shaded region denotes the extracellular-gate region; dashed arrows highlight the z-positions of the bottlenecks . DOI: http://dx . doi . org/10 . 7554/eLife . 11189 . 02310 . 7554/eLife . 11189 . 024Figure 9—figure supplement 1 . Radius pore profile of 1KPL ( CLC structure determined at pH 4 . 6 ) . The extracellular gate is narrow in the Salmonella CLC ( StCLC ) structure determined at pH 4 . 6 . The pore radius profiles of the Cl- transport tunnel for ClC-ec1 ( blue ) and StCLC ( orange ) along the z-axis ( membrane normal ) . Shown are the profiles for subunit 1; the results for subunit 2 are very similar and thus not shown . The z-position of the central Cl--binding site is chosen as the origin of the z-axis . The shaded region denotes the extracellular-gate region; dashed arrows highlight the z-positions of the bottlenecks . StCLC exhibits an additional bottleneck towards the extracellular side of the ion-permeation pathway See also Figure 1—figure supplement 1 for a comparison of these two structures . DOI: http://dx . doi . org/10 . 7554/eLife . 11189 . 024 To test the idea that additional gate-opening motions occur in the absence of Gluex , the computational analysis discussed above was applied to characterize and analyze the collective motions of channel-like ClC-ec1 . The analysis revealed that there are fewer collective motions that can open the extracellular gate after the cross-link is introduced to the channel-like mutant ( P=0 . 001–0 . 070 ) ( Figure 10A ) , whereas the intracellular gate was not significantly affected ( P=0 . 354 ) ( Figure 10B ) . This result is consistent with that obtained in the WT background . Taken together , our MD results suggest that the cross-link at residue 417 hinders the opening of the extracellular gate – beyond the Gluex motions – in both the WT and channel-like ClC-ec1 . 10 . 7554/eLife . 11189 . 025Figure 10 . Cross-linking at 417 impedes opening of the extracellular but not the intracellular gate in channel-like ClC-ec1 , as detected by computational analysis . ( A ) Opening motions of the extracellular gate . The number ( N ) of collective motions that lead to distance changes ( Δr > 1 . 5 Å ) at each of the extracellular-gate residue pairs was determined from analysis of MD simulations for WT and cross-linked ( “x-link” ) channel-like ClC-ec1 , as described in the text . The data are shown in a box-and-whisker plot where the whiskers denote minimum and maximum of the data and the box denotes the range of 25th percentile to 75th percentile of the data when sorted . The horizontal line in the box denotes the median of the data . ( B ) The number ( N ) of collective motions that lead to distance changes ( Δr > 1 . 5 Å ) at the intracellular gate pair 107–445 is not significantly different between WT and cross-linked channel-like ClC-ec1 . DOI: http://dx . doi . org/10 . 7554/eLife . 11189 . 025 How are motions at Helix P transmitted to the extracellular gate ? Visual inspection reveals an obvious potential transduction pathway: Helix N , which forms part of the extracellular gate ( Figure 1B , Figure 8A ) , makes direct contacts to Helix P through side-chain packing of conserved residues in each Helix ( Figure 11A , B ) . We hypothesized that disrupting these contacts would disrupt transduction of Helix-P motions to the extracellular gate , thereby abolishing the inhibitory effect of the Helix-P cross-link . To test this hypothesis , we generated Helix-N mutants F357A and L361A , in which the inter-helical coupling of motion is expected to be weakened by removing bulky side chains contributing to the contact area . The mutant transporters are slow compared to WT but retain the ability to couple Cl-/H+ exchange ( Figure 3—figure supplement 1 ) . Strikingly , the D417C cross-link only weakly inhibits L361A activity and completely fails to inhibit F357A ( Figure 11C , D ) . The sluggish turnover of the F357A mutant suggests that it might be insensitive to the D417C cross-link because it is already maximally inhibited . To evaluate this possibility , we examined another slow mutant , A404L . A404 lines an intracellular “portal” for water ( and hence H+ ) entry into the transporter ( Han et al . , 2014 ) . This residue is located at the N-terminal end of Helix P ( Figures 7A , 11A ) , which does not contact Helix N . We found that the activity of the A404L mutant , despite being similarly sluggish to F357A , is reduced further yet by the D417C cross-link ( Figure 11E , F ) . Thus , the lack of sensitivity of F357A to the D417C cross-link appears due to the weakened interaction with Helix P and not to its already-low turnover . These results provide strong support for the hypothesis that Helix-P motions are transmitted to the extracellular gate via side-chain contacts to Helix N . 10 . 7554/eLife . 11189 . 026Figure 11 . Helix P is coupled to the extracellular gate via Helix N . ( A ) Side view of ClC-ec1 , in stereo . Conserved residues L411 and M415 in Helix P ( blue ) make direct contact with conserved residues F357 and L361 in Helix N ( yellow ) . ( B ) Close-up of Helices P and N . ( C ) Detection of inter-subunit disulfide cross-links by non-reducing SDS-PAGE in Helix-N mutants D417C/L361A ( top ) and D417C/F357A ( bottom ) . ( D ) Effect of cross-linking on activity . Left: Representative data traces showing Cl--transport activity of D417C/L361A and D417C/F357A . Right: Summary data showing Cl--transport activity as a function of disulfide cross-linking . Each data point represents individual data points as described in Figure 4 . Purple , yellow and blue each represent data obtained from a separate protein preparation . ( E ) Detection of inter-subunit disulfide cross-links on D417C/A404L ( F ) Effect of cross-linking on activity of D417C/A404L . Purple , yellow and blue represent data obtained from separate protein preparations . DOI: http://dx . doi . org/10 . 7554/eLife . 11189 . 026
Our results describe a previously unidentified protein conformational state and suggest a new framework for understanding the CLC transport mechanism , introducing two key concepts . First , the structure of the E148Q mutant , with the side chain rotated away from Sext ( Figure 1C ) represents an “outward-facing occluded” ( OFoccluded ) state ( Stein and Litman , 2014 ) , in which bound Cl- does not have full access to the extracellular solution . Second , H+ binding promotes an “outward-facing open” ( OFopen ) state , involving conformational rearrangement of Helices N and P ( Figure 11A ) , that widens the extracellular ion-permeation pathway in comparison to the known crystal structures . The first clue to conformational change at Helix P came from our NMR studies of Y419 , on the short P/Q linker , where unambiguous changes in both chemical shift and solvent accessibility of 19F-labeled Y419 are observed when the pH is lowered from 7 . 5 to 4 . 5 ( Figure 3B–D ) . At pH 7 . 5 , the lack of accessibility is consistent with the crystal structure of the occluded conformational state , which depicts Y419 in a buried position . At pH 4 . 5 , the increased accessibility of Y419 indicates a conformational state different from that captured in crystal structures . This state ( with Y419 exposed to solution ) is observed in channel-like ClC-ec1 at both pH 7 . 5 and 4 . 5 ( Figure 3E ) . This shift in equilibrium distribution of conformational states for channel-like ClC-ec1 is useful because it enables comparison of the disulfide cross-linking of the two states , which must be done at a pH that is amenable to disulfide bond formation ( 7 . 5 rather than 4 . 5 ) . In the WT background , cross-links near Y419 , at D417C , form readily ( Figure 4C ) , as expected based on the crystal structure of the occluded conformational state ( Figure 5A–C ) . In contrast , in the channel-like E148A/Y445S background , D417C is resistant to cross-linking ( Figure 4E ) . These results suggest that the pH-dependent conformational change detected by NMR involves a change in inter-subunit proximity of D417 residues in addition to the change in solvent accessibility of Y419 . DEER experiments confirm such pH-dependent change at D417 ( Figure 4G ) . Inter-subunit cross-linking of D417C restricts the conformational transition to the OFopen state and inhibits activity . The inhibition occurs not only in the WT background but also in uncoupled E148A , Y445S , and E148A/Y445S ( channel-like ) backgrounds ( Figures 4 and 6 ) . Therefore , the conformational change being restricted is something other than the localized movements of side-chain gates , as these gates ( E148 and Y445 ) are missing altogether in the uncoupled transporters . To gain insight into how conformational change near the subunit interface affects activity , we performed MD simulations on WT and channel-like ClC-ec1 , with and without the D417C crosslink . We found that the major motions of both WT and channel-like involve opening of the extracellular vestibule and that these opening motions are dampened by the cross-link at D417 ( Figures 8 , 10 ) . Further , our mutagenesis experiments show that removing side-chain interactions between Helices N and P eliminates the effect of the cross-link on Cl- transport ( Figure 11 ) . Therefore , we conclude that rearrangement of these helices facilitates a widening of the extracellular ion-permeation pathway . The residual activity remaining with maximal cross-linking at D417 ( ranging from 0–300 s-1 , Table 1 ) suggests that the OFoccluded state may allow some minimal level of Cl- flux . However , an alternative interpretation is that the OFoccluded is completely impermeant to Cl- and that the residual transport observed with the cross-link is either ( 1 ) not distinguishable from zero ( due to compounding uncertainties in the various steps involved in the experimental measurement , including quantification of the fraction cross-linked ) or ( 2 ) occurs because the cross-link does not completely prevent movement of Helix N and opening of the extracellular vestibule to the OF state . We favor the alternative interpretation as it is in keeping with the general principles of transporter function , in which protein conformational change plays a key role in sustaining coupling stoichiometry . In support of this idea , we note that Helix N motions have been strongly implicated not only in ClC-ec1 ( the results presented here ) but also in the mammalian antiporter CLC-4 ( Osteen and Mindell , 2008 ) . Experiments on this homolog identified an inhibitory Zn2+-binding site at the top of Helix N that appears to transmit conformational change to the Cl--permeation pathway at the other end of Helix N ( Osteen and Mindell , 2008 ) . While it is clear that rearrangement of Helices N and P is required for opening the extracellular vestibule , the precise molecular details of this rearrangement remain to be determined . Nevertheless , several pieces of information suggest that the overall motions , though long-range in effect , may involve rearrangements/reorientations of only a few Angstroms in magnitude . First , the cross-linking of Y419C , just 5 Å away from D417C , does not inhibit function ( Figure 4—figure supplement 1 ) . Second , any large movement of Helix P would likely have a major effect on water entry via the narrow portal that is the rate-limiting barrier for formation of water wires and H+ transport ( Lim et al . , 2012; Han et al . , 2014 ) . Since our computational analysis indicates that cross-linking does not significantly affect water entry ( Figure 7 ) , Helix-P motion may involve only a small tilt or rotation that exerts a “lever-arm” effect on Helix N and the Cl--entryway . Third , previous inter-subunit cross-linking studies targeting Helices I and Q , and the H-I and I-J loops showed that simultaneously cross-linking these regions had no significant effect on function ( Nguitragool and Miller , 2007 ) and therefore argue against a major restructuring of the inter-subunit interface . Together , these results suggest that the rearrangements at Helices N and P are likely small in magnitude and do not involve the entire inter-subunit interface . This conclusion is in line with computational studies using normal-mode and functional-mode analysis , which showed the subunit interface remaining largely intact even as other regions of ClC-ec1 underwent global conformational changes to alternately expose Cl-- and H+-binding sites during the exchange process ( Miloshevsky et al . , 2010; Krivobokova et al . , 2012 ) . One of the mobile helices identified in these computational studies was Helix R , which has also been pinpointed in experimental studies of H+-dependent conformational change ( Bell et al . , 2006; Abraham et al . , 2015 ) . Since Helix R extends from the center of the protein ( where Y445 coordinates Cl- , Figure 1A ) out to the cytoplasmic solution ( Figure 1B ) , the H+-dependent conformational change characterized here , while not large in magnitude , may extend well beyond the immediate region around Helices P and N . To integrate the OFopen state into a model of the CLC transport cycle , we build on the model of Basilio et al . ( Basilio et al . , 2014 ) . Starting with the OFoccluded state ( State 1 in Figure 12A , reflecting the state captured in the E148Q crystal structure , Figure 1B , C ) , a conformational change generates the OFopen state ( State 2 ) . This conformational change is pH dependent ( Figures 3 , 4 ) but need not be promoted solely by the protonation of Gluex , as suggested by previous observations of H+-dependent conformational changes in Gluex mutants ( Bell et al . , 2006; Elvington et al . , 2009 ) . The conformational change allows 2 Cl- ions to exit to the extracellular side ( State 3 ) . Entry of the protonated Gluex into the vacated permeation pathway ( State 4 ) facilitates transfer of one H+ to the intracellular side , via water wires and the internal H+-transfer site Gluin ( Figure 1A ) ( Accardi et al . , 2005; Lim and Miller , 2009; Lim et al . , 2012; Han et al . , 2014 ) . Upon unbinding of H+ , the protein adopts the apo occluded conformation ( State 5 ) which can then undergo conformational change to the inward-facing state ( State 6 , ( Basilio et al . , 2014 ) ) . Binding of 2 Cl- from the intracellular side knocks Gluex out of the Sext-binding site ( State 7 ) , which then allows H+ binding from the extracellular side ( back to State 1 ) . This revised model is completely consistent with previous experimental observations , and the addition of new conformational states adds potentially key control points to the mechanism . First , the extracellular occlusion in State 7 assures no extra Cl- slips through during the step in which Cl- binds from the intracellular side . Second , we hypothesize that the OFopen state lowers Cl- affinity and promotes Cl- release , as suggested by the increase in Cl--binding affinity observed when formation of the OFopen state is inhibited by the D417C cross-link ( Figure 5D , E ) . 10 . 7554/eLife . 11189 . 027Figure 12 . Revised model of the CLC transporter mechanism . ( A ) CLC transporter cycle . The OFoccluded state ( 1 ) undergoes a conformational change to OFopen ( 2 ) . This step is pH-dependent but may be promoted by protonation of residues other than Gluex ( see Discussion ) . Two Cl- ions leave ( 3 ) and then entry of the protonated Gluex into the permeation pathway ( 4 ) facilitates H+-transfer to the inside ( via Gluin , Figure 1B ) ( 5 ) . Conformational change to the inward-facing state ( 6 ) allows 2 Cl- ions to enter from the intracellular side , knocking Gluex out of the pathway ( 7 ) . The cycle is reversible , with protonation favoring conformational change to the OFopen state . ( B ) Channel-like CLC states . The crystal structure of channel-like ClC-ec1 reveals a narrow constriction at the extracellular-gate region , depicted at left . However , results here demonstrate that the major conformation adopted in solution more closely resembles the OFopen state ( equilibrium shifted to right ) . This finding is consistent with the high Cl- throughput observed in channel-like ClC-ec1 . DOI: http://dx . doi . org/10 . 7554/eLife . 11189 . 027 Our revised model also sheds light on the mechanism of channel-like ClC-ec1 . Previously , it was recognized that the narrow pathway depicted by the crystal structures of channel-like ClC-ec1 is not sufficiently wide to allow rapid ion conduction and that protein dynamics ( either breathing or conformational change ) must play an important part in the mechanism of ion conduction ( Jayaram et al . , 2008 ) . Our results clarify the issue by showing that channel-like ClC-ec1 populates a conformation different from that seen in the crystal structure and exhibiting similarities to the new OFopen state characterized in these studies . In this state , the region of the narrowest constriction – just above Sext – is significantly widened ( Figure 12B ) . The population of this state explains why channel-like ClC-ec1 can conduct Cl- rapidly at pH 7 . 5 . The long-range conformational change described here improves our understanding of CLC mechanisms by providing a first glimpse of an “outward-facing open” CLC conformational state and its mechanistic implications . In future studies , it will be important to investigate the transition between the OFopen , OFoccluded and inward-facing conformational state ( s ) . Using a cross-linking strategy , Basilio et al . showed that transition to the inward-facing state involves motion of the intracellular half of Helix O . This motion is thought to be limited in scope , as it is relayed directly to the intracellular gate via a steric interaction between intracellular-gate residue Y445 ( Figure 2A ) and Helix O residue I402 ( Basilio et al . , 2014 ) . Nevertheless , since Helix O also makes direct contacts to Helices N and P ( studied here ) , it seems likely that intracellular and extracellular gate-opening motions will be linked through the interaction of these three helices . Understanding these interactions will be critical to providing a molecularly detailed view of the CLC transport mechanism .
Expression and purification of unlabeled ClC-ec1 WT and mutant proteins was performed as documented in detail ( Accardi and Miller , 2004 ) except that the final purification step was by size exclusion chromatography on a Superdex gel filtration column ( Walden et al . , 2007 ) rather than ion-exchange chromatography . Point mutations introduced by conventional PCR methods were confirmed by sequencing . D417C constructs were made in a previously characterized cysteine-less background C85A/C302A/C347S ( Nguitragool and Miller , 2007 ) , which here is referred to as the “WT background” . For preparing ClC-ec1 under reducing conditions , 20 mM β-mercaptoethanol ( β-ME ) and 1 mM dithiothreitol ( DTT ) ( Fisher Scientific , Pittsburgh , PA ) were added to cell pellets during resuspension , and 1 mM DTT was included in subsequent purification steps . DTT was removed in the final purification step over a Superdex 200 size exclusion column . To measure turnover rates in flux assays , ClC-ec1 variants were reconstituted into liposomes by dialysis ( Walden et al . , 2007 ) into buffer R ( 300 mM KCl , 40 mM Na-citrate , pH 4 . 5 ) using 0 . 2 – 5 μg protein per mg of E . coli polar lipids ( Avanti Polar Lipids , Alabaster , AL ) . For the high-turnover channel-like variant , the lower end of this range ( 0 . 2 µg protein per mg lipids ) was used . For experiments to determine stoichiometry , protein to lipid ratio was 0 . 4–10 μg protein per mg lipid ( with higher ratios used for low-turnover mutants ) . Reconstituted liposomes were subjected to 4 freeze-thaw cycles and were extruded through 400-nm filters 15 times using an Avanti Mini-Extruder . Liposomes were buffer-exchanged through Sephadex G-50 spin columns ( Basilio and Accardi , 2015 ) into flux-assay buffer ( 300 mM K-isethionate , 50 µM KCl , buffered with 2 or 40 mM Na-citrate pH 4 . 5 ) . ( The 2 mM Na-citrate buffer was used in experiments in which Cl- and H+ transport were measured in parallel; the 40 mM Na-citrate buffer was used in experiments in which only Cl- transport was measured . ) Transport was initiated by addition of 2 µg/mL valinomycin ( for dual Cl-/H+-transport measurements ) or 3 µg/mL CCCP + 7 µg/mL valinomycin ( for Cl--transport measurements ) ( Han et al . , 2014 ) . At the end of each flux-assay experiment , total liposomal Cl- was determined by disrupting the liposomes with Triton X-100 ( 0 . 01%; from a 10% stock solution ) ; flux-assay traces shown in Figures 4 , 6 and 11 show normalization to this value . Transport turnover rates were calculated by measuring the initial velocity of the Cl- and/or H+ transport ( Walden et al . , 2007 ) . Stoichiometry was determined from the ratio of the Cl- to the H+ turnover rate . Flux assays were performed in sets of 20–40 samples; within each set , an assay was discarded if the total liposomal [Cl-] ( a measure of the yield of reconstituted liposomes , which affects the accuracy of the unitary-turnover calculation ) was >30% outside of the mean . Flux-assay measurements were performed on at least 4 samples for each condition . This sample size and selection method is based on previous experience with flux-assay measurements ( Howery et al . , 2012; Han et al . , 2014 ) . 19F-Tyr labeling was performed as described ( Elvington et al . , 2009 ) . Labeled ClC-ec1 was purified into Buffer A ( 150 mM NaCl , 10 mM HEPES ( Fisher Scientific , Pittsburgh , PA ) , pH 7 . 5 and 5 mM n-decyl β-D maltopyranoside ( DM ) ( Anatrace , Maumee , OH ) , then concentrated to approximately 50 µM . E . coli polar lipids were added in a 1:80 lipid:detergent molar ratio to the BuriedOnly construct to enhance stability ( Elvington et al . , 2009 ) . The Y419Only construct was more stable without the addition of lipids . 10% D2O was added prior to NMR experiments . Samples ( ~300 µL starting volumes ) were placed in the outer tube of Shigemi symmetrical microtubes in order to reduce the volume of sample required for data acquisition . The Shigemi tube insert was not used so as to avoid generating froth from adjusting the plunger in the detergent containing sample . Data were collected using a 5 mm H/F probe on a Bruker Avance 500 MHz spectrometer running Topspin version 1 . 3 with variable temperature control . Data represent acquisition of 30 – 50k transients at 470 MHz; 12 kHz spectral width; 45° pulse; 0 . 17s acquisition time; 1 . 8 – 2 . 8s relay cycle; 20°C; 15 Hz linebroadening; referenced to TFA . The pH of the samples was lowered to 4 . 5 using a 1 M citric acid solution ( EMD Millipore , Billerica , MA ) and raised to 7 . 5 using a 1M Tris-acetate pH 9 . 0 solution . TEMPOL ( 4-Hydroxy-2 , 2 , 6 , 6-tetramethylpiperidine 1-oxyl , Fluka Analytical , Ronkonkoma , NY ) was added to the sample by carefully weighing out and adding the solid reagent required to attain a final concentration of 100 mM in the NMR sample . All procedures were carried out at room temperature ( 21–23ºC ) . Stock solutions of CuP at 10x were made from 1:3 mixtures of CuSO4 ( aqueous ) ( MCB Reagents , Cincinnati , OH ) and 1 , 10-phenanthroline ( in ethanol ) ( Sigma-Aldrich , St . Louis , MO ) . ClC-ec1 eluted from the Superdex 200 column in Buffer A was diluted to 0 . 2 mg/mL ( 1 . 9 μM homodimer; 3 . 8 µM Cys-containing subunits ) before addition of CuP . After an hour of incubation , 1 mM Na-EDTA ( Fisher Scientific , Pittsburgh , PA ) was added to terminate the cross-linking reaction . Cross-linking was visualized using SDS/PAGE ( 4–15% gradient gels ) and staining with Coomassie brilliant blue ( TCI America , Portland , OR ) . Cross-linking was documented using an Odyssey Infrared Imaging System ( LI-COR Biosciences ) using the 700 nm channel . ClC-ec1 band intensities were quantified using NIH ImageJ software . Un-cross-linked ClC-ec1 runs as a monomer ( apparent molecular weight ~36 kD ) and cross-linked mutant as a dimer ( apparent molecular weight ~64 kD ) . The fraction cross-linked was calculated based on the relative intensities of the dimer and monomer bands . For channel-like ClC-ec1 , which exhibited a low efficiency of cross-linking , free thiols were quantified colorimetrically ( Life Technologies Thiol and Sulfide quantification kit , T6060 ) . During reconstitution into liposomes , most CuP-treated samples were dialyzed in buffer containing 1 mM DTT in order to avoid additional cross-linking during the dialysis step; this level of DTT was sufficiently low that it did not reduce D417C disulfide bonds that had already been formed . Mutant proteins D417C/F357A and D417C/L361A were reconstituted in the absence of DTT when crosslinked with 100 µM CuP , as an extra precaution to avoid disulfide-bond reduction in these samples . For preparing D417C ClC-ec1 ( WT and channel-like backgrounds ) for DEER experiments , 20 mM β-ME was added to cell pellets during resuspension . β-ME was removed during washing and elution from the cobalt column . Proteins eluted from the cobalt column were incubated with 50x molar excess of the paramagnetic spin label MTSSL ( Enzo Life Sciences , Farmingdale , NY ) that was dissolved in small volume of dimenthylformamide ( DMF ) ( Fisher Scientific , Pittsburgh , PA ) such that the final DMF concentration was < 0 . 1% . The protein solution was sealed under argon and mixed by slow rotation ( ~20 rpm ) for 2 hr at room temperature . The remaining steps of the purification were identical to our usual ClC-ec1 purifications . Thus , the 6-His tag was then removed by a one-hour incubation with endoprotease Lys-C ( Roche Diagnostics , Indianapolis , IN ) . The ClC-ec1 samples were then purified from the cleaved 6-His tag and excess MTSSL by size exclusion chromatography on a Superdex 200 column . Glycerol ( 23% v/v ) was added to the protein solution as cryoprotectant . This was achieved by adding an 80% ( v/v ) glycerol stock solution ( prepared in buffer A ) to the purified protein . The samples were then concentrated to a final concentration of 50–100 µM , and E . coli polar lipids were added at 1:80 lipid:detergent molar ratio . A stock solution of 25 mM citrate was used to adjust the sample at pH 7 . 5 to pH 4 . 5 . Functional assays were performed on EPR samples that had been exposed to pH 4 . 5 for one hour before reconstitution . CW-spectra were collected on a Bruker EMX at 10 mW power with a modulation amplitude of 1 . 6G . Spectra were normalized to the double integral . DEER experiments were carried out using a standard four-pulse protocol ( Jeschke , 2012 ) . Samples were maintained at 83K . DEER distributions were obtained from fitting the DEER decays to a sum of Gaussian distributions ( Brandon et al . , 2012; Mishra et al . , 2014; Stein et al . , 2015 ) . For crystallization , the D417C mutant was put into a deletion construct ( ΔNC ) lacking N-terminal residues 2–16 and C-terminal residues 461–464 ( Lim et al . , 2012 ) . Purified ΔNC-D417C was cross-linked with 100 µM CuP for 1 h , incubated with excess Fab fragment ( Dutzler et al . , 2003 ) for 30 min , then purified by size exclusion chromatography ( Superdex 200 ) into buffer containing 100 mM NaCl , 5 mM DM , 10 mM Tris ( Fisher Scientific , Pittsburgh , PA ) , pH 7 . 5 . The complex was concentrated to 10–12 mg/mL and mixed with 30% PEG 400 ( Hampton Research , Aliso Viejo , CA ) , 0 . 075 M K/Na-tartrate ( Fluka Analytical , Ronkonkoma , NY ) , 0 . 1 M Tris HCl ( MP Biomedicals , Santa Ana , CA ) ( pH 9 . 0 ) . Crystals were grown by the sitting drop method for 2–4 weeks at 20oC and were directly harvested from the reservoir , flash frozen and stored in liquid N2 . Diffraction data were collected to 0 . 9795 Å at the BL12-2 beamline ( SLAC ) and processed using XDS ( Kabsch , 2010 ) . Phases were obtained by molecular replacement with the WT protein in complex with Fab ( PDB 1OTS ) using the MOLREP program ( Vagin and Teplyakov , 2010 ) . Refinement was done using the refmac program ( Murshudov et al . , 1997 ) . Atomic coordinate and structure factors are deposited in the Protein Data Bank under accession code 5HD8 . ITC was carried out using a MicroCal VP-ITC instrument . Chloride binding to WT and mutants were carried out as described previously ( Picollo et al . , 2009; Howery et al . , 2012 ) . Briefly , ClC-ec1 ( WT or D417C or D417 cross-linked using 100 µM CuP ) was purified over a Superdex 200 size exclusion column pre-equilibrated with Buffer B ( 100 mM K+-Na+-tartrate , 20 mM HEPES , 5 mM DM , pH 7 . 5 ) and then concentrated to 25–50 µM . Percent cross-link following treatment with 100 μM CuP was 92 . 0 ± 0 . 6% ( n=2 ) . The injection syringe was filled with Buffer B containing 20 mM KCl . Each experiment consisted of 30 10-µL injections of the Cl--containing solution at 5 min intervals , to achieve a final molar ratio of 50–160 . The chamber was kept at 25oC with constant stirring at 350 rpm . All solutions were filtered and degassed before use . ITC data were fit to a single-site isotherm as described with Origin 7 MicroCal program . The ClC-ec1 crystal structure at 2 . 51 Å ( PDB ID: 1OTS ) ( Dutzler et al . , 2003 ) was used to prepare for the MD simulations of all the systems studied in the present work – WT , D417C , channel-like ( E148A/Y445S ) , and D417C/channel-like . The system setup for the WT ClC-ec1 is detailed in our previous work ( Han et al . , 2014 ) . In short , to have the protein hydrated properly , all the crystallographic water molecules were maintained and 49 additional water molecules were added using DOWSER ( Zhang and Hermans , 1996 ) . One additional water molecule was placed between Gluex ( E148 ) and the Cl- ion bound to the central ion-binding site of ClC-ec1 ( Figure 1 ) in order to stabilize the two closely ( within ~4 Å ) positioned negative charges , as suggested in previous simulation studies ( Bostick and Berkowitz , 2004; Cohen and Schulten , 2004; Wang and Voth , 2009 ) . Gluex ( E148 ) and Gluin ( E203 ) were both deprotonated , while E113 was modeled in its protonated form according to previous Poison-Boltzmann electrostatic calculations ( Faraldo-Gomez and Roux , 2004 ) . The protein was embedded into a POPE lipid bilayer , fully equilibrated TIP3P water ( Jorgensen et al . , 1983 ) and buffered in 150 mM NaCl , resulting in a 105 × 105 × 110 Å3 box with ~110 , 000 atoms . The mutant systems were constructed on the basis of that of the WT . For each mutant , residue substitutions were done using the Mutator plugin of VMD ( Humphrey et al . , 1996 ) . Disulfide bonds were constructed by introducing geometric restraints on two cysteine residues , including a distance restraint between the sulfur atoms and angular restraints involving Cβ atom of either cysteine and the two sulfur atoms . To avoid structural disruption of the protein due to sudden introduction of restraints , the disulfide restraints were turned on gradually over 20-ns simulations . Note that the systems prepared as such are not significantly different from the cross-linked D417C crystal structure . In fact , during the equilibrium simulation of the mutant containing the disulfide bond ( see below ) , the RMSD of the protein to the D417C crystal structure is on average ~1 . 6 Å , even smaller than its RMSD ( ~1 . 9 Å ) to the WT crystal structure that the simulation started from . All MD simulations were carried out with NAMD 2 . 9 ( Phillips et al . , 2005 ) using the CHARMM-CMAP ( Mackerell et al . , 2004 ) and CHARMM36 force fields ( Klauda et al . , 2010 ) to model the proteins and lipids , respectively . The particle mesh Ewald ( PME ) ( Darden et al . , 1993 ) method was used to calculate long-range electrostatic forces without truncation . All simulation systems were subjected to Langevin dynamics and the Nosé-Hoover Langevin piston barostat ( Nose , 1984; Hoover , 1985 ) for constant pressure ( P = 1 atm ) and temperature ( T = 310 K ) ( NPT ) . Each system was energy-minimized for 5 , 000 steps , followed by a 1-ns MD run with positions of all protein atoms and oxygen atoms of the crystallographic water molecules restrained . Each system was simulated without any restraints for ~300 ns . The collective motions of the protein were analyzed through principal component analysis ( PCA ) of the equilibrium MD trajectories ( Amadei et al . , 1993 ) . Specifically , we first constructed the covariance matrix C of Cα atoms of select parts of the proteins for each subunit based on equilibrium MD trajectories . The covariance matrix C was calculated as cij = < ( xin - <xin> ) ( xjn - <xjn> ) > , where Xn={xin} are the coordinates of Cα atoms of select parts of protein in the nth sampled structure and the brackets <> denote the averages over all the sampled structures . The first 50 ns of each MD trajectory were discarded to remove any initial bias . Only the transmembrane helical regions were selected for this analysis as they define the overall architecture of the protein and most relevant to the functionally relevant global motions . We then derived orthonormal eigenvectors R={Rk} of the covariance matrix C . Each eigenvector Rk={rik} defines relative movement ( rik ) of each select atom in a collective motion of the protein represented by the eigenvector . The 20 eigenvectors with the largest eigenvalues were chosen for further analysis . These eigenvectors correspond to the collective motions that account for >75% of protein motion observed in the simulations . Following the approach by Bahar and co-workers ( Isin et al . , 2008 ) , conformational deformation driven by a given collective motion can be calculated according to the associated eigenvector Rk as follows: ( 1 ) X'=X0±ARk where X0 and X’ denote the coordinates of the reference structure and the structure of the protein deformed by the collective motion , and A is an arbitrary scaling factor determining the extent of structural deformation to be examined . The value of A is related to the RMSD between the reference and the deformed structures through the relationship RMSD = A/M1/2 , where M is the number of atoms selected to calculate RMSD ( here M=538 , the number of Cα atoms located in the transmembrane helical region of the protein ) . To make a meaningful comparison of all collective motions investigated , the value of A was chosen such that the structure of the protein is altered by each motion to the same extent , targeting always a total RMSD of 3 . 5 Å with respect to the original structure . Thus , the distance change ( Δr ) between two sites ( xi and xj ) of interest ( Figure 8B ) can be calculated according to ∆r = |x'i -x'j| - |xi-xj| . Finally , we quantified the protein’s ability of opening its gates via collective motions by counting the dominant collective motions that involved an increase in the distance between residues lining the gates by Δr > 1 . 5 Å . To achieve a statistical estimate of such counts , each ~300-ns simulation trajectory of the homodimer was divided evenly into three time blocks ( Rapaport , 2004 ) , each being analyzed through the procedure described above , providing a dataset of six segments ( three time blocks for each subunit x 2 subunits ) . Statistical comparisons between datasets were made using the Wilcoxon-Mann-Whitney test ( Mann and Whitney , 1947 ) . | Cells have transporter proteins on their surface to carry molecules in and out of the cell . For example , the CLC family of transporters move two chloride ions in one direction at the same time as moving one hydrogen ion in the opposite direction . To be able to move these ions in opposite directions , transporters have to cycle through a series of shapes in which the ions can only access alternate sides of the membrane . First , the transporter adopts an 'outward-facing' shape when the ions first bind to the transporter , then it switches into the 'occluded' shape to move the ions through the membrane . Finally , the transporter takes on the 'inward-facing' shape to release the ions on the other side of the membrane . However , structural studies of CLCs suggest that the structures of these proteins do not change much while they are moving ions , which suggests that they might work in a different way . Khantwal , Abraham et al . have now used techniques called “nuclear magnetic resonance” and "double electron-electron resonance" to investigate how a CLC from a bacterium moves ions . The experiments suggest that when the transporter adopts the outward-facing shape , points on the protein known as Y419 and D417 shift their positions . Chemically linking two regions of the CLC prevented this movement and inhibited the transport of chloride ions across the membrane . Khantwal , Abraham et al . then used a computer simulation to model how the protein changes shape in more detail . This model predicts that two regions of the transporter undergo major rearrangements resulting in a gate-opening motion that widens a passage to allow the chloride ions to bind to the protein . Khantwal , Abraham et al . ’s findings will prompt future studies to reveal the other shapes and how CLCs transition between them . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2016 | Revealing an outward-facing open conformational state in a CLC Cl–/H+ exchange transporter |
Autophagy is a conserved cellular process involved in the elimination of proteins and organelles . It is also used to combat infection with pathogenic microbes . The intracellular pathogen Legionella pneumophila manipulates autophagy by delivering the effector protein RavZ to deconjugate Atg8/LC3 proteins coupled to phosphatidylethanolamine ( PE ) on autophagosomal membranes . To understand how RavZ recognizes and deconjugates LC3-PE , we prepared semisynthetic LC3 proteins and elucidated the structures of the RavZ:LC3 interaction . Semisynthetic LC3 proteins allowed the analysis of structure-function relationships . RavZ extracts LC3-PE from the membrane before deconjugation . RavZ initially recognizes the LC3 molecule on membranes via its N-terminal LC3-interacting region ( LIR ) motif . The RavZ α3 helix is involved in extraction of the PE moiety and docking of the acyl chains into the lipid-binding site of RavZ that is related in structure to that of the phospholipid transfer protein Sec14 . Thus , Legionella has evolved a novel mechanism to specifically evade host autophagy .
Autophagy is an evolutionarily conserved intracellular degradation process in eukaryotes , which is essential for cellular homeostasis in response to various environmental and cellular stresses . During autophagy , a double-membrane structure , termed as phagophore ( or isolation membrane ) engulfs the cytosolic components and closes to form an autophagosome , which subsequently fuses with a lysosome , resulting in degradation of internal contents ( Mizushima and Komatsu , 2011 ) . Autophagosome formation is the key process in autophagy . The biogenesis of autophagosomes is believed to be dependent on the formation of lipidated Atg8 . Microtubule-associated protein light chain 3 ( LC3 ) and GABARAP family proteins are the orthologs of yeast Atg8 in mammalian cells . Lipidated and membrane-associated Atg8/LC3 has been used as a bona fide marker of autophagosomes and progression of autophagy ( Fujita et al . , 2008; Sou et al . , 2008 ) . Production of lipidated Atg8/LC3 is controlled by two ubiquitin-like conjugation systems . Newly synthesized Atg8/LC3 is processed by a protease , Atg4 , to expose a C-terminal glycine . The resulting Atg8/LC3 is conjugated to phosphatidylethanolamine ( PE ) through a ubiquitin-like conjugation reaction controlled by Atg7 , Atg3 and the Atg12-Atg5:Atg16 complex . The Atg12-5:16 complex is generated by another ubiquitin-like conjugation system controlled by Atg7 and Atg10 . Atg4 releases lipidated Atg8/LC3 from the surface of autophagosomes ( Kimura et al . , 2007; Kirisako et al . , 2000; Tanida et al . , 2004; Xie et al . , 2008 ) . However , the role of Atg8-PE and its regulators in autophagosome formation remain poorly understood . Autophagy also serves as a defense mechanism against invading pathogens ( termed xenophagy ) ( Deretic , 2011; Levine , 2005 ) . Xenophagy recognizes bacteria through autophagy receptors that contain two crucial domains , the ubiquitin-binding domain ( UBD ) and LC3-interacting region ( LIR ) motifs , which are important for cargo recognition and interaction with the LC3 proteins , respectively . The receptor binds to the ubiquitinated pathogen through its UBD and recruits it to the autophagosome membrane via the interaction of the LIR motifs to LC3 proteins ( Kirkin et al . , 2009; Korac et al . , 2013; Mostowy et al . , 2011; Thurston et al . , 2009; von Muhlinen et al . , 2012; Wild et al . , 2011 ) . LIR motifs are also present in receptor and scaffold proteins involved in other selective autophagy processes and play an essential role in recruiting components of the autophagy machinery to phagophores ( Klionsky and Schulman , 2014; Noda et al . , 2010 ) . However , it is not clear how LIR motifs selectively recognize mammalian Atg8 family members . Some bacteria have evolved specific mechanisms to avoid autophagy or even hijack the autophagy machinery in order to survive in the cell ( Choy and Roy , 2013 ) . The pathogenic bacterium Legionella pneumophila , which causes Legionnaire’s disease , manipulates the core machinery of autophagosome formation to evade host autophagy . L . pneumophila inhibits autophagy by injecting an effector protein called RavZ into the cytoplasm . RavZ functions as a cysteine protease and irreversibly deconjugates mammalian Atg8s from PE to inhibit autophagosome formation ( Choy et al . , 2012 ) . Unlike Atg4 that cleaves the amide bond between terminal glycine and PE , RavZ cleaves the amide bond before glycine . As a consequence , the RavZ-cleaved Atg8 proteins cannot be relipidated , leading to inhibition of autophagy . RavZ represents an interesting pathogenic effector , functional characterization of which will shed light on the mechanism of autophagosome biogenesis . To date , how RavZ recognizes and deconjugates LC3-PE is not known . This is largely due to the previously insurmountable difficulties in recombinant preparation and handling of lipidated LC3 proteins . Herein , semi-synthetic LC3 proteins make it possible to elaborate the mechanism of RavZ function . We have used chemical methods to produce LC3 proteins with different C-terminal modifications , enabling the analysis of structure-function relationships of LC3-deconjugation by RavZ , allowing formulation of a membrane extraction model . We find that RavZ extracts LC3-PE from membranes and then deconjugates C-terminal Gly-PE . We show that the second N-terminal LIR motif ( LIR2 ) is required for RavZ activity and RavZ:LC3 interaction . The crystal structures of RavZ:LC3 and LIR2:LC3 complexes and interaction analysis suggest RavZ initially recognizes LC3 mainly via its LIR2 motif . We identify the lipid-binding site ( LBS ) of RavZ , which shows a similar fold to that of the LBS of yeast phospholipid transfer proteins ( Sec14 family ) . The LBS involves a highly dynamic and hydrophobic helix α3 that is engaged in association with the membrane and plays an essential role in extraction of the conjugated PE from the membrane . Therefore , by a combination of chemical , biophysical and cell biological approaches , we gain insights into a novel mode of host-pathogen interaction .
Advances in protein ligation methods have provided a powerful tool for studying post-translational modified proteins ( Dawson and Kent , 2000; Hackenberger and Schwarzer , 2008; Vila-Perelló and Muir , 2010 ) . In order to address the mechanism of Legionella effector RavZ function in host autophagy , we sought to produce LC3 proteins with various C-terminal modifications by expressed protein ligation ( EPL ) and direct aminolysis of protein thioesters ( Figure 1 ) . Previously , we reported the semi-synthesis of lipidated protein LC3-PE using a combination of lipidated peptide synthesis and EPL ( Yang et al . , 2013 ) . An MBP-assisted solubilization strategy was used to facilitate ligation under folding conditions and to solubilize the lipidated protein without detergents and membranes . 10 . 7554/eLife . 23905 . 003Figure 1 . Semisynthesis of LC3 proteins with various C-terminal modifications . ( A ) Strategy for the semisynthesis of lipidated LC3 proteins using expressed protein ligation ( EPL ) . ( B ) Strategy for the semisynthesis of LC3 proteins containing soluble fragments of PE by direct aminolysis . DOI: http://dx . doi . org/10 . 7554/eLife . 23905 . 00310 . 7554/eLife . 23905 . 004Figure 1—figure supplement 1 . Synthesis of peptides and PE fragments . ( A ) Synthesis of C-terminal lipidated peptides of LC3 . ( B ) Chemical synthesis of compounds 8 and 10 . DOI: http://dx . doi . org/10 . 7554/eLife . 23905 . 004 The C-terminal peptides containing DPPE ( 16:0 ) ( 1 , 2-dipalmitoyl-sn-glycero-3-phosphoethanolamine ) , DHPE ( 6:0 ) ( 1 , 2-dihexanoyl-sn-glycero-3-phosphoethanolamine ) and 1-hexadecanol ( C16 ) were ligated with MBP-LC31–114-thioester ( Figure 1—figure supplement 1 and Supplementary file 1 ) . Direct aminolysis strategy ( Payne et al . , 2008; Yi et al . , 2010 ) was used to produce LC3 proteins with modification of the soluble PE fragments , including ethanolamine ( EA , 1 ) , phosphoethanolamine ( pEA , 2 ) , glycerophosphoethanolamine ( GpEA , 8 ) and diacetyl glycerophosphoenthanolamine ( DAGpEA , 10 ) ( Figure 1; Supplementary file 2 and Supplementary file 3 ) . First of all , we examined whether RavZ can act on semisynthetic LC3-PE . RavZ only cleaved LC3-PE but not pro-LC3 , whereas Atg4 cleaved both substrates ( Figure 2—figure supplement 1A ) . In keeping with previous studies , Atg4B treatment led to LC31–120 , whereas RavZ-mediated deconjugation resulted in LC31–119 ( Figure 2—figure supplement 1B ) . In a previous report , RavZ deconjugated LC3-PE from membranes ( Choy et al . , 2012 ) . However , our results suggest that RavZ can cleave LC3-PE without the requirement for membranes . The question then arises as to how RavZ recognizes LC3-PE . Based on the results shown above , two possible hypotheses could be made . First , RavZ would recognize a certain soluble fragment derived from the PE rather than the lipid chain , which could not be ‘seen’ by RavZ when it is buried in the membrane ( Figure 2A ) . Second , the fatty acid chain would be required for binding to RavZ . In this case , LC3-PE would have to be extracted from the membrane by RavZ before proteolytic cleavage could occur . To distinguish these scenarios , a structure-function relationship study of RavZ-mediated deconjugation is required ( Figure 2B ) . The semisynthetic LC3 proteins make it possible to perform such an analysis . LC3 proteins with different C-terminal modifications were subjected to RavZ and Atg4B treatment . The measurements showed that RavZ cannot cleave LC3 proteins containing soluble fragments derived from PE , while Atg4B is active toward these proteins ( Figure 2—figure supplement 2A ) . RavZ only cleaves PE-modified LC3 proteins , with a preference for long fatty acid chain . Enzyme kinetic data show that the catalytic efficiency of RavZ for LC3-PE ( 16:0 ) ( kcat/Km = 1140 M−1·s−1 ) is 12 times higher than that for LC3-PE ( 6:0 ) ( kcat/Km = 93 . 8 M−1·s−1 ) ( Figure 2B; Figure 2—figure supplement 2B–D ) . LC3-C16 cannot be processed by RavZ , suggesting that LC3 with a single fatty acid chain is not sufficient for hydrolysis by RavZ ( Figure 2—figure supplement 2E ) . In contrast , Atg4B can cleave all modified LC3 proteins tested ( Figure 2C; Figure 2—figure supplement 2A and E ) . Taken together , in contrast to Atg4B activity that does not require any specific structure downstream of the C-terminal glycine , RavZ activity is strictly dependent on conjugated PE structures . Therefore , it is conceivable that RavZ contains a lipid-binding site and can extract LC3-PE from the membrane before cleavage . This extraction model is further confirmed in later experiments . 10 . 7554/eLife . 23905 . 005Figure 2 . Structure-function relationship of LC3-deconjugation by RavZ . ( A ) Schematic diagram of proteo-membrane containing LC3-PE . The fatty acid chains of PE are buried in the lipid bilayer , serving as the membrane anchor . ( B ) LC3 proteins containing different fragments derived from the PE . Ethanolamine , EA; phosphoethanolamine , pEA; glycerophosphoethanolamine , GpEA and diacetyl glycerophosphoenthanolamine , DAGpEA . ( C ) In vitro cleavage of semisynthetic LC3 proteins by Atg4B and RavZ . DOI: http://dx . doi . org/10 . 7554/eLife . 23905 . 00510 . 7554/eLife . 23905 . 006Figure 2—figure supplement 1 . In vitro cleavage of LC3 proteins with various C-terminal modifications . ( A ) In vitro LC3 cleavage assay by SDS-PAGE . RavZ only cleaves LC3-PE but not pro-LC3 , while Atg4B processes both of them . ( B ) RavZ and Atg4 deconjugate LC3-PE at different sites . ESI-MS spectra of MBP-LC3-PE before and after Atg4B and RavZ treatments were shown . MBP-LC3-PE , Mw calculated 60078 , found 60075; MBP-LC31–120 , Mw calculated 59391 , found 59402; MBP-LC31–119 , Mw calculated 59344 , found 59344 . ( C ) The C-terminal lipidated peptide of LC3-PE is insufficient for RavZ proteolysis . The C-terminal lipidated peptide of LC3-PE and chimeric PE-modified Rab7 were subjected to RavZ treatment . The lipidated peptides were solved in the HEPES buffer containing detergent ( 30 mM HEPES 7 . 4 , 50 mM NaCl , 2 mM DTT , 0 . 1% Triton X-100 ) . RavZ ( 2 µM ) was added to the peptide solution ( 2 µM ) and chimeric PE-modified Rab7 ( 7 µM ) . After overnight incubation at 37°C , the reactions were subjected to LC-MS . No cleavage products were observed in these conditions . Right panel shows ESI-MS spectra of EGFP-Rab7 thioester , EGFP-Rab7-PE chimeric protein and RavZ-treated EGFP-Rab7-PE . Chimeric EGFP-Rab7-PE protein , Mw calculated 50500 , found 50485 . DOI: http://dx . doi . org/10 . 7554/eLife . 23905 . 00610 . 7554/eLife . 23905 . 007Figure 2—figure supplement 2 . Structure-function relationship study of LC3-deconjugation by RavZ . ( A ) ESI-MS spectra of modified LC3 ( S1–S4 ) before ( black line ) and after Atg4B ( red line ) or RavZ treatment ( blue line ) . 0 . 7 µM LC3 proteins were incubated with 0 . 7 µM Atg4B or RavZ for 8 hr at 37°C . ( B ) Kinetics of RavZ-medicated cleavage reactions with LC3-PE ( 16:0 ) and LC3-PE ( 6:0 ) ( S5 ) . 7 µM LC3-PE ( 16:0 ) or LC3-PE ( 6:0 ) was incubated with RavZ ( 0 . 7 µM ) at 37°C . The reaction was quenched at different time points and subjected to SDS-PAGE . The graphs in the lower panel show reaction progress . The red solid line shows the fitting to a single-exponential equation to obtain t1/2 . ( C ) Enzyme kinetics of RavZ reaction . Varying concentrations of LC3-PE ( 16:0 ) or LC3-PE ( 6:0 ) were incubated with RavZ ( 0 . 7 µM ) for 8 min or for 140 min at 37°C , respectively . The reaction was quenched and subjected to SDS-PAGE . In the lower panel , the reaction rates were plotted against substrate concentrations . The red solid line shows the fitting to Michaelis-Menten equation . ( D ) Summary of Michaelis-Menten kinetic parameters . n = 3 independent experiments . Mean and SD are presented . ( E ) In vitro cleavage assay of LC3-C16 ( S6 ) . LC3-C16 was treated with Atg4B or RavZ ( 0 . 7 µM ) at 37°C for 2 hr and subjected to SDS-PAGE . ( F ) Kinetics of RavZ20-331-medicated cleavage reaction with LC3-PE ( 16:0 ) . The same conditions were used as in ( B ) . ( G ) The exponential decay curve showing the half-time ( t1/2 ) of LC3-PE cleavage activity of RavZ20-331 . n = 3 independent experiments . Mean and SD are presented . DOI: http://dx . doi . org/10 . 7554/eLife . 23905 . 007 Interestingly , RavZ did not process PE-peptides containing 6 , 7 or 11 C-terminal amino acid residues . To further confirm the findings , we ligated the CQETFG-PE peptide to the C-terminus of Rab7 GTPase . Again , RavZ cannot hydrolyze the chimeric Rab7-PE protein ( Figure 2—figure supplement 1C ) . These results suggest that a C-terminal PE-modified peptide is not sufficient for substrate recognition and cleavage by RavZ . The structure-function relationship studies prompted us to examine the extraction model of RavZ . To this end , we used protease-deficient RavZC258A in an LC3-PE extraction assay ( Figure 3A ) . MCF-7 cells stably expressing GFP-LC3 were subjected to starvation to induce autophagy . The membrane fraction of cells was incubated with different concentrations of RavZC258A protein . The supernatant was precipitated using TCA/DOC ( trichloroacetic acid/sodium deoxycholate ) . The soluble proteins in supernatant and the membrane-associated proteins were visualized by immunoblotting with anti-LC3 antibody . Our results showed that RavZC258A treatment decreased the level of membrane association of both endogenous LC3-II ( lipidated LC3 ) and GFP-LC3-II in a dose-dependent manner . Accordingly , the level of soluble LC3-II and GFP-LC3-II increased with increasing concentration of RavZC258A . These results suggest that RavZC258A extracts lipidated LC3 from membranes . 10 . 7554/eLife . 23905 . 008Figure 3 . RavZ extracts LC3-PE from membranes . ( A ) LC3-PE extraction assay . Endogenous LC3-PE and GFP-LC3-PE on membranes from GFP-LC3 stable cells ( 20 µg ) were treated with increasing concentrations of recombinant RavZC258A . After incubation for 2 hr at 37°C , the membrane and soluble fractions were separated by ultracentrifuge and subjected to western blotting with LC3-specific antibody . Upper panel shows the schematic diagram of LC3-PE extraction by RavZC258A . ( B ) Confocal microscopy of GFP-LC3 stable cells transfected with mCherry-RavZ constructs . Cells were treated for 2 hr under starvation medium ( EBSS ) . A complete set of images for GFP-LC3 cells transfected with all mCherry-RabZ constructs ( WT , C258A , and their mutants ) are shown in Figure 6—figure supplement 2A and B . ( C ) Quantification of GFP-LC3 puncta shown in ( B ) . n = 21–32 cells . Mean and SD are presented; ***p<0 . 001 , *p<0 . 05 . ( D ) Effect of RavZ on LC3-II level in cells . GFP-LC3 stable cells were transfected with wild-type RavZ ( WT ) and RavZC258A and cultured either in growth or in starvation medium . ( E–G ) Quantification of ratio of endogenous LC3-II to LC3-I ( E ) , GFP-LC3-II to GFP-LC3-I ( F ) and p62 to actin ( G ) in ( D ) . n = 3 independent experiments . Mean and SD are presented; ***p<0 . 001 , *p<0 . 05 , ns: not significant . ( H ) Validation of LC3-PE extraction by RavZ in the cell . GFP-LC3 stable cells were transfected with RavZC258A . After starvation , cells were lysed . Membrane fractionation was performed by ultracentrifuge . The membrane and soluble fractions were subjected to western blotting . DOI: http://dx . doi . org/10 . 7554/eLife . 23905 . 00810 . 7554/eLife . 23905 . 009Figure 3—figure supplement 1 . Membrane binding and α3 are required for extraction activity of RavZ . ( A ) Confocal microscopy of HeLa cells co-transfected with GFP-LC3 and mCherry-RavZ constructs after starvation for 2 hr in EBSS . ( B ) Quantification of GFP-LC3 puncta in ( A ) . n = 25–30 cells . Mean and SD are presented; ***p<0 . 001 , ns: not significant . ( C ) GFP-LC3 expression levels in stable and transient cells by western blot . ( D ) Fluorescence microscopy of GFP-LC3 transiently transfected cells . The percentage of transfection efficiency = ( number of cells stained with fluorescent/total number of cells per field ) X 100 . ( E ) Quantification of the ratio of transient and stable GFP-LC3 expression level to actin in ( C ) . n = 3 . Mean and SD are presented . ( F ) LC3-PE extraction assay . GFP-LC3-PE on membranes from GFP-LC3 stable cells ( 20 µg ) was treated with increasing concentrations of recombinant RavZC258A , α3-m3 ( C258A ) mutant or 1-331 ( C258A ) fragment . After incubation for 2 hr at 37°C , the membrane and soluble fractions were separated by ultracentrifuge and subjected to Western blotting with LC3-specific antibody . ( G ) Quantification of the LC3-II level in membrane and supernatant fractions in ( F ) . n = 3 independent experiments . Mean and SD are presented . DOI: http://dx . doi . org/10 . 7554/eLife . 23905 . 009 To further evaluate the extraction activity of RavZ in vivo , GFP-LC3 stable cells were transfected with mCherry , mCherry-RavZ wt or mCherry-RavZC258A . Both RavZ wt and RavZC258A significantly inhibited GFP-LC3 puncta formation ( Figure 3B and C ) . However , the inhibitory effect of RavZC258A was not observed in a previous report , where GFP-LC3 and RavZC258A were transiently expressed in the cell ( Choy et al . , 2012 ) . We repeated the experiment using cells transiently expressing GFP-LC3 . We also find that RavZ wt but not RavZC258A inhibits GFP-LC3 puncta formation under these conditions , probably due to the high expression level of GFP-LC3 in transient cells ( Figure 3—figure supplement 1A–1E ) . RavZ wt but not RavZC258A led to decrease in LC3-II level in GFP-LC3 cells ( Figure 3D–3F ) . Moreover , lipidated LC3 was found in the soluble fraction in RavZC258A-expressing cells , which is in contrast to the control , where lipidated LC3 is only found in the membrane fraction ( Figure 3H ) . Both RavZ wt and RavZC258A significantly inhibited p62 degradation , a marker for autophagy flux ( Figure 3D and G ) . Therefore , the inhibitory effect of RavZC258A is a result of extraction of lipidated LC3 molecules from membranes , leading to accumulation of LC3-II in the cytosol and reduction of membrane-localized LC3 proteins . This scenario is quite different from the effect of overexpression of Atg4BC74A ( a protease-deficient mutant ) , which inhibits LC3 lipidation by sequestration of unlipidated LC3 through formation of stable complexes in the cytosol ( Fujita et al . , 2008 ) . Since the lipidated LC3 level is not changed in RavZC258A-expressing cells , RavZC258A inhibits autophagy by sequestrating lipidated LC3 proteins rather than unlipidated LC3 proteins . These findings indicate that RavZ does not interfere with autophagy machinery upstream of Atg8/LC3 but specifically processes lipidated LC3 proteins . The importance of the LIR motif for recognition of Atg8/LC3 in selective autophagy prompted us to identify LIR motifs in RavZ . The LIR motif is composed of conserved W/Y/FxxL/I/V sequence . The binding of LIR motifs with Atg8/LC3 proteins is quite conserved , with two key hydrophobic residues playing an essential role in interaction with the hydrophobic pocket of the Atg8s ( Ichimura et al . , 2008; Noda et al . , 2008 , 2010 ) . Three potential LIR motifs of RavZ were identified by the iLIR online server ( Kalvari et al . , 2014 ) ( Figure 4A ) . The first and second LIR motifs are located at the N-terminal region of RavZ ( residues 14–19 , LIR1; and 27–32 , LIR2 ) , while the third motif is in the C-terminal region ( residues 433–438 , LIR3 ) . To identify which LIR motif is involved in RavZ activity and to map the catalytic domain of RavZ , RavZ was dissected into fragments and subjected to the in vitro LC3-PE cleavage assay and to measurement of binding with LC3 by a fluorescence polarization assay ( Figure 4A; Figure 4—figure supplement 1B ) . Truncation of LIR1 ( 20–502 ) and LIR3 ( 1–431 ) did not influence RavZ activity and led to only up to ca . twofold reduction of binding affinity with LC3 , whereas simultaneous truncation of LIR1 and LIR2 ( 33–502 and 55–487 ) completely abolished cleavage activity and reduced the binding affinity with LC3 by 2–4 fold ( Figure 4A; Figure 4—figure supplement 1B ) . Mutation of the key aromatic residue ( F29A ) of LIR2 led to dramatic reduction of RavZ activity , whereas mutation of LIR1 ( F16A ) did not affect RavZ activity . The RavZF16/F29A mutant showed the same effect as RavZF29A ( Figure 4C and D ) . These results demonstrate that LIR2 is crucial for RavZ activity . However , fragments containing one of the LIR motifs can still form stable complexes with LC3 . In contrast , RavZ fragments ( 55–430 and 55–331 ) without LIR motifs completely lost their binding ability to LC3 ( Figure 4A; Figure 4—figure supplement 1A and B ) . Therefore , LIR2 is crucial for both binding and cleavage activities . Indeed , RavZ fragment 20–331 that contains only LIR2 can cleave LC3-PE with only twofold decrease in binding affinity to LC3 . Thus , RavZ20-331 represents the minimal catalytic domain . Prompted by these results , we tested binding of the LIR2 peptide ( DIDEFDLLEGDE ) to LC3 by ITC and binding of the fluorescein-labeled LIR2 peptide to LC3 by fluorescence polarization , leading to dissociation constants ( Kd ) of 360 nM and 550 nM , respectively , which is comparable with that of RavZ:LC3 complex ( Kd = 260 nM ) ( Figure 4—figure supplement 1C and D ) . Therefore , the LIR2 motif of RavZ contributes the major binding energy for RavZ:LC3 interaction . 10 . 7554/eLife . 23905 . 010Figure 4 . The LIR motif is essential for RavZ:LC3 binding and cleavage activity of RavZ . ( A ) Schematic diagram of RavZ fragments showing LIR motifs and the catalytic residue Cys258 . Binding and cleavage activities of the fragments are shown . ( B ) In vitro LC3-PE cleavage assay of the RavZ fragments . ( C ) In vitro LC3-PE cleavage assay of RavZ containing LIR mutations . The LC3-PE protein ( 7 µM ) was treated with RavZ mutants ( 700 nM ) at 37°C for 10 min or 20 min and then resolved by SDS-PAGE . ( D ) Quantification of the LC3-PE cleavage in ( C ) . n = 3 independent experiments . Mean and SD are presented; **p<0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 23905 . 01010 . 7554/eLife . 23905 . 011Figure 4—figure supplement 1 . LIR2 is required for RavZ:LC3 interaction . ( A ) Gel filtration analysis of complex formation of LC31–119 with RavZ1–502 or RavZ55-430 . Orange line indicates elution profile of the mixture of RavZ and LC3 , blue and green lines indicate profiles of RavZ and LC3 alone , respectively . Gray dashed line indicates elution profile of protein markers ( from left to right ) : 670 , 158 , 44 , 17 and 1 . 3 kDa . The black line underneath the profile indicates collected fractions for SDS-PAGE analysis shown below . ( B ) Interaction analysis of RavZ fragments with LC31–119 by fluorescence polarization . Dissociation constants are shown in the table ( right ) . ( C ) Interaction analysis of the FITC-LIR2 peptide with LC3 by fluorescence polarization . Dissociation constants are shown in the table ( right ) . ( D ) Interaction analysis of LIR2 with LC3 using ITC technique . Upper panel is thermographic binding isotherm and lower panel is the fit of the binding . * The Kd was obtained from three independent ITC experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 23905 . 011 Next , we solved the structures of two RavZ fragments , 1–487 and 20–502 . Crystal structures of the fragments are essentially identical , solved in space group I422 . RavZ structures of both fragments are resolved from residues 47 to 433 and 48 to 432 , respectively , whereas the missing regions at the N- and C-termini ( residues 1–47 and 433–502 ) might be flexible and disordered . Our RavZ structures are identical to the reported RavZ structure ( PDB:5CQC ) with an RMSD of 0 . 33 ( Figure 5—figure supplement 2A ) ( Horenkamp et al . , 2015 ) . RavZ shows two major domains , the N-terminal catalytic domain ( residues 49–320 , catalytic triad: C258 , H176 , D197 ) and the C-terminal domain ( 329-423 ) . In line with the crystal structure , RavZ1–331 showed similar LC3-binding affinity and catalytic efficiency to the wild-type RavZ ( Figure 4A; Figure 2—figure supplement 2F , Figure 2—figure supplement 2G and Figure 4—figure supplement 1B ) , suggesting the catalytic domain is sufficient for deconjugation of LC3-PE . The C-terminal domain has been recently shown to be involved in association with phosphatidylinositol 3-phsphate ( PI3P ) and be essential for membrane binding ( Horenkamp et al . , 2015 ) . Interestingly , although the catalytic domain is sufficient for cleavage , the extraction activity is reduced ( Figure 3—figure supplement 1F and G ) . Since the catalytic domain does not bind to membranes ( Horenkamp et al . , 2015 ) , membrane binding of RavZ is required for extraction of LC3-PE in vivo . To understand how RavZ recognizes LC3 before extraction , we solved the crystal structure of the RavZ1–431:LC31–119 complex at 2 . 5 Å ( Supplementary file 4 ) . The complex shows 1:1 stoichiometry . However , the complex displays only a relatively small binding interface ( 300 Å2 ) . Nevertheless , the structure of RavZ in the complex shows a well-defined resolution of flexible loops , which are missing in the free-structure . The loop ( 249–253 ) , loop ( 278–287 ) and the α8-α9 loop ( 346–356 ) are now visible ( Figure 5—figure supplement 2B ) . Importantly , the N-terminal loop containing the LIR2 motif ( residues 25 to 48 ) is resolved in the structure . LIR2 of RavZ interacts with the α2-β3 loop of LC3 instead of the canonic LIR-binding site that involves hydrophobic pockets located in β1 , β2 and α3 ( Noda et al . , 2008 , 2010 ) . Further analysis of the crystal contacts between symmetry-related RavZ molecules ( RavZa and RavZb ) shows that LIR2 of RavZa interacts extensively with RavZb , involving the α3 loop and the β6-β7 loop ( Figure 5A ) . Comparing with the free-RavZ structure , the α3 loop in the complex moves ca . 9 Å inward , suggesting a dynamic nature of this region ( Figure 5B ) . Key hydrophobic interactions include F29 ( RavZa ) with Y211 ( RavZb ) , L31 ( RavZa ) with I170 ( RavZb ) , L208 ( RavZb ) and F212 ( RavZb ) . It is conceivable that the crystal packing led to a low-energy binding configuration involving interaction of LIR2 with the α3 loop of its symmetric mate . Such an interaction is less likely to exist outside the crystal , because the affinities of the α3 mutant ( Y211D/F212D/Y216D ) and the wild-type RavZ toward LC3 are identical ( Figure 6—figure supplement 2C ) . The result indicates that there is no defined LC3-binding site on RavZ except for the LIR2 loop . 10 . 7554/eLife . 23905 . 012Figure 5 . Structural basis of RavZ:LC3 interaction . ( A ) Crystal structure of the RavZ:LC3 complex . Catalytic domain , the PIP3-binding domain and the N-terminal loop containing LIR2 of RavZa are shown in blue , teal and pink , respectively . The LC3 molecule is shown in green with its α2 and the α2–β3 loop colored in orange . The symmetry-related RavZ ( RavZb ) molecule is shown in pale cyan with its α3 and the the β6–β7 loop colored in yellow . ( B ) Binding interface between LC3 with RavZa and RavZb . Free RavZ structure ( gray ) is superimposed with RavZb to show the conformational change of α3 . Hydrogen bonds are shown as red dashed lines . ( C ) Crystal structure of the LIR2:LC3 complex . LC3 is colored in green and the LIR2 peptide is coloured in pink . Hydrogen bonds are shown as red dashed lines . ( D ) Interaction of LIR2 with LC3 . LIR peptide is shown as pink stick and LC3 is displayed as surface . Two hydrophobic binding pockets ( HP1 and HP2 ) are shown in pale yellow . The 2Fo–Fc map of the LIR2 motif is contoured density at α of 1 . 0 . DOI: http://dx . doi . org/10 . 7554/eLife . 23905 . 01210 . 7554/eLife . 23905 . 013Figure 5—figure supplement 1 . LIR2 is required for RavZ:LC3 interaction . ( A ) Interaction analysis of the FITC-LIR2 peptide with LC3 , GABARAPL1 and GATE16 by fluorescence polarization . Dissociation constants are shown in the table ( below ) . ( B ) Inhibition of RavZ activity by the LIR2 peptide . The LC3-PE protein ( 7 µM ) was incubated with different concentrations of LIR2 peptide for 1 hr on ice and was subjected to RavZ ( 0 . 7 µM ) treatment for 10 min at 37°C . The reaction was resolved by SDS-PAGE . The data were fitted to a sigmoidal equation , leading to IC50 value . n = 3 independent experiments . Mean and SD are presented . ( C ) Competitive titration assay by fluorescence polarization measurement . 1 . 25 µM FITC-LIR2 was titrated with 6 . 25 µM LC3 and then with increasing concentrations of RavZ . Full recovery of fluorescence polarization signal for FITC-LIR2 was observed ( orange columns ) . In the control experiment , LC3 was omitted ( green unfilled columns ) . No binding between FITC-LIR2 and RavZ was observed . DOI: http://dx . doi . org/10 . 7554/eLife . 23905 . 01310 . 7554/eLife . 23905 . 014Figure 5—figure supplement 2 . Structural analysis of free RavZ and RavZ:LC3 complex . ( A ) Structural comparison of free RavZ with the published RavZ ( PDB: 5CQC ) . Catalytic domain and PI3P-binding domain of our RavZ structures are colored in blue and deep-teal , respectively . Published RavZ structure is colored in gray , α3 hydrophobic residues and PI3P binding residues are shown as stick models and colored in orange and magenta , respectively . ( B ) Structural comparison of free RavZ ( deep-teal ) , RavZ in complex with LC3 ( pale cyan ) and published RavZ ( gray ) . The missing loops in the published RavZ that are found in our free RavZ and complexed RavZ are colored in orange and red , respectively . ( C ) Structural comparison of the docking structure of RavZ:LC3 complex and the crystal structure of Atg4B:LC3 . RavZ is shown in deep-teal , Atg4B is shown in light pink , LC3RavZ is shown in green and LC3Atg4B is shown in light orange . N-terminal LIR2RavZ and N-terminal LIRATG4B are indicated in red and blue , respectively . The catalytic triad is shown as a stick model . ( D ) Crystal packing of the LIR2-fused LC3 protein . LC3 molecules in chain A and chain B are shown in green and cyan , respectively . LIR2 is colored in magenta . Black dashed line indicates asymmetric unit . DOI: http://dx . doi . org/10 . 7554/eLife . 23905 . 014 To prove whether LIR2 interacts with the LIR-binding site in LC3 , we solved two structures of an LIR2-LC3 fusion in two crystallization conditions . Both structures are identical and solved in the same space group ( Supplementary file 4 ) . There are two molecules ( molecule A and B ) of the LIR2-LC3 fusion in one asymmetric unit with LIR2 of each molecule interacting with its symmetry-related LC3 molecule ( Figure 5—figure supplement 2D ) . The hydrophobic interaction of LIR2 with LC3 is conserved and adopts the classical LIR-LC3-binding mode ( Klionsky and Schulman , 2014; Noda et al . , 2010 ) . The first aromatic residue , F29 , of LIR2 interacts with the first hydrophobic pocket ( HP1 ) consisting of I23 , P32 , L53 , F108 and the alkyl sidechain of K51 and the second aromatic residue L32 interacts with the second hydrophobic pocket ( HP2 ) consisting of F52 , V54 , P55 , V58 , L63 and I66 ( Figure 5C and D ) . The salt bridge interactions of residues LIR2E28 with LC3K51 and LIR2D30 with LC3R70 are observed in both structures . We further investigate whether RavZ shows preference on binding to individual Atg8 family proteins , we tested binding of LIR2 peptide with three Atg8 members ( representing two subfamilies ) , LC3B , GABARAPL1 and GATE16 . No significant difference in binding affinities was observed ( Figure 5—figure supplement 1A ) . Therefore , RavZ has no preference in binding to Atg8 proteins . To verify the interaction of LIR2 with LC3 seen in the crystal structure , we mutated the interacting residues on LC3 , hydrophobic residues ( F52 and L53 ) and charged residues ( K51 and R70 ) , to alanine . The mutants were subjected to binding analysis with the LIR2 peptide by fluorescence polarization . F52A mutation led to the reduction of LIR2 binding by about sixfold ( Kd = 3 . 14 µM ) , while R70A mutation reduced binding affinity about 15-fold ( Kd = 8 . 10 µM ) . L53A and K51A mutation led to only residual effect on binding to LIR2 . No binding of the free LIR2 peptide with the LIR2-LC3 fusion protein was observed under these conditions , suggesting that the LIR2 peptide competes with the conjugated LIR2 peptide for binding to the LIR-binding site of LC3 ( Figure 4—figure supplement 1C ) . Moreover , RavZ and the LIR2 peptide bind to LC3 in a competitive manner , as shown by the competitive titration and enzymatic inhibition assay ( Figure 5—figure supplement 1B and C ) . Therefore , RavZ recognizes LC3 mainly via its N-terminal LIR2 motif by binding to the LIR-binding site in LC3 . In order to understand the thermodynamic basis of RavZ extraction , we sought to determine the binding affinity of RavZC258A with LC3-PE and LC31–119 by microscale thermophoresis ( MST ) technique . To make LC3-PE soluble in solution without detergent , MBP tag was left intact . The measurements showed that RavZC258Abinds to MBP-LC3-PE and MBP-LC31–119 with dissociation constants ( Kd ) of 23 ± 4 nM and 69 ± 5 nM , respectively , which suggests that RavZ binds to lipidated LC3 in three times higher affinity than unlipidated LC3 ( Figure 6—figure supplement 1D ) . Therefore , the thermodynamic driving force for RavZ extraction is modest but still favorable . Based on our model , there should be a lipid-binding site ( LBS ) in RavZ . The amino acid sequence of RavZ does not give any hint of the LBS . We superimposed the free RavZ structure with the phospholipid-binding domain of Sec14 proteins ( PDB: 1AUA and 3B74 ) , which are the major yeast phosphatidylinositol ( PtdIns ) /phosphatidylcholine ( PtdCho ) transfer proteins and are essential for lipid metabolism ( Schaaf et al . , 2008; Welti et al . , 2007 ) . Structural alignment indicated that part of the N-terminal catalytic domain of RavZ ( α2–α4 , β6–β9 ) shows a similar fold to that of the core-lipid-binding domain of Sec14 , consisting of 5 β-strands and 4 α-helices ( α7–α10 ) ( Schaaf et al . , 2008 ) ( Figure 6A ) . The β sheets of Sec14 serve as the floor of hydrophobic binding pocket , along helices α7-α10 that gate the Sec14 pocket , forming extensive van der Waals contacts with the bound fatty acid chains . Helices α9–α10 act as lid helices , which move toward helix α8 upon lipid binding to form a closed conformation to capture lipid into the binding site ( Figure 6—figure supplement 1C ) ( Welti et al . , 2007 ) . Similarly in RavZ , helices α2 ( corresponding to α7 of Sec14 ) and α4 ( corresponding to α8 of Sec14 ) along β sheets form a large hydrophobic interface ( Figure 6A and B ) . However , the hydrophobic pocket is closed in RavZ , suggesting that a conformational change may be required to accommodate the fatty acid chains . A lid helix in RavZ is not clearly observed based on the structural alignment . However , due to the highly dynamic and hydrophobic nature of α3 , it is possible that α3 may resemble the function of the lid helix and may undergo substantial movements to interact and stabilize the lipid moiety in the LBS upon binding of PE . 10 . 7554/eLife . 23905 . 015Figure 6 . Identification of the lipid-binding site ( LBS ) of RavZ . ( A ) Structural alignment between the lipid-binding sites of yeast Sec14 homolog ( Shf1 ) ( PDB: 3B74 ) and RavZ . RavZ is colored in pale yellow , while Shf1 is shown as a green transparent cartoon and its secondary elements are indicated . The phosphatidylethanolamine ( PE ) is shown as stick model with carbon atoms colored in magenta . ( B ) Structure of N-terminal catalytic domain of RavZ showing the LBS of RavZ and the predicted lipid-binding residues . The α3 loop and α4 are colored in orange and light pink , respectively . The residues involved in lipid binding are highlighted and displayed as side chain sticks . ( C ) In vitro LC3-PE cleavage assay of RavZ mutants . ( D ) Effect of RavZ mutants on LC3-II level in the cell . GFP-LC3 stable cells were transfected with RavZ mutants and subjected for starvation for 2 hr . ( E ) Quantification of the ratio of endogenous LC3-II to LC3-I ( left ) and GFP-LC3-II to GFP-LC3-I ( right ) shown in ( D ) . n = 3 independent experiments . Mean and SD are presented; ***p<0 . 001 , **p<0 . 01 , ns: not significant . ( F ) Effect of RavZ mutants on GFP-LC3 puncta formation in the cell as shown in Figure 6—figure supplement 2A . n = 25–37 cells . Mean and SD are presented; ***p<0 . 001 , **p<0 . 01 , ns: not significant . ( G ) Effect of RavZC258A mutants on GFP-LC3 puncta formation in the cell as shown in Figure 6—figure supplement 2B . n = 22–35 cells . Mean and SD are presented; ***p<0 . 001 , **p<0 . 01 , ns: not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 23905 . 01510 . 7554/eLife . 23905 . 016Figure 6—figure supplement 1 . Structural analysis of the lipid-binding site of free RavZ , free Sec14 and PE-bound Sfh1 . ( A ) Structural comparison of N-terminal catalytic domain of RavZ with NEDP1 . RavZ and NEDP1 are colored in pale yellow and pale green , respectively . The α3 loop of RavZ is shown in orange . ( B ) Surface representation of LBS of free RavZ and free Sec14 ( PDB: 1AUA ) . The lipid-binding cavity of Sec14 is observed . The PE group is shown in the cavity by alignment with the PE-bound Sfh1 ( 3B74 ) . ( C ) Structural comparison of free Sec14 ( PDB: 1AUA ) and PE-bound Sfh1 . The lid helix ( α9-α10 ) of free Sec14 and PE-bound Sfh1 is colored in deep teal and green , respectively . ( D ) Interaction analysis of RavZC258A with MBP-LC3-PE ( 16:0 ) or MBP-LC31–119 using MST technique . DOI: http://dx . doi . org/10 . 7554/eLife . 23905 . 01610 . 7554/eLife . 23905 . 017Figure 6—figure supplement 2 . Functional characterization of the lipid-binding site of RavZ . ( A ) Confocal microscopy of GFP-LC3 stable cells transfected with mCherry-RavZ constructs . Before imaging , cells were treated for 2 hr under starvation medium ( EBSS ) . ( B ) Confocal microscopy of GFP-LC3 stable cells transfected with mCherry-RavZ C258A constructs . Before imaging , cells were treated for 2 hr under starvation media ( EBSS ) . ( C ) Binding of RavZ LBS mutants with LC3 by fluorescence polarization . Dissociation constants are shown in the table . ( D ) Circular dichroism ( CD ) spectra of RavZ and its variants . DOI: http://dx . doi . org/10 . 7554/eLife . 23905 . 017 In order to prove the predicted LBS of RavZ , we mutated a set of hydrophobic residues on α2 , α3 , α4 , the α4-β9 loop and β7 to the negatively change residue Asp ( Figure 6B and C ) . The In vitro LC3-PE cleavage assay showed that most of the mutants affect the cleavage activity of RavZ . The double mutation ( Y211D/F212D , α3-m2 ) or triple mutation ( Y211D/F212D/Y216D , α3-m3 ) in α3 completely abolished the cleavage activity . A single mutation scan shows that F211D , Y212D and L208D dramatically reduce the cleavage activity , whereas the Y216D mutant largely retains activity . Triple mutation of the gate residues on α4 ( L224D/I288D/L232D ) and the α4-β9 loop ( F242D/L239D/F237D ) led to significant reduction in cleavage activity . To confirm the floor residues of the hydrophobic pocket of RavZ , we mutated hydrophobic residues on α2 ( L139D/L143D ) and β7 ( L180D/I182D ) . Again , these mutations significantly reduced cleavage activity ( Figure 6C ) . To evaluate the deconjugation activity of RavZ mutants in cells , GFP-LC3 cells were transfected with different mCherry-RavZ mutant constructs . An additional mutant RavZF163D ( F163 on β6 located inside of the hydrophobic cavity ) was also included in the assay . The level of LC3-II was evaluated by Western blotting ( Figure 6D ) and GFP-LC3 puncta assay ( Figure 6F ) . Cells expressing RavZ mutants displayed a significant increase in the ratio of endogenous LC3-II/LC3-I as well as the ratio of GFP-LC3-II/GFP-LC3-I and in the number of GFP-LC3 puncta , compared to those expressing wild-type RavZ ( Figure 6E and F; Figure 6—figure supplement 2A ) , suggesting that deficient activity of RavZ mutants results in decreased deconjugation of LC3-PE in cells . Moreover , the α3-m2 mutant completely abolished RavZ deconjugation activity in vivo . These results are in line with those obtained by the in vitro LC3-PE cleavage assay . To further examine the extraction activity of RavZ mutants , GFP-LC3 cells were transfected with mCherry-RavZC258A mutant constructs . In keeping with the deconjugation results , a significantly higher number of GFP-LC3 puncta was observed in cells expressing RavZC258A mutants than those expressing RavZC258A , indicating less extraction activity of these mutants in vivo ( Figure 6G; Figure 6—figure supplement 2B ) . Consistently , the α3-m2 and α3-m3 mutants completely lost extraction activity ( Figure 6G; Figure 3—figure supplement 1F and G ) . Moreover , all RavZ mutants bind to LC3 with similar Kd values to those of the wild-type protein ( Figure 6—figure supplement 2C ) and display similar secondary structure with comparable α helix content to the wild-type protein ( Figure 6—figure supplement 2D ) . Therefore , the mutations do not significantly affect protein folding and binding to LC3 . Taken together , we conclude that the LBS of RavZ , consisting of α2 , α3 , α4 , α4–β9 loop and β sheets ( β6–β9 ) , accommodates the fatty acid chains of PE . The LBS plays an essential role in extraction and cleavage of LC3-PE .
In previous studies , the substrate LC3-PE was produced on liposomes by recombinant reconstitution of the ubiquitin-like conjugation system ( i . e . mammalian Atg7 , Atg3 and Atg8s ) ( Choy et al . , 2012 ) . The question then arises whether RavZ activity requires membranes or not . However , production of liposome-free LC3-PE and manipulation of LC3-PE structures are difficult using the recombinant approach . In this study , the semisynthetic approach makes it possible to address these inherent problems in the analysis of post-translationally modified proteins . We have used chemical approaches to prepare LC3 proteins with various modifications in a membrane-free manner . Using semisynthetic LC3 proteins modified with various PE fragments , we analyzed the structure-function relationship of LC3 deconjugation by RavZ . We demonstrate that RavZ activity is strictly dependent on the lipid structure of the substrate and RavZ cleaves LC3-PE without requirement for membranes . These observations indicate an extraction model of RavZ function , which is further proved by the extraction assay using protease-deficient RavZ . This is the first-time identification of such an action mode for host-pathogen interactions . Although the N-terminal catalytic domain is sufficient to bind and cleave LC3-PE in solution , its extraction activity and thereby its deconjugating activity in vivo are attenuated ( Figure 6E–G; Figure 3—figure supplement 1F and G ) . Therefore , association of RavZ with membranes via its C-terminal PI3P-binding domain is also required for efficient extraction in vivo , probably by targeting RavZ to the autophagosome membrane ( Horenkamp et al . , 2015 ) . L . pneumophila may have evolved both functions simultaneously , that is extraction and cleavage . The former determines the specificity and the latter confers the turnover . Without retrieval of LC3-PE from membranes RavZ cannot perform proteolytic activity . However , although extraction alone is sufficient to inhibit autophagy , it requires a much higher amount of RavZ molecules because it is driven by stoichiometric binding between RavZ and LC3-PE . The proteolytic function of RavZ drives turnover of LC3-PE molecules , so that less than stoichiometric amounts of RavZ molecules are needed . It is possible that the extraction model may be present in other host-pathogen interactions , such as Shigella effector IpaJ with N-myristoylated Arf GTPases and Yersinia effector YopT with prenylated RhoA GTPases , since these effectors also recognize the lipid group ( Burnaevskiy et al . , 2013 , 2015; Shao et al . , 2002 , 2003 ) . We have demonstrated that the LIR2 motif ( residue 27–32 ) plays an essential role in RavZ activity and RavZ:LC3 interaction . Interestingly , the LIR motif at the N-terminal tail of Atg4B is involved in regulation of Atg4B activity . Binding of the LIR motif with the second non-substrate LC3 molecule leads to an open conformation , which is required for deconjugation of LC3 . Atg4B lacking the N-terminal tail shows higher processing activity ( Satoo et al . , 2009 ) . It is not clear whether such interaction occurs in vivo . However , the LIR2 motif of RavZ is required for initial recognition of LC3 . The LIR-LC3 binding constitutes the major interaction between RavZ and LC3 before extraction of the PE moiety from the membrane . Therefore , LIR2 is involved in recognizing and orienting the LC3 molecule to facilitate subsequent extraction and cleavage of LC3-PE . According to this model , it is not surprising that RavZ does not process PE-peptides containing C-terminal amino acid residues of LC3 and the Rab7-CQETFG-PE chimeric protein , due to the lack of LIR binding . In contrast , Shigella effector IpaJ and Yersinia effector YopT can process the lipidated peptides containing an N-myristoylated glycine and a C-terminally prenylated polybasic sequence , respectively , suggesting a unique manner of recognizing substrate for RavZ ( Burnaevskiy et al . , 2013 , 2015; Shao et al . , 2002 , 2003 ) . Based on this finding , we are able to inhibit RavZ activity by the LIR2 peptide ( IC50 = 43 µM ) ( Figure 5—figure supplement 1B ) , suggesting that suppression of LIR-LC3 binding could be beneficial for attenuating inhibition of host autophagy by L . pneumophila . Through structural alignment with yeast lipid transfer proteins and mutagenesis studies , we have identified the LBS of RavZ , involving α2 , α3 , α4 , the α4–β9 loop and β sheets ( β6–β9 ) . The LBS is required for both extraction and cleavage of LC3-PE . Therefore , LIR2:LC3 and LBS:PE interactions may constitute two major binding interfaces for RavZ:LC3-PE complex , which place the C-terminal tail of LC3 at the correct position of the active site of RavZ for cleavage . However , such a LBS is not observed in Atg4B ( Figure 5—figure supplement 2C ) , consistent with the observation that Atg4B has no specificity toward structures C-terminal to the scissile bond . This could be one of the explanations for the distinct modes of action of Atg4 and RavZ . The fold of the N-terminal catalytic domain of RavZ is closely related to cysteine proteases in the ubiquitin-like ( Ubl ) -specific protease ( Ulp ) family that is specific for de-conjugating Ubl proteins ( Mossessova and Lima , 2000; Shen et al . , 2005 ) . A similar LBS fold is also found in Ulp proteins ( NEDP1 , PBD: 2BKR ) , but this fold is lack of the α3 loop ( Figure 6—figure supplement 1A ) . It should be noted that the LBS of RavZ is closed without showing a binding pocket for lipid , whereas a clear hydrophobic binding cavity that can accommodate fatty acid chains is observed in free Sec14 structures ( Figure 6—figure supplement 2B ) . Because the PE moiety involves interaction with residues located deep in the LBS fold , there must be a conformational change to open the lipid-binding site , for example , by outward movement of α4 . Because α4 is connected to β9 via the highly dynamic α3 loop , such a movement could be made possible . These findings are in keeping with the important role of the α3 loop in RavZ activity ( Figure 6C–6F; Figure 3—figure supplement 1F and G ) . The hydrophobic and dynamic features of α3 are in agreement with this proposed function . The association of α3 with the membrane , as shown previously ( Horenkamp et al . , 2015 ) , may facilitate initial interaction of α3 with the conjugated PE in the membrane . Subsequently , the lipid-binding site of RavZ is open and able to accommodate the PE moiety that is dug out of the membrane by α3 , which moves toward the LBS . The α3 may serve as a lid for the lipid-binding pocket to confer the binding of fatty acid chains . Intracellular molecules with analogous feature , that is extraction of lipidated proteins , are GDP-dissociation inhibitors ( GDIs ) and the GDI-like molecule PDEδ , which serve as recycling factors for prenylated Ras GTPase family proteins ( Ras , Rho and Rab ) between membranes and the cytosol ( Ismail et al . , 2011; Rak et al . , 2003; Tnimov et al . , 2012 ) . Interestingly , the lipid-binding site of isolated RabGDI is also closed and undergoes a conformational change involving an outward movement of an α helix to open the LBS for accommodating the prenyl moiety ( Rak et al . , 2003 ) . The opening of the LBS is induced by binding to a Rab molecule rather than the prenyl group per se ( Ignatev et al . , 2008; Zhao et al . , 2016 ) . Therefore , RavZ has evolved a GDI-like mechanism to extract LC3-PE from membranes . Taken together , we propose a working model of RavZ function as shown in Figure 7 . 10 . 7554/eLife . 23905 . 018Figure 7 . Working model of RavZ-mediated LC3-PE deconjugation on the membrane . RavZ ( deep teal ) recognizes LC3 molecule ( green ) on the membrane via its LIR2 motif ( magenta ) . RavZ targets to autophagosome membrane by interaction of its C-terminal domain with PI3P ( orange ) and association of the α3 helix ( the helix inserted into the membrane ) with membranes . α3 facilitates extraction of the PE moiety ( yellow ) from the membrane and docking of the fatty acid chains into the lipid-binding site of RavZ . The interaction of the PE moiety with LBS and the LIR2:LC3 binding orient the C-terminal tail into the active site for cleavage . RavZ:LC3 complex was generated by molecular docking using ZDOCK server: an automatic protein docking server ( http://zdock . umassmed . edu ) . The LC31–120 structure from Atg4B:LC3 complex ( PDB: 2Z0D ) was docked onto the structure of RavZ20-502 ( residues 48–432 ) . The C-terminal residues ( 115-120 ) of LC31–120 and catalytic residues ( C258 , H176 and D197 ) located in the active site of RavZ were selected as the binding residues . The model was selected from top 10 scoring models . Binding of N-terminal LIR2 loop of RavZ with LC3 was generated by superimposition of the LIR2:LC3 structure onto LC3 in the docking structure . DOI: http://dx . doi . org/10 . 7554/eLife . 23905 . 018
MCF7 cells ( HTB-22 ) and HeLa ( CCL-2 ) cells were obtained from ATCC . Human microtubule-associated protein light chain 3B ( LC3 ) were cloned into pEGFP-C1 vector ( Clontech ) and pEGFP-C1-LC3 plasmid was used to generate human MCF-7 stable cell lines by single-cell Fluorescence-Activated Cell Sorting ( FACS ) . They were tested negative for mycoplasma contamination . The cells were grown in minimum essential medium ( MEM ) ( Sigma-Aldrich Cat# M4655 ) supplemented with 10% fetal bovine serum ( FBS ) , 1% sodium pyruvate and 1% non-essential amino acids ( NEAA ) . G418 ( 200 µg/mL ) for the stable cell line . Cell lines were cultured at 37°C in 5% CO2 . For starvation , cells were first washed with PBS three times and incubated in EBSS ( Sigma Cat# E3024 ) for 2 hr . Chemicals for peptide and compound synthesis were obtained from Acros , Aldrich , Avanti , Fluka , Santa Cruz or Novabiochem and used without further purification . Antibodies used in the study were rabbit anti-LC3B ( Cell Signaling Technology Cat# 2775 , RRID:AB_915950 ) , anti-Cherry ( Abcam Cat# ab125096 , RRID:AB_11133266 ) and mouse anti-β-actin ( Millipore Cat# MAB1501 , RRID:AB_2223041 ) . HRP-conjugated secondary antibodies used for WB were anti-mouse ( Dako Cat# P022602 , RRID:AB_579516 ) and anti-rabbit ( Millipore Cat# AQ132P , RRID:AB_92785 ) . Peptides were synthesized employing an Fmoc-based solid phase peptide synthesis strategy using 2-chlorotrityl chloride resin . Protected peptide ( 0 . 016 mmol ) was lipidated by 1-hexadecanol ( 16C ) , 1 , 2-dipalmitoyl-sn-glycero-3-phosphoethanolamine ( DPPE ) or 1 , 2-dihexanoyl-sn-glycero-3-phosphoethanolamine ( DHPE ) . All peptides were characterizated by high-resolution mass spectra ( HR-MS ) ( Supplementary file 1 ) . HR-MS were measured on a Thermo Orbitrap coupled to a Thermo Accela HPLC system using electrospray ionization ( ESI ) . The small nucleophiles , ethanolamine 1 and phosphoethanolamine 2 were purchased from Sigma . Other small nucleophiles , glycerophosphoethanolamine 8 and diacetyl glycerophosphoethanolamine 10 were synthesized as shown in Figure 1—figure supplement 1B . Subsequently , these compounds were characterizated by NMR and HR-MS ( Supplementary file 2 ) . 1H NMR spectra were recorded on a Varian Mercury Plus 300 MHz spectrometer ( 300 MHz ) , a Bruker DRX 400 ( 400 . 13 MHz ) spectrometer . NMR spectra were calibrated to the solvent signals of DMSO-d6 ( δ = 2 . 50 and 40 . 45 ppm ) . All reactions were carried out under an inert atmosphere in dry solvents unless otherwise noted . The lipidated proteins were achieved by performing express protein ligation ( EPL ) of recombinant LC3-thioester proteins with synthetic lipidated peptides according to our previous method ( Yang et al . , 2013 ) . LC3 proteins containing soluble fragments of PE lipid were produced using direct aminolysis strategy of protein thioester by a small-molecule nucleophile ( Payne et al . , 2008; Yi et al . , 2010 ) . Ligation of the DPPE ( 16:0 ) or 16-carbon hexanoyl chain modified peptide with MBP-LC31–114 thioester was performed as previously reported ( Yang et al . , 2013 ) . The ligation of DHPE ( 6:0 ) modified peptide with MBP-LC31–114 thioester was performed in absence of the detergent . After incubation overnight at room temperature , the ligated product was purified by size exclusion chromatography ( Superdex 200 10/300 GL ) . Aminolysis of LC31–120-thioester by small-molecule nucleophile was carried out by the following procedure . 50 µM LC31–120-MESNA thioester was incubated with 0 . 5 M small molecule nucleophiles in the buffer ( 100 mM NaH2PO4 , pH 7 . 2 , 50 mM NaCl , 100 mM MPAA ) for 48 hr at 4°C . The reaction was subjected to LC-MS analysis . Finally , the solution was dialyzed against dialysis buffer to remove unreacted compound . Plasmids containing RavZ fragments or LC3 proteins were transformed into E . coli BL21 ( DE3 ) -LIR cells . Protein expression was induced with 0 . 2–0 . 4 mM IPTG and carried out at 20°C overnight . Purification was performed with Äkta prime plus chromatography purification system ( GE Healthcare Life Sciences ) . The cells were harvested and resuspended in breaking buffer containing 1X protease cocktail ( Roche Life Science Cat# 05056489001 ) or 1 mM PMSF . Cells were then lysed by a Microfluidizer ( Microfluidics ) . All proteins used in this study contain affinity tags ( MBP , GST or chitin-binding domain ( CBD ) ) with an extra 6xHis tag fused to its N-terminus . The proteins were initially purified by Ni-NTA-affinity purification using HisTrap HP column ( GE Healthcare Life Sciences ) eluted with gradient of 0–100% of 500 mM imidazole . To release the protein from the affinity tags , the fusion proteins carrying precision protease or TEV protease cleavage site were cleaved by the corresponding proteases overnight at 4°C . For the chitin fusion proteins with intein tag were cleaved by adding powdered MESNA to a final concentration of 0 . 5 M and incubation overnight at room temperature to produce the protein thioesters . The cleaved tags were removed with HisTrap HP column , proteins were further purified with size exclusion chromatography using HiLoad 16/60 Superdex 200 . For analytical gel filtration analysis of the complex formation . The experiments were performed on Superdex 200 10/300 GL equipped with ÄKTAFPLC system . Protein samples were previously purified with size exclusion chromatography as a last purification step . Complex formation was done by mixing each of RavZ fragment and LC31–119 with molar ratio of 1:1 . 5 and incubated at 4°C overnight . 500 µL of proteins or protein complexes were injected to pre-equilibrated column with buffer containing 50 mM HEPES pH 7 . 5 , 50 mM NaCl and 2 mM DTE . The size exclusion chromatography was performed with a flow rate of 0 . 5 ml/min for 35 min . The standard protein markers containing ( thyroglobulin ( 670 kDa ) , globulin ( 158 kDa ) , ovalbumin ( 44 kDa ) , myoglobin ( 17 kDa ) and vitamin B12 ( 1 . 3 kDa ) ) were run on the same column under identical condition . For crystallization screening , proteins were screened for crystallization conditions with various crystallization screening kits from QIAGEN . Protein and screening buffer were automatically mixed in ratio 1:1 with 0 . 1 µL of each by mosquito in crystallization screening plate . The crystallization plates were incubated in the Rock Imager ( Formulatrix ) , an automated imaging system at 277 . 15 K and 293 . 15 K . RavZ1–430/LC3B1–119 complex was crystallized in 0 . 17 M ammonium acetate , 0 . 085 M tri sodium citrate pH 5 . 6 , 25 . 5% PEG4000 and 15% glycerol at 277 . 15 K . RavZ1–487 and RavZ20-502 were crystallized in 16% PEG3350 , 0 . 2 M BaCl2 and 0 . 1 M MES pH 5 . 6 at 277 . 15 K . LIR2-LC31–119 fusion protein was crystallized in two conditions , as indicated with low-salt and high-salt . The low-salt condition contains 0 . 1 M citric acid anhydrous pH 4 . 0 and 1 . 6 M ammonium sulfate at 293 . 15 K and high-salt condition contains 0 . 1 M sodium acetate pH 4 . 5 and 3 M NaCl at 277 . 15 K . The crystallization conditions were further optimized by refining concentration of precipitant , salt and pH . The data were collected at 100 K at PXII-XS10SA beamline in the Swiss Light Source ( SLS ) Villingen . Data was indexed , processed and scaled with XDS ( Kabsch , 1993 ) . RavZ ( PDB: 5CQC ) and LC3B ( PDB: 2ZOD ) were used as molecular replacement search models . Molecular replacement and structural refinement were done with PHENIX ( Adams et al . , 2010 ) and rebuilt in COOT ( Emsley and Cowtan , 2004 ) . Validation was done with Molprobity ( Chen et al . , 2010 ) . X-ray data collection and refinement statistics are listed in Supplementary file 4 . LC-MS analysis was performed on an Agilent 1100 series chromatography system equipped with an LCQ electrospray mass spectrometer ( Finnigan , San Jose ) using Jupiter C4 columns ( 5 μm , 15 × 0 . 46 cm , 300 Å pore-size ) from Phenomenex ( Aschaffenburg , Germany ) . For LC-separations a gradient of buffer B ( 0 . 1% formic acid in acetonitrile ) in buffer A ( 0 . 1% formic acid in water ) with a constant flow-rate of 1 mL/min was employed . Upon sample injection , a ratio of 20% buffer B was kept constant for 4 min . Elution was achieved using a linear gradient of 30–80% buffer B in buffer A for 5–15 min followed by a steep gradient ( 70–90% buffer B ) for 15–17 min . The column was extensively flushed for 17–19 min with 90–10% buffer B . Data evaluation was carried out using the Xcalibur software package and MagTran software programs was used for deconvolution of ESI mass spectra ( Supplementary file 3 ) . CD spectroscopy was acquired using a Jasco J-815 CD spectrometer equipped with a JASCO PTC-423S temperature controller . Proteins were diluted to 0 . 125 mg/ml in 50 mM phosphate buffer pH 7 . 5 containing 50 mM NaCl and 2 mM DTE . Protein samples were filled in 0 . 1 cm quartz cuvette and scan from 195 nm to 260 nm at 25°C , scanning speed was 20 nm/min and CD spectra were accumulated three times . Firstly , CD spectra of protein samples were subtracted by spectra of buffer and smoothed using a FFT filter . The unit of millidegree ( θ ) was conversed to mean residue molar ellipticity [θ]MRW , λ . The ratio of secondary structure elements was calculated with the software CDNN . The database consisting of 33 reference proteins was used in the deconvolution analysis . For imaging , MCF7 cells stably expressing GFP-LC3 or HeLa cells were plated at a density of 5 . 0 × 104 cells per well on µ-Slide 4 Well ( Ibidi ) . GFP-LC3 stable cells were transfected with 250 ng of mCherry-RavZ constructs using X-treme GENE HP DNA transfection reagent . For co-transfection , HeLa cells were co-transfected with each 125 ng of GFP-LC3 and mCherry-RavZ constructs . After 20 hr , live cell imaging was performed in MEM without phenol red ( Thermo Fisher Cat# 51200046 ) or EBSS ( Sigma Cat# E3024 ) by using an inverted confocal microscope Leica TCS SP2 or SP5 AOBS equipped with a 63×/1 . 4 HCX Plan Apo oil immersion lens and a temperature-controlled hood at 37°C and 5% CO2 . Quantification of the area of GFP-LC3 puncta was performed through Analyze Particles function of ImageJ . For immunoblotting , GFP-LC3 stable cells were plated at a density of 5 . 0 × 105 cells per well in a 6‐well plate . Cells were transfected with 1 μg of each plasmid using X-treme GENE HP DNA transfection reagent ( Roche Life Science Cat# 06365244001 ) . For immunoblotting , total cell lysates were prepared by adding cells to RIPA buffer ( 50 mM Tris pH 7 . 8 , 150 mM NaCl , 1% Triton X-100 , 1% sodium deoxycholate and 0 . 1% sodium dodecyl sulfate ( SDS ) ) . Equal amount of proteins was resolved by SDS-PAGE , transferred to PVDF membranes , and incubated with primary antibody overnight at 4°C . After washing 3 times with TBST buffer , membranes were incubated with secondary antibodies conjugated with horseradish peroxidase for 60 min . Signals were visualized with ECL Prime Western Blotting Detection Reagent ( Amersham Cat# 10600001 ) or SuperSignal Western Blot Enhancer ( Thermo Fisher Cat# 34095 ) using UltraCruz Autoradiography Films ( Santa Cruz Cat# sc-201696 ) . MCF7 cell membranes were prepared as described ( Pylypenko et al . , 2006; Rak et al . , 2003 ) . GFP-LC3 stable cells were starved in EBSS for 2 hr to induce autophagy . The cells were collected and resuspended in the fractionation buffer ( FB buffer , 250 mM Sucrose , 20 mM HEPES ( 7 . 4 ) , 10 mM KCl , 1 . 5 mM MgCl2 , 1 mM EDTA , 1 mM DTT and protease inhibitor cocktail ) on ice for 30 min . The cell solution was homogenized 80 times using Dounce homogenizer . The cell lysate was cleared by centrifugation at 540 g for 5 min . The extract was then loaded onto a 1 mL cushion of sucrose ( 60% in RB ) , and centrifuged at 100 , 000 g for 1 hr in a Optima MAX-XP ultracentrifuge equipped with rotor TLA_100 . 4 ( Beckman Coulter ) . The buffer–sucrose interface was collected in a minimal volume , and the protein concentration was determined by the Bradford assay . Different concentrations of RavZC258A protein were incubated with the membrane fraction ( 20 µg of membrane protein ) in the final assay volume of 100 µL at 37°C for 2 hr . 500 µL of FB buffer was added to each reaction and mixed . Diluted assay mixtures were centrifuged for 1 hr at 100 , 000 g using the ultracentrifuge equipped with rotor TLA_100 . 1 ( Beckman Coulter ) . Precipitation of the supernatant was done using TCA/DOC ( trichloroacetic acid/sodium deoxycholate ) protocol . Briefly , DOC ( final concentration 125 µg/ml ) and TCA ( final concentration 6% ) were added to the protein fraction , subsequently . The resulting solution was mixed and incubated on ice for 20 min , and then was centrifuged 15 min in max speed ( 14 , 680 rpm ) . The pellet was washed by cold acetone and dissolved in 1X SDS sample buffer . The samples were resolved on 13% SDS–PAGE gel . Membrane-associated and extracted endogenous LC3-II and GFP-LC3-II were visualized by immunoblotting with anti-LC3 antibody . For detection of cytosolic LC3-PE level in the RavZC258A transfected cells , GFP-LC3 stable cells were plated to 6 cm dish at a density of 1 . 6 × 106 cells per dish . The cells were transfected with 3 μg of plasmid pmCherry-RavZC258A using X-treme GENE HP DNA transfection reagent . The pmCherry vector also was used as a control . After 24 hr , cells were starved and collected , and resuspended in the FB buffer on ice for 30 min . The cell solution was homogenized 80 times using Dounce homogenizer . The cell lysate was cleared by centrifugation at 540 g for 5 min . The extract ( total protein fraction ) was then centrifuged at 100 , 000 g for 1 hr in an Optima MAX-XP ultracentrifuge equipped with rotor TLA_100 . 1 ( Beckman Coulter ) . The supernatant ( cytosolic fraction ) was collected and the pellet ( membrane fraction ) was resuspended in the RIPA buffer . The protein concentration was determined by the Bradford assay . LC3 proteins were visualised by immunoblotting with anti-LC3 antibody . Results were expressed as mean ± standard deviation of means ( SD ) . The data were assessed by one-way analysis of variance ( ANOVA ) followed by the Fisher LSD's post hoc comparisons . In all statistical comparisons , differences with p<0 . 05 were considered significance . Statistical analysis was performed in Origin 9 . Atomic coordinates and structure factors have been deposited in the Protein Data Bank under the following accession codes 5MS2 ( RavZ1–431/LC3B ) , 5MS5 ( low-salt RavZ LIR2-LC3B ) , 5MS6 ( high-salt RavZ LIR2-LC3B ) , 5MS7 ( RavZ20-502 ) and 5MS8 ( RavZ1–487 ) . | Organisms have to fight bacteria and other disease-causing microbes on a daily basis . To stay on top of the game , cells use many ways to defend themselves . Autophagy , for example , is a process that breaks down unwanted or damaged molecules inside cells , which has also been linked to fighting infections caused by disease-causing microbes . During this process , structures called autophagosomes engulf the molecules or microbes , and digest them with the help of enzymes . Forming an autophagosome is a complex process that requires several steps and molecules . First , cells create a growing membrane sac that collects the debris and eventually seals to form the autophagosome . One of the key proteins that expands the membrane is the protein LC3 , which is located on the both sides of the autophagosome’s membrane . LC3 must be linked to an oily molecule ( or lipid ) , known as PE for short , in order to interact with this membrane . The combined molecule LC3-PE then provides a docking station for receptor proteins that collect and deliver the debris , including microbes , into the growing autophagosome . A specific part of these receptors called LIR , short for LC3-interacting region motif , connects the receptors to LC3 . Some bacteria have evolved mechanisms to avoid autophagy or can even hijack the process to survive in the host cell . For example , the bacterium Legionella pneumophila , which causes Legionnaire’s disease , manipulates the main molecular pathways involved in autophagy to avoid being digested by the host cell . The bacterium injects a protein called RavZ into the cell , which splits the lipid component from LC3-PE . It was still unknown , however , how RavZ can recognize and split LC3-PE . Yang et al . created LC3 proteins with different PE fragments to study the molecular pathways underlying this process . The experiments revealed that RavZ also contains LIR motifs that it uses to recognize and attach to LC3 . After RavZ binds , it extracts LC3-PE from the membrane of autophagosomes and integrates the PE part into its own lipid-binding site . RavZ then splits LC3-PE by removing the lipid component of the protein . Building on this knowledge , Yang et al . were able to experimentally prevent RavZ from breaking up LC3-PE . This suggests that hindering RavZ from binding by blocking its LIR motif could represent a potential pharmaceutical approach to stop L . pneumophila from avoiding autophagy . A next step will be to confirm if blocking RavZ could indeed support autophagy . It will also be useful to find out if other microbes use the same mechanisms as L . pneumophila to avoid autophagy . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2017 | Elucidation of the anti-autophagy mechanism of the Legionella effector RavZ using semisynthetic LC3 proteins |
Long Interspersed Nuclear Element-1 ( LINE-1 , L1 ) is a mobile genetic element active in human genomes . L1-encoded ORF1 and ORF2 proteins bind L1 RNAs , forming ribonucleoproteins ( RNPs ) . These RNPs interact with diverse host proteins , some repressive and others required for the L1 lifecycle . Using differential affinity purifications , quantitative mass spectrometry , and next generation RNA sequencing , we have characterized the proteins and nucleic acids associated with distinctive , enzymatically active L1 macromolecular complexes . Among them , we describe a cytoplasmic intermediate that we hypothesize to be the canonical ORF1p/ORF2p/L1-RNA-containing RNP , and we describe a nuclear population containing ORF2p , but lacking ORF1p , which likely contains host factors participating in target-primed reverse transcription .
Sequences resulting from retrotransposition constitute more than half of the human genome and are considered to be major change agents in eukaryotic genome evolution ( Kazazian , 2004 ) . L1 retrotransposons have been particularly active in mammals ( Furano et al . , 2004 ) , comprising ~20% of the human genome ( Lander et al . , 2001 ) ; somatic retrotransposition has been widely implicated in cancer progression ( Lee et al . , 2012; Tubio et al . , 2014 ) and may even play a role in neural development ( Muotri et al . , 2005 ) . Despite the magnitude of their contributions to mammalian genomes , L1 genes are modest in size . A full-length L1 transcript is ~6 knt long and functions as a bicistronic mRNA that encodes two polypeptides , ORF1p and ORF2p ( Ostertag and Kazazian , 2001 ) , which respectively comprise a homotrimeric RNA binding protein with nucleic acid chaperone activity ( Martin and Bushman , 2001 ) and a multifunctional protein with endonuclease and reverse transcriptase activities ( Mathias et al . , 1991; Feng et al . , 1996 ) . Recently , a putative primate-specific third ORF , named ORF0 , has been identified on the Crick strand of the L1 gene; this ORF encodes a 71 amino acid peptide and may generate insertion-site-dependent ORFs via splicing ( Denli et al . , 2015 ) . ORF1p and ORF2p are thought to interact preferentially with the L1 RNA from which they were translated ( in cis ) , forming a ribonucleoprotein ( RNP ) ( Kulpa and Moran , 2006; Taylor et al . , 2013 ) considered to be the canonical direct intermediate of retrotransposition ( Hohjoh and Singer , 1996; Kulpa and Moran , 2005; Martin , 1991; Kulpa and Moran , 2006; Doucet et al . , 2010 ) . L1 RNPs also require host factors to complete their lifecycle ( Suzuki et al . , 2009; Peddigari et al . , 2013; Dai et al . , 2012; Taylor et al . , 2013 ) and , consistent with a fundamentally parasitic relationship ( Beauregard et al . , 2008 ) , the host has responded by evolving mechanisms that suppress retrotransposition ( Goodier et al . , 2013; Arjan-Odedra et al . , 2012; Goodier et al . , 2012; Niewiadomska et al . , 2007 ) . It follows that as the host and the parasite compete , L1 expression is likely to produce a multiplicity of RNP forms engaged in discrete stages of retrotransposition , suppression , or degradation . Although L1 DNA sequences are modestly sized compared to typical human genes , L1 intermediates are nevertheless RNPs with a substantially sized RNA component; e . g . larger than the ~5 knt 28S rRNA ( Gonzalez et al . , 1985 ) and approximately three to four times the size of a ‘typical’ mRNA transcript ( Lander et al . , 2001; Sommer and Cohen , 1980 ) . Therefore , it is likely that many proteins within L1 RNPs form interactions influenced directly and indirectly by physical contacts with the L1 RNA . We previously reported that L1 RNA comprised an estimated ~25% of mapped RNA sequencing reads in ORF2p-3xFLAG affinity captured fractions ( Taylor et al . , 2013 ) . We also observed that the retention of ORF1p and UPF1 within affinity captured L1 RNPs was reduced by treatment with RNases ( Taylor et al . , 2013 ) . In the same study we observed that two populations of ORF2p-associated proteins could be separated by split-tandem affinity capture ( ORF2p followed by ORF1p ) , a two-dimensional affinity enrichment procedure ( Caspary et al . , 1999; Taylor et al . , 2013 ) . Initial characterization of these two L1 populations by western blotting suggested that discrete L1 populations were likely primed for function in different stages of the lifecycle . We therefore expected additional uncharacterized complexity in the spectrum of L1-associated complexes present in our affinity enriched fractions . In this study , we have used quantitative mass spectrometry ( MS ) to investigate the proteomic characteristics of endogenously assembled ectopic L1-derived macromolecules present in an assortment of affinity-enriched fractions . We revisited RNase treatment and split-tandem affinity capture approaches and complemented them with RNA sequencing , enzymatic analysis , and in-cell localization of ORF proteins by immunofluorescence microscopy ( see also the companion manuscript by Mita et al . , 2018 ) . We additionally explored proteomes associated with catalytically-inactivated ORF2p point mutants and monitored the rates of protein exchange from L1 macromolecules in vitro . Taken together , our data support the existence of a variety of putative L1-related protein complexes .
Figure 1 ( panels A-C ) illustrates the approach and displays the findings of our assay designed to reveal which proteins depend upon the presence of intact L1 RNA for retention within the obtained L1 RNPs . Briefly , metabolically-labeled affinity captured L1s were treated either with a mixture of RNases A and T1 — thus releasing proteins that require intact RNA to remain linked to ORF2p and the affinity medium — or BSA , as an inert control . After removing the fractions released by the RNase or BSA treatments , the proteins remaining on the affinity media were eluted with lithium dodecyl sulfate ( LDS ) , mixed together , and then analyzed by MS . Proteins released , and so depleted , by RNase treatment were thus found to be more abundant in the BSA-treated control . The results obtained corroborate and extend our previous findings: ORF1p and UPF1 exhibited RNase-sensitivity ( Taylor et al . , 2013 ) . We also observed that ZCCHC3 and MOV10 exhibited RNase-sensitivity to a level similar to ORF1p . The remaining I-DIRT significant proteins were RNase-resistant in this assay . With the exception of the PABPC1/4 proteins ( and ORF2p itself , see Discussion ) , the I-DIRT significant proteins ( colored nodes , Figure 1C ) that were resistant to RNase treatment ( nearest the origin of the graph ) classify ontologically as nuclear proteins ( GO:0005634 , p ≈ 3 × 10−4 , see Materials and methods ) . These same proteins were previously observed as specific L1 interactors in I-DIRT experiments targeting ORF2p but not in those targeting ORF1p; in contrast , the proteins that demonstrated RNase-sensitivity: ORF1 , MOV10 , ZCCHC3 , and UPF1 were observed in both ORF1p and ORF2p I-DIRT experiments ( Table 1 ) . Stated another way , the proteins released upon treating an affinity captured ORF2p fraction with RNases are among those that can also be obtained when affinity capturing ORF1p directly , while those that are RNase-resistant are not ORF1p interactors ( Taylor et al . , 2013 ) . The ORF1p-linked , I-DIRT significant , RNase-sensitive proteins were too few to obtain a high confidence assessment of ontological enrichment; but , when combined with the remaining proteins exhibiting sensitivity to RNase treatment ( black nodes , Figure 1C ) , they together classified as 'RNA binding' ( GO:0003723 , p ≈ 1 × 10−11 ) . This analysis also revealed a statistically significant overrepresentation of genes associated with the exon junction complex ( EJC , GO: 0035145 , p ≈ 1 × 10−6 , discussed below ) . Hence , the overlapping portion of the ORF1p- and ORF2p-associated interactomes appeared to depend upon intact L1 RNA . Host-encoded proteins segregated into groups that responded differentially to RNase treatment , with a substantial population of RNase-resistant interactors linked to both ORF2p and the nucleus . This observation led to the hypothesis that our ORF2p-3xFLAG affinity captured L1s constitute a composite purification of at least , but not limited to , ( 1 ) a population of L1-RNA-dependent , ORF1p/ORF2p-containing L1 RNPs , and ( 2 ) an ORF1p-independent nuclear population associated with ORF2p . While effects of PABPC1 , MOV10 , and UPF1 on L1 activity have been described ( Arjan-Odedra et al . , 2012; Taylor et al . , 2013; Dai et al . , 2012 ) , effects of ZCCHC3 on L1 remained uncharacterized . ZCCHC3 is an RNA-binding protein associated with poly ( A ) + RNAs ( Castello et al . , 2012 ) but otherwise little is known concerning its functions . Notably , in a genome-wide screen , small interfering ( si ) RNA knockdown of ZCCHC3 was observed to increase the infectivity of the Hepatitis C , a positive sense RNA virus ( Li et al . , 2009 ) ; and ZCCHC3 was observed to copurify with affinity captured HIV , a retrovirus , at a very high SILAC ratio ( >10 ) , supporting the specificity of this interaction ( Engeland et al . , 2014 ) . We therefore explored the effects on L1 mobility both of over-expression and siRNA knockdown of ZCCHC3 . Over-expression of ZCCHC3 reduced L1 retrotransposition to ~10% that observed in the control , consistent with a negative regulatory role for ZCCHC3 in the L1 lifecycle; small interfering RNA ( siRNA ) knockdown of ZCCHC3 induced a modest increase in retrotransposition compared to a scrambled control siRNA ( ~1 . 9x ± 0 . 1; Supplementary file 2 ) . Moreover , although not among our I-DIRT hits ( see Discussion ) , the presence of EJC components ( MAGOH , RBM8A , EIF4A3 , UPF1 ) among the RNase-sensitive fraction of proteins intrigued us , given that L1 genes are intronless . We speculated that L1s may use EJCs to enhance nuclear export , evade degradation by host defenses , and/or aggregate with mRNPs within cytoplasmic granules . For this reason we carried out a series of siRNA knockdowns of these EJC components and other physically or functionally related proteins found in the affinity captured fraction ( listed in Supplementary file 2 ) . siRNA knockdowns of RBM8A and EIF4A3 caused inviability of the cell line . We found that knocking-down MAGOH or the EJC-linked protein IGF2BP1 ( Jønson et al . , 2007 ) reduced retrotransposition by ~50% , consistent with a role in L1 proliferation; although these knockdowns also caused a reduction in viability of the cell line ( see Discussion ) . To further test our hypothesis and better characterize the components of our L1 fraction , we conducted split-tandem affinity capture . Figure 1 ( panels D-F ) illustrates the approach and displays the findings of the assay , which physically separated ORF1p/ORF2p-containing L1 RNPs from a presumptive 'only-ORF2p-associated' population . Briefly , metabolically-labeled L1s were affinity captured by ORF2p-3xFLAG ( first dimension ) and the obtained composite was subsequently further fractionated by α-ORF1p affinity capture ( second dimension , or split-tandem capture ) , resulting in α-ORF1p-bound and unbound ( supernatant ) fractions . The bound fraction was eluted from the affinity medium with LDS ( elution ) . The supernatant and elution fractions were then mixed and analyzed by MS to ascertain proteomic differences between them . The α-ORF1p elution contained the population of proteins physically linked to both ORF2p and ORF1p , whereas the supernatant contained the proteins associated only with ORF2p ( and , formally , those which have dissociated from the ORF1p/ORF2p RNP ) . The results corroborated our previous observations that: ( i ) almost all of the ORF1p partitioned into the elution fractions , ( ii ) a quarter of the ORF2p ( ~26% ) followed ORF1p during the α-ORF1p affinity capture , ( iii ) roughly half of the UPF1 ( ~55% ) followed ORF1p , and ( iv ) most of the PCNA ( ~87% ) remained in the ORF1p-depleted supernatant fraction ( Figure 1F , and consistent with prior estimates based on protein staining and western blotting [Taylor et al . , 2013] ) ; thus ( v ) supporting the existence of at least two distinct populations of L1-ORF-protein-containing complexes in our affinity purifications . The population eluted from the α-ORF1p affinity medium ( Figure 1D , far right gel lane , and nodes located in the upper right of the graph , panel F ) is consistent with the composition of the ORF1p/ORF2p-containing L1 RNP suggested above . Our split-tandem separation segregated the constituents of the L1 fraction comparably to the RNase-sensitivity assay , both in terms of which proteins co-segregated with ORF1p/ORF2p ( compare Figure 1C and F , blue nodes , upper right of graphs ) as well as those which appear to be linked only to ORF2p ( compare Figure 1C and F , magenta nodes , lower left of the graphs ) . The ORF1p/ORF2p RNPs obtained by split-tandem capture included putative in vivo interactions associated with both α-ORF1p and α-ORF2p I-DIRT affinity capture experiments; whereas the unbound , ORF1p-independent fraction includes proteins previously observed as significant only in α-ORF2p I-DIRT experiments ( Table 1 ) . Analysis of the nodes whose degree of ORF1p association was similar to that of UPF1 ( blue nodes exhibiting ≥55% ORF1p co-partitioning , Figure 1F ) revealed that they map ontologically to a ‘cytoplasmic ribonucleoprotein granule’ classification ( GO:0036464 , p ≈ 6 × 10−8; see Discussion ) . In contrast , all sixteen proteins exhibiting ORF1p co-partitioning approximately equal to or less than that of ORF2p were predominantly found in the supernatant fraction and were enriched for cell-compartment-specific association with the nucleus ( GO:0005634 , p ≈ 4 × 10−5; Figure 1F: all magenta nodes ≤36% ) . These two fractions therefore appear to be associated with different cell compartments , reaffirming our postulate: the ORF1p/ORF2p-containing population is a cytoplasmic intermediate related to the canonical L1 RNP typically ascribed to L1 assembly in the literature , and the predominantly ORF2p-associated population comprises a putative nuclear interactome . From the same analysis , we noted that PURA , PURB , PCNA , and TOP1 which all partition predominantly with nuclear L1 , exhibited an ontological co-enrichment ( termed 'nuclear replication fork , ' GO:0043596 , p ≈ 3 × 10−4 ) . The nodes representative of PURA , PURB , and PCNA appeared to exhibit a striking proximity to one another , suggesting highly similar co-fractionation behavior potentially indicative of direct physical interactions . In an effort to examine this possibility , we graphed the frequency distribution of the proximities of all three-node-clusters observed within Figure 1F , revealing the likelihood of the PURA/PURB/PCNA cluster to be p=3 . 2×10−7 ( see Appendix 1 ) . We therefore concluded that PURA , PURB , PCNA , and ( perhaps at a lower affinity ) TOP1 , likely constitute a physically associated functional module interacting with L1 . In further support of this assertion , we noted that known functionally linked protein pairs PABPC1/PABPC4 ( cytoplasmic ) ( Jønson et al . , 2007; Katzenellenbogen et al . , 2007 ) and HSPA8/HSPA1A ( nuclear ) ( Jønson et al . , 2007; Nellist et al . , 2005 ) also exhibited comparable co-partitioning by visual inspection , and statistical testing of these clusters revealed the similarity of their co-partitioning to be significant at p ≈ 0 . 001 for the former , and p ≈ 0 . 0002 for the latter . The observed variation in co-partitioning behavior between the different proteins comprising the nuclear L1 fraction might reflect the presence of multiple distinctive ( sub ) complexes present within this population . To validate our hypothesis that these proteins are associated with ORF2p in the nucleus , possibly engaged with host genomic DNA , we carried out ORF2p-3xFLAG affinity capture from chromatin-enriched sub-cellular fractions and found that the co-captured proteins we identified ( Supplementary file 3 ) overlapped with those described above as nuclear interactors , including: PARP1 , PCNA , UPF1 , PURA , and TOP1 . We previously demonstrated that silencing PCNA expression adversely affects L1 retrotransposition ( Taylor et al . , 2013 ) , in this study we found that knocking down TOP1 approximately doubled retrotransposition frequency , while a more modest 1 . 4x increase effect was observed for PURA , and no substantial effect was observed for PURB , compared to a scrambled siRNA control . In contrast , over-expression of PURA reduced retrotransposition to ~20% of the expected level ( Supplementary file 2 ) . IPO7 was also observed among the putative ORF2p co-factors within the chromatin enriched fraction , congruent with its matching behavior in Figure 1C and F . Notably , IPO7 functions as a nuclear import adapter for HIV reverse transcription complexes ( Fassati et al . , 2003 ) . Several other proteins that were observed did not previously exhibit I-DIRT specificity ( Supplementary file 3 ) . Because the L1 RNA is an integral component of proliferating L1s , and because we observed that interactions between ORF2p , ORF1p , and some host proteins were sensitive to treatment with RNases , we sought to characterize the RNAs present in our samples . We extracted RNAs from each of the three fractions produced by split-tandem affinity capture ( Figure 1D ) and carried out RNA sequencing; Figure 2A displays the sequence coverage observed across the entirety of our synthetic L1 construct in each fraction , revealing a normalized ~2 fold difference in abundance between the elution and supernatant fractions . Synthetic L1s constituted ~60% of the mapped , annotated sequence reads in the fractions eluted from the α-FLAG and α-ORF1p affinity media , and ~30% of the reads in the ORF1p-depleted supernatant fraction; sequencing reads mapping to protein coding genes made up the majority of the remaining annotated population in all fractions . We observed that a substantial number of reads mapped to unannotated regions of the human genome , in particular in the supernatant fraction , enriched for putative nuclear L1 complexes; the breakdown of mapped and annotated sequencing reads is summarized in Figure 2B and expanded in Supplementary file 4 . Retrotransposition-competent L1 RNPs form in cis , with ORF proteins binding to the L1 RNA that encoded them ( ‘cis preference’ ) , presumably at the site of translation in the cytoplasm ( Kulpa and Moran , 2006; Wei et al . , 2001 ) . Given that ORF1/2p partitioned to the split-tandem elution fraction ( cytoplasmic ) along with the greater fraction of L1 RNA , yet only ORF2p and a lesser portion of the L1 RNA were observed in the supernatant ( nuclear ) , an important consideration regarding these fractions is: to what extent they contain L1 macromolecules capable of proliferation . To address this question , we performed the LINE-1 element amplification protocol ( LEAP ) on split-tandem affinity captured fractions ( Figure 2C; Supplementary file 4 ) , including a ΔORF1 construct ( pLD561 ) as a control ( Taylor et al . , 2013 ) . LEAP is currently the best biochemical assay for functional co-assembly of L1 RNA and proteins ( Kulpa and Moran , 2006 ) ; it measures the ability of ORF2p to amplify its associated L1 RNA by reverse transcription . To execute LEAP on the α-ORF1p affinity captured fraction , we developed a competitive di-peptide elution reagent based on the linear peptide sequence used to generate the α-ORF1p 4H1 monoclonal antibody: residues 35–44 in ORF1p ( [Khazina et al . , 2011; Taylor et al . , 2013]; see Materials and methods ) . We were thus able to assay the partitioning of enzymatic activity within the different populations of copurifying proteins in a split-tandem affinity capture experiment . Our data showed robust LEAP activity in both nuclear and cytoplasmic split-tandem supernatant and elution fractions . We note that our 3xFLAG eluted fractions have been shown to possess ~70 fold higher specific activity than L1 RNPs obtained by sucrose cushion velocity sedimentation ( Taylor et al . , 2013 ) , hence the activity levels detected far exceed those obtained by sedimentation . Although our proteomic and biochemical analyses supported the existence of distinctive nuclear and cytoplasmic L1 populations , our prior immunofluorescence ( IF ) analyses did not reveal an apparent nuclear population , leading us to revisit IF studies . Previously , IF of ORF1p and ORF2p in HeLa and HEK-293T cells yielded two striking observations: ( i ) ORF2 expression was seemingly stochastic , with ORF2p observed in ~30% of cells; and ( ii ) while ORF1p and ORF2p co-localized in cells that exhibited both , we did not observe an apparent nuclear population of either protein ( Taylor et al . , 2013 ) . Subsequently , we noted an absence of mitotic cells from these preparations . Reasoning that these cells were lost due to selective adherence on glass slides , and noting that cell division has been reported to promote L1 transposition ( Xie et al . , 2013; Shi et al . , 2007 ) , we repeated the assays using puromycin-selected Tet-on HeLa cells grown on fibronectin coated coverslips . The results are shown in Figure 3 . The modified IF assay corroborated our prior results in that nearly all the cells exhibited cytoplasmic ORF1p and a minority subset of ~1/3rd also exhibited co-localized cytoplasmic ORF2p ( Figure 3A , top row ) . We also observed an uncommon and previously unrecognized subpopulation of cells , consisting of pairs exhibiting nuclear localized ORF2p ( Figure 3A , middle row ) ; because these cells occurred in proximal pairs , we presumed them to have recently gone through mitosis . Statistical analysis of microscopy images displaying cells with nuclear localized ORF2p confirmed their proximities to be significantly closer than those of randomly selected cells ( Figure 3B; Supplementary file 5 ) . Expression of ORF2 in the absence of ORF1 ( ΔORF1; pLD561 ) resulted in the majority of cells exhibiting cytoplasmic ORF2p , consistent with our previous work ( Taylor et al . , 2013 ) . We did not observe instances of nuclear ORF2p using the ΔORF1 construct ( Figure 3A , bottom row ) , suggesting that ORF1p is required for ORF2p nuclear localization ( see Discussion ) . In a separate study , including more detailed analyses of ORF protein localization , Mita et al . , 2018 observed that both ORF proteins enter the nucleus of HeLa cells during mitosis , however , nuclear ORF1p does not seem to be physically associated with nuclear ORF2p ( see Discussion ) . Taken together , the data obtained from the modified IF experiments aligned well with our proteomic and biochemical data; L1 expression resulted in at least two distinct populations: cytoplasmic complexes containing both ORF1p and ORF2p , and nuclear complexes containing ORF2p while potentially lacking ORF1p . Based on the hypothesis that our composite purifications contain bona fide nuclear intermediates , we decided to explore the effects of catalytic point mutations within the ORF2p endonuclease and reverse transcriptase domains , respectively . We reasoned that such mutants may bottleneck L1 intermediates at the catalytic steps associated with host gDNA cleavage and L1 cDNA synthesis , potentially revealing protein associations that are important for these discrete aspects of target-primed reverse transcription ( TPRT ) , the presumed mechanism of L1 transposition ( Luan et al . , 1993; Feng et al . , 1996; Cost et al . , 2002 ) . For this we used an H230A mutation to inactivate the endonuclease activity ( EN-/pLD567 ) , and a D702Y mutation to inactivate the reverse transcriptase activity ( RT-/pLD624 ) ( Taylor et al . , 2013 ) . Figure 4 illustrates the approach and displays the findings of our assay . Broadly , while we observed comparable RNA-level properties between samples ( Figure 4B , Supplementary file 4 ) , our findings revealed several classes of distinctive protein-level behaviors ( Figure 4C ) . Two classes of behavior appeared to be particularly striking: ( 1 ) the yield of constituents of cytoplasmic L1s was reduced , relative to WT , by the EN- mutation , yet elevated by the RT- mutation ( Figure 4C , left side ) ; and ( 2 ) numerous constituents of nuclear L1s were elevated in yield by the EN- mutation but reduced or nominally unchanged , relative to WT , by the RT- mutation ( Figure 4C , right side ) . With respect to the second group , IPO7 , NAP1L4 , NAP1L1 , FKBP4 , HSP90AA1 , and HSP90AB1 were all elevated in the EN- mutants , potentially implicating these proteins as part of an L1 complex ( or complexes ) immediately preceding DNA cleavage . Notably , there is a third class of proteins , including PURA/B , PCNA , TOP1 , and PARP1 , that all respond similarly to both EN- and RT- mutants compared to WT , exhibiting reduced associations with the mutant L1s; although , the RT- mutant showed a larger effect size on the PURA/B proteins . These data suggest that cleavage of the host genomic DNA by ORF2p fosters associations between L1 and this third class of proteins , but that interactions with PURA/B may be further enhanced by L1 cDNA production . Other nuclear L1 proteins: HSPA8 , HAX1 , HSPA1A , TUBB , and TUBB4B were increased in both mutants . To better visualize the range of behaviors exhibited by our proteins of interest , and the population at large , we cross-referenced the relative enrichments of each protein detected in both experiments , shown in Figure 4D . We noted the same striking trend mentioned above , that two seemingly opposite behavioral classes of interactors could also be observed globally among all proteins associating with ORF2p catalytic mutants ( see Figure 4C , left side and right side , and Figure 4D ) , creating the crisscross pattern displayed ( see also Figure 4—figure supplement 1 ) . Notably , the pattern observed appears to track with the relative behavior of ORF1p , which , along with other cytoplasmic L1 factors is elevated in RT- mutants and reduced in EN- mutants . We therefore speculate that the sum of observed interactomic changes include effects attributable directly to the catalytic mutations as well as potential indirect effects resulting in increased cytoplasmic RNPs ( including ORF1p ) in the RT- mutant . We next decided to measure the in vitro dynamics of proteins copurifying with affinity captured L1s , reasoning that proteins with comparable profiles are likely candidates to be physically linked to one another or otherwise co-dependent for maintaining stable interactions with L1s . To achieve this , we first affinity captured heavy-labeled , affinity-tagged L1s and subsequently incubated them , while immobilized on the medium , with light-labeled , otherwise identically prepared cell extracts from cells expressing untagged L1s ( Luo et al . , 2016 ) . In this scenario , heavy-labeled proteins present at the zero time point are effectively ‘infinitely diluted’ with light-labeled cell extract . The exchange of proteins , characterized by heavy-labeled proteins decaying from the immobilized L1s and being replaced by light-labeled proteins supplied by the cell extract , was monitored by quantitative MS . These experiments were conducted using constructs based on the naturally occurring L1RP sequence ( Dai et al . , 2014; Taylor et al . , 2013; Kimberland et al . , 1999 ) . Figure 5 illustrates the approach and displays the findings of our assay . We observed three distinctive clusters of behaviors ( Figure 5B , C ) . Notably , ORF1p , ZCCHC3 , and the cytoplasmic poly ( A ) binding proteins clustered together , forming a relatively stable core complex . Exhibiting an intermediate level of relative in vitro dynamics , UPF1 and MOV10 clustered with TUBB , TUBB4B , and HSP90AA1 . A third , and least stable , cluster consisted of only nuclear L1 interactors . Having observed coordinated and distinctive behaviors exhibited by groups of L1 interacting proteins across several distinctive biochemical assays , we then integrated the data and calculated the behavioral similarity of the I-DIRT-significant interactors , producing a dendrogram; Figure 6 displays their relative similarities . A cluster containing the putative cytoplasmic L1 components ( MOV10 , UPF1 , ZCCHC3 , PABC1/4 , ORF1p ) was observed , as was a cluster containing PURA/B , PCNA , TOP1 , PARP1 , aligning with our assessments of the separated datasets ( Figures 1 , 4 and 5 ) . In addition to these , we also observed three distinctive clusters derived from the nuclear L1 interactome . We believe that this is likely to reflect the presence of a collection of distinctive macromolecules .
ORF1p , MOV10 , UPF1 , and ZCCHC3 are released from L1 RNPs by treatment with RNases ( Figure 1 ) , indicating the importance of the L1 RNA in the maintenance of these interactions . In this context , the L1 ORF and poly ( A ) binding proteins support L1 proliferation ( Kulpa and Moran , 2006; Dai et al . , 2012; Wei et al . , 2001 ) , whereas ZCCHC3 ( Supplementary file 2 ) and MOV10 ( Goodier et al . , 2012; Arjan-Odedra et al . , 2012 ) function in repressive capacities . Although UPF1 might also be expected to operate in a repressive capacity through its role in nonsense mediated decay ( NMD ) , we previously demonstrated that UPF1's role does not apparently resemble that of canonical NMD and it acts as an enhancer of retrotransposition despite negatively affecting L1 RNA and protein levels , supporting the possibility of repressive activity in the cytoplasm and proliferative activity in the nucleus ( Taylor et al . , 2013 ) . Notably , MOV10 has been implicated in the recruitment of UPF1 to mRNA targets through protein-protein interactions ( Gregersen et al . , 2014 ) . However , we observed that MOV10 exhibited a greater degree of RNase-sensitivity than UPF1 , indicating that , if MOV10 directly modulates the UPF1 interactions with L1 , a sub-fraction of UPF1 exhibits a distinct behavior ( UPF1 is ~62% as sensitive to RNase treatment as MOV10 , Figure 1C ) . Bimodal UPF1 behavior can also be seen in split-tandem capture experiments , only about half of the UPF1 exhibited ORF1p-like partitioning with the canonical L1 RNP ( Figure 1F ) . Moreover , UPF1 was recovered with L1s affinity captured from fractionated chromatin ( further discussed below ) , and only about half of the UPF1 exhibits ORF1p-like partitioning with the canonical L1 RNP ( Figure 1F ) . Presumably , the RNase-sensitive fraction , released in concert with MOV10 , is the same fraction observed in cytoplasmic L1s obtained by split-tandem capture . In contrast , PABPC1 and C4 exhibit strong ORF1p-like partitioning ( comparable to MOV10 ) , but appear wholly insensitive to RNase treatment . This is most likely due to the fact that neither RNase A nor T1 cleave RNA at adenosine residues ( Volkin and Cohn , 1953; Yoshida , 2001 ) ; hence poly ( A ) binding proteins may not be ready targets for release from direct RNA binding by the assay implemented here ( or generally , using these ribonucleases ) . Failure to release ORF2p into the supernatant upon RNase treatment is expected due to its immobilization upon the affinity medium ( Dai et al . , 2014 ) . However , we note that ORF2p binding to the L1 RNA has also been proposed to occur at the poly ( A ) tail ( Doucet et al . , 2015 ) , raising the related possibility of close physical association on the L1 RNA between ORF2p and PABPC1/4 in cytoplasmic L1 RNPs . ORF1p , PABPC1/4 , MOV10 , ZCCHC3 , and UPF1 , all behaved comparably in response to EN- and RT- catalytic mutations , decreasing together in EN- mutants , and increasing together in RT- mutants ( Figure 4C ) . Moreover , when the exchange of proteins within L1 RNPs was monitored directly , PABPC1/4 and ZCCHC3 exhibited nearly identical stability , well above the background distribution; UPF1 and MOV10 also exhibited comparable kinetics to one another , falling into an intermediary stability cluster ( Figure 5B , C ) . RNase-sensitivity was displayed by numerous proteins not previously identified as putative L1 interactors ( Table 1 , Figure 1; [Taylor et al . , 2013] ) . A known limitation of I-DIRT ( and many SILAC-based analyses ) is that it cannot discriminate non-specific interactors from specific but rapidly exchanging interactors ( Wang and Huang , 2008; Luo et al . , 2016; Smart et al . , 2009 ) . Our samples likely contain rapidly exchanging , physiologically relevant factors that were not revealed by I-DIRT under the experimental conditions used . With this in mind , we note members of the exon junction complex ( EJC ) , RBM8A ( Y14 ) , EIF4A3 ( DDX48 ) , and MAGOH , are among our RNase-sensitive constituents , all exhibiting a similar degree of RNase-sensitivity ( Figure 1C , labeled black dots ) . Crucially , these proteins are physically and functionally connected to UPF1 ( reviewed in [Schweingruber et al . , 2013] ) , and physically to MOV10 ( Gregersen et al . , 2014 ) , both validated L1 interactors . We therefore hypothesize that EJCs may constitute bona fide L1 interactors missed in our original screen . This may seem unexpected because canonical L1 RNAs are thought not to be spliced , but this assumption has been challenged by one group ( Belancio et al . , 2006 ) , and splicing-independent recruitment of EJCs has also been demonstrated ( Budiman et al . , 2009 ) . Perhaps more compelling , EJC proteins exhibited a striking similarity in RNase-sensitivity to MOV10 ( Figure 1C ) . EIF4A3 has been suggested to form an RNA-independent interaction with MOV10 ( Gregersen et al . , 2014 ) , and MOV10 is a known negative regulator of L1 , making it attractive to speculate that these proteins were recruited and released in concert with MOV10 and/or UPF1 . Ectopically expressed canonical L1 RNPs have been shown to accumulate in cytoplasmic stress granules ( Doucet et al . , 2010; Goodier et al . , 2010 ) , and our observation of UPF1 , MOV10 , and MAGOH in the RNase-sensitive fraction is consistent with this characterization ( Jain et al . , 2016 ) . However , the additional presence of EIF4A3 and RBM8A suggested that our RNPs may instead overlap with IGF2BP1 ( IMP1 ) granules , reported to be distinct from stress granules ( Jønson et al . , 2007; Weidensdorfer et al . , 2009 ) . Consistent with this possibility , we observed IGF2BP1 , YBX1 , DHX9 , and HNRNPU within the mixture of co-captured proteins ( Supplementary file 1 ) . We did not , however , observe canonical stress granule markers G3BP1 or TIA1 ( Goodier et al . , 2007; Jain et al . , 2016; Doucet et al . , 2010 ) . Surprisingly , siRNA knockdown of IGF2BP1 substantially reduced L1 retrotransposition; however , we note that the cytotoxicity associated with knocking-down EJC components may confound interpretation ( Supplementary file 2 ) . Given the result obtained , IGF2BP1 appears to support L1 proliferation . Consistent with an established function ( Bley et al . , 2015; Weidensdorfer et al . , 2009 ) , IGF2BP1 granules may sequester and stabilize L1 RNPs in the cytoplasm , creating a balance of L1 supply and demand that favors proliferation over degradation . Although human L1 does not contain a known IRES , it is known that ORF2 is translated by a non-canonical mechanism ( Alisch et al . , 2006 ) , and IGF2BP1 may promote this ( Weinlich et al . , 2009 ) . The fraction eluted from the α-ORF1p medium contained the population of proteins physically linked to both ORF2p and ORF1p and greatly resembled the components released upon RNase treatment , hence these linkages primarily occur through the L1 RNA ( or are greatly influenced by it ) . In contrast , the supernatant from the α-ORF1p affinity capture contained the proteins we speculate to be associated with ORF2p , but not ORF1p; moreover , fully intact RNA does not appear to be essential to the maintenance of these interactions . An exciting alternate interpretation to direct protein-protein linkage is that some of the L1 RNAs in this population may be at least partially hybridized to L1 cDNAs , which would render them RNase resistant: at the salt concentration used in our RNase assay ( 0 . 5 M; Figure 1C ) , RNase A is unlikely to cleave the RNA component of DNA/RNA hybrids ( Halász et al . , 2017; Wyers et al . , 1973 ) , and such activity is not expected of RNase T1 . This interpretation is supported by several pieces of indirect evidence: ( 1 ) the presence of well-known DNA binding factors ( Figure 1 ) ; ( 2 ) the presence of several of these same factors ( PARP1 , PCNA , PURA , and TOP1 ) in ORF2p-3xFLAG affinity captured from enriched chromatin ( Supplementary file 3 ) ; ( 3 ) The pronounced decrease in stable in vivo co-assembly of TOP1 , PCNA , PARP1 , PURA , and PURB in affinity captured L1 fractions harboring ORF2p EN- and RT- mutations ( Figure 4 ) , with a greater effect in RT- mutations; and ( 4 ) our L1 preparations exhibit RT activity ( Figure 2C , in vitro; as well as in vivo [Taylor et al . , 2013] ) . If true , linkage of subcomplexes via DNA/RNA hybrids would further support the nuclear origin of much of this fraction; further study is needed . Notable within this group of putative nuclear interactors was the PURA/PURB/PCNA cluster ( Figure 1F ) , with TOP1 also in close proximity , ontologically grouping to the nuclear replication fork ( GO:0043596 ) . Separately , a few physical and functional connections have been shown for PURA/PURB ( Knapp et al . , 2006; Kelm et al . , 1999; Mittler et al . , 2009 ) , PCNA/TOP1 ( Takasaki et al . , 2001 ) , and PURA/PCNA ( Qin et al . , 2013 ) . Notably , PURA , PURB , and PCNA have been independently linked to replication-factor-C/replication factor-C-like clamp loaders ( Kubota et al . , 2013; Havugimana et al . , 2012 ) . Given that we also observe tight clustering of protein pairs known to be physically and functionally linked , e . g . PABPC1/4 ( Jønson et al . , 2007; Katzenellenbogen et al . , 2007 ) and HSPA8/1A ( Jønson et al . , 2007; Nellist et al . , 2005 ) , and because we have established PCNA as a positive regulator of L1 retrotransposition ( Taylor et al . , 2013 ) , we propose that the [PURA/B/PCNA/TOP1] group is a functional sub-complex of nuclear L1 . In addition , although it does not cluster as closely to the [PURA/B/PCNA/TOP1] group , PARP1 is found within the putative nuclear L1 population and is functionally linked with PCNA , specifically stalled replication forks ( Bryant et al . , 2009; Min et al . , 2013; Ying et al . , 2016 ) . Further tying them together , these proteins all also exhibited substantial affinity capture yield decreases in response to mutations that abrogated ORF2p EN or RT activity ( Figure 4 ) . This is compelling because these ORF2p enzymatic activities are required in order for it to manipulate DNA and traverse the steps of the L1 lifecycle that benefit from physical association with replication forks . One caveat to this interpretation is that , while knocking down PCNA reduced L1 retrotransposition ( Taylor et al . , 2013 ) , no such effect was observed for TOP1 or PURA/B , which led instead to mild increases in L1 activity ( Supplementary file 2 ) . These proteins may be physically assembled within a common intermediate , but functionally antagonistic . HSP90 proteins were also observed in this fraction , and are also linked with stalled replication forks ( Arlander et al . , 2003; Ha et al . , 2011 ) , but exhibited a distinctive response to catalytic mutants , accumulating in EN- mutants while exhibiting a modest decrease in RT- mutants . The recruitment of the ORF2p/PCNA complex to stalled replication forks has been also proposed by Mita et al . , 2018 . As mentioned above , we previously speculated that an RNase-insensitive fraction of L1-associated UPF1 may support retrotransposition in conjunction with PCNA in the nucleus ( Azzalin and Lingner , 2006; Taylor et al . , 2013 and Mita et al . , 2018 ) . In contrast to other PCNA-linked proteins , catalytic inactivation of ORF2p did not robustly affect the relative levels of co-captured UPF1 , and UPF1 behaved in a distinct manner during tandem capture . The equivocal behavior of UPF1 in several assays ( Figures 1 , 4 and 5 ) supports UPF1’s association with both the putative cytoplasmic and nuclear L1 populations , the latter being additionally supported by the association of UPF1 with ORF2p-3xFLAG captured from chromatin ( Supplementary file 3 ) . NAP1L4 , NAP1L1 , FKBP4 , HSP90AA1 , and HSP90AB1 ( Baltz et al . , 2012; Castello et al . , 2012; Simon et al . , 1994; Rodriguez et al . , 1997; Peattie et al . , 1992 ) are associated with RNA binding , involved in protein folding and unfolding , and function as nucleosome chaperones . An interesting possibility is that they have a nucleosome remodeling activity that may be required to allow reverse transcription to begin elongating efficiently , or for assembly of nucleosomes on newly synthesized DNA . An obvious need is the continued validation of putative interactors by in vivo assays . Genetic knockdowns coupled with L1 insertion measurements by GFP fluorescence ( Ostertag et al . , 2000 ) provide a powerful method to detect effects on L1 exerted by host factors . However , this approach can sometimes be limited by cell viability problems associated with important genes; it is therefore critical to control for this ( Supplementary file 2 ) . IF and high-resolution microscopy may be useful to demonstrate co-localization of putative L1-associated proteins and may also be informative , warranting effort to identify appropriate antibodies and assay conditions . Bolstered by our analytical successes , RNA-sequencing , LEAP , and RNase-based affinity proteomics appear as notably high-value assays for further application-specific expansion and refinement . Throughout this and our prior study ( Taylor et al . , 2013 ) we have used comparable in vitro conditions for the capture and analysis of L1 interactomes . However , we are aware that this practice has enforced a single biochemical ‘keyhole’ through which we have viewed L1-host protein associations . It is important to expand the condition space in which we practice L1 interactome capture and analysis in order to expand our vantage point on the breadth of L1-related macromolecules ( Hakhverdyan et al . , 2015 ) . In concert with this , we must develop sophisticated , automated , reliable , low-noise methods to integrate biochemical , proteomic , genomic , and ontological data; the first stages of which we have attempted in the present study . Although we have used I-DIRT to increase our chances of identifying bona fide interactors ( Tackett et al . , 2005; Taylor et al . , 2013 ) , it is clear , and generally understood , that some proteins not making the significance cut-off will nevertheless prove to be critical to L1 activity ( Byrum et al . , 2011; Luo et al . , 2016; Joshi et al . , 2013 ) , such as demonstrated by our unexpected findings with IGF2BP1 ( Supplementary file 2 ) . Through further development , including reliable integration with diverse , publicly available interactome studies , we hope to enable the detection of extremely subtle physical and functional distinctions between ( sub ) complexes and their components , considerably enhancing reliable exploration and hypothesis formation . Furthermore , it is striking that no structures of assembled L1s yet exist; these are missing data that are likely to provide a profound advance for the mechanistic understanding of L1 molecular physiology . However , we believe that with the methods presented here , endogenously assembled ORF1p/ORF2p/L1-RNA-containing cytoplasmic L1 RNPs can be prepared at sufficiently high purity and yield ( Figure 1F ) to enable electron microscopy studies . Importantly , we have shown that our affinity captured fractions are enzymatically active for reverse transcription of the L1 RNA ( Figure 2C; ( Taylor et al . , 2013 ) ) , providing some hope that cryo-electron microscopy could be used to survey the dynamic structural conformations of L1s formed during its various lifecycle stages ( Takizawa et al . , 2017 ) .
Freestyle-293 medium lacking Arginine and Lysine was custom-ordered from Life Technologies , and heavy or light amino acids plus proline were added at the same concentrations previously described ( Taylor et al . , 2013 ) , without antibiotics . Suspension-adapted HEK-293TLD were spun down , transferred to SILAC medium and grown for >7 cell divisions in heavy or light medium . On day 0 , four ( 4 ) 1L square glass bottles each containing 200 ml of SILAC suspension culture at ~2 . 5 × 106 cells/ml were transfected using 1 μg/ml DNA and 3 μg/ml polyethyleneimine ‘Max’ 40 kDa ( Polysciences , Warrington , PA , #24765 ) . A common transfection mixture was made by pre-mixing 800 μg DNA and 2 . 4 mL of 1 mg/ml PEI-Max in 40 ml Hybridoma SFM medium ( Life Technologies , Grand Island , NY , #12045–076 ) and incubating for 20 min at room temperature ( RT ) ; 10 ml of the mixture was added to each bottle . On day 1 , cells ( 200 ml ) were split 1:2 . 5 ( final two bottles each containing 250 mL ) without changing the medium . On day 3 , the cells were induced with 1 µg/ml doxycycline , and on day four the cells were harvested and extruded into liquid nitrogen . Aliquots were tested by western blot and the per-cell expression of both ORFs was indistinguishable from puromycin-selected material described previously ( Appendix 1 ) ; transfection efficiency was assessed at >95% by indirect immunofluorescence of expressed ORF proteins . The median lysine and arginine heavy isotope incorporation levels for cell lines presented in this study were >90% , determined as previously described ( Taylor et al . , 2013 ) . Four sets of 200 mg of light ( L ) and heavy ( H ) pLD401 transfected cell powders , respectively , were extracted 1:4 ( w:v ) with 20 mM HEPES-Na pH 7 . 4 , 500 mM NaCl , 1% ( v/v ) Triton X-100 ( extraction solution ) , supplemented with 1x protease inhibitors ( Roche , Indianapolis , IN , #11836170001 ) . After centrifugal clarification , all of the L and H supernatants were pooled , respectively , and then split , resulting in two sets of cleared L and H extracts equivalent to duplicate 400 mg samples from each SILAC cell powder . These four samples were each subjected to affinity capture upon 20 μl α-FLAG magnetic medium . After binding and washing , one set of L and H samples were treated with a control solution consisting of 2 μl of 2 mg/ml BSA ( Thermo Fisher Scientific , Waltham , MA , #23209 ) and 50 μl extraction solution , v:v ( Ctrl ) ; the other set of L and H samples was treated with a solution of 2 μl 2 mg/ml RNase A/5000 u/ml RNase T1 ( Thermo Fisher Scientific #EN0551 ) and 50 μl extraction solution , v:v ( RNase ) . Samples were then incubated 30 min at RT with agitation , the supernatant was removed , and the medium was washed three times with 1 ml of extraction solution . The retained captured material was eluted from the medium by incubation with 40 μl 1 . 1x LDS sample loading buffer ( Life Technologies #NP0007 ) . To enable quantitative comparisons of fractions , the samples were combined , respectively , as follows: 30 ul each of the LRNase with HCtrl , and 30 ul each of the LCtrl with HRNase . These samples were reduced , alkylated and run until the dye front progressed ~6 mm on a 4–12% Bis-Tris NuPAGE gel ( Life Technologies , as per manufacturer’s instructions ) . The gels were subsequently subjected to colloidal Coomassie blue staining ( Candiano et al . , 2004 ) and the sample regions ( ‘gel-plugs’ ) excised and processed for MS analyses , as described below . 400 mg of light ( L ) and heavy ( H ) pLD401 transfected cell powders , respectively , were extracted and clarified as above . These extracts were subjected to affinity capture on 20 μl α-FLAG magnetic medium , 30 min at 4°C , followed by native elution with 50 μl 1 mg/ml 3xFLAG peptide ( 15 min , RT ) . 45 μl of the elution were subjected to subsequent affinity capture upon 20 μl α-ORF1 magnetic medium , resulting in a 45 μl supernatant ( Sup ) fraction depleted of ORF1p . Finally , the material was eluted ( Elu ) from the α-ORF1p medium in 45 μl 2 . 2x LDS sample loading buffer by heating at 70°C for 5 min with agitation . To enable quantitative comparisons of fractions the samples were combined , respectively , as follows: 28 μl each of the LSup with HElu , and 28 μl each of the LElu with HSup . These samples were then prepared as gel-plugs ( as above ) and processed for MS analyses , as described below . Gel plugs were excised , cut into 1 mm cubes , de-stained , and digested overnight with enough 3 . 1 ng/μl trypsin ( Promega , Madison , WI , #V5280 ) in 25 mM ammonium bicarbonate to cover the pieces . In RNase-sensitivity and split-tandem SILAC analyses based on pLD401 , as well as in vitro protein exchange experiments based on pMT302 and pMT289 , an equal volume of 2 . 5 mg/ml POROS R2 20 µm beads ( Life Technologies #1112906 ) in 5% v/v formic acid , 0 . 2% v/v TFA was added , and the mixture incubated on a shaker at 4°C for 24 hr . Digests were desalted on Stage Tips ( Rappsilber et al . , 2007 ) , eluted , and concentrated by vacuum centrifuge to ~10 μl . ~3 μl were injected per LC-MS/MS analysis . RNase-sensitivity and split-tandem samples were loaded onto a PicoFrit column ( New Objective , Woburn , MA ) packed in-house with 6 cm of reverse-phase C18 material ( YMC∗Gel ODS-A , YMC , Allentown , PA ) . Peptides were gradient-eluted ( Solvent A = 0 . 1 M acetic acid , Solvent B = 0 . 1 M acetic acid in 70% v/v acetonitrile , flow rate 200 nl/min ) into an LTQ-Orbitrap-Velos or an LTQ-Orbitrap-XL mass spectrometer ( Thermo Fisher Scientific ) acquiring data-dependent CID fragmentation spectra . In vitro exchange samples were loaded onto an Easy-Spray column ( ES800 , Thermo Fisher Scientific ) and gradient-eluted ( Solvent A = 0 . 1% v/v formic acid in water , Solvent B = 0 . 1% v/v formic acid in acetonitrile , flow rate 300 nl/min ) into an Q Exactive Plus mass spectrometer ( Thermo Fisher Scientific ) acquiring data-dependent HCD fragmentation spectra . In SILAC experiments comparing inactivated ORF2p catalytic mutants to WT ( based on pLD401 [WT] , pLD567 [EN-] , and pLD624 [RT-] ) peptides were extracted from the gel in two 1 hr incubations with 1 . 7% v/v formic acid , 67% v/v acetonitrile at room temperature with agitation . Digests were partially evaporated by vacuum centrifugation to remove acetonitrile , and the aqueous component was desalted on Stage Tips . Peptides were loaded onto an Easy-Spray column ( ES800 , Thermo Fisher Scientific ) and gradient-eluted ( Solvent A = 0 . 1% v/v formic acid in water , Solvent B = 0 . 1% v/v formic acid in acetonitrile , flow rate 300 nl/min ) into an Orbitrap Fusion Tribrid mass spectrometer ( Thermo Fisher Scientific ) acquiring data-dependent fragmentation spectra ( either CID spectra alone , or CID and HCD spectra ) . Raw files were submitted to MaxQuant ( Cox and Mann , 2008 ) version 1 . 5 . 2 . 8 for protein identification and isotopic ratio calculation . Searches were performed against human protein sequences ( UP000005640 , April 2016 ) , custom L1 ORF1p and ORF2p protein sequences , common exogenous contaminants , and a decoy database of reversed protein sequences . Search parameters included fixed modification: carbamidomethyl ( C ) ; variable modification: Arg10 , Lys8 , methionine oxidation; razor and unique peptides used for protein quantitation; requantify: enabled . Contaminants , low-scoring proteins and proteins with one razor+unique peptides were filtered out from the MaxQuant output file ‘proteingroups . txt’ . The list of contaminants was uploaded from the MaxQuant web-site ( http://www . coxdocs . org/; ‘contaminants’ ) . Additionally , proteins with the ‘POTENTIAL CONTAMINANT’ column value ‘+' were filtered out . Proteins with at least two razor+unique peptides were retained for the analysis . H/ ( H + L ) and L/ ( H + L ) values were derived from unnormalized ‘ratio H/L’ values and were used for plotting label-swapped RNase-sensitivity and split-tandem data . Unnormalized ‘ratio H/L’ values were used to calculate H/ ( H + L ) in ORF2p catalytic mutant comparisons and in vitro exchange experiments . These values have been referred to as ‘affinities’ within the Supplementary Materials . Normalization and clustering procedures applied to data presented in the figures ( Supplementary file 1 ) are detailed below and also in Appendix 1 . Raw and processed data are available via ProteomeXchange with identifier PXD008542 . To plot RNase-sensitivity affinity capture results ( Figure 1C ) , these data were normalized such that proteins that did not change upon treatment with RNases are centered at the origin . The mean value and standard deviation were calculated using the distribution of distances from the origin . The distance threshold for p-value=0 . 001 was calculated using the R programming language . A circle with radius equal to the threshold was plotted . Points with distances higher than the threshold were marked as black . To plot split-tandem affinity capture results ( Figure 1F ) , these data were normalized such that the ORF1p affinity was set to one and the distribution median was maintained . Probabilities associated with selected clusters were calculated based on the frequency distributions of 2- and 3-node clusters present in the data . To plot EN- and RT- mutant affinity capture results ( Figure 4C ) , the matrix of detected proteins for each experiment ( EN- and RT- ) was filtered to retain only proteins detected in at least two replicate experiments . The difference between the affinity value of ORF2p and 0 . 5 value was calculated for each experiment . The affinities of each protein were shifted by the calculated difference . To determine the statistical significance of differentially co-captured proteins between EN- or RT- and WT , respectively , we used a 1-sample t-test and applied Benjamini-Hochberg p-value correction . To determine the statistical significance of differentially co-captured proteins between EN- and RT- we used an unpaired t-test and applied Benjamini-Hochberg p-value correction . To plot in vitro dynamics ( Figure 5B , C ) , only proteins which were identified at all time points were used . The cosine similarity method was used to calculate distances between proteins , and hierarchical clustering was used to visualize these distances . To integrate and plot the combined data ( Figure 6 ) , we calculated Euclidean and cosine distances for each I-DIRT-significant protein pair present in each experiment . Euclidean distances were rescaled to the range ( 0 , 0 . 9 ) . Proteins not detected in any common experiments were assigned a Euclidian distance of 1 after rescaling . The total distance between protein pairs was calculated as d = log ( ( rescaled Euclidean distance ) * ( cosine distance ) ) . This distance was rescaled to the range ( 0 , 1 ) . Hierarchical clustering was used to visualize the calculated distances . Genes corresponding to the proteins previously reported as significant by I-DIRT ( Taylor et al . , 2013 ) were tested for statistical overrepresentation using the default settings provided by http://www . panthnerdb . org ( Mi et al . , 2017 , 2013 ) , searches were conducted using GO complete molecular function , biological process , and cellular compartment: all results are compiled in Supplementary file 6 . RNA fractions were obtained from fractions of L1 macromolecules isolated from pLD401 expressing cells by split-tandem affinity capture ( Figure 1D ) and from pLD567 and pLD624 expressing cells by affinity capture ( Figure 4 ) . The fractions were produced as described above , except few adjustments to favor RNA extraction . Identical stock solutions were used for making buffers but were diluted to working concentration with nuclease-free water ( Thermo Fisher Scientific #4387936 ) and supplemented with RNasin ( Promega , Cat . # N2511 ) – 1:250 during sample extraction and 3xFLAG peptide elution , and 1:1000 during affinity media washing . 600 mg of cell powder was used per preparation , extracted as 3 × 200 mg and pooled after centrifugal clarification , producing ~3 ml of extract . The pooled extracts were combined with magnetic affinity medium from 30 µl of slurry . 75 µl of 1 mg/mL 3xFLAG peptide was used for elution . ½ of the sample was saved for RNA extraction ( input ) and the other ½ was carried forward to split-tandem IP , using 15 µl α-ORF1 affinity medium slurry . RNAs were extracted from input , α-ORF1 supernatant fractions , as well as directly from the α-ORF1 affinity medium ( elution ) with 500 µl of TRIzol ( Thermo Fisher Scientific #15596026 ) , following the manufacturer’s instructions . Aqueous TRIzol extracts were re-extracted in an equal volume of chloroform , and the aqueous phase was again removed; 1 µl ( ~15 ug ) of GlycoBlue ( Thermo Fisher Scientific #AM9516 ) and 2 ul of RNasin were added to this and mixed before combining with 250 µl of isopropanol and incubating for 10’ on ice to precipitate RNA . Alcohol precipitates were centrifuged at 20 k RCF for 30’ @ 4°C and the pellets were washed twice with 500 µl of cold 70% ethanol , then air dried for 5’ at RT and re-solubilized in 100 µl of nuclease-free water . Extracted RNAs in water were then further purified and concentrated using a Qiagen RNeasy MinElute Cleanup Kit ( #74204 ) following the manufacturer’s instructions , and eluted in 14 µl of nuclease-free water . 5 µl of purified RNA was used directly in RNA fragmentation . Libraries were prepared with unique barcodes and were pooled at equimolar ratios . The pool was denatured and sequenced on Illumina NextSeq 500 sequencer using high output V2 reagents and NextSeq Control Software v1 . 4 to generate 75 bp single reads , following manufacturer’s protocols ( #15048776 , Rev . E ) . Human genome hg19 GRCh37 . 87 ( FASTA ) and annotation ( GTF file ) were downloaded from ENSEMBL ( ftp://ftp . ensembl . org/pub/grch37/release-90 ) and reference FASTA and GTF files were created by combining the human genome and ORFeus-Hs from pLD401 ( Taylor et al . , 2013 ) ; Supplementary file 7: ORFeus-Hs_pLD401 . gbk ) . To map sequencing reads onto the reference genome and produce differential gene expression analysis: ( 1 ) FASTAQ files were trimmed via trimmomatic ( Bolger et al . , 2014 ) using the following parameters: -phred33 -threads 8 , LEADING:3 TRAILING:3 SLIDINGWINDOW:4:16 MINLEN:25; ( 2 ) mapping was performed via STAR ( Dobin et al . , 2013 ) version 2 . 5 . 3a ( https://github . com/alexdobin/STAR ) using the following parameters: -runThreadN 8 , --quantMode GeneCounts , --outSAMtype BAM SortedByCoordinate , --outFilterMatchNmin 30; ( 3 ) the results were output to one binary alignment map file for each sample matched to the reference; ( 4 ) genes with the coverage of 10 or more reads in at least three experiments were selected; and ( 5 ) data was normalized using the ‘DESeq2’ ( Love et al . , 2014 ) R package version 1 . 14 . 1 . Raw and normalized mapped , annotated reads are described in Supplementary file 4 . FASTAQ files are available through Gene Expression Omnibus at NCBI: GSE108270 . We generated an N-terminally acetylated , C-terminally amidated version of the ORF1p peptide ( MENDFDELRE ) as a di-peptide composed of repeats of the same sequence linked by a four-unit polyethylene glycol moiety; which was used to elute ORF1p-containing complexes from α-ORF1p medium at a concentration of approximately 2 mM ( Appendix 1; Supplementary file 4 ) . Peptides were synthesized by standard Fmoc solid-phase synthesis methods ( Kates and Albericio , 2000 ) ; the incorporation of a PEG spacer into the peptide sequence was accomplished using N-Fmoc-amido- ( PEG ) n-acid building blocks . 400 mg of cryogenically milled L1-expressing cells ( pLD401 and pLD561 ) were subjected to split-tandem affinity capture as described above , but with native elution from α-ORF1p medium and included the addition of RNasin ( Promega #N2515 ) at 1:500 v/v to the extraction buffer; 1x protease inhibitors and 1:200 v/v RNasin were also added to the 3xFLAG peptide and ORF1p-derived di-peptide solutions . For α-FLAG affinity capture , competitive elution was achieved using 60 μl of 1 mg/ml 3xFLAG peptide . Of this , 20 μl were held aside ( Input ) , 40 μl were carried forward to α-ORF1p affinity capture . The ORF1p-depleted fraction was retained ( Sup ) and the captured material was eluted with 40 μl ORF1p di-peptide ( Elu ) . Half of each fraction ( Input , Sup , Elu ) was set aside for protein analysis ( Supplementary file 4 ) and to the other half , glycerol was added to 25% v/v ( using a 50% v/v glycerol solution ) ; the latter were subsequently analyzed for enzymatic activity by LEAP . Raw data resulting from these assays is located in Supplementary file 4 . For LEAP , 2 μl from each of the above-described fractions were used in a 50 μl reaction , and 1 μl of each LEAP assay was used in SYBR Green qPCR ( carried out in triplicate ) as previously described ( Taylor et al . , 2013 ) . As controls , ( 1 ) an untagged L1RP construct was used in a ‘mock purification , ’ and ( 2 ) pLD401-derived ‘Input’ was heated at 100°C for 5 min and then added to the reaction mix , respectively . Neither produced detectable activity ( Supplementary file 4 ) . A second LEAP analysis was later carried out on an independently prepared set of fractions , prepared as above , stored frozen −80°C in 25% v/v glycerol . Tet-on HeLa M2 cells ( Hampf and Gossen , 2007 ) ( a gift from Gerald Schumann ) , were transfected and selected with 1 µg/ml puromycin for three days . Puromycin-resistant cells were plated on coverslips pre-coated for 1–2 hr with 10 µg/ml fibronectin in PBS ( Life Technologies ) . 8–16 hr after plating , L1 was induced with 1 µg/ml doxycycline . 24 hr later , cells were fixed in 3% paraformaldehyde for 10 min . Fixative was then quenched using PBS containing 10 mM glycine and 0 . 2% w/v sodium azide ( PBS/gly ) . The cells were permeabilized for 3 min in 0 . 5% Triton X-100 and washed twice with PBS/gly . Staining with primary and secondary antibodies was done for 20 min at room temperature by inverting coverslips onto Parafilm containing 45 ml drops of PBS/gly supplemented with 1% BSA , mouse α-FLAG M2 ( Sigma , 1:500 ) , rabbit α-ORF1 JH73 ( 1:4000 ) ( Taylor et al . , 2013 ) , Alexa Fluor 488 conjugated α-mouse IgG ( Life Technologies , 1:1000 ) , and Alexa Fluor 568 conjugated α-rabbit IgG ( Life Technologies , 1:1000 ) . DNA was stained prior to imaging with Hoechst 33285 ( Life Technologies , 0 . 1 µg/ml ) . Epifluorescent images were collected using an Axioscop microscope ( Zeiss , Jena , Germany ) equipped for epifluorescence using an ORCA-03G CCD camera ( Hamamatsu , Japan ) . For each microscope field , nuclei were identified and spatially located using a custom script in ImageJ , consisting of Otsu thresholding and watershed transformation of DAPI signal to segment each of the nuclei . ORF2p positive nuclei were differentiated from ORF2p negative nuclei by using another thresholding script for the ORF2p fluorescence channel and cross-registering the associated nuclei; all ORF2p positive nuclei were then hand-verified and then coordinates were converted into microns . The number of ORF2p+ nuclei per field , x , and a corresponding random distribution of x nuclei was calculated by randomly and repeatedly ( n = 1000 ) selecting x nuclei among all nuclei . The random distribution was used to calculate Bonferroni corrected p-values for the pairwise distances between ORF2p+ nuclei . The distribution of ORF2p+ inter nuclei distances was then compared to the distribution of random inter-nuclei distances using Welch’s t-test . The custom scripts used to select nuclei and calculate statistics , extracted data , calculated distances , p-values , and raw images are presented in the supplement ( Supplementary file 5; Figure 3—source data 1 ) . | Our genome consists of about two percent genes , while around 60 to 70 percent are made up of hundreds of thousands of copies of very similar DNA sequences . These repeats have accumulated over time due to specific genetic elements called transposons . Transposons are often referred to as ‘jumping genes’ , as they can move within the genome and thereby create mutations that may lead to cancer or other genetic diseases . LINE-1 is the only remaining active transposon in humans , and it expands by copying and pasting itself to new locations . To do so , it is first transcribed into RNA – the molecules that help to make proteins – and then converted back into identical DNA sequences . In a never-ending battle , our cells have been fighting to keep LINE-1 and its ancestors from replicating , and so evolved various defense mechanisms . Yet , LINE-1 has learned to circumvent these barriers , and continues to replicate and cause disease . Our understanding of these defenses and of how LINE-1 evades them is limited . Previous research has shown that the LINE-1 RNA and its two encoded proteins , called ORF1p and ORF2p , interact with a series of other proteins , with which they can form different types of complexes . Now , Taylor , Altukhov , Molloy et al . used human embryonic kidney cells grown in the laboratory with different LINE-1 mutations to identify how they affect the bound proteins and RNAs . The results showed that LINE-1 can form at least two different sets of complexes with other proteins . The complex containing ORF1p and ORF2p and several other proteins was located in the cytoplasm , the fluid that fills the cells . However , the experiments also revealed a new complex in the cell nucleus , which contained ORF2p and proteins involved in DNA replication and repair , but not ORF1p . The results suggest ORF1p delivers RNPs to the nucleus around the time the cell divides . Another group of researchers has looked more closely at what happens during cell division . A next step will be to study how exactly LINE-1 contributes to cancer . In the future , overactive LINE-1 proteins could be targeted to kill cancer cells , to identify cancer early , or to see if the cancer has come back . LINE-1 may also provide clues on how the genome has evolved . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"computational",
"and",
"systems",
"biology"
] | 2018 | Dissection of affinity captured LINE-1 macromolecular complexes |
Fibroblast growth factor-21 ( FGF21 ) is a hormone secreted by the liver during fasting that elicits diverse aspects of the adaptive starvation response . Among its effects , FGF21 induces hepatic fatty acid oxidation and ketogenesis , increases insulin sensitivity , blocks somatic growth and causes bone loss . Here we show that transgenic overexpression of FGF21 markedly extends lifespan in mice without reducing food intake or affecting markers of NAD+ metabolism or AMP kinase and mTOR signaling . Transcriptomic analysis suggests that FGF21 acts primarily by blunting the growth hormone/insulin-like growth factor-1 signaling pathway in liver . These findings raise the possibility that FGF21 can be used to extend lifespan in other species .
Caloric restriction without malnutrition is a proven means of inhibiting aging in species ranging from worms to nonhuman primates ( Masoro , 2005; Bishop and Guarente , 2007; Kenyon , 2010 ) . The effect of caloric restriction on longevity appears to be mediated by multiple nutrient-sensing pathways including those involving insulin and insulin-like growth factor ( IGF-1 ) , target of rapamycin ( TOR ) , AMP kinase and sirtuins . Pharmacologic and/or genetic manipulation of these pathways increases longevity to varying degrees , suggesting the feasibility of drugs that increase lifespan in the absence of caloric restriction ( Bishop and Guarente , 2007; Kenyon , 2010; Barzilai et al . , 2012 ) . Fibroblast growth factor-21 ( FGF21 ) is an atypical FGF that functions as an endocrine hormone ( Potthoff et al . , 2012 ) . In mice , FGF21 is strongly induced in liver in response to prolonged fasts through a peroxisome proliferator-activated receptor α-dependent mechanism . FGF21 in turn elicits diverse aspects of the adaptive starvation response . Among these , FGF21 increases insulin sensitivity and causes a corresponding decrease in basal insulin concentrations; FGF21 increases hepatic fatty acid oxidation , ketogenesis and gluconeogenesis; and , FGF21 sensitizes mice to torpor , a hibernation-like state of reduced body temperature and physical activity ( Potthoff et al . , 2012 ) . FGF21 also blocks somatic growth by causing GH resistance , a phenomenon associated with starvation . Transgenic ( Tg ) mice overexpressing FGF21 are markedly smaller than wild-type mice and have a corresponding decrease in circulating IGF-1 concentrations despite having elevated growth hormone ( GH ) levels ( Inagaki et al . , 2008 ) . Conversely , FGF21-knockout mice grow more than wild-type mice under conditions of nutrient deprivation ( Kubicky et al . , 2012 ) . In liver , FGF21 inhibits the GH signaling pathway by blocking JAK2-mediated phosphorylation and nuclear translocation of the transcription factor , STAT5 . This suppresses the transcription of Igf1 and other GH/STAT5-regulated genes ( Inagaki , et al . , 2008 ) . Thus , FGF21-mediated repression of the GH/IGF-1 axis provides a mechanism for blocking growth and conserving energy under starvation conditions . Dwarf mice , including the pituitary loss-of-function Ames and Snell strains and GH receptor/GH binding protein-knockout mice , are the longest living mouse mutants discovered to date , living up to ∼70% longer than their wild-type counterparts ( Liang et al . , 2003; Bartke and Brown-Borg , 2004; Brown-Borg and Bartke , 2012 ) . Interestingly , FGF21-Tg mice share a number of phenotypes with these long-lived mice including small size , enhanced insulin sensitivity and a blunted GH/IGF-1 signaling axis . In this report , using FGF21-Tg mice , we examine the consequences of chronic FGF21 exposure on lifespan .
We previously described FGF21-Tg mice in which the FGF21 transgene is selectively expressed in hepatocytes under the control of the apoE promoter ( Inagaki et al . , 2007 , 2008 ) . Circulating concentrations of FGF21 are ∼5–10-fold higher in the FGF21-Tg mice than under fasted conditions . Younger FGF21-Tg mice ( <8-month-old ) have significant decreases in serum insulin , IGF-1 , glucose , triglycerides and cholesterol and in hepatic triglyceride levels ( Inagaki et al . , 2007 , 2008 ) . Similar effects on insulin , glucose , triglycerides and cholesterol levels were seen in FGF21-Tg mice in which the FGF21 transgene was under the control of the albumin promoter ( Kharitonenkov et al . , 2005 ) . We examined whether these and additional metabolic parameters were altered in groups of older ( 26–27-month-old ) wild-type and FGF21-Tg mice . There were no differences in food intake , physical activity , oxygen consumption or respiratory exchange ratio ( Figure 1A–D ) . Although both male and female FGF21-Tg mice weighed less than their wild-type littermates , there were no differences in their percent fat and lean mass ( Figure 1E , F ) . Accordingly , plasma leptin levels were not significantly changed in FGF21-Tg mice ( Table 1 ) . Plasma adiponectin concentrations were significantly higher in male FGF21-Tg mice ( Table 1 ) , with increases in both the monomeric and more active oligomeric forms ( Figure 1G ) . There was no significant change in either the total amount or the different forms of adiponectin in females ( Table 1 , Figure 1G ) . Plasma ketone body levels were significantly higher in female but not male FGF21-Tg mice ( Table 1 ) . Hepatic triglyeride concentrations were lower in female but not male FGF21-Tg mice ( Table 1 ) . Plasma and hepatic cholesterol concentrations were unchanged in FGF21-Tg mice ( Table 1 ) . As expected ( Wei et al . , 2012 ) , FGF21-Tg mice had reduced bone mass ( Figure 1H ) . 10 . 7554/eLife . 00065 . 003Figure 1 . Metabolic parameters in aging wild-type and FGF21-transgenic mice . ( A ) Food intake , ( B ) physical activity , including total horizontal and ambulatory activity , ( C ) respiratory exchange ratio ( VCO2/VO2 ) and ( D ) oxygen consumption ( VO2 ) data for 30-month-old male wild-type ( WT ) and FGF21-transgenic ( Tg ) mice ( n=6/group ) housed singly in metabolic cages . For ( D ) , the area under the curve ( AUC ) data are shown in the right panel . ( E ) Body weights of male and female mice measured at 6 , 12 and 26 months . Measurements were done on all surviving mice in the cohorts ( n=27–54/group ) . ( F ) Fat and lean mass percentages were measured in 26-month-old mice ( n=5–6/group ) . ( G ) Adiponectin oligomer forms , including high molecular weight ( HMW ) , medium molecular weight ( MMW ) and low molecular weight ( LMW ) forms , were measured in plasma from 26- to 27-month-old male and female WT and FGF21-Tg mice . Representative western blots using plasma from single animals are shown together with the ratios of the different adiponectin forms in FGF21-Tg and WT plasma ( n=4/group ) . ( H ) Quantification of trabecular bone volume by μCT analysis using tibiae from 33- to 35-month-old male WT and FGF21-Tg mice ( n=4–5/group ) . All data are presented as the mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 00065 . 00310 . 7554/eLife . 00065 . 004Table 1 . Plasma and hepatic parametersDOI: http://dx . doi . org/10 . 7554/eLife . 00065 . 004MaleFemaleWTTgpWTTgpPlasma IGF-1* ( ng/mL ) 381 . 52±42 . 31250 . 1±13 . 760 . 03427 . 94±56 . 35171 . 84±11 . 710 . 009 Insulin* ( ng/mL ) 0 . 90±0 . 130 . 45±0 . 090 . 040 . 63±0 . 120 . 34±0 . 040 . 08 Glucose* ( mg/dL ) 155±9 . 21111 . 6±8 . 030 . 008181 . 75±19 . 77121 . 4±7 . 500 . 048 Ketone bodies* ( μM ) 118 . 38±30 . 51221 . 02±38 . 100 . 07140 . 61±17 . 46294 . 56±48 . 100 . 03 Adiponectin ( μg/mL ) 9 . 43±1 . 6523 . 42±2 . 650 . 00624 . 13±2 . 8231 . 71±7 . 970 . 42 Leptin ( ng/mL ) 3 . 20±1 . 024 . 14±1 . 090 . 554 . 80±1 . 827 . 46±2 . 890 . 47 Triglycerides ( mg/dL ) 75 . 14±9 . 2351 . 82±8 . 630 . 1150 . 28±10 . 3121 . 79±6 . 230 . 06 Cholesterol ( mg/dL ) 83 . 20±10 . 1881 . 45±2 . 320 . 8879 . 82±9 . 4470 . 56±13 . 580 . 60Liver Triglycerides ( mg/g ) 11 . 62±2 . 236 . 67±0 . 750 . 1122 . 42±3 . 7510 . 33±0 . 680 . 04 Cholesterol ( mg/g ) 4 . 29±0 . 504 . 07±0 . 200 . 714 . 52±0 . 224 . 54±0 . 130 . 94Measurements were made using 26–27-month-old mice ( n=4–5 mice/group ) . *4 hr fasting data . We next examined glucose homeostasis in older mice . Under 4 hr fasted conditions , plasma glucose concentrations were lower in both male and female FGF21-Tg mice , while plasma insulin concentrations were significantly lower in male FGF21-Tg mice ( Table 1 ) . Plasma IGF-1 concentrations were also lower in both male and female FGF21-Tg mice , with the decrease particularly striking in females ( Table 1 ) . In oral glucose tolerance tests done in mice fasted for 16 hr , there was no difference in glucose excursion between wild-type and FGF21-Tg mice , but the FGF21-Tg mice had significantly lower insulin levels in response to the glucose challenge ( Figure 2A ) . In insulin tolerance tests done in male mice fasted for 4 hr , plasma glucose concentrations decreased more in FGF21-Tg mice than in wild-type mice ( Figure 2B ) . 10 . 7554/eLife . 00065 . 005Figure 2 . FGF21-transgenic mice have increased insulin sensitivity . 26–27-month-old wild-type ( WT ) and FGF21-transgenic ( Tg ) male and female mice ( n=5/group ) were subjected to ( A ) oral glucose tolerance tests or ( B ) insulin tolerance tests . Plasma glucose and insulin levels were measured as indicated . Quantification of the area under the curve ( AUC ) is shown in the panels to the right . ( C , D ) Hyperinsulinemic-euglycemic clamp studies were performed on 6- to 8-month-old male WT and FGF21-Tg mice ( n=6/group ) . Insulin was infused at 2 mU/kg/min and ( C ) blood glucose ( upper panel ) was clamped at 120 mg/dL during the steady-state period ( t=110–150 min ) using a variable glucose infusion rate ( lower panel ) . The glucose infusion rate in FGF21-Tg mice is statistically different ( p<0 . 05 ) from WT mice at all points after t=0 min . ( D ) Rates of hepatic glucose production ( upper panel ) and whole-body glucose disposal ( lower panel ) in WT ( n=6 ) and FGF21-Tg ( n=3 ) mice during the basal and steady-state periods of the clamp . DOI: http://dx . doi . org/10 . 7554/eLife . 00065 . 005 These data suggested that the FGF21-Tg mice have increased insulin sensitivity . To address this directly , hyperinsulinemic-euglycemic clamp experiments were performed . The exogenous glucose infusion rate required to maintain euglycemia under clamp conditions was markedly higher in FGF21-Tg mice than in wild-type mice , demonstrating enhanced whole-body insulin sensitivity ( Figure 2C ) . Glucose tracer kinetic analysis revealed that insulin-stimulated suppression of hepatic glucose production and activation of whole-body glucose disposal were significantly greater in FGF21-Tg mice than in wild-type mice ( Figure 2D ) . Clamp insulin levels were similar between wild-type and FGF21-Tg mice ( 7 . 0±2 . 0 and 8 . 3±2 . 3 ng/mL , respectively; not significantly different ) . Together , these data demonstrate that glucose tolerance and whole-body insulin sensitivity are dramatically increased in FGF21-Tg mice . Given the effects of long-term FGF21 exposure on carbohydrate and lipid parameters , especially insulin and IGF-1 concentrations , we measured the lifespan of FGF21-Tg mice . Longevity was significantly extended in FGF21-Tg mice compared to wild-type littermates ( hazard ratio=0 . 22 [0 . 15 , 0 . 34] , p=2 . 7e−12 ) , with a 36% increase in the median survival time for FGF21-Tg mice ( Figure 3A ) . Although there was no difference in longevity between male and female wild-type mice , the difference in lifespan between male and female FGF21-Tg mice was statistically significant ( hazard ratio=2 . 42 [1 . 37 , 4 . 26] , p=0 . 0023 ) ( Figure 3B , C ) . Cox proportional-hazards regression analysis shows that FGF21 reduced the risk of death by 65% in males ( hazard ratio=0 . 35 [0 . 20 , 0 . 60] , p=0 . 00017 ) and 88% in females ( hazard ratio=0 . 12 [0 . 059 , 0 . 25] , p=1 . 8e−08 ) ( Figure 3B , C ) . Overall , there was a strong interaction between the presence of the FGF21 transgene and sex ( p=0 . 01 ) . Notably , at the time of this analysis >30% of the age-matched female FGF21-Tg mice were still alive at 44 months of age . 10 . 7554/eLife . 00065 . 006Figure 3 . FGF21 extends lifespan . ( A–C ) Kaplan–Meyer survival curves for wild-type ( WT ) and FGF21-transgenic ( Tg ) mice are shown . ( A ) Combined male and female data; ( B ) male data; ( C ) female data . ( D ) Median survival time ( at 50th percentile ) and maximum lifespan ( at 95th percentile ) for each cohort . Hazard ratios ( HR ) and 95% confidence intervals are shown for Tg vs WT mice . DOI: http://dx . doi . org/10 . 7554/eLife . 00065 . 006 Since FGF21 is induced by fasting and elicits diverse aspects of the adaptive starvation response , we examined whether chronic FGF21 exposure mimics nutrient deprivation with respect to changes in gene expression . Comprehensive transcriptome analysis was performed by microarray using RNA from liver , gastrocnemius muscle and epididymal white adipose tissue of wild-type and FGF21-Tg mice and mice subjected to either caloric restriction or a 24 hr fast . Using a false discovery rate <0 . 10 and fold change >2 as criteria , we found that expression of 33 , 8 and 22 genes was changed in liver , muscle and adipose of FGF21-Tg mice , respectively ( Figure 4A ) . Many more genes were regulated by caloric restriction or fasting than by the FGF21 transgene in all three tissues . As expected , Fgf21 was strongly induced in liver by fasting . Surprisingly , however , Fgf21 was not induced by the caloric restriction regimen ( Figure 4B ) . Likewise , there was no increase in plasma FGF21 concentrations in response to caloric restriction ( data not shown ) . While the molecular basis for this differential regulation of Fgf21 by fasting and caloric restriction is not yet known , these data indicate that FGF21 is not an endogenous mediator of the caloric restriction response . Notably , 30 of the 33 genes with changed expression in liver of FGF21-Tg mice were also regulated by caloric restriction , while 20 of these genes were regulated by fasting ( Figure 4B ) . Eight of the genes with altered expression in liver of FGF21-Tg mice ( highlighted in red in Figure 4B; see Discussion ) are similarly regulated in long-lived dwarf mice ( Swindell , 2007 ) . In contrast , there was little overlap in genes regulated by FGF21 and either caloric restriction or fasting in muscle or adipose . These data suggest that FGF21 may extend lifespan by regulating a small subset of genes also regulated by caloric restriction in liver . 10 . 7554/eLife . 00065 . 007Figure 4 . Genes regulated by FGF21 and caloric restriction overlap in liver . ( A ) Venn diagrams showing overlap of genes significantly regulated in liver , muscle and adipose tissue of FGF21-transgenic ( Tg ) vs wild-type ( WT ) mice ( FDR<0 . 10 , >twofold regulation ) compared to the same gene expression analysis in calorically restricted ( CR ) vs ad libitum ( AL ) or fasted vs AL mice . ( B ) Heat map of genes significantly regulated in liver of FGF21-Tg vs WT ( FDR<0 . 10 , >twofold regulation ) compared to expression of the same liver gene set regulated by fasting or CR . Microarray analysis was performed using liver , epididymal white adipose tissue and gastrocnemius muscle from wild-type and FGF21-Tg male mice and male C57BL/6J mice subjected to 60% caloric restriction for 2 weeks or a 24 hr fast . All mice used in these studies were 3 months old at the end of the study . DOI: http://dx . doi . org/10 . 7554/eLife . 00065 . 007 Increases in AMP kinase and sirtuin activity and decreases in mTOR activity are associated with increased longevity ( Bishop and Guarente , 2007; Kenyon , 2010 ) . To begin to assess whether these pathways are affected by FGF21 , we measured phosphorylated and total levels of AMP kinase and the mTOR targets S6 and 4E-BP1in liver , muscle and adipose tissue of male and female wild-type and FGF21-Tg mice . We also determined mitochondrial DNA content in liver as a downstream measure of AMP kinase activity . Phospho-AMP kinase levels were not increased in tissues from FGF21-Tg mice ( Figure 5A–C ) . Consistent with these data , mitochondrial DNA content was unchanged in liver ( Figure 5D ) . While Phospho-S6 and phospho-4E-BP1 levels were decreased in muscle of male FGF21-Tg mice ( Figure 5B ) , they were unchanged in muscle of longer-lived FGF21-Tg females or in liver or adipose from either sex ( Figure 5A–C ) . We also did not observe increases in NAD+ concentrations ( Figure 5E ) or the mRNA levels of Sirtuins 1–7 ( data not shown ) in liver of FGF21-Tg mice , suggesting that sirtuin activity is unlikely to be increased . Taken together , these data suggest that FGF21 may increase longevity through a mechanism independent of the AMP kinase , mTOR and sirtuin pathways . 10 . 7554/eLife . 00065 . 008Figure 5 . Evaluation of markers of AMP kinase , mTOR and sirtuin pathway activity in FGF21-Tg mice . Phosphorylated levels of AMP kinase , S6 , and 4E-BP1 in ( A ) liver , ( B ) gastrocnemius muscle , and ( C ) epididymal white adipose tissue; ( D ) mitochondrial DNA content and ( E ) NAD+ concentrations in liver of 26_28-month-old male and female wild-type ( WT ) and FGF21-transgenic ( Tg ) mice ( n=4/group except for female adipose tissue , where n=2/group; all data are presented as the mean ± SEM; *p<0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00065 . 008
In this report , we demonstrate that chronic exposure of mice to the starvation hormone , FGF21 , increases median survival time by ∼30% and ∼40% in males and females , respectively , without decreasing food intake . The increase in lifespan extension is comparable to that achieved by caloric restriction ( Turturro et al . , 1999 ) . While the FGF21-mediated increase in longevity is less than the 50–70% increase seen in pituitary loss-of-function Ames mice and GH receptor/GH binding protein-knockout mice ( Brown-Borg et al . , 1996; Coschigano et al . , 2000 ) , it is similar to that seen in other loss-of-function dwarf models including hypopituitary Snell mice ( Flurkey et al . , 2002 ) and mice lacking pregnancy-associated plasma protein-A ( PAPP-A ) , a protease that increases IGF-1 activity ( Conover et al . , 2010 ) . The FGF21-Tg mice share other phenotypic similarities with long-lived dwarf mice including small size , reduced circulating insulin and IGF-1 concentrations , increased circulating adiponectin levels and female infertility . Like the Ames and Snell dwarf mice , female FGF21-Tg mice live longer than males ( Liang et al . , 2003; Brown-Borg and Bartke , 2012 ) . These similarities together with our previous finding that FGF21-Tg mice are GH resistant ( Inagaki et al . , 2008 ) suggest that FGF21 may increase lifespan by inhibiting the GH/IGF-1 signaling pathway . Because FGF21 also regulates other pathways that impact metabolism , it is not surprising that some of the effects that are seen in dwarf mice , such as increased adiposity in GH receptor-knockout mice ( Masternak et al . , 2012 ) , are not recapitulated in FGF21-Tg mice . In microarray studies , FGF21 modulated more genes in liver than in either muscle or adipose . Interestingly , nearly all of these hepatic genes ( 30 of 33 ) were similarly regulated by caloric restriction , a proven method of extending lifespan in mice and other species ( Masoro , 2005; Bishop and Guarente , 2007; Kenyon , 2010 ) . In contrast , there was little or no overlap in the genes regulated by FGF21 and caloric restriction in muscle or adipose . A remarkable result of our studies is that FGF21 increased longevity to a similar extent as caloric restriction while regulating a much smaller set of genes in liver . These data suggest that FGF21 may extend lifespan by acting as a selective caloric restriction mimetic in liver . An important caveat , however , is that there may be changes in other tissues or post-transcriptional changes that were missed in this analysis . We note that caloric restriction did not increase circulating FGF21 levels , indicating that endogenous FGF21 does not mediate the longevity effects of caloric restriction . Using microarray analysis , Swindell ( 2007 ) previously defined a set of 43 candidate longevity genes based upon their similar regulation in different dwarf mouse strains or between dwarf strains and caloric restriction . Eight of these genes ( Fmo3 , Igfals , Hes6 , Alas2 , Cyp4a12b , Mup4 , Serpina12 , Hsd3b5 ) are among those co-regulated by FGF21 and caloric restriction in liver in our current study , and four others ( Igf1 , Igfbp1 , Socs2 and Mup3 ) were previously shown by more sensitive quantitative real-time PCR to be regulated in liver of FGF21-Tg mice ( Inagaki et al . , 2008 ) . Strikingly , 26 of the 33 genes regulated by FGF21 in liver were previously shown to be regulated in a similar manner in mice lacking STAT5 activity ( Barclay et al . , 2011 ) . Thus , FGF21 may increase longevity by suppressing the GH signaling axis in liver in a manner similar to that of caloric restriction . Interestingly , GH induces FGF21 in liver through a STAT5-dependent mechanism ( Barclay et al . , 2011; Chen et al . , 2011; Yu et al . , 2012 ) . This regulatory relationship between GH and FGF21 is likely to be important in controlling GH activity and coordinating energy homeostasis during starvation . The ability to increase lifespan adds to a growing list of other beneficial effects that pharmacologic administration of FGF21 has in mammals , including insulin sensitization , normalization of glycemia , and reduced body weight in obese animals ( Potthoff et al . , 2012 ) . These attributes have made FGF21 an intriguing therapeutic target . In contrast to these positive effects , however , we previously showed that younger ( <8-month-old ) FGF21-Tg mice and adult mice administered recombinant FGF21 for 14 days have reduced bone mass ( Wei et al . , 2012 ) . In our current study , similar reductions in bone mass were seen in older mice . The FGF21-dependent decrease in bone mass is caused in part by an increase in the differentiation of marrow adipocytes and corresponding decrease in osteoblast differentiation ( Wei et al . , 2012 ) . In addition , FGF21 inhibits GH action directly at the growth plate ( Wu et al . , 2012 ) , and long-lived PAPP-A-knockout mice have a decrease in bone mineral density ( Tanner et al . , 2008 ) , suggesting that changes in GH/IGF-1 signaling are also likely to be involved in FGF21-induced decreases in bone mass . While it remains to be determined whether FGF21 causes bone loss in humans , it is likely that this adverse effect will have to be overcome if FGF21 is to be used clinically to combat aging . In summary , we show that chronic exposure to FGF21 , a naturally-occurring hormone and potent insulin sensitizer , markedly extends lifespan in mice through a mechanism that may involve suppression of the GH/IGF-1 signaling axis in liver . Regarding other hormones that influence lifespan , we note that the extracellular domain of Klotho , a membrane protein that regulates phosphate metabolism and serves as a co-receptor for FGF23 but not FGF21 , can be cleaved to yield a circulating peptide that blocks insulin and IGF-1 signaling . This hormonal activity may contribute to Klotho’s lifespan-extending effect in mice ( Kurosu et al . , 2005 ) . However , unlike FGF21 , Klotho causes insulin resistance , which may limit its utility as a therapeutic agent . We conclude that FGF21 could potentially be used as a hormone therapy to extend lifespan in mammals .
FGF21-Tg mice and wild-type littermates ( Inagaki et al . , 2007 ) were maintained on C57Bl/6J background in a specific-pathogen-free facility with 12:12 light:dark cycle , and fed 2916 global diet ( Harlan ) . Littermates were housed in the same cage with a maximum number of five mice per cage . Cages were changed every week and nesting material was provided . Animals were sacrificed if they had wounds from fighting , developed severe dermatitis , tumors or other signs of morbidity . Caloric restriction experiments were performed with C57Bl/6J mice . Eight-week-old male mice were individually caged and fed ad libitum . Food intake was measured for 1 week . For the following 2 weeks , each mouse was fed at 4 p . m . every day with food equal to 40% of the daily consumption in the first week . Body weight and total body fat composition were measured 1 hr before feeding . The control group received food ad libitum . At the end of the study , mice from the caloric restricted group , ad libitum fed group , and a 24 hr fast group were sacrificed between 2 and 4 p . m . and blood and tissues were collected for analysis . All animal experiments were approved by the Institutional Animal Care and Research Advisory Committee at the University of Texas Southwestern Medical Center . Mice were individually housed in CLAMS system metabolic cages and were allowed to acclimate for 1 day . Oxygen consumption , CO2 production , activity and food intake data were collected for 3 days as described ( Kalaany et al . , 2005 ) . Body fat and lean mass were measured using an EchoMRI-100 ( Echo Medical Systems , LLC ) . Bone mass was determined by μCT using tibiae as described ( Wei et al . , 2012 ) . Blood was collected into EDTA-coated tubes ( Starstedt , Newton , NC ) . Plasma was separated by centrifugation and assayed using commercially-available kits for IGF-1 ( Alpco ) , total cholesterol ( Thermo Scientific ) , triglycerides ( Thermo Scientific ) , total ketone bodies ( Wako Chemicals ) , leptin and total adiponectin ( Millipore ) . The oligomeric forms of adiponectin were measured essentially as described ( Waki et al . , 2003 ) . Briefly , 0 . 5 μL of plasma from individual mice was subjected to SDS-PAGE under non-reducing and non-heat-denaturing conditions and western blot analysis performed using anti-adiponection mouse polyclonal antiserum ( a gift from Philipp Scherer ) . Glucose tolerance tests were performed on overnight fasted mice . Twenty percent D-glucose ( Sigma ) ( 1 g/kg body weight ) was administered by oral gavage . At 0 , 20 , 40 , 60 , and 120 min after administration , blood was collected by tail vein bleeding . Glucose levels were measured by enzymatic assay ( Wako Chemicals ) . Insulin levels were measured by ELISA ( Crystal Chem ) . Insulin tolerance tests were performed on mice after a 4 hr fast . 0 . 75 U/kg of human insulin ( Sigma ) was intraperitoneally injected into mice . At 0 , 20 , 40 , 60 , and 90 min after injection , tail vein blood was drawn and glucose levels measured using a One Touch Ultra glucometer ( Life Scan ) . Insulin sensitivity was evaluated in 6- to 8-month-old male WT and FGF21-Tg mice ( n=6/group ) by performing hyperinsulinemic-euglycemic clamp experiments as described ( Ayala et al . , 2006; Berglund et al . , 2009 ) . Briefly , 5 days prior to the clamp studies , infusion catheters were implanted in the right jugular vein under isofluorane anesthesia , tunneled subcutaneously , and exteriorized at the back of the neck . On the morning of the experiment , mice were transferred to sterile cages with bedding and water to begin a 4 hr fast . At t=−90 min , water was removed and a primed continuous infusion of HPLC-purified [3-3H]glucose ( 5 μCi bolus + 0 . 05 μCi/min ) was started for the assessment of glucose turnover . Blood samples were obtained from the cut tail at t=−15 and −5 min for the measurement of basal blood glucose ( AlphaTRAK glucometer , Abbott , Chicago , IL ) . At t=0 min , a primed continuous infusion of insulin ( 6 mU + 2 . 0mU/kg/min; Humulin ) was started to induce hyperinsulinemia , and the [3-3H]glucose infusion rate was doubled ( 0 . 1 μCi/min ) to minimize alterations in specific activity . Blood glucose was measured every 10 min thereafter , while 50% dextrose was infused at a variable rate to maintain target glycemic levels at 120 mg/dL . Glucose turnover was calculated during the basal ( t=−15 to −5 min ) and steady-state ( t=110–150 min ) periods of the clamp using Steele's steady-state equation ( Steele et al . , 1956 ) . Liver triglyceride and cholesterol were extracted from 0 . 1 g of liver tissue by the Folch method as described ( Kalaany et al . , 2005 ) . Total cholesterol or triglycerides were measured by colorimetric enzymatic assays ( Thermo Scientific ) . Liver and adipose tissue proteins were extracted as described ( Inagaki et al . , 2008; Dutchak et al . , 2012 ) . Gastrocnemius protein was extracted in 140 mM NaCl , 10 mM Tris-HCl ( pH 8 . 1 ) , 1 mM CaCl2 , 1 mM MgCl2 , 10% glycerol , 1% NP-40 with Complete protease inhibitor cocktail ( Roche diagnostics ) . 30 mg of whole liver lysate was resolved on a SDS-polyacrylamide gels and electrotransferred to a PVDF membrane ( Amersham ) . The membrane was then hybridized with antibodies against total AMPK , phospho-AMPK ( T172 ) , total S6 and phospho-S6 ( S240/244 ) ( Cell Signaling ) and total and phospho-4E-BP1 ( T70 ) ( Cell Signaling ) . Protein was detected by chemiluminescence ( ECL kit , Amersham ) . Results were quantified by densitometry using ImageJ Software ( NIH ) . Genomic DNA was isolated from liver using a DNeasy Blood and Tissue kit ( Qiagen ) . The mitochondrial gene COX-1 was measured by quantitative real-time PCR with β-globin used as a control as described ( Sahin et al . , 2011 ) . Frozen liver tissues were extracted and NAD+ levels were measured using HPLC as described ( Ramsey et al . , 2009 ) . Survival time was calculated from the date of birth until the last date of the observation . Mice that were still alive at the end of study or sacrificed during the study were censored at the last date of the observation . Survival curves were estimated using the Kaplan–Meier curves ( Kaplan and Meier , 1958 ) and were compared using the log-rank test . The univariate and multivariate survival analyses were performed using Cox proportional-hazards analysis ( Collett , 2003 ) . Median survival time was calculated as the shortest survival time for which the survivor function is ≤0 . 5 . Maximum lifespan was calculated as the shortest survival time for which the survivor function is ≤0 . 95 . RNA was prepared from liver , epididymal white adipose and gastrocnemius muscle from 3-month-old wild-type and FGF21-Tg mice . RNA was reverse transcribed into cRNA and biotin-UTP labeled using the Illumina TotalPrep RNA Amplification Kit ( Ambion ) and hybridized to the Illumina mouseRefseq-8v2 Expression BeadChips . Hybridizations were done in triplicate with each replicate containing cRNA from 1 to 3 different mice . Image data were converted into unnormalized Sample Probe Profiles using the Illumina BeadStudio software and raw expression data were processed using the Model-Based Background Correction method ( Xie et al . , 2009 ) . All gene expression values were log2 transformed and quantile normalized . Significance Analysis of Microarrays ( Tusher et al . , 2001 ) was used to identify differentially expressed genes among different groups using a false discovery rate of less than 10% and fold-change >2 as criteria . Results were analyzed by a Student’s unpaired t-test using GraphPad Prism ( GraphPad Software , Inc . ) . All data are presented as the mean ± SEM . The microarray data have been deposited in the NCBI Gene Expression Omnibus with accession number GSE39313 . | In 1934 , in a famous experiment at Cornell University , it was discovered that laboratory mice could live twice as long as expected if they were fed a low-calorie diet that included enough nutrients to avoid malnutrition . This phenomenon has since been observed in species ranging from worms to primates , but not in humans . Reducing calorie intake leads to longer lives by modifying a number of the biochemical pathways that sense nutrients , including pathways that involve insulin and various other biomolecules . Chemical and genetic methods can also increase longevity by modifying these pathways , which suggests that it might be possible to develop drugs that can increase lifespan without reducing calorie intake . Mice , humans and other creatures respond to prolonged fasting through a number of adaptive changes that include mobilizing and burning fatty acids . The liver has an important role in this response , secreting a hormone called fibroblast growth factor-21 ( FGF21 ) that coordinates these processes among tissues . Previous experiments on transgenic mice with high levels of this hormone have shown that it suppresses the activity of growth hormone and reduces the production of insulin-like growth factor , which prevents growth and can lead to hibernation-like behavior . Here Zhang et al . compare groups of wild-type mice and transgenic mice with high levels of FGF21 . They find that the transgenic mice have a longer median survival time than wild-type mice ( 38 months vs 28 months ) , and that the transgenic female mice on average live for 4 months longer than their male counterparts . However , unlike in other examples of increased longevity , they find that decreased food intake is not required . Instead , they find that transgenic mice eat more food than wild-type mice , yet remain profoundly insulin-sensitive . The results suggest that the longer survival times are caused by a reduction in the production of insulin-like growth factor , but they also suggest that the mechanism responsible for the increased longevity is independent of the three pathways that are usually associated with such increases . Further research is needed to understand this mechanism in greater detail and could , perhaps , pave the way for the use of FGF21-based hormone therapy to extend lifespan without the need for a low-calorie diet . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression"
] | 2012 | The starvation hormone, fibroblast growth factor-21, extends lifespan in mice |
The ability to decrypt volatile plant signals is essential if herbivorous insects are to optimize their choice of host plants for their offspring . Green leaf volatiles ( GLVs ) constitute a widespread group of defensive plant volatiles that convey a herbivory-specific message via their isomeric composition: feeding of the tobacco hornworm Manduca sexta converts ( Z ) -3- to ( E ) -2-GLVs thereby attracting predatory insects . Here we show that this isomer-coded message is monitored by ovipositing M . sexta females . We detected the isomeric shift in the host plant Datura wrightii and performed functional imaging in the primary olfactory center of M . sexta females with GLV structural isomers . We identified two isomer-specific regions responding to either ( Z ) -3- or ( E ) -2-hexenyl acetate . Field experiments demonstrated that ovipositing Manduca moths preferred ( Z ) -3-perfumed D . wrightii over ( E ) -2-perfumed plants . These results show that ( E ) -2-GLVs and/or specific ( Z ) -3/ ( E ) -2-ratios provide information regarding host plant attack by conspecifics that ovipositing hawkmoths use for host plant selection .
Insects rely on olfaction in most aspects of life: volatile signals guide them to food sources , mating partners and oviposition hosts . Especially for herbivorous insects , plant volatiles provide important cues to locate and identify appropriate host plants for their offspring . Upon herbivory , plants respond with an increased release and de novo synthesis of several volatile compounds from their vegetative tissues ( Mumm and Dicke , 2010 ) . These so-called herbivore induced plant volatiles can provide significant information to the surrounding environment as composition and abundance reflect several biotic and abiotic factors ( Takabayashi et al . , 1995; De Moraes et al . , 1998; Gouinguené et al . , 2001; Schuman et al . , 2009; Hare , 2010 ) . Due to the context dependent composition of plant volatile signals , the ability to detect and discriminate volatile compounds is crucial for insects to generate appropriate behavioral responses . In insects and more specifically in the hawkmoth Manduca sexta ( Lepidoptera/Sphingidae ) , olfactory sensory neurons ( OSNs ) located on the antennae detect odorant molecules ( Kalinová et al . , 2001; Shields and Hildebrand , 2001; Fraser et al . , 2003; Spaethe et al . , 2013 ) and convey this information to the antennal lobe ( AL ) , the first olfactory processing center . The AL of M . sexta females consists of about 70 structural and functional subunits called olfactory glomeruli ( Grosse-Wilde et al . , 2011 ) . OSNs expressing the same receptor , and thus responding to the same set of odorants , converge onto the same glomerulus in the AL ( Gao et al . , 2000; Vosshall , 2000 ) as has been demonstrated for Drosophila melanogaster and indirectly also in several moth species ( Hansson , 1997 ) . Spatio-temporal patterns of neuronal activity representing sensory input to the AL can be visualized by optical imaging methods ( Hansson et al . , 2003; Skiri et al . , 2004; Carlsson et al . , 2005; Silbering and Galizia , 2007 ) enabling identification of compound- and blend-specific responses in the AL of M . sexta ( Hansson et al . , 2003; Bisch-Knaden et al . , 2012; Kuebler et al . , 2012 ) . Green leaf volatiles ( GLVs ) constitute a large group of herbivore-induced plant volatiles characterized by a C6-backbone . While emitted only in trace amounts from healthy , undamaged plant tissue , they are emitted instantly after cell disruption ( Turlings et al . , 1995; D’Auria et al . , 2007 ) . GLVs are generated from C18-fatty acids via the enzymes lipoxygenase ( LOX ) and hydroperoxide lyase ( HPL; Allmann et al . , 2010 ) . One of the most abundant GLVs , ( Z ) -3-hexenal , originates from the cleavage of α-linolenic acid through the activity of HPL and it partly rearranges to ( E ) -2-hexenal . Both alkenals can be further metabolized by an alcohol dehydrogenase ( ADH ) and alcohol acyltransferase ( AAT; D’Auria et al . , 2007 ) to the corresponding alcohols and their esters ( Matsui , 2006 ) . GLVs have been assigned various plant defense-associated functions by directly inhibiting phytopathogens ( Hamilton-Kemp et al . , 1992; Nakamura and Hatanaka , 2002; Prost et al . , 2005 ) and repelling several herbivore species ( De Moraes et al . , 2001; Kessler and Baldwin , 2001; Vancanneyt et al . , 2001; Zhang and Schlyter , 2004 ) . Remarkably , GLVs also function as indirect plant defenses by attracting foraging predators and host-seeking parasitoids to the plant and its attacker ( Kessler and Baldwin , 2001; Shiojiri et al . , 2006; Halitschke et al . , 2008; Schuman et al . , 2012 ) reminiscent of the role of other herbivore induced plant volatiles . Due to their ubiquity and instant release , GLVs are thought to act as nonspecific signals of plant damage ( Hatanaka et al . , 1987; Hoballah et al . , 2002 ) . We recently showed that an enzymatic component of the oral secretions ( OS ) of M . sexta larvae adds an herbivory-specific feature to the GLV signal . Mechanically damaged leaves of Nicotiana attenuata released large amounts of ( Z ) -3-GLVs and low amounts of ( E ) -2-GLVs . However , when the plant was attacked by M . sexta caterpillars or when puncture wounds of plant leaves were treated with M . sexta’s OS , the amount of ( E ) -2-GLVs released increased , while the amount of ( Z ) -3-GLVs decreased , resulting in a distinct change in the ( Z ) -3/ ( E ) -2-ratio of GLV emissions . This herbivore-induced change in the ( Z ) -3/ ( E ) -2-ratio attracted the generalist hemipteran predator Geocoris spp . , which decreased the herbivore load on the plant by feeding on herbivore eggs ( Allmann and Baldwin , 2010 ) . Our discovery of a ( 3Z ) : ( 2E ) -enal isomerase in the OS of M . sexta larvae raises many questions . Why does Manduca produce an enzyme that generates volatiles which betray the insect to its enemies , and why did evolution not select against this isomerase ? The enzyme might be maladaptive and therefore is , or will be , under negative selection . The occurrence of this specific isomerase activity in at least two other lepidopteran species ( Allmann and Baldwin , 2010 ) however , suggests that it may have a beneficial function that outweighs the larva’s net costs of maintaining such an enzyme . It is well known that plants exchange information above ground by releasing volatiles into the air ( Baldwin , 2010 ) , which can be perceived by insects as well . Insects can use plant derived volatiles for communication by giving the herbivore induced volatile blend a ‘personal’ note—in our case , by converting ( Z ) -3-GLVs to their structural isomers and by changing the ( Z ) -3/ ( E ) -2 ratio . Which message could M . sexta larvae thereby communicate ? In this study we hypothesized that the altered GLV emission might serve to reduce the number of competitors on their host plant by informing conspecific ovipositing moths that this plant is already occupied and , possibly , receiving increased predation . Reduced oviposition of Manduca moths in response to feeding damage has been shown in field experiments with M . quinquemaculata ( Kessler and Baldwin , 2001 ) as well as under laboratory conditions with M . sexta ( Spaethe et al . , 2013 ) . Deviating from the previous study , we chose Datura wrightii ( Solanaceae ) for our experiments . Datura is a highly preferred host plant of both M . sexta and the congeneric M . quinquemaculata for both nectar feeding ( Alarcón et al . , 2008; Kessler , 2012 ) and oviposition ( Spaethe et al . , 2013 ) . Its distribution covers southwestern USA ( Avery , 1959; Munz , 1973 ) overlapping with the occurrence of both Manduca species . The perennial shrub is repeatedly described to quickly regrow leaves after herbivore damage ( Bronstein et al . , 2009; Reisenman et al . , 2010 , 2013 ) . Laboratory experiments failed to find reduced oviposition on damaged D . wrightii ( Reisenman et al . , 2013; Spaethe et al . , 2013 ) suggesting flexibility in oviposition choice of Manduca females . As the previously examined N . attenuata ( Gaquerel et al . , 2009 ) , D . wrightii , respond to Manduca herbivore attack by emitting GLVs ( Hare and Sun , 2011 ) . While we investigated GLV emission during the day when focusing on the diurnal egg predator Geocoris ssp . , Manduca moths oviposit at twilight and night ( Madden and Chamberlin , 1945; Lingren et al . , 1977 ) . Therefore , we decided to collect volatiles during these times instead . We expected the shift to occur also during the night , as in several plant species GLV emission has been shown to occur also in the dark period ( Loughrin et al . , 1994; Arimura et al . , 2008 ) , and the respective shift in the ( Z ) -3/ ( E ) -2-ratio is caused by M . sexta oral secretions and not by the plant itself ( Allmann and Baldwin , 2010 ) . However , volatile emissions vary with light regime ( Halitschke et al . , 2000; De Moraes et al . , 2001; Gouinguene and Turlings , 2002; Morker and Roberts , 2011 ) , and we therefore chose two nocturnal light conditions differing by moonlight intensity to examine whether light intensity affects GLV emission in D . wrightii . We performed functional imaging in the antennal lobe of female M . sexta moths asking whether ( Z ) -3- and ( E ) -2-structural isomers of any of the tested GLVs can be discriminated by the olfactory system . In classical host recognition experiments the Colorado potato beetle Leptinotarsa decemlineata has been shown to recognize and avoid altered ratios of ( Z ) -3- and ( E ) -2-GLVs emitted by its host plant Solanum tuberosum ( Visser and Avé , 1978 ) . Furthermore , enantioselectivity has been reported for projection neurons in the female AL of M . sexta in response to ( + ) - and ( − ) -linalool ( Reisenman et al . , 2004 ) . Thus , we hypothesized that M . sexta females would be able to differentiate between ( Z ) -3 and ( E ) -2-isomers of at least one GLV . If so , ovipositing M . sexta should avoid plants with increased levels of ( E ) -2-GLVs as they indicate host plants with increased larval feeding competition and predation risk ( Allmann and Baldwin , 2010 ) . Here we show by combining field studies with neurophysiological imaging techniques that ( i ) OS-induced D . wrightii plants have altered ( Z ) -3/ ( E ) -2-ratios also during the night under both laboratory and field conditions , ( ii ) Manduca females detect and discriminate the ( Z ) -3- and ( E ) -2-isomers and ( iii ) show ovipositional preference for high ( Z ) -3/ ( E ) -2-GLV ratios .
To investigate whether application of M . sexta’s OS onto wounded leaves of Datura wrightii plants causes a similar shift in the ( Z ) -3/ ( E ) -2-ratio as observed in Nicotiana attenuata , we compared the emissions of mechanically wounded D . wrightii plants that were treated with either water as a control ( w + w ) or with M . sexta’s OS ( w + OS ) in growth chamber experiments . During the day , application of OS onto wounds caused a significant decrease in the ( Z ) -3/ ( E ) -2-ratio of the GLVs released from Datura plants compared with control plants ( Figure 1A , day ) . 10 . 7554/eLife . 00421 . 003Figure 1 . Diurnal changes in the emitted ( Z ) -3/ ( E ) -2-ratios of GLVs in Datura wrightii plants . ( Z ) -3/ ( E ) -2-ratios of GLVs in Datura wrightii plants represented as box plots . ( A ) Growth chamber experiment: a single not yet fully developed leaf of each D . wrightii plant was mechanically wounded and treated with water ( w + w ) or M . sexta OS ( w + OS ) during three different light conditions to mimic day , sunset , and night . ( B ) Field experiment: Three single previously undamaged leaves per plant were chosen and randomly assigned to a treatment ( control , w + w or w + OS ) . Values of the control leaf were subtracted from the values of treated leaves . As ( Z ) -3-hexenal was not detectable in any of the field samples ( E ) -2-hexenal values are displayed in ng*cm−2*2h−1 ( adsorbents used in field collection are not accountable for the absence of ( Z ) -3-hexenal; Table 6 ) . For visual simplifications ( Z ) -3/ ( E ) -2-ratios <1 are represented as their negative reciprocal . Values above ‘1’ ( red dotted line ) thus represent treatment-groups that produced more of the ( Z ) -3-isomer and values below ‘1’ represent treatment-groups that produced more of the ( E ) -2-isomer . Asterisks indicate significant differences between treatments ( A: Mann–Whitney U test , **p≤0 . 01 , *p≤0 . 05; n = 5 ) , ( B: Wilcoxon signed-rank test , *p<0 . 05; n = 8 ) . ADH: alcohol dehydrogenase; AAT: alcohol acyl-transferase . The median is represented as a line in each box , box outlines mark the 25% and 75% percentiles; outliers are depicted as circles ( if value > 1 . 5× the interquartile range ) . For raw data , see F1AB_AllmannSpaethe2012_volatiles . xlsx ( Dryad: Allmann et al . , 2012 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00421 . 003 Since Manduca moths are crepuscular and nocturnal insects ( Theobald et al . , 2010 ) , we repeated the experiment under low light and no-light conditions to mimic sunset and night ( Figure 2 ) . The ( Z ) -3/ ( E ) -2-ratio of the aldehydes differed significantly between treatments also at sunset and night light intensities ( Figure 1A , sunset ) . However , the ( Z ) -3/ ( E ) -2-ratio of w + w treated plants also decreased with decreasing light intensities , which was mainly caused by increased ( E ) -2-hexenal emissions ( Figure 3 and Table 1 ) . Correspondingly , treatment-dependent differences in ( Z ) -3/ ( E ) -2-ratios for the alcohol and the hexenyl acetate decreased under lower light conditions and were not found during the night ( Figure 1A , sunset , night ) . 10 . 7554/eLife . 00421 . 004Figure 2 . Light conditions during laboratory volatile collection . Light composition and intensity changed within 24 hr to simulate day , sunset and night condition . Photosynthetically active radiation ( PAR , μmol photons*m−2*s−1 , orange line ) was measured for every light composition and ranged from 0 . 39 ± 0 . 01 SE at night to 138 . 37 ± 0 . 09 SE at full day conditions . Blue lines denote PAR values measured in the field during the respective volatile collection event ( during the night samplings , PAR was below detection limit ) . For the graph values were logarithmized . Grey areas denote volatile collection events; respective light spectra are shown on the right . For representational reasons time scale starts at 2 am . Flight activity , related to nectar feeding and oviposition ( Madden and Chamberlin , 1945; Lingren et al . , 1977 ) , is indicated on top of the graph . For raw data , see F2_AllmannSpaethe2012_light . xlsx ( Dryad: Allmann et al . , 2012 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00421 . 00410 . 7554/eLife . 00421 . 005Figure 3 . Total amounts of GLVs released from Datura wrightii plants at different times of the day in laboratory and field experiments . Mean release of major GLVs from Datura wrightii plants at different times of the day and at different light intensities . Grey and white bars represent ( Z ) -3- and ( E ) -2-GLVs , respectively . Single leaves were mechanically damaged and volatiles were trapped for 2 hr immediately after wounds had been treated with either water ( w + w ) or with M . sexta’s OS ( w + OS ) . ( A ) GLV emissions of D . wrightii plants under controlled light conditions in a growth chamber . Light conditions are explained in this figure . Quantities are given in nmol/g fresh mass ( FM ) /2 hr; n = 5 . ( B ) GLV emissions of D . wrightii plants naturally grown in the field . Quantities are given in pmol/cm2/2 hr; n = 8 . For an approximate comparison between ( A ) and ( B ) : 50 cm2 leaf area ≈ 1 g FM . Colored bars mark the emission of aldehydes ( light green ) , alcohols ( green ) and acetates ( dark green ) . For raw data , see F1AB_AllmannSpaethe2012_volatiles . xlsx ( Dryad: Allmann et al . , 2012 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00421 . 00510 . 7554/eLife . 00421 . 006Table 1 . GLV emission of Datura wrightii plants in the growth chamber during the first 2 hr after w + w or w + OS treatment with 100% light ( day ) , 20–10% light ( sunset ) or 0% light ( night ) DOI: http://dx . doi . org/10 . 7554/eLife . 00421 . 006ClassCommon nameRTvolatile release in µg / g leaf fresh massw + ww + OSDayAldehyde ( Z ) -3-hexenal8 . 540 . 64 ± 0 . 2930 . 097 ± 0 . 027 ( E ) -2-hexenal10 . 490 . 22 ± 0 . 1090 . 7 ± 0 . 17Alcohol ( Z ) -3-hexenol14 . 981 . 30 ± 0 . 5111 . 06 ± 0 . 275 ( E ) -2-hexenol15 . 570 . 058 ± 0 . 0340 . 195 ± 0 . 034Hexenylester ( Z ) -3-hexenyl acetate13 . 281 . 59 ± 0 . 4421 . 92 ± 0 . 244 ( E ) -2-hexenyl acetate13 . 750 . 017 ± 0 . 0040 . 105 ± 0 . 018 ( Z ) -3-hexenyl butyrate17 . 070 . 028 ± 0 . 0090 . 051 ± 0 . 016 ( E ) -2-hexenyl butyrate17 . 440 . 01 ± 0 . 0020 . 017 ± 0 . 004SunsetAldehyde ( Z ) -3-hexenal8 . 541 . 62 ± 0 . 50 . 26 ± 0 . 118 ( E ) -2-hexenal10 . 491 . 28 ± 0 . 7751 . 69 ± 0 . 697Alcohol ( Z ) -3-hexenol14 . 981 . 62 ± 0 . 4330 . 93 ± 0 . 308 ( E ) -2-hexenol15 . 570 . 45 ± 0 . 3150 . 44 ± 0 . 183Hexenylester ( Z ) -3-hexenyl acetate13 . 281 . 62 ± 0 . 4311 . 28 ± 0 . 511 ( E ) -2-hexenyl acetate13 . 750 . 18 ± 0 . 120 . 158 ± 0 . 067 ( Z ) -3-hexenyl butyrate17 . 070 . 039 ± 0 . 0110 . 031 ± 0 . 003 ( E ) -2-hexenyl butyrate17 . 440 . 013 ± 0 . 0040 . 01 ± 0 . 001NightAldehyde ( Z ) -3-hexenal8 . 541 . 71 ± 0 . 7320 . 28 ± 0 . 118 ( E ) -2-hexenal10 . 492 . 43 ± 0 . 5211 . 22 ± 0 . 697Alcohol ( Z ) -3-hexenol14 . 981 . 18 ± 0 . 350 . 81 ± 0 . 308 ( E ) -2-hexenol15 . 570 . 79 ± 0 . 140 . 37 ± 0 . 183Hexenylester ( Z ) -3-hexenyl acetate13 . 280 . 63 ± 0 . 2680 . 71 ± 0 . 511 ( E ) -2-hexenyl acetate13 . 750 . 093 ± 0 . 040 . 083 ± 0 . 067 ( Z ) -3-hexenyl butyrate17 . 070 . 036 ± 0 . 0020 . 033 ± 0 . 003 ( E ) -2-hexenyl butyrate17 . 440 . 01 ± 0 . 0020 . 014 ± 0 . 001Mean ( ±SEM; n = 5 ) release of GLVs in D . wrightii plants . A single not yet fully developed leaf of each plant was mechanically wounded and treated with water ( w + w ) or M . sexta OS ( w + OS ) during the day ( A , 100% light ) , sunset ( B , 20–10% light ) and night ( C , 0% light ) . Volatiles are listed by chemical classes and in order of their retention time . To evaluate whether w + w and w + OS treated plants release GLVs in distinguishable ratios under normally variable conditions found in nature , we trapped volatiles during daylight and repeatedly at night from a native D . wrightii population in the Utah desert during the 2011 field season . We performed the experiments on two different days using eight plants for each sampling . Three equally sized leaves of each plant were selected and randomly assigned to one of the treatments ( control , w + w or w + OS ) . Similar to previous experiments with N . attenuata ( Allmann and Baldwin , 2010 ) we were unable to detect ( Z ) -3-hexenal in any of the samples . During the day the application of OS to the wounds caused a significant increase in ( E ) -2-hexenal emissions compared with w+w treated leaves ( Figure 1B , day , and Table 2 , day ) . As seen from the climate chamber experiment , average ( Z ) -3/ ( E ) -2-ratio of the hexenyl acetates decreased ( Figure 1B , day ) , but this change was not significant . 10 . 7554/eLife . 00421 . 007Table 2 . GLV emission of native Datura wrightii plants in the field ( 2011 ) during the first 2 hr after w + w or w + OS treatment; during day ( 1:30–3:30 pm ) , first or second night ( 0–2 am ) DOI: http://dx . doi . org/10 . 7554/eLife . 00421 . 007ClassCommon nameRTVolatile release in ng/cm2 leafControlw + ww + OSDayAldehyde ( E ) -2-hexenal10 . 870 . 062 ± 0 . 0061 . 02 ± 0 . 2332 . 43 ± 0 . 597Alcohol ( Z ) -3-hexenol15 . 380 . 137 ± 0 . 0671 . 21 ± 0 . 282 . 07 ± 0 . 465 ( E ) -2-hexenol15 . 970 . 248 ± 0 . 0350 . 368 ± 0 . 0880 . 57 ± 0 . 148Hexenylester ( Z ) -3-hexenyl acetate13 . 660 . 26 ± 0 . 08311 . 1 ± 1 . 88112 . 3 ± 2 . 067 ( E ) -2-hexenyl acetate14 . 130 . 01 ± 0 . 0020 . 87 ± 0 . 3961 . 34 ± 0 . 564 ( Z ) -3-hexenyl butyrate17 . 440 . 011 ± 0 . 0020 . 22 ± 0 . 1810 . 19 ± 0 . 142 ( E ) -2-hexenyl butyrate17 . 80 . 004 ± 0 . 0010 . 007 ± 0 . 0020 . 006 ± 0 . 001First nightAldehyde ( E ) -2-hexenal10 . 870 . 103 ± 0 . 01324 . 6 ± 7 . 84422 . 5 ± 5 . 312Alcohol ( Z ) -3-hexenol15 . 380 . 032 ± 0 . 0069 . 6 ± 2 . 0288 . 2 ± 3 . 734 ( E ) -2-hexenol15 . 970 . 296 ± 0 . 0234 . 12 ± 0 . 9552 . 89 ± 0 . 855Hexenylester ( Z ) -3-hexenyl acetate13 . 660 . 165 ± 0 . 02810 . 7 ± 3 . 62111 . 53 ± 4 . 291 ( E ) -2-hexenyl acetate14 . 130 . 009 ± 0 . 0011 . 14 ± 0 . 3711 . 15 ± 0 . 306 ( Z ) -3-hexenyl butyrate17 . 440 . 007 ± 0 . 0010 . 022 ± 0 . 0080 . 04 ± 0 . 022 ( E ) -2-hexenyl butyrate17 . 80 . 002 ± 00 . 006 ± 0 . 0020 . 007 ± 0 . 003Second nightAldehyde ( E ) -2-hexenal10 . 870 . 055 ± 0 . 0094 . 7 ± 1 . 8779 . 5 ± 4 . 009Alcohol ( Z ) -3-hexenol15 . 380 . 034 ± 0 . 0184 . 0 ± 1 . 2252 . 94 ± 0 . 522 ( E ) -2-hexenol15 . 970 . 177 ± 0 . 0210 . 99 ± 0 . 4271 . 47 ± 0 . 554Hexenylester ( Z ) -3-hexenyl acetate13 . 660 . 089 ± 0 . 0244 . 8 ± 2 . 1149 . 4 ± 4 . 708 ( E ) -2-hexenyl acetate14 . 130 . 01 ± 0 . 0020 . 74 ± 0 . 5051 . 77 ± 0 . 972 ( Z ) -3-hexenyl butyrate17 . 44bld . 0 . 032 ± 0 . 0190 . 039 ± 0 . 013 ( E ) -2-hexenyl butyrate17 . 8bld . 0 . 005 ± 0 . 0030 . 007 ± 0 . 002Mean ( ±SEM; n = 5 ) release of GLVs in D . wrightii plants in nature . A single not yet fully developed leaf of each plant was mechanically wounded and treated with water ( w + w ) or M . sexta OS ( w + OS ) during the day ( A , 1:30–3:30 pm ) and during night ( B , first night , C , second night , 0–2 am ) . Volatiles are listed by chemical classes and in order of their retention time; bld . : below the limit of detection . During the first night-experiment ( first night , average temperature 17 . 6 ± 0 . 7°C , wind speed 1 . 1 ± 0 . 8 m/s , waxing crescent lunar illumination with 9% of the moon illuminated ) , plants of both treatments released very high but similar amounts of ( E ) -2-hexenal , and the ( Z ) -3/ ( E ) -2-ratios of the alcohols and hexenyl acetates were low , but did not differ between treatments , resembling the results of the night trapping in the growth chamber ( Figure 1B , first night , and Table 2 , first night ) . During the second experiment ( second night; average temperature 24 . 6 ± 0 . 8°C , wind speed 0 . 7 ± 0 . 8 m/s , full moon ) , approximately 2 weeks later , w + OS-treated plants released significantly higher amounts of ( E ) -2-hexenal ( twofold increase compared with w + w treated plants ) and the ( Z ) -3/ ( E ) -2-ratios of the hexenols and hexenyl acetates were significantly lower compared with mechanically wounded plants that were treated with water only ( Figure 1B , second night ) . To evaluate if female M . sexta moths are physiologically able to discriminate between ( Z ) -3- and ( E ) -2-GLVs and between different ( Z ) -3/ ( E ) -2-ratios we performed functional calcium imaging in the antennal lobes ( AL ) of females . Odor-evoked calcium changes in response to exposure to the pure ( E ) -2- and ( Z ) -3-isomers of hexenal , hexenol and hexenyl acetate led to activity in discrete regions corresponding to specific glomeruli in the AL of M . sexta females ( Figure 4A , B ) . Aldehyde and alcohol structural isomers activated one single specific region ( region of interest 2 [ROI 2] , green ) , with significantly stronger responses to the ( E ) -2- compared with ( Z ) -3-isomers ( Figure 4B ) . ( Z ) -3-hexenyl acetate and its structural isomer activated three different regions in the female AL: a significantly ( Z ) -3-specific ( ROI 3 , blue ) , a significantly ( E ) -2-specific ( ROI 4 , pink ) and an isomer-unspecific region ( ROI 1 , grey , Figure 4B ) . The differences in activation patterns caused by stimulations with ( Z ) -3- or ( E ) -2-hexenyl acetate ( Figure 4C ) strongly suggest that the two odors activated OSNs expressing different sets of odorant receptor types on the female antennae . Of all tested GLVs , hexenyl acetate was the only compound eliciting isomer-specific responses in the AL , therefore we focused on ( Z ) -3- and ( E ) -2-hexenyl acetate for all further experiments . 10 . 7554/eLife . 00421 . 008Figure 4 . Calcium activity patterns of the ( Z ) -3- and ( E ) -2-isomers in the M . sexta antennal lobe ( AL ) . ( A ) View onto the AL ( marked by outline ) of a Manduca sexta female after bath application with the calcium-sensitive dye calcium-green-AM . Stimulations with the six tested GLVs resulted in the activation of four regions in the AL most probably corresponding to single glomeruli ( four ROIs , regions of interest ) . ( B ) Representative false color-coded images show calcium responses in the AL after odor stimulation . Images are individually scaled to the strongest activation ( given by the max value in each image ) . Time traces show activity of ROI 1 , 2 , 3 and 4 ( n = 10 ) in response to odor stimulation ( 2 s; grey bar ) . Error bars represent standard errors of means . For hexenal and hexenol , stimulations with the ( E ) -2-isomers activated ROI 2 significantly stronger than did stimulations with the ( Z ) -3-isomers ( Wilcoxon signed-rank test: hexenal: p<0 . 01 , hexenol: p<0 . 05 ) . ΔF: change in fluorescence; F: background fluorescence . For raw data , see F4B_AllmannSpaethe2012_timetracesGlvs . xlsx ( Dryad: Allmann et al . , 2012 ) . ( C ) Comparison of response pattern similarity for repeated stimulations of one structural isomer ( ( Z ) -3 vs ( Z ) -3 Or ( E ) -2 vs ( E ) -2 , white boxes ) and for both structural isomers ( ( E ) -2 vs ( Z ) -3 , grey boxes ) ; sample size is given above the boxes ( Mann–Whitney U test: hexenal: p>0 . 05; hexenol: p>0 . 05 , hexenyl acetate: p<0 . 001 ) . For raw data , see F4C_AllmannSpaethe2012_correlationcoefficientsGlvs . xlsx ( Dryad: Allmann et al . , 2012 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00421 . 008 As plants do not emit isomerically pure odors but rather mixtures , we studied AL representation of the acetate structural isomers in more detail by stimulating the antenna with blends of ( Z ) -3- and ( E ) -2-hexenyl acetate in different ratios ( given as Z/E: 100/0 , 80/20 , 50/50 , 20/80 , 0/100 ) . In ROI 3 ( blue ) calcium signals evoked by ( Z ) -3-hexenyl acetate-containing mixtures were significantly higher compared with stimulations with pure ( E ) -2-hexenyl acetate , which in turn did not differ from the mineral oil control ( Figure 5A , B ) . For the ( E ) -2-specific ROI 4 ( pink ) stimulation with pure ( Z ) -3-hexenyl acetate led to significantly lower calcium responses when compared with pure ( E ) -2-hexenyl acetate and the 20/80 ratio , but was not different from stimulation with mineral oil ( Figure 5B ) . Calcium responses of the unspecific ROI 1 ( in grey ) did not differ between the structural isomers and their mixtures ( Figure 5B ) . 10 . 7554/eLife . 00421 . 009Figure 5 . Female antennal lobe ( AL ) shows isomer-specific calcium responses to ( Z ) -3- and ( E ) -2-hexenyl acetate . ( A ) Representative false color-coded images show calcium responses in the AL after odor stimulation with isomeric mixtures of a total dose of 250 ng . Images are individually scaled to the strongest activation ( given by the max value in each image ) . Time traces show activity of ROI 1 , 3 and 4 ( n = 10 ) in response to odor stimulation ( 2 s; grey bar ) . Error bars represent standard error of mean . For raw data , see F5A_AllmannSpaethe2012_timetraceshexenylacetate . xlsx ( Dryad: Allmann et al . , 2012 ) . ( B ) Change in fluorescence in ROI 1 , 3 and 4 to the pure structural isomers and their mixtures , normalized to the highest activation in every animal . Filled boxes represent responses significantly different from the mineral oil ( MO ) control; different letters denote significantly different calcium responses ( Kruskal–Wallis and Dunn’s multiple comparison test ) . For raw data , see F5BCE_AllmannSpaethe2012_imaginghexenylacetate . xlsx ( Dryad: Allmann et al . , 2012 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00421 . 009 When comparing odor-evoked activation by different ( Z ) -3/ ( E ) -2-ratios in ROI 3 and 4 , stimulations with pure structural isomers as well as the 20% ( Z ) -3/80% ( E ) -2 mixture led to significantly different levels of neural activity in these ( E ) -2/ ( Z ) -3-specific regions ( Figure 6A ) . Activation patterns differed significantly for pure ( E ) -2-hexenyl acetate compared with the 50/50 and 80/20 ( Z ) -3/ ( E ) -2 mixtures as well as for pure ( Z ) -3-hexenyl acetate compared with the 20/80 mixture ( Figure 6A ) . However , no differences were found between the isomeric mixtures ( 20/80; 50/50; 80/20 ) . 10 . 7554/eLife . 00421 . 010Figure 6 . Isomer-specific regions show different response characteristics . ( A ) Both isomer-specific regions ROI 3 and ROI 4 are shown as ratios of ROI activation ( ROI 3/ROI 4; for ROI 4 > ROI 3: −1/ratio ) at stimulations with 250 ng . Asterisks indicate significant differences from 1 , the ratio at which activation would be equal for ROI 3 and 4 ( Wilcoxon signed-rank test , 100/0 , 0/100: p<0 . 01 , 20/80: p<0 . 05 ) . Structural isomers and their mixtures were tested with Kruskal–Wallis and Dunn’s multiple comparison test , different letters denote significantly different ratios . ( B ) Calcium signals of ROI 3 ( x-axis ) and ROI 4 ( y-axis ) ( % norm ΔF/F , separated by axes ) in response to odor stimulation ( colored boxes ) and the solvent mineral oil ( grey box ) . Points denote the median values , box outlines mark the 25% and 75% percentiles . For raw data , see F5BCE_AllmannSpaethe2012_imaginghexenylacetate . xlsx ( Dryad: Allmann et al . , 2012 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00421 . 010 In addition to the isomer-specificity for both hexenyl acetates , ROI 3 and 4 displayed different response characteristics ( Figure 6B ) . The level of activation of the ( Z ) -3-hexenyl acetate-specific ROI 3 ( x-axis ) was solely dependent on the presence of the ( Z ) -3-isomer and did not change with various amounts of it in the isomeric mixtures ( ranging from 50 ng in 20/80 to 250 ng in 100/0 ) . In contrast , the calcium signal in ROI 4 ( y-axis ) increased gradually with increasing percentage of the ( E ) -2-isomer up to 80% in the isomeric mixtures . Thus , ROI 4 is able to convey information about the ratio of ( Z ) -3- and ( E ) -2-hexenyl acetate in a mixture . To test whether female Manduca moths use the herbivory-specific shift in ( Z ) -3/ ( E ) -2-ratio to choose appropriate host plants for their offspring , we performed oviposition assays in the field during the 2010 field season ( Figure 7A ) . We selected two native populations of D . wrightii plants located close to the Lytle Preserve research station ( Santa Clara , UT ) . On each experimental day we tested two mixes that contained either only ( Z ) -3 or only ( E ) -2-GLVs or both structural isomers but in different ratios . Since our calcium imaging data suggested that M . sexta possesses ( Z ) -3- and ( E ) -2-hexenyl acetate specific glomeruli ( and thereby OSNs ) we also tested these two compounds separately ( Table 3 gives composition of each GLV-mixture ) . Experiments were done in a paired design ( Figure 7A ) to minimize the volatile ‘noise’ caused by , for example , different numbers of flowers , different grades of damage or different plant ages . 10 . 7554/eLife . 00421 . 011Figure 7 . Manduca moths laid more eggs on the ( Z ) -3- than on the ( E ) -2-scented side of the plant . ( A ) The effect of different GLV-mixes on the oviposition behavior of female Manduca moths was tested during the 2010 field season on native Datura wrightii plants . On each experimental day , two different mixes were tested in a paired design . Mixes used on different experimental days are plotted above the timeline . The detailed composition of each mixture is described in Table 3 . ( B ) Difference in number of eggs oviposited per plant . Higher oviposition rates were observed for the ( Z ) -3-scented side of the D . wrightii plants . Treatment pairs with no oviposited eggs were excluded prior to the statistical analysis ( Wilcoxon signed-rank test ) . The median is represented as a line in each box , box outlines mark the 25% and 75% percentiles; outliers are depicted as circles ( if value > 1 . 5× the interquartile range ) . For raw data , see F6B_AllmannSpaethe2012_oviposition . xlsx ( Dryad: Allmann et al . , 2012 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00421 . 01110 . 7554/eLife . 00421 . 012Table 3 . GLV-mixtures used for oviposition assay in the fieldDOI: http://dx . doi . org/10 . 7554/eLife . 00421 . 012Compounds ( common names ) ( Z ) -3/ ( E ) -2-mix 1:1; ≈ w + OS ( µg/µl lanolin ) ( Z ) -3/ ( E ) -2-mix 9:1; ≈ w + w ( µg/µl lanolin ) ( Z ) -3-GLVs ( Z ) -3-hexenal ( 50% in triacetin ) 5 . 09 . 0 ( Z ) -3-hexenol5 . 09 . 0 ( Z ) -3-hexenyl acetate0 . 050 . 09 ( Z ) -3-hexenyl butyrate0 . 050 . 09 ( E ) -2-GLVs ( E ) -2-hexenal5 . 01 . 0 ( E ) -2-hexenol5 . 01 . 0 ( E ) -2-hexenyl acetate0 . 050 . 01 ( E ) -2-hexenyl butyrate0 . 050 . 01Triacetin per 10 mL mix ( derived from ( Z ) -3-hexenal ) , μl51 . 2592 . 2Triacetin added per 10 mL mix , μl40 . 950Total amount of triacetin per 10 mL mix , μl92 . 292 . 2Compounds ( common names ) Only ( Z ) -3-mix ( µg/µl lanolin ) Only ( E ) -2-mix ( µg/µl lanolin ) ( Z ) -3-GLVs ( Z ) -3-hexenal ( 50% in triacetin ) 10 . 00 . 0 ( Z ) -3-hexenol10 . 00 . 0 ( Z ) -3-hexenyl acetate0 . 10 . 0 ( Z ) -3-hexenyl butyrate0 . 10 . 0 ( E ) -2-GLVs ( E ) -2-hexenal0 . 010 . 0 ( E ) -2-hexenol0 . 010 . 0 ( E ) -2-hexenyl acetate0 . 00 . 1 ( E ) -2-hexenyl butyrate0 . 00 . 1Triacetin per 10 ml mix ( derived from ( Z ) -3-hexenal ) , μl102 . 50Triacetin added per 10 ml mix , μl0102 . 5Total amount of triacetin per 10 ml mix , μl102 . 5102 . 5Compounds ( common names ) ( Z ) -3-hexenyl acetate ( µg/µl lanolin ) ( E ) -2-hexenyl acetate ( µg/µl lanolin ) ( Z ) -3-hexenyl acetate5 . 00 . 0 ( E ) -2-hexenyl acetate0 . 05 . 0 When plants were augmented with isomerically pure mixtures that consisted of either ( Z ) -3 or ( E ) -2-GLVs ( aldehyde , alcohol , hexenyl acetate and hexenyl butyrate , Table 3 ) Manduca moths laid significantly more eggs on the ( Z ) -3 than on the ( E ) -2-scented side of the plant ( mean ± SEM number of eggs oviposited per plant side: ( Z ) -3-isomers 1 . 0 ± 0 . 2 , ( E ) -2-isomers 0 . 5 ± 0 . 1 , Figure 7B ) . When we compared two GLV mixes that contained all tested ( Z ) -3 and ( E ) -2-GLVs in a balanced isomeric ratio ( 1:1 ) or in a high ( Z ) -3 vs ( E ) -2-ratio ( 9:1 ) , significantly more eggs were oviposited on the sides of the plants that were scented with the higher ( 9:1 ) ( Z ) -3/ ( E ) -2-ratio ( 9:1-ratio 1 . 8 ± 0 . 2; 1:1-ratio 0 . 6 ± 0 . 2; Figure 7B ) . Finally , when we compared the two different hexenyl acetates , on average one additional egg per plant was oviposited on sides scented with the ( Z ) -3-isomer ( ( Z ) -3-hexenyl acetate 1 . 8 ± 0 . 3 , ( E ) -2-hexenyl acetate 0 . 9 ± 0 . 1; Figure 7B ) .
Here we demonstrate that the ( Z ) -3/ ( E ) -2-ratio of the GLV bouquet emitted from D . wrightii plants differs depending on the presence or absence of M . sexta larval oral secretions at sites of simulated feeding-damage . As OS-specific changes in the ( Z ) -3/ ( E ) -2-ratio were detectable during one of the two nights in field experiments , this volatile signal may be encountered by ovipositing Manduca females searching for host plants . Functional imaging experiments revealed that M . sexta females detect ( Z ) -3- and ( E ) -2-hexenyl acetate with distinct OSN populations leading to discrete activation patterns in the AL . In field experiments Manduca females laid fewer eggs on plants scented with GLV mixtures with increased proportions of ( E ) -2-GLVs . Our initial laboratory studies indicated that OS-induced changes in the GLV-profile of Datura wrightii plants are detectable during day and night , but they also revealed that light plays a role for the magnitude of this change in the signal . It has been shown that darkening can cause a temporary burst of GLVs in plants ( Graus et al . , 2004; Brilli et al . , 2011 ) . Furthermore , in Nicotiana attenuata the lipoxygenase NaLOX2 , which specifically provides oxygenated fatty acids for the GLV-pathway , has its highest transcript levels during the night ( Allmann and Baldwin , 2010 ) , and , while this might explain the overall increase in GLVs with decreasing light intensities , it does not explain the specific increase in ( E ) -2-GLVs in w + w treated plants ( Figure 3 and Table 1 , night ) . ( 3Z ) : ( 2E ) -enal isomerase activity has been found in crude extracts of some plants ( alfalfa and soybean; Takamura and Gardner , 1996; Noordermeer et al . , 1999 ) , but not in N . attenuata ( Allmann and Baldwin , 2010 ) , and it needs to be determined whether Datura plants possess such an enzyme with a nocturnal peak activity . Circadian rhythm is also known to affect volatile emission ( Goodspeed et al . , 2012 ) and might therefore be another factor involved in the variation in GLV emission found between the light and dark period . Most research on herbivore induced plant volatiles has been done in laboratory studies under controlled conditions ( Hunter , 2002; Kigathi et al . , 2009 ) . While these studies provide useful information about the influence of single stress factors , they often fail to include biotic and abiotic stresses that influence volatile production under natural conditions ( Kigathi et al . , 2009 ) . To evaluate the importance of these stresses , we repeated our trapping experiments in the field using native populations of D . wrightii plants . Night-GLV emissions were measured at two different dates; while ( Z ) -3/ ( E ) -2-ratios were the same in w + w and w + OS treated plants during the first experimental night , shortly after a new moon , they differed significantly during the full moon , the second experimental night . Although quantitative differences in light intensities between the two experimental nights were not detectable with the instruments available on site , they were obvious to the human eye . The releases of several volatile compounds are known to exhibit diurnal photoperiodicity in their quantitative but also qualitative emission patterns ( Loughrin et al . , 1994; Turlings et al . , 1995 ) . In cotton , acyclic terpenes like β-farnesene and β-ocimene were emitted in a diurnal fashion , while GLVs and few terpenes did not show such a clear diurnal pattern ( Loughrin et al . , 1994 ) . Diurnal-rhythm-dependent emission has also been observed in N . tabacum after feeding by larvae of Heliothis virescens , M . sexta and Helicoverpa zea , as these plants released larger amounts of ( E ) -2-hexenal during the night and emitted other GLVs exclusively in the dark period ( De Moraes et al . , 2001 ) . Experiments with lima beans revealed that leaves damaged during the scotophase responded with an almost immediate nocturnal emission of ( Z ) -3-hexenyl acetate , while the main emission of β-ocimene was delayed and peaked during the photophase ( Arimura et al . , 2008 ) . These studies affirm that light plays an important regulatory role in volatile emissions . Due to our sample size , it remains to be shown by further experiments whether moonlight is sufficient to regulate volatile emissions . The herbivore-induced volatile blend comprises several groups of compounds such as GLVs , terpenoids and/or aromatics , all of which have been shown to mediate plant-insect interactions ( Mumm and Dicke , 2010 ) . GLVs , which were investigated in the present study , seem to play an important role in volatile ‘communication’ as almost every green plant releases them upon various stress conditions . Furthermore , GLVs are released instantly from plant tissue upon damage , independent of the time of day ( Turlings et al . , 1995; D’Auria et al . , 2007 ) , while terpenoids are released with a delay ( Kant et al . , 2004 ) and several terpenoids not at all during the night as they are linked to photosynthesis ( Arimura et al . , 2008 ) . This makes GLVs an important cue for ovipositing Manduca moths as they are active during sunset and night ( Theobald et al . , 2010 ) and thus need to rely on signals that are released during the scotophase . The use of volatile blends for host location by insects depends heavily on the ability to detect and process olfactory signals . The insect’s olfactory system is highly sophisticated and enables detection of odors at very low concentrations ( Hansson et al . , 1999; Tanaka et al . , 2009 ) . However , in a redolent world , insects must distinguish host odors from a high background noise . Plant volatiles are detected by OSNs and these can be tuned to highly specific ( Bruce and Pickett , 2011 ) or to ubiquitous host plant compounds ( Hansson et al . , 1999; Bruce et al . , 2005 ) . We found that stimulations with hexenal- and hexenol-structural isomers led to activation of a distinct region in the AL ( ROI 2 , Figure 4A , B ) . However , calcium signals evoked by ( E ) -2-GLVs were significantly stronger compared with those evoked by ( Z ) -3-GLVs ( Figure 4B ) . This difference in activation intensity is likely a result from different binding affinities of the structural isomers to the olfactory receptor expressed by OSNs targeting the activated glomerulus ( Hallem et al . , 2004; Hooper et al . , 2009 ) . Of all tested compounds , only the ( Z ) -3- and ( E ) -2-isomers of hexenyl acetate activated two different discrete regions in the AL of M . sexta females ( Figures 4B , C and 5A , B ) strongly suggesting different isomer-specific OSN populations on the insect antenna . This leads to the proposition that for M . sexta females , changes in the volatile emission of ( Z ) -3- and ( E ) -2-GLVs might primarily be detected via hexenyl acetate . Given that all other tested GLVs activated only ROI 2 , the investment in isomer-specific receptors and consequently glomeruli to detect and process ubiquitous GLV compounds such as ( Z ) -3- and ( E ) -2-hexenyl acetate indicates the importance of the information content transferred by these compounds and their respective ratios . Specific responses to different types of green leaf volatiles have been reported both at physiological ( Hansson et al . , 1999; Røstelien et al . , 2005 ) and behavioral levels ( Reinecke et al . , 2005 ) . For hexenyl acetate , AL activation patterns elicited by stimulations with mixtures of both structural isomers were purely additive suggesting no mixture interaction at the OSN and AL input levels , which is consistent with other studies ( Deisig et al . , 2006; Carlsson et al . , 2007; Silbering and Galizia , 2007; Kuebler et al . , 2012 ) . When comparing the ratio of ROI activation , we did not find any difference between the mixtures ( Figure 6A ) . This result is not surprising when taking the different response characteristics of ROI 3 and 4 into account . Calcium activity of ROI 4 in response to mixtures with increasing percentages of the ( E ) -2-isomer were dose-dependent , whereas the activation of ROI 3 to the same mixtures resembled more an ‘on–off’ mechanism and was thus solely dependent on the presence of the ( Z ) -3-isomer , leading to a constant bias towards ( Z ) -3-hexenyl acetate ( Figures 5B and 6B ) . We can , however , not neglect the possibility that the logarithmic , dose-dependent phase in the neural dynamics of the neurons innervating ROI 3 lies at a concentration range below what was tested here . The different response characteristics of ROI 3 and 4 might mirror the relevance of the odors for M . sexta females . ( Z ) -3-hexenyl acetate is a rather ubiquitously occurring plant volatile , which is released in large amounts after damage irrespective of its origin ( Arimura et al . , 2008; Mumm and Dicke , 2010 ) . Electrophysiological experiments revealed that this compound elicited many responses in OSNs on the female M . sexta antenna: 60% of the tested sensilla ( Spaethe et al . , 2013 ) as well as 21 of 34 cells in the female AL ( Kuebler et al . , 2011 ) responded to this compound . ( E ) -2-hexenyl acetate , in contrast , has rarely been reported in insect-plant interactions ( Whitman and Eller , 1990; Quiroz and Niemeyer , 1998; Williams et al . , 2010 ) , aside from its release among other GLVs after larval feeding of M . sexta on N . attenuata ( Allmann and Baldwin , 2010 ) as an indication of actual larval damage . Thus , the presence of each structural isomer contains specific information but at different levels of resolution and in different contexts . In the case of ( Z ) -3-hexenyl acetate , the detected signal might also be relevant in long-range host location and host choice . Information about ( E ) -2-hexenyl acetate gained by ROI 4 in a dose-dependent fashion should , on the other hand , be most valuable at a short distance to the plant , possibly to locate the best spot for oviposition depending on the actual amounts emitted by different plant sites or to choose among neighboring plants with different levels of ( conspecific ) herbivory . Numerous studies suggest that the ratio of plant volatiles is an important component of the olfactory signal ( Visser and Avé , 1978; Bruce et al . , 2009; Cha et al . , 2011 ) . Visser and Avé ( 1978 ) found several GLVs playing important roles in host recognition of Leptinotarsa decemlineata . Augmenting the volatile emission of a potato host plant with the single GLV components ( Z ) -3- or ( E ) -2-hexenol , ( E ) -2-hexanal or 1-hexanol resulted in a disruption of the orientation of L . decemlineata to the potato plant . Further studies found neurons specifically responding to these GLVs both at the periphery and in the AL of L . decemlineata ( De Jong and Visser , 1988a , 1988b ) . In the case of the oriental fruit moth , Grapholita molesta , the ratio of a minor compound to the remaining components of a plant-derived synthetic mixture determined behavioral acceptance of this mixture , which could be associated with the response of two glomeruli in the female AL ( Piñero et al . , 2008; Najar-Rodriguez et al . , 2010 ) . We found that ovipositing Manduca moths distinguished between different ( Z ) -3/ ( E ) -2-ratios and that they used these volatile cues to choose oviposition sites associated with less feeding competition and predation . However , independent of whether the mixtures tested were rather complex in their composition ( 9:1 vs 1:1 ratios ) , or less complex ( only ( Z ) -3-GLVs vs only ( E ) -2-GLVs ) , or consisted of only a single compound ( ( Z ) -3-hexenyl acetate vs ( E ) -2-hexenyl acetate ) ovipositing Manduca moths continuously made a choice and always preferred the side of the plant that smelled more of ( Z ) -3-GLVs , or less of ( E ) -2-GLVs ( Figure 7B ) . From our results we cannot conclude whether a complex GLV bouquet of different ratios provides more reliable information than do single compounds , but our results demonstrate that by adding a single component to the volatile bouquet of native D . wrightii plants one can alter the choice of ovipositing Manduca moths . How did this behavior evolve ? It has been shown that M . sexta moths learn to feed from flowers of non-hosts due to olfactory conditioning ( Riffell et al . , 2008 , 2013 ) . Experience could as well shape female oviposition choice as it has been shown in other moth species ( Rietdorf and Steidle , 2002; Olsson et al . , 2006 ) . During oviposition M . sexta females might encounter the herbivory-specific signal , but can never experience the reward of oviposition success associated with it . A M . sexta larva feeding on plants , however , is continuously surrounded by GLVs emitted from wounded plants and more importantly encounters almost continuously a low ( Z ) -3/ ( E ) -2-ratio caused by its own oral secretions . The retention of odor memory learned at the larval stage onto the adult stage has been shown to occur in M . sexta ( Blackiston et al . , 2008 ) . However , in such a case you would rather expect a preference for OS-elicited bouquets , as the larva grew up in these . More experiments are needed to solve whether experience and learning are involved in the avoidance of herbivory-specific ( Z ) -3/ ( E ) -2-ratios . Almost every green plant releases volatiles in highly variable amounts and compositions . This makes it a challenge for host searching insects to simultaneously extract useful information while flying through the odor plumes from multiple sources . Our results show that the AL , the first odor processing center of the insect brain , has the capacity to resolve the composition of GLV blends as emitted by highly relevant host plants . Correspondingly , gravid females make an informed choice . They prefer oviposition sites with reduced predation and competition risks for their offspring , as indicated by the plant volatile bouquet . Future work will reveal whether increasing amounts of ( E ) -2-GLVs or rather changes in the ( Z ) -3/ ( E ) -2-ratio at the background of other host odors provide crucial information for female Manduca moths .
Datura wrightii seeds were initially purchased from B & T World Seeds ( Paguignan , France ) and subsequently harvested from plants propagated in the glasshouse . Plants were grown in 2 l pots in the glasshouse ( 23–25°C , 50–70% humidity , 16 hr light supplemented by Philips Sun-T Agro 400 W Na-vapor bulbs , 350–500 µmol/m2/s1 photosynthetic photon flux at plant level ) and used for experiments 35–0 days after sowing . For field experiments we used native populations of similar sized D . wrightii plants , which were located close to the Lytle Preserve research station . Wild plants at the field site and plants grown from purchased seeds showed high morphological similarity . For all treatments , plants were wounded with a fabric pattern wheel to punch three rows of holes on each side of the midrib . Wounded leaves were immediately treated with 20 μl of deionized water ( w + w ) or with 1:3 ( vol/vol ) diluted M . sexta oral secretions ( w + OS ) , which were pipetted directly onto the wounded leaf and gently dispersed across the surface . The OS was collected from third to fifth instar caterpillars which were fed on D . wrightii plants , and OS was stored under Argon at −20°C until usage . Volatile collections were performed in a growth chamber ( temperature 23–25°C , humidity 50–60% ) on shelves equipped with diode arrays of white ( approximately 420–690 nm ) , red ( 630–690 nm ) and UV ( 380–420 nm ) . Diode arrays were programmed to simulate daylight , twilight and night conditions accordingly regarding both light intensity and spectral composition ( 16:8 hr light/dark cycle , Figure 2 ) . D . wrightii plants were placed in the chamber two days prior the experiment to acclimatize . On the experimental day 1 single leaf per plant was enclosed immediately after treatment between two 50-mL food-quality plastic containers ( Huhtamaki , Bad Bertrich , Germany ) secured with miniature claw-style hair clips . Ambient air was pulled through the collection chamber and a glass tube ( ARS , Inc . , Gainsville , FL; www . ars-fla . com ) packed with glass wool and 20 mg of Super Q ( Alltech , Düsseldorf , Germany; www . alltech . com ) . Airflow was created by a vacuum pump ( model DAAV114-GB; Gast Mfg , Benton Harbour , MI; www . gastmfg . com ) as described by Halitschke et al . ( 2000 ) . For each time point and each treatment we trapped volatiles from five replicate plants . Directly after volatile sampling , we determined the fresh mass ( FM ) of each trapped leaf for further calculations . In the field , we selected eight plants of approximately similar size for each measurement in a 10-m radius . For each plant , we estimated the total leaf damage and we counted the number of flowers . To account for differences in volatile emissions caused by different degrees of leaf damage we selected three equal sized leaves of each plant and randomly assigned each leaf to one of the treatments ( control , w + w or w + OS ) . Each leaf was photographed to calculate the leaf area . We subsequently subtracted the amounts of volatiles emitted from untreated control leaves from those emitted from treated leaves of the same plant . We used a Li-COR Li-250A light meter with a Li-190SA quantum sensor ( http://www . licor . com ) to measure the photosynthetic active radiation during the different trapping periods . Weather data during the volatile collection were obtained from weather station KUTSTGEO6 located in St . George , UT ( www . wunderground . com ) . The first two trappings were performed on the 3 and 4 of June , soon after new moon . During the day volatiles were sampled at an average light intensity of 1450 µmol/s/m2 . During the night samplings , the light intensity was below the detection limit . The second trapping was performed in the night from the 14 to the 15 of June . Although it was a bright night ( full moon ) the average light intensity remained below the detection limit . On the experimental day , we enclosed single leaves directly after elicitation in polystyrene chambers fitted with holes at both ends . Air was pulled through the chamber and subsequently through a single-use charcoal trap ( Orbo M32; Sigma-Aldrich , Seelze , Germany ) as described in Kessler and Baldwin ( 2001 ) . Charcoal traps were equipped with MnO2-coated copper gauze as ozone scrubbers ( OBE Corporation , Fredericksburg , TX ) to prevent oxidation of volatiles . In all experiments , volatiles were trapped for 2 hr immediately after elicitation . Both charcoal and SuperQ traps were eluted with 250 μl dichloromethane ( DCM ) into a GC vial after spiking each SuperQ trap with 320 ng and each charcoal trap with 240 ng tetralin ( Sigma-Aldrich , Seelze , Germany ) as an internal standard . Samples were analyzed on an Agilent 7890A gas chromatograph ( Agilent Technologies , CA ) with the injection port kept at 230°C , operated in split-less mode and connected to an Agilent 5975C mass spectrometer . One microliter of each sample was injected on a polar column ( Innowax; 30 m , 0 . 25 mm ID , 0 . 25-µm film thickness; J&W Scientific , Folsom , CA ) operated under a constant He flow of 1 . 1 ml/min . The GC oven was programmed to hold 40°C for 5 min , to increase the temperature at 5°C/min to 130°C , then increasing temperature at 30°C/min to a maximum of 240°C . The maximum temperature was held for 15 min . The transfer line to the MS was kept at 260°C . The MS was operated in electron impact mode ( 70 eV ) with the ion source at 230°C and the quadrupole at 150°C . The detector monitored selected ions ( SIM ) : hexenals: m/z 55 , 69 , 83; hexenols: m/z 55 , 57 , 67 , 82; hexenyl acetates: m/z 67 , 71 , 82; tetralin: m/z 104 , 132 . Retentions times for each GLV were ascertained using standards of ( Z ) -3-hexenal , ( E ) -2-hexenal , ( Z ) -3-hexenol , ( E ) -2-hexenol , ( Z ) -3-hexenyl acetate , ( E ) -2-hexenyl acetate , ( Sigma-Aldrich , Seelze , Germany ) and quantifications were done after normalization to the peak of IS tetralin with calibration curves for each compound ( 33 , 10 , 5 , 1 , 0 . 5 , and 0 . 1 ng; n = 3 replicates ) using single ion traces ( hexenal m/z 83 , hexenol and hexenyl actetate m/z 82 ) . Emission rates were calculated based on fresh mass or surface area of the sampled leaves . ( Z ) -3/ ( E ) -2-ratios were calculated for each sample dividing the amount of the ( Z ) -3-GLV by the amount of its corresponding ( E ) -2-isomer . For visual simplifications ( Z ) -3/ ( E ) -2-ratios <1 were depicted as their negative reciprocal . M . sexta females were reared as described in Grosse-Wilde et al . ( 2011 ) . Pupae were kept separately in paper bags at 25°C and 70% relative humidity under a 16:8 hr light/dark cycle . Naïve adult females were used in functional imaging experiments 2–4 days post emergence . Moths were restrained in 15-ml Falcon tubes with the head exposed and fixed with dental wax ( Surgident; Heraeus Kulzer , Dormagen , Germany ) . The head capsule was opened and all tissues covering the antennal lobes were carefully removed . The brain was bathed with Calcium Green-2 AM ( 30 μmol; Invitrogen , Darmstadt , Germany , http://www . invitrogen . com ) containing physiological saline solution ( Christensen and Hildebrand , 1987 ) with 6% Pluronic F-127 , ( Invitrogen ) for 90 min at 4°C . After staining the brain was rinsed several times with Ringer’s solution to remove remaining dye . For imaging we used a Till Photonics imaging system ( Martinsried , Germany ) equipped with a CCD camera ( Sensicam; PCO Imaging ) connected to an upright microscope ( Olympus BX51WI ) . Monochromatic excitation light was given at 475 nm ( 500 nm SP; Xenon arc lamp , Polychrome V ) and fluorescence was detected with a LP515 emission filter and transmitted by a DCLP490 dicroic filter . The set-up was controlled by software Tillvision 4 . 0 ( Till Photonics ) . Images were taken with a water immersion objective ( Olympus , 10×/0 . 30 ) . Four-fold symmetrical binning resulted in image sizes of 344 × 260 pixels with one pixel corresponding to an area of 2 . 5 × 2 . 5 µm ( 10× magnification ) . Odors were chosen based on the results of a previous study ( Allmann and Baldwin , 2010 ) and on volatile collections of D . wrightii plants performed for this study ( Figure 1; ( Z ) -3- and ( E ) -2- hexenal , hexenol and hexenyl acetate [Sigma Aldrich , Seelze , Germany] ) . Odors were diluted in mineral oil and used in doses of 25 , 250 , and 2500 ng for the comparison of pure structural isomers . Hexenyl acetate was additionally tested as percentage mixtures of its ( Z ) -3- and ( E ) -2-isomers ranging from 0/100% , 20/80% , 50/50% , 80/20% to 100/0% ( vol/vol ) in doses of 250 and 1250 ng . 6 µl of the odorant mixtures were pipetted prior the experiment on a filter paper ( Whatman , http://www . whatman . com/ ) in glass pipettes using doses of 25 , 250 , 1250 ( isomeric mixtures ) and 2500 ng , respectively . The same volume of mineral oil served as a control stimulus . The stimulus loaden pipette and a second empty pipette were inserted in parallel into a glass tube , which delivered a constant flow of clean humidified air ( 1 l/min ) along one antenna . A continuous clean airstream ( 0 . 1 l/min ) was directed through the empty pipette and switched to the odor-containing pipette ( Syntech Stimulus Controller CS-55 ) during stimulation , thus preventing any change in total flow during the experiment . Every stimulation experiment lasted for 10 s , recording 2 s pre- and 6 s post-stimulus and 2 s of odor stimulation . Inter-stimulus time of at least 1 min was chosen to reduce adaptation effects . Every odor was presented first in the lower concentration . The sequence of the stimulations was changed from animal to animal . In some females ( hexenal n = 10 , hexenol n = 14 , hexenyl acetate n = 8 ) , the odors were repeatedly measured to test for the reproducibility of the evoked activity patterns within an animal ( Figure 4C ) . All stimulation experiments were recorded with 4 Hz resulting in a series of 40 consecutive frames , which were analyzed with custom written software ( IDL; ITT Visual Informations Solutions ) . Data were corrected for background fluorescence , bleaching of the dye and movement during the measurement ( Sachse and Galizia , 2002 , 2003 ) . A spatial median filter of 5 pixels was applied to reduce shot noise . Odor responses represented as change in fluorescence ( ΔF/F ) at spatially distinct activity spots were analyzed at the spot center in an area of the size of a small to medium-sized glomerulus ( 60 × 60 µm ) . Time traces of ΔF/F were smoothed by averaging three successive frames for each activity spot . The maximum ΔF/F value after stimulus onset was averaged with the pre- and postmaximum value . For every animal the odor responses were normalized to the maximal response and were taken into account if they reached ≥30% of the maximal value in this animal in at least one activity spot . Due to the lack of a glomerular map in the M . sexta AL observed activity regions for the tested odors could not be directly compared between animals . Thus , activation patterns for every isomeric pair of hexenal , hexenol and hexenyl acetate were used to calculate correlation coefficients providing a relative measurement of similarity ( Bisch-Knaden et al . , 2012 ) . Repeated stimulations with the same structural isomer and the correlation coefficients thereof were used as control . To compare activity patterns between the different isomeric mixtures of hexenyl acetate we calculated the difference in activity of both isomer-specific glomeruli resulting from the ratio of activity in the ( Z ) -3-specific and the ( E ) -2-specific glomerulus . For visual simplifications values below 1 ( representing cases in which the ( E ) -2-specific glomerulus was more active than the ( Z ) -3-specific glomerulus ) were displayed as their negative reciprocal and all values were presented on a scale without the range between ‘−1’ and ‘1’ ( Figure 6A ) . Experiments were done between 26 May and 1 July 2010 in southwestern Utah . This area is part of the native habitat of the tobacco and tomato hawkmoths Manduca sexta and M . quinqemaculata . Eggs of both species were counted for this experiment . We selected between 15 and 17 plants of two native populations of D . wrightii plants , which were located close to the Lytle Preserve research station . All plants were carefully inspected and oviposited Manduca eggs were removed prior the experiment . On each experimental day two mixes were tested in a paired design: every evening before sunset ( at around 5 pm ) cotton swabs were dipped into the GLV-scented lanolin pastes and stuck onto two opposing branches of one Datura plant ( Figure 7A ) . This paired design was chosen to minimize the effect that different numbers of flowers or different grades of leaf damage might have on the oviposition behavior of the moths . On the next day freshly laid Manduca eggs were counted in a defined area on the plant , approx 30 cm around the scented cotton swabs and afterwards removed . Treatment sides were switched every day . Plants with no oviposited eggs ( isomers N = 36 , ratios N = 15 , acetates N = 18 ) were excluded prior to the statistical analysis . The GLV-scented lanolin pastes were prepared by warming up lanolin and adding different GLV-mixtures to the liquefied lanolin paste shortly before it solidified again . The GLV mixes used are described in Table 3 . A comparison of emission rates and ( Z ) -3/ ( E ) -2-ratios emitted from cotton swabs and w + OS and w + w treated D . wrightii plants is shown in Tables 4 and 5 . 10 . 7554/eLife . 00421 . 013Table 4 . Average ( ±SD ) GLV emissions of GLV-mixtures used for the field bioassays ( cotton swab , after Allmann and Baldwin , 2010 ) and of native Datura wrightii plants in the field during the first 2 hr after w + w or w + OS treatment; second night ( 0–2 am ) DOI: http://dx . doi . org/10 . 7554/eLife . 00421 . 013Common namesVolatile release in µg/30 minCotton swab ( after Allmann and Baldwin , 2010 ) D . wrightii leaf9:1 GLV mix1:1 GLV mixw+ww+OS ( Z ) -3-hexenol9 . 8 ± 13 . 217 . 1 ± 9 . 710 . 20 ± 0 . 170 . 15 ± 0 . 07 ( Z ) -3-hexenyl acetate0 . 18 ± 0 . 210 . 15 ± 0 . 180 . 24 ± 0 . 300 . 47 ± 0 . 67 ( E ) -2-hexenal1 . 3 ± 2 . 144 . 3 ± 7 . 200 . 24 ± 0 . 270 . 48 ± 0 . 57 ( E ) -2hexenol1 . 3 ± 1 . 848 ± 11 . 390 . 05 ± 0 . 060 . 07 ± 0 . 08 ( E ) -2-hexenyl acetate0 . 06 ± 0 . 070 . 16 ± 0 . 200 . 04 ± 0 . 070 . 09 ± 0 . 1410 . 7554/eLife . 00421 . 014Table 5 . Average ( Z ) -3/ ( E ) -2-ratios of GLV-mixtures used for the field bioassays ( cotton swab , after Allmann and Baldwin , 2010 ) and of native Datura wrightii plants in the field during the first 2 hr after w+w or w+OS treatment; 2nd night ( 0–2am ) DOI: http://dx . doi . org/10 . 7554/eLife . 00421 . 014Common names ( Z ) -3/ ( E ) -2-ratio of emitted GLVsCotton swab ( after Allmann and Baldwin , 2010 ) D . wrightii leaf9:1 GLV mix1:1 GLV mixw + ww + OSHexenol8 . 371 . 074 . 442 . 38Hexenyl acetate3 . 240 . 9025 . 9415 . 67Emissions of D . wrightii were adjusted from leaf surface ( cm2 ) to fresh mass ( g ) scale by the rough estimate of 50 cm2 = 1 g and represent the emission of two medium sized leaves . 10 . 7554/eLife . 00421 . 015Table 6 . Comparison of trapping capability of adsorbents used in laboratory ( SuperQ ) and field volatile collectionsDOI: http://dx . doi . org/10 . 7554/eLife . 00421 . 015Compound namePeak area × 106 ± SDRTActivated charcoalSuperQ ( Z ) -3-hexenal2 . 400 . 90 ± 0 . 135 . 4 ± 1 . 01 ( E ) -2-hexenal3 . 579 . 3 ± 3 . 336 . 9 ± 1 . 21 ( Z ) -3-hexenol3 . 3510 . 1 ± 3 . 077 . 6 ± 1 . 49 ( E ) -2-hexenoln . d . n . d . ( Z ) -3-hexenyl acetate12 . 250 . 56 ± 0 . 280 . 47 ± 0 . 22 ( E ) -2-hexenyl acetate12 . 680 . 52 ± 0 . 030 . 59 ± 0 . 04Mean ( ±SD; n = 6 ) peak areas of GLVs emitted from N . attenuata plants in the glass house . Of each plant two equally sized leaves were mechanically wounded . Subsequently , volatiles were collected for 1 hr with traps filled with either SuperQ or activated charcoal . Traps were eluted with 250 μl Dichloromethane and measured on a GC-MS equipped with a BR-5ms column ( Bruker , 15 m , 0 . 25 mm ID , 25 μm ) . | Plants have developed a variety of strategies to defend themselves against herbivorous animals , particularly insects . In addition to mechanical defences such as thorns and spines , plants also produce compounds known as secondary metabolites that keep insects and other herbivores at bay by acting as repellents or toxins . Some of these metabolites are produced on a continuous basis by plants , whereas others—notably compounds called green-leaf volatiles—are only produced once the plant has been attacked . Green-leaf volatiles—which are also responsible for the smell of freshly cut grass—have been observed to provide plants with both direct protection , by inhibiting or repelling herbivores , and indirect protection , by attracting predators of the herbivores themselves . The hawkmoth Manduca sexta lays its eggs on various plants , including tobacco plants and sacred Datura plants . Once the eggs have hatched into caterpillars , they start eating the leaves of their host plant , and if present in large numbers , these caterpillars can quickly defoliate and destroy it . In an effort to defend itself , the host plant releases green-leaf volatiles to attract various species of Geocoris , and these bugs eat the eggs . One of the green-leaf volatiles released by tobacco plants is known as ( Z ) -3-hexenal , but enzymes released by M . sexta caterpillars change some of these molecules into ( E ) -2-hexenal , which has the same chemical formula but a different structure . The resulting changes in the ‘volatile profile’ alerts Geocoris bugs to the presence of M . sexta eggs and caterpillars on the plant . Now Allmann et al . show that adult female M . sexta moths can also detect similar changes in the volatile profile emitted by sacred Datura plants that have been damaged by M . sexta caterpillars . This alerts the moths to the fact that Geocoris bugs are likely to be attacking eggs and caterpillars on the plant , or on their way to the plant , so they lay their eggs on other plants . This reduces competition for resources and also reduces the risk of newly laid eggs being eaten by predators . Allmann et al . also identified the neural mechanism that allows moths to detect changes in the volatile profile of plants—the E- and Z- odours lead to different activation patterns in the moth brain . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Material",
"and",
"methods"
] | [
"ecology",
"neuroscience"
] | 2013 | Feeding-induced rearrangement of green leaf volatiles reduces moth oviposition |
In eukaryotes , most integral membrane proteins are synthesized , integrated into the membrane , and folded properly in the endoplasmic reticulum ( ER ) . We screened the mutants affecting rhabdomeric expression of rhodopsin 1 ( Rh1 ) in the Drosophila photoreceptors and found that dPob/EMC3 , EMC1 , and EMC8/9 , Drosophila homologs of subunits of ER membrane protein complex ( EMC ) , are essential for stabilization of immature Rh1 in an earlier step than that at which another Rh1-specific chaperone ( NinaA ) acts . dPob/EMC3 localizes to the ER and associates with EMC1 and calnexin . Moreover , EMC is required for the stable expression of other multi-pass transmembrane proteins such as minor rhodopsins Rh3 and Rh4 , transient receptor potential , and Na+K+-ATPase , but not for a secreted protein or type I single-pass transmembrane proteins . Furthermore , we found that dPob/EMC3 deficiency induces rhabdomere degeneration in a light-independent manner . These results collectively indicate that EMC is a key factor in the biogenesis of multi-pass transmembrane proteins , including Rh1 , and its loss causes retinal degeneration .
In eukaryotes , most integral membrane proteins are synthesized , integrated into the membrane , and folded properly in the endoplasmic reticulum ( ER ) . Molecular chaperones and folding enzymes are required for the folding of the integral membrane proteins in the ER . A comprehensive approach in yeast to identify genes required for protein folding in the ER identified the ER membrane protein complex ( EMC ) , which comprises six subunits ( Jonikas et al . , 2009 ) . Another report studying the comprehensive interaction map of ER-associated degradation ( ERAD ) machinery revealed that EMC contains four and three additional subunits in mammals and Drosophila , respectively ( Christianson et al . , 2011 ) . The deletions of each emc1–6 gene causes the unfolded protein response ( UPR ) , presumably caused by the accumulation of misfolded proteins ( Jonikas et al . , 2009 ) . Meanwhile , a recent study showed that EMC also facilitates lipid transfer from ER to mitochondria ( Lahiri et al . , 2014 ) . In photoreceptors , the massive biosynthesis of rhodopsin demands chaperones in the ER . In the vertebrate retina , rhodopsin interacts with the ER degradation enhancing α-mannosidase-like 1 ( EDEM1 ) protein and a DnaJ/Hsp40 chaperone ( HSJ1B ) ( Chapple and Cheetham , 2003; Kosmaoglou et al . , 2009 ) . Meanwhile , in Drosophila photoreceptors , rhodopsin 1 ( Rh1 ) sequentially interacts with chaperones calnexin99A ( Cnx ) , NinaA , and Xport before exiting from the ER ( Colley et al . , 1991; Rosenbaum et al . , 2006 , 2011 ) . Defects in rhodopsin biosynthesis and trafficking cause retinal degeneration in both Drosophila and humans; more than 120 mutations in the rhodopsin gene are associated with human retinitis pigmentosa ( Mendes et al . , 2005; Xiong and Bellen , 2013 ) . The overwhelming majority of these mutations lead to misfolded rhodopsin , which aggregates in the secretory pathway ( Hartong et al . , 2006 ) . Thus , it is important to understand the mechanisms underlying the folding and trafficking of rhodopsin as well as retinal degeneration caused by misfolded rhodopsin . In zebrafish the partial optokinetic response b ( pob ) a1 mutant exhibits red cone photoreceptor degeneration ( Brockerhoff et al . , 1997; Taylor et al . , 2005 ) . The localization of overexpressed zebrafish Pob protein in cultured cells in the early secretory pathway including the ER and Golgi body indicates that Pob is involved in red cone rhodopsin transport ( Taylor et al . , 2005 ) . The zebrafish pob gene is the homolog of a subunit of EMC , EMC3 . Here we report the function of dPob , Drosophila pob homolog , on Rh1 maturation , photoreceptor maintenance , and expression of other multi-pass membrane proteins .
Retinal mosaic screening using the FLP/FRT method and two-color fluorescent live imaging was used to identify the genes essential for Rh1 maturation and transport ( Satoh et al . , 2013 ) . For selected lines exhibiting defects in Rh1 accumulation in the live imaging screening , the immunocytochemical distribution of Rh1 was investigated to evaluate the phenotype with respect to transport and morphogenesis ( Table 2 , Satoh et al . , 2013 ) . Among them , CG6750e02662 ( Kyoto stock number: 114504 ) exhibits severe Arrestin2::GFP and Rh1 reduction in rhabdomeres ( Figure 1A , C ) with normal ommatidial organization . CG6750e02662 has an insertion of a piggyBac transposon right downstream of the stop codon of CG6750 ( Figure 1B ) . The phenotype was reverted by the precise excision of the piggyBac transposon or transgenically-expressed CG6750 ( data not shown ) ; this indicates Rh1 reduction is caused by reduced CG6750 gene function . CG6750 shares 65% identity and 82% similarity with zebrafish pob and 27% identity and 44% similarity with yeast EMC3 . Because CG6750 is likely to be the homolog of zebrafish pob , we designated CG6750 as ‘dPob’ and analyzed its functions in Rh1 transport and retinal morphogenesis . 10 . 7554/eLife . 06306 . 003Figure 1 . Identification of CG6750 as an essential gene for rhodopsin 1 ( Rh1 ) biosynthesis . ( A ) Observation of fluorescent protein localizations in CG6750e02662 mosaic retinas by the water immersion technique . RFP ( red ) indicates wild-type photoreceptors ( R1–R8 ) . Arrestin2::GFP ( green ) shows endogenous Rh1 localization in R1–R6 peripheral photoreceptors . ( B ) Schematic drawing of CG6750 and insertion/deletion mutants . The dPob-null mutant allele , dPob∆4 , was created by the recombination of two FRTs on dPobf07762 and dPobCB−0279−3 using an FRT/FLP-based deletion method . ( C , D ) Immunostaining of dPobe02662 ( C ) and dPob∆4 ( D ) retinas expressing RFP as a wild-type cell marker ( magenta ) by anti-Rh1 antibody ( green ) . Asterisks show mutant cells . Scale bar: 5 μm ( A , C , D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06306 . 003 To address the possibility that the severe reduction of Rh1 protein in dPobe02662 mutant is caused by the reduction of mRNA , Rh1 mRNA was quantified in whole-eye clones of the mutant . When compared with control FRT40A whole-eye clone , relative mRNA levels normalized to Act5C were , Rh1: 0 . 51 ( n = 4 , S . D . = 0 . 24 ) ; trp: 0 . 31 ( n = 4 , S . D . = 0 . 17 ) ; and Arr2: 0 . 49 ( n = 4 , S . D . = 0 . 24 ) . Thus , the great reduction of the Rh1 protein level in dPobe02662 clones could not be interpreted by the reduction of mRNA . As expected from the position of the insertion , dPob was still weakly expressed in dPobe02662 homozygous photoreceptors ( Figure 2B , C ) , so it was classified as a hypomorphic allele . To further investigate the function of dPob , dPob∆4 , a null mutant allele lacking the entire coding sequence of dPob , was created using an FRT/FLP-based deletion method ( Figure 1B ) ( Parks et al . , 2004 ) . Unlike dPobe02662 , which gives escapers up to the late pupal stage , dPob∆4 flies were lethal in the first instar larval stage . Immunostaining of dPob∆4 mosaic retinas shows a great reduction of Rh1 in dPob∆4 homozygous photoreceptors , similar to dPobe02662 homozygous photoreceptors ( Figure 1D ) . 10 . 7554/eLife . 06306 . 004Figure 2 . Construction of antisera against dPob . ( A ) Immunoblotting of wild-type ( +/+ ) and dPobe02662 homozygous ( −/− ) extracts from whole larvae using antiserum against dPob N- and C-terminal polypeptides . ( B ) Immunostaining of a dPobe02662 mosaic retina expressing RFP ( red ) as a wild-type cell marker ( not shown ) by rat anti-dPob-C1 antiserum ( blue ) and phalloidin ( green ) . Asterisks show dPobe02662 homozygous photoreceptors . ( C , D ) Immunostaining of wild-type retinas by anti-dPob ( green ) and anti-NinaA ( C ) or anti-HDEL ( D ) antisera . Scale bar: 5 μm ( B–D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06306 . 004 Next , antisera against dPob ( Figure 2 ) were created to investigate dPob localization in fly photoreceptors . Four antisera ( three against the N-terminal and one against the C-terminal ) recognized a single ∼27 kD band in wild-type head homogenates by immunoblotting ( Figure 2A ) . This band was greatly reduced in dPobe02662 homozygous head homogenates , indicating that these four antisera recognized dPob and that the molecular weight of dPob is ∼27 kD . In immunostaining dPobe02662 mosaic retinas , two of the C-terminal antisera ( dPob-C1 and dPob-C3 ) produced similar staining patterns in the cytoplasm of wild-type cells which were reduced in dPobe02662 homozygous photoreceptors ( Figure 2B and Figure 3B ) , indicating that these two antisera recognized dPob in tissue . Because dPob-C3 antiserum had the highest reactivity , we used it in further experiments . Anti-dPob reactivity co-localized with ER markers NinaA and HDEL ( Figure 2C , D ) , indicating ER localization of dPob in fly photoreceptors . 10 . 7554/eLife . 06306 . 005Figure 3 . dPob stabilizes rhodopsin 1 ( Rh1 ) apoprotein . ( A ) Immunostaining of a dPob∆4 mosaic retina from a fly reared in vitamin A ( VA ) -deficient medium by anti-Rh1 antibody . Asterisks show dPob∆4 homozygous photoreceptors . ( B–D ) Immunostaining of a wild-type ( B ) , ninaAp263 ( C ) , or dPob∆4 ( D ) ommatidium of flies reared in normal vitamin A-containing medium . ( E ) Immunostaining of a dPobe02662 mosaic retina in ninaAp263 homozygous mutant background from a fly reared in normal medium . Asterisks show dPob∆4 homozygous photoreceptors . Scale bar: 5 μm ( A–E ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06306 . 005 Rh1 comprises opsin ( an apoprotein ) and 11-cis retinal ( a chromophore ) . Without the chromophore , newly synthesized Rh1 apoprotein accumulates in the ER as an N-glycosylated immature form ( Ozaki et al . , 1993 ) . To investigate whether dPob is essential for the accumulation of Rh1 apoprotein in the ER , dPob∆4 mosaic retinas were observed in flies reared in medium lacking vitamin A , the source of the chromophore ( Figure 3A ) . Rh1 apoprotein was greatly reduced in dPob∆4 photoreceptor cells , indicating that dPob is essential for the early stage of Rh1 biosynthesis before chromophore binding in the ER . NinaA , the rhodopsin-specific peptidyl-prolyl-cis-trans-isomerase , is a known Rh1 chaperone . In contrast to dPob deficiency , which lacks both Rh1 apoprotein and mature Rh1 ( Figure 3D ) , loss of NinaA causes accumulation of Rh1 apoprotein in the ER similar to that observed in the chromophore-depleted condition ( Colley et al . , 1991 ) ( Figure 3C ) . To investigate the epistatic interaction between dPob and NinaA for Rh1 synthesis , Rh1 apoprotein was observed in the dPob∆4/ninaAp263 double mutant . Rh1 apoprotein was greatly reduced in dPob∆4/ninaAp263 double-mutant photoreceptors , similar to that in the dPob∆4 single mutant ( Figure 3E ) . This indicates that dPob is epistatic to NinaA . Cnx is also an Rh1 chaperone and is known to be epistatic to NinaA . Rh1 apoprotein is greatly reduced in both the cnx1 mutant and cnx1/ninaAp269 double mutant ( Rosenbaum et al . , 2006 ) , suggesting that dPob functions in the same stage or a stage close to that in which Cnx functions . The null mutant of dPob shows a characteristic phenotype with no detectable protein expression of Rh1 and very weakened expression of other multiple-transmembrane domain proteins such as Na+K+-ATPase in the mosaic retina ( see below ) . We did not find any other mutant lines with such a phenotype in the course of mosaic screening among 546 insertional mutants described previously ( Satoh et al . , 2013 ) . To explore other mutants showing phenotypes similar to the dPob null mutant , we examined a collection of 233 mutant lines deficient in Rh1 accumulation in photoreceptor rhabdomeres obtained in an ongoing ethyl methanesulfonate ( EMS ) mutagenesis screening . The detail of the screening will be published elsewhere; at present the Rh1 accumulation mutant collection covers three chromosome arms , approximately 60% of the Drosophila melanogaster genome . Under the assumption of a Poisson distribution of the mutants on genes , the collection stochastically covers more than 80% of genes in those arms . The distribution of Rh1 and Na+K+-ATPase was examined for 55 lines of mutants on the right arm of the third chromosome , 93 lines of mutants on the right arm of the second chromosome , and 85 mutants on the left arm of the second chromosome . Among them , only two lines—665G on the right arm of the third chromosome and 008J on the right arm of the second chromosome—showed a dPob null-like phenotype in the mean distribution of Rh1 and Na+K+-ATPase in the mosaic retina ( Figure 4A , C ) . 10 . 7554/eLife . 06306 . 006Figure 4 . Loss of rhodopsin 1 ( Rh1 ) apoprotein in EMC1 and EMC8/9 deficiency . Immunostaining of a EMC1655G mosaic retina ( A , B ) or a EMC8/9008J mosaic retina ( C , D ) reared in normal ( A , C ) and vitamin A-deficient media ( B , D ) . Asterisks show EMC1655G or EMC8/9008J homozygous photoreceptors . RFP ( red ) indicates wild-type photoreceptors ( R1–R8 ) . ( A , C ) Na+K+-ATPase , green; Rh1 , blue; RFP , red . ( B , D ) Rh1 , green; RFP , magenta . Scale bar: 5 μm ( A–D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06306 . 006 Meiotic recombination mapping and RFLP analysis ( Berger et al . , 2001 ) were used to map the mutations responsible for the dPob-like phenotype of 008J and 655G . Close linkage of the mutation responsible for the dPob-like phenotype of 655G indicated that the responsible gene is located close to the proximal FRT . Since CG2943 gene , the potential Drosophila homolog of EMC1 , is also close to the proximal FRT , CG2943 was recognized as a candidate of the responsible gene of 655G . As expected , Df ( 3R ) BSC747 , which is lacking the CG2943 gene , failed to complement the lethality of 655G . Targeted re-sequencing in the vicinity of CG2943 revealed that 655G has a two-base deletion at 3R:3729838-3729839 which causes a frame-shift mutation of CG2943 , causing185aa deletion from I730 to C-terminus adding polypeptide of RTVRGQESGKQQCLEFLASSANAPRGAPVLYTAHNS . The only membrane-spanning helix of CG2943 is lost in this frame-shift mutation . RFLP analysis narrowed down the cytology of the responsible gene of 008J to 58D2−59D11 . Whole genome re-sequencing revealed that the 008J chromosome obtained three unique mutations in the mapped region compared with the starter stock: one silent mutation on CG30274 at 2R:18714026 , a missense mutation on MED23 ( E329K ) at 2R:18777637 , and one nonsense mutation on CG3501 at 2R:18770005 which turns Q40 to a stop codon . Complementation with the deficiencies over the MED23 ( BSC783 , BSC784 ) excluded the missense mutation on MED23 from the candidate mutation responsible for the dPob-like phenotype . The amino acid sequence of CG3501 shows 38% and 39% identity to the human EMC8 and EMC9 , respectively , and no other gene similar to EMC8/9 was found in the Drosophila genome . Based on these results , we identified 655G and 008J as a loss of functional mutation of EMC1 and EMC8/9 of Drosophila and named these alleles EMC1655G and EMC8/9008J . We investigated whether EMC1 and EMC8/9 are necessary for the accumulation of Rh1 apoprotein in the ER using EMC1655G and EMC8/9008J mosaic retinas reared in medium lacking vitamin A ( Figure 4B , D ) . Rh1 apoprotein was greatly reduced in both EMC1655G and EMC8/9008J photoreceptor cells , indicating that EMC1 and EMC8/9 are also essential for the early stage of Rh1 biosynthesis , like dPob . To investigate if EMCs form a complex and bind to Rh1 apoprotein , we performed a co-immunoprecipitation assay ( Figure 5 ) . Since C-terminally tagged dPob protein did not predominantly localize to the ER in vivo ( data not shown ) , GFP-tagged EMC1 protein ( EMC1::GFP ) was used as the bait . A protein-trap line expressing GFP-tagged sec61alpha protein ( sec61::GFP ) which localizes in the ER membrane was used as a negative control . Since the overall expression level of EMC1::GFP was strong , hs-Gal4 driver was used to activate UAS:EMC1::GFP for most of the experiments . To analyze the interaction between EMC1 and Rh1 apoprotein , Rh1-Gal4 driver was also used because the expression of EMC1::GFP was stronger in the photoreceptors ( data not shown ) . For the Rh1-Gal4 experiment , flies were reared in a medium lacking vitamin A to accumulate Rh1 apoprotein in the ER . Membrane fraction was recovered from the adult heads , the membrane proteins extracted by CHAPS from the adult head membrane fraction were bound to anti-GFP magnetic beads , and the elutions were analyzed by immunoblotting with antibodies against GFP , Rh1 , dPob , and Cnx . 10 . 7554/eLife . 06306 . 007Figure 5 . Co-immunoprecipitation of EMC1::GFP with dPob and calnexin ( Cnx ) . Immunoblotting of precipitates with anti-GFP antibody from the head extract was prepared from Rh1-Gal4/UAS-EMC1::GFP or sec61::GFP flies reared in a vitamin A ( VA ) -deficient medium ( left ) or heat shock ( hs ) -Gal4/UAS-EMC1::GFP or sec61::GFP flies reared in a vitamin A-containing normal medium ( right ) . The mature form of rhodopsin 1 ( Rh1 ) is accumulated in the rhabdomeres in normal medium but not in vitamin A-deficient medium . Instead of the mature form , an N-glycosylated immature form of Rh1 with a larger molecular weight accumulated in the endoplasmic reticulum of flies reared in the vitamin A-deficient medium . In both input extracts prepared from Rh1-Gal4/UAS-EMC1::GFP or sec61::GFP flies there is a band with the same position as EMC1GFP; this band will be the protein cross-reacting to anti-GFP antibody . DOI: http://dx . doi . org/10 . 7554/eLife . 06306 . 007 EMC1::GFP and sec61::GFP were concentrated in the immunoprecipitated extract from flies expressing either in the photoreceptor or in the whole head . dPob was co-immunoprecipitated with EMC1::GFP much more strongly than with sec61::GFP . Cnx was also well co-immunoprecipitated with EMC1::GFP but was barely detectable with sec61::GFP . However , Rh1 was not co-immunoprecipitated with EMC1::GFP from vitamin A-deficient photoreceptors accumulating immature Rh1 apoprotein in the ER . These results indicate that dPob and EMC1 are in a complex in vivo , as shown in yeast , and Cnx can also be associated with the complex , which is consistent with the result of epistatic analysis; the stage at which dPob works on the expression of Rh1 apoprotein is close to that of Cnx . Despite the requirement for the expression of Rh1 and co-localization with immature Rh1 apoprotein in the ER , EMC1 does not stably bind to Rh1 , indicating that the EMC complex is only temporarily associated with Rh1 apoprotein . To investigate the substrate specificity of EMC/dPob , we investigated the expressions of secreted or transmembrane proteins other than Rh1 in dPob∆4 mosaic retinas . In dPob∆4 photoreceptors , multi-pass membrane proteins , the alpha subunit of Na+K+-ATPase ( Figure 6A ) and transient receptor potential ( TRP ) ( Figure 6B ) , were greatly reduced and neither anti-Rh3 nor anti-Rh4 staining was detected ( Figure 6C , D ) . On the other hand , the type I single-pass membrane proteins Crb ( Figure 6B ) and DE-Cad ( Figure 6E ) were localized normally in the stalks and adherence junctions in dPob∆4 photoreceptors . Similarly , a type II single-pass membrane protein Nrt ( Figure 6G ) and a type VI single-pass membrane protein Syx1A ( Figure 6H ) were localized normally in Golgi units and on the plasma membrane in Pob∆4 photoreceptors . Eys , a secreted protein that expands the inter-rhabdomeric space ( IRS ) ( Husain et al . , 2006; Zelhof et al . , 2006 ) , was also secreted normally in dPob∆4 ommatidia , as expected from the near-normal size of the IRS ( Figure 6I ) . Two other type I single-pass membrane proteins expressed in retinal cone cells , Neuroglian ( Nrg ) and Fasiclin III ( FasIII ) , exhibited normal localization in contact sites between cone cells and cone cell feet ( Figure 6J , K ) . Only one type II single-pass membrane protein , the beta subunit of Na+K+-ATPase ( Nrv ) , showed deficient expression in Pob∆4 photoreceptors ( Figure 6F ) . As alpha and beta subunits of Na+K+-ATPase are assembled into a heterodimer within the ER and then transported to the plasma membrane , the absence of Nrv in Pob∆4 photoreceptors can be interpreted as a consequence of the lack of the multi-pass alpha subunit . These results indicate that dPob is essential for the normal biosynthesis of multi-pass membrane proteins but not for single-pass membrane proteins or secreted proteins . 10 . 7554/eLife . 06306 . 008Figure 6 . Essential role of dPob in the biosynthesis of multi-pass transmembrane proteins . Immunostaining of a dPob∆4 mosaic retina ( A–H ) or a dPobe02662 mosaic retina ( I ) . Asterisks show dPob homozygous photoreceptors . ( A ) Na+K+-ATPase , green; Rh1 , magenta . ( B ) Crb , green; TRP1 , magenta . ( C , D ) Rh3 ( C ) and Rh4 ( D ) , green; RFP ( wild-type cell marker ) , magenta . Although the boundary between dPob∆4 and wild-type cells is unclear , all green signals are attached to RFP-expressing cell bodies , indicating that mutant R7 cells do not express Rh3 ( C ) or Rh4 ( D ) . ( E ) DE-Cad staining . ( F ) Nrv , the beta subunit of Na+K+-ATPase , green; dMPPE , magenta . ( G ) Nrt staining . ( H ) Syx1A staining . ( I ) Eys staining . ( J ) Nrg , blue; F-actin , red; GFP-nls ( wild-type cell marker ) , green . ( K ) FasIII staining . ( L ) Na+K+-ATPase , green; Rh1 , magenta . Scale bar: 2 μm ( A , B ) , 10 μm ( C , D ) , 2 μm ( E–I ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06306 . 008 EMC1655G- and EMC8/9008J-deficient photoreceptors show similar substrate specificity to dPob∆4-deficient photoreceptors ( Figure 6 and Figure 7 ) . In both mutants , accumulation of the membrane proteins with multiple transmembrane domains , Na+K+-ATPase ( Figure 4A , C ) , Rh3 , Rh4 and TRP ( Figure 7A , C ) , on the plasma membrane are greatly reduced in the photoreceptors . However , a type I single-pass transmembrane protein , Crb , is localized intensively in the stalks in EMC1655G or EMC8/9008J mutant photoreceptors ( Figure 7B , D ) . A type II single-pass membrane protein , Nrt , and a type VI single-pass membrane protein , Syx1A , is localized normally in Golgi units and on the plasma membrane in EMC1655G and EMC8/9008J photoreceptors , respectively ( Figure 7C , F ) . Eys was also secreted normally and formed a near-normal size of IRS in EMC1655G or EMC8/9008J mutant ommatidia ( Figure 7B , D ) . Similar to Pob∆4 photoreceptors , a type II single-pass membrane protein , the beta subunit of Na+K+-ATPase ( Nrv ) was not detected in the plasma membrane of EMC1655G or EMC8/9008J photoreceptors ( data not shown ) . 10 . 7554/eLife . 06306 . 009Figure 7 . Essential role of EMC1 and EMC8/9 in the biosynthesis of multi-pass transmembrane proteins . Immunostaining of a EMC1655G mosaic retina ( A , B , C ) or a EMC8/9008J mosaic retina ( D , E , F ) . ( A , D ) Left: Rh3 , middle: Rh4 , right: TRP in green , RFP in magenda . ( B , E ) Eys in green , Crb in blue , and RFP , wild-type cell marker in red . ( C , F ) Left: dMPPE , middle: Nrt , right: Syx1A in green , RFP in magenda . Scale bar: 10 μm ( left and middle in A , D ) , 5 μm ( right in A , D ) , 5 μm ( B , C , E , F ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06306 . 009 We observed the expression of dMPPE ( Cao et al . , 2011 ) , a Golgi luminal metallophosphoesterase , anchored by a type II transmembrane helix in the N-terminal region and another transmembrane helix in the C-terminal region . dMPPE was expressed normally in Pob∆4 , EMC1655G , and EMC8/9008J mutant photoreceptors ( Figures 6F , 7C , F ) . As two transmembrane helices of dMPPE are separated from each other by the enzymatic domain , these two helices might not associate but behave as two separate transmembrane helices . The EMC requirement for proteins with two transmembrane helices therefore remains unclear . Electron microscopic observation of thin sections of late pupal flies showed massive amplification of the ER membrane in both dPobe02662 and dPob∆4 photoreceptors ( Figure 8A–C ) despite the substantial reduction in immature Rh1 apoprotein . In dPobe02662 photoreceptors the ER maintains its sheet structures: the number and length of the sheets was greatly increased but their lumens were almost normal with slight swelling and the sheets were aligned at a regular distance . Meanwhile , in dPob∆4 photoreceptors the ER sheet structures were no longer maintained and the cytoplasmic space was filled with ER membrane with a larger luminal space . Golgi bodies were also swollen and dilated , and sometimes vesiculated ( Figure 8A–C , insets ) . Moreover , concordant with the reduction in Rh1 , the rhabdomeres in dPob mutant photoreceptors were quite small and thin but the adherence junctions and basolateral membrane exhibited normal morphology . ER membrane amplification and rhabdomere membrane reduction therefore represent the most prominent phenotype in dPob-deficient photoreceptors . 10 . 7554/eLife . 06306 . 010Figure 8 . Endoplasmic reticulum membrane amplification and unfolded protein response ( UPR ) induced in dPob∆4 photoreceptor . ( A–C ) Electron microscopy of late pupal photoreceptors: wild-type ( A ) , dPobe02662 ( B ) , and dPob∆4 photoreceptors ( C ) . Arrow indicate adherens junctions . Insets show Golgi bodies . ( D , E ) Immunostaining of a dPobe02662 mosaic retina . dPob is shown in green and KDEL ( D ) or HDEL ( E ) are shown in magenta . Asterisks show dPob∆4 homozygous photoreceptors . Scale bar: 1 μm ( A–C ) , 5 μm ( D , E ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06306 . 010 The massive amplification of the ER membrane in both dPobe02662 and dPob∆4 photoreceptors prompted us to quantify the amounts of residual ER proteins using anti-KDEL and HDEL antibodies . KDEL and HDEL sequences are signals for ER retention , and Drosophila ER resident chaperones including Hsp70–3 and PDI contain these sequences ( Bobinnec et al . , 2003; Ryoo et al . , 2007 ) . Corresponding to ER membrane amplification , anti-HDEL and anti-KDEL staining were greatly increased in dPob-deficient photoreceptors ( Figure 8D , E ) . Accumulation of unfolded proteins in the ER invokes the UPR , which includes activation of the transcription of chaperones and related genes , suppression of translation and enhanced degradation of unfolded protein . The UPR is regulated by some unique intracellular signal transduction pathways . Therefore , mutants lacking the function of a gene essential for folding or degradation of unfolded protein probably exhibit UPR . In fact , the yeast Pob homolog , EMC3 , was identified by screening of mutants exhibiting upregulated UPR . ER amplification and chaperone induction , which we observed in dPob-deficient photoreceptors , are also common outcomes of UPR . We therefore examined whether UPR is induced in dPob-deficient photoreceptors . First we used the Xbp1:GFP sensor , which is an established method for detecting UPRs in flies ( Ryoo et al . , 2007 ) . During UPR , Ire1 catalyzes an unconventional splicing of a small intron from the xbp1 mRNA , enabling translation into an active transcription factor ( Yoshida et al . , 2001 ) . Using this mechanism , Xbp1:GFP sensor , a fused transcript of Drosophila Xbp1 and GFP translated only after the unconventional splicing by Ire1 , can be used as a reporter of one of the UPR transduction pathways ( Ryoo et al . , 2007 ) . In both dPob∆4 and dPobe02662 mutant mosaic retinas expressed Xbp1:GFP sensor in all R1−6 photoreceptors , and Xbp1:GFP fusion proteins were detected in the dPob mutant photoreceptors but not in the wild-type ( Figure 9A and data not shown ) . Next , we examined the level of eukaryotic translation Initiation Factor 2α ( eIF2α ) phosphorylation because UPR is well known to induce eIF2α phosphorylation to attenuate protein translation on the ER membrane in a transduction pathway independent from IreI/Xbp1 ( Ron and Walter , 2007; Cao and Kaufman , 2012 ) . Anti-phospho-eIF2α signals were stronger in both dPob∆4 and dPobe02662 photoreceptors than in wild-type photoreceptors ( Figure 9B and data not shown ) . These results indicate that UPR is induced in the dPob-deficient photoreceptors , similar to EMC mutant . 10 . 7554/eLife . 06306 . 011Figure 9 . Unfolded protein response ( UPR ) induced in dPob∆4 photoreceptor . ( A ) Projection image from the Z-series section with a 1 μm interval of dPob∆4 mosaic retina expressing RFP ( magenta ) as a wild-type cell marker and Xbp1:GFP as a UPR sensor . The Xbp1:GFP signal ( green ) is enhanced by immunostaining using anti-GFP antibody . Asterisks show dPob∆4 homozygous photoreceptors . ( B ) Immunostaining of a dPob∆4 mosaic retina expressing RFP ( magenta ) as a wild-type cell marker . Phosphorylated eukaryotic translation Initiation Factor 2α is shown in green . Asterisks show dPob∆4 homozygous photoreceptors . DOI: http://dx . doi . org/10 . 7554/eLife . 06306 . 011 Because the synthesis of many membrane proteins was affected in dPob mutant cells , we observed the phenotype of dPob mutant throughout the developmental processes of photoreceptors . Despite the lack of many membrane proteins , ommatidial formation was not affected in dPob∆4 photoreceptors in mosaic retina; adherence junctions formed normally ( Figure 6E ) and the apical membrane was well differentiated into stalks and rhabdomeres ( identified with Crb and phosphorylated moesin , respectively ) ( Figure 6B and data not shown ) ( Karagiosis and Ready , 2004 ) . The IRS was formed normally and rhabdomeres were still separated by IRSs ( Figure 8A–C ) . We observed dPob∆4 mosaic retinas at 58% and 75% pupal development ( pd ) by electron microscopy ( Figure 10A , B ) . The wild-type photoreceptors at 58% pd had already begun to amplify the rhabdomere membranes . The rhabdomeres were shorter in dPob∆4 photoreceptors than in wild-type photoreceptors , but the difference in their appearance was subtle at this stage . Until 75% pd , the microvilli of wild-type rhabdomeres were ∼0 . 5 μm long and packed tightly . However , the microvilli of dPob∆4 rhabdomeres at 73% pd retained almost the same length and appearance as those at 58% pd , which is the same as the dPob∆4 rhabdomeres of the late pupal retina ( Figures 10A , B and 8C ) . ER membrane expansion and dilation were already apparent at 58% pd . These results indicate that dPob does not inhibit overall photoreceptor development and morphogenesis but does affect microvilli elongation and rhabdomere formation . 10 . 7554/eLife . 06306 . 012Figure 10 . Development and degeneration of dPob∆4 photoreceptor rhabdomeres . Electron microscopy of pupal and adult dPob∆4 mosaic retinas . Asterisks show dPob∆4 homozygous photoreceptors . Scale bar: 1 μm . ( A , B ) dPob∆4 mosaic ommatidia from 58% pupal development ( A ) and 73% pupal development ( B ) under constant light ( L ) condition . ( C–F ) dPob∆4 mosaic ommatidia from flies reared in complete darkness ( D ) ( C , E ) or under 12 hr light/12 hr dark conditions ( D , F ) . Ommatidia from 3-day-old ( C , D ) and 17-day-old ( E , F ) flies . ( D , inset ) dPob∆4 R5 photoreceptor rhabdomere at higher magnification . DOI: http://dx . doi . org/10 . 7554/eLife . 06306 . 012 Because zebrafish pob was identified as the responsible gene of poba1 mutant which exhibits red cone photoreceptor degeneration ( Brockerhoff et al . , 1997; Taylor et al . , 2005 ) , we investigated photoreceptor degeneration of the dPob null mutant . Three-day-old dPob∆4 mosaic retinas from flies reared under dark or 12 hr light/12 hr dark cycles were observed by electron microscopy ( Figure 10C , D ) . In both conditions the rhabdomeres of dPob∆4 photoreceptors invaginated into the cytoplasm , indicating that dPob-deficient rhabdomeres undergo retinal degeneration in a light-independent manner , like Rh1 null mutants ( Kumar and Ready , 1995 ) . No microvilli or invaginations were observed in 17-day-old dPob∆4 mosaic retinas , suggesting most invaginated microvilli had degraded before day 17 ( Figure 10E , F ) . Such rhabdomere degeneration was observed not only in R1–6 peripheral photoreceptors but also in R7 central photoreceptors . Therefore , dPob is an essential protein for maintenance of retinal structure , similar to the zebrafish pob gene .
The present study shows that dPob , the Drosophila homolog of a subunit of EMC , EMC3 , localizes in the ER and is essential for Rh1 accumulation of the rhabdomeres . The deficiency of each of two other EMC subunits , EMC1 and EMC8/9 , also shows absence of Rh1 on the rhabdomeres . Mammalian EMC8 and EMC9 were identified together with EMC7 and EMC10 by high-content proteomics strategy ( Christianson et al . , 2011 ) . Unlike EMC1−6 subunits , EMC8 and EMC9 do not have a transmembrane helix or signal peptide and no experimental data have been reported to show the functions of these subunits . We observed that Drosophila EMC8/9-deficient cells lack accumulation of Rh1 apoprotein in the ER and impaired biosynthesis of the multi-pass transmembrane proteins . These phenotypes in EMC8/9 deficiency are indistinguishable from those in dPob and EMC1 mutant cells , suggesting that EMC8/9 work together with EMC1 and dPob . This is the first functional study of the additional subunits of EMC , which are lacking in yeast . We found that null mutants of EMC subunits are defective in expressing the multi-pass transmembrane proteins rhodopsins , TRP , and the alpha subunit of Na+K+-ATPase , which have seven , six , and eight transmembrane helices , respectively . In contrast , the EMC null mutants adequately express type I , type II , or type IV single-pass membrane proteins . Our observation on the substrate specificity of EMC is mostly consistent with previous reports . Jonikas et al . ( 2009 ) found that EMC mutants and a strain overexpressing a misfolded transmembrane protein , sec61-2p or KWS , had a similar genetic interaction pattern and suggested that EMC works as a chaperone for transmembrane proteins . A recent study in Caenorhabditis elegans using a hypomorphic EMC6 allele and RNAi knock-down of emc1–6 genes showed results partially consistent with our study; at least two pentameric Cys-loop receptors , AcR and GABAA , consisting of subunits with four transmembrane helices , were significantly decreased in the hypomorphic EMC6 mutants but GLR-1 , a tetrameric AMPA-like glutamate receptor with four transmembrane helices and a type I single-pass transmembrane EGF receptor , was not affected ( Richard et al . , 2013 ) . Despite its four transmembrane helices , GLR-1 was normally expressed in the hypomorphic emc6 mutant of the nematode; however , these results may indicate that the residual activity of EMC was sufficient for the expression of GLR-1 . The degree of requirement of EMC activity can vary for each membrane protein . In fact , in a dPob hypomorphic allele , dPobe02662 , near-normal expression of Na+K+-ATPase was detected ( Figure 6I ) despite a severe reduction in a dPob null allele , dPob∆4 . Overall , the results observed in the dPob null mutant does not conflict with previous studies but rather clarifies the role of EMC in the biosynthesis of multi-pass transmembrane proteins . Because of the limited availability of antibodies , we could not show a clear threshold for the number of transmembrane helices in the substrates for EMC activity . In total , the data presented to date indicate that EMC affects the expression of membrane proteins with four or more transmembrane helices . Co-immunoprecipitation of dPob/EMC3 and Cnx by EMC1 indicates that EMC components and Cnx can form a complex . The photoreceptors of an amorphic mutant of Cnx show complete loss of Rh1 apoprotein ( Rosenbaum et al . , 2006 ) , just as shown in dPob , EMC1 or EMC8/9 mutants . Moreover , both Cnx and EMC3 are epistatic to the mutant of the rhodopsin-specific chaperone , NinaA , which accumulates Rh1 apoprotein in the ER . These results indicate that EMC and Cnx can work together in the Rh1 biosynthetic cascade prior to NinaA . Cnx , the most studied chaperone of N-glycosylated membrane proteins , recognizes improperly folded proteins and facilitates folding and quality control of glycoproteins through the calnexin cycle , which prevents ER export of misfolded proteins ( Williams , 2006 ) . One possible explanation for our result is that the EMC-Cnx complex is required for multi-pass membrane proteins to be incorporated into the calnexin cycle . If the EMC-Cnx complex is a chaperone of Rh1 , physical interaction is expected between ER-accumulated Rh1 apoprotein and the EMC-Cnx complex . Indeed , it is reported that Cnx is co-immunoprecipitated with Drosophila Rh1 ( Rosenbaum et al . , 2006 ) . However , in this study , Rh1 apoprotein accumulated in the chromophore-depleted photoreceptor cells was not co-immunoprecipitated with EMC1 . Thus , even if EMC is a Rh1 chaperone , our result indicates that EMC is unlikely to be working in the calnexin cycle or acting as a buffer of properly folded Rh1 apoprotein ready to bind the chromophore 11-cis retinal . In addition to preventing the export of immature protein by the calnexin cycle , Cnx is also known to recognize the nascent polypeptides co-translationally ( Chen et al . , 1995 ) . The dual role of Cnx might explain the observations that both dPob/EMC3 and Cnx are epistatic to another ER resident chaperone , NinaA , whereas Cnx but not the EMC-Cnx complex binds to Rh1 . These results imply that the EMC-Cnx complex is more likely to be involved in the earlier processes such as membrane integration or co-translational folding than in the folding of fully translated membrane-integrated Rh1 apoprotein . In spite of the absence of Rh1 apoprotein , UPR is much more upregulated in the EMC3 null mutant than in the NinaA null mutant which accumulates Rh1 apoprotein in the ER . The elevated UPR without accumulation of Rh1 apoprotein in the dPob mutant photoreceptor can be explained either by the quick degradation of Rh1 apoprotein or by accumulation of the single-pass membrane proteins abandoned by the multi-pass binding partner . Newly synthesized secreted proteins co-translationally translocate across the membrane through the translocons Sec61 in eukaryotic ER or SecYEG in the plasma membrane of bacteria . The translocons also mediate integration of the transmembrane helix of the integral membrane protein into the lipid bilayer ( Park and Rapoport , 2012 ) . In bacteria , mitochondria and chloroplasts , YidC/Oxa1/Alb3 proteins specifically facilitate insertion , folding , and assembly of many transmembrane proteins ( Wang and Dalbey , 2011 ) . In the ER membrane of eukaryotes , in addition to the translocon , other components such as translocon-associated protein/signal sequence receptor ( TRAP/SSR ) complex and translocating chain-associating membrane protein ( TRAM ) complex are required for the membrane insertion of the transmembrane helix . Most of the newly synthesized multi-pass membrane proteins are co-translationally integrated into the ER membrane through the translocon complex . Although the mechanism of this process is yet to be fully understood , it is assumed that only one or two transmembrane helices can be stored in the translocon channel and the lateral gate and that the next set of newly synthesized transmembrane helices displace them ( Rapoport et al . , 2004; Cymer et al . , 2014 ) . In the case of nascent chain of bovine rhodopsin , translocon associates with transmembrane helices sequentially , and TRAM temporarily associates with the second transmembrane helix ( Ismail et al . , 2008 ) . EMC may be involved in these co-translational membrane integration or co-translational folding processes . Zebrafish pob was identified as the responsible gene of poba1 mutant , which exhibits red cone photoreceptor degeneration ( Brockerhoff et al . , 1997; Taylor et al . , 2005 ) . Because only red cone photoreceptors degenerated in zebrafish poba1 mutant , pob is postulated as a gene with a red cone-specific function . However , the identification of the poba1 mutation as hypomorphic together with pob expression in all photoreceptors , as well as its localization in the early secretory pathway , suggests that Pob has a general function rather than being red cone-specific ( Taylor et al . , 2005 ) . We found that dPob-deficient rhabdomeres undergo retinal degeneration in a light-independent manner , like Rh1 null mutants ( Kumar and Ready , 1995 ) . Rhabdomere degeneration was observed not only in R1–6 peripheral photoreceptors but also in R7 central photoreceptors . Our results indicate that dPob is an essential protein for the maintenance of retinal structure , similar to the zebrafish pob gene .
Flies were reared at 20–25°C in 12 hr light/12 hr dark cycles and fed standard cornmeal/glucose/agar/yeast food unless noted otherwise . Vitamin A-deficient food contained 1% agar , 10% dry yeast , 10% sucrose , 0 . 02% cholesterol , 0 . 5% propionate , and 0 . 05% methyl 4-hydroxybenzoate . UAS-Xbp1::GFP was a gift from H Ryoo at New York University and other Drosophila stocks obtained from Bloomington Stock Center ( BL ) or the Kyoto Drosophila Genetic Resource Center ( KY ) are referred to with their respective sources and stock numbers . dPob deletion mutants were made using a standard induced FLP/FRT recombination method ( Parks et al . , 2004 ) . Trans-heterozygous PBac ( WH ) f07762 ( BL19109 ) and P ( RS3 ) CB−0279−3 ( KY123106 ) males carrying hs-FLP ( BL6876 ) were heat treated three times at 37°C for 1 hr at larval stages . SM6a-balanced offspring were genotyped using PCR to select the recombinant carrying both the proximal side of PBac ( WH ) f07762 and the distal side of P ( RS3 ) CB−0279−3 with the following primers: 5′-CTCCTTGCCAGCTTCTGC-3′ and 5′-TCGCTGTCTCACTCAGACTCA-3′ for P ( RS3 ) CB−0279−3 , and 5′–CCACCGAAGAGGCCTACTATT-3′ and 5′-TCCAAGCGGCGACTGAGATG-3′ for PBac ( WH ) f07762 . The entire coding region of the dPob gene was amplified from a cDNA clone LD37839 ( DGRC: Drosophila Genomics Resource Center , Bloomington , IN , USA ) and cloned into pTW ( DGRC ) to construct pP{UAST-dPob} . To construct pP{UAST-EMC1::GFP} , the entire coding region of CG2943 except the stop codon was amplified from a cDNA clone LD19064 ( DGRC ) and cloned into pTWG ( DGRC ) . Plasmids were injected into embryos by BestGene Inc . ( Chino Hills , CA , USA ) to generate transgenic lines . Fluorescent proteins expressed in photoreceptors were imaged by water-immersion technique . y w ey-FLP;CG6750e02662 FRT40A/ CyO y+ ( KY114504 ) was mated with w;P3RFP FRT40A/SM1;Rh1-Arrestin2::GFP eye-FLP/TM6B ( Satoh et al . , 2013 ) . Late pupae of the siblings with GFP-positive RFP mosaic retina were attached to the slide glass using double-sided sticky tape and the pupal cases around the heads were removed . The pupae were chilled on ice , embedded in 0 . 5% agarose , and observed using an FV1000 confocal microscope equipped with a LUMPlanFI water-immersion 40× objective ( Olympus , Tokyo , Japan ) . Arrestin2::GFP specifically binds to activated rhodopsin ( Satoh et al . , 2010 ) . Rh1 was activated by a 477 nm solid-state laser to bind Arr2:GFP and GFP . The wild-type marker P3RFP is DsRed gene under the control of three Pax3 binding sites and labels photoreceptors ( Bischof et al . , 2007 ) . The precise method of screening , whole genome re-sequencing , will be described elsewhere . Briefly , second or third chromosomes carrying P-element vector with FRT on 40A , 42D , or 82B ( Berger et al . , 2001 ) were isogenized and used as the starter strains . EMS was fed to males in a basic protocol ( Bökel , 2008 ) and mosaic retinas were generated on F1 or F2 . The estimated number of lethal mutations introduced per chromosome arm was 0 . 8–1 . 8 . The mutants were screened based on the distribution of Arr2-GFP by confocal live imaging under water-immersion lens using 3xP3-RFP as the wild-type marker , as previously described for the screening of insertional mutants ( Satoh et al . , 2013 ) . Meiotic recombination mapping was carried out by the standard method ( Bökel , 2008 ) . Briefly , to allow meiotic recombination between the proximal FRT , the phenotype-responsible mutation and a distal miniature w+ marker , flies carrying isogenized chromosome of 008J and 655G were crossed with flies with isogenized P{EP755} and P{EP381} which carry miniature-w+ marker , respectively . Female offspring carrying the mutated chromosome and the miniature-w+-marked chromosome were crossed with males carrying FRT42D , P3RFP , and Rh1Arr2GFP . The resulting adult offspring with w+ mosaic , which means maternally inherited both FRT and w+ , were observed using live imaging to judge whether the mutation responsible for the dPob-like phenotype had been inherited . The recovered flies were individually digested in 50 µl of 200 ng/µl Proteinase K in 10 mM Tris-Cl ( pH 8 . 2 ) , 1 mM EDTA , and 25 mM NaCl at 55°C for 1 hr and heat inactivated at 85°C for 30 min and at 95°C for 5 min . 0 . 5 µl of the digested solution were used as the template of PCR amplification for RFLP analysis according to the method described in the FlySNP database ( Chen et al . , 2008; http://flysnp . imp . ac . at/index . php ) . The mutation responsible for the dPob-like phenotype of 008J was mapped between SNP markers 1417 and 1518 defined in the FlySNP database . For the whole genome re-sequencing of the 008J mutant , the second chromosome was balanced over a balancer , CyO , P{Dfd-GMR-nvYFP} ( Bloomington stock number 23230 ) to facilitate the isolation of homozygous embryo . Using REPLI-G single cell kit ( QIAGEN , Hilden , Germany ) , the genomic DNA was amplified from two 008J homozygous embryos independently . A sequencing library was prepared using Nextera DNA sample preparation kit ( Illumina , San Diego , CA , USA ) for each embryo and 2 × 250 bp reads were obtained using MiSeq v2 kit ( Illumina ) . Reads were mapped to release five of the Drosophila melanogaster genome using BWA 0 . 7 . 5a . The RFLP-mapped region of 008J was covered by reads with an average depth of 23 . 2× and width of 99 . 5% . Mapped reads were processed using picard-tools 1 . 99 and Genome Analysis Tool Kit 2 . 7-2 ( GATK , Broad Institute , Cambridge , MA , USA ) . SNVs and Indels were called using Haplotypecaller in GATK . SNVs and Indels were subtracted by the ones of the isogenized starter stock to extract the unique variants in 008J and annotated using SnpSift ( Cingolani , 2012 ) . The point mutation on 2R:18770005 was verified by capillary sequencing of PCR-amplified fragment using 5′ GTCGCGGTCACACTTTCTAG 3′ and 5′ CTGCAGCGTCATCAGTTTGT 3′ as primers . For targeted re-sequencing of 655G , a region including CG2943 was amplified from a heterozygous fly of the 655G mutant chromosome and the starter chromosome using KOD FX Neo DNA polymerase and 5′ TTTTGTTCTTGTTGGGCGACTCCTTTTCCGTCTC 3′ and 5′ AGGCTGTGTCTTTGTTGTTTTGGCGTTGTCGTC 3′ as primers . Reads covering the CG2943 gene region at a depth of 2213–6436 were obtained using MiSeq and mapped , as described above . The sequence was confirmed by capillary sequencing and PCR using 5′ GCAAGAATCCCATCGAGCAT 3′ and 5′ CCTTCTTCACGTCCCTGAGT 3′ as primers . Fragments of cDNA encoding V28-D104 ( dPob-N ) or G173-S247 ( dPob-C1 ) of dPob were amplified from a cDNA clone , LD37839 ( Drosophila Genomics Resource Center , Bloomington , IN , USA ) and cloned into pDONR-211 using Gateway BP Clonase II and then into pET-161 expression vector using Gateway LR Clonase II ( Life Technologies , Carlsbad , CA , USA ) . The fusion proteins with 6xHis-tag were expressed in BL21-Star ( DE3 ) ( Life Technologies ) and purified using Ni-NTA Agarose ( QIAGEN ) . To obtain antisera , rabbits were immunized six times with 300 µg dPob-N fusion protein ( Operon , Tokyo , Japan ) and three rats were immunized six times with 125 µg dPob-C1 fusion protein ( Biogate , Gifu , Japan ) . Antisera against Drosophila Cnx were raised by immunizing a rabbit four times with 400 to 200 µg of synthetic peptide corresponding to C-terminal 24 amino acids of Cnx99a protein conjugated to KLH ( Sigma Aldrich Japan , Tokyo , Japan ) . Immunoblotting was performed as described previously ( Satoh et al . , 1997 ) . The antibodies used were as follows: rabbit anti-dPob–N-terminal ( dPob-N ) ( 1:2000 concentrated supernatant ) ( made by the authors of this paper ) , three rat anti-dPob–C-terminal antibodies ( dPob-C1-3 ) ( 1:2000 concentrated supernatant ) ( made by the authors of this paper ) as primary antibodies . HRP-conjugated anti-rat or anti-rabbit IgG antibody ( 1:20 , 000 , Life Technologies ) was used as a secondary antibody . For co-immunoprecipitation , 1:2000 rabbit anti-dPob-N , 1:2000 rabbit anti-Cnx99A , 1:2000 rabbit anti-GFP ( Life Technologies ) , mouse anti-Rh1 monoclonal antibody 4C5 , and detected by biotinylated secondary antibodies followed by HRP-conjugated avidin . Signals were visualized using enhanced chemiluminescence ( Clality Western blotting ECL Substrate; BioRad , Hercules , CA , USA ) and imaged using ChemiDoc XRS+ ( BioRad ) . Fixation and staining were performed as described previously ( Satoh and Ready , 2005 ) . The primary antisera were as follows: rabbit anti-Rh1 ( 1:1000 ) ( Satoh et al . , 2005 ) , chicken anti-Rh1 ( 1:1000 ) ( Satoh et al . , 2013 ) , mouse monoclonal anti-HDEL ( 1:100 ) ( Santa Cruz Biotechnology , Dallas , TX , USA ) , mouse monoclonal anti-KDEL ( 1:100 ) ( Assay Designs , Ann Arbor , MI , USA ) , rabbit anti-NinaA ( 1:300 ) ( gift from Dr Zuker , Colombia University ) , mouse monoclonal anti-Na+K+-ATPase α subunit ( 1:500 ascite ) ( DSHB , Iowa City , IA , USA ) , rat monoclonal anti-DE-cad ( 1:20 supernatant ) ( DSHB ) , mouse monoclonal anti-Syx1A ( 1:20 supernatant ) ( DSHB ) , mouse monoclonal anti-Nrt ( 1:20 supernatant ) ( DSHB ) , mouse monoclonal anti-Nrv ( 1:20 supernatant ) ( DSHB ) , mouse monoclonal anti-FasIII ( 1:20 supernatant ) ( DSHB ) , mouse monoclonal anti-Nrg ( 1:20 supernatant ) ( DSHB ) , mouse monoclonal anti-Chp ( 24B10 ) ( 1:20 supernatant ) ( DSHB ) , rat anti-Crb ( gift from Dr Tepass , University of Toronto ) , rabbit anti-TRP ( gift from Dr Montell , Johns Hopkins University ) , rabbit anti-dMPPE ( 1:50 ) ( gift from Dr Han , Southeast University ) , and rabbit anti-phosphorylated eIF2α ( 1:300 ) ( Cell Signaling Technologies , Danvers , MA , USA ) . The secondary antibodies used were anti-mouse , rabbit , rat , and chicken IgG labeled with Alexa Fluor 488 , 568 , and 647 ( 1:300 ) ( Life Technologies ) and Cy2 ( 1: 300 ) ( GE Healthcare Life Sciences , Pittsburgh , PA , USA ) . Samples were examined and images recorded using a FV1000 confocal microscope ( 60× , 1 . 42-NA lens; Olympus , Tokyo , Japan ) . To minimize bleed-through , each signal in double- or triple-stained samples was imaged sequentially . Images were processed in accordance with the guidelines for proper digital image handling using ImageJ and/or Adobe Photoshop CS3 . The EMC1 gene was cloned into a P-element vector pTWG using the Gateway System ( Life Technologies ) to express EMC1 protein-tagged GFP on the C-terminus under control of upstream activation sequence ( UAS ) . Transgenic lines were generated by the BestGene Inc . ( Chino Hills , CA , USA ) . UAST-EMC1-GFP ( 1M ) , a line carrying the transgene on the second chromosome , was crossed to Rh1-Gal4 line to express EMC1-GFP in the photoreceptor or to hs-Gal4 line to express EMC1-GFP in the whole body . A protein-trap line , Sec61alpha [ZCL0488] which constitutively expresses GFP-tagged Sec61alpha protein , was used as a control . To accumulate rhodopsin in the ER , flies were reared in the vitamin A-deficient medium in a Rh1-driven experiment . For heat-shock driven expression , newly eclosed adult fly flies were incubated at 37°C for 45 min a day before preparation . Within 0–1 days after eclosion , flies were frozen with liquid nitrogen and stored at −80°C . The heads were collected by sieving in liquid nitrogen , ground to powder and homogenized in buffer ( 50 mM Tris-Cl , 500 mM NaCl , pH 7 . 5 ) containing 1:200 Protein inhibitor cocktail VI ( Calbiochem , San Diego , CA , UAS ) using BioMasher II ( Wako Pure Chemical , Osaka , Japan ) with motor drive . Debris was removed by centrifugation at 950×g for 5 min and the membrane was precipitated by centrifugation at 21 , 500×g for 15 min . Approximately 30 µl of membrane pellet were solubilized by 130 µl of 1% CHAPS and placed on ice for 1 hr , and the insoluble membrane was removed by centrifugation at 21 , 500×g for 30 min . The extract was diluted fivefold by the buffer and 50 µl of Anti-GFP-Magnetic beads ( MBL , Nagoya , Japan ) were added and mixed by mild rotation for 18 hr . The magnetic beads were rinsed with 2× 100 μl of 0 . 1% CHAPS in buffer and the bound protein was extracted by incubation in 20 µl SDS-PAGE Sampling Buffer ( BioRad ) for 5 min at room temperature and an equal amount of Sampling Buffer with 2-mercaptoethanol was then added . The extracts were heat denatured for 5 min at 37°C . SDS-PAGE and immunoblotting was performed as described above . Electron microscopy was performed as described previously ( Satoh et al . , 1997 ) . Samples were observed on a JEM1200 or JEM1400 electron microscope ( JEOL , Tokyo , Japan ) . Whole-eye mutant clones were generated using the FRT/GMR-hid method ( Stowers and Schwarz , 1999 ) . Both eyes were dissected from two adult flies per sample and cDNA was reverse-transcribed using SuperPrep Cell Lysis and RT Kit for qPCR ( Toyobo , Osaka , Japan ) according to the manufacturer’s instructions . Eyes with whole-eye clones of FRT40A were used as a control to obtain the relative standard curves . qPCR reactions were performed using the StepOne real-time PCR system ( Life Technologies ) and KOD SYBR qPCR Mix ( Toyobo , Osaka , Japan ) , according to the manufacturers’ instructions . PCR condition was 98°C for 2 min , followed by 40 cycles at 98°C for 15 s , 55°C for 15 s , and 68°C for 45 s , and a melt curve stage of 95°C for 30 s , 60°C for 1 min , and 0 . 3°C/s increments to 98°C , with primers of Rh1: ( ninaE-qF1:5′-GTGGACACCATACCTGGTC-3′ and ninaE-qR1:5′-GCGATATTTCGGATGGCTG-3′ ) , Arr2: ( Arr2-qF1:5′-AAGGATCGCCATGGTATCG-3′ and Arr2-qR1:5′-TACGAGATGACAATACCACAGG-3′ ) , TRP: ( Trp-qF2:5′-GAATACACGGAGATGCGTC-3′ and Trp-qF2:5′-CTCGAGTTCCATGGATGTG-3′ ) , Act5C: ( 5′-GCTTGTCTGGGCAAGAGGAT-3′ and 5′-CTGGAACCACACAACATGCG-3′ ) . The relative expression levels were normalized by Act5C . | The membranes that surround cells contain many proteins , and those that span the entire width of the membrane are known as transmembrane proteins . Rhodopsin is one such transmembrane protein that is found in the light-sensitive ‘photoreceptor’ cells of the eye , where it plays an essential role in vision . Transmembrane proteins are made inside the cell and are inserted into the membrane surrounding a compartment called the endoplasmic reticulum . Here , they mature and ‘fold’ into their correct three-dimensional shape with help from chaperone proteins . Once correctly folded , the transmembrane proteins can be transported to the cell membrane . Incorrect folding of proteins can have severe consequences; if rhodopsin is incorrectly folded the photoreceptor cells can die , leading to blindness in humans and other animals . Experiments carried out in zebrafish have shown that the chaperone protein Pob is required for the survival of photoreceptor cells . Pob is part of a group or ‘complex’ of chaperone proteins in the endoplasmic reticulum called the EMC complex . This suggests that the EMC complex may be involved in folding rhodopsin , but the details remain unclear . Here , Satoh et al . studied the role of the EMC complex in the folding of rhodopsin in fruit flies . This involved examining hundreds of flies that carried a variety of genetic mutations and that also had low levels of rhodopsin . The experiments show that dPob—the fly version of Pob—and two other proteins in the EMC complex are required for newly-made rhodopsin to be stabilized . If photoreceptor cells are missing proteins from the complex , the light-sensitive structures in the eye degenerate . Rhodopsin is known as a ‘multi-pass’ membrane protein because it crosses the membrane multiple times . Satoh et al . found that the EMC complex is also required for the folding of other multi-pass membrane proteins in photoreceptor cells . The next challenge will be to reveal how the EMC complex is able to specifically target this type of transmembrane protein . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology"
] | 2015 | dPob/EMC is essential for biosynthesis of rhodopsin and other multi-pass membrane proteins in Drosophila photoreceptors |
Parkinson’s disease ( PD ) is a common neurodegenerative disorder without effective disease-modifying therapeutics . Here , we establish a chemogenetic dopamine ( DA ) neuron ablation model in larval zebrafish with mitochondrial dysfunction and robustness suitable for high-content screening . We use this system to conduct an in vivo DA neuron imaging-based chemical screen and identify the Renin-Angiotensin-Aldosterone System ( RAAS ) inhibitors as significantly neuroprotective . Knockdown of the angiotensin receptor 1 ( agtr1 ) in DA neurons reveals a cell-autonomous mechanism of neuroprotection . DA neuron-specific RNA-seq identifies mitochondrial pathway gene expression that is significantly restored by RAAS inhibitor treatment . The neuroprotective effect of RAAS inhibitors is further observed in a zebrafish Gaucher disease model and Drosophila pink1-deficient PD model . Finally , examination of clinical data reveals a significant effect of RAAS inhibitors in delaying PD progression . Our findings reveal the therapeutic potential and mechanisms of targeting the RAAS pathway for neuroprotection and demonstrate a salient approach that bridges basic science to translational medicine .
As the most common movement disorder and the second most common neurodegenerative disorder , Parkinson’s disease ( PD ) affects 0 . 3% of the general population , with a majority of cases being sporadic ( Obeso , 2017 ) . The hallmark of PD is a selective vulnerability of substantia nigra dopamine ( DA ) neurons among others ( Gonzalez-Rodriguez et al . , 2020 ) , accompanied by both motor and non-motor features , as well as prominent Lewy body pathology and mitochondrial dysfunction ( Bose and Beal , 2016; Tarakad and Jankovic , 2020 ) . As one of the world’s fastest growing neurological disorders , the economic cost of PD is estimated to be at least $51 . 9 billion a year in the United States ( Yang , 2020 ) . The current drug therapies available for PD provide only symptomatic relief by enhancing the dopaminergic action , decreasing metabolism of DA , or replacing the natural form of DA with exogenous drugs ( Haddad et al . , 2017 ) . While it provides benefits in improving motor symptoms in the short term , chronic therapy can result in motor fluctuation and dyskinesia . Furthermore , many patients experience a wearing-off effect where levodopa loses its efficacy even with dosing adjustments . Despite potentially promising efforts ( Dawson and Dawson , 2019 ) , there is yet to be a therapy that can halt or slow down disease progression . In recent years , the drug discovery pipeline for neurological diseases has been stagnant , with a phase I to approval rate being only 8 . 4% from 2006 to 2015 ( Wong et al . , 2019 ) which is lower than the average approval rate of all therapeutic indications . One of the reasons for such lack of success can be attributed to the fact that conventional approach of target-based drug discovery is difficult in the setting of neurological diseases because of the complex etiology and biological pathways involved . Phenotypic screening provides a promising opportunity . In particular , the whole organism-based drug discovery has been successfully applied to model organisms ( Szabo , 2017 ) . Larval zebrafish , being a vertebrate that can fit in 96-well plates , offers multiple advantages including genomic and anatomical conservation in addition to high throughput capabilities ( Guo , 2009 ) . Screens have uncovered therapeutic leads that are currently in clinical testing and/or shed light on biological mechanisms ( Macrae and Peterson , 2015; Zon and Peterson , 2005; Baraban et al . , 2013; Matsuda , 2018 ) . Here , we report a DA neuron-based neuroprotective drug discovery pipeline for PD , from assay development to small molecule screening , hit target validation , mechanisms , and ultimately , evolutionary conservation of neuroprotection across species including humans . Due to the late onset and variable phenotypic expressivity of genetic PD models and the highly toxic nature of neurotoxins ( e . g . MPTP ) to researchers , no good in vivo assay systems exist that are suitable for high content screening of neuro-protectives . We therefore first established an inducible chemogenetic DA neuron ablation model in larval zebrafish . This model expressed the E . coli nitroreductase ( NTR ) controlled by the promoter of tyrosine hydroxylase ( th ) , a rate-limiting enzyme in DA synthesis . Addition of the commonly used and safe-to-handle antibiotic , metronidazole ( MTZ ) , caused robust DA neuron loss . By showing that DA neuron loss is preceded by mitochondrial DNA damage and ensuing mitochondrial dysfunction , we demonstrated the validity and relevance of this model to PD for small molecule screening purpose . Using this model to screen >1400 bioactive small molecules , we uncovered a series of compounds that protected against DA neuron loss by inhibiting different proteins in the renin-angiotensin-aldosterone system ( RAAS ) , a pathway classically known for regulating vasoconstriction and water homeostasis ( Bader , 2010 ) . Genetic validation and molecular characterization revealed that the angiotensin receptor 1 ( AGTR1 ) acted cell autonomously in DA neurons , the inhibition of which restored the expression of mitochondrial pathway genes disrupted by neurotoxic insults . Furthermore , we showed that RAAS inhibitors were neuroprotective in a zebrafish 1-methyl-4-phenyl-1 , 2 , 3 , 6-tetrahydropyridine ( MPTP ) model and a zebrafish model of Gaucher disease , a lysosomal storage disorder with strong comorbidity of PD ( Riboldi and Di Fonzo , 2019 ) . The AGTR1 inhibitor olmesartan was also protective in a Drosophila pink1-deficient PD model ( Yang et al . , 2006 ) . Finally , utilizing the Parkinson’s Progression Marker Initiative ( PPMI ) database ( Marek , 2018 ) , we performed a clinical informatics analysis to uncover that RAAS inhibitors significantly slowed down PD progression . Together , our results delineate a powerful approach for neuroprotective small molecule drug discovery that leverages whole organism screening and cross-species validation .
Genetic PD models in rodents generally have weak , variable , and late onset degeneration phenotypes ( Dawson et al . , 2010 ) . Modeling neurodegeneration in zebrafish is a promising approach ( Paquet , 2009; Flinn , 2013; Ana Lopez , 2017; Zhang , 2017 ) , but neuroprotective screening based on direct imaging of DA neuron integrity has not been reported , in large part because high content screening needs assays that are sufficiently robust , sensitive , and scalable . Neurotoxins such as MPTP are highly toxic to experimenters and not scalable . We have therefore used an inducible chemogenetic DA neuron ablation model , employing the nitroreductase-metronidazole ( NTR-MTZ ) system ( Williams , 2015; Pisharath and Parsons , 2009; Curado et al . , 2008 ) : NTR was expressed as a transgene in tyrosine hydroxylase ( TH+ ) DA neurons to convert the pro-drug MTZ ( a commonly used antibiotic ) to the toxic nitroso radical form in vivo . Twenty-four hours ( hrs ) after adding MTZ to 5 days post fertilization ( dpf ) larval zebrafish , we observed at 6 dpf robust DA neuronal loss in the ventral forebrain region ( Figure 1A ) , the homologous group to mammalian substantia nigDA ( Rink and Wullimann , 2001 ) . The specificity of ventral forebrain DA neuron labeling in this transgenic line has been previously validated ( Liu , 2016 ) . Although the NTR-MTZ system has been used for cell ablation and noted to induce apoptotic cell death ( Chen et al . , 2011 ) , the underlying mechanisms of cell death are not well understood . Previous reports suggest that MTZ as an antibiotic targets bacterial DNA ( Edwards , 1979 ) . In vertebrate cells , two organelles containing DNA are the nucleus and the mitochondria . A semi-quantitative PCR assay based on the notion that DNA lesions block DNA polymerase progression ( Furda et al . , 2012; Yakes and Van Houten , 1997 ) , was used to detect nuclear and mitochondrial DNA integrity . Total DNAs were extracted from DA neurons in DMSO control and 4 . 5 mM MTZ-treated transgenic larvae ( 8 hr after MTZ treatment , when DA neurons remain morphologically intact ) . Anterior brains ( rostral to the mid-hindbrain boundary ) were dissected and acutely dissociated . The mCherry+ DA neurons were manually aspirated under a fluorescent microscope . PCR of equally long-length ( ~10 kb ) nuclear or mitochondrial DNA products was carried out using primers as previously described ( Furda et al . , 2012; Yakes and Van Houten , 1997 ) . This data uncovered a significant damage of mitochondrial DNA but not nuclear DNA in MTZ-exposed individuals ( Figure 1B ) . The nuclear DNA was unaffected possibly due to protection by nucleosomes; this observation also supports the notion that the time of our DNA integrity assessment precedes that of overt neurodegeneration . We next performed in vivo time-lapse imaging , which uncovered mitochondrial dysfunction in morphologically intact DA neurons after MTZ treatment . These include reduced mitochondrial number , increased mitochondrial length , decreased motility and velocity . Interestingly , in MTZ-treated DA neurons , mobile mitochondria moved exclusively in retrograde direction toward neuronal soma , suggesting that they are targeted for repair and/or mitophagy ( Figure 1C–H; Figure 1—videos 1; 2 ) . Together , mitochondrial defects are observed prior to DA neurodegeneration in the NTR-MTZ model , suggesting that mitochondrial DNA damage and ensuing mitochondrial dysfunction is likely a cause rather than a consequence of DA neuron degeneration . Given the observed mitochondrial deficits prior to neurodegeneration in the NTR-MTZ model , we next wondered whether enhancing the activity of genes functioning in mitochondrial quality control would protect against neurodegeneration in the model . Homozygous parkin mutations account for the majority of early onset autosomal recessive PD ( Kitada , 1998 ) . The parkin gene encodes a conserved E3 ubiquitin ligase that promotes mitochondrial quality control ( Pickrell and Youle , 2015 ) . The zebrafish Parkin protein is approximately 80% identical to the human counterpart in functionally relevant domains . We therefore synthesized mRNAs encoding full-length human parkin gene or EGFP ( control ) and micro-injected them into one-cell stage Tg[fuguth:gal4-uas:GFP; uas-NTRmCherry] embryos . Expressivity of microinjected mRNAs was verified by observing the GFP fluorescent reporter following egfp mRNA injection . Because of the short-lived nature of all microinjected mRNAs , we treated control or parkin mRNA-injected embryos with MTZ at an earlier stage , 30 hr post fertilization ( hpf ) , and subsequently imaged for mCherry at 50 hpf ( rather than treatment at 5 dpf and imaging at 6 dpf as shown in Figure 1A ) . This regimen of MTZ administration similarly led to DA neuron loss . Increased expression of Parkin significantly protected DA neurons ( Figure 1I ) . Moreover , increased expression of pink1 and DJ-1 , two other genes associated with mitochondrial quality control ( Pickrell and Youle , 2015 ) , was also neuro-protective ( Figure 1J–K ) . In the case of pink1 , the kinase dead mutant form failed to show protective effects , further validating the specificity of our assay system ( Figure 1J ) . Human α-synuclein ( either the wildtype or the A53T mutant form ) , associated with a dominant form of PD ( Polymeropoulos , 1997 ) , did not show any protection under the 9 mM MTZ treatment condition; mis-expression of the A53T mutant form of α-synuclein significantly worsened DA neuron integrity under the milder 4 . 5 mM MTZ treatment condition ( Figure 1L–M ) . This is consistent with the toxic nature of α-synuclein when mis-expressed . These results suggest that DA neuron degeneration in the NTR-MTZ model can be alleviated by enhancing the activity of mitochondrial quality control genes and aggravated by the activity of A53T mutant form of α-synuclein . We have previously described a high throughput in vivo brain imaging method in zebrafish larvae ( Liu , 2016 ) . Using this method and the established DA neuron ablation model as described above , we screened >1400 bioactive compounds ( SelleckChem ) that are part of the UCSF Small Molecule Discovery Center ( SMDC ) bioactive screening set . Many have validated biological and pharmacological activities , with demonstrated safety and effectiveness in preclinical and clinical research , and some are FDA-approved therapeutics . A dual-flashlight plot of Brain Health Score ( BHS ) and strictly standardized mean difference ( SSMD ) score were generated as previously described ( Liu , 2016 ) , in order to quantitatively document the effects of each compound on DA neuron integrity ( Figure 2A ) . In addition to the DMSO-positive control and MTZ-only negative control , N-acetyl cysteine ( NAC ) was also used as a positive control . NAC is an over-the-counter available supplement that works primarily by restoring body’s natural antioxidant glutathione for proposed DA improvement in PD ( Monti , 2019 ) . In our screen , NAC had an SSMD score of 0 . 826 . Around 100 out of 1403 compounds had the same or better SSMD scores than NAC ( ranged from 0 . 826 to 6 . 832 ) . Validation of these candidate hit compounds is ongoing . Among them , compounds that target different components of the renin-angiotensin-aldosterone system ( RAAS ) in particular caught our attention: First , compounds that target three different components of the RAAS signaling pathway ( Figure 2—figure supplement 1 ) were all among the top 100 candidate hits [e . g . Olmesartan ( angiotensin receptor 1-AGTR1 inhibitor ) with the SSMD scores of 1 . 649 - ranked 14 , aliskiren ( renin inhibitor ) ’s SSMD score was 1 . 540 - ranked 20 , imidapril ( ACE inhibitor ) ’s SSMD score was 0 . 938 - ranked 69] . Additionally , a Wilcoxon rank-sum test comparing all 13 RAAS pathway inhibitors from the primary screen with the entire screening library uncovered a significantly higher SSMD score for RAAS Inhibitors ( Figure 2B ) . Multiple hits targeting the same pathway thus built strong confidence in these compounds and in the involvement of RAAS pathway in DA neuron degeneration . Furthermore , literature search has uncovered reports of RAAS pathway inhibition for neuroprotection against PD and Alzheimer’s disease ( AD ) ( Grammatopoulos , 2007; Muñoz , 2006; Nelson et al . , 2013 ) . However , given the well-established role of these inhibitors in vascular remodeling and blood pressure control , it is conventionally thought that the neuroprotective effects of RAAS inhibition are due to their vascular actions , despite that RAAS pathway expression is detected in the CNS ( Wright and Harding , 2011 ) . It therefore remains unclear how RAAS inhibition might be neuroprotective . Finally , since RAAS inhibitors are commonly used anti-hypertensives , it offers an opportunity to investigate their role in PD patients via retrospective clinical data analysis . Taken together , we have therefore chosen RAAS inhibitors for in-depth analysis in this study . We next performed secondary hit validation for olmesartan , captopril ( another ACE inhibitor similar to imidapril ) , and aliskiren . all of them showed significant DA neuron neuroprotection , either by automated fluorescence intensity quantification ( Figure 2C–D ) or blinded manual neuronal counting ( Figure 2—figure supplement 2 ) . For dose-response curves of these compounds , we implemented dual imaging ( imaging both before and after MTZ treatment ) to further control for possible individual variability in DA neuron fluorescent intensity ( Figure 2—figure supplement 2B , D ) . With PD being highly associated with motor symptoms and occurring mostly in adult populations , we next determined whether RAAS inhibitors are capable of restoring motor function in adult Tg[fuguth:gal4-uas:GFP; uas-NTRmCherry] zebrafish treated with MTZ . Prolonged treatment of MTZ with or without RAAS inhibitors was performed over the period of two weeks in adult transgenic zebrafish , accompanied by locomotor behavioral tracking ( Figure 2E ) . As a positive control , we used levodopa , a gold standard symptomatic drug that can restore motor function in PD patients by increasing DA release from surviving neurons . Compared to vehicle controls , MTZ-treated animals showed a progressive decline of locomotive ability for the first 5 days post MTZ treatment and then reached steady low levels . Co-administration of levodopa 1 day after MTZ did not prevent initial locomotor decline , but was able to subsequently restore locomotor function , and interestingly , a hyper-locomotor state was reached at Day 14 . The chronic use of levodopa leading to hyperactivity is previously reported in mouse studies ( Gellhaar et al . , 2015 ) also , uncontrolled involuntary muscle movement such as dyskinesia is a common side effect of chronic levodopa use in humans ( Espay , 2018 ) . In contrast to levodopa , administration of the RAAS inhibitor olmesartan 1 day after MTZ fully ameliorated locomotor defects ( Figure 2F ) . The bioavailability of RAAS inhibitors in the brain and their ability to cross the blood brain barrier ( BBB ) are generally poor in humans ( Michel et al . , 2013 ) , but it is not known whether olmesartan enters the brain and the conversion of the pro-drug to its active carboxylate form takes place in zebrafish . We therefore performed mass spectrometry after 14-day treatment of adult zebrafish with olmesartan medoxomil ( the pro-drug form of olmesartan ) . The presence of active olmesartan was detected in the adult zebrafish brain ( Figure 2G ) . In addition to locomotor improvements , significant DA neuron protection was also observed in the adult brain setting ( Figure 2—figure supplement 2E-F ) . To verify whether the chemical inhibitors indeed target RAAS signaling components to exert their neuroprotective effects in zebrafish , we knocked down the angiotensin receptor 1 ( agtr1 ) gene activity . Two genes ( agtr1a and agtr1b ) encode agtr1 in vertebrates including zebrafish . We designed morpholino ( MO ) antisense oligonucleotides that inhibited protein translation ( ATG MO ) ( Figure 2—figure supplement 3A ) , and micro-injected them into one-cell stage embryos . Using an agtr1 antibody , we verified the protein knockdown in agtr1a and 1b double morphants at 6 dpf ( Figure 2—figure supplement 3B ) . The morphants appeared morphologically normal . At 5 dpf , control and morphants were treated with 9 mM MTZ for 24 hr . Significant DA neuron protection was observed in agtr1b and agtr1a&1b double morphants , the extent of which was comparable to olmesartan treatment ( Figure 2H , Figure 2—figure supplement 3C ) . These data suggest that inhibition of agtr1 protects against DA neuron degeneration . In our primary screen and secondary validations , prophylactic dosing was used ( i . e . screening compounds were added earlier than MTZ ) . In order to determine whether the neuroprotective benefits of RAAS inhibitors can be observed after MTZ treatment ( i . e . mimicking therapeutic dosing ) , we carried out MTZ ablation 8 hrs or 24 hrs prior to administering olmesartan ( Figure 2—figure supplement 4A ) . In both cases , olmesartan showed significant neuroprotection , suggesting that RAAS inhibitors can be beneficial even after the onset of neurotoxic insults ( Figure 2—figure supplement 4B ) . The RAAS pathway as a peptidergic system is composed of ligands and G-protein-coupled receptors ( GPCRs ) classically known to regulate blood pressure and salt retention ( Bader , 2010 ) . RAAS inhibitors are widely used drugs for treating high blood pressure . In recent years , RAAS signaling expression is detected outside vasculature and in the central nervous system ( Wright and Harding , 2011 ) . To understand the potential contribution of neuronal RAAS to DA neuron degeneration , we first performed qPCR on DA neurons purified by Fluorescence-Activated Cell Sorting ( FACS ) from the anterior brains of 5 day old Tg[fuguth:gal4-uas:GFP; uas:NTRmCherry] larvae ( Figure 3A; Supplementary file 1 ) . qPCR results uncovered enriched expression of pro-renin receptor , angiotensinogen , agtr1a , agtr1b , and Angiotensin Converting Enzyme ( ace ) , whereas the expression of renin and ace2 was undetectable in DA neurons compared to the rest of non-DA cells ( Figure 3B ) . To address whether RAAS inhibition is cell autonomously required in DA neurons for neuroprotection , we performed CRISPR-mediated conditional knockout ( Auer et al . , 2014 ) of agtr1a and agtr1b in DA neurons ( Figure 3C ) . Eight sgRNAs for each gene were designed and screened to identify those with high knockout efficiency , by microinjecting sgRNA and Cas9 protein into one-cell stage embryos followed by sanger sequencing and analysis using the ICE ( Inference of CRISPR Edits ) software ( Figure 3—figure supplement 1A , B ) . The Gal4-UAS system was then used to selectively express the GFP-tagged Cas9 enzyme under the control of th1 promoter . DNA constructs containing both the UAS-Cas9 and U6 promoter-driven high-efficiency sgRNAs were delivered into one-cell stage Tg[th1:gal4; uas:NTRmCherry] embryos , followed by quantification of DA neuronal integrity . The efficacy of agtr1a and agtr1b-targeting sgRNAs was verified by genotyping DA neurons isolated by manual aspiration ( Figure 3—figure supplement 1C ) . We found that DA neurons conditionally expressing Cas9 and effective agtr1a and agtr1b-targeting sgRNAs ( i . e . yellow neurons , expressing Cas9-GFP and NTR-mCherry ) were better preserved than those expressing control scrambled sgRNAs following MTZ treatment ( Figure 3D–E ) . These results suggest that inhibition of agtr1 in DA neurons is cell autonomously neuroprotective . Given that NTR-MTZ-induced DA neuron degeneration does not occur in human patient settings , we next tested whether RAAS inhibitors are neuroprotective in other models relevant to human PD . Mutations in the glucocerebrosidase ( gba1 ) gene cause Gaucher disease ( GD ) , the most common genetic risk factor for PD ( Riboldi and Di Fonzo , 2019 ) . The zebrafish genetic model for GD has a weak and later-onset DA neurodegeneration phenotype ( Keatinge , 2015 ) . Chemical inhibition of GBA using conduritol B-epoxide ( CBE ) has been successfully used to model the disease in both mice ( Vardi , 2016 ) and zebrafish ( Artola , 2019 ) . CBE exhibits some selectivity for GBA1 but can also inhibit lysosomal α-glucosidase ( GAA ) , non-lysosomal glucosylceramidase ( GBA2 ) , and lysosomal β-glucuronidase ( GUSB ) . We first observed that CBE dose-dependently reduced DA neuron integrity and locomotor activity in larval zebrafish , with 500 μM in the medium yielding significant results ( Figure 4—figure supplement 1 ) . We next tested whether olmesartan exerted a protective effect against CBE-induced DA neuron loss and locomotor deficit . Levodopa was used in comparison . While both levodopa and olmesartan ameliorated locomotor deficits induced by CBE ( Figure 4A–B ) , only olmesartan rescued TH immunoreactivity deficits induced by CBE treatment ( Figure 4C–D ) . Furthermore , CBE preferentially damaged TH neurons as revealed by the double immunofluorescent staining of TH and 5HT ( serotonin ) ( Figure 4C and E ) . We also tested whether RAAS inhibitors might protect against the neurotoxin and mitochondrial complex I inhibitor 1-methyl-4-phenyl-1 , 2 , 3 , 6-tetrahydropyridine ( MPTP ) . MPTP is a prodrug to MPP+ , and both have been shown to damage DA neurons in larval zebrafish ( Bretaud et al . , 2004; Lam et al . , 2005 ) . Zebrafish were therefore treated with 1 mM MPP+ ( dissolved in 0 . 02% Tween-80 to facilitate membrane penetration ) from 1 to 3 dpf , and RAAS inhibitors or vehicle ( 0 . 02% DMSO ) were administered 4 hrs prior to MPP+ treatment . Imaging of ventral forebrain DA neurons showed that MPP+ treatment significantly reduced DA neuron intensity , and treatment with Olmesartan and captopril significantly protected against DA neuronal loss ( Figure 4—figure supplement 2 ) . Together , these results demonstrate that RAAS inhibitors are not only neuroprotective in the synthetic NTR-MTZ model but is also neuroprotective in a Gaucher disease model and a MPP+ model . These findings reinforce the validity of the NTR-MTZ synthetic model for neuroprotective small molecule screening . To further understand the molecular basis underlying the neuroprotective effects of RAAS inhibitors , we carried out DA neuron-specific RNA-seq ( Figure 5A ) . Tg[th1:gal4; uas:NTRmCherry] larvae were treated with vehicle , CBE , MTZ , olmesartan , CBE+ olmesartan , or MTZ+ Olmesartan for a defined time window , followed by FACs purification of DA neurons from anterior brains and cell type-specific RNA-seq . MTZ and CBE models were theretofore referred to as the neurotoxic models . Upon annotating the sequence reads with the GRCz11 genome assembly , normalizing the read counts , and plotting all the significant gene expression changes ( α = 0 . 05 , FDR = 0 . 1 ) ( Figure 5—figure supplement 1 ) , we noted that the two neurotoxic models shared significant overlap and formed distinct clusters compared to the DMSO- or olmesartan alone control groups . Furthermore , treatment of both neurotoxic models with olmesartan restored transcriptomic expression to levels that were similar to controls , especially on the transcriptomes up regulated in the neurotoxic models ( Figure 5B ) . The expression of 1248 genes were commonly altered in the two neurotoxic models compared to vehicle controls ( Figure 5C ) , while the expression of 507 genes were commonly altered by olmesartan co-treatment in comparison to each of the neurotoxic insult alone ( α = 0 . 05 , FDR = 0 . 1 ) ( Figure 5D ) . The expression of RAAS pathway genes prorenin receptor ( PRR , gene name atp6ap2 ) , agtr1b , and ace2 were significantly upregulated in the MTZ treated group compared to the control ( padj = 0 . 001 , 0 . 032 , and 0 . 015 respectively ) . The atp6ap2 was also significantly upregulated in the CBE-treated group compared to the control ( padj <0 . 001 ) . Pathway enrichment analysis with the Reactome and KEGG pathway database showed 28 significantly altered pathways in the neurotoxic models when compared to vehicle controls ( p < 0 . 01 ) . Interestingly , the differentially expressed genes common to both neurotoxic models showed high significance ( LogP = –2 . 55 ) for the PD KEGG pathway ( ID: hsa05012 ) . These results further reinforce the notion that the neurotoxic models used in this study are relevant to PD . Cluster analysis of the gene ontology and pathways using g:Profiler , DAVID ( version 6 . 8 ) ( Huang et al . , 2009 ) , and Metascape ( Zhou , 2019 ) GO enrichment revealed distinct ontology clusters that were altered in both neurotoxic models compared to controls . Importantly , several pathways related to the mitochondrial function such as ATP synthesis , oxidative stress , and electron transport chain showed the highest significance values ( Figure 5E ) . Olmesartan treatment , when compared to the neurotoxic models , significantly affected the clusters related to mitochondrial function , including respiratory electron transport , oxidative phosphorylation , ATP metabolic process , and inorganic cation transport , ( Figure 5F ) . Given the prominence of mitochondrial pathway gene alterations in the neurotoxic models and by olmesartan , we further examined the molecular nature of these genes . As described earlier , preferential mitochondrial DNA damage was observed in DA neurons of the NTR-MTZ model prior to neuronal loss ( Figure 1B ) . There is also a strong link to mitochondrial dysfunction in lysosomal storage diseases ( Plotegher and Duchen , 2017 ) . Thus , disruption of mitochondrial gene expression is possibly causal to DA neuron degeneration in these models . The differentially expressed genes related to mitochondrial function were further divided into up-regulated and down-regulated categories ( Table 1 ) . Fourteen genes that were significantly up regulated in the neurotoxic models behaved oppositely upon olmesartan treatment ( genes highlighted in blue in Table 1 ) . They function in the mitochondrial electron transport chain ( e . g . Complex I , III , IV , and V ) and TOM-TIM complex critical for protein translocation through the mitochondrial membrane . One gene , trim3 , which was significantly down-regulated in the neurotoxic models , was up-regulated by olmesartan co-treatment ( highlighted in red in Table 1 ) . Trim3 ( Tripartite motif containing 3 ) , with reported ubiquitin ligase activity , is found to be down-regulated in PD patient plasma ( Dong , 2019 ) and can attenuate apoptosis via activating PI3K/AKT signaling pathway in PD models ( Dong , 2020 ) . Many of these mitochondrial pathways were no longer significantly altered when comparing the olmesartan+ CBE or olmesartan+ MTZ groups to the vehicle control group ( Supplementary file 1 ) . Taken together , these findings suggest that active AGTR1 receptor is necessary for upregulating the expression of mitochondrial electron transport pathway genes and downregulating trim3 in both neurotoxic models . Inhibiting its activity can help restore normalcy of these pathways , leading to neuroprotection . Drosophila offers a plethora of genetic PD models in which DA neuronal loss is evident ( West et al . , 2015; Lu and Vogel , 2009 ) . The conserved PINK1-Parkin pathway that directs mitochondrial quality control ( MQC ) has been originally delineated in flies ( Yang et al . , 2006; Yang et al . , 2003; Greene , 2003; Park , 2006; Clark , 2006 ) . These models have been used in genetic and pharmacological testing for genes and agents that offer neuroprotection ( Wang , 2006 ) . Although the RAAS pathway similar to vertebrates has not been fully described in Drosophila , genes encoding the angiotensin converting enzymes are detected in this species ( Coates , 2000 ) . Recently , it has also been reported that RAAS inhibitors rescue memory defects in a Drosophila Alzheimer’s disease model ( Lee et al . , 2020 ) . We therefore tested olmesartan in the Drosophila pink1 model , which recapitulates key features of PD including mitochondrial dysfunction , aberrant mitochondrial morphology , DA neuron and muscular degeneration . The most robust phenotype of the pink1 mutant flies is the degeneration of their indirect flight muscle caused by the accumulation of dysfunctional and morphologically aberrant mitochondria . This results in flies with collapsed thorax ( thoracic indentation ) and abnormal wing posture , manifested as droopy or held-up wings as opposed to the straight wings in control animals . Treatment of pink1 mutant flies by feeding them with food containing 100 μM olmesartan resulted in significant rescue of wing posture ( Figure 6A–C ) and the thoracic indentation ( Figure 6D–F ) . The abnormal mitochondrial morphology and neuronal loss phenotypes in DA neurons were also rescued ( Figure 6G–J ) . Collectively , these results suggest that olmesartan’s protective effect is conserved across species . Since RAAS inhibitors are commonly used anti-hypertensives , this provided us with an opportunity to ask whether the neuroprotective benefits of RAAS inhibitors shown in zebrafish and Drosophila can be observed in human PD patients . We used the Parkinson’s Progression Markers Initiative ( PPMI ) database , which includes a total of 423 de novo PD patients , 308 of which had complete data ( accurate medication and medical history records for each visit throughout the longitudinal study , no missing records on age , gender , duration of PD , and high visit compliance with no more than three missing records for motor assessment score ) . The de novo PD patients refer to subjects who have a diagnosis of PD for two years or less and are not taking any PD medications at the time of enrollment . Among them , 96 patients were on RAAS inhibitors ( RAAS ) while 212 patients were not ( non-RAAS ) . Among the non-RAAS cohort , 42 patients were hypertensive and taking other medications such as calcium channel blockers or diuretics for the management of hypertension ( Figure 7A ) . Using this dataset , we sought to compare PD progression in patients on RAAS inhibitors to those who were not . At present , there are no accepted progression biomarkers for PD ( Espay et al . , 2017 ) . The Unified Parkinson’s Disease Rating Scale ( UPDRS ) , while widely utilized , suffers from limitations including subjectivity and ambiguities in the written text ( Movement Disorder Society Task Force on Rating Scales for Parkinson’s Disease , 2003 ) . Because of the wear-off and debilitating side effects of levodopa after prolonged use ( Espay , 2018 ) , clinicians delay the prescription of levodopa to PD patients until absolutely necessary for treating debilitating motor symptoms . Therefore , we used the Time-to-Levodopa as a quantifiable and objective parameter to measure disease progression . After propensity score matching between RAAS and no-RAAS cohorts ( Figure S8A-C ) , our analysis found that the patients on RAAS inhibitors had a significantly delayed onset of levodopa therapy compared to the patients not on RAAS inhibitors ( difference , –5 . 8; 95% CI –11 . 26 to –0 . 4254; p = 0 . 035 ) ( Figure 7B ) . To control for hypertension as a variable , we compared patients on RAAS inhibitors to those on other classes of anti-hypertensive medications . This analysis also uncovered a significant effect of RAAS inhibitors in delaying the onset of levodopa therapy as shown in the Kaplan Meier curve ( p = 0 . 032 ) ( Figure 7C ) . The UPDRS scores part I , II , and III were also examined for a subset of patients within each cohort whom were levodopa-naive for at least 3 years; this subset of patients was chosen since levodopa use can significantly influence the UPDRS scores . The UPDRS part I score , which examines the mentation , behavior , and mood , showed a significantly lower score in the cohort on RAAS inhibitors compared to the cohorts not on RAAS inhibitors or those taking other hypertensive medications over the course of 5 years ( difference: NO RAAS vs RAAS = 0 . 289 , Other HTN vs RAAS = 0 . 266 , p = 0 . 017 ) ( Figure 7D ) . The UPDRS part 2 and part 3 , which examine the activities of daily living and motor function respectively , did not show significance in patients taking RAAS inhibitors when compared to other cohorts ( p = 0 . 82 ) ( Figure S8E-F ) . Taken together , these data suggest that inhibition of RAAS signaling slows down disease progression in human PD patients .
For most neurodegenerative disorders , there exist no disease-modifying therapeutics . Here , we validated the NTR-MTZ-based chemogenetic DA neuron ablation model in zebrafish by showing that DA neuronal loss is preceded by preferential mitochondrial DNA damage and ensuing mitochondrial dysfunction . We then used this system to conduct a high-content DA neuron imaging-based chemical screen and identified the inhibitors of Renin-Angiotensin ( RAAS ) system to be significantly neuroprotective via cell-autonomous effects of the angiotensin receptor one in DA neurons . Dopamine neuron-specific RNA-seq further revealed the molecular action of RAAS signaling in mitochondrial gene regulation . Finally , inhibition of RAAS signaling was neuroprotective across species . Together , this study identifies RAAS inhibitors as promising therapeutics for slowing down PD progression and highlights a new approach composed of high content screening in zebrafish , cross-species validation , and examination of human clinical data to uncover previously unrecognized neuroprotective agents and underlying mechanisms . Although MTZ has historically been widely prescribed as an antibiotic , the underlying mechanism of cell death in vertebrates has however remained elusive . By investigating the integrity of mitochondrial and nuclear DNAs , performing live imaging of mitochondria , and DA neuron-specific RNA-seq , we link the mode of NTR-MTZ-mediated cell death to mitochondrial dysfunction for the first time to our knowledge . This property , together with its safe and scalable nature , makes the NTR-MTZ-mediated cell ablation model in larval zebrafish a valuable small molecule discovery tool for disorders where mitochondrial dysfunction is a prevalent underlying pathophysiological mechanism . Despite the mechanistic relevance of the NTR-MTZ model to PD in the context of mitochondrial dysfunction , this model has several limitations , including not being able to recapitulate the etiology of PD and the time course of neurodegeneration . Given these limitations , it is advisable that additional validation in etiologically relevant models as we have done in this study are to be carried out . Our unbiased small molecule screen together with cross-species validations has revealed that inhibition of RAAS signaling is significantly neuroprotective for DA neurons in the context of animal PD models and moreover human PD patients . Hits that target different components of the RAAS pathway ( e . g . renin , ACE , and agtr1 ) were uncovered from the primary screen and validated in secondary screening . The AGTR1 inhibitor olmesartan was further shown to be neuroprotective in adult zebrafish , in a zebrafish GD model and a MPP+ model , and Drosophila pink1-deficient model . The GD model was created by inhibiting the GBA protein with a chemical inhibitor CBE , which preferentially inhibit GBA1 but can also inhibit other related enzymes including GBA2 . Despite this limitation , CBE treatment has been previously shown to recapitulate disease phenotypes in both mice and zebrafish ( Vardi , 2016; Artola , 2019 ) . This chemically induced GD model presents several advantages over the genetic model of GD ( Keatinge , 2015 ) : First , it shows significant DA neuronal loss at larval stages , whereas the genetic GD model only exhibit weak and variable deficit of DA neurons at adult stages . Second , it can be conveniently combined with transgenic lines that label DA neurons . Its conditional nature further allows us to gain temporal control over the access of DA neurons prior to degeneration . The finding that RAAS inhibitors are also neuroprotective in the Drosophila pink1-deficient model suggests that the underlying neuroprotective mechanisms are deeply conserved across evolution . Intriguingly , literature search has uncovered reports of RAAS inhibitors’ neuroprotective effects in various animal models of neurodegeneration , but the mechanisms were not well characterized in these studies ( Grammatopoulos , 2007; Muñoz , 2006 ) . Our findings corroborate with these reports and further demystify the actions of RAAS inhibitors , by showing that agtr1a and 1b act cell autonomously in DA neurons beyond their conventional actions on the vasculature systems . The DA neuron-specific RNA-seq has identified mitochondrial pathway genes and trim3 , the expression of which was perturbed in the neurotoxic models and restored by RAAS inhibition . While the up-regulation of mitochondrial electron transport chain gene expression in the neurotoxic models appears counter-intuitive to the concept of neurodegeneration as a consequence of bioenergetic crisis , it is worth noting that the RNA-seq was carried out prior to overt neurodegeneration . Although upregulation of electron transport chain gene expression could be a compensatory response to mitochondrial damage or dysfunction , it could also lead to dysregulated metabolic pathways resulting in neurodegeneration ( Area-Gomez et al . , 2019 ) . Future experiments to alter the expression of these genes either individually or in combination followed by evaluating DA neuron states will help further verify the cause-effect relationships . Several lines of evidence suggest that RAAS signaling via active agtr1 may play a direct role in promoting neurodegeneration via disrupting mitochondrial gene regulation: First , among the pathways that are commonly altered in morphologically intact DA neurons from both neurotoxic models , those that regulate mitochondrial function have the highest significance values . The ~40 mitochondria-related genes were mostly nuclear genes that encode proteins ranging from electron transport chain subunits such as NADH ubiquinone , cytochrome-c oxidase , and ATP synthase to mitochondrial translocation machinery . The notion that olmesartan restored normalcy to a large extent suggests that AGTR1 is the culprit in mediating these gene expression changes . Second , AGTR1 expression is detected in the mitochondria of a variety of cell types including DA neurons ( Valenzuela , 2016; Abadir , 2011 ) , lending support that it may have a direct role in regulating mitochondrial function . Indeed , AGTR1 is reported to promote reactive oxygen species production and activate MAP kinase pathway that leads to the activation of transcription factors including NF-kB and AP-1 , and P53 in the context of vascular senescence ( Min et al . , 2009 ) . Intriguingly , significant upregulation of AP-1 was observed in both neurotoxic models , suggesting the following possible model: Neurotoxic insults activate AGTR1 , which activates kinase signaling cascades that upregulate the expression of transcription factors such as AP-1 , in turn increasing the expression of mitochondrial electron transport chain genes . Such altered mitochondrial gene expression is associated with increased reactive oxygen species production leading to mitochondrial dysfunction and neuronal death . Future follow-up studies on the genes and pathways discovered in this cell type-specific RNA-seq dataset shall provide deeper insights into how active AGTR1 perturbs mitochondrial function and aggravates neurodegeneration . The evolutionarily conserved actions of RAAS inhibitors together with their prevalent use for anti-hypertension in PD patients prompted us to examine and subsequently discover their significant effect in slowing down PD progression . Previous studies that interrogate electronic health records ( EHR ) data reported mixed results regarding the effect of RAAS inhibitors on the incidence of PD ( Lee et al . , 2014; Warda et al . , 2019 ) . Given the diverse and complex etiology of PD , this is not surprising . In this study , rather than evaluating the incidence of PD , we focused on PD progression using the Time-to-Levodopa therapy as an innovative criterion in addition to the commonly used UPDRS scores . A significant effect of RAAS inhibitors was detected in delaying the time to levodopa therapy . This marker for disease progression can also be applied to other EHR or clinical data where exam metrics are incomplete due to inadequate hospital protocols , or text mining is difficult due to variations in note taking practices by healthcare workers . Our sample size of 308 PD patients is relatively modest . It would therefore be of interest to further expand this analysis to a larger patient population . It is also worth noting that the blood-brain-barrier ( BBB ) permeability of currently available RAAS inhibitors in humans vary from compounds to compounds and are generally poor for compounds such as olmesartan and losartan ( Michel et al . , 2013; Unger , 2003 ) . This may contribute to the modest neuroprotective effects observed in the clinical data ( e . g . , UPDRS part 2 and part three scores showing no significant improvements ) . With expanded patient population , it may be possible to evaluate and compare the BBB profile of RAAS inhibitors and their extent of neuroprotection . Given the cell autonomous mechanisms that we have discovered through animal studies , we postulate that RAAS inhibitors with better BBB penetrating ability will possibly have a higher neuroprotective effect . RAAS pathway components are broadly expressed in the CNS , suggesting that its inhibition could be broadly neuroprotective , not only for PD , but also for other neurodegenerative diseases .
This study was designed to identify neuroprotective small molecules for Parkinson’s disease ( PD ) . A chemo-genetic DA neuron degeneration model employing the NTR-MTZ system was first characterized to uncover mitochondrial dysfunction as a plausible cause of cell death . A whole organism DA neuron imaging-based small molecule screen employing such transgenic zebrafish was then carried out . By screening 1 , 403 bioactive small molecule compounds , the RAAS pathway inhibitors were identified to be significantly neuroprotective . Their neuroprotective actions were further validated in multiple animal models and in human PD patients . Cell type-specific CRISPR and RNA-seq revealed a DA neuron-autonomous regulation of mitochondrial function as a mechanism underlying the neuroprotective effects of RAAS inhibitors . In vivo studies employing zebrafish were approved by the Institutional Animal Care Use Committee at University of California , San Francisco ( Approval Number: AN179000 ) . Use of patient data in the PPMI database was approved by the Michael J Fox Foundation PPMI Data and Publications Committee . No statistical methods were used to predetermine sample size . The sample size ( n ) for each experimental group was indicated in the figure legends . The compound treatment , image collection , and data analysis , for the compound screening , manual counting for secondary hit validation of RAAS inhibitors , western blot of morpholino injections , mass spectrometry of adult fish brains , and adult zebrafish behavior studies were performed in a blinded manner . For all other experiments , the investigators were not blinded to allocation during experiments and outcome assessment . All the experiments were replicated at least two independent times . Zebrafish were raised on a 14:10 hr light/dark cycle and maintained in the zebrafish facility according to the University of California San Francisco Institutional Animal Care and Use Committee standards . Embryos were raised in Blue Egg Water ( 2 . 4 g CaSO4 , 4 g IO Salt , 600 μl of 1 % Methylene per 20 L ) . The following transgenic lines were used: Tg[fuguth:gal4-uas:GFP; uas-NTRmCherry] ( for in vivo drug screening , hit validation , MO injection , and behavioral assessment ) ( Liu , 2016 ) ; Tg[UAS:mtPAGFP:mtDsRed2] for imaging mitochondrial dynamics , kindly provided by Dr . Edward Burton’s lab ( Dukes , 2016 ) ; Tg[th1:gal4; uas:NTRmCherry] ( for CBE double immunofluorescence staining of TH and 5HT , conditional CRISPR knockout of agtr1a and agtr1b , DA neuron specific RNA-seq ) . Tg[th1-gal4] was kindly provided by Dr . Jiulin Du’s lab ( Li , 2015 ) . Morpholino ( MO ) antisense oligonucleotides that inhibit protein translation were designed for agtr1a and agtr1b ( Figure 2—figure supplement 3A ) and purchased from Gene Tools , LLC . 0 . 5 mM agtr1a and agtr1b MO working solution was mixed with 1% phenol red and micro-injected into 1 cell stage embryos ( estimated 1–4 nls per embryo ) . At 5 dpf , control and morphants were treated with 9 mM MTZ for 24 hr and confocal imaging was performed with brightfield and DsRed channel at 6dpf . Eight-bit images were cropped to isolate the diencephalic region of the brain and the DA neuron intensity was quantified with ImageJ . For western blotting , the 6dpf larvae with DMSO or MTZ treatment were collected after performing confocal imaging . Thirty larvae for each group were homogenized in 150 uL of SDS sample buffer and boiled for 10 minutes at 99°C and transferred to ice . The samples were centrifuged for 1 min at 12 , 000 rpm and the supernatant was transferred to a new tube with 5 x SDS protein loading buffer . The samples were loaded into Mini-PROTEAN TGX Gels ( cat# 4561083 ) and run at 180 V for 50 min . Transblotting was done using the Trans-Blot Turbo Transfer System ( cat# 1704150 ) and washed with PBS . Primary antibodies were incubated at 4°C overnight . For the anti-rabbit agtr1 antibody ( Proteintech 25343–1-AP ) , 1:500 was used; for the anti-mouse beta actin control ( Sigma A5441 ) , 1:2000 dilution was used . Horseradish Peroxidase conjugated secondary antibodies were used ( Abcam ab6721 and ab6728 ) with 2 hr incubation . After washing off the secondary antibodies with PBS , the western blot was visualized with the iBright CL750 Imaging System ( Invitrogen A44116 ) . The expected bands of 37 kda for anti-beta actin and 50 kda for anti-AGTR1 were identified and analyzed with imageJ using the ‘Mean Grey Value’ measurement tool . The NTR/MTZ model with NTR expressed in DA neurons was used for drug screening , secondary hit validation , and mechanistic studies of DA neuron degeneration . 4 . 5 mM , 9 mM , or 10 mM MTZ were used in larval zebrafish with varying time courses ( ranging from 8 to 48 hr ) to achieve different goals ( e . g . pre- , mild , moderate , or severe DA neuron loss ) . Five mM MTZ was used for prolonged treatment in adult zebrafish . Drug screening was performed in 96-well plates with the bioactive compound library from SelleckChem obtained from the UCSF Small Molecule Discovery Center ( SMDC ) . Ten µM of compounds were dissolved in blue egg water containing 0 . 2% DMSO for a total volume of 200 µL . Tg[fuguth:gal4-uas:GFP; uas:NTRmCherry] were treated with 200 µM 1-phenyl 2-thiourea ( PTU ) on 1dpf and at 3dpf , larvae were transferred to 96-well plates containing the screening compounds or 0 . 2% DMSO ( positive control ) . Four hr later , 4 . 5 mM MTZ was added to compound-containing wells as well as wells that serve as negative control . Treatment lasted for 48 hrs . At 5dpf , the larvae were imaged with brightfield and TexRed channels . The multi-pose method ( Liu , 2016 ) was used to image DA neurons in vivo using In Cell Analyzer 2000 . The images were analyzed using a custom CellProfiler ( Jones , 2008 ) pipeline that masks the eyes and auto-detects the DA neurons to calculate the Brain Health Score ( BHS ) and SSMD score as previously described ( Liu , 2016 ) . In brief , we compared automated methods for neuronal counting vs . total fluorescence intensity measure , and have found them to be strongly correlated , with the latter revealing the most significant difference between positive control ( vehicle-treated ) and negative control ( MTZ-treated ) . This is likely because neurodegeneration is often initiated at the level of neuronal processes , followed by the loss of cell bodies . Therefore , we have chosen to quantifying fluorescence intensity in our study . For secondary hit validations , 40 µL of 1 . 5% agarose was added to ensure that the larvae were embedded in a dorsal down position for confocal imaging before and after MTZ treatment . The live confocal imaging was conducted using In Cell Analyzer 6000 with DsRed and FITC channels with 200 ms exposure time . For manual counting of DA neurons , experimenters were blinded to the treatment conditions . Individual larval zebrafish were mounted in 1 . 5% agarose . Ventral forebrain DA neurons were observed under a Zeiss epi-fluorescent microscope and counted on one side of the brain in an identical manner across all larval zebrafish . For the dose response studies , concentrations of the RAAS inhibitors were prepared from a series of fold dilutions . The RAAS pathway inhibitors used in the study including olmesartan , aliskiren , captopril , were purchased from Sigma-Aldrich ( cat #144689-63-4 , 62571-86-2 , 173334-58-2 ) . Metronidazole and NAC were purchased from Selleck Chemicals ( cat# S1907 , S1623 ) . CBE was purchased from Sigma Aldrich ( cat# 6090-95-5 ) . MPP+ was purchased from Millipore Sigma ( cat # 36913-39-0 ) . For all adult and larval behavior assays , animals were individualized and incubated in their home tanks in a 26°C behavior room overnight for habituation . Six-well plates were used to house individual larva in each well with 5 mL of total volume per well . The wells were placed on a lightbox and the videos were recorded from a top-down view . For the adult behavior experiments , the fish were individually housed in their home tank with 500 mL of system water . For the 2-week duration of the adult behavior test , they were fed with flakes in the morning and replaced with fresh water containing the test compounds daily . The recordings were taken from the top view for 5 min . The total distance moved for the 5-min duration was analyzed through the EthoVision XT software using the dynamic subtraction algorithm with detection limits between 10 and 100 pixels . For larval fish , the static subtraction algorithm was used with detection limits between 10 and 40 pixels . Tg[fuguth:gal4-uas:GFP; uas-NTRmCherry] were treated with PTU ( 1:100 ) ( 200 µM ) on 1dpf . At 5dpf , the larvae were treated with 4 . 5 mM MTZ for 8 hr and immediately transferred to HBSS ( Ca/Mg Free ) Buffer ( Gibco 14170120 ) and the brains anterior to the mid/hindbrain boundary were acutely dissected and dissociated with TrypLE ( Gibco 12604013 ) for 30 min . DA neurons were collected via mouth pipetting and the genomic DNA was extracted using extraction buffer ( 10 mM Tris pH 8 . 2 , 10 mM EDTA , 200 mM NaCl , 0 . 5 % SDS , 200 µg/ml proteinase K ) . The nuclear DNA was PCR amplified using the primer sequences: Forward 5’ to 3’ AGAGCGCGATTGCTGGATTCAC , Reverse 5’ to 3’ GTCCTTGCAGGTTGGCAAATGG and the mitochondrial DNA was PCR amplified using the primer sequences: Forward 5’ to 3’ TTAAAGCCCCGAATCCAGGTGAGC , Reverse 5’ to 3’ GAGATGTTCTCGGGTGTGGGATGG . The target base pair sizes are 10 . 7 kb and 10 . 3 kb , respectively . The PCR was performed with the QIAGEN Long-Range PCR Kit ( cat# 206402 ) optimized for long-range amplification of genomic DNA . The PCR was performed with an initial denaturation step at 94°C for 1 min , 24 cycles ( nuclear DNA ) or 19 cycles ( mitochondrial DNA ) of 94°C for 15 s , 69°C for 45 s , and 72°C for 30 s , with final extension at 72°C for 10 min . The DNA integrity was evaluated by gel electrophoresis ( 2% agarose ) and the bands were analyzed with ImageJ using the ‘Calibrate’ function to determine the optical density of the molecular weight standard , the nuclear DNA , and mitochondrial DNA bands . Transgenic zebrafish Tg[th1:gal4; uas:NTRmCherry] were crossed with Tg[UAS:mtPAGFP:mtDsRed2] and treated with PTU ( 1:100 ) ( 200 µM ) at 1dpf . The larvae were screened for th1-NTRmCherry on 4dpf and were treated with either 0 . 2% DMSO ( control ) or 4 . 5 mM MTZ for 16 hr . The larvae were embedded with 1% low melting point agarose ( Sigma 39346-81-1 ) ( 1:100 tricaine added , 0 . 168 µg/mL ) in 35 mm glass bottom dishes ( Corning ) . The PA-GFP was activated with the Nikon 40 x WI objective DAPI channel for 1 min . Upon successful activation , the mitochondria were observable under GFP . Live imaging was performed with 10 s intervals for a total of 10 min . The imaging movies were processed with ImageJ and IMARIS software ( version 9 . 7 ) where the xyz coordinates of the mitochondria movements were obtained . The values were exported to a custom MATLAB script to calculate total displacement , velocity , and direction . Zebrafish were treated with 10 μM of olmesartan medoxomil ( Sigma Aldrich cat# 144689-63-4 , the pro-drug form of olmesartan ) for 14 days . The drug was freshly dissolved in the system water and administered daily . On day 14 , adult zebrafish were dissected to collect the body and the brain which were then pooled to obtain approximately 125 mg per sample ( n = 10 males , 10 females ) . Brain samples were prepared by addition of PBS at a 1:5 ratio and homogenizing ( Bertin Precellys 24 ) . Homogenates were mixed with acetonitrile and methanol ( 1:1 v/v ) containing 0 . 05 µg/mL niflumic acid as an internal standard before filtering with Captiva ND plates ( 0 . 2 um ) into water and analyzed for the active olmesartan ( Sigma Aldrich cat# 144689-24-7 ) with a QTRAP 5500 tandem mass spectrometer ( Sciex ) coupled with a Nexera X2 series UHPLC ( Shimadzu ) . Zebrafish treated with CBE ( a chemical inhibitor of GBA ) and the RAAS inhibitor olmesartan were tested for both locomotor behavior and confocal imaging of DA neurons . Initially , CBE concentrations ranging from 100 μM , 500 μM , and 1 mM were used to treat embryonic and larval zebrafish from 1dpf to 5dpf with fresh compounds dissolved in Blue Egg Water changed daily to determine that 500 μM is the optimal concentration for the study ( Figure 4—figure supplement 1 ) . Prior to treatment with CBE , olmesartan , or levodopa , 5dpf larvae were embedded in 96-well glass bottom plates with 1 . 5% agarose and blue egg water and imaged using a 20 x objective under the InCell 6000 confocal microscope . CBE ( 500 μM ) , olmesartan ( 10 μM ) or levodopa ( 500 μM ) were added to the agarose-embedded larvae . Twenty-four hr later , the 6 dpf larvae were again imaged and the before vs . after TH intensity was quantified using ImageJ . For behavioral recording , 5 dpf larvae were treated with CBE , olmesartan , or levodopa for 24 hr in six-well plates and behavior was analyzed using Ethovision XT using the methodologies described above . Treatment conditions were used similar to what has been previously described ( Lam et al . , 2005 ) . In brief , 10 μM olmesartan in 0 . 02% DMSO was administered to 1 dpf zebrafish . Four hr later , 1 mM MPP+ in 0 . 02% Tween-80 was added . The treatment lasted till 3 dpf . Positive control was treated with vehicle only ( 0 . 02% Tween-80 , 0 . 02% DMSO ) , and negative control was treated with MPP+ only . DA neurons were imaged and quantified as described above . Newly eclosed PINK1B9; TH-Gal4> UAS-mito-GFP male flies were raised on instant fly food ( Carolina ) or instant fly food containing 100 μM olmesartan . Flies were transferred to fresh vials daily . After two weeks , the flies were scored for wing posture or examined under dissecting microscope for thoracic indentation . Afterwards , flies were dissected for DA neuron staining . At least seven individuals were examined for each condition . Dissected brain tissue samples were briefly washed with 1 x PBS and fixed with 4% formaldehyde in 1 x PBS containing 0 . 25% Triton X-100 for 30 min at room temperature . Fixed samples were subsequently blocked with 1 x PBS containing 5% normal goat serum and incubated for 1 hr at room temperature followed by incubation with primary antibodies at 4°C overnight . The primary antibodies used were: chicken anti-GFP ( 1:5000 , Abcam ) , and rabbit anti-TH ( 1:1000 , Pel-Freez ) . After three washing steps with 1 x PBS/0 . 25% Triton X-100 each for 15 min at room temperature , the samples were incubated with Alexa Fluor 594- and Alexa Fluor 488-conjugated secondary antibodies ( 1:500 , Molecular Probes ) for 3 hr at room temperature and subsequently mounted in SlowFade Gold ( Invitrogen ) . Samples were observed under a Leica SP8 confocal microscope and fluorescent confocal images were processed using Photoshop . FACs was performed on the BD FACSaria III Cell Sorter with 488 nm , 561 nm and 638 nm channels . To ensure high accuracy of cell sorting , DA neurons were sorted with both 488 and 561 nm channels from Tg[fuguth:gal4-uas:GFP; uas:NTRmCherry] . The dead cells stained with DAPI ( 1 ng/mL ) were sorted with the 405 nm channel . The FACs-sorted cells were immediately processed for RNA extraction ( Ambion ) and converted to cDNA for qPCR . Transcript sequences were obtained from the Ensembl genome browser for zebrafish ( GRCz11; https://uswest . ensembl . org/Danio_rerio/Info/Index ) as shown in column 6 . The primers were designed with NCBI primer blast ( https://www . ncbi . nlm . nih . gov/tools/primer-blast/ ) spanning a product length between 70bp to 200 bp while minimizing self 3’ complementary score . All primers were validated with gel electrophoresis prior to qPCR . The Ct values were compared relative to eef1a1 as a housekeeping Gene ( Supplementary file 1 ) . For RNA-seq , approximately 500 DA neurons were collected per sample with biological triplicates using the FACs procedure described above . RNA was extracted using the Lexogen SPLIT RNA extraction kit ( cat 008 ) and the quality was assessed in the Agilent 2 , 100 Bioanalyzer ( cat# G2939BA ) . The library was prepared using the Lexogen QuantSeq 3’ mRNA-Seq Library Prep Kit FWD for Illumina ( cat# 015 ) . The libraries were quality controlled using Agilent 2100 Bioanalyzer and pooled at 20 mL of 3 ng/mL concentration . The RNA-seq was performed on the Illumina HiSeq 4000 ( cat SY-401–4001 ) , with single end 50 bp , generating 350 million reads per lane . FastQC was performed for quality check and all sequences showed high per base sequence quality with greater than 75% uniquely mapped reads aligned against GRCz11 ( Figure 5—figure supplement 1A ) . The count normalization was performed using the DESeq2’s median of ratios method which accounts for sequencing depth and RNA composition . This normalization method allows for gene count comparisons between samples , which is suitable for comparing differential gene expression across different sample groups with high sensitivity and specificity ( Li et al . , 2020; Bullard et al . , 2010 ) . To visualize the similarity of our samples , initially a sample-level QC was performed using Principal Component Analysis ( PCA ) as shown in Figure 5—figure supplement 1B . Each dot represents a sample from the respective group . The raw counts for each gene were modeled and the log2 fold changes were shrunken and differential gene expression analysis was performed using DESeq2 in R with an α level of 0 . 05 and FDR of 0 . 1 . The gene set was then annotated and converted from the zebrafish ensemble Gene ( ENSG ) to Homo sapiens ENTREZ gene ID with gProfiler . The pathway analysis was conducted on DAVID for the KEGG pathway maps and Metascape for enriched ontology clusters . Antibody staining was performed as previously described ( Yang et al . , 2012 ) . In brief , 6 dpf larvae were fixed in 4% PFA overnight , washed with PBS , and their brains were dissected . After 24 hr of dehydration in 100 % methanol overnight followed by rehydration , the samples were incubated in primary antibodies for 72 hr at 4°C with the rabbit anti-5HT ( Immunostar cat#20080 ) and mouse anti-TH ( Immunostar cat#22941 ) primary antibodies . The brains were then subjected to secondary antibody labeling , using Alexa Fluor 488 anti-mouse ( cat A-11001 ) and Alexa Fluor 568 anti-rabbit ( cat A-11036 ) . The brains were stored in 75% Glycerol and mounted for confocal imaging . The confocal imaging was taken on the Nikon Ti inverted fluorescence microscope with CSU-W1 large field of view using Apo LWD 40 x/1 . 15 water immersion lens under GFP and RFP channels . The sgRNA sequences were predicted and designed based on the CHOPCHOP and CRISPRscan database ( Figure 3—figure supplement 1A ) . The sgRNAs were synthesized and co-injected with Cas9-LS protein ( UC Berkeley , https://qb3 . berkeley . edu/facility/qb3-macrolab/ ) into one-cell stage embryos . The genomic DNAs were extracted and sequenced . Upon different designs of sgRNAs ( eight for each gene ) , a maximal knockout efficiency of 58% and 65% were obtained for agtr1a and agtr1b respectively , based on the analysis with the Synthego ICE software ( version 2 . 0 , https://ice . synthego . com/#/ ) . The plasmid backbone used for the conditional knockout construct was the Tol2-pUAS:Cas9T2AGFP;U6:sgRNA1;U6:sgRNA2 , in which UAS elements drive the expression of Cas9 and GFP linked via the T2A peptide and two sgRNA targets can be simultaneously used . The BsaI and BsmBI restriction sites were used for the sgRNA target sequence cloning as previously described ( Auer et al . , 2014 ) . After cloning , the obtained plasmid construct was co-injected with Tol2 transposase mRNA ( Kawakami et al . , 2000 ) into one-cell stage embryos derived from Tg[th1:gal4; UAS:NTRmCherry] . To validate successful knockout of the genes , after live imaging of DA neurons under both GFP and RFP channels , the zebrafish brains were dissociated with TrypLE Express for 30 min and mouth-pipetting was used to collect the GFP+NTRmCherry+ DA neurons for PCR and sequencing ( Figure 3—figure supplement 1B ) . The PPMI is an observational clinical study providing a comprehensive database for clinical , imaging , and biological data . The PPMI repository contains clinical data with subject demographics , comprehensive medication history , UPDRS motor assessments , and non-motor assessments . The data were downloaded and accessed in May 2019 . Initially 423 de novo PD patients , defined as subjects with a diagnosis of PD with two years or less who are not taking any PD medications , were identified , in which 115 patients had missing information or withdrew from the study making the total included subjects to be 308 patients . Among the 308 patients , 96 patients were taking either ACE inhibitors or ARBs ( RAAS group ) while 212 patients were not taking ACE inhibitors or ARBs ( no RAAS group ) . Based on the medication history , the average Time-to-Levodopa was compared between the two groups . Among the 212 patients , 42 patients had a diagnosis of hypertension ( ICD code R03 . 0 ) and were taking other medications for the management of their hypertension . A total of 170 patients were neither hypertensive nor taking other blood pressure medications . For the patient cohorts , propensity score matching was used to match the covariates including age , gender , race , smoking , caffeine consumption , alcohol consumption , and history of head injury ( Figure 7—figure supplement 1A-C ) between the cohorts taking RAAS inhibitors ( RAAS group ) and not taking RAAS inhibitors ( no RAAS group ) . For the UPDRS motor assessment analysis , a subset of patients from each cohort not taking levodopa for at least three years from their initial PPMI enrollment were selected to remove the possible confounding effects of levodopa on motor improvement . Part 1 ( non-motor experiences of activities of daily living ) , part 2 ( motor experiences of activities of daily living ) , and part 3 ( motor examination ) were assessed ( Figure 7—figure supplement 1E , F ) . The imaging data from screening studies and behavior studies were analyzed by unpaired t-test using R program and Graphpad Prism software and expressed as means ± SEM unless otherwise stated . Wilcoxon rank sum test was used for the analysis of the high throughput screening database . Differential gene expression analysis of the RNA-seq data was done using the DESeq2 package in R and the fold changes of gene expressions were evaluated with Wald test at an α of 0 . 05 and FDR 0 . 1 . Clinical data analysis of the PPMI database on Time-to-Levodopa was conducted with a Log-rank Mantel-cox test; the UPDRS motor scores were analyzed with nonparametric one-way analysis of variance ( ANOVA ) and post-hoc Tukey Test . | Parkinson’s disease is caused by the slow death and deterioration of brain cells , in particular of the neurons that produce a chemical messenger known as dopamine . Certain drugs can mitigate the resulting drop in dopamine levels and help to manage symptoms , but they cause dangerous side-effects . There is no treatment that can slow down or halt the progress of the condition , which affects 0 . 3% of the population globally . Many factors , both genetic and environmental , contribute to the emergence of Parkinson’s disease . For example , dysfunction of the mitochondria , the internal structures that power up cells , is a known mechanism associated with the death of dopamine-producing neurons . Zebrafish are tiny fish which can be used to study Parkinson’s disease , as they are easy to manipulate in the lab and share many characteristics with humans . In particular , they can be helpful to test the effects of various potential drugs on the condition . Here , Kim et al . established a new zebrafish model in which dopamine-producing brain cells die due to their mitochondria not working properly; they then used this assay to assess the impact of 1 , 403 different chemicals on the integrity of these cells . A group of molecules called renin-angiotensin-aldosterone ( RAAS ) inhibitors was shown to protect dopamine-producing neurons and stopped them from dying as often . These are already used to treat high blood pressure as they help to dilate blood vessels . In the brain , however , RAAS worked by restoring certain mitochondrial processes . Kim et al . then investigated whether these results are relevant in other , broader contexts . They were able to show that RAAS inhibitors have the same effect in other animals , and that Parkinson’s disease often progresses more slowly in patients that already take these drugs for high blood pressure . Taken together , these findings therefore suggest that RAAS inhibitors may be useful to treat Parkinson’s disease , as well as other brain illnesses that emerge because of mitochondria not working properly . Clinical studies and new ways to improve these drugs are needed to further investigate and capitalize on these potential benefits . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience",
"genetics",
"and",
"genomics"
] | 2021 | A zebrafish screen reveals Renin-angiotensin system inhibitors as neuroprotective via mitochondrial restoration in dopamine neurons |
The most frequent missense mutations in familial Parkinson’s disease ( PD ) occur in the highly conserved LRRK2/PARK8 gene with G2019S mutation . We previously established a fly model of PD carrying the LRRK2-G2019S mutation that exhibited the parkinsonism-like phenotypes . An herbal medicine , Gastrodia elata Blume ( GE ) , has been reported to have neuroprotective effects in toxin-induced PD models . However , the underpinning molecular mechanisms of GE beneficiary to G2019S-induced PD remain unclear . Here , we show that these G2019S flies treated with water extracts of GE ( WGE ) and its bioactive compounds , gastrodin and 4-HBA , displayed locomotion improvement and dopaminergic neuron protection . WGE suppressed the accumulation and hyperactivation of G2019S proteins in dopaminergic neurons and activated the antioxidation and detoxification factor Nrf2 mostly in the astrocyte-like and ensheathing glia . Glial activation of Nrf2 antagonizes G2019S-induced Mad/Smad signaling . Moreover , we treated LRRK2-G2019S transgenic mice with WGE and found that the locomotion declines , the loss of dopaminergic neurons , and the number of hyperactive microglia were restored . WGE also suppressed the hyperactivation of G2019S proteins and regulated the Smad2/3 pathways in the mice brains . We conclude that WGE prevents locomotion defects and the neuronal loss induced by G2019S mutation via glial Nrf2/Mad signaling , unveiling a potential therapeutic avenue for PD .
Parkinson’s disease ( PD ) is a highly prevalent neurodegenerative disorder characterized by the loss of dopaminergic neurons in the substantia nigra projecting to the striatum of the basal ganglion , representing a circuit involved in motor planning and coordination . As a consequence , PD is associated with motor abnormalities , bradykinesia , hypokinesia , rigidity , and resting tremor . Currently , the most frequently applied pharmacological treatment , levodopa ( L-DOPA ) , exerts limited motor improvement and elicits negative side effects ( Ray Chaudhuri et al . , 2018 ) . Hence , identifying and developing alternative or complementary treatments may assist in mitigating PD progression . PD is a multicausal disease with a complicated etiology , including familial inheritance . More than 20 PARK genes have been genetically linked to PD , a number that is increasing ( Houlden and Singleton , 2012 ) . Missense mutations in PARK8 , or Leucine-rich repeat kinase 2 ( LRRK2 ) , induce characteristic PD symptoms and pathologies such as loss of dopaminergic neurons and the appearance of Lewy bodies ( Martin et al . , 2014 ) . Notably , the most commonly observed mutation , dominant G2019S , among familial PD cases is located in the kinase domain of Lrrk2 , which augments its kinase activity via auto- and hyperactivation ( Sheng et al . , 2012 ) . The hyperactive G2019S mutant protein alters several cellular processes , including vesicle trafficking , microtubule dynamics , autophagy , mitochondrial function ( Martin et al . , 2014 ) , and , most commonly , increases susceptibility to oxidative stress that contributes to neuronal degeneration ( Angeles et al . , 2011; Nguyen et al . , 2011 ) . These indicate that regulation of the hyperactivation of G2019S mutant protein appears to be a disease-modifying strategy . Glia provide structural and metabolic supports to neurons and regulate synaptic transmissions , so they are important for the function and survival of dopaminergic neurons ( Lin et al . , 1993; Sofroniew and Vinters , 2010 ) . Dysfunction in two major glial types , astrocytes and microglia , contribute to the onset and progression of both sporadic and familial PD ( Kam et al . , 2020 ) . Astrocytes and microglia of postmortem PD brains exhibit pathological lesions and hyper-immunoactivity ( Miklossy et al . , 2006 ) . A clinical trial involving downregulation of microglial oxidative stress highlights the significance of glia to PD ( Jucaite et al . , 2015 ) . Moreover , Lrrk2 regulates the inflammatory response in microglia and the autophagy-lysosome pathway in astrocytes , with the G2019S mutation altering the size and pH of lysosomes ( Henry et al . , 2015; Moehle et al . , 2012 ) . Expression of G2019S mutant protein in neurons was previously shown to induce the secretion of Glass bottom boat ( Gbb ) /bone morphogenetic protein ( BMP ) signal that , in turn , upregulates Mothers against decapentaplegic ( Mad ) /Smad signaling in glia , which prompts feedback signals to promote neuronal degeneration ( Maksoud et al . , 2019 ) . These studies indicate that G2019S mutant protein alters the homeostasis and interaction between neurons and glia , contributing to PD pathogenesis . However , whether any dietary or pharmacological treatment blockading this neuron-glia interaction beneficiary to G2019S-induced PD is unclear . Given that up to 70% of human PARK genes are conserved in the Drosophila genome , Drosophila is frequently used as a PD model for studying gene function , such as the PARK1/SNCA ( Chen and Feany , 2005 ) and PARK8/LRRK2 ( Liu et al . , 2008 ) . Genetic and molecular linkage between PINK1/PARK6 and Parkin/PARK2 was first established in Drosophila ( Clark et al . , 2006 ) . When overexpressed in Drosophila dopaminergic neurons , LRRK2 transgenes carrying G2019S or other dominant mutations induce dopaminergic neuron loss and locomotion impairment , two age-dependent symptoms of PD ( Lin et al . , 2010; Liu et al . , 2008 ) . The G2019S model has further been used to screen a collection of FDA-approved drugs to suppress these PD phenotypes . Thus , Drosophila represents an amenable model of PD for genetic , molecular , and pharmacological study of potential therapeutic interventions . Strikingly , lovastatin was found to prevent dendrite degeneration , dopaminergic neuron loss , and impaired locomotion , and , critically , a lovastatin-involved Nrf2 pathway proved neuroprotective ( Lin et al . , 2016a ) . Nevertheless , whether the Nrf2-mediated neuroprotection is cell- or non-cell-autonomous remains elusive . Traditional Chinese Medicine ( TCM ) is often used as an alternative or dietary treatment for human diseases , including PD ( Kim et al . , 2012; Li et al . , 2017 ) . Although the results were inconclusive , some TCM could display adjuvant effects when used in combination with L-DOPA , reducing the L-DOPA dosage required in long-term treatments and relieving non-motor symptoms ( Kim et al . , 2012 ) . As a prominent component in TCM , Gastrodia elata Blume ( GE; Orchidaceae ) has been used to treat neurological disorders for centuries ( Chen and Sheen , 2011 ) . GE has been shown to exert neuroprotective , anti-inflammatory , and antioxidative effects in neurodegenerative disease models ( Jang et al . , 2015 ) . The major bioactive compounds in GE include gastrodin and 4-hydroxybenzyl alcohol ( 4-HBA ) , both of which display pharmacological effects on neurobiological and psychological disorders ( Chen et al . , 2016; Kumar et al . , 2013 ) . Additionally , gastrodin and 4-HBA have been reported to activate the Nrf2 signaling in dopaminergic neurons and astrocytes , respectively ( Jiang et al . , 2014; Luo et al . , 2017 ) , highlighting a potential benefit of incorporating GE in PD treatments . However , the effects and mechanisms underlying how GE moderate Lrrk2-G2019S PD remain unclear . In the present study , we treated G2019S animals with water extract of GE ( WGE ) , and its bioactive compounds , gastrodin and 4-HBA . We have investigated the impact of WGE treatment on PD in restoring locomotion and protecting dopaminergic neurons in the Drosophila G2019S model . We identified two distinct pathways induced by WGE in the model , that is , suppression of Lrrk2 protein accumulation and hyperphosphorylation in neurons , and activation of the Nrf2 pathway in glia , particularly in astrocyte-like and ensheathing glia . We show that WGE-induced Nrf2 activation antagonizes the Gbb-activated Mad signaling in glia , contributing to neuronal protection . WGE also suppressed the hyperactivation of G2019S proteins and antagonized Smad2/3 signaling in a LRRK2-G2019S mouse model , which restored locomotion , protected dopaminergic neurons , and regulated the microglia hyperactivation . Conservation of the pathways impacted by WGE treatment in both the Drosophila and mouse G2019S models implies the beneficial effects of GE and represent a reliable and effective complementary therapy for PD .
We employed the GAL4-UAS system to express the human G2019S mutant of Lrrk2 by the Ddc-GAL4 driver ( Ddc>G2019S ) in dopaminergic neurons and then assessed the anti-geotactic climbing activity of adult flies . We observed that locomotion of Ddc>G2019S flies was affected significantly relative to control flies expressing human wild-type Lrrk2 ( Ddc>Lrrk2 ) ( Figure 1A , Figure 1—figure supplement 1A ) . At weeks 1 and 2 , more than 80% of Ddc>G2019S flies could successfully climb above an 8 cm threshold , a proportion comparable to that of Ddc>Lrrk2 flies . However , the success rate declined to ~40% at week 3 , ~ 20% at week 4 , and to less than 10% at weeks 5 and 6 . These proportions are significantly lower than the ~80% at week 3 , ~60% at week 4 , and ~40% at weeks 5 and 6 displayed by Ddc>Lrrk2 flies . Although both Ddc>G2019S and Ddc>Lrrk2 flies failed to reach the 8 cm mark at weeks 7 and 8 , Ddc>Lrrk2 flies could still climb the wall , whereas almost all Ddc>G2019S flies could not ( Figure 1—figure supplement 1A ) . We also tested the climbing activity of another control expressing GFP in dopaminergic neurons ( Ddc>mCD8-GFP ) . Both Ddc>Lrrk2 and Ddc>mCD8-GFP flies showed comparable climbing activities in the first six weeks , and a significant number of Ddc>mCD8-GFP flies were still able to climb above the 8 cm mark at week 7 . Therefore , we used the Ddc>Lrrk2 line as a control for Ddc>G2019S flies in subsequent experiments to dissect the specific mode of pathogenicity of the G2019S mutation . Next , we examined the effect of feeding flies with WGE as a dietary supplement at different concentrations ( 0 . 1 , 0 . 5 , or 1 . 0% , w/w ) . WGE treatment of Ddc>G2019S flies at all three doses elicited a significant improvement in their climbing ability ( Figure 1A ) . Strikingly , the lowest concentration ( 0 . 1% ) of WGE proved the most effective , with Ddc>G2019S flies still performing well at climbing ( i . e . , comparably to Ddc>Lrrk2 control flies ) in weeks 5 and 6 . The higher doses of WGE ( 0 . 5 and 1 . 0% ) still exerted beneficial effects at weeks 3 and 4 , albeit not as significantly as the 0 . 1% dose , but had no beneficial effect in weeks 5 and 6 . As 0 . 1% is the lowest among the three doses tested , we lowered the dose of WGE to 0 . 02% and found that 0 . 02% WGE was less effective than 0 . 1% starting at week 3 till week 6 , suggesting that 0 . 1% is the optimal dose in restoring the climbing activity of Ddc>G2019S flies ( Figure 1—figure supplement 1B ) . We also tested the WGE effect on the Ddc>G2019S flies that were fed with regular food without WGE for 3 weeks . At week 4 , these Ddc>G2019S flies also showed a significant improvement in their climbing ability compared to the age-matched Ddc>G2019S flies fed continuously on regular food ( Figure 1—figure supplement 2 ) . The effect of improving climbing activity in the WGE-fed Ddc>G2019S flies was reduced at week 5 and diminished at week 6 , suggesting that WGE feeding starting at earlier stages is important for long-term locomotion improvement . Gastrodin and 4-HBA are two major phenolic compounds in GE displaying neuropharmacological effects ( Zhan et al . , 2016 ) . Feeding Ddc>G2019S flies with food containing gastrodin ( 0 . 1 mM ) equivalent to the amount in 0 . 1% WGE also restored locomotion of Ddc>G2019S flies in weeks 2–4 , though its impact diminished to nonsignificant levels at weeks 5 and 6 . However , increasing the gastrodin dose 10-fold ( 1 . 0 mM ) resulted in improved climbing activity at weeks 5 and 6 ( Figure 1B ) . Similarly , the equivalent 0 . 1 mM of 4-HBA , the aglyconic form of gastrodin and the bioactive form in the brain ( Wu et al . , 2017 ) , was sufficient to restore the climbing ability of Ddc>G2019S flies , and a 10-fold dose at 1 . 0 mM had an even better effect ( Figure 1C ) . These results indicate that both gastrodin and 4-HBA are primary bioactive compounds in GE that prevent locomotion decline in Ddc>G2019S flies , and higher doses are more beneficial to reach the effect as WGE did . Success in the antigravity wall-climbing assay also requires an immediate response to startle knockdown . Accordingly , we performed a second assay , free-walking in an open arena , to assess improved locomotion . Consistently , free-walking by Ddc>G2019S flies was greatly impaired , with total walking distance reduced to less than 20% that displayed by control Ddc>Lrrk2 flies ( Figure 1D ) . Moreover , Ddc>G2019S flies displayed centrophobism , that is , they avoided walking into the central open space . Strikingly , 0 . 1% WGE feeding greatly extended walking distance and suppressed the centrophobism of Ddc>G2019S flies . Together , these two assays strongly indicate that the defective locomotion exhibited by Ddc>G2019S flies is greatly improved by feeding them with 0 . 1% WGE . Because the improvement on the locomotion was more effective in flies fed with 0 . 1% WGE than the pure compounds , we therefore fed the flies with 0 . 1% WGE in the following experiments . Expression of G2019S mutant protein has been shown to induce a gradual loss of dopaminergic neurons in the adult fly brain , contributing to impaired locomotion ( Lin et al . , 2010; Liu et al . , 2008 ) . Several clusters of dopaminergic neurons have been identified in the adult brain of Drosophila . Here , we focused on the PPL1 , PPL2 , PPM1/2 , and PPM3 clusters that have well-defined roles in modulating locomotion ( Mao and Davis , 2009 ) to assess the effect of WGE treatment . We detected reduced numbers of dopaminergic neurons in the PPL1 , PPL2 , PPM1/2 , and PPM3 clusters of Ddc>G2019S flies relative to Ddc>Lrrk2 controls , which increased in severity from weeks 2 to 6 ( Figure 2 , Figure 2—figure supplement 1A–C ) . Consistently , feeding Ddc>G2019S flies with 0 . 1% WGE restored numbers of dopaminergic neurons in these clusters to the levels observed in controls ( Figure 2 , Figure 2—figure supplement 1A–C ) . Thus , concomitant rescue of locomotion and dopaminergic neuron populations in Ddc>G2019S flies indicates that WGE treatment likely promotes dopaminergic neuron survival to restore locomotion . The enhanced survival of dopaminergic neurons due to WGE treatment implies that WGE induces neuroprotective mechanisms in Ddc>G2019S flies . The G2019S mutation causes Lrrk2 hyperphosphorylation , protein accumulation , and aberrant cellular signaling ( Price et al . , 2018 ) . Therefore , we explored if WGE-induced neuroprotection is responsible for abrogating these processes in G2019S flies . We detected comparable levels of Lrrk2 proteins in 3-day-old adult brains pan-neuronally expressing wild-type Lrrk2 ( elav>Lrrk2 ) or G2019S mutant protein ( elav>G2019S ) ( Figure 3A ) . However , phosphorylation levels at the Ser1292 autophosphorylation site were higher in the elav>G2019S flies compared to elav>Lrrk2 ( Figure 3A and B ) . This outcome was also observed at week 4 ( Figure 3F and G ) , consistent with the idea that Lrrk2 is hyperactivated upon G2019S mutation . Hyperphosphorylation in elav>G2019S flies led to Lrrk2 protein accumulation , as determined by total Lrrk2 levels at weeks 2 and 4 ( Figure 3D and E ) . However , both hyperphosphorylation and protein accumulation were suppressed upon feeding elav>G2019S flies with 0 . 1% WGE ( Figure 3D–G ) . In the elav>Lrrk2 control , WGE feeding had no effect on levels of wild-type Lrrk2 or Ser1292 phosphorylation ( Figure 3—figure supplement 1A–C ) . We further examined the G2019S mutant-activated downstream effector Rab10 , the phosphorylation status of which can serve as an indicator of Lrrk2 kinase activity ( Karayel et al . , 2020 ) . We observed that levels of phosphorylated Rab10 were increased in elav>G2019S flies , but this phenotype was suppressed by WGE feeding ( Figure 3F and H ) . Hence , feeding flies with WGE suppresses G2019S-induced Lrrk2 protein phosphorylation , accumulation , and signaling . The Akt/GSK3β/Nrf2 signaling axis has been shown to promote survival of dopaminergic neurons and ameliorate motor dysfunction in PD models ( Lin et al . , 2016b ) . We assayed the phosphorylation status of Akt at Ser505 in 3-day-old adult fly head extracts and found reduced levels of pAkt in elav>G2019S flies relative to elav>Lrrk2 controls ( Figure 3A and C ) . This pAkt reduction persisted at weeks 2 and 4 , but was abrogated by feeding elav>G2019S flies with 0 . 1% WGE ( Figure 4A and B ) . We also examined phosphorylation levels of Nrf2 at Ser40 and GSK3β at Ser9 , both of which were reduced upon expression of G2019S mutant protein and were equally offset by WGE feeding ( Figure 4C and D ) . Induction of Akt/GSK3β/Nrf2 signaling activates expression of heme oxygenase 1 ( HO-1 ) , and we observed diminished levels of this latter protein in elav>G2019S flies , which could be rescued by WGE feeding ( Figure 4C and D ) . Therefore , WGE feeding restores the Akt/GSK3β/Nrf2 signaling activity compromised by the G2019S mutation . Restoration of Akt/GSK3β/Nrf2 signaling activity by WGE treatment prompted us to test by genetic assays if that signaling pathway mediates the WGE mode of action . We focused on the downstream effector Nrf2 encoded by cap-n-collar ( cnc ) in Drosophila . Intriguingly , neither overexpression ( UAS-cncC-FL2 ) nor RNAi knockdown ( UAS-cncTRiP ) of Nrf2 in dopaminergic neurons had an impact on the climbing ability of Ddc>G2019S flies ( Figure 5A ) . Also , WGE feeding still rescued locomotion of Ddc>G2019S flies with Nrf2 knockdown ( Ddc>G2019S; cncTRiP ) to levels comparable to WGE-fed control flies ( Ddc>G2019S; mCD8-GFP ) without Nrf2 knockdown ( Figure 5A ) . Thus , the Nrf2 pathway activation by WGE that appears to be effective in protecting neurons and restoring locomotion is likely exerted in cells other than neurons . Nrf2 activation may be examined by assessing GFP signal from the ARE-GFP reporter that harbors Nrf2 binding sites and responds to Nrf2 activation ( Chatterjee and Bohmann , 2012 ) . ARE-GFP adult fly brains displayed low basal GFP signals ( Figure 5—figure supplement 1A ) , but feeding with 0 . 1% WGE elicited many GFP-positive signals in Repo-positive glia ( arrowheads in Figure 5—figure supplement 1B ) , evidencing that glia may be the primary cell type in which Nrf2 is activated . Next , we investigated if WGE-induced Nrf2 activity in glia is effective in promoting locomotion in G2019S flies . To this end , we employed the LexA-LexAop system to overexpress wild-type Lrrk2 or G2019S mutant protein in dopaminergic neurons ( Ddc-LexA>Lrrk2 and Ddc-LexA>G2019S , respectively ) and used the GAL4 driver to manipulate Nrf2/CncC activity in glia ( repo>cncC-FL2 for overexpression and repo>cnc-RNAi for knockdown ) . As a first step , we validated the age-dependent locomotion decline of Ddc-LexA>G2019S flies that was severer than Ddc-LexA>Lrrk2 and could be rescued by 0 . 1% WGE feeding ( Figure 5—figure supplement 2 ) . We then assayed the climbing ability of Ddc-LexA>G2019S flies exhibiting repo-GAL4-driven Nrf2 overexpression in glia ( Ddc-LexA>G2019S; repo>cncC-FL2 ) . Significantly , the Ddc-LexA>G2019S; repo>cncC-FL2 flies performed better in the climbing assay than Ddc-LexA>G2019S; repo-GAL4 without Nrf2 overexpression ( Figure 5B ) . We also performed Nrf2 knockdown in glia ( Ddc-LexA>G2019S; repo>cnc-RNAi ) , which had little impact on the already declined climbing activity in Ddc-LexA>G2019S; repo-GAL4 ( Figure 5B ) . Importantly , WGE feeding could not rescue locomotion deficits of glial Nrf2-knockdown flies ( Ddc-LexA>G2019S; repo>cnc-RNAi ) ( Figure 5B , compare to Ddc-LexA>G2019S; repo-GAL4 with WGE feeding ) . Thus , glial overexpression of Nrf2 partially restores locomotion of Ddc-LexA>G2019S flies , and glial depletion of Nrf2 abolishes the ability of WGE to rescue impaired locomotion . We stained dopaminergic neurons of 6-week-old adult fly brains and confirmed that numbers of TH-positive dopaminergic neurons in the PPL1 cluster were reduced in the Ddc-LexA>G2019S; repo-GAL4 flies compared to Ddc-LexA>Lrrk2; repo-GAL4 controls ( Figure 5C and D ) . Importantly , numbers of dopaminergic neurons in the PPL1 cluster were restored upon glial overexpression of Nrf2 in the Ddc-LexA>G2019S; repo>cncC-FL2 flies . In contrast , glial Nrf2 knockdown had little impact on the already reduced dopaminergic neurons . Taken together , these results support that glial Nrf2 is compromised in Ddc-LexA>G2019S flies , and glial expression of Nrf2 protects the dopaminergic neurons . Although WGE feeding suppressed hyperactivity of G2019S mutant protein ( Figure 3D–G ) , it was not clear if WGE-mediated Nrf2 activation in glia could directly suppress mutant protein hyperactivity in dopaminergic neurons . To test this possibility , we compared the levels of total Lrrk2 protein and phosphorylated Lrrk2 protein in Ddc-LexA>G2019S with and without glial Nrf2 overexpression ( Ddc-LexA>G2019S; repo-GAL4 and Ddc-LexA>G2019S; repo>cncC ) . We observed comparable levels of Lrrk2 and pLrrk2 in the control and Nrf2 overexpression lines ( Figure 6—figure supplement 1A–C ) . Thus , the neuroprotective effects of Nrf2 activity are unlikely to operate through modulation of Lrrk2 levels or activity . Five types of glia with different morphologies and functions have been identified in the fly brain ( Freeman , 2015 ) . We decided to identify specific subtypes of glia that may mediate the WGE-induced Nrf2 activity endowing neuronal protection , given that dysfunctional astrocytes and microglia have been linked to onset and progression of both sporadic and familial PD ( Kam et al . , 2020 ) . In Ddc-LexA>G2019S flies , GAL4-driven cnc knockdown in astrocyte-like ( Ddc-LexA>G2019S; alrm>cnc-RNAi ) or ensheathing glia ( Ddc-LexA>G2019S; R56F03>cnc-RNAi ) abolished the improved locomotion elicited by WGE treatment , recapitulating the effect of pan-glial Nrf2 knockdown ( Ddc-LexA>G2019S; repo>cnc-RNAi ) ( Figure 6A ) . This outcome was not observed when we used GAL4 drivers to knock down cnc in cortex ( np2222 ) , perineurial ( np6293 ) , or subperineurial ( moody ) glia . We confirmed the involvement of astrocyte-like and ensheathing glia by means of Nrf2 overexpression in astrocyte-like ( Ddc-LexA>G2019S; alrm>cncC-FL2 ) or ensheathing ( Ddc-LexA>G2019S; R56F03>cncC-FL2 ) glia , with both treatments improving the climbing activity of Ddc-LexA>G2019S flies not subjected to WGE feeding ( Figure 6B ) . These analyses indicate that WGE feeding induces Nrf2 activity in astrocyte-like and ensheathing glia , which mitigates the reduced locomotion displayed by G2019S mutant-expressing flies . Next , we assayed Nrf2-regulated ARE-GFP expression in astrocyte-like and ensheathing glia of flies expressing G2019S mutant protein and subjected to WGE treatment . We focused on the astrocyte-like and ensheathing cells located adjacent to dopaminergic neurons . In control flies expressing wild-type Lrrk2 , we detected basal levels of GFP signal in mCherry-positive astrocyte-like ( alrm-GAL4 ) or ensheathing glia ( R56F03 ) ( Figure 6C–F ) . We detected lower levels of GFP signal in these cells when G2019S mutant protein was expressed in dopaminergic neurons . However , upon feeding with 0 . 1% WGE , we observed higher levels of GFP signal in both types of glia . Changes in the intensities of GFP signals were also detected in TH-positive dopaminergic neurons , although the levels were lower than in glia ( Figure 6—figure supplement 2A and B ) . Thus , overexpression of G2019S mutant protein in dopaminergic neurons elicited reduced Nrf2 signaling activity , but WGE feeding restored or further enhanced Nrf2 activities in these two glial subtypes . G2019S mutant protein in dopaminergic neurons has been shown previously to enhance the expression of the proprotein convertase Furin 1 ( Fur1 ) that processes the BMP signaling molecule Gbb for maturation and release , leading to activation of the BMP signaling pathway in glia ( Maksoud et al . , 2019 ) . We confirmed that finding by removing one copy of Mad that encodes the pathway’s downstream effector to restore locomotion in Ddc>G2019S flies ( Figure 7—figure supplement 1 ) . To address if WGE-induced Nrf2 activation could antagonize BMP signaling activity in glia , first we assessed expression of the phosphorylated Mad ( pMad ) activated by BMP signaling . In glia of Ddc>G2019S adult fly brains , pMad levels were higher than in Ddc>Lrrk2 brains and they could be suppressed by WGE treatment ( Figure 7A and B ) . Similar to the previous report ( Maksoud et al . , 2019 ) , glial overexpression of Mad ( UAS-Mad ) or constitutively active type I receptor Tkv ( UAS-tkvQ253D ) was sufficient to impair locomotion , even without expressing G2019S mutant protein in neurons . However , impaired locomotion was rescued in both cases by 0 . 1% WGE feeding ( Figure 7C and D ) . In addition , numbers of dopaminergic neurons in the PPL1 cluster were reduced upon glial overexpression of Mad or tkvQ253D and they were restored by 0 . 1% WGE treatment ( Figure 7E–H ) . Together , these analyses indicate that activation of BMP signaling in glia recapitulates the phenotypes observed in flies overexpressing G2019S mutant protein in dopaminergic neurons and , furthermore , that these effects can be suppressed by WGE treatment . Next , we explored if WGE-induced Nrf2 activation antagonizes BMP signaling in glia . Glial Nrf2 overexpression in Ddc-LexA>G2019S flies ( Ddc-LexA>G2019S; repo>cncC-FL2 ) partially rescued the locomotory impairment caused by G2019S mutation ( Figure 5B ) , and removing one copy of Mad ( Ddc-LexA>G2019S , Mad+/-; repo>cncC-FL2 ) further enhanced this effect ( Figure 8A ) . Indeed , this outcome was equivalent to removing one copy of Mad but without Nrf2 overexpression ( Ddc-LexA>G2019S , Mad+/-; repo-GAL4 ) , suggesting that Nrf2 functions mainly to antagonize Mad activity ( Figure 8A ) . Consistently , Nrf2 overexpression in glia or WGE treatment suppressed the upregulation of pMad levels in glia caused by G2019S mutant protein overexpression in dopaminergic neurons ( Ddc-LexA>G2019S; repo-GAL4 ) ( Figure 8B and C ) . Depletion of Nrf2 from glia ( Ddc-LexA>G2019S; repo>cnc-RNAi ) , even in the presence of WGE treatment , maintained high pMad levels in glia ( Figure 8B and C ) . Thus , modulation of Nrf2 activity , either by genetic manipulation or by WGE treatment , has an antagonistic effect on pMad levels in glia . We also addressed if Mad modulates expression of the Nrf2 target ARE-GFP . We observed diminished GFP signal in PPL1-surrounding glia of Ddc-LexA>G2019S adult fly brains relative to the Ddc-LexA>Lrrk2 control . Heterozygosity of Mad in Ddc-LexA>G2019S flies ( Ddc-LexA>G2019S , Mad+/- ) restored the level of GFP signal , implying that Mad modulates targeted gene expression induced by Nrf2 activity ( Figure 8D and E ) . To further study the effect of WGE on G2019S mutation-induced neurodegeneration in a mammalian system , we fed LRRK2-G2019S transgenic mice with WGE starting at the age of 8 . 5 months , that is , prior to onset of impairments in locomotion and dopaminergic neurons ( Chou et al . , 2014 ) . We quantified three locomotor activities from video-tracking paths of an open-field test , that is , accumulative moving distance , average velocity , and percentage of time moving ( Figure 9A and B ) . At the age of 8 . 5 months , three groups—non-transgenic ( nTg ) , transgenic LRRK2-G2019S , and WGE-fed transgenic LRRK2-G2019S mice—presented comparable locomotor activities . At 9 . 5 months of age , the LRRK2-G2019S mice displayed clearly impaired locomotion , which was statistically significant at the age of 11 . 5 months relative to nTg littermates ( Figure 9B ) , consistent with a previous report ( Chen et al . , 2012 ) . Importantly , the LRRK2-G2019S mice fed with WGE showed improved locomotion throughout the 3-month treatment period , an outcome that was statistically significant at 11 . 5 months ( Figure 9A and B ) . WGE treatment also suppressed the centrophobism displayed by LRRK2-G2019S mice ( Figure 9A ) . We also analyzed the gait of these three groups of mice ( Figure 9C ) . Similar to our findings from the open-field test , stride length of LRRK2-G2019S mice was significantly reduced at 11 . 5 months , but it was restored to the level of nTg mice by WGE feeding ( Figure 9C and D ) . Collectively , these analyses show that WGE feeding is an effective means of restoring G2019S mutation-induced locomotory declines in this mouse model of PD . Since the nigral-striatal system contributes to locomotor function , we counted the number of TH-positive dopaminergic neurons in the substantia nigra . In comparison to nTg littermates , the number of dopaminergic neurons in 11 . 5-month-old LRRK2-G2019S mice was significantly reduced , but WGE treatment for 3 months abrogated this loss of dopaminergic neurons ( Figure 10A and B ) . In the Ddc>G2019S Drosophila model , glia mediate the protective effects of WGE on dopaminergic neurons . We found that activated microglia marked by ionized calcium-binding adapter molecule 1 ( Iba-1 ) were increased in the substantia nigra of LRRK2-G2019S mice , but this increase was suppressed by WGE treatment ( Figure 10C and D , Figure 10—figure supplement 1 ) . We further analyzed levels of activated LRRK2 ( pLRRK2 ) in lysates isolated from the nigra striatum . Levels of pLRRK2 normalized to LRRK2 signal were higher in LRRK2-G2019S mice than in nTg littermates , but WGE treatment partially abrogated that outcome ( Figure 10E and F ) . Since G2019S mutation enhanced Mad signaling activation in the fly brain ( Figure 8B and C ) , we also tested this scenario in the mouse model . Immunoblots revealed that the ratio of phosphorylated Smad2 to total Smad2 ( pSmad2/Smad2 ) was significantly elevated in LRRK2-G2019S mice , but this increase was suppressed by WGE treatment ( Figure 10E and G ) . Moreover , although the level of pSmad3/Smad3 was not significantly enhanced by G2019S mutant protein expression , it was suppressed by WGE treatment ( Figure 10E and G ) . Thus , WGE feeding suppresses G2019S mutation-induced microglia activation and Smad signaling in the substantia nigra of the LRRK2-G2019S mouse model of PD . In summary , the potential mechanisms of WGE involved in the modulation of the neuron-glial interaction are proposed ( Figure 11 ) . WGE regulates the hyperactivation of G2019S mutant protein in dopaminergic neurons and antagonizes the Mad signaling by activating the Nrf2 pathway in glia . Both actions provide neuroprotection .
Motor dysfunction in Drosophila neurodegeneration models has been frequently evaluated by means of negative-geotaxis assay that measures the insect innate response . The assay begins with a sudden external stimulation , which initially inhibits spontaneous locomotion but is followed by climbing behavior . The entire response requires motor circuit coordination and muscle tone regulation , two processes that progressively decline with age and that are impacted by neuropathological insults ( Grotewiel et al . , 2005 ) . In contrast , the open-arena walking assay allows free exploration without disturbance , representing an assay for locomotor activities that can reveal deficits like bradykinesia ( Chen et al . , 2014 ) . Our G2019S flies exhibited reduced locomotor activity in both assays . Moreover , the G2019S flies also exhibited centrophobism-like behavior in the open arena ( Figure 1D ) . Centrophobism is indicative of emotional abnormalities , such as anxiety and depression , both of which are often associated with PD ( Kulisevsky et al . , 2008 ) , and was also displayed by the LRRK2-G2019S mice ( Figure 9A ) . WGE treatment suppresses centrophobism in both fly and mouse PD models and has important implications for tackling major symptoms of PD and even non-PD-related depression , as reported for rodent models ( Lin et al . , 2018; Lin et al . , 2016b ) . Thus , WGE treatment exerts beneficial effects in both of our PD models . We found that the 0 . 1% dosage of WGE is optimal in suppressing age-dependent locomotion decline in G2019S flies , with higher and lower doses being less effective ( Figure 1A , Figure 1—figure supplement 1B ) . Although an inverted U-shaped drug response is common , a plausible explanation for the diminished effectiveness of higher doses is that WGE downregulates the day-time but not night-time locomotor activity of flies ( Jo et al . , 2017 ) . In mice , higher WGE doses have sleep-promoting effects by activating adenosine A1/A2A receptors in the ventrolateral preoptic area ( Zhang et al . , 2012 ) . Thus , dosage level is critical to the beneficial effects of GE in both PD models . Gastrodin and 4-HBA are considered the principal active components in GE ( Zhan et al . , 2016 ) . Although both compounds can cross the blood-brain barrier ( Wu et al . , 2017 ) , the capability of gastrodin to do so is relatively poor compared to aglyconic 4-HBA due to the glucose moiety ( Lin et al . , 2007 ) . Gastrodin is quickly metabolized to 4-HBA and undetermined metabolites in the brain ( Lin et al . , 2008 ) , perhaps explaining the lower effectiveness of gastrodin in G2019S flies . However , it is likely that the combination of gastrodin with other components in GE might exert optimal beneficial effects . Expression of G2019S mutant protein impacts different clusters of dopaminergic neurons in the fly brain that are known for their connectivity and function . Activation of two specific mushroom body ( MB ) -projection dopaminergic neurons in the PPL1 cluster inhibits climbing performance ( Sun et al . , 2018 ) . Mutations in the circadian gene Clock ( Clk ) cause PPL1 dopaminergic neuron degeneration , accelerating impaired age-associated climbing ability ( Vaccaro et al . , 2017 ) . Dopaminergic neurons in the PPL2 cluster extend processes to the calyx of the MB , which has been linked to climbing activity ( Sun et al . , 2018 ) . In contrast , PPM3 neurons project to the central complex , activation of which enhances locomotion ( Kong et al . , 2010 ) . A significant reduction in dopaminergic neurons in the PPM1/2 cluster was only found at week 4 when impaired locomotion of G2019S flies was prominent . In a PD fly model involving SNCA overexpression and aux knockdown , dopaminergic neurons in the PPM1/2 cluster were selectively degenerated and this phenotype was accompanied by impaired locomotion in relatively young adult flies ( Song et al . , 2017 ) . We postulate that the age-dependent impaired locomotion displayed by G2019S flies could be caused by gradual and differential loss of dopaminergic neurons in these clusters , thereby affecting different aspects of locomotion . However , further study is needed to test that hypothesis . Expression of the G2019S mutant protein induces Lrrk2 auto- and hyperphosphorylation , as well as protein accumulation , together enhancing cellular Lrrk2 activity and causing aberrant downstream signaling ( Sheng et al . , 2012 ) . We have shown here that neuronal expression of Lrrk2-G2019S reduced Akt phosphorylation ( Figure 4A and B ) . Consistently , hyperactivated G2019S mutant protein impaired interaction with and phosphorylation of Akt , resulting in compromised signaling and accelerated neurodegeneration ( Ohta et al . , 2011; Panagiotakopoulou et al . , 2020 ) . However , WGE feeding restored downstream Akt signaling by suppressing G2019S mutant protein hyperactivation . Rab10 , one of the best-characterized substrates for Lrrk2 , mediates several of Lrrk2’s cellular functions ( Karayel et al . , 2020 ) . In Drosophila , Rab10 and Lrrk2-G2019S synergistically affect the activity of dopaminergic neurons , mediating deficits in movement ( Fellgett et al . , 2021; Petridi et al . , 2020 ) . We have shown that WGE treatment downregulated levels of Lrrk2 accumulation and phosphorylated Rab10 ( Figure 3F–H ) , thus alleviating their synergistic toxicity . Several kinase inhibitors have been developed to block the kinase activity of Lrrk2 , including of both wild-type Lrrk2 and the G2019S mutant , which could affect endogenous Lrrk2 activity ( Sheng et al . , 2012 ) . Instead , WGE treatment modulates the phosphorylation status and protein level of the G2019S mutant but not those of wild-type Lrrk2 . The new hydrogen bond created at the Ser2019 autophosphorylation site may provide a docking site for some chemicals in WGE , representing a possible explanatory mechanism that warrants further study ( Lang et al . , 2015 ) . The antioxidation and detoxification factor Nrf2 is a target of Akt activation . Nrf2 phosphorylation and HO-1 expression levels revealed that Nrf2 is inactivated in G2019S flies , but it was activated by WGE treatment ( Figure 4C and D ) . Intriguingly , our genetic data indicate that Nrf2 primarily functions in the glia of G2019S flies , with Nrf2 depletion from glia eliminating the beneficial effects of WGE and glial Nrf2 activation partially substituting for WGE feeding ( Figure 5B ) . Cortical neurons express much lower levels of Nrf2 than astrocytes owing to hypo-acetylation and transcriptional repression of the Nrf2 promoter ( Bell et al . , 2015 ) . Moreover , neurons express greater amounts of Cullin 3 , the scaffold component of the E3 ubiquitin ligase that targets Nrf2 for proteasomal degradation ( Jimenez-Blasco et al . , 2015 ) . Both those mechanisms render neuronal Nrf2 inert to activation . Nrf2 activation in astrocytes maintains neuronal integrity and function against oxidative insults in response to stress by supplying antioxidants such as glutathione and HO-1 ( Kraft et al . , 2004; Vargas and Johnson , 2009 ) . Previous study showed that 4-HBA triggers glia to secrete HO-1 via the Nrf2 pathway , protecting neurons from hydrogen peroxide in the primary culture ( Luo et al . , 2017 ) . In PD models in which wild-type or mutant α-synuclein is overexpressed , activation of neuronal Nrf2 ( Barone et al . , 2011; Skibinski et al . , 2017 ) or astrocytic Nrf2 ( Gan et al . , 2012 ) proved neuroprotective . In a previous study , lovastatin treatment provides neuroprotection in the G2019S-induced PD model , also through the Akt/Nrf2 pathway ( Lin et al . , 2016b ) . As activation of neuronal Nrf2 plays a nonconventional role in promoting developmental dendrite pruning ( Chew et al . , 2021 ) , it remains interesting to further study the cell types that mediate the action of lovastatin . By genetically manipulating the G2019S fly model , we have shown that WGE-induced Nrf2 activation in glia but not in neurons protects dopaminergic neurons from degeneration and ameliorates impaired locomotion . Astrocyte-like and ensheathing glia are two major types of glia in the Drosophila nervous system , surrounding and also extending long processes into neuropils of the brain . These astrocyte-like glia exhibit a morphology and function similar to those of mammalian astrocytes , including reuptake of neurotransmitters and phagocytosis of neuronal debris ( Freeman , 2015; Tasdemir-Yilmaz and Freeman , 2014 ) . Ensheathing glia of varying morphologies encase axonal tracts and neuropils , regulating neuronal excitability and participating in phagocytosis and injury-induced inflammation ( Doherty et al . , 2009; Otto et al . , 2018 ) . Thus , given their proximity to neurons and similar functions , it is not surprising that both types of glia collectively mediate the protective effects of WGE . Communication between neurons and glia maintains homeostasis , yet also confers the disease state during neurodegeneration . In the Drosophila G2019S model , upregulation of the BMP ligand Gbb in dopaminergic neurons activates Mad/Smad signaling in glia , which promotes neuronal degeneration via a feedback mechanism ( Maksoud et al . , 2019 ) . Surprisingly , although the number of dopaminergic neurons in the fly brain is relatively small , the upregulated pMad signal spreads throughout the brain ( Figure 7B ) , suggesting that BMP can be disseminated over long distances . In PD patients , higher levels of TGF-β1 have been detected in the striatum and ventricular cerebrospinal fluid ( Vawter et al . , 1996 ) . Thus , members of the TGF-β1 superfamily such as TGF-β1 and BMP signaling molecules may represent indicators of neuronal degeneration . Accordingly , disrupting the glia-to-neuron feedback mechanism may sustain neuronal survival . In glia , we found that WGE treatment downregulated the pMad levels that had been increased in the G2019S flies ( Figure 7A and B ) . Nrf2 activation in glia also suppressed the enhanced levels of pMad in G2019S flies ( Figure 8B and C ) . Indeed , our genetic assays indicate that the Nrf2 and Mad pathways interact in the glia of G2019S flies . Thus , WGE exerts its beneficial effects by activating Nrf2 to antagonize the Mad activity that would otherwise contribute to the degeneration of dopaminergic neurons . As a transcriptional activator , Nrf2 induces expression of the inhibitory component Smad7 to form inactive Smad complexes ( Song et al . , 2019 ) and the phosphatase subunit PPM1A to alter Smad2/3/4 phosphorylation and DNA binding ( Lin et al . , 2006 ) . Given that these components are conserved in Drosophila , Nrf2 may employ similar pathways to block Mad signaling in glia . That glial Nrf2 activation protects neurons is evidenced by our observations of enhanced HO-1 expression ( Figure 4C ) and increased numbers of dopaminergic neurons ( Figure 5C and D ) . These results support that the role of Nrf2 in glia is to induce expression of antioxidation building blocks , such as phase-II detoxification enzymes , and to enhance inflammatory processes ( Hirrlinger and Dringen , 2010; Rojo et al . , 2010 ) . In a model of fibrosis , TGF-β/Smad2/3 suppressed expression of the ARE-luciferase reporter and glutathione ( Ryoo et al . , 2014 ) . Moreover , Nrf2 knockdown was shown to reduce expression of the antioxidative enzyme NAD ( P ) H quinone dehydrogenase 1 ( NQO1 ) , thereby elevating cellular oxidative stress and upregulating TGF-β/Smad targeted gene expression ( Prestigiacomo and Suter-Dick , 2018 ) . Hence , we propose that WGE promotes Nrf2 activation to antagonize the Smad signaling in glia that is induced by dopaminergic neuron-secreted BMP/Gbb signal during degeneration . In our study , the LRRK2-G2019S mice show locomotor defects and dopaminergic loss at the age of 11 . 5 months . A previous study shows only earlier signs of defects , the reduction of the dopamine level and release at the age of 12 months in the LRRK2-G2019S mice ( Li et al . , 2010 ) , which could be contributed by the genetic background ( FVB/NJ vs . C57BL/6J ) . Nevertheless , we have demonstrated that feeding these mice with WGE rescues their locomotor coordination , suppresses their centrophobism , and recovers their numbers of dopaminergic neurons and hyperactivated microglia ( Figure 9 and Figure 10 A – D ) . Significantly , we found that the activity of the TGF-β/Smad2/3 pathway was elevated in nigrostriatal brain lysates , and this activity was also suppressed by WGE treatment ( Figure 10 ) . Collectively , these results from fly and mouse PD models indicate that the effectiveness of WGE is likely mediated through conserved Nrf2/Mad pathways ( Figure 11 ) . Our findings contribute to our mechanistic understanding of PD and provide potential therapeutic strategies that incorporate the traditional herbal medicine GE .
All fly stocks were maintained on standard cornmeal-based food medium at 25°C . Drosophila stocks sourced from the Bloomington Drosophila Stock Center ( Indiana University , Bloomington , USA ) were UAS-mCD8-GFP ( #5137 ) , elac-GAL4 ( #8760 ) , repo-GAL4 ( #7215 ) , Ddc-LexA ( #54218 ) , UAS-cncTRIP ( #25984 ) , Tub-GAL80ts ( #7108 ) , alrm-GAL4 ( #67032 ) , R56F03-GAL4 ( #39157 ) , UAS-tkvQ253D ( #36536 ) , and MadK00237 ( #10474 ) . NP2222-GAL4 ( #112830 ) was from the Drosophila Genomics Resource Center and NP6293-GAL4 ( #105188 ) was from the Kyoto Stock Center . Other stocks include UAS-Flag-LRRK2-WT ( Lin et al . , 2010 ) , UAS-Flag-LRRK2-G2019S ( Lin et al . , 2010 ) , Ddc>GAL4 ( Sang et al . , 2007 ) , ARE-GFP ( Sykiotis and Bohmann , 2008 ) , UAS-cncC-FL2 ( Sykiotis and Bohmann , 2008 ) , moody-GAL4 ( Bainton et al . , 2005 ) , and UAS-Mad ( Takaesu et al . , 2006 ) . The two LexAop fly lines—LexAop-LRRK2-WT and LexAop-LRRK2-G2019S—were generated in this study . In brief , the cDNAs for LRRK2-WT and LRRK2-G2019S were isolated from the pDEST53-LRRK2-WT and pDEST53-LRRK2-G2019S plasmids ( Addgene , MA ) for subcloning into LexAop plasmids ( Addgene ) , which were for microinjection ( Fly facility , University of Cambridge , UK ) . The transgenes were site landed at an attP site on the 2nd chromosome ( 25C6 ) . For temperature-shift assay of GAL80ts flies , parental flies were maintained at 19°C and allowed to mate , before collecting the F1 adults and shifting them to 29°C to inactivate GAL80 . Authentication of GE and preparation of WGE were as described previously ( Lin et al . , 2018; Lin et al . , 2016a ) . WGE ( KO DA Pharmaceutical Co . Ltd . , Taoyuan , Taiwan ) , gastrodin ( Wuhan YC Fine Chemical Co . , Wuhan , China ) , and 4-HBA ( Sigma-Aldrich , Darmstadt , Germany ) were added to freshly prepared cornmeal-based fly food at indicated final concentrations ( w/w ) . For experiments , 1- to 3-day-old post-eclosion flies were collected and transferred to fresh food medium twice per week . A negative geotaxis climbing assay was performed to assess locomotor activity , and it was conducted according to a previous study with minor modification ( Madabattula et al . , 2015 ) . Cohorts of 35 flies from each genotype were assayed weekly for six consecutive weeks . Success rates were calculated as the percentage of flies that could climb above the 8 cm mark of a 20 cm cylinder within 10 s . The free-walking assay protocol was conducted based on a previous report with minor modification ( Chen et al . , 2014 ) . Cohorts of eight flies were habituated on a 10 cm agar-filled dish for 30 min . The dishes were gently tapped to encourage the flies to walk , which was video-taped for 5 min . Movement tracks were processed in ImageJ and quantified using the Caltech multiple fly tracker ( Ctrax ) . The protocol for immunostaining whole-mount adult brains was essentially as described previously ( Lin et al . , 2010; Maksoud et al . , 2019 ) . Adult fly brains for each genotype were dissected at the indicated timepoints for immunostaining with the following primary antibodies: mouse anti-TH ( Immunostar , 22941 , 1:1000 ) , mouse anti-repo ( Hybridoma Bank DSHB , 8D12 , 1:500 ) , chicken anti-GFP ( Abcam , ab13970 , 1:10 , 000 ) , and rabbit anti-phospho-Smad3 ( Ser423/425 ) ( Abcam , ab52903 , 1:250 ) ( Smith et al . , 2012 ) . Fluorophore-conjugated secondary antibodies were FITC-conjugated goat anti-mouse IgG ( Jackson ImmunoResearch , AB_2338589 , 1:500 ) , Alexa Fluor 488-conjugated goat anti-mouse IgG ( Invitrogen , A28175 , 1:500 ) , Cy3-conjugated goat anti-mouse IgG ( Jackson ImmunoResearch , AB_2338680 , 1:500 ) , Alexa Fluor 488-conjugated goat anti-rabbit IgG ( Invitrogen , A27034 , 1:500 ) , and Cy5-conjugated goat anti-rat IgG ( Invitrogen , A10525 , 1:500 ) . Phalloidin-TRITC ( Sigma-Aldrich , P1951 , 1:5000 ) that binds F-actin was also used for counterstaining . Immunofluorescence signals were acquired under confocal microscopy ( ZEISS LSM 710 , Germany ) . Adult brain extracts were prepared according to a previously described protocol ( Lin et al . , 2010 ) . In brief , ~80 fly heads for each genotype were isolated for extract preparation . Equivalent amounts of samples ( 30 μg/20 μL/well ) were resolved by SDS-PAGE for immunoblotting . The following primary antibodies were used: rabbit anti-human LRRK2 ( Abcam , ab133474 , 1:1000 ) ; rabbit anti-phospho-LRRK2 ( Ser1292 ) ( Abcam , ab203181 , 1:500 ) ; rabbit anti-Akt ( Cell Signaling , #4691 , 1:1000 ) ; rabbit anti-Drosophila phospho-Akt Ser505 ( Cell Signaling , #4054 , 1:500 ) ; rabbit anti-Nrf2 ( Thermo Fisher Scientific , 710574 , 1:1000 ) ; rabbit anti-phospho-Nrf2 ( Ser40 ) ( Thermo Fisher Scientific , PA5-67520 , 1:1000 ) ; mouse anti-HO-1-1 ( Thermo Fisher Scientific , MA1-112 , 1:1000 ) ; rabbit anti-GAPDH ( GeneTex , GTX100118 , 1:5000 ) ; and rabbit anti-alpha tubulin ( Cell Signaling , #2144 , 1:10 , 000 ) , followed by blotting with secondary antibodies peroxidase-conjugated goat anti-rabbit IgG ( Jackson ImmunoResearch , AB_2307391 , 1:7500 ) and peroxidase-conjugated goat anti-mouse IgG ( Jackson ImmunoResearch , AB_10015289 , 1:7500 ) . The antioxidant response element ( ARE ) -GFP reporter assay was a modification of the protocol from a previous study ( Sykiotis and Bohmann , 2008 ) . ARE-GFP flies of 1 week old ( for Figure 5—figure supplement 1 ) or 6 weeks old ( Figure 6 ) were fed with regular food or food containing 0 . 1% WGE prior to brain dissection and GFP immunostaining . To quantify ARE-GFP in the 6-week-old adult fly brain , confocal images were processed in ImageJ . Mean GFP fluorescence intensities in mCherry-labeled glial cells and TH-positive dopaminergic neurons of the PPL1 cluster were quantified and normalized as GFP/mCherry . Transgenic LRRK2-G2019S mice were purchased from the Jackson Laboratory ( JAX stock #009609 , Bar Harbor , ME ) and maintained at the animal center of the National Taiwan University Hospital ( NTUH ) . Non-transgenic ( nTg ) and heterozygous transgenic LRRK2-G2019S mice were obtained by crossing heterozygous LRRK2-G2019S mice with wild-type FVB/NJ mice ( JAX stock #001800 ) . Mice at 8 . 5 months old were assigned to one of three groups ( 5–6 mice per group ) : nTg , LRRK2-G2019S , and LRRK2-G2019S fed with WGE ( 0 . 5 g/kg body weight per day ) ( Lin et al . , 2018 ) for 3 months . We employed two behavioral tests to assay mouse motor function , that is , an open-field assay to assess spontaneous locomotor activity and CatWalk XT gait analysis to assay coordination . Behavioral experiments were conducted blind to genotype , as described previously ( Lin et al . , 2020 ) . After 3 months of WGE treatment , mice were sacrificed at the age of 11 . 5 months . The substantia nigra and striatum were dissected out . The substantia nigra was subjected to immunostaining , as described previously ( Lin et al . , 2020 ) . Anti-tyrosine hydroxylase ( TH ) ( Millipore , AB152 , 1:200 ) and anti-ionized calcium-binding adapter molecule 1 ( Iba-1 ) ( GeneTex , GTX100042 , 1:200 ) were used as primary antibodies for 24 hr at 4°C . Secondary antibodies were DyLight 488 goat anti-rabbit 1:300 and Alexa Fluor 546 goat anti-rabbit at 1:200 ( 25°C for 1 hr ) . Mounting medium with DAPI ( GeneTex , GTX30920 ) was used as a counterstain . Frozen nigrostriatal brain tissues were homogenized and mixed with lysis buffer to determine protein content and for immunoblotting , as described previously ( Lin et al . , 2020 ) . The membrane was incubated overnight at 4°C with the following primary antibodies: anti-LRRK2 ( Abcam , ab133474 , 1:5000 ) , anti-phospho-LRRK2 ( Ser1292 , Abcam , ab203181 , 1:1000 ) , anti-Smad2 ( Cell Signaling , #5339 , 1:1000 ) , anti-phospho-Smad2 ( Ser465/467 , Cell Signaling , #3108 , 1:1000 ) , anti-Smad3 ( Cell Signaling , #9523 , 1:1000 ) , anti-phospho-Smad3 ( Ser423/425 , Abcam , ab52903 , 1:1000 ) , and anti-beta actin ( Sigma-Aldrich , A5441 , 1:5000 ) . After washing , peroxidase-conjugated goat anti-rabbit IgG ( GeneTex , GTX213110-01 , 1:5000 ) or peroxidase-conjugated goat anti-mouse IgG ( GeneTex , GTX213111-01 , 1:5000 ) were used as secondary antibodies . All statistical analyses were carried out in GraphPad Prism 6 ( La Jolla , CA ) . Data are presented as mean ± standard error ( SEM ) . Statistical analysis was performed using either Student t-test or one-way analysis of variance ( ANOVA ) followed by Tukey’s multiple comparison test . p<0 . 05 were considered indicative of significance . For exact n numbers , p-values , F-values , t-values , and degrees of freedom of each statistical test , please see the statistical information in Source data 1 . | Parkinson’s disease is a brain disorder that leads to tremors and difficulties with balance and coordination . These symptoms are caused by the loss of neurons which release a chemical messenger that is needed to regulate movement called dopamine . Most treatments for this disease work by boosting levels of dopamine in the brain , but this can lead to severe side effects and these drugs often become less effective over time . A traditional Chinese medicine called Gastrodia elata Blume ( or GE for short ) has previously been reported to relieve symptoms of Parkinson’s disease in both human and animal studies when administered as a decoction or formula . However , it is unclear how GE protects dopamine-producing neurons and if this mechanism involves another type of brain cell known as glia that has also been linked to Parkinson’s disease . To investigate , Lin et al . studied fruit flies and mice that carry a genetic mutation that produces the symptoms and molecular characteristics of Parkinson’s disease . The experiments showed that when the flies and mice were fed food containing water extracts of GE , they experienced less difficulties moving and had a higher number of intact dopamine-producing neurons . Lin et al . found that GE switched on a protein in glial cells located near dopamine-producing neurons . Activation of this protein , called Nrf2 , inhibited a signaling pathway in degenerating neurons that leads to the disease state . As a result , less dopamine-producing neurons were damaged and the animals’ coordination and balance were maintained . These findings suggest that GE could potentially provide an alternative or complementary therapy for Parkinson’s disease , although it still needs to be studied further in humans . If the same effect is observed , the specific compounds in GE that have this protective effect could be isolated and analyzed to see if they could be used for treatment . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2021 | Glial Nrf2 signaling mediates the neuroprotection exerted by Gastrodia elata Blume in Lrrk2-G2019S Parkinson’s disease |
Leaf senescence is an essential part of the plant lifecycle during which nutrients are re-allocated to other tissues . The regulation of leaf senescence is a complex process . However , the underlying mechanism is poorly understood . Here , we uncovered a novel and the pivotal role of Arabidopsis HDA9 ( a RPD3-like histone deacetylase ) in promoting the onset of leaf senescence . We found that HDA9 acts in complex with a SANT domain-containing protein POWERDRESS ( PWR ) and transcription factor WRKY53 . Our genome-wide profiling of HDA9 occupancy reveals that HDA9 directly binds to the promoters of key negative regulators of senescence and this association requires PWR . Furthermore , we found that PWR is important for HDA9 nuclear accumulation . This study reveals an uncharacterized epigenetic complex involved in leaf senescence and provides mechanistic insights into how a histone deacetylase along with a chromatin-binding protein contribute to a robust regulatory network to modulate the onset of plant aging .
Age-dependent organ and tissue dysfunction is detrimental to all organisms . Leaf senescence is an integral part of the plant lifecycle . Although efficient senescence is important to increase plant viability in the next generation , premature senescence often causes a decrease in the yield and quality of crops ( Avila-Ospina et al . , 2014; Distelfeld et al . , 2014; Guo and Gan , 2014 ) . Thus , the knowledge of mechanisms underlying leaf aging has profound implications in many biotechnological applications , including increasing plant productivity and preventing post-harvest loss during transportation and storage . Regulation of leaf senescence is a complex process controlled by developmental and environmental signaling pathways ( Lim et al . , 2007; Woo et al . , 2013; Schippers , 2015 ) . Many senescence-associated genes ( SAGs ) and transcription factors have been identified ( Gepstein et al . , 2003; Buchanan-Wollaston et al . , 2005; Breeze et al . , 2011; Guo and Gan , 2012 ) . However , their in vivo function in senescence remains largely unknown . For the relatively well-studied senescence-associated transcription factors , current knowledge of their function is mostly derived from knockout mutants , transgenic overexpressing plants , or identification of downstream target genes . Little is known how these transcription factors are regulated and function mechanistically in the global control of leaf senescence . Epigenetic modification is an important gene regulatory mechanism in eukaryotic organisms and plays critical roles in diverse biological processes , including genome stability and integrity , normal growth and development , diseases , and stress responses ( Kawashima and Berger , 2014; Matzke and Mosher , 2014; Pikaard and Mittelsten Scheid , 2014 ) . Histone ( de ) acetylation plays important roles in genome expression , organization , and function through the coordinated activities of histone acetyltransferases and deacetylases ( Haberland et al . , 2009; Wang et al . , 2014; Verdin and Ott , 2015 ) . While acetylation is often associated with active transcription , histone deacetylases ( HDACs ) are generally considered transcriptional repressors that remove acetylation and induce chromatin compaction ( Verdin and Ott , 2015 ) . HDACs are highly conserved enzymes in eukaryotes . The flowering plant Arabidopsis thaliana has eighteen annotated histone deacetylases that are grouped into three families: twelve RPD3-like ( REDUCED POTASSIUM DEPENDENCE 3 ) , two SIR2 ( SILENT INFORMATION REGULATOR 2 ) , and four plant-specific HD2 ( HISTONE DEACETYLASE 2 ) based on sequence similarity and cofactor dependence ( Pandey et al . , 2002 ) . Genetic studies have revealed the critical function of HDACs in crosstalk between plant genomes and the environment in plant responses to diverse stresses at the cellular and organismal levels ( Krogan and Long 2009; Kim et al . 2012; Pikaard and Mittelsten Scheid , 2014; Wang et al . , 2014 ) . Certain HDACs ( e . g . HDA6 and HDA19 ) also function in genome integrity and gene silencing ( Murfett et al . , 2001; Probst et al . , 2004; May et al . , 2005; Earley et al . , 2006; Vaillant et al . , 2007; Tanaka et al . , 2008; Pontvianne et al . , 2013 ) . Interplays between HDACs and other epigenetic modifications have also been documented . For example , HDA6 is important for DNA methylation ( Aufsatz et al . , 2002; Earley et al . , 2010; To et al . , 2011; Liu et al . , 2012; Blevins et al . , 2014; Stroud et al . , 2013 ) . Functional disruption of HDACs often causes pleiotropic abnormalities in plant growth and development ( Krogan and Long , 2009; Kim et al . , 2012; Pikaard and Mittelsten Scheid , 2014; Wang et al . , 2014 ) . For instance , early studies using an antisense approach to knockdown histone deacetylases suggest a potential role of histone deacetylation in leaf senescence ( Tian and Chen , 2001 ) . HDA6 is implicated in leaf senescence by downregulation of two SAGs in the loss-of-function mutants ( Wu et al . , 2008 ) . However , the underlying mechanism through which HDA6 and other HDACs act in leaf senescence is unknown . Histone deacetylase 9 ( HDA9 ) is a RPD3 type deacetylase , closely related to mammalian HDAC3 ( Pandey et al . , 2002 ) . Previous genetic mutational studies have established important roles of HDA9 in flowering ( Kim et al . , 2013; Kang et al . , 2015 ) , seed germination ( van Zanten et al . , 2014 ) , and salt and drought stress ( Zheng et al . , 2016 ) . However , the composition , regulation , and mechanistic action of HDA9 are unknown . In this study , we report the identification and characterization of a previous uncharacterized repressive complex containing HDA9 and a SANT domain-containing protein POWERDRESS ( PWR ) as novel regulators of leaf senescence . HDA9 promotes the onset of age-related and dark-induced leaf senescence by regulating the expression of genes involved in senescence . Our genome-wide profiling of HDA9 occupancy reveals that HDA9 directly binds to the promoters of key negative regulators and this binding requires PWR . PWR physically interacts with HDA9 and loss-of-function pwr mutants phenocopy hda9 , indicating that this complex is biologically relevant . Furthermore , we demonstrate that PWR is important for HDA9 nuclear accumulation as HDA9 protein level is significantly reduced in the nucleus in pwr mutants . Thus , we propose that PWR acts at multiple levels to regulate HDA9 function . To our knowledge , this is the first genome-wide study on targeting and regulating mechanism for a histone deacetylase in plants . Together , this study reveals an uncharacterized epigenetic complex involved in leaf senescence and provides mechanistic insights into how a histone deacetylase along with a chromatin-binding protein contribute to a robust regulatory network to promote the onset of plant aging .
To gain mechanistic insights into HDA9 action , we identified the protein complex associated with HDA9 by performing immunoaffinity purification followed by multidimensional protein identification technology mass spectrometry ( IP-MS ) . We generated Arabidopsis transgenic plants expressing HDA9-3xFLAG driven by the native HDA9 promoter in hda9 mutant background ( pHDA9::HDA9-3xFLAG/hda9 , abridged as HDA9-FLAG , Figure 1—figure supplement 1A ) . HDA9-FLAG rescued the dwarf phenotype of hda9 ( Figure 1—figure supplement 1B ) ( Kang et al . , 2015 ) , suggesting that HDA9-FLAG is functional in vivo . As a control , the same purification was performed in parallel with wild type Col-0 ( WT ) . Our IP-MS analysis revealed 51 unique HDA9 peptides and also identified a peptide corresponding to a known HDA9-interacting protein AHL22 ( Yun et al . , 2012 ) ( Figure 1A , Figure 1—source data 1 ) . Besides HDA9 , the most enriched protein in our MS is a SANT domain-containing protein POWERDRESS ( PWR ) with 27 unique peptides ( Figure 1A ) . To confirm the PWR interaction , we generated transgenic plants expressing PWR-3xFLAG driven by its endogenous promoter ( pPWR::PWR-3xFLAG , abridged as PWR-FLAG ) in WT background . Reciprocal IP-MS of PWR-FLAG also purified HDA9 ( Figure 1A , Figure 1—source data 1 ) . To further validate the HDA9-PWR interaction , we performed co-immunoprecipitation ( co-IP ) experiments in F1 Arabidopsis plants expressing both HA-tagged HDA9 and FLAG-tagged PWR . When we pulled down PWR with anti-FLAG beads , we detected the co-precipitation of HDA9 with an anti-HA antibody ( Figure 1B ) . 10 . 7554/eLife . 17214 . 003Figure 1 . HDA9 interacts with PWR and WRKY53 . ( A ) Summary of partial proteins associated with HDA9 and PWR identified by affinity purification and mass spectrometry analysis . Coverage indicates the percentage of full-length protein covered by identified unique peptides . Unique peptides indicate the number of identified peptides that are mapped to an individual protein . ( B ) Co-immunoprecipitation of HDA9 and PWR using Arabidopsis F1 plants expressing both HDA9-HA and PWR-FLAG . ( C ) In vitro pull down assay of GST-WRKY53 and HDA9-FLAG . Two technical replicates were performed ( rep1 and rep2 ) . GST protein serves as a control . DOI: http://dx . doi . org/10 . 7554/eLife . 17214 . 00310 . 7554/eLife . 17214 . 004Figure 1—source data 1 . List of proteins identified by IP-MS in HDA9 and PWR . DOI: http://dx . doi . org/10 . 7554/eLife . 17214 . 00410 . 7554/eLife . 17214 . 005Figure 1—figure supplement 1 . HDA9-FLAG protein is functional in Arabidopsis . ( A ) Determination of HDA9 protein level in transgenic Arabidopsis expressing pHDA9::HDA9-3xFLAG/hda9 ( HDA9-FLAG ) . Top panel , schematic diagram of HDA9-FLAG protein; bottom panel , detection of HDA9-FLAG protein with FLAG antibody . Coomassie staining of large subunit of rubisco as a loading control . ( B ) HDA9-FLAG rescued the dwarf phenotype of hda9 mutant . DOI: http://dx . doi . org/10 . 7554/eLife . 17214 . 005 Our IP-MS also showed the co-purification of the WRKY53 transcription factor with HDA9 ( Figure 1A ) . WRKY53 is induced at the early stage of leaf senescence and promotes the onset of senescence ( Miao and Zentgraf , 2007 ) . To confirm HDA9-WRKY53 interaction , we expressed and purified GST tagged full-length WRKY53 protein from E . coli , incubated with HDA9 protein purified from Arabidopsis HDA9-FLAG transgenic plants , and performed an in vitro GST pull down assay . HDA9-FLAG was pulled down by GST-WRKY53 but not GST itself ( Figure 1C ) , suggesting that WRKY53 interacts with HDA9 . The physical association of PWR with HDA9 led us to propose that PWR is important for HDA9 activity and function in vivo . Previous studies revealed that HDA9 is critical for deacetylation of H3K9 ( H3K9ac ) and H3K27 ( H3K27ac ) in vivo ( Kim et al . , 2013; van Zanten et al . , 2014 ) . Accordingly , our immunoblotting assays revealed increased H3K9ac and H3K27ac levels in hda9 mutant , but not in the HDA9-FLAG complementation plants ( Figure 2A , Figure 2—figure supplement 1 ) . Given that PWR physically interacts with HDA9 , we examined whether PWR is important for H3K9ac and H3K27ac deacetylation . Similar as hda9 , loss-of-function pwr mutant induces global H3K9 and H3K27 hyperacetylation ( Figure 2B ) . We next investigated the genetic interaction between HDA9 and PWR by generating an hda9 pwr double mutant and found substantial increases of H3K9ac and H3K27ac in hda9 pwr double mutants compared to WT ( Figure 2B ) . Consistent with our data that PWR functions together with HDA9 ( Figure 1 ) , no significant differences in the increase of H3K9ac and H3K27ac levels were noted between single and hda9 pwr double mutants ( Figure 2B ) . 10 . 7554/eLife . 17214 . 006Figure 2 . Loss-of-function hda9 and pwr mutants induce H3K9 and H3K27 hyperacetylation . ( A ) Immunoblots of histone acetylation marks in hda9 , pwr , and HDA9-FLAG early senescent leaves . ( B ) Immunoblots of histone acetylation marks in hda9 , pwr , and hda9 pwr early senescent leaves . ( C ) Overlap of H3K27ac increased peaks in hda9 and pwr identified by ChIP-seq . Fisher’s exact test was used to calculate the p-value . ( D ) Genomic distribution of H3K27ac increased peaks in hda9 and pwr . ( E ) Metaplots of the H3K27ac distribution on genes in WT , hda9 , and pwr . Black bar in the X-axis represents all genes in the genome . TSS , Transcription Start Sites; TTS , Transcription Terminal Sites; −2K and +2K represent 2 kb upstream of TSS and 2 kb downstream of TTS , respectively . The Y-axis represents read density of H3K27ac ChIP-seq . ( F ) Browser snapshots of representative loci with increased H3K27ac in hda9 and pwr . DOI: http://dx . doi . org/10 . 7554/eLife . 17214 . 00610 . 7554/eLife . 17214 . 007Figure 2—figure supplement 1 . HDA9 preferentially removes acetylation on histone H3 tail in vivo . Two biological replicates ( rep1 and rep2 ) were performed . DOI: http://dx . doi . org/10 . 7554/eLife . 17214 . 007 To identify the specific hyperacetylated regions in hda9 and pwr , we performed H3K27ac chromatin immunoprecipitation followed by sequencing ( ChIP-seq ) in hda9 and pwr mutants . Consistent with the immunoblotting , we identified 11 , 372 and 7687 H3K27ac increased peaks in hda9 and pwr , respectively ( Figure 2C ) . We found that ~90% ( 6901 peaks ) of pwr hyperacetylated peaks overlapped with those increased peaks in hda9 ( Figure 2C ) , suggesting that PWR and HDA9 target at similar genomic regions . Genomic distribution analysis of these hyperacetylated peaks showed that most of them are located in genic regions ( 97% for hda9 and 96% for pwr , respectively ) ( Figure 2D ) , near the transcription start sites ( Figure 2E , F ) . Together , these results suggest that PWR and HDA9 mediate deacetylation of H3K27ac at similar genomic regions . HDACs are generally considered transcriptional co-repressors associated with silent genes . To identify the in vivo binding pattern of HDA9 , we determined the genomic occupancy of HDA9 using ChIP-seq in plants expressing HDA9-FLAG . The ChIP-seq was performed in parallel with WT . HDA9 is highly enriched in gene-rich euchromatic regions , but depleted in repeat-rich centromeric heterochromatin ( Figure 3A ) . Further analysis identified a total of 9489 binding peaks ( p=1e-03 ) corresponding to 8232 genes ( Figure 3B , Figure 3—source data 1A ) . The majority of HDA9 binding peaks ( 6515 or approximately 69% ) were located in promoter regions ( Figure 3B ) . Similar observations were reported in a genome-wide profiling of mammalian HDACs ( Wang et al . , 2009 ) . We next examined the relationship between HDA9 binding and gene expression levels . We divided all 28 , 000 Arabidopsis genes equally into five groups based on their expression levels and correlated them with HDA9 binding . Surprisingly , we found that HDA9 is preferentially enriched in the promoters of active genes but not silent genes ( Figure 3C ) . We also found that HDA9 bound genes showed significantly higher expression than the average expression levels of all genes ( Figure 3D ) . To further examine the relationship between HDA9 and active genes , we correlated HDA9 binding with DNase I hypersensitive sites that are generally associated with accessible chromatin states ( Zhang et al . , 2012 ) . We found co-localization between HDA9 binding and DNase I hypersensitive sites in gene promoters ( Figure 3E , Figure 3—figure supplement 1A ) . Thus , HDA9 is associated with active genes . 10 . 7554/eLife . 17214 . 008Figure 3 . HDA9 binds to promoters of active genes . ( A ) Chromosomal views of HDA9 distribution on five chromosomes . The Y-axis represents the log2 value of HDA9-FLAG ChIP-seq reads relative to those of untagged WT control . Chr1 , Chr2 , Chr3 , Chr4 , and Chr5 represent chromosomes 1 to 5 , respectively . Black triangles indicate the location of centromeric regions . ( B ) Genomic distribution of HDA9 binding peaks . ( C ) Metaplots of HDA9 binding levels on genes . Total genes were divided evenly into five groups based on their expression level in WT . Top 20% indicates the 20% genes with highest expression level , 81%–100% indicates the 20% genes with lowest expression level . The Y-axis represents the log2 value of HDA9-FLAG ChIP-seq reads relative to those of untagged WT control . −2K and +2K represent 2 kb upstream and downstream of TSS , respectively . ( D ) Box plots of the average expression levels of HDA9 bound genes and total genes . The Y-axis indicates log10 value of FPKM + 1 . FPKM , Fragments Per Kilobase of transcript per million mapped reads . Bars within the boxes represent the mean values . ***p<0 . 001 . ( E ) Metaplots of HDA9 binding on previously identified DH ( DNase I Hypersensitive ) sites in HDA9-FLAG and untagged WT control . Black bar in the X-axis represents DH sites . The Y-axis represents the read density of HDA9-FLAG ChIP-seq reads . ( F ) Metaplots of H3K27ac on HDA9 bound genes and non-HDA9 bound genes in WT , hda9 , and pwr . Black bar in the X-axis represents genes . The Y-axis represents read density of H3K27ac ChIP-seq reads . ( G ) Representative DNA motifs identified in HDA9 binding sites by DREME . DOI: http://dx . doi . org/10 . 7554/eLife . 17214 . 00810 . 7554/eLife . 17214 . 009Figure 3—source data 1 . HDA9 binds to active genes . ( A ) List of HDA9 bound genes . ( B ) Genes showing HDA9 binding and upregulation in hda9 . ( C ) Chromatin related proteins showing HDA9 binding . DOI: http://dx . doi . org/10 . 7554/eLife . 17214 . 00910 . 7554/eLife . 17214 . 010Figure 3—figure supplement 1 . HDA9 binds to open chromatin regions with known DNA motifs . ( A ) Representative snapshots show the correlations between HDA9 binding sites and DNase I hypersensitive sites . ( B ) DNA motifs identified in HDA9 binding peaks by DREME . Motifs with E value less than 1 . 0e-7 , peak number more than 1200 were shown . E value and peak numbers are listed at bottom of each motif . ( C ) Analysis of expression level of HDA9 bound genes . Left panel , all genes are equally divided into 5 groups based on their expression level . Middle panel , distribution of HDA9 bound genes in different groups . Right panel , distribution of genes with HDA9 binding and upregulation in hda9 in different groups . ( D ) GO analysis of genes with HDA9 binding but no significant expression change in hda9 . DOI: http://dx . doi . org/10 . 7554/eLife . 17214 . 010 To determine the biological significance of our defined HDA9 binding sites , we correlated the HDA9 binding genes with their H3K27ac levels in hda9 . By comparing the H3K27ac levels over HDA9 and non-HDA9 bound genes , we found that HDA9 bound genes showed much higher increased H3K27ac levels in hda9 compared to non-HDA9 bound genes ( Figure 3F ) . Similarly , HDA9 binding is highly correlated with pwr induced H3K27 hyperacetylation ( Figure 3F ) , suggesting that PWR and HDA9 mediate deacetylation of H3K27ac at similar genomic regions . We next searched for putative DNA-binding motifs for HDA9 binding peaks using the DREME algorithm ( Bailey , 2011 ) and identified seven significantly enriched consensus motifs ( cut off p<1e-07 and minimum of 1200 peaks ) ( Figure 3—figure supplement 1B ) . Among them , 1233 HDA9 binding peaks ( 13% ) showed the significant enrichment of G-box ( CACGTG ) motif ( p=1 . 8e-129 ) ( Figure 3G ) that is present mostly in the promoter regions of genes and recognized by transcription factors ( Menkens et al . , 1995 ) . PIF4/5 is one of the G-box binding proteins previously shown to be involved in leaf senescence ( Sakuraba et al . , 2014 ) . In addition , we identified a W-box motif ( TTGAC/T ) , recognized by WRKY family transcription factors ( Rushton et al . , 2010 ) , as another putative HDA9 recognition motif ( 1328 peaks , p=1 . 9e-8 ) ( Figure 3G ) , consistent with our observation that WRKY53 is co-purified with HDA9 ( Figure 1A , C ) . The HDA9-WRKY53 interaction and enrichment of WRKY binding motif in HDA9 binding sites led us to investigate a potential role of HDA9 in leaf senescence . We examined the leaf yellowing in hda9 T-DNA knockout mutants and WT plants . After five weeks of growth , we observed the tips of older leaves in WT yellowed sooner than those of hda9 ( Figure 4A ) . By measuring the number of days from germination to leaf tip yellowing , we found that hda9 mutant leaves became senescent at 39 days , significantly later than WT ( 35 days , p=4e-06 ) and the HDA9-FLAG complementation plants ( 37 days , p=0 . 025 ) ( Figure 4B ) . Consistent with the role of HDA9 in the early stage of senescence , we found slightly elevated HDA9 protein in early senescent leaves ( Figure 4—figure supplement 1 ) . 10 . 7554/eLife . 17214 . 011Figure 4 . HDA9 and PWR act in the same pathway to promote leaf senescence . ( A ) Phenotypic analysis of leaves from 5-week-old plants of wild type ( WT ) , hda9 , and HDA9-FLAG plants expressing HDA9-FLAG driven by the native HDA9 promoter in hda9 mutant background . Rosette leaves were numbered from bottom to top with the first leaf being the oldest and 20th being the youngest . ( B ) Quantification of days from germination to onset of leaf senescence in WT , hda9 and HDA9-FLAG . Points ( round , square , or triangle ) represent the number of days for an individual plant to reach onset of senescence . Error bars represent standard deviation for at least 30 tested plants . ( C ) Dark treatment of the 5th and 6th leaves detached from 3-week-old plants of hda9 , pwr , and hda9 pwr double mutants . ( D ) Chlorophyll content measurement of leaves from ( C ) . Error bars represent a standard deviation for three biological replicates . ( E ) Leaf senescence phenotype of 5-week-old pwr mutant . The oldest ten leaves are shown . ( F ) Quantification of days from germination to the onset of leaf senescence in WT and pwr . Points ( round or square ) represent the number of days for an individual plant to reach onset of senescence . Error bars represent standard deviations for at least 30 tested plants . Student’s t-tests were used to calculate the p values . *p<0 . 05 , ***p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 17214 . 01110 . 7554/eLife . 17214 . 012Figure 4—figure supplement 1 . HDA9-FLAG shows elevated protein accumulation in early senescent leaves . Western blots of HDA9-FLAG protein levels in young leaf ( YL ) , mature leaf ( ML ) , early senescent leaf ( ES ) , and late senescent leaf ( LS ) . Numbers at the bottom represent the intensity of HDA9-FLAG signals normalized to those of histone H3 . DOI: http://dx . doi . org/10 . 7554/eLife . 17214 . 012 Leaf senescence can also be induced by environmental stresses such as darkness ( Lim et al . , 2007 ) . To examine the role of HDA9 in dark-induced leaf senescence , we analyzed leaf yellowing of the fifth and sixth rosette leaves detached from the plants of WT and hda9 . Dark-induced leaf yellowing was attenuated in the hda9 mutant relative to WT ( Figure 4C ) . Consistent with the visible phenotype , hda9 mutant leaves had greater total chlorophyll content than WT after dark treatment ( Figure 4D ) . Thus , HDA9 promotes the onset of both age-related and dark-induced leaf senescence . Given that PWR physically interacts with HDA9 , we examined whether PWR also contributes to leaf senescence . Loss-of-function pwr mutants phenocopy hda9 , showing delayed senescence of both naturally aged and dark-treated leaves ( Figure 4C–F ) . Furthermore , we found that there was no noticeable difference in the degree of delayed leaf yellowness between the hda9 pwr double and the respective single mutants ( Figure 4C ) . The similar leaf senescence phenotypes of hda9 , pwr , and hda9 pwr mutants were further supported by their similar retention of chlorophyll content ( Figure 4D ) . These observations indicate that HDA9 and PWR act in the same pathway to promote leaf senescence . The onset of leaf senescence is often accompanied by increased expression of senescence-associated genes ( SAGs ) and decreased expression of senescence downregulated genes ( SDGs ) ( Gepstein et al . , 2003; Breeze et al . , 2011; Brusslan et al . , 2015 ) . Consistent with the delayed leaf senescence phenotype in hda9 , we found the downregulation of hallmark SAGs including SENESCENCE4 ( SEN4 ) , SAG12 , and SAG113 in hda9 ( Figure 5A ) . To further examine the role of HDA9 in senescence , we performed whole transcriptome analysis by mRNA sequencing ( RNA-seq ) in hda9 mutant and WT of two week-old young leaves ( YL ) as well as early senescence ( ES ) leaves . A previous study identified differential regulation of 3474 SAGs and 2849 SDGs during different stages of leaf senescence ( Breeze et al . , 2011 ) . Although the expression of 30% of the SAGs ( 1023 out of 3474 ) increased in senescing leaves of both hda9 and WT compared to the young leaves ( Figure 5B ) , the fold change was significantly less in hda9 compared to WT ( Figure 5C ) . Similarly , decreased expression levels of 575 SDGs ( 20% of 2849 ) were impaired in hda9 senescing leaves ( Figure 5—figure supplement 1A , B ) , suggesting that HDA9 regulates the expression of senescence-related genes to promote leaf senescence . Besides SAGs and SDGs , many abscisic acid ( ABA ) response genes known to promote the onset of leaf senescence , were significantly downregulated ( Figure 5—figure supplement 1C ) , further supporting a role for HDA9 in leaf senescence . 10 . 7554/eLife . 17214 . 013Figure 5 . HDA9 and PWR regulate expression of the same group of genes involved in leaf senescence . ( A ) Expression of senescence marker genes in hda9 by quantitative RT-PCR . Error bars represent a standard deviation from two biological replicates . ( B ) Heatmaps show expression of senescence-associated genes ( SAGs ) in young leaf ( YL ) and early senescence leaf ( ES ) in WT and hda9 . The color bar on the right indicates the Z-score . ( C ) Boxplots show the less increased expression of SAGs in hda9 than WT in ES . The Y-axis represents log2 value of fold change of expression levels of SAGs between ES and YL . ( D ) Overlap of upregulated genes in hda9 and HDA9 bound genes . ( E ) Scatter plots show the expression of HDA9 bound genes in WT and hda9 . ( F ) Overlap of differentially expressed genes in RNA-seq of hda9 and pwr . Fisher’s exact test was used to calculate the p-value . ( G ) Quantitative RT-PCR confirming the upregulation of WRKY57 and APG9 in hda9 and pwr . Relative expression was calculated as relative to ACTIN7 , and then normalized to WT . Error bars represent a standard deviation from two biological replicates . *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 17214 . 01310 . 7554/eLife . 17214 . 014Figure 5—source data 1 . Differentially expressed genes in hda9 . DOI: http://dx . doi . org/10 . 7554/eLife . 17214 . 01410 . 7554/eLife . 17214 . 015Figure 5—source data 2 . Differentially expressed genes in pwr . DOI: http://dx . doi . org/10 . 7554/eLife . 17214 . 01510 . 7554/eLife . 17214 . 016Figure 5—source data 3 . List of overlapped genes showing differential expression in both hda9 and pwr . DOI: http://dx . doi . org/10 . 7554/eLife . 17214 . 01610 . 7554/eLife . 17214 . 017Figure 5—figure supplement 1 . HDA9 and PWR regulate similar group of genes involved in leaf senescence . ( A ) Heatmaps show expression of senescence downregulated genes ( SDGs ) in young leaf ( YL ) and early senescence leaf ( ES ) in WT and hda9 . The color bar on the right indicates the Z-score . ( B ) Boxplots show the less decreased expression of SDGs in hda9 than WT in ES . The Y-axis represents log2value of fold change of expression levels of SDGs between ES and YL . Bars within the boxes represent the mean values . ***p<0 . 001 . ( C ) Heatmaps show expression of selected ABA responsive genes in YL and ES in WT and hda9 . The color bar on the left indicates the Z-score . ( D ) A volcano plot of differentially expressed genes in ES in hda9 . The Y-axis represents log10 of p value and the X-axis represents log2value of fold change in hda9 compared to WT . Genes were considered significantly downregulated or upregulated if p<0 . 05 . ( E ) GO analysis of differentially expressed genes in hda9 . ( F ) Venn diagram of genes with HDA9 binding and upregulated in hda9 and genes with increased H3K27ac in hda9 . ( G ) Schematic symbols of candidate genes for molecular analysis selected from HDA9 bound genes with upregulation in hda9 . ΔAlthough the increased expression of WRKY57 is not defined as significant ( p<0 . 05 ) in hda9 ( p=0 . 07 ) , we confirmed the upregulation with RT-qPCR in both hda9 and pwr . ( H ) A volcano plot of differentially expressed genes in ES in pwr . The Y-axis represents log10 of p value and the X-axis represents log2value of fold change in pwr compared to WT . Genes were considered significantly downregulated or upregulated if p<0 . 05 . ( I ) GO analysis of upregulated and downregulated genes in both hda9 and pwr . DOI: http://dx . doi . org/10 . 7554/eLife . 17214 . 017 We identified 782 upregulated and 656 downregulated genes in hda9 early senescent leaves compared to WT ( Figure 5—figure supplement 1D , Figure 5—source data 1 ) . These differentially expressed genes showed enrichment in senescence related pathways including jasmonic acid ( JA ) and ABA response ( Figure 5—figure supplement 1E ) . To further explore the biological significance of HDA9 binding , we examined the correlation between HDA9 binding and altered gene expression in hda9 . We found 222 genes ( ~28% ) were shown to be bound by HDA9 and significantly upregulated in hda9 ( Figure 5D , Figure 3—source data 1B ) , suggesting that they are directly regulated by HDA9 . Of the 222 genes , 151 ( ~68% ) showed increased H3K27ac in hda9 , indicating a positive correlation between gene upregulation and H3K27ac increase of HDA9 targets ( p=1 . 2e-10 ) ( Figure 5—figure supplement 1F ) . Interestingly , the majority of HDA9 bound genes did not show significant expression change in hda9 ( Figure 5D , E ) . Similar observations are reported in human ( Wang et al . , 2009 ) and maize ( Yang et al . , 2016 ) . It is possible that HDA9 is not a primary regulator of transcription at the majority of its targets . Another possibility is that other HDACs or chromatin/transcriptional repressors function together with HDA9 to regulate gene expression , and thus loss of HDA9 itself is insufficient to release the transcriptional repression of its targets . Given the function of HDA9 in leaf senescence , we sought to investigate whether HDA9 directly regulate genes involved in this process . By searching the 222 genes with HDA9 binding and upregulation in hda9 , we found 11 genes with potential or known functions in senescence ( Figure 5—figure supplement 1G ) , including catalase that protects cells from oxidative damage ( CAT1 ) ( Du et al . , 2008 ) , autophagy proteins that delay senescence and programmed cell death ( APG9 , ATG2 , ATG8E and ATG13 ) ( Hanaoka et al . , 2002; Yoshimoto et al . , 2004; Suttangkakul et al . , 2011; Wang et al . , 2011b ) , proteins that negatively regulate ABA signaling pathway known to promote senescence ( NPX1 , PLL5 , AFP2 , AFP4 ) ( Schweighofer et al . , 2004; Huang and Wu , 2007; Garcia et al . , 2008; Kim et al . , 2009b ) , BIK1 that negatively regulates the salicylic acid ( SA ) signaling pathway ( Veronese et al . , 2006 ) , and WRKY57 ( a WRKY family transcription factor ) acts as a negative regulator of JA to prevent leaf senescence ( Jiang et al . , 2014 ) . To further dissect the functional relationship of PWR with HDA9 in leaf senescence , we performed RNA-seq in pwr and identified 887 upregulated and 860 downregulated genes relative to WT in ES leaves ( Student’s t test , p<0 . 05 ) ( Figure 5—figure supplement 1H , Figure 5—source data 2 ) . Of the affected genes in hda9 , 277 out of 782 upregulated genes ( 38% ) and 354 out of 656 downregulated genes ( 54% ) showed up- or downregulation in pwr , respectively ( Figure 5F , Figure 5—source data 3 ) . The number is much larger than expected by chance ( Fisher’s exact test , p<2 . 2e-16 ) . GO analysis of the overlapping genes showed enrichment in developmental and environmental stress and ABA signaling pathways ( Figure 5—figure supplement 1I ) , indicating that PWR and HDA9 regulate the expression of the same group genes in leaf senescence . NPX1 , one of the HDA9 direct targets , also showed significant upregulation in pwr ( Figure 5—source data 3 ) . Besides NPX1 , we also found upregulation of APG9 and WRKY57 in pwr . However , they are not defined to be significant based on our significance criteria . To further examine their expression , we performed RT-qPCR with additional biological replicates and confirmed that APG9 and WRKY57 were significantly upregulated in pwr ( Figure 5G ) . All together , these results together with the physical interaction of HDA9 and PWR ( Figure 1 ) support the notion that PWR and HDA9 act in the same pathway to promote leaf senescence . We have confirmed PWR as a functional partner of HDA9 ( Figures 4 and 5 ) . To further dissect the molecular mechanism of PWR on HDA9 function , we examined whether PWR directly binds the same targets as HDA9 . We performed ChIP-qPCR in PWR-FLAG plants and found that 9 of the 11 randomly chosen HDA9 bound loci showed significant enrichment of PWR ( Figure 6—figure supplement 1A ) . Furthermore , PWR specifically binds to the same genomic regions within APG9 , WRKY57 , and NPX1 where HDA9 are enriched ( Figure 6A ) . Next , we asked whether PWR affected histone acetylation on the same targets of HDA9 . We performed ChIP-qPCR and found that H3K27ac levels were significantly increased in hda9 and pwr mutants at WRKY57 , APG9 , and NPX1 ( Figure 6B ) . To further examine whether pwr induced H3K27 hyperacetylation is correlated with HDA9 genome-wide binding , we compared the H3K27ac levels of HDA9 bound genes over non-HDA9 bound genes , and found that HDA9 bound genes showed a significantly higher increase of H3K27ac relative to non-HDA9 bound genes in pwr ( p<2 . 2e-16 ) ( Figure 6C ) . Together , these results suggest that PWR binds to the same genomic regions as HDA9 at HDA9 targets . 10 . 7554/eLife . 17214 . 018Figure 6 . HDA9 nuclear accumulation and chromatin association are dependent on PWR . ( A ) ChIP-qPCR shows that PWR is enriched at the same genomic regions of HDA9 targets . Upper panel illustrates snapshots of HDA9 binding at APG , NPX1 , and WRKY57 . ChIP-qPCR value of PWR was normalized to WT control . Primer positions are indicated with P1 , P2 , and P3 . Error bars represent a standard deviation from two biological replicates . ( B ) ChIP-qPCR shows H3K27ac levels at WRKY57 , APG9 , and NPX1 in WT , hda9 , and pwr mutants . ChIP-qPCR value of H3K27ac was normalized to WT control . Error bars represent a standard deviation from two biological replicates . ( C ) Box plots of H3K27ac levels on HDA9 bound genes and non-HDA9 bound genes in WT , hda9 , and pwr . The Y-axis represents FPKM of H3K27ac ChIP-seq reads . Student’s t test , ***p<0 . 001 . ( D ) ChIP-qPCR shows HDA9 enrichment on WRKY57 , APG9 , and NPX1 in HDA9-FLAG and HDA9-FLAG/pwr plants . TA3 is a transposable element that serves as a negative control . Error bars represent a standard deviation from two biological replicates . *p<0 . 05 , **p<0 . 01 . ( E ) Detection of HDA9-FLAG protein in total ( T ) , cytoplasmic ( C ) , and nuclear ( N ) extracts in HDA9-FLAG/hda9 and HDA9-FLAG/hda9 pwr . DOI: http://dx . doi . org/10 . 7554/eLife . 17214 . 01810 . 7554/eLife . 17214 . 019Figure 6—figure supplement 1 . PWR is required for HDA9 recruitment to targets . ( A ) ChIP-qPCR for PWR enrichment on randomly chosen HDA9 targets . CK1 ( AT1G70380 ) , CK2 ( AT2G31425 ) , and TA3 ( AT1TE46405 ) are negative controls without HDA9 binding . Error bars represent a standard deviation from two biological replicates . ( B ) Western blot of HDA9-FLAG protein in WT and pwr background . ( C ) ChIP-qPCR for HDA9 enrichment on randomly chosen HDA9 targets in HDA9-FLAG/hda9 and HDA9-FLAG/hda9 pwr . Error bars represent a standard deviation for two biological replicates . Student’s t-test , *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 17214 . 019 Given the loss-of-function pwr mutation induced global H3K9 and H3K27 hyperacetylation ( Figure 2 ) , one possible role for PWR is the recruitment of HDA9 to target loci . To test this hypothesis , we crossed an HDA9-FLAG line into the pwr mutant and confirmed that the overall HDA9 protein level was not affected in pwr ( Figure 6—figure supplement 1B ) . We then performed ChIP-qPCR to determine the chromatin association of HDA9 in the absence of PWR . We found that the enrichment of HDA9 at WRKY57 , APG9 , NPX1 , and four other randomly selected loci was substantially decreased in pwr ( Figure 6D , Figure 6—figure supplement 1C ) , suggesting that HDA9 binding to these targets requires PWR . HDA9 needs to be imported in the nucleus for its histone deacetylation activity . The abolishment of HDA9 chromatin association in pwr promotes us to examine whether PWR is important for HDA9 nuclear accumulation . We performed a nuclear-cytoplasmic fractionation assay and found that HDA9 was present both in the cytoplasm and in the nucleus ( Figure 6E ) . Interestingly , HDA9 accumulation in the nucleus was greatly reduced in pwr mutants compared to the plants with PWR ( Figure 6E ) . The similar accumulation of HDA9 in the total extracts of WT and pwr indicates that PWR is important for HDA9 nuclear accumulation .
Leaf senescence is a complex process regulated by multiple pathways . In this study , we found that many ABA-responsive genes were downregulated in hda9 ( Figure 5—figure supplement 1C ) , indicating that the ABA signaling pathway is impaired in hda9 during leaf senescence . This is consistent with a previous study showing insensitivity to ABA-mediated seed dormancy and germination inhibition in hda9 ( van Zanten et al . , 2014 ) . ABA is known to promote leaf senescence ( Jibran et al . , 2013; Khan et al . , 2014 ) and loss-of-function of the ABA receptor PYL8 or PYL9 causes delayed leaf senescence ( Lee et al . , 2015; Zhao et al . , 2016 ) . Our ChIP-seq and RNA-seq analyses reveal that neither PYL8 nor PYL9 is a primary target of HDA9 . Instead , we found that HDA9 directly binds and represses two negative regulators of the ABA signaling pathway , NPX1 and AFP4/TMAC2 ( Figure 3—source data 1B ) . NPX1 and AFP4/TMAC2 are proposed to act as negative regulators in ABA signaling because their overexpression reduces plant sensitivity to ABA ( Huang and Wu , 2007; Kim et al . , 2009b ) . Based on these findings , we propose that HDA9 promotes leaf senescence in part by repressing the negative regulators of the ABA signaling pathway ( Figure 7 ) . Besides ABA , one negative regulator of the JA pathway , WRKY57 , is also shown to be a direct target of HDA9 ( Figures 5 and 6 , Figure 3—source data 1B ) . WRKY57 is a transcription factor that represses JA-induced leaf senescence ( Jiang et al . , 2014 ) , suggesting that HDA9 may also regulate leaf senescence through the JA signaling pathway . Autophagy is an intracellular process for protein degradation and is associated with leaf longevity ( Avila-Ospina et al . , 2014 ) . APG9 is an essential component of the plant autophagy pathway and apg9 mutants display early leaf senescence ( Hanaoka et al . , 2002 ) . Consistently , we found APG9 and several other autophagy genes ( ATG8E , ATG2 and ATG13 ) are directly targeted and repressed by HDA9 ( Figures 5 and 6 , Figure 3—source data 1B ) . Leaf senescence results in many cellular metabolic changes , including accumulation of oxidation products and reduction of antioxidant enzymes such as catalase ( Griffiths et al . , 2014 ) . We found a significant enrichment of HDA9 at CATALASE1 ( CAT1 ) and CATALASE3 ( Figure 3—source data 1 ) , encoding enzymes that decompose hydrogen peroxide ( Du et al . , 2008 ) . Our RNA-seq analysis also revealed a significant increase of CAT1 expression in hda9 ( Figure 5—source data 1B ) , suggesting that HDA9 represses CAT1 to allow the accumulation of oxidation during leaf senescence . Interestingly , CAT1 is reported to be a direct target of WRKY53 ( Miao et al . , 2004 ) , consistent with the HDA9 and WRKY53 interaction ( Figure 1 ) . Thus , HDA9 appears to act in multiple pathways to promote leaf senescence . A common theme is that HDA9 directly targets and represses the expression of negative regulators , which in turn , promotes leaf senescence ( Figure 7 ) . HDACs are generally considered as transcriptional co-repressors associated with silent genes . Unexpectedly , our genome-wide occupancy study reveals a predominant enrichment of HDA9 at the actively transcribed genes , and that the binding levels are positively correlated with gene expression ( Figure 3 ) . Similar genome-wide localization patterns were reported for HDAC1 , HDAC2 , HDAC3 , and HDAC6 in human ( Wang et al . , 2009 ) and HDA101 in maize ( Yang et al . , 2016 ) . Unlike maize HDA101 that mostly binds transcription start sites ( Yang et al . , 2016 ) , HDA9 is preferentially enriched in the promoters ( proximal to transcription start sites ) analogous to mammalian HDAC1 and HDAC3 ( Wang et al . , 2009 ) . The positive correlation between HDA9 bound regions and DNase I hypersensitive sites also supports this observation ( Figure 3E , Figure 3—figure supplement 1A ) . Different from the mammalian and maize studies , we found only a very small fraction of HDA9-enriched genes are silent genes ( ~5% ) ( Figure 3—figure supplement 1C ) . Remarkably , none of the 222 HDA9-bound genes that were upregulated in hda9 are inactive genes ( Figure 3—figure supplement 1C ) . The precise mechanism why HDA9 binds to the promoters of actively expressed genes is unclear . One explanation could be that HDA9 may be recruited to the promoters of active genes to prevent promiscuous cryptic transcription . Another possibility is that HDA9 may compete with more active HDAs for binding to the similar genomic regions . It is also unclear why silent genes are not the preferred targets of HDA9 . One possibility is that silent genes in Arabidopsis tend to have high promoter DNA methylation and thus may be repressed by DNA methylation ( unpublished data ) . Another difference is that our HDA9 affinity purification and MS analysis failed to detect Pol II subunits , which were reported to associate with HDAC in mammals ( Wang et al . , 2009 ) . We co-purified WRKY53 transcription factor with HDA9 . WRKY53 is known to act either as a transcriptional activator or transcriptional repressor in leaf senescence ( Miao et al . , 2004; Miao and Zentgraf , 2007; Miao et al . , 2013 ) . Supporting the relevance of HDA9-WRKY53 interaction , we found that HDA9 binding peaks are significantly enriched for WRKY binding motifs ( Figure 3G ) . It is possible that WRKY53 recruits HDA9 to active genes to remove the acetylation marks added by HATs to maintain their proper expression levels during senescence . Analogous to reports on mammalian and maize HDACs ( Wang et al . , 2009; Yang et al . , 2016 ) , our ChIP-seq and RNA-seq analyses also reveal that the majority of HDA9 bound genes do not alter gene expression in the absence of HDA9 . Although the mechanism is unclear , several possibilities could account for this pattern . First , the ultimate gene expression level is determined by the combined actions of multiple redundant repressors or activators , exemplified by yeast Rpd3 and chromatin remodeling enzyme Isw2 ( Fazzio et al . , 2001 ) . It is possible that other HDACs ( e . g . HDA6 and HDA19 ) or transcriptional repressors function redundantly with HDA9 to regulate gene expression , and thus loss of HDA9 itself is insufficient to release the transcriptional repression of its targets . This may also provide an explanation for the low number of mis-regulated genes in hda6 ( To et al . , 2011; Blevins et al . , 2014 ) . In support of this notion , loss-of-function hda6 mutations induce delayed leaf senescence and repress the expression of SAG12 and SEN4 ( Wu et al . , 2008 ) . The molecular basis of HDA6 in leaf senescence is unclear . It will be interesting to explore the precise relationship between HDA9 with HDA6 in targeting and regulating gene expression in leaf senescence as well as other biological processes . Second , our GO analysis reveals that HDA9-bound genes that have no significant change in RNA transcripts in hda9 are enriched for various developmental processes and environmental responses ( Figure 3—figure supplement 1D ) . It is possible that the association of HDA9 with the promoters of these developmental genes can rapidly induce the repression of their expression in response to certain internal and/or external signals . Indeed , many photosynthesis-related genes are reported to be upregulated in hda9 seeds during imbibition ( van Zanten et al . , 2014 ) . Third , HDA9 may act to regulate chromatin structure rather than acting through transcriptional regulation . We found that HDA9 binds 177 chromatin factors , including SWI/SNF chromatin remodelers , ATP-dependent helicases , methyl DNA binding proteins , Tudor/PWWP/MBT superfamily proteins , and WD-40 proteins ( Figure 3—source data 1C ) . Fourth , HDA9 may be recruited to the promoters of active genes to prevent promiscuous cryptic transcription as suggested by mammalian and maize studies ( Wang et al . , 2009; Yang et al . , 2016 ) . Finally , some of the HDA9 enriched loci may not be its bona fide targets . Further experiments will be required to determine the precise function and mechanism of HDA9 in the regulation of its target genes . Our results suggest that HDA9 , PWR and the WRKY53 transcription factor form a previously unknown complex that promotes leaf senescence . This complex may be analogous to HDAC3-SMRT/N-CoR repressor complex in animals . Although the function of HDAC3-SMRT/N-CoR in various cellular processes ( e . g . development , differentiation , and diseases ) has been well studied ( Karagianni and Wong , 2007; Shahbazian and Grunstein , 2007 ) , its role in cellular senescence and aging remains unclear in mammals . Our ChIP assay showed that PWR and HDA9 are enriched at the same loci , and HDA9 binding to these loci requires PWR in vivo ( Figure 6A , D , Figure 6—figure supplement 1C ) . This is consistent with the role of SMRT/N-CoR in targeting HDAC3 to chromatin ( Guenther et al . , 2000; Karagianni and Wong , 2007; You et al . , 2013 ) . SMRT/N-CoR has an additional function in stimulating HDAC3 deacetylase activity in vitro ( Guenther et al . , 2001; Watson et al . , 2012 ) . Despite extensive testing , we have been unable to find in vitro conditions that allow the robust HDA9 deacetylase activity . Thus , it remains to be determined whether PWR promotes HDA9 catalytic activity . On the other hand , we discovered a potential role of PWR in HDA9 nuclear localization ( Figure 6E ) . Contrast to a previous study showing that HDA9 predominantly localized in the nucleus ( Kang et al . , 2015 ) , we found a significant fraction of HDA9 protein is present in the cytoplasm ( Figure 6E ) . Such difference in observation may be due to the fact that mature leaves were examined in our experiment whereas the 10-day-old seedlings were used in the previous study . It will be interesting to examine whether HDA9 localizes differently in the cell during the different developmental stages . PWR was previously reported to promote the expression of several miRNA genes ( Yumul et al . , 2013 ) . MiRNA is also implicated in leaf senescence ( Kim et al . , 2009a; Humbeck , 2013; Li et al . , 2013; Huo et al . , 2015 ) . We wonder whether the induction of senescence by PWR also partially depends on miRNA pathways . Although we cannot completely rule out this possibility , several lines of evidence suggest it is less likely . First , the expression of miRNA genes reported in ( Yumul et al . , 2013 ) is not affected in hda9 mutants according to our RNA-seq analysis ( Figure 5—source data 1 ) . Second , hda9 pwr double mutant shows no further delay in leaf senescence than any of the hda9 or pwr single mutants , suggesting that PWR and HDA9 act through the same pathway in promoting leaf senescence . Additionally , senescence-related genes regulated by HDA9 and PWR are largely the same group of genes ( Figure 5—source data 3 ) . Thus , PWR promotes leaf senescence likely through its functional interaction with HDA9 . Taken together , our data provide molecular insights into the function and mechanism of how HDA9-PWR-WRKY53 complex integrates and coordinates multiple signaling pathways to regulate global gene expression during leaf senescence .
Arabidopsis thaliana ecotype Columbia-0 ( Col-0 ) was used as wild type ( WT ) for all experiments . The T-DNA insertion lines of hda9 ( SALK_007123 ) and pwr ( SALK_071811C ) were obtained from the Arabidopsis Biological Resource Center ( ABRC ) . Seeds were sown in soil and kept at 4°C for 2 days before transferring to 24 hr constant light at 22°C . The full-length cDNA of WRKY53 was amplified and cloned into GST tagged protein expression vector pGOOD , modified from pGEX-6P by adding 6XHIS tag at the C terminus . Genomic DNA of PWR and HDA9 with their 1 kb promoters were amplified and cloned into pENTR/D-TOPO . These constructs were recombined into the pEarleyGate302 binary vectors ( Du et al . , 2012 ) to create epitope-tagged FLAG or HA fusions and were transformed by agrobacterium-mediated infection into hda9 mutants or wild type plants . Detailed information for primers can be found in Supplementary file 1 . Rosette leaves were detached from 2-week-old plants and placed on moisturized filter paper in petri dishes . The dishes were kept in dark or constant light at 22°C for 4–5 days . Chlorophyll was extracted from leaves of dark-treated or untreated controls using 80% acetone . Briefly , the leaves were crushed in 1 ml 80% acetone and kept in dark at 4°C overnight . The chlorophyll content was determined as described previously ( Inskeep and Bloom , 1985 ) and then normalized to leaf fresh weight . For RNA-seq , RNA was extracted from the young leaves ( YL ) of 10-day-old seedlings and leaves of early senescent plants ( ES ) grown in constant light at 22°C . Total RNA was extracted with Trizol reagent ( ThermoFisher ) and treated with DNase I ( NEB ) . One microgram RNA was reverse-transcribed into cDNA with SuperScript III ( ThermoFisher ) followed by quantitative PCR assay with SYBR Green Master Mix using CFX96 Real-Time System 690 ( Bio-Rad , Hercules , CA ) . Relative transcript level to ACTIN7 was calculated with the 2-ΔCT method ( Livak and Schmittgen , 2001 ) . Detailed information for primers can be found in Supplementary file 1 . FLAG and HA epitope tags were detected with horseradish peroxidase ( HRP ) conjugated anti-FLAG ( Sigma , A8592 ) and anti-HA ( Roche , 12013819001 ) , respectively . The α-tubulin antibody is from cell signaling ( 3878 ) . The following histone antibodies were used: H3K9ac ( Millipore , 07–352 ) , H3K27ac ( Active Motif , 39133 ) , H3ac ( Active Motif , 39139 ) , H3 ( Abcam , ab1791 ) , H4K8ac ( Millipore , 07–328 ) , H4K12ac ( Millipore , 07–595 ) , H4K16ac ( Millipore , 07–329 ) , H4ac ( Active Motif , 39243 ) , H4 ( Abcam , ab7311 ) . All western blots were developed using ECL Plus Western Blotting Detection System ( GE healthcare , RPN2132 ) . RNA-seq and ChIP-seq libraries were constructed using a TruSeq RNA Library Preparation Kit ( Illumina , #RS-122–2002 ) and the Ovation Ultralow DR Multiplex System ( NuGEN , #0330 ) , respectively . Libraries were sequenced on a HiSeq2000 in the UW-Madison Biotechnology Center . Reads were aligned to the Arabidopsis reference genome ( TAIR10 ) using Bowtie2 ( v2 . 1 . 0 ) with default parameters . Reads that mapped to identical positions in the genome were collapsed into one read . Tophat ( 2 . 0 . 8b ) and Cufflink ( 2 . 1 . 1 ) were used for differential expression analysis ( Trapnell et al . , 2012 ) . The genes showing a p<0 . 05 were considered as significantly differentially expressed genes . Two biological replicates were performed for RNA-seq . Gene Ontology analysis was performed using agriGO ( http://bioinfo . cau . edu . cn/agriGO/ ) . MACS ( 1 . 4 . 2 ) was used for peak calling with p=1e-03 . BEDTools ( 2 . 17 . 0 ) and custom PERL scripts were used for further analysis . Increased H3K27ac peaks were identified by calling peaks in hda9 or pwr over WT with p=1e-03 . In HDA9 binding profiling , log2 value of normalized HDA9 ChIP reads divided by WT reads was calculated and binned in 100 kb intervals . DNase I hypersensitive ( DH ) sites were from ( Zhang et al . , 2012 ) . All statistical analysis and figures were done using R ( 3 . 2 . 3 ) . The total reads obtained for each sample are listed in Supplementary file 2 . Histone ChIP was performed as previously described ( Lu et al . , 2015 ) . A two-gram mixture of mature leaves and early senescent leaves were ground into powder in liquid nitrogen and cross-linked in Nuclei Isolation Buffer I ( 10 mM Hepes pH 8 , 1M Sucrose , 5 mM KCl , 5 mM MgCl2 , 5 mM EDTA , 0 . 6% Triton X-100 , 0 . 4 mM PMSF , and protease inhibitor cocktail tablet [Roche , 14696200] ) with 1% formaldehyde for 20 min at room temperature . The homogenate was filtered through two layers of miracloth ( Millipore , 475855 ) and pelleted by centrifuging at 4000 rpm for 25 min at 4°C . The pellet was washed with Nuclei Isolation Buffer II ( 0 . 25 M sucrose , 10 mM Tris-HCl pH 8 , 10 mM MgCl2 , 1% Triton X-100 , 1 mM EDTA , 5 mM β-mercaptoethanol , 0 . 4 mM PMSF , protease inhibitor cocktail tablet ) , then resuspended with Nuclear Lysis Buffer ( 50 mM Tris-HCl pH 8 , 10 mM EDTA , 1% SDS , 0 . 4 mM PMSF , protease inhibitor cocktail tablet ) and kept on ice for 10 min . The lysates was diluted 10-fold with ChIP Dilution Buffer ( 1 . 1%Triton X-100 , 1 . 2 mM EDTA , 16 . 7 mM Tris-HCl pH 8 , 167 mM NaCl , 0 . 4 mM PMSF , and protease inhibitor cocktail tablet ) and sheared by sonication . After centrifugation at 5000 rpm for 10 min , the supernatant was incubated with 5 μg antibody and 40 μl magnetic protein A/G beads ( Life Technologies ) overnight with rotation at 4°C . After sequential washes with Low salt Buffer ( 150 mM NaCl , 0 . 1% SDS , 1% Triton X-100 , 2 mM EDTA , 20 mM Tris-HCl pH 8 ) , High Salt Buffer ( 500 mM NaCl , 0 . 1% SDS , 1% Triton X-100 , 2 mM EDTA , 20 mM Tris-HCl pH 8 ) , LiCl Buffer ( 0 . 25M LiCl , 1% NP-40 , 1% sodium deoxycholate , 1 mM EDTA , 10 mM Tris-HCl pH 8 ) and TE Buffer ( 10 mM Tris-HCl pH 8 , 1 mM EDTA ) , the DNA-protein complex was eluted with ChIP Elution Buffer ( 1% SDS , 0 . 1M NaHCO3 ) and reverse cross-linked at 65°C for over 6 hr . After proteinase K and RNase treatment , DNA was purified by standard phenol–chloroform method for qPCR . Antibodies used in histone acetylation ChIP were the same as used in western blot . HDA9-FLAG and PWR-FLAG ChIP were slightly modified from ( Du et al . , 2012 ) . Nuclei were isolated from the two grams of a mixture of mature leaves and early senescent leaves and cross-linked in vitro using the same method as described above . After wash with Nuclei Isolation Buffer II , the nuclei was resuspended with IP Binding Buffer ( 50 mM Tris-HCl pH 8 , 150 mM NaCl , 5 mM MgCl2 , 5% Glycerol , 0 . 1% NP-40 , 1 mM PMSF , and protease inhibitor cocktail tablet ) and sheared by sonication . After centrifugation at 5000 rpm for 10 min , the supernatant was incubated with Anti-FLAG M2 magnetic beads ( Sigma , M8823 ) overnight with rotation at 4°C . After wash with IP Binding Buffer containing 500 mM NaCl , the protein-DNA complex was eluted with ChIP Elution Buffer and reverse cross-linked at 65°C for 6 hr . After proteinase K and RNase treatment , DNA was purified by standard phenol–chloroform method for qPCR analysis or sequencing . Affinity purification and mass spectrometry analysis of HDA9-FLAG and PWR-FLAG were performed as previously described ( Du et al . , 2012 ) . Approximately twenty grams of leaves from HDA9-FLAG , PWR-FLAG or WT ( negative control ) were ground into powder and homogenized in 80 ml IP Binding Buffer ( 50 mM Tris-HCl pH 8 , 150 mM NaCl , 5 mM MgCl2 , 5% Glycerol , 0 . 1% NP-40 , 1 mM DTT , 1 mM PMSF , and protease inhibitor cocktail tablet ) . After centrifugation at 10 , 000 g for 15 min , the supernatant was incubated with anti-FLAG M2 magnetic beads ( Sigma , M8823 ) with rotation at 4°C for 3 hr . The bead-bound complex was washed 4 times with IP Binding Buffer at 4°C for 5 min each . Bound protein was released by two 10 min incubations with Elution Buffer ( 50 mM Tris-HCl pH 8 , 150 mM NaCl , 5 mM MgCl2 , 5% Glycerol , 0 . 5 mM DTT , 1 mM PMSF , and protease inhibitor cocktail tablet ) containing 150 ng/μl 3×FLAG peptide ( Sigma , F4799 ) . The eluted protein complexes were precipitated with trichloroacetic acid , further digested with Trypsin , and analyzed on an Orbitrap mass spectrometer ( LTQ Velos , ThermoFisher Scientific ) . HPLC separation employed a 100 x 365 μm fused silica capillary micro-column packed with 20 cm of 1 . 7μm-diameter , 130 Angstrom pore size , C18 beads ( Waters BEH ) , with an emitter tip pulled to approximately 1 μm using a laser puller ( Sutter instruments ) . Peptides were loaded on-column at a flow-rate of 400 nL/min for 30 min and then eluted over 120 min at a flow-rate of 300 nl/min with a gradient of 2% to 30% acetonitrile , in 0 . 1% formic acid . Full-mass profile scans were performed in the FT orbitrap between 300–1500 m/z at a resolution of 60 , 000 , followed by ten MS/MS HCD scans of the ten highest intensity parent ions at 42% relative collision energy and 7500 resolution , with a mass range starting at 100 m/z . Dynamic exclusion was enabled with a repeat count of two over the duration of 30 s and an exclusion window of 120 s . For data analysis , the acquired precursor MS and MS/MS spectra were searched against a Mus musculus protein database ( Uniprot reviewed canonical database , containing 16 , 639 sequences ) using SEQUEST , within the Proteome Discoverer 1 . 3 . 0 . 339 software package ( ThermoFisher Scientific ) . Masses for both precursor and fragment ions were treated as mono-isotopic . Oxidized methionine ( +15 . 995 Da ) and the gly-gly footprint on lysine ( +114 . 043 Da ) were allowed as dynamic modifications and carbamidomethylated cysteine ( +57 . 021 Da ) was searched as a static modification . The database search permitted for up to two missed trypsin cleavages and ion masses were matched with a mass tolerance of 10 ppm for precursor masses and 0 . 1 Da for HCD fragments . The output from the SEQUEST search algorithm was validated using the Percolator algorithm . The data were filtered using a 1% false discovery rate ( Rohrbough et al . , 2006 ) , based on q-Values , with a minimum of two peptide matches required for confident protein identification . Co-immunoprecipitation was performed in 1 . 5 g F1 Arabidopsis plants co-expressing HDA9-HA and PWR-FLAG . Total extracts were incubated with 25 µl FLAG magnetic beads for PWR-FLAG purification and the HDA9-HA was detected by using anti-HA-Peroxidase High Affinity 3F10 antibody ( Roche , 13800200 ) . GST tag only and GST-WRKY53 proteins were induced with 200 μM IPTG ( Isopropyl β-D-1-thiogalactopyranoside ) for three hours at 37°C . HDA9-FLAG protein was purified from HDA9-FLAG transgenic plants using IP methods described above . Purified HDA9-FLAG and GST tagged protein were incubated in the pull down buffer ( 20 mM Tris-HCl 8 . 0 , 200 mM NaCl , 1 mM EDTA , 0 . 5% Nonidet P-40 ) for one hour then incubated with Glutathione Sepharose 4B beads ( GE Healthcare , 17075601 ) for one hour at 4°C . After washed with pull down buffer for three times with 5 min per wash , protein-bead complex was boiled in SDS loading buffer , subjected to SDS-PAGE gel , and detected with anti-GST and anti-FLAG antibodies . Mature leaves of HDA9-FLAG/hda9 and HDA9-FLAG/hda9 pwr plants were used . Nuclear-cytoplasmic fractionation of HDA9-FLAG was performed as previously described ( Wang et al . , 2011a ) . RNA-seq and ChIP-seq data were deposited into GEO with the accession number GSE80056 . | The leaves of many plants turn yellow in the fall as nutrients are recycled to prepare for the winter months . However , if leaves age and yellow too early , it can limit how much energy the plant can harvest from light . Thus , it is crucial for plants to know when they should start the leaf aging process . This is also important for plant biologists because premature leaf yellowing can reduce both the yield and quality of crop plants . Certain aging-related genes tightly control when and how leaves age . Like in many other organisms , plant DNA is packaged around proteins called histones . As such , one of the ways that plants regulate the activity of their genes is by chemically modifying the DNA or histones to alter how tightly the DNA is packaged . For example , to switch particular genes off , enzymes known as histone deacetylases remove an acetyl group from their histones . However , it is not clear how these enzymes know which genes to modify and how this helps to make sure that leaf aging happens at the appropriate time . Chen et al . studied a histone deacetylase called HDA9 in a flowering plant named Arabidopsis . The experiments show that the HDA9 enzyme plays an important role in ensuring the leaves turn yellow at the right time . Without HDA9 , the leaf aging process is delayed . HDA9 also needs the help of another protein called PWR that instructs HDA9 to remove acetyl groups from the histones of specific aging-associated genes in order to switch these genes off . The next challenge is to understand how HDA9 and PWR sense developmental and environmental signals to trigger the histone modifications . It will also be important to decipher how this enzyme works with other regulators to trigger leaf aging at the right time . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression",
"plant",
"biology"
] | 2016 | POWERDRESS interacts with HISTONE DEACETYLASE 9 to promote aging in Arabidopsis |
Cell behavior is controlled through spatio-temporally localized protein activity . Despite unique and often contradictory roles played by Src-family-kinases ( SFKs ) in regulating cell physiology , activity patterns of individual SFKs have remained elusive . Here , we report a biosensor for specifically visualizing active conformation of SFK-Fyn in live cells . We deployed combinatorial library screening to isolate a binding-protein ( F29 ) targeting activated Fyn . Nuclear-magnetic-resonance ( NMR ) analysis provides the structural basis of F29 specificity for Fyn over homologous SFKs . Using F29 , we engineered a sensitive , minimally-perturbing fluorescence-resonance-energy-transfer ( FRET ) biosensor ( FynSensor ) that reveals cellular Fyn activity to be spatially localized , pulsatile and sensitive to adhesion/integrin signaling . Strikingly , growth factor stimulation further enhanced Fyn activity in pre-activated intracellular zones . However , inhibition of focal-adhesion-kinase activity not only attenuates Fyn activity , but abolishes growth-factor modulation . FynSensor imaging uncovers spatially organized , sensitized signaling clusters , direct crosstalk between integrin and growth-factor-signaling , and clarifies how compartmentalized Src-kinase activity may drive cell fate .
Regulation of cell fate and behavior is achieved through complex and interconnected signaling networks acting in concert ( Cai et al . , 2014; Devreotes et al . , 2017; Ridley et al . , 2003 ) . For precise signaling , activities of key signaling proteins are tightly compartmentalized in the cell , both spatially and temporally ( Depry et al . , 2015; Gulyani et al . , 2011; Komatsu et al . , 2011 ) . Especially important to cellular controls are signaling nodes that function downstream of multiple receptor classes and help in signal integration . Src family kinases ( SFKs ) are such key signaling nodes; activated by cell adhesion receptors , integrins , receptor tyrosine kinases ( RTKs , including growth factor receptors ) and Gprotein-coupled receptors ( GPCRs ) among others ( Abram and Courtneidge , 2000; Giannone and Sheetz , 2006; Grande-García et al . , 2007; Parsons and Parsons , 2004; Thomas and Brugge , 1997 ) . Src-kinases critically influence cell fate; regulating cell shape , migration and adhesion , survival and growth , stemness and differentiation making them important therapeutic targets in multiple diseases ( Chetty et al . , 2015; Gujral et al . , 2014; Kim et al . , 2009; Lewis-Tuffin et al . , 2015; Nygaard et al . , 2014; Saad , 2009; Timpson et al . , 2001; Zhang et al . , 2013; Zhang et al . , 2014 ) . Despite their importance , intracellular activity patterns of individual Src family kinases with spatial and temporal precision are still unclear . Visualization of active kinases in cells is essential to understand how Src kinases integrate signals and regulate multiple , and sometimes opposing processes , with fidelity and precision . Currently available fluorescent SFK sensors have severe limitations . Most available biosensors , especially genetically encoded ones , do not directly report the intracellular distribution of active kinases since they rely on detecting the phosphorylation of a ‘pseudosubstrate’ peptide ( Liao et al . , 2012; Ouyang et al . , 2008; Ouyang et al . , 2019; Seong et al . , 2011; Wang et al . , 2005 ) . As a result , biosensor readout – extent of phosphorylation of ‘sensor’ peptides – can be affected by the activities of kinases as well as cellular phosphatases . Confounding readouts further , it has been reported that sensors can get trapped in an ‘ON’ state owing to strong intramolecular interactions between phosphorylated substrate peptides and their respective recognition motifs ( Komatsu et al . , 2011; Regot et al . , 2014 ) ; and therefore are unable to report the turning OFF of kinase activity . Another major drawback of current sensors is that multiple kinases , especially closely related ones , can phosphorylate these substrate-based sensors in a promiscuous manner leading to a lack of specificity . Study of SFK activity is especially complicated by the presence of multiple closely related Src family members , including three ubiquitously expressed kinases c-Src , c-Yes and Fyn . While there is some functional redundancy , individual SFKs also perform critical , exclusive roles in the cell ( Zhang et al . , 2014; Kuo et al . , 2005; Lowe et al . , 1993; Lowell and Soriano , 1996; Marchetti et al . , 1998; Molina et al . , 1992; Palacios-Moreno et al . , 2015 ) . Therefore , despite decades of study , it is not always clear which kinase ( s ) is activated in a given cellular context ? In this regard , fluorescent biosensors that report the activation of individual Src kinases in live cells would be extremely valuable in clarifying kinase activity and function . However , such specific biosensors for individual Src kinases are still limited ( Gulyani et al . , 2011; Koudelková et al . , 2019; Paster et al . , 2009; Stirnweiss et al . , 2013 ) , with current sensors generally giving only a pan-SFK readout with limited spatial and temporal resolution . Here , we present a fluorescent biosensor specific for the SFK-Fyn . Fyn , is ubiquitously expressed and regulates cell migration , epithelial to mesenchymal transition ( EMT ) , cancer metastasis , immune-response , axonal guidance and patterning , and synaptic functions ( Gujral et al . , 2014; Du et al . , 2016; Lewin et al . , 2010; Meriane et al . , 2004; Posadas et al . , 2016; Salter and Kalia , 2004 ) . Interestingly , among several SFKs implicated , recent evidence shows Fyn may specifically control EMT and metastatic progression ( Gujral et al . , 2014 ) . Overall , Fyn is a critical player in multiple cell/tissue types , integrating signaling through multiple receptor classes ( Palacios-Moreno et al . , 2015; Martín-Ávila et al . , 2016; Yadav and Denning , 2011 ) and controlling cell fate . However , there is currently no tool , including specific antibodies , to visualize Fyn activity in cells . As a consequence , there is no information on intracellular Fyn dynamics in live cells . A Fyn biosensor would bridge this gap and offer a method to address Fyn functional complexity as a key signaling node . Generation of biosensors is often limited by a lack of naturally available binders that can detect active states of proteins in real-time . To address this , we have developed a strategy of using artificially engineered binders that recognize the active form of target proteins in live cells ( Gulyani et al . , 2011 ) . Like other SFKs ( Arold et al . , 2001; Boggon and Eck , 2004; Kinoshita et al . , 2006; Noble et al . , 1993; Sicheri and Kuriyan , 1997 ) , Fyn is maintained in an inactive or closed form through multiple intramolecular interactions ( Thomas and Brugge , 1997 ) . Activation of the kinase involves disruption of these interactions leading to an open conformation , with more exposed surfaces ( Gulyani et al . , 2011; Young et al . , 2001 ) . Taking advantage of this , we screened a combinatorial protein library to isolate a binding protein ( F29 ) that specifically binds to a region of Fyn ( SH3 domain ) that is more exposed in the open conformation . For screening , we chose a library generated by mutagenizing the Sso7d protein - a highly stable protein derived from the hyperthermophilic archaeon Sulfolobus solfataricus . We have previously shown that Sso7d mutant-libraries can be used to isolate binding proteins for a wide range of targets ( Gera et al . , 2012; Gera et al . , 2011; Gocha et al . , 2017; Hussain et al . , 2013 ) . Also , since Sso7d has no known interactions with mammalian intracellular proteins , we reasoned that Sso7d would serve as an ideal ‘inert’ scaffold for generating an intracellular biosensor . Our analysis shows that F29 indeed binds specifically to Fyn , with little to no cross-reactivity with other SFKs . Nuclear magnetic resonance ( NMR ) spectroscopy analysis of F29 in complex with the target reveals the structural basis of the specificity of F29 binding to Fyn . An efficient readout of the molecular recognition of the target by the binding protein is central to constructing an effective biosensor . Here , we used our new binder/F29 to construct a genetically encoded Fyn biosensor ( FynSensor ) based on fluorescence resonance energy transfer ( FRET ) . FynSensor is robust , sensitive and faithfully reports Fyn activation and regulation . We also show how FynSensor expression in cells is minimally perturbing , with no measurable changes in either cellular morphodynamics or downstream signaling . Significantly , FynSensor enables direct visualization of active Fyn in live cells , revealing Fyn activation to be spatially compartmentalized , polarized and pulsatile . To the best of our knowledge , this is the first reported visualization of activated Fyn in live cells; notably , unavailability of suitable antibodies currently precludes even routine detection of active Fyn . Strikingly , FynSensor imaging shows the presence of spatially localized signaling clusters sensitive to integrin and growth factor signaling; and reveals how growth factor response in cells is influenced by integrin-dependent protein activity . More broadly , FynSensor shows how spatially compartmentalized activation of key signaling proteins may lead to efficient signal integration and precise control of cell physiology . Further , our results provide a framework for systematic development of intracellular biosensors for specific Src-family and other kinases in general .
Specifically recognizing active Fyn is critical for constructing an intracellular sensor . Activation of SFKs is accompanied by large conformational changes; the active kinase adopts a more open conformation with significantly reduced intramolecular interactions ( Thomas and Brugge , 1997 ) . Specifically , prior evidence ( Gulyani et al . , 2011; Young et al . , 2001 ) suggests that the SH3 domain in SFKs is more accessible in the active conformation relative to the closed , inactive state ( Figure 1A ) . Therefore , a binding protein targeting the SH3 domain of Fyn can be expected to preferentially bind the active form of the kinase ( Gulyani et al . , 2011 ) . Such an active state binder can then be engineered as a fluorescent biosensor for reporting intracellular Fyn activation ( Figure 1A , B ) . Accordingly , we sought to generate a binding protein specific for Fyn SH3 domain . To this end , we screened a yeast surface display library of Sso7d mutants ( library generated by randomizing ten Sso7d residues ) , using magnetic screening and fluorescence associated cell sorting ( FACS ) , adapting methods described previously ( Gera et al . , 2011; Gera et al . , 2013 ) ( Figure 1—figure supplement 1A , B ) . Stringent negative selection steps were included to eliminate binders that exhibited cross-reactivity to the SH3 domains of seven other SFKs , specifically c-Src , c-Yes , Fgr , Blk , Lck , Lyn , and Hck . The pool of yeast cells obtained after magnetic selection were further subjected to multiple rounds of FACS to enrich Fyn SH3-binding yeast cells ( Figure 1—figure supplement 1C ) . Single clones were then picked and assessed for specific binding to Fyn-SH3 . One such clone ( referred to as F29 hereafter ) showed specific binding to Fyn-SH3 domain , with little or no cross-reactivity with SH3 domains from the other highly homologous SFKs mentioned above ( Figure 1C and Figure 1—figure supplement 2 ) . Sequence of F29 and its comparison to wild-type Sso7d is shown in Figure 1D . We further examined if F29 retained binding to Fyn-SH3 when removed from the context of the yeast cell surface . F29 fused with glutathione-S-transferase ( GST-F29 ) was recombinantly expressed and purified , and immobilized on glutathione sepharose beads . Immobilized GST-F29 could pull down 6xHistidine ( 6xHis ) -tagged Fyn-SH3 , as detected by immunoblotting using an anti-6xHis antibody . In contrast , no signal was seen when the Fyn SH3 pulldown experiment was done with GST-saturated beads , or when the SH3 domain of Src was pulled down using GST-F29 beads ( Figure 1—figure supplement 3 , Figure 1E ) . Taken together these results confirm that F29 binds specifically to the Fyn-SH3 . We further used pulldown experiments to estimate the binding affinity ( KD ) of the interaction between F29 and Fyn-SH3 . Briefly , immobilized GST-F29 was incubated with varying concentrations of Fyn-SH3 , and the amount of bead-bound Fyn-SH3 was quantified . Upon fitting the data to a monovalent binding isotherm , the apparent KD of the F29-Fyn-SH3 interaction was estimated as 60 ± 16 nM ( Figure 1F , Figure 1—source data 1 ) . F29 specifically binds Fyn SH3 while showing little or no cross-reactivity to SH3 domains of other highly homologous SFKs . To investigate the molecular basis of the specificity of F29 binding to Fyn-SH3 , we used NMR spectroscopy to determine the structure of F29 in complex with Fyn-SH3 . For this , we first analyzed the structure of isolated F29 ( Figure 2 ) . The 15N-edited Heteronuclear Single Quantum Correlation ( HSQC ) spectrum of purified F29 ( isolated from E . coli grown in 13C , 15N-labeled media ) showed well-dispersed peaks , indicating proper folding of the protein despite extensive mutagenesis of the Sso7d scaffold ( Figure 2—figure supplement 1A ) . Complete chemical shift assignments of the backbone and side chain atoms using a series of 2D and 3D NMR experiments were then used to calculate a structure by the CS-ROSSETTA algorithm ( van der Schot et al . , 2013 ) . The structure of F29 superimposed well with Sso7d ( Figure 2—figure supplement 2A ) indicating a conservation of the Sso7d fold in the binder . To obtain the structure of Fyn-SH3:F29 complex , 15N-edited HSQC spectra of 15N-labeled free Fyn SH3 domain ( residues 87 to 139 of h-Fyn ) as well as in complex with F29 ( unlabeled ) were acquired . The differences between chemical shifts in the free and bound Fyn-SH3 provided the Chemical Shift Perturbations ( CSPs ) in Fyn-SH3 due to F29 binding ( Figure 2—figure supplement 1B , D ) . High CSPs point to specific Fyn SH3 residues likely present at the binding interface in the complex with F29 ( Figure 2—figure supplement 1D ) . Similarly , the CSPs in15N-labeled F29 upon binding Fyn-SH3 were also obtained ( Figure 2—figure supplement 1A , C ) ; these values help identify the F29 binding interface in the complex . A 3D 15N-edited Nuclear Overhauser Effect Spectroscopy ( NOESY ) experiment , performed on a complex of 15N , 2H-labeled Fyn-SH3 and unlabeled F29 , led to the unambiguous assignment of ~7 intermolecular NOEs from the spectra . These NOEs helped identify through-space interfacial interactions between residues . Using the structure of Fyn-SH3 ( pdb: 3UA6 ) , the structure of F29 obtained above , and the CSPs as ambiguous restraints , and NOEs as unambiguous restraints , the structure of Fyn-SH3:F29 complex was solved by Haddock ( de Vries et al . , 2010; van Zundert et al . , 2016 ) . After analysis , the most populated cluster had 200 structures ( Supplementary file 1A ) and the lowest energy structure and its surface representation are shown in Figure 2—figure supplement 2B and Figure 2A , B . The contact residues at the interface of the complex are shown in Figure 2C , D . At the binding interface several non-covalent interactions including hydrophobic-aromatic , hydrogen bonds and salt bridges can be observed ( Supplementary file 1B ) . Eight of the ten randomized residues as well as five originals ( SSo7d scaffold ) amino acids of F29 show interactions with twelve amino acids of Fyn SH3 ( Figure 2E ) . Analysis of the binding interface reveals that residues Phe22 , Thr24 , Tyr26 , and Phe31 , 32 , 43 of F29 show multiple interactions with the Fyn SH3 domain , with SH3 residues Tyr137 , Asn136 , Tyr132 , Trp119 , Arg96 and Tyr91 featuring prominently in these interactions . Interestingly , Lys9 ( an original Sso7d residue ) and Arg33 ( randomized residue ) on the F29 binder showed exclusive salt bridge interactions with Asp118 and Asp99 of Fyn SH3 domain , respectively ( Figure 2E ) . SH3 domains naturally interact with proteins with poly-proline motifs ( Li , 2005 ) . To gain insight into how F29 binds and specifically recognizes Fyn SH3 , we compared the F29-Fyn SH3 complex with structures of FYN-SH3 bound to poly-Proline peptides ( pdb: 4EIK and pdb: 3UA7 ) ( Figure 2—figure supplement 1E , F ) . The buried surface area of F29:Fyn SH3 complex is much larger ( 800 Å2 , ∆G = −7 kcal/mol ) compared to that of the poly-Pro:Fyn SH3 complex ( pdb: 3UA7 , 330 Å2 , ∆G = −5 kcal/mol ) . The number of contacts is also considerably larger in the F29 ( 84 contacts ) versus the poly-Proline ( 40 contacts ) . Moreover , the orientation of the poly-Prolines at the interface differs from the orientation of the F29 interfacial beta-sheet by ~30o . Hence , although the F29 interface overlaps with the poly-Proline interface , it is significantly distinct in terms of contacts , surface area and orientation . Residues of Fyn-SH3 that are involved in binding F29 as well as Poly-Pro-1/2 polypeptides are shown in Figure 2—figure supplement 1G . We then used site-directed mutagenesis to examine contributions of key interfacial residues to F29-Fyn SH3 binding . A salt bridge between the residue R33 of F29 and D99 of Fyn SH3 appears prominent . Indeed , the introduction of an R33A mutation in F29 resulted in reduced binding to Fyn-SH3 domain , as evidenced by pull-down analysis ( Figure 2—figure supplement 3A , Figure 2—figure supplement 3—source data 1 ) . These results confirm that R33 in F29 and D99 in Fyn-SH3 contribute substantially to the F29-Fyn-SH3 binding interaction . Interestingly , a Pro41Ala mutant of F29 also showed reduced binding to Fyn-SH3 ( Figure 2—figure supplement 3A ) . Although the NMR analyses do not show Pro41 to form direct interfacial contacts with SH3 , the reduced binding of this mutant may be attributed to perturbation of local structure . Proline41 forms a critical hinge preceding β3 strand ( Figure 2D ) ; mutation of the proline likely disrupts the structure the β-sheet that contains multiple key binding contacts with Fyn-SH3 . These structural analyses also offer insight into the molecular basis for the specificity of F29 binding to Fyn-SH3 over highly homologous SH3 domains of other SFKs . While the binding interface in the Fyn-SH3-F29 complex reveals multiple interactions ( Figure 2D , E ) , the residue D99 in Fyn-SH3 that forms a salt bridge with R33 of F29 is unique to Fyn among the ubiquitously expressed SFKs as well as Lyn and Fgr ( Figure 2—figure supplement 3B ) . Therefore , we hypothesized that D99 in Fyn-SH3 may contribute to the binding specificity of F29 for Fyn-SH3 vis-à-vis to at least some of the SFK SH3 domains . To test this hypothesis , we generated a mutant version of the SH3 domain from Src ( Src-SH3 ) wherein the native threonine was replaced with aspartic acid ( T99D ) . Strikingly , T99D Src-SH3 , but not wild-type Src-SH3 , showed detectable binding to F29 in pulldown assays ( Figure 2—figure supplement 3C ) . These results suggest that D99 in Fyn-SH3 contributes to the binding specificity of F29 for Fyn over other SFKs , notably Src . Nevertheless , specificity of F29 for Fyn-SH3 over Yes-SH3 ( containing E99 ) ( Figure 1C , E ) and Lyn/Fgr ( containing D99 , Figure 1—figure supplement 2 ) , suggests that binding specificity results from a unique combination of multiple interactions at the binding interface that may require further investigation . F29 efficiently and specifically binds Fyn SH3 in vitro . We then examined if F29 is able to recognize full-length , cellular Fyn and if it preferentially binds the active form of Fyn . To test this , either WT Fyn kinase or the CA mutant ( Takeuchi et al . , 1993 ) ( active and open conformation ) was exogenously expressed in human embryonic kidney ( HEK-293T ) cells and the cell lysate treated with immobilized GST-F29 . Fyn pulldown and immunoblotting showed that the CA mutant was more efficiently pulled down by F29 as compared to the WT Fyn kinase ( Figure 3A , Figure 3—source data 1 ) . On the contrary , GST-alone control beads showed no significant pulldown of either WT or CA Fyn . These data indicate that F29 preferentially binds active Fyn; the SH3 domain targeted by F29 is more accessible in the CA mutant compared to the WT Fyn kinase . Further , similar GST-pulldown and immunoblot analysis showed that immobilized F29 ( GST-F29 ) , but not GST-alone control , was able to bind and pull-down endogenous or native Fyn from cell lysates ( Figure 3B ) . These results confirm efficient binding of F29 to cellular Fyn . Taken together , our results so far show F29 to be a promising candidate for generating an intracellular biosensor for active Fyn . Accordingly , we conceptualized a biosensor design based on FRET wherein the binding of F29 to active Fyn would result in an increase in FRET between suitably placed donor and acceptor fluorophores ( Figure 3C ) . To construct a genetically encoded biosensor , Fyn was labeled with a FRET donor fluorescent protein ( mCerulean ) while F29 was tagged with a FRET acceptor protein ( mVenus ) . While F29 already shows high specificity to Fyn , our design further ensures that FRET signal can arise only from labeled F29 binding to labeled active Fyn . For generating the fluorescently labeled Fyn , we chose to introduce mCerulean ( FRET donor ) between the unique ( UD ) and the SH3 domains of full-length Fyn kinase ( Paster et al . , 2009; Stirnweiss et al . , 2013 ) ( Figure 3C ) . The choice of this insertion position is designed to increase the probability of efficient FRET when acceptor-labeled F29 binds the exposed SH3 domain in active Fyn . Another critical design consideration was to minimize any perturbation caused due to mCerulean insertion into Fyn . To this end , we selected a flexible poly- ( glycine-serine ) peptide linker ( Trinh et al . , 2004 ) to insert mCerulean within Fyn ( also see Materials and method and Figure 3—figure supplement 1 for details on molecular engineering ) . We then examined the suitability of the designed biosensor constructs for cellular imaging . Immunoblotting analysis on cell lysates showed robust intracellular expression of full-length mCerulean-Fyn fusion proteins ( hereby referred to as ‘mCer-Fyn’ ) , confirming that the fusion proteins remain intact and resistant to proteolysis in cells ( Figure 3—figure supplement 2A–I ) . Further , mCer-Fyn fusions expressed in cells showed expected fluorescence; with spectra from cells closely resembling unmodified mCerulean ( Figure 3—figure supplement 2C ) . Additionally , the mVenus-F29 fusion biosensor protein also expressed well , was resistant to proteolysis ( Figure 3—figure supplement 3A ) and showed the expected fluorescent signatures resembling unmodified mVenus ( Figure 3—figure supplement 3B ) . Importantly , we then tested if labeled Fyn kinase retains activity and is regulated appropriately . Src family kinases possess autocatalytic activity; a tyrosine residue ( Y420 in h-Fyn ) in the activation loop is phosphorylated when the protein gets activated ( Cooper and MacAuley , 1988; Barker et al . , 1995 ) . Significantly , immunoblot analysis showed that mCer-Fyn retains autocatalytic activity similar to the wild-type unmodified protein in cells . Further quantification of Fyn Y-420 autophosphorylation levels , show no significant differences between mCer-Fyn and unlabeled wt-Fyn ( Figure 3—figure supplement 2A , B , Figure 3—figure supplement 2—source data 1 ) . Importantly , these data show that mCerulean insertion is minimally disruptive to kinase activity and that the engineered kinase appears to be regulated similar to unmodified Fyn . To further test if our Fyn fusion ( mCer-Fyn ) is functional and can accurately reflect cellular Fyn dynamics , we examined if mCer-Fyn can modulate downstream signaling and functionally replace endogenous , wild-type Fyn . For this we carefully assayed Fyn effects on downstream signaling , as measured through extracellular signal-related kinase ( ERK ) phosphorylation and performed Fyn knockdown rescue in cells , comparing both unlabeled as well as Fyn-fusion ( mCer-Fyn ) constructs . We first established that RNA-i leads to a specific knockdown of cellular Fyn protein ( Figure 3—figure supplement 4A ) but not the closely related Src and Yes kinases in HEK293T cells ( Figure 3—figure supplement 4B , C , Figure 3—figure supplement 4—source data 1 ) . This Fyn knockdown significantly attenuates the levels of ERK phosphorylation in cells ( Figure 3—figure supplement 5A , C , Figure 3—figure supplement 5—source data 1 ) . This is consistent with earlier reports documenting Fyn’s role in modulating ERK activity and downstream signaling ( Wary et al . , 1998 ) . Notably , expressing either mCer-Fyn or wt , unlabeled Fyn in ‘Fyn-knockdown’ cells significantly rescues this reduction in ERK phosphorylation . Further , the extent of increase in this ERK phosphorylation , mediated through ectopic Fyn expression in ‘Fyn-knockdown’ cells , was observed to be the same for wild-type , unlabeled Fyn as well as mCer-Fyn ( Figure 3—figure supplement 5B–D , Figure 3—figure supplement 2—source data 1 ) . These results unequivocally demonstrate that our labeled Fyn ( mCer-Fyn ) is regulated appropriately , ‘signaling competent’ and can functionally replace native , untagged Fyn; thereby making it suitable for faithfully reporting cellular Fyn dynamics . Biosensor variants were then tested for efficacy in reporting intracellular Fyn activity . For this , we expressed biosensor constructs in HEK-293T cells and recorded fluorescence spectra from live cells ( see Materials and methods ) . When cells co-expressing FRET donor ( mCer-Fyn ) and FRET acceptor ( mVenus-F29 ) were excited at 435 nm ( mCerulean/donor excitation ) , emission peaks were seen at ~475 nm ( mCerulean ) as well as at ~525 nm corresponding to the peak emission wavelength of the acceptor fluorophore ( mVenus ) ( Figure 3D ) . Appearance of acceptor emission ( mVenus , ~525 nm ) on donor excitation ( 435 nm ) clearly shows that FRET occurs between mCer-Fyn and mVenus-F29 . This confirms that mVenus-F29 can recognize intracellular Fyn in live cells and this recognition can be reported through a robust FRET signature . We then tested if this FRET signal is sensitive to the relative stoichiometry of the target ( Fyn ) and the binder F29 as would be expected with a binding-induced FRET . Indeed , increasing the amount of intracellular mVenus-F29 while maintaining constant expression of mCer-Fyn , led to a concomitant increase in FRET signal , followed by saturation of this FRET increase . Here , the saturation in FRET signal suggests that even increasing concentrations of F29 binder in cells does not cause an inadvertent or artefactual activation of Fyn kinase , especially since F29 binder expression in cells does not lead to any measurable changes in autocatalytic activity of Fyn ( Y-420 autophosphorylation ) ( Figure 3—figure supplement 6 , Figure 3—figure supplement 6—source data 1 ) . Importantly , the FRET signal in cells is specific and dependent on mVenus-F29 binding to mCer-Fyn , since making a single point mutation in F29 abolishes the FRET response . If the non-binding variant of F29 ( mVenus-F29-P41A ) is co-expressed with mCer-Fyn in cells , little or no FRET is seen , even at higher concentrations of the non-binding control ( Figure 3D left inset , Figure 3—source data 2 ) . Further , the FRET signal was higher in cells co-expressing mVenus-F29 and constitutively active Fyn , relative to wild-type Fyn ( Figure 3D right inset , Figure 3—source data 3 ) . This shows that F29 binding and biosensor readout ( FRET response ) is indeed sensitive to the activation status of Fyn ( Figure 3D ) . Fyn kinase has two acylation marks ( a myristoyl and a palmitoyl group ) that make it preferentially localized to the cell membrane ( van't Hof and Resh , 1999 ) . Therefore , we reasoned that localization of F29 to the cell membrane would increase its proximity to membrane-bound Fyn , resulting in an increase in FRET signal and consequently greater dynamic range of the biosensor . To test this hypothesis , we added a myristoylation ( ‘myr’ ) sequence ( MGSSKSKPKDPS ) to the F29 binder ( Victor and Cafiso , 1998 ) . Indeed , addition of a myristoylation signal led to a substantial increase in FRET signal ( Figure 3E , Figure 3—source data 4 ) . To further test the fidelity of this enhanced FRET response observed with myristoylated-F29 , we tested the non-binding variant ( P41A ) of the myr-mVenus-F29 binder . Notably , myristoylated non-binding mVenus-F29 P41A showed little or no FRET signal when co-expressed with mCer-Fyn ( Figure 3E ) . Taken together , these results clearly show that the FRET signal observed with the F29-based Fyn biosensor is indeed due to specific binding of F29 to the active form of Fyn , and not just due to non-specific membrane localization and incidental proximity . The biosensor constructs –mCer-Fyn and myr-mVenus-F29 ( referred to as binder ) – are collectively referred to as FynSensor hereafter . Src family kinases , including Fyn are known to strongly regulate cell adhesion as well as the cytoskeleton . Several SFK substrates and interacting partners are reported to be actin binding proteins , regulators of Rho family GTPases and other proteins that help remodel and regulate the actomyosin network ( Ridley et al . , 2003; Etienne-Manneville and Hall , 2001; Huveneers and Danen , 2009; Roca-Cusachs et al . , 2012 ) . We therefore reasoned that measuring the cell morphodynamics ( Gulyani et al . , 2011; Hodgson et al . , 2016 ) would be a sensitive way to examine if moderate-to-low expression of labeled Fyn and the F29 binder significantly and measurably perturb the cell . First , the expression levels of FynSensor constructs labeled Fyn ( mCer-Fyn ) as well as mVenus-F29 were quantified in U2OS cells to ensure that the binder expression causes no artefactual activation of Fyn kinase ( levels of pTyr416/total Fyn; Figure 3—figure supplement 6 ) . We then quantitatively examined temporal changes in cell area and perimeter of cells expressing either FynSensor ( mCer-Fyn:myr-mVenus-F29 , DNA ratio 1:2 , also see Figure 3D ) or the binder alone or a control construct ( myr-tagged-mVenus ) in the adherent osteosarcoma U2OS cells at the expression levels specified . We find no significant change in cellular morphodynamics as a result of FynSensor expression the binder alone as compared to controls cells ( Figure 3—figure supplement 7 , Figure 3—figure supplement 7—source data 1 ) . We further confirmed that the expression of Fyn-binder does not perturb the overall intracellular localization of labeled Fyn ( Figure 3—figure supplements 8 and 9 , Figure 3—figure supplement 9—source data 1 ) . Immunofluorescence data also shows that labelled Fyn ( mCer-Fyn ) localization is similar to endogenous Fyn ( Figure 3—figure supplement 9A ) . While FynSensor and F29 expression had no effect on cell morphodynamics and Fyn localization , we also examined if FynSensor expression significantly perturbs downstream signaling . For this , we again used the extent of ERK phosphorylation as a sensitive readout of Fyn signaling . We find that neither expression of FynSensor nor binder alone in adherent osteosarcoma U2OS cells under the conditions used has any measurable effect on ERK phosphorylation levels ( Figure 3—figure supplement 10 , Figure 3—figure supplement 10—source data 1 ) . Similarly , expression of FynSensor constructs ( labeled Fyn + binder ) in C2C12 mouse myoblasts ( Figure 3—figure supplement 11 , Figure 3—figure supplement 11—source data 1 ) or Fyn-knockdown HEK-293 cells did not significantly change levels of p-ERK ( Figure 3—figure supplement 5 ) . These data collectively show that FynSensor expression under these specified conditions does not significantly perturb downstream signaling . In light of these data showing little or no cellular perturbation , we have used the same FynSensor expression conditions for all further biosensor imaging studies . Our results show that FynSensor produces a FRET signal upon specific binding of F29 to Fyn-SH3 , resulting in a sensitive readout of active intracellular Fyn . We further evaluated FynSensor for its ability to report spatial patterns of active Fyn in single , living cells . To investigate the spatial FRET response generated by FynSensor , we first employed the ‘fluorescence recovery after acceptor photobleaching ( APB ) ’ method . APB analysis is a robust method of confirming and quantifying FRET in cells ( Karpova and McNally , 2006 ) , where recovery of donor fluorescence on APB is seen as a strong indicator of proximity-induced FRET . Energy transfer from the donor to the acceptor molecules results in quenching of donor fluorescence , and this quenching is relieved when proximal acceptor molecules undergo photobleaching . Indeed , when adherent osteosarcoma U2OS cells expressing FynSensor were illuminated with increasing doses of 514 nm laser light ( wavelength corresponding to acceptor absorption; mVenus photobleaching ) , we observed a substantial , dose-dependent recovery of donor ( mCerulean ) fluorescence . This confirms FRET between mCer-Fyn and mVenus-F29 . Significantly , APB-induced recovery of donor fluorescence was greater at cell edges ( Figure 4A , B , Figure 4—source data 1 ) , suggesting spatial variation in Fyn activity . Notably , in control cells expressing a mutant form of labeled-F29 ( P41A ) , acceptor photobleaching showed no such dose-dependent recovery of donor fluorescence ( Figure 4C , Figure 4—source data 2 ) . Since the P41A mutant lacks the ability to bind Fyn-SH3 , these data show that APB-induced recovery of donor fluorescence and the FRET response are due to F29 binding to target and not due to incidental proximity or optical artifacts . Finally , when cells were treated with a known pharmacological inhibitor of Src family kinase including Fyn , SU6656 ( Blake et al . , 2000 ) , we observed a significant reduction in the amount of donor fluorescence being recovered ( Figure 4—figure supplement 1 , Figure 4—figure supplement 1—source data 1 ) . This reduction in FRET ( F29 binding to Fyn ) on inhibitor treatment is interesting and points to allosteric changes in Fyn conformation on inhibitor binding . While ATP-competitive inhibitors like SU6656 directly bind the catalytic domain , they can also stabilize an inactive conformation of Src family kinases and modulate the accessibility of its key regulatory domains to ligands/binding proteins ( Krishnamurty et al . , 2013 ) . Therefore , our data ( reduced FynSensor FRET ) suggests that indeed Fyn likely adopts a closed conformation in response to inhibitor treatment , thereby reducing SH3 domain accessibility/binding to F29 . The sensitivity of FynSensor FRET to inhibitor binding in fact suggests that the labeled kinase shows allosteric conformational changes similar to those seen with untagged SFKs , further indicating that the biosensor faithfully reports Fyn conformational dynamics . Taken together , these results further confirm that the FynSensor readout is the result of specific binding of F29 binding to Fyn-SH3 , and selectively reports the active form of Fyn ( Figure 4B , C and Figure 4—figure supplement 1 ) . The FynSensor readout also shows hitherto uncharacterized spatial patterns in intracellular active Fyn in live cells , with greater Fyn activity closer to the cell edge . We used FynSensor to examine the spatial and temporal dynamics of Fyn activity , especially in light of the important role played by Fyn as a signaling node functioning downstream of distinct receptors classes . We reasoned that ability to specifically image the active conformation of a single SFK would provide new insight into how signal integration may occur . We first examined Fyn activation dynamics in serum-starved cells plated on fibronectin ( FN ) that show constitutive protrusion retraction cycles . For imaging activity dynamics , we used ratiometric sensitized emission ( normalized acceptor emission due to donor excitation ) to quantify FRET signals from the sensor ( Gordon et al . , 1998 ) ( Materials and methods ) . Ratiometric sensitized emission measurements allow facile and rapid imaging of FRET responses in cells with both spatial and temporal resolution . Sensitized emission ( FRET ) imaging with FynSensor in FN-plated serum-starved U2OS cells revealed spatially localized Fyn activity , with clear intracellular zones showing higher total FRET ( FRETT ) . Such intracellular zones showing enhanced FynSensor FRETT were consistently observed in all cells in the data set ( Figure 5A–I ) . Since all cells spontaneously appear to show regions of differential Fyn activity , we employed quantitative image analysis to probe these patterns of Fyn activity . An automated cell quadrant analysis of intracellular FRETT levels provides clear evidence of differential and compartmentalized Fyn activity . Briefly , every cell imaged was divided into quadrants in an automated manner and FRET response for each quadrant was analyzed over time . These analyses show substantial and significant differences in Fyn activation levels ( FynSensor FRETT ) between intracellular quadrants ( designated as ‘low’ and ‘high’ FRET quadrants for consistency and comparison ) ( Figure 5A–II , Figure 5—source data 1 ) . Importantly , the spatial patterns of Fyn activation revealed through FynSensor imaging are robust and are unlikely to be an artifact of imaging , since a single point mutation in the binder substantially abolishes the FRET response seen . When mCer-Fyn is co-expressed with the non-binding P41A mutant of F29 , little or no FRET signal is seen in single cells , clearly showing that FynSensor FRET is dependent on F29 binding to active Fyn ( Figure 5B , Figure 5—source data 2 , Figure 5—figure supplement 1 , Figure 5—figure supplement 1—source data 1 , Figure 5—video 1 ) . Overall , the spontaneous compartmentalizing of Fyn activity was observed in all cells and is a striking finding revealed through FynSensor imaging . ( Figure 5A ) . We demonstrate that the spatially enhanced FRETT patterns observed are unlikely to be an artifact of an unequal or spatially-constrained intracellular distribution of the F29 binder . Figure 5—figure supplement 2 ( Figure 5—figure supplement 2—source data 1 ) shows that while the FynSensor FRET levels are non-uniform and spatio-temporally patterned , the fluorescently labeled F29 ( myr-mVenus-F29 ) is homogeneously distributed in cells . While increased recruitment of Fyn may contribute to the establishment of regions of ‘high Fyn activity’ in serum-starved , FN-plated cells , our data analysis shows that differential Fyn activity observed cannot solely be ascribed to increased localization of Fyn . Figure 5—figure supplement 3 ( Figure 5—figure supplement 3—source data 1 ) show that intracellular regions showing higher FRETT also show significantly higher levels of donor-normalized FRET ( donor-normalized FRET = FRETT/mCer-Fyn fluorescence ) , which normalizes the overall activity for protein levels . This is also illustrated ( Figure 5—figure supplement 4 , Figure 5—figure supplement 4—source data 1 ) by examples of regions that have almost similar levels of the kinase ( mCer-Fyn localization as visualized in the donor channel in regions marked 1 and 2 ) but show significantly different levels of kinase activity ( FRET index image , regions 1 and 2 ) , reaffirming the point that increases in FRET signals are just not due to increase in kinase localization . We surmised that spatially enhanced Fyn activity seen in serum-starved cells on fibronectin may be caused by differential integrin signaling . To test if this Fyn activation in serum-starved cells , as visualized through FynSensor , is indeed sensitive to integrin signaling , we treated cells with an inhibitor of focal adhesion kinase ( FAK ) , PF-562271 ( Mills et al . , 2015; Stokes et al . , 2011 ) and examined FynSensor response ( Figure 5C–I , Figure 5—video 2 ) . Focal adhesion kinase ( FAK ) is a critical transducer of integrin signaling and is known to partner SFKs in mediating cellular responses ( Renshaw et al . , 1999; Cheng et al . , 2014 ) . Strikingly , when serum-starved , FN-plated U2OS cells were treated with PF-562271 , there was an immediate and drastic reduction in the FynSensor FRET response ( Figure 5C II-IV , Figure 5—source data 3 ) . This reduction in activity is accompanied by a substantial dampening of the temporal changes seen in levels of active Fyn in cells ( see below ) . These results clearly demonstrate that localized Fyn activity , reported through FynSensor , is sensitive to focal adhesion kinase activity; and supports the model that compartmentalized Fyn activity arises through differential integrin signaling in serum-starved adherent cells . To confirm that these spatial patterns of Fyn activity are not just limited to any one specific cell-type , we performed FynSensor imaging in mouse C2C12 myoblasts apart from U2OS osteosarcoma cells . Notably , FynSensor again shows Fyn activity to be spatially patterned in the serum-starved C2C12 myoblast cells ( Figure 6A ) . Quantitative quadrant analysis confirms differential Fyn activity ( FRETT ) in distinct intracellular zones ( Figure 6A–II , Figure 6—source data 1 ) . To assess if cells with moderately overexpressed levels of Fyn kinase also show similar spatial-activation patterns , we performed imaging experiments in HEK-293T cells where the endogenous Fyn has been depleted through RNA-i and then complemented with FynSensor ( Figure 3—figure supplement 9 , Figure 6—figure supplement 1 , Figure 6—figure supplement 1—source data 1 ) . Biosensor imaging in these HEK cells indeed showed activation patterns very similar to those previously observed in U2OS and C2C12 cells ( Figure 6B , Figure 6—source data 2 ) . The observation of similar , spatially enhanced patterns of Fyn activity in these very distinct cell types shows that FynSensor responses are highly robust and is suggestive of conserved spatially modulated signaling mechanisms . To further confirm that the conserved patterns of Fyn activity are not impacted by changes in FynSensor expression levels , we systematically examined FynSensor FRET profiles as a function of expression . In all three cell types examined , the spatial patterns remain conserved independent of the precise levels of labeled Fyn ( mCer-Fyn ) as well as the binder ( mVenus-F29 ) ( Figure 6C , Figure 6—source data 3 , Figure 6—figure supplement 2 , Figure 6—figure supplement 2—source data 1 and Figure 6—figure supplement 2—source data 2 ) . Another consistent and striking feature revealed by FynSensor is the temporal bursts of Fyn activity . FRETT levels show clear oscillations over time in serum-starved cells plated on FN and imaged ( Figure 7A , Figure 7—source data 1 , Figure 7—figure supplement 1 , Figure 7—figure supplement 1—source data 1 , Figure 7—video 1 ) . Interestingly , these oscillations in Fyn activity revealed through FynSensor imaging are seen in multiple cell types and appears to be a conserved feature of Fyn activation dynamics . FRETT profiles show temporal oscillations in C2C12 myoblast ( Figure 7B , Figure 7—source data 2 , Figure 7—figure supplement 2 , Figure 7—figure supplement 2—source data 1 ) , Fyn-KD HEK-293T ( Figure 7C , Figure 7—source data 3 , Figure 7—figure supplement 3 , Figure 7—figure supplement 3—source data 1 ) as well as U2OS cells ( Figure 7A ) expressing FynSensor . We also tested if these Fyn activity pulses arise due to any sudden changes in kinase localization . Figure 5—figure supplement 3B–I clearly shows that along with the FynSensor FRETT profiles , even the donor/mCerulean normalized FRET levels show pulsatile behavior . This shows that pulses observed are not due to fluctuations in Fyn concentration levels but arise due to direct and rapid regulation/modulation of Fyn activity ( Figure 5—figure supplement 3B , Figure 5—figure supplement 3—source data 2 ) . Since the pulsatile nature of Fyn activity was found to be conserved across cell types , we specifically examined the duration/frequency of these pulses in U2OS , C2C12 and HEK-Fyn KD cells . Figure 7D , Figure 7—source data 4 shows the mean of the dominant time-period of FynSensor FRET pulses in all three cell-types , determined through power-spectrum density ( PSD ) analysis performed on the FRETT time-traces . The PSD analysis reveals the mean of dominant time-period to be ~3 . 5 min in three very distinct cell types , again highlighting a conserved feature of Fyn activity dynamics . There is increasing evidence to suggest that signaling modules , including growth factor responses in cells may show pulsatile behavior ( Coster et al . , 2017; Warmflash et al . , 2012; Weber et al . , 2010 ) . Bursts of activity are expected to bear signatures of complex positive and negative ( inhibitory ) feedback loops that are integral part of growth factor and other signaling modules ( Avraham and Yarden , 2011; Sparta et al . , 2015; Albeck et al . , 2013 ) . Interestingly , temporal oscillations are highly pronounced in regions of spatially-enhanced Fyn activity as can be seen by plotting FynSensor FRETT against time as well as distance from the celledge . The time-distance 3-D plot of Fyn activity shows bursts of activity in a tight intracellular zone proximal to the edge . ( Figure 7E–I , II , Figure 7—source data 5 ) . Fyn kinase signals downstream of multiple receptor classes , including integrin and receptor tyrosine kinases ( RTKs/Growth factor receptors ) . Based on prior evidence ( DeRita et al . , 2017; Edick et al . , 2007; Samayawardhena et al . , 2007; Parsons and Parsons , 1997 ) , we hypothesized that Fyn , functioning as a signaling node may be involved in dynamically integrating signals downstream of integrins and RTKs ( Lehembre et al . , 2008; Kinnunen et al . , 1998; Arias-Salgado et al . , 2005 ) . We used the FynSensor to investigate this putative signal integration . For this , we examined the effect of platelet-derived growth factor ( PDGF ) on FN-plated , serum-starved U2OS cells and visualized active Fyn using FynSensor FRET . As discussed earlier , serum-starved FN-plated cells already showed spatially enhanced and temporally regulated Fyn activity , consistent with constitutive cell polarization and Fyn activation downstream of spatially regulated integrin signaling ( Figure 5A ) . However , after stimulation with PDGF , FynSensor cells ( but not the P41A non-binding control Figure 8—video 1 ) showed a significant increase in Fyn activity as indicated by enhanced FRET signal ( Figure 8A ) . Strikingly , despite a global PDGF stimulation , the increase in Fyn activity observed was spatially localized . It appeared that on stimulation , the FRETT signal , preferentially increased in regions that already showed higher FRET in un-stimulated cells ( Figure 8A; Figure 8—video 1 , resulting in highly compartmentalized Fyn activity patterns . To quantitatively examine these activity patterns and signal modulation , we again employed automated image analysis . An automated intracellular quadrant analysis confirms differential FynSensor FRET levels across distinct intracellular zones , post-PDGF stimulation ( Figure 8B–I , II , Figure 8—source data 1 ) . These data demonstrate that cells remain ‘polarized’ with respect to Fyn activity despite a global PDGF stimulation . Notably , quadrant image analysis also confirmed a preferential PDGF-induced enhancement of Fyn activity in pre-activated intracellular zones . Figure 8B–II shows that PDGF-induced enhancement in FynSensor FRET is greater in intracellular quadrants already showing higher FRETT pre-stimulation . Overall , this is a powerful demonstration of the ability of FynSensor to reveal hitherto unexplored aspects of Fyn signaling dynamics . FynSensor shows Fyn activity to be compartmentalized even in serum-starved cells . This is likely due to differential integrin signaling , since Fyn activity is significantly attenuated upon FAK inhibition . This spatially constrained Fyn activity could be further increased through growth factor stimulation , with greater enhancement seen in pre-activated areas despite a global stimulation . These results provide a direct illustration of dynamic and intracellular localized signaling crosstalk between integrins and growth factor receptors; visualized through the activation ( conformational change ) of a non-receptor tyrosine kinase that get activated by each of these receptor classes . While results so far imply that differential integrin signaling appears to spatially restrict the effect of growth factor , we asked if appropriate integrin signaling is required for growth factor mediated modulation of Fyn activity . For this , we inhibited focal adhesion kinase activity and specifically tested if growth factor is still able to modulate Fyn activity levels . Interestingly , when FynSensor U2OS expressing cells are treated with PF-562271 ( FAK inhibitor ) , not only are the FynSensor FRET levels drastically reduced , the cells become insensitive to growth factor stimulation as measured through Fyn activity . PDGF-stimulation of inhibitor treated cells show little or no change in FynSensor FRET levels ( Figure 8C , Figure 8—source data 2 , Figure 8—figure supplement 1 , Figure 8—figure supplement 1—source data 1 , Figure 8—video 2 ) . These data indeed confirm a robust , spatio-temporally modulated and functional crosstalk between integrin and growth factor signaling . FynSensor reveals striking spatio-temporal patterns of Fyn activity . Since Fyn is a key regulator of cell physiology , including cytoskeleton remodeling and adhesion dynamics , we also asked how Fyn activity correlates with membrane motility . For this we acquired images at faster acquisition speeds ( ~35 s/frame ) and performed quantitative image analysis using the Fiji plugin ADAPT ( Barry et al . , 2015 ) and measured both membrane motility and FRETT along the periphery of the cell , over the timecourse of the experiment ( Figure 9A I-II , Figure 9—source data 1 ) . Intriguingly , a plot of membrane motility versus FRETT levels shows that the two parameters tend to be inversely correlated ( Figure 9B ) . This is an interesting trend suggesting that intracellular regions showing higher Fyn activity is likely to show reduced overall membrane motility changes . This link between Fyn activity and dampened cell membrane oscillations points to a ‘poised’ cell membrane when/where Fyn is active . Such a notion is consistent with Fyn’s role in regulating cell-matrix adhesions and may need to be probed further ( Figure 9C ) . It is remarkable that visualizing the conformational dynamics of a membrane-bound , non-receptor kinase reveals pulsatile patterns , which can be modulated through integrin and growth-factor signaling .
Direct visualization of protein activity is essential in order to gain a quantitative understanding of dynamic signaling networks that govern cell behavior . Despite the critical roles played by Src family kinases ( SFKs ) in regulating physiology ( Chetty et al . , 2015; Gujral et al . , 2014; Kim et al . , 2009; Lewis-Tuffin et al . , 2015; Nygaard et al . , 2014; Saad , 2009; Timpson et al . , 2001; Zhang et al . , 2013; Zhang et al . , 2014 ) , specific tools/sensors to image activity of individual kinases in live cells and tissues are not available . This is particularly important as individual SFKs can perform overlapping but specific , even seemingly opposing roles , to control cellular output ( Zhang et al . , 2014; Kuo et al . , 2005; Lowe et al . , 1993; Lowell and Soriano , 1996; Marchetti et al . , 1998; Molina et al . , 1992; Palacios-Moreno et al . , 2015 ) . Therefore , understanding the specific activity patterns of individual kinases assumes considerable importance . Our work clearly addresses this issue and establishes a platform for developing new biosensors for visualizing activation of individual Src kinases . Using combinatorial library screening and protein engineering , we develop a biosensor for the critical SFK , Fyn . This specific biosensor reports the activation dynamics of Fyn in live cells , with no interference from other kinases . Fyn kinase is a major regulator of multiple cellular processes and has emerged as a key player in various disease pathologies ( Nygaard et al . , 2014; Bhaskar et al . , 2005; Schenone et al . , 2011; Chin et al . , 2005 ) , but this is the first direct visualization of Fyn activity in cells . Interestingly , currently there are no tools for imaging active Fyn , with even specific antibodies not being available . Our approach is extremely general and can be used for building sensors for other Src family kinases that are critical players in homeostasis and diseases . In this light , our work on addressing the specificity of biosensor binding using NMR structural analysis is likely to be significant . While our targeted screening and binding analysis yields a new binder ( F29 ) that is highly specific for Fyn ( Figure 1 ) , the NMR structure of F29 bound to Fyn SH3 provides fascinating new insights into the molecular basis of specificity ( Figure 2 , Figure 2—figure supplement 1 ) . We identify key interactions that mediate binding and further use this insight to switch specificity of binding ( Figure 2 ) . For instance , our Fyn-specific binder F29 does not or weakly bind Src-SH3 ( Figure 1C , E ) . However , based on predictions from structural analyses , making just a single residue change in Src SH3 allows it to now bind F29 better ( Figure 2—figure supplement 3C ) . These findings offer the possibility of generating specific binders targeting various Src family kinases , leading to new biosensors and tools . More broadly , we showcase how targeted screening of combinatorial protein libraries can be deployed to generate genetically encoded biosensors for visualizing active conformations of signaling proteins . This is a generally applicable strategy for biosensor development . With this method , it is possible to develop sensors even for ‘difficult’ targets , where structural information is limited and conventional tools like antibodies do not exist . We demonstrate that FynSensor is able to report on the activation status of Fyn in cells . Newly engineered binder F29 binds the active form of Fyn and this binding leads to a FRET response that can be measured in live cells ( Figures 3–9 ) . Our sensor design and extensive control experiments ensures high sensitivity while minimizing perturbation . We have shown that the fluorescently labeled Fyn retains kinase activity , is fully functional and behaves like the unmodified kinase in terms of its regulation , localization and ability to modulate downstream signaling ( Figure 3—figure supplements 2 and 5–11 ) . Importantly , the fluorescently labeled Fyn is also able to complement the endogenous Fyn in RNAi-knockdown rescue experiments . For instance , in knockdown-rescue experiments , labeled Fyn is able to rescue downstream ERK phosphorylation similar to untagged Fyn , showing that our tagging of Fyn does not perturb its regulation and signaling function . We also demonstrate that under our imaging conditions ( low-to-moderate expression of FynSensor ) in two very distinct cell types , biosensor does not perturb either downstream signaling or cellular morphodynamics , reiterating the efficacy of our design ( Figure 3—figure supplements 6–11 ) . When the non-binding control version of the F29 binder ( F29P41A ) is used in cells , the FRET response is abolished , showing FRET to be dependent on F29 recognizing activated Fyn ( Figure 5B , Figure 5—figure supplement 1 ) . Further , an inhibitor of kinase activity also significantly attenuates FRET response , showing the biosensor readout to reflect kinase activity ( Figure 5C ) . Overall , we show that the FynSensor response is specific and reflects cellular activity of Fyn . A significant advantage of FynSensor is that it is highly specific and directly reports on the active conformation of Fyn with high spatial and temporal precision , unlike previously reported kinase sensors . This direct visualization of the active form of a single kinase , within a critically important yet complex family of kinases , has led to new insights into signaling dynamics and regulation . FynSensor imaging reveals Fyn activity to be localized and temporally modulated ( Figures 4 , 5 , 6 , 7 and 8 ) , with greater activity closer to the cell edge . Further , as serum-starved , fibronectin ( FN ) -plated cells undergo constitutive cycles of protrusion and retraction , we observe intracellular zones of high Fyn activity and even spontaneous cell polarization measured through levels of active Fyn . Critically , we show that this compartmentalized Fyn activity is dependent on integrin signaling . Inhibition of focal adhesion kinase , a key mediator of integrin signaling , abolishes or greatly attenuates Fyn activity . When FN-plated cells with spatially localized Fyn activity are treated with platelet-derived growth factor ( PDGF ) , we observe a greater increase in Fyn activity in and around the zones that were already pre-activated . This is indeed remarkable . Despite global PDGF treatment , its effect is highly localized and is dependent on intracellular zones that are established through integrin signaling . Our results suggest a model wherein localized/differential integrin activation in cells undergoing constitutive protrusion-retraction cycles not only helps establish signaling zones/compartments with higher Fyn activity , but also sensitizes these zones to be more responsive to growth factor signaling ( Figure 10 ) . This could be through either preferential localization or pre-sensitization of PDGF-receptors and/or associated signaling components ( Figure 10 ) . Thus , compartmentalized integrin signaling is not only maintained but also ‘functionally enhanced’ through growth factor stimulation . We further show that integrin signaling , not only spatially constraints Fyn activity and subsequent growth factor response , but that appropriate integrin signaling is required for growth factor modulation of Fyn activity . Inhibiting focal adhesion kinase activity not only dramatically attenuates Fyn activity in FN-plated , serum-starved cells , it also renders Fyn insensitive to any further stimulation through PDGF ( Figure 8C ) . Our results offer a striking demonstration of spatially localized crosstalk between integrin and growth factor signaling and shows that integrin can localize and regulate the effect of even globally applied growth factors in activating downstream signaling ( Figure 8 ) . This illustrates how by visualizing the dynamic activation of a key signaling node ( Fyn ) , it is possible to directly visualize signaling crosstalk between receptors in cells . Prior work has specifically implicated Fyn for its ability to integrate information emanating from different cell-surface receptors ( Colognato et al . , 2002; Colognato et al . , 2004 ) . For instance , Fyn is plays a crucial role in cell survival and myelination of neurons by oligodendrocytes , specifically by integrating integrin signaling with growth factor signaling . Oligodendrocyte precursors are able to survive and subsequently wrap neurons with a myelin sheath , even in low growth factor conditions through the additive effects of integrin and RTK signaling ( Laursen et al . , 2009; Laursen et al . , 2011; Laursen and Ffrench-Constant , 2007; Sperber and McMorris , 2001; Schäfer et al . , 2016 ) . This signal integration is mediated through Fyn activity ( Laursen et al . , 2009; Schäfer et al . , 2016 ) . It is striking and highly significant that our biosensor reveals evidence of such signal integration , in single cell imaging experiments in-vitro . This shows that compartmentalized signaling crosstalk between different receptor classes may be a key feature of signaling systems ( Figure 10 ) . Biosensor imaging also shows Fyn activity signals to be pulsatile . Recent work has shown that oscillatory signaling patterns are key hallmarks of several signaling modules , including regulation by transcription factors ( Imayoshi et al . , 2013; Isomura and Kageyama , 2014 ) as well as MAP kinases ( Weber et al . , 2010; Sparta et al . , 2015; Albeck et al . , 2013; Antoine-Bertrand et al . , 2016; Shankaran et al . , 2009 ) . Generally , pulses arise due to the presence of negative feedback loops , which may also be accompanied by additional modulators . Interestingly , there is some evidence that switching temporal activity patterns can alter cellular fate; for instance , pulsatile ERK activity leads to one output while persistent activity leads to another . We observe robust and spontaneous activity pulses of Fyn in very distinct cell types even in serum-starved cells plated on FN ( Figure 7 ) , and these pulses are further modulated by growth factor stimulation . This behavior strongly suggests the presence of rapidly activated , negative feedback loops that directly modulate Fyn activity . Fyn activity pulses may be functionally significant in light of its demonstrated role in regulating cell migration , adhesion and actomyosin remodeling . In migration , there is an intrinsic periodicity in cycles of actin polymerization/remodelling/depolymerisation , periods of membrane remodelling as well adhesion assembly and disassembly . A protein that regulates several of these processes is conceptually much more likely to have pulses of activity versus sustained activity . In this light , it is remarkable that we are able to pick up ‘activity pulses’ at specific locations in the cell . This would need to be explored further in the context of directed migration and control of cell proliferation . Further , its indeed striking that membrane motility is reduced in the regions/zones where Fyn is more active ( Figure 9 ) . Spatially-localized pulsatile activation may likely cause transient but limited stabilization of signaling clusters which would be difficult to achieve with sustained activation . These observations taken together suggest that Fyn activity may be correlated with a ‘poised’ membrane state and formation and stabilization of specific integrin-FAK dependent signaling clusters , which can be further tuned by growth factor signaling . Such specific control of adhesion and actomyosin dynamics may be important in Fyn’s ability to mediate directed cell migration , growth control and tissue invasion . Context dependent and spatio-temporally regulated Fyn activity may provide a new perspective on understanding Fyn’s roles in regulating diverse and sometimes opposing cellular functions . Fyn activity profiles also provide a direct readout of highly localized signaling crosstalk between distinct receptor classes . Overall , this work sets the stage for detailed investigations of such functional signaling crosstalk in numerous cellular contexts and offers a new paradigm for direct visualization of signaling dynamics mediated by Fyn and other Src kinases .
Note: Please see Supplementary file 3 . Primers used in study were procured from Bioserve Biotechnologies ( India ) . The oligonucleotide synthesis scale was 25 nmol . Primers up to 50 nucleotide lengths were desalted , however , longer primers were PAGE purified . All DNA sequencings were performed on Illumina- MiSeq DNA Sequencer machine at the DNA sequencing facility at NCBS . Fyn SH3 domain was encoded in a pET-14b plasmid , so that a 6X-His-SUMO-ppx-SH3 fusion protein can be produced . The plasmid was transformed into E . coli BL21- ( DE3 ) cells , and grown in LB to produce unlabeled protein . 13C , 15N-labeled protein was produced by growing the E . coli in vitamin supplemented M9 minimal media with13C6H12O6 and ( Lewis-Tuffin et al . , 2015 ) NH4Cl as the sole source for carbon and nitrogen , respectively . Labeled/Unlabeled Fyn-SH3 domain was purified as follows: Cells were grown up to OD ~0 . 8 , followed by induction with 1 mM IPTG . Four hours post induction , cells were harvested by centrifugation at 4°C , then washed once with Buffer-B ( 50 mM sodium phosphate buffer , pH 8 . 0 , 300 mM NaCl ) , supplemented with 0 . 1% ( v/v ) Triton X-100 ) , and then stored at –80°C . Frozen cells were re-suspended in Buffer-B supplemented with 2 mM phenylmethylsulfonyl fluoride ( 1:10 , wt . /vol . ) . The suspension was sonicated at 4°C for 30 min for lysis , followed by centrifugation at 4°C . The resultant supernatant was allowed to bind Nickel beads ( Thermo Fisher ) , washed with 20 mM imidazole , and eluted with 100–200 mM imidazole . Fractions containing the fusion protein were pooled , and the imidazole was removed by dialysis against Buffer-B for overnight at 4°C . The fusion protein ( 3 mg ) was cleaved by Precission Protease ( 1 mg ) in an overnight 1 ml reaction Buffer-B under mild rotating conditions at 4°C . Cleaved Fyn-SH3 was purified by gel-filtration . Fyn-SH3 fractions were pooled and dialyzed against Buffer-C ( 50 mM sodium phosphate buffer , pH 8 . 0 ) overnight at 4°C . The dialyzed protein was applied to a Mono Q column equilibrated in Buffer-C . The protein was eluted with a linear gradient from 0% to 100% of Buffer-D ( 50 mM sodium phosphate buffer , pH 8 . 0 with 500 mM NaCl ) . Fractions containing pure Fyn SH3 were pooled and concentrated after dialysis in Buffer-B ( 50 mM sodium phosphate buffer , pH 8 . 0 , 300 mM NaCl ) . p-GST-ppx-L15-F29 plasmid was transformed into E . coli BL21 ( DE3 ) cells . The proteins were over-expressed in suitable media as given above . Post lysis , the supernatant was passed through a column of glutathione-Sepharose beads ( GE-Healthcare ) in Buffer-C . After extensive washing , the bound GST-ppx-L15-F29 fusion protein was eluted with 15 mM glutathione in Buffer-C . Fractions containing the fusion protein were pooled and allowed dialyze overnight at 4°C in Buffer-C . The fusion protein was cleaved by Precission Protease . Cleaved F29 was purified by size exclusion chromatography . Fractions containing the F29 were pooled and concentrated up to 1mM for NMR studies . NMR Studies- All NMR spectra were acquired at 298K on an 800 MHz Bruker Avance III spectrometer ( NMR facility-NCBS ) . Assignment of backbone resonances of the proteins were carried out using HNCACB and CBCACONH experiment . 1H and 13C resonance assignments of side chain atoms in F29 were obtained by collecting H ( CC ) CONH and ( H ) CC ( CO ) NH spectra , respectively . A 3D 15N-edited NOESY experiment was carried out on a complex of 15N , 2H-labeled Fyn-SH3 and unlabeled F29 . All the experimental data were processed using NMR pipe ( Delaglio et al . , 1995 ) and TOPSPIN3 . 2 software . Analysis of NMR data was carried out by using Sparky ( Kneller and Kuntz , 1993 ) . Backbone assignments were obtained by PINE-NMR ( Lee et al . , 2009 ) and confirmed manually . Structural model of F29 was calculated by CS-ROSETTA ( Lange et al . , 2012 ) . NMR titration was performed by titrating unlabeled F29 ( ligand ) to a sample of 15N-labelled Fyn-SH3 domain ( protein ) . At least six different protein:ligand ratios were collected ranging from 1:0 . 5 to 1:4 . 1H-15N HSQC spectra were taken at each titration point . The reverse-titration was done by adding unlabeled Fyn-SH3 to labeled F29 . Structure calculation- The structure calculation of the Fyn-SH3/F29 complex was performed using HADDOCK ( van Zundert et al . , 2016 ) . The input structure of Fyn-SH3 is from pdb: 3UA6 , and that of F29 is the output lowest energy structure from CS-ROSETTA . The ambiguous restraints were obtained from the CSPs observed in the NMR titration data . Unambiguous restraints obtained from the NOESY spectra were included in the structure calculation . All restraints were used during the docking steps . The interface of F29 and Fyn-SH3 were kept semi-flexible during simulated annealing and the water refinement . Pull-down experiments were performed to validate the interaction of binder F29 and Fyn SH3 domain . In the plasmid p-GST-ppx-L15-F29 , we have made P41A and R33A mutations in F29 ORF using site directed mutagenesis to yield plasmid p-GST-ppx-L15-F29-P41A and p-GST-ppx-L15-F29-R331 with primers SN 10 , 11 and 12 , 13 respectively . ( 1 ) Y=Bmax* XKD+X Wherein: X is ligand concentration ( Fyn SH3 , nM ) Y is amount of bound protein Bmax is the maximum binding capacity ( same units as the Y-axis ) KD is the equilibrium dissociation constant ( same units as the X-axis , concentration ) . Glutathione sepharose beads saturated with GST-F29 and GST protein ( bait protein ) were used to pull-down the endogenous and exogenously expressed Fyn kinase ( prey protein , regulatable; WT as well as open active CA ) from the cell lysate of HEK-293T cells . We have followed an established protocol ( Sambrook and Russell , 2006 ) with modifications as described below . We have constructed intermolecular FRET-donor and acceptor biosensor using h-Fyn gene and binder ( F29 ) , respectively . In the FRET-donor , mCerulean ( λex=435 nm , λem=475 nm ) and in FRET-acceptor , mVenus ( λex=515 nm , λem=525 nm ) were used . The source of mCerulean and mVenus were from pTriEx-mCerulean-Rac1 WT and pTriEx-mVenus-CBD constructs , respectively ( gifted from Prof . Klaus Hahn's Lab University of North Carolina Chapel Hill ) . The h-Fyn gene source was from pRK5-c-Fyn ( a gift from Dr . Filippo Giancotti , Addgene plasmid # 16032 ) . Biosensors were constructed using gene fusion , quick change mutagenesis , and overlap extension PCR methods ( Kunkel et al . , 1987; Zheng et al . , 2004; Bryksin and Matsumura , 2010 ) . The key steps are shown in Figure 3—figure supplement 1 and details of primers and constructs used in the study can be found in Supplementary file 2 . Biosensors were either made in their original plasmid or in form of PCR fusion-product and were sub-cloned in the pTriEx-4neo vector ( Novagen ) under NcoI and BamH1 site . The strategies used during biosensors construction are summarized here in a step wise manner . The over-expression , protein integrity and activity of the biosensor was analyzed in live cells . Following methods have been adapted: The endogenous Fyn kinase protein of HEK-239T cells was knocked-down ( KD ) using Retrovirus-mediated RNA interference ( sh-RNA ) as described earlier ( Zhang et al . , 2009 ) . pRetroSuper-shFyn construct has been used for endogenous Fyn knockdown which was a gift from Dr . Joan Massague ( Addgene plasmid # 26985 ) . In this case the Fyn kinase ( XM_017010653 . 1 ) 3’UTR region: 5'—GAACTTCCATGGCCCTCAT—3’ was used as shRNA target sequence . For scrambled shRNA control , we have used pSUPER retro puro Scr shRNA , which was a gift from John Gurdon ( Addgene plasmid # 30520 ) where the scrambled shRNA sequence: 5’—GCGAAAGATGATAAGCTAA—3’was used . For packaging of the retrovirus we have used AmphoPack-293 Cell Line ( Gift from Dr Reety Arora/Prof Jyotsna Dhawan , inStem ) . ~60–70% confluent AmphoPack-293 cells ( grown in 10 cm culture plate , with 10 ml DMEM supplemented with 10% FBS ) were transfected with 10 μg of shRNA DNA construct using Lipofectamine LTX and Plus reagent according to the manufacturer's instruction ( Thermo-Fisher Scientific ) in BSL2 . Post-transfection the spent media containing packaged retrovirus were collected at 48 and 72 hr . The collected media ( ~18 ml ) was filtered using 0 . 45 μm filter and used to infect the HEK-293T cells ( 10 cm plate , ~70–80% confluent , grown in DMEM and 10% FBS ) in presence of polybrene ( 10 μg/ml ) . The plate was further incubated for 24 hr and replaced with fresh media having 4 μg/ml of puromycin ( Sigma ) . After 48 hr , only ~3 to 4 cells ( per 10 cm plate ) were found viable in both cases . The media containing puromycin was changed as per requirement until these cells reached ~90% confluency . Finally , the puromycin selected cells were passaged in the media without puromycin and grown . The cells were tested for mycoplasma and found to be negative . The final batches from each case were frozen and stored in liquid N2 . The knockdown of Fyn kinase was verified by immunoblot using specific antibodies . Earlier it has been shown that downstream ERK phosphorylation can be regulated through/is sensitive to Src family kinase ‘Fyn’ signaling ( Wary et al . , 1998 ) . Therefore , ERK phosphorylation offers a sensitive measure of the signaling state . Here we aim to show- ( a ) cellular Fyn levels regulate downstream ERK signaling , ( b ) FynSensor Fyn is able to rescue downstream ERK signaling in Fyn-knockdown cells efficiently , and ( c ) FynSensor , labeled Fyn and F29 binder do not perturb downstream ERK signaling in multiple cell lines . Cellular Fyn levels regulate downstream signaling - In order to show that ERK phosphorylation is sensitive to Fyn signaling , we used stable cells expressing Fyn sh-RNA ( Fyn KD HEK cells ) as well as scrambled shRNA ( control HEK cells ) in this experiment . The cells were seeded at a density of ~3×105/ml in 35 mm tissue culture dishes . After 24 hr , the cells were incubated in media containing 1% serum . 16 hr post incubation , the cells were washed with 1X DPBS and lysed in 200 μl of RIPA lysis buffer . The cell lysates were resolved on 10% SDS gel followed by transfer on a PVDF membrane . Blot region corresponding to 54 to 90 kDa was probed with anti-Fyn antibody , 29 to 54 kDa with anti-p44/42 MAPK ( Erk1/2 ) antibody and anti-Phospho-p44/42 MAPK ( Erk1/2 ) ( Thr202/Tyr204 ) antibody . Vinculin was used as loading control ( 116 kDa ) . The band intensities from at least three such blots were analysed and the levels of phospho-/total ERK were compared using GraphPad Prism . FynSensor Fyn is able to rescue downstream ERK signaling in Fyn-knockdown cells efficiently - In the FynKD-HEK cells , the ERK phosphorylation rescue assay was performed by transiently expressing vector ( pTriex4 Neo ) alone or unlabeled WT Fyn kinase or , FynSensor {FRET donor ( mCer-Fyn ) + FRET acceptor ( myr-mVenus-F29 ) } in reduced serum ( 1% ) media for 18 hr . Using cell lysates from these different conditions , we have performed the immunoblot experiment using specific antibodies as described above . The band intensities from at least three such blots were analysed and the levels of phospho-/total ERK were compared using GraphPad Prism . FynSensor , labeled Fyn and F29 binder do not perturb downstream ERK signalling in multiple cell lines: U2OS cells - Cells were seeded at a density of ~3×105/ml cells/ml in 35 mm culture dishes containing 2 ml media . After 24 hr , cells were transiently transfected with vector , or myr-mVenus or myr-mVenus-F29 ( FRET acceptor ) or Fyn kinase donor ( mCer-Fyn ) or FynSensor . After 18 hr of transient protein expression , cells were lysed using RIPA buffers . The lysates were resolved on to 10% SDS gel followed by transfer on a PVDF membrane . Blot region corresponding to 54 to 90 kDa was probed with anti-Fyn antibody , 29 to 54 kDa with anti-p44/42 MAPK ( Erk1/2 ) antibody and anti-Phospho-p44/42 MAPK ( Erk1/2 ) ( Thr202/Tyr204 ) antibody . Vinculin antibody was used as loading control ( 116 kDa ) . The band intensities from at least three such blots were analysed and the levels of phospho-/total ERK were compared using GraphPad Prism . Mouse myoblast C2C12 cells - Cells were seeded at a density of ~3×105/ml cells/ml in 35 mm culture dishes containing 2 ml media . After 24 hr , cells were transiently transfected with vector , or myr-mVenus or myr-mVenus-F29 ( FRET acceptor ) or Fyn kinase donor ( mCer-Fyn ) or FynSensor . After 18 hr of transient protein expression , cells were lysed using RIPA buffers . The lysates were resolved on to 10% SDS gel followed by transfer on a PVDF membrane . Blot region corresponding to 54 to 90 kDa was probed with anti-Fyn antibody , 29 to 54 kDa with anti-p44/42 MAPK ( Erk1/2 ) antibody and anti-Phospho-p44/42 MAPK ( Erk1/2 ) ( Thr202/Tyr204 ) antibody . β-actin was used as loading control ( 43 kDa ) . Protocol for immunofluorescence assay: Glass-bottom imaging dishes ( Nunc Cat . No . 150682 ) were coated with 10 μg/ml of fibronectin ( FN ) solution . Briefly , the 1 mg/ml stock was diluted to the desired concentration in DPBS and the solution was added to the plates and incubated at 37°C for an hour . Post incubation , plates were washed with DPBS and stored for immediate use at 4°C . Post 16 hr of transfection the HEK 293T Fyn KD ( with FynSensor WT , or unlabeled WT Fyn ) or HEK 293T control cells were plated on glass coverslip coated with FN and allowed to attach for 1 hr . The cells were then fixed and used for immunofluorescence studies ( see Materials and methods above ) . Alexa Fluor-647 conjugated Fyn kinase protein complex was excited using 640 nm laser . In order to capture the spatial expression of Fyn , several optical slices along the z-axis were collected . The slice with maximum intensity for the emission corresponding to 640 nm excitation channel ( specific for Alexa Fluor-647 conjugated Fyn kinase protein ) was chosen , its intensity value was measured using Fiji . The data was analyzed using GraphPad Prism and shown in Figure 3—figure supplement 9 . U2OS cells transiently expressing the FynSensor ( WT-mCer-Fyn+mVenus-F29 ) were plated on FN coated glass bottomed dish and imaged on the Olympus IX83 inverted microscope coupled to the cellTIRF module fitted with an Okolab stage-top live-cell incubator to maintain cells at 37OC at 5% CO2 . Cells were excited using the 445 nm laser and imaged using a 100X objective ( Olympus UAPON 100X oil TIRF objective , NA = 1 . 49 ) at a penetration depth of 100 nm . Images were captured on the Evolve 512 Delta EMCCD camera at an exposure of 400 ms . Images were acquired using the Olympus cellSens software and processed using Fiji . The Acceptor photo-bleaching experiment on SU6656 treated cells was carried on the Olympus FV1000 confocal microscope using a 63X oil immersion objective . FRET-donor ( mCer-Fyn ) was excited using the 405 nm laser and emission from ( 425-500 ) nm was collected . Acceptor ( myr-mVenus-F29 ) was excited using the 515 nm laser and emission from ( 519-619 ) nm was collected . For bleaching the acceptor , the 515 nm laser was used at a high laser dose ( 1 . 38 mW for 27 s ) . Post bleaching , both the donor and acceptor channel images were acquired . The data analysis was done as described in the Materials and methods section . Cells transfected with FynSensor were serum starved and plated as mentioned in Materials and methods . Cells were then treated with 2 µM of SU6656 inhibitor ( 2 mM stock in 100% DMSO ) . Post-inhibitor addition , bleaching of the acceptor was carried on . DMSO controls were treated and imaged in the same way . For single-cell imaging experiments , transient transfections in all the cells ( U2OS , C2C12 and Fyn-KD-HEK 293T ) were carried out using Lipofectamine 3000 according to the manufacturer's instruction ( Life Technologies ) . Cells were checked for expression of plasmid of interest after 18 hr post transfection by fluorescence microscopy . Cells were then starved for 4–6 hr prior to being plated on glass-bottomed dishes coated with desired density of FN ( see Materials and methods ) and allowed to adhere for 30–40 min prior to imaging . Serum-starved cells plated on FN were imaged for 25 min ( acquisition settings are mentioned in following section ) to record basal Fyn activity ( activity in response to integrin activation/engagement ) in a serum-starved state . For stimulation experiments , cells were initially imaged for 5–6 min to record basal Fyn activity . PDGF at a final concentration of 10 ng/ml ( 20 μg/ml stock in 1X DPBS ) was then added onto the cells and images for an additional 15–16 min were acquired . For faster imaging , serum-starved cells plated on FN were imaged for 10–15 min at an inter frame gap of ~35 s ( acquisition settings are mentioned in the following section . For treatment of cells with FAK inhibitor ( PF-562271 ) , serum-starved cells plated on FN were imaged for an initial period of 10 min and the inhibitor PF-562271 was added at a final concentration of 10 µM ( 10 mM stock in 100% DMSO ) . Post inhibitor addition cells were imaged for an additional 10 min . The cells were then treated with PDGF ( 10 ng/ml ) and imaged for another 10 min . All imaging experiments ( unless mentioned ) were carried on the Olympus FV3000 microscope equipped with an PLANAPO N 60X oil immersion objective with a NA = 1 . 42 attached to a stage-top live-cell incubator to maintain cells at 37°C with 5% CO2 ( Tokai Hit ) . The microscope has two High Sensitivity Detectors ( HSDs ) . The microscope also has a Z-Drift compensator and a motorized XY-stage which allows for multi-point-time-lapse imaging . Dex-Dem images were acquired using an excitation wavelength of 405 nm and emission was collected on HSD1 and the bandwidth of collection was from ( 437-499 ) nm . Aex-Aem images were acquired using an excitation wavelength of 514 nm and emission was collected on HSD2 and the bandwidth of collection was ( 528-628 ) nm . Dex-Aem images were acquired using donor excitation and emission was same as acceptor emission . Voltage and gain settings were kept same while imaging the same cell prior to and after any kind of treatment ( Bleaching/addition of PDGF/addition of inhibitor ) . Two methods were used to study FRET response in cells . The details of the methods used are listed below: Acceptor Photo-bleaching Method-For cells expressing FynSensor or non-binding mutant , prebleach and post-bleach Dex-Dem , and Aex-Aem images were acquired as described above . Acceptor bleaching was performed using 514 nm laser at 2X , 5X and 10 X light dosage for 10 s each ( X = 0 . 069 mW ) ( refer to Figure 4B & C ) and Dex-Dem , and Aex-Aem images acquired . For determining the amount and pattern of the increase in donor fluorescence post acceptor photo-bleaching , a customized MATLAB program was used . Dex-Dem ( donor channel ) images were used for this analysis . All the images were subjected to background subtraction and a thresholding operation before analysis . A user-defined global threshold was first used to generate a binary mask using the donor channel pre-bleach image of each cell . This was then used to determine the boundary of the cell using Canny’s edge detection algorithm ( Canny , 1986; Mathworks , 2019a ) . A centroid of each cellular mask was also determined . A ‘linescan’ algorithm was used to quantify the variation of donor fluorescence recovery with distance from the cell periphery . For this , lines from every point on the periphery of the cell to the centroid were generated . Donor fluorescent intensities ( from raw images ) along these lines were calculated , and averages of all intensities were obtained as a function of distance from the periphery . For a given cell , the donor intensity values before and after bleaching ( Pre-bleachDonor and Post-bleachDonor ) were thus determined and used in the following equation to calculate percentage recovery of donor fluorescence as per established methods ( Karpova et al . , 2003 ) . ( 2 ) Donorrecovery ( % ) = ( 1−IPre−bleachDonorIPost−bleachDonor ) ∗100where , I Pre-bleachDonor - donor fluorescence intensity prior to bleaching . I Post-bleachDonor - donor fluorescence intensity after bleaching . For generating the representative ∆Donor images , in Figure 4A the prebleach donor channel image was subtracted from the postbleach donor channel image of the same cell ( ∆Donor = Post bleachDonor-Pre-bleachDonor ) using MATLAB . Donor Sensitized emission method-The Donor Sensitized Emission ( SE ) method records the FRET signal by measuring the amount of acceptor emission after excitation of the donor . The method used here has been modified from the original method as described earlier ( Gordon et al . , 1998 ) and requires the Dex-Dem , Dex-Aem and Aex-Aem images for calculation of FRET . To calculate the levels of both donor and acceptor bleed-through and direct acceptor cross-excitation using donor excitation , ‘donor only’ and ‘acceptor only’ samples were imaged in the following way: All the image analyses were done using MATLAB and plots have been generated using Graph Pad Prism 5 , unless otherwise specified . The customized codes written for the purpose of image analysis are available on GitHub . All images acquired were converted to 8-bit format prior to processing . FRET measurements were done on cells expressing both donor and acceptor fluorophores of FynSensor using Equation 3 , followed by a median filtering with a 2 × 2 pixel block around each pixel ( Lim , 1990 ) . The Fiji plugin ‘FRET analyzer' ( Hachet-Haas et al . , 2006 ) was used to determine the bleed through values used in Equation 3 from donor-only and acceptor-only cells . FRET index images are shown using ImageJ ( NIH ) Fire LUT and scaled by an arbitrary number only for visual representations . Briefly , for our sensitized FRET experiments , the following equation was used to generate the FRET index images: ( 3 ) FRETindeximage=IFRET− ( αDIDonor ) − ( αAIAcceptor ) Wherein: αD – is the mean of the slope of the bleed-through signal of the donor alone ( here 0 . 55 ) , αA - is the mean of the slope of the bleed-through signal of the acceptor alone ( here 0 . 028 ) , IFRET - is the FRET channel image , IDonor – is the Donor channel image , and , IAcceptor – is the Acceptor channel images . FRET index images depict total FRET ( FRETT ) Calculations used are consistent with other literature reports on intermolecular FRET measurements , like the Fiji plugin ‘FRET analyzer' ( Hachet-Haas et al . , 2006 ) and another very recent report ( Oldach et al . , 2018 ) . For cellular area estimation , acceptor channel image time series ( Aex-Aem ) for each cell was processed , ( global thresholding , smoothening by erosion operation and filling holes using imfill and removing small area particles by bwareafilt ) . Cellular area processed images ( binary images with a value of 1 inside cell and 0 outside ) hence generated , were used as masks to further process the FRET index images ( or ‘YFP channel’ or ‘CFP channel images as the case may be ( Figure 5—figure supplement 2A and Figure 5—figure supplement 3B ) . Cell centroids were determined on the binary masks using regionprops function . With respect to the centroid , a vertical and horizontal line was drawn to divide each cell into four quadrants as shown in Figure 5A–I . Mean fluorescent intensity ( either FRET or ‘CFP’ or ‘YFP’ ) in each quadrant was calculated using Equation 4; wherein sum of the fluorescent intensity values from all pixels in a given quadrant is divided by total number of pixels in the quadrant . The mean intensity is followed over time and the variation is plotted in Figure 7A–II , B–II , C–II . ( 4 ) Meanintensity= ∑i=1nPixelValueinn = no . of pixels in the specific quadrant For testing if Fyn activity is spatially enhanced , mean FRETT values ( see above ) for every quadrant in each cell was analyzed over time . Max-FRETT-HFQ ( HFQ: high-FRET quadrant ) refers to the maximum value of mean quadrant FRETT seen in the time series , and was determined for each cell . Max-FRETT-LFQ ( LFQ:low FRET quadrant’ ) was the corresponding mean FRETT value for lowest FRET quadrant at that specific time point , for the given cell . Max-FRETT-HFQ and Max-FRETT-LFQ were averaged over multiple cells for both serum-starved as well as PDGF-stimulated cells . For calculating PDGF-induced enhancement in Fyn activity , difference in the FRET intensity levels ( Max-FRETT-HFQ ) before and after PDGF stimulation were calculated . Also calculated were differences in Max-FRETT-LFQ before and after PDGF . Power spectral density ( PSD ) ( Proakis John , 1996 ) of the time series was computed using periodogram function . Prior to subjecting the FRETT time traces to frequency decomposition analysis , we removed any quadratic trend using the detrend function in line with established procedures ( Mitov , 1998; Colak , 2009 ) . The frequency against which maximum power ( Proakis John , 1996; Mathworks , 2019b ) is obtained , known as dominant frequency in the time series , was used to calculate the time-period of FRETT intensity oscillations in each quadrant . The frequency is converted to time period based on the sampling frequency , and is plotted in the Figure 7D . The maximum FRET intensity region of cell membrane was identified through line scanning . An arc of arbitrary length unit ( 40 pixels here ) on the membrane encompassing this region of maximum FRET intensity was selected , and a sector was constructed which spanned the membrane at the cell edge and extended towards the centroid of the cell . The intensity in this sector was measured over time and plotted as a function of distance from the cell edge ( Figure 7E ) . Donor channel images ( Dex-Dem ) were used to generate a cell mask and calculate the cellular area at each time point ( see above ) . Two consecutive cell mask images were used to determine the temporal change in the cellular area ( ΔAi=Ai+1-Ai ) . The absolute value of this temporal change in cell area was then normalized for the initial cell area ( |ΔAi|/A1 ) to account for differently sized cells . This was then used to calculate the mean fractional area change to reflect the cumulative area changes over a period of time ( Equation 5a ) . Similarly , cumulative cell perimeter changes were determined using the modulus of change in perimeter between consecutive cell frames , normalized for the initial cell boundary length ( Equation 5b ) . ( 5a ) MeanFractionalAreaChange=∑1n−1|ΔAi|A1n−1 ( 5b ) MeanFractionalPerimeterChange=∑1n−1|ΔPi|P1n−1 In the above equation , n is total numbers of frames recorded . Once mean fractional area and perimeter changes were determined for a single cell , averages from multiple such cells were then calculated for each of the parameters in each set has been plotted Figure 3—figure supplement 7 . We have used the open source plugin ADAPT , developed by Barry et al . ( 2015 ) quantitatively examine the correlation between membrane motility and FRET intensity . For this purpose , we use the acceptor channel image ( Aex-Aem ) as ‘reference channel’ for demarcating the cell boundary . FRET index image ( obtained as per Equation 3 ) is used as the ‘signal channel’ ( FRET intensity/Fyn activity ) . The values used for resolution ( micron/pixel ) and frames/min are 0 . 2 and 2 , respectively . The ‘signal’ signifying Fyn activity approximately 1 µm from the membrane and the ‘velocity’ at the membrane along the cell periphery and over time , are obtained using the plugin . The overall ‘membrane motility’ map is obtained by performing a modulus operation on the velocity map . 3D plot ( Figure 9A ) shows changes in ‘signal’ and ‘membrane motility’ over time and position along cell boundary for a single cell . To compute a global correlation between FRET intensity ( Fyn activity ) and ‘membrane motility’ a one-to-one mapping of these two arrays extracted from ADAPT plugin , was performed . For the purpose of quantification , we have binned the FRET intensity values into bins of 3 intensity units [ ( 9-12 , 12-15 , 15-18 ) ] and so on . The curve-fitting was done using the polynomial cubic equation ( f = y0+a*x+b*x^2+c*x^3 ) on SigmaPlot ( see Figure 9B ) . Key resource table: Provided as Supplementary file 3 of this manuscript . | Cells contain networks of signaling proteins that can respond to a variety of cues from the surrounding environment . Often the cell’s response to these cues is not just controlled by the level of protein , but by changing the activity of signaling proteins . For example , a signaling protein in humans and other mammals known as Fyn regulates a number of different processes , including when a cell grows , dies , or develops a specialist role . Defects in the activity of Fyn are associated with several diseases in humans including cancer and Alzheimer’s disease . However , it remains unclear how Fyn contributes to these diseases , or how the protein is able to precisely coordinate responses to multiple different cues in healthy individuals . This is largely because there are no readily available tools that are able to specifically detect where and when this protein is active in cells . Researchers often use fluorescent proteins called biosensors as tools to detect where specific proteins are located in living cells over time . Now , Mukherjee , Singh et al . have developed a new biosensor named FynSensor to monitor the active form of Fyn in mammalian cells . Microscopy imaging of FynSensor in several different cell types showed that although Fyn was present everywhere , it was only active in certain areas . In these areas the protein switched between an active and inactive state , with clear ‘pulses’ of signaling activity lasting a couple of minutes in response to specific cues . These areas of high Fyn activity behaved like signaling hubs in which several different cues integrate together before Fyn triggers an appropriate cell response . These results shed light on how Fyn is able to precisely control many different processes in cells . In the future , FynSensor could be used to rapidly screen for drug-like molecules to treat cancer , Alzheimer’s disease and other conditions linked with defects in Fyn activity . Furthermore , the FynSensor could be adapted to allow researchers to study other signaling proteins in humans and other animals . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"cell",
"biology"
] | 2020 | A Fyn biosensor reveals pulsatile, spatially localized kinase activity and signaling crosstalk in live mammalian cells |
Planarian neoblasts are pluripotent , adult somatic stem cells and lineage-primed progenitors that are required for the production and maintenance of all differentiated cell types , including the germline . Neoblasts , originally defined as undifferentiated cells residing in the adult parenchyma , are frequently compared to embryonic stem cells yet their developmental origin remains obscure . We investigated the provenance of neoblasts during Schmidtea mediterranea embryogenesis , and report that neoblasts arise from an anarchic , cycling piwi-1+ population wholly responsible for production of all temporary and definitive organs during embryogenesis . Early embryonic piwi-1+ cells are molecularly and functionally distinct from neoblasts: they express unique cohorts of early embryo enriched transcripts and behave differently than neoblasts in cell transplantation assays . Neoblast lineages arise as organogenesis begins and are required for construction of all major organ systems during embryogenesis . These subpopulations are continuously generated during adulthood , where they act as agents of tissue homeostasis and regeneration .
Neoblasts are planarian adult somatic stem cells that exhibit levels of plasticity and pluripotency comparable to embryonic and induced pluripotent stem cells ( Elliott and Sánchez Alvarado , 2013; Rink , 2013; Wagner et al . , 2011 ) . In flies , fish , mice and humans , adult somatic stem cells are fate-restricted , sustaining production of cell lineage ( s ) in resident tissues ( Fuchs and Segre , 2000; Wagers and Weissman , 2004 ) . Although embryonic stem cells cultured ex vivo remain capable of producing a diversity of tissue types from different germ layers , such plasticity is typically lost from most somatic cells as development proceeds . In contrast , the planarian neoblast population is wholly responsible for the production of all differentiated cell types in these bilaterally symmetric , triploblastic animals ( Baguñà and Auladell , 1989 ) . In fact , transplantation of a single neoblast into a stem cell deficient host was sufficient for rescue and long-term reconstitution ( Wagner et al . , 2011 ) , confirming the pluripotency of planarian somatic stem cells . Despite longstanding discussion of the similarities between neoblasts and embryonic stem cells , a comparison explicitly stated in the original definition of the term ( Randolph , 1892 ) , the provenance of neoblasts during embryogenesis was unknown . Neoblasts are abundant and widely distributed across the anteroposterior axis , occupying the parenchymal space surrounding the gut ( Reddien et al . , 2005 ) . All neoblasts contain chromatoid bodies ( Auladell et al . , 1993; Hay and Coward , 1975; Hori , 1982; Morita et al . , 1969 ) and express nuage genes , including piwi-1 , and factors implicated in germ cell identity , genome surveillance and post-transcriptional regulation of gene expression ( Guo et al . , 2006; Palakodeti et al . , 2008; Reddien et al . , 2005; Rouhana et al . , 2010 , 2012; Salvetti et al . , 2005; Shibata et al . , 1999; Solana et al . , 2009; Wagner et al . , 2012; Yoshida-Kashikawa et al . , 2007 ) . Neoblasts are the only cycling somatic cells in adults ( Baguñà , 1976; Newmark and Sánchez Alvarado , 2000; Orii et al . , 2005; Salvetti et al . , 2000 ) ; quiescent neoblasts were not observed in BrdU pulse chase experiments ( Newmark and Sánchez Alvarado , 2000 ) . Mounting evidence suggests that the neoblast population contains pluripotent stem cells as well as cycling , lineage-primed progenitors ( Reddien , 2013 ) . Heterogeneous expression of developmental transcription factors ( TFs ) in neoblasts has been reported and likely reflects the diversity of lineage-primed progenitors within the compartment ( Adler et al . , 2014; Cowles et al . , 2013; Currie and Pearson , 2013; Lapan and Reddien , 2011 , 2012; März et al . , 2013; Pearson and Sánchez Alvarado , 2010; Scimone et al . , 2014 , 2011; van Wolfswinkel et al . , 2014; Wenemoser et al . , 2012 ) . Schmidtea mediterranea ( Smed ) freshwater flatworms are stable diploids that exist as two biotypes: asexual animals that reproduce by fission , and obligate cross-fertilizing hermaphrodites that reproduce sexually ( Newmark and Sánchez Alvarado , 2002; Newmark et al . , 2008 ) . Both biotypes mount robust regeneration responses following amputation , and similarly rely on neoblasts for homeostatic maintenance and regeneration of all tissues . The asexual clonal line CIW4 ( C4 ) has received the most scrutiny in studies examining the molecular mechanisms underlying regeneration , neoblast maintenance , pluripotency and lineage commitment ( Newmark and Sánchez Alvarado , 2002 ) . However , neoblasts are ever-present in C4 animals , precluding investigation of their developmental origin . Neither a normal table for Smed embryonic development nor functional studies have been reported . Our work establishes Smed as a developmental model system and leverages the novel , unexploited context of embryogenesis to hone the molecular and operational definition of the planarian neoblast . We generated a molecular staging resource for Smed embryogenesis that associates unique gene expression signatures with chronological age , embryo morphology , representative images and written summaries of key developmental events to holistically describe and define prototypes for each stage . We also provide an atlas of molecular markers describing temporary embryonic tissue types and definitive organ system development . These data , found in the supplementary material , are also searchable online at https://planosphere . stowers . org . We investigated the developmental origin of neoblasts during Smed embryogenesis and show that early embryonic cells are molecularly and functionally distinct from the adult neoblast population . Pluripotent neoblasts and lineage-dedicated progenitors arise as organogenesis begins . Our results suggest that the framework for understanding cell fate specification and organ formation during Smed embryogenesis diverges radically from existing developmental paradigms . Here , in a bilaterally symmetric , triploblastic animal not thought to undergo gastrulation ( Cardona et al . , 2005; Le Moigne , 1963; Sánchez Alvarado , 2003; Stevens , 1904 ) , heterogeneous expression of key developmental regulators within a pluripotent , cycling blastomere population generates the panoply of lineage-dedicated progenitors required for organogenesis . Moreover , neoblasts perpetuate embryonic developmental programs during adulthood , where they are required for continued maintenance and rebuilding of tissues during homeostasis and regeneration .
Smed flatworms are direct developers: newborn hatchlings grow and mature into adult worms without an intervening larval stage ( Sánchez Alvarado , 2003 ) . At hatching , juveniles are sexually immature but otherwise possess a body plan grossly similar to that of adult hermaphrodites ( Sánchez Alvarado , 2003; Wang et al . , 2007 ) . Smed embryos undergo an evolutionarily divergent mode of development that bears little resemblance to the ancestral Spiralian cleavage programs utilized by many Lophotrochozoans . Smed embryos are ectolecithal: yolk is not contained within oocytes , but rather is produced by somatic vitellaria ( yolk glands ) arrayed ventrolaterally beneath the testes ( Chong et al . , 2011; Steiner et al . , 2016; Stevens , 1904 ) . Oocytes are fertilized internally by sperm from a partner . Zygote ( s ) are packaged , along with yolk cells , into an egg capsule in the genital atrium ( Figure 1A ) ( Chong et al . , 2011; Hyman , 1951; Newmark et al . , 2008; Stevens , 1904 ) . 10 . 7554/eLife . 21052 . 003Figure 1 . A molecular staging series for Smed embryogenesis informed by single embryo RNA-Seq . ( A ) Top: Cartoon depicting the reproductive system of a sexually mature Smed hermaphrodite . Ventral view . G , gonopore; O , ovary; OD , oviduct; P , penis papilla; SD , sperm duct . Oocytes are fertilized internally and zygote ( s ) are packaged with yolk produced by vitellogenic gland cells into a developing egg capsule in the genital atrium ( purple ) . Capsules are laid through the gonopore . Bottom: Brightfield image of a live Smed hermaphrodite . Anterior: left . Dorsal view . White asterisk: pharynx . ( B ) Developmental timeline and staging designations for Smed embryogenesis at 20°C . Timeline: days ( d ) post egg capsule deposition . Gray bars and letters C–I indicate time windows , and corresponding panels ( C–I ) , for RNA-Seq samples . Double-headed arrows: time windows for stages ( S ) S1–S8 . ( C–I ) Brightfield images of live embryos harvested for RNA-Seq ( top ) , hematoxylin- and eosin-stained sections ( middle ) , and heat maps for enriched transcripts ( bottom ) . Scale bars: 100 µm . Yellow arrowheads: temporary embryonic pharynx . Black arrowheads: definitive pharynx . Heat maps depict cohorts of enriched transcripts at indicated stages . ( J ) Principal component analysis demonstrates clustering of replicates and separation of developmental time points in expression space . ( K ) Correlation matrix for single embryo sequencing replicates . Total transcripts with a row sum >1 CPM: 31 , 248 . ( C–K ) Y , yolk . 2 , S2 . 3 , S3 . 4 , S4 . 5 , S5 . 6 , S6 . 7 , S7 . 8 , S8 . DOI: http://dx . doi . org/10 . 7554/eLife . 21052 . 00310 . 7554/eLife . 21052 . 004Figure 1—source data 1 . Molecular staging resource for Smed embryogenesis . Tab 1 ( staging overview ) : an overview of the molecular staging resource materials for S1–S8: developmental time interval ( days post-egg capsule deposition , 20°C ) ; names of single embryo RNA-Seq replicates; references to representative images ( live brightfield and histological cross-sections ) ; number of enriched transcripts; references to Figure 1 supplement files containing mean centered expression and raw RPKM profiles across embryogenesis for enriched transcripts; references to Figure 1 source data files ( excel spreadsheets ) containing enriched transcripts , organized by cluster membership , with RPKM profiles across development , BLASTx-based annotations , GO analysis , and short written descriptions of each stage ( S2–S8 ) . Additional tabs are included for: ( 1 ) pairwise comparison overview; ( 2 ) mixed stage reference overview; ( 3 ) lists of enriched transcripts for S2–-S8 , compiled from both the pairwise and mixed stage reference analysis; ( 4 ) GO triage criteria; ( 5 ) categories and lists of biological process ( BP ) GO IDs , manually curated from the statistically significant hits for S2–S8-enriched transcripts; ( 6 ) summary table containing the number and percentage of enriched transcripts ( S2–S8 ) assigned to BP GO ID categories . DOI: http://dx . doi . org/10 . 7554/eLife . 21052 . 00410 . 7554/eLife . 21052 . 005Figure 1—source data 2 . Stage-2-enriched transcripts from pairwise and/or mixed stage reference comparisons . Criteria for inclusion are indicated in Figure 1—source data 1 , as well as the legends for Figure 1—figure supplements 2–3 . Tabs in this excel file contain: ( 1 ) pairwise comparison data ( if applicable ) , ( 2 ) mixed stage reference comparison data , ( 3 ) cluster membership ( see Figure 1C ) , average RPKM values across embryogenesis ( Y–S8 ) , and in C4 and SX adults , as well as best BLASTx hits ( E < 0 . 001 ) versus the NR , Swiss-Prot , C . elegans , D . melanogaster , D . rerio , X . tropicalis , M . musculus and H . sapiens RefSeq databases , ( 4 ) GO analysis: manually curated and categorized biological process ( BP ) GO IDs and ( 5 ) GO analysis: unabridged results . See also Figure 1—figure supplement 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 21052 . 00510 . 7554/eLife . 21052 . 006Figure 1—source data 3 . Stage-3-enriched transcripts from pairwise and/or mixed stage reference comparisons . Criteria for inclusion are indicated in Figure 1—source data 1 , as well as the legends for Figure 1—figure supplements 2–3 . Tabs in this excel file contain: ( 1 ) pairwise comparison data ( if applicable ) , ( 2 ) mixed stage reference comparison data , ( 3 ) cluster membership ( see Figure 1D ) , average RPKM values across embryogenesis ( Y–S8 ) , and in C4 and SX adults , as well as best BLASTx hits ( E < 0 . 001 ) versus the NR , Swiss-Prot , C . elegans , D . melanogaster , D . rerio , X . tropicalis , M . musculus and H . sapiens RefSeq databases , ( 4 ) GO analysis: manually curated and categorized biological process ( BP ) GO IDs , and ( 5 ) GO analysis: unabridged results . See also Figure 1—figure supplement 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 21052 . 00610 . 7554/eLife . 21052 . 007Figure 1—source data 4 . Stage-4-enriched transcripts from pairwise and/or mixed stage reference comparisons . Criteria for inclusion are indicated in Figure 1—source data 1 , as well as the legends for Figure 1—figure supplements 2–3 . Tabs in this excel file contain; ( 1 ) pairwise comparison data ( if applicable ) , ( 2 ) mixed stage reference comparison data , ( 3 ) cluster membership ( see Figure 1E ) , average RPKM values across embryogenesis ( Y–S8 ) , and in C4 and SX adults , as well as best BLASTx hits ( E < 0 . 001 ) versus the NR , Swiss-Prot , C . elegans , D . melanogaster , D . rerio , X . tropicalis , M . musculus and H . sapiens RefSeq databases , ( 4 ) GO analysis: manually curated and categorized biological process ( BP ) GO IDs , and ( 5 ) GO analysis: unabridged results . See also Figure 1—figure supplement 6 . DOI: http://dx . doi . org/10 . 7554/eLife . 21052 . 00710 . 7554/eLife . 21052 . 008Figure 1—source data 5 . Stage-5-enriched transcripts from pairwise and/or mixed stage reference comparisons . Criteria for inclusion are indicated in Figure 1—source data 1 , as well as the legends for Figure 1—figure supplements 2–3 . Tabs in this excel file contain; ( 1 ) pairwise comparison data ( if applicable ) , ( 2 ) mixed stage reference comparison data , ( 3 ) cluster membership ( see Figure 1F ) , average RPKM values across embryogenesis ( Y–S8 ) , and in C4 and SX adults , as well as best BLASTx hits ( E < 0 . 001 ) versus the NR , Swiss-Prot , C . elegans , D . melanogaster , D . rerio , X . tropicalis , M . musculus and H . sapiens RefSeq databases , ( 4 ) GO analysis: manually curated and categorized biological process ( BP ) GO IDs , and ( 5 ) GO analysis: unabridged results . See also Figure 1—figure supplement 7 . DOI: http://dx . doi . org/10 . 7554/eLife . 21052 . 00810 . 7554/eLife . 21052 . 009Figure 1—source data 6 . Stage-6-enriched transcripts from pairwise and/or mixed stage reference comparisons . Criteria for inclusion are indicated in Figure 1—source data 1 , as well as the legends for Figure 1—figure supplements 2–3 . Tabs in this excel file contain; ( 1 ) pairwise comparison data ( if applicable ) , ( 2 ) mixed stage reference comparison data , ( 3 ) cluster membership ( see Figure 1G ) , average RPKM values across embryogenesis ( Y–S8 ) , and in C4 and SX adults , as well as best BLASTx hits ( E < 0 . 001 ) versus the NR , Swiss-Prot , C . elegans , D . melanogaster , D . rerio , X . tropicalis , M . musculus and H . sapiens RefSeq databases , ( 4 ) GO analysis: manually curated and categorized biological process ( BP ) GO IDs , and ( 5 ) GO analysis: unabridged results . See also Figure 1—figure supplement 8 . DOI: http://dx . doi . org/10 . 7554/eLife . 21052 . 00910 . 7554/eLife . 21052 . 010Figure 1—source data 7 . Stage-7-enriched transcripts from pairwise and/or mixed stage reference comparisons . Criteria for inclusion are indicated in Figure 1—source data 1 , as well as the legends for Figure 1—figure supplements 2–3 . Tabs in this excel file contain; ( 1 ) pairwise comparison data ( if applicable ) , ( 2 ) mixed stage reference comparison data , ( 3 ) cluster membership ( see Figure 1H ) , average RPKM values across embryogenesis ( Y–S8 ) , and in C4 and SX adults , as well as best BLASTx hits ( E < 0 . 001 ) versus the NR , Swiss-Prot , C . elegans , D . melanogaster , D . rerio , X . tropicalis , M . musculus and H . sapiens RefSeq databases , ( 4 ) GO analysis: manually curated and categorized biological process ( BP ) GO IDs , and ( 5 ) GO analysis: unabridged results . See also Figure 1—figure supplement 9 . DOI: http://dx . doi . org/10 . 7554/eLife . 21052 . 01010 . 7554/eLife . 21052 . 011Figure 1—source data 8 . Stage-8-enriched transcripts from pairwise and/or mixed stage reference comparisons . Criteria for inclusion are indicated in Figure 1—source data 1 , as well as the legends for Figure 1—figure supplements 2–3 . Tabs in this excel file contain; ( 1 ) pairwise comparison data ( if applicable ) , ( 2 ) mixed stage reference comparison data , ( 3 ) cluster membership ( see Figure 1I ) , average RPKM values across embryogenesis ( Y–S8 ) , and in C4 and SX adults , as well as best BLASTx hits ( E < 0 . 001 ) versus the NR , SwisSchmidteas-Prot , C . elegans , D . melanogaster , D . rerio , X . tropicalis , M . musculus and H . sapiens RSeq databases , ( 4 ) GO analysis: manually curated and categorized biological process ( BP ) GO IDs , and ( 5 ) GO analysis: unabridged results . See also Figure 1—figure supplement 10 . DOI: http://dx . doi . org/10 . 7554/eLife . 21052 . 01110 . 7554/eLife . 21052 . 012Figure 1—source data 9 . Molecular fate mapping resource . Short summaries describing molecular markers for temporary embryonic tissues and definitive organ systems . Written descriptions accompany Figure 1—figure supplement 11 ( primitive ectoderm ) , Figure 1—figure supplement 12 ( temporary embryonic pharynx ) , Figure 1—figure supplement 13 ( gut ) , Figure 1—figure supplement 14 ( definitive pharynx ) , Figure 1—figure supplement 15 ( definitive epidermis ) , Figure 1—figure supplement 16 ( nervous system ) , Figure 1—figure supplement 17 ( muscle ) , Figure 1—figure supplement 18 ( protonephridia ) and Figure 1—figure supplement 19 ( eyes ) . DOI: http://dx . doi . org/10 . 7554/eLife . 21052 . 01210 . 7554/eLife . 21052 . 013Figure 1—figure supplement 1 . Histological cross-sections of Stage one embryos . ( A–D ) Four independent examples of the single cell stage ( S1 ) , either metaphase II arrested oocytes or zygotes , in paraffin embedded , hematoxylin- and eosin-stained sectioned egg capsules fixed at 1–2 days post egg capsule deposition . The entire capsule is shown in ( D ) . Black arrowheads: metaphase II arrested oocytes or zygotes , surrounded by a corona of yolk cells . Scale bars: 25 µm ( A–C ) ; 100 µm ( D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 21052 . 01310 . 7554/eLife . 21052 . 014Figure 1—figure supplement 2 . MA plots for pairwise comparisons . ( A ) S2 vs Y . ( B ) S2 vs S3 . ( C ) S3 vs S4 . ( D ) S4 vs S5 . ( E ) S5 vs S6 . ( F ) S6 vs S7 . ( G ) S7 vs S8 . Upregulated transcripts: red . Downregulated transcripts: green . Criteria for flagging differentially expressed transcripts: adjusted p-value<1e-5 , log2 ratio ≥2 . 322 ( five-fold upregulation ) or log2 ratio ≤−2 . 322 ( five-fold downregulation ) , average scaled RPKM value ≥1 . 0 at indicated time point , transcript has ≥1 ORF . Transcripts derived from transposase or retroviruses were omitted . S2–S8 , Stages 2–8 . Y , yolk . DOI: http://dx . doi . org/10 . 7554/eLife . 21052 . 01410 . 7554/eLife . 21052 . 015Figure 1—figure supplement 3 . MA plots for enriched transcripts identified in mixed-stage reference comparisons . ( A ) S2 , ( B ) S3 , ( C ) S4 , ( D ) S5 , ( E ) S6 , ( F ) S7 and ( G ) S8-enriched transcripts ( red ) identified using the following criteria: S2–S5 enriched transcripts: p adj <1e-5; S6–S8-enriched transcripts: p adj <1e-20 . All stages: log2 ratio ≥2 . 322 ( five-fold upregulation ) , average scaled RPKM value ≥1 . 0 at indicated time point , transcript has ≥1 ORF . Transcripts derived from transposase or retroviruses were omitted , as were S2 , S3 and S4 transcripts also upregulated in yolk . S2–S8: Stages 2–8 . DOI: http://dx . doi . org/10 . 7554/eLife . 21052 . 01510 . 7554/eLife . 21052 . 016Figure 1—figure supplement 4 . Mean centered expression and average RPKM profiles for S2-enriched transcripts . Mean centered expression and average RPKM profiles for S2-enriched transcripts , organized by cluster membership presented in Figure 1C and Figure 1—source data 2 . S2–S8 Stages 2–8 . Y , yolk . DOI: http://dx . doi . org/10 . 7554/eLife . 21052 . 01610 . 7554/eLife . 21052 . 017Figure 1—figure supplement 5 . Mean centered expression and average RPKM profiles for S3-enriched transcripts . Mean centered expression and average RPKM profiles for S3-enriched transcripts , organized by cluster membership presented in Figure 1D and Figure 1—source data 3 . S2–S8 , Stages 2–8; Y , yolk . DOI: http://dx . doi . org/10 . 7554/eLife . 21052 . 01710 . 7554/eLife . 21052 . 018Figure 1—figure supplement 6 . Mean centered expression and average RPKM profiles for S4-enriched transcripts . Mean centered expression and average RPKM profiles for S4-enriched transcripts , organized by cluster membership presented in Figure 1E and Figure 1—source data 4 . S2–S8 , Stages 2–8; Y , yolk . DOI: http://dx . doi . org/10 . 7554/eLife . 21052 . 01810 . 7554/eLife . 21052 . 019Figure 1—figure supplement 7 . Mean centered expression and average RPKM profiles for S5-enriched transcripts . Mean centered expression and average RPKM profiles for S5-enriched transcripts , organized by cluster membership presented in Figure 1F and Figure 1—source data 5 . S2–S8 , Stages 2–8; Y , yolk . DOI: http://dx . doi . org/10 . 7554/eLife . 21052 . 01910 . 7554/eLife . 21052 . 020Figure 1—figure supplement 8 . Mean centered expression and average RPKM profiles for S6-enriched transcripts . Mean centered expression and average RPKM profiles for S6-enriched transcripts , organized by cluster membership presented in Figure 1G and Figure 1—source data 6 . S2–S8 , Stages 2–8; Y , yolk . DOI: http://dx . doi . org/10 . 7554/eLife . 21052 . 02010 . 7554/eLife . 21052 . 021Figure 1—figure supplement 9 . Mean centered expression and average RPKM profiles for S7-enriched transcripts . Mean centered expression and average RPKM profiles for S7-enriched transcripts , organized by cluster membership presented in Figure 1H and Figure 1—source data 7 . S2–S8: Stages 2–8; Y , yolk . DOI: http://dx . doi . org/10 . 7554/eLife . 21052 . 02110 . 7554/eLife . 21052 . 022Figure 1—figure supplement 10 . Mean centered expression and average RPKM profiles for S8-enriched transcripts . Mean centered expression and average RPKM profiles for S8-enriched transcripts , organized by cluster membership presented in Figure 1I and Figure 1—source data 8 . S2–S8 , Stages 2–8; Y , yolk . DOI: http://dx . doi . org/10 . 7554/eLife . 21052 . 02210 . 7554/eLife . 21052 . 023Figure 1—figure supplement 11 . Molecular markers for the primitive ectoderm . ( A ) Average RPKM values per embryo for primitive ectoderm markers gelsolin-like ( SMED30014940 , blue ) and spondin1-like ( SMED30032088 , red ) during embryogenesis . S2–S8 , Stages 2–8; Y , yolk . ( B–C ) The number of primitive ectoderm cells remained constant while embryo volume increased during S3–S4 . Primitive ectoderm cell nuclei were scored in SPIM reconstructed S3 and S4 embryos . ( B ) During S3 , the average number of primitive ectoderm nuclei per embryo was 21 . 5 ± 2 . 9 . During S4: , the average number of primitive ectoderm nuclei per embryo was 22 ± 1 . 4 . n = 5 embryos . Unpaired t-test , two tailed p value = 0 . 72 . ( C ) Average S3 embryo volume was 7 . 4 × 107 µm3 . Average S4 embryo volume was 1 . 5 × 108 µm3 . Unpaired t-test , two tailed p value = 0 . 01 . Embryo volumes were calculated by generating a masked surface in Imaris . Twenty-two S3 embryos and five S4 embryos were scored . Error bars represent the standard deviation of the mean . ( D–E ) gelsolin-like ( D ) and spondin1-like ( E ) expression ( blue ) during embryogenesis ( S2-S8 ) . Black arrowheads: temporary embryonic pharynx . Red arrowheads: definitive pharynx . Yellow arrows: primitive ectoderm cells . A , aboral hemisphere; O , oral hemisphere; V , ventral . Scale bars: 100 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 21052 . 02310 . 7554/eLife . 21052 . 024Figure 1—figure supplement 12 . Molecular markers for the temporary embryonic pharynx . ( A–B ) The temporary embryonic pharynx is innervated by PC-2+ neurons ( A ) and contains mhc-1+ radial muscle fibers ( B ) . For simplicity , only S4 is shown . The image shown in ( B ) is also shown in Figure 1—figure supplement 17C . ( C ) Average RPKM values per embryo for the temporary embryonic pharynx markers venom allergen-like ( VAL-like; SMED30015313 ) , macrophage expressed gene one like-1 ( MPEG1-like 1; SMED30000139 ) , MPEG1-like 2 ( SMED30034696 ) and netrin-like ( SMED30023593 ) during embryogenesis . S2–S8 , Stage 2–8; Y , yolk . ( D–G ) Expression of temporary embryonic pharynx specific markers VAL-like ( D , S3–S7 ) , netrin-like ( E , S2–S7 ) , MPEG1-like-1 ( F , S3–S4 ) and MPEG1-like-2 ( G , S3–S4 ) . ( A–B , D–G ) Anterior: top ( S6–S8 ) . Black arrowheads: temporary embryonic pharynx . Red arrowheads: definitive pharynx . A , aboral hemisphere; D , dorsal; O , oral hemisphere; V , ventral . Scale bars: 100 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 21052 . 02410 . 7554/eLife . 21052 . 025Figure 1—figure supplement 13 . Molecular markers for the developing gut . ( A , E , H ) Average RPKM values per embryo for ( A ) embryonic gut transcripts cathepsin L1-like ( CTSL-like: SMED30023322 ) , lysosomal alpha glucosidase-like ( LYAG-like: SMED30028442 , SMED30008977 ) , macrophage-expressed gene one like-3 ( MPEG1-like-3: SMED30015696 ) , ( E ) gamma ( γ ) class neoblast transcripts ( gata456a , hnf4 , prox-1 , nkx2 . 2 ) , and ( H ) transcripts with enriched expression in the adult gut ( Forsthoefel et al . , 2012; Vu et al . , 2015; Wurtzel et al . , 2015 ) . S2–S8 , Stage 2–8; Y , yolk . Early embryonic gut transcript expression was validated by WISH on staged embryo collections . ( H ) Adult gut-enriched transcripts with enriched expression during S5 and/or S6 ( top , n = 146 ) , or S7 and/or S8 ( bottom , n = 292 ) . Adult gut-enriched transcripts are flagged in Figure 1—source data 5–8 . 74% ( n = 1 , 112 ) of the intestinal phagocyte-enriched transcripts reported in ( Forsthoefel et al . , 2012 ) were identified in the smed20140614 transcriptome; 129 ( 11% ) of the cross-referenced transcripts were enriched during S5 , S6 , S7 and/or S8 . 90% ( n = 425 ) of the gut-enriched transcripts reported in ( Wurtzel et al . , 2015 ) were identified in the smed20140614 transcriptome; 44% ( n = 186 ) of the cross-referenced transcripts were enriched during S5 , S6 , S7 and/or S8 . ( B–D ) CTSL-like ( B ) , LYAG-like ( C ) and MPEG1-like-3 ( D ) expression ( blue ) in the temporary embryonic pharynx ( S2–S4 ) , four primitive gut cells abutting the temporary embryonic pharynx ( S4 ) , and yolk-laden gut cells forming an irregular lattice beneath the embryonic wall ( S5–S6 ) . Expression of these markers was downregulated as branching morphogenesis proceeded during S7 . ( F–G ) gata456a ( F ) and hnf4 ( G ) expression ( blue ) during embryogenesis , S2–S8 . Staining was detected in the presumptive temporary embryonic pharynx ( S2 ) , and was later detected in scattered parenchymal cells from S5 onwards . Expression of both markers became more prominent in the developing gut over time , especially after branching morphogenesis was underway during S7–S8 . ( I ) porcn-A expression ( blue ) during embryogenesis , S2–S8 . Hazy , faint expression was detected in the gut during S5–S6 , with increasing signal following the initiation of branching morphogenesis during S7–S8 . ( B–D , F–G , I ) Anterior: top ( S6–S8 ) . A , aboral hemisphere; D , dorsal; O , oral hemisphere; V , ventral . Black arrowheads: temporary embryonic pharynx . Black arrows: primitive gut cells . Red arrowheads: definitive pharynx . Scale bars: 100 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 21052 . 02510 . 7554/eLife . 21052 . 026Figure 1—figure supplement 14 . :Molecular markers for the definitive pharynx . ( A ) WISH developmental time course using foxA1 riboprobes ( blue ) , S3–S8 . foxA1 expression was consistently detected in the embryonic pharynx lumen during S3–S5 ( black arrowheads ) . Anterior: top ( S6–S8 ) . Black arrowheads: temporary embryonic pharynx . Red arrowheads: definitive pharynx . A , aboral hemisphere; D , dorsal; O , oral hemisphere; V , ventral . Scale bars: 100 µm . ( B ) Average RPKM values per embryo for the definitive pharynx markers foxA1 , meis , laminin and npp-1 during embryogenesis ( Adler et al . , 2014; Scimone et al . , 2014 ) . S2-S8 , Stage 2–8; Y , yolk . DOI: http://dx . doi . org/10 . 7554/eLife . 21052 . 02610 . 7554/eLife . 21052 . 027Figure 1—figure supplement 15 . Molecular markers for the definitive epidermis . ( A , C , F ) Average RPKM values per embryo for: ( A ) zeta ( ζ ) class neoblast transcripts ( C ) Category 2 and Category 3transcripts , and ( F ) Category 4 and 5 transcripts ( Eisenhoffer et al . , 2008; Pearson and Sánchez Alvarado , 2010; Tu et al . , 2015; van Wolfswinkel et al . , 2014; Wagner et al . , 2012; Zhu et al . , 2015 ) . S2–S8 , Stage 2–8; Y , yolk . Transcripts shown had enriched expression during S5 , S6 , S7 and/or S8 and are flagged in Figure 1—source data 5–8 . ( B , D–E , G–I ) WISH with riboprobes complementary to: ( B ) p53 , ( D ) prog-1 , ( E ) AGAT-1 , ( G ) zpuf-6 , ( H ) vim-3 and ( I ) crocc ( blue ) , S3–S8 . Anterior: top ( S6–S8 ) . D , dorsal; V , ventral . Red arrowheads: definitive pharynx . Red arrows: ciliated protonephridial tubules . Scale bars: 100 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 21052 . 02710 . 7554/eLife . 21052 . 028Figure 1—figure supplement 16 . Molecular markers for the developing nervous system . ( A , C ) Average RPKM values per embryo for validated and putative adult neural progenitor transcripts ( Cowles et al . , 2013; Currie and Pearson , 2013; Lapan and Reddien , 2012; März et al . , 2013; Monjo and Romero , 2015; Scimone et al . , 2014; Wenemoser et al . , 2012 ) ( A ) and adult neural classifier transcripts identified in single cell sequencing experiments ( Wurtzel et al . , 2015 ) ( C ) , Neural transcripts that showed enriched expression during S5 , S6 , S7 and/or S8 are flagged in Figure 1—source data 5–8 . 90% ( n = 533 ) of the neural-enriched transcripts reported by Wurtzel et al . ( 2015 ) were identified in the smed20140614 transcriptome; 60% ( n = 323 ) of the cross-referenced transcripts were enriched during S5 , S6 , S7 and/or S8 . ( B , D ) Expression of the neural progenitor marker pax6a ( B ) and the neural marker synaptotagmin ( syt , D ) ( blue ) , S2–S8 . Anterior: top ( S6–S8 ) . A , aboral hemisphere; D , dorsal; O , oral hemisphere; V , ventral . Black arrowheads: temporary embryonic pharynx . Red arrowheads: definitive pharynx . Cyan arrows: cephalic ganglia . Cyan arrowheads: ventral nerve cords . Scale bars: 100 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 21052 . 02810 . 7554/eLife . 21052 . 029Figure 1—figure supplement 17 . Molecular markers for the developing musculature . ( A , C ) Expression of the muscle progenitor marker myoD and the mature muscle marker mhc-1 , S2–S8 . Anterior: top ( S6–S8 ) . A , aboral hemisphere; D , dorsal; O , oral hemisphere; V , ventral . Black arrowheads: temporary embryonic pharynx . Red arrowheads: definitive pharynx . Cyan arrows: cephalic ganglia . Cyan arrowheads: ventral nerve cords . Scale bars: 100 µm . ( B , D ) Average RPKM values per embryo for the putative muscle progenitor marker myoD ( B ) and transcripts enriched in adult muscle ( D ) . S2–S8 , Stages 2–8; Y , yolk . Transcripts in ( D ) showed enriched expression during S5 , S6 , S7 and/or S8 ( n = 166 transcripts , or 42% of the muscle cell-enriched transcripts reported in Wurtzel et al . ( 2015 ) ) . Muscle-enriched transcripts are flagged in Figure 1—source data 5–8 . DOI: http://dx . doi . org/10 . 7554/eLife . 21052 . 02910 . 7554/eLife . 21052 . 030Figure 1—figure supplement 18 . Molecular markers for the developing excretory system . ( A , C ) Average RPKM per embryo for transcripts expressed in protonephridia progenitors ( pou2/3 , six1/2–2 , sal1 , eya , osr ) ( Scimone et al . , 2011 ) ( A ) or differentiated protonephridia ( Rink et al . , 2011; Scimone et al . , 2011; Wurtzel et al . , 2015 ) ( C ) . Transcripts shown were enriched during S5 , S6 , S7 and/or S8 and are flagged in Figure 1—source data 5–8 . ( B , D ) WISH developmental time course with riboprobes complementary to the protonephridial progenitor and tubule cell marker pou2/3 ( B ) or the non-ciliated tubule marker CAVII-1 ( D ) ( blue ) , S2–S8 . Anterior: top ( S6–S8 ) . D , dorsal; L , lateral . Red arrowheads definitive pharynx . Scale bars: 100 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 21052 . 03010 . 7554/eLife . 21052 . 031Figure 1—figure supplement 19 . Molecular markers for the developing eyes . ( A , C ) Average RPKM per embryo for transcripts required for eye progenitor specification ( ovo , six-1/2 and eya [purple] ) , photoreceptor neuron differentiation ( otxA [ ( red] ) , or pigment cup differentiation ( sp6-9 and dlx [ ( blue] ) ( A ) , or with enriched expression in adult eye tissue ( Lapan and Reddien , 2012 ) ( C ) . S2–S8 , Stages 2–8; Y , yolk . Transcripts shown were enriched during S5 , S6 , S7 and/or S8 , and are flagged in Figure 1—source data 5–8 . ( B , D–E ) WISH developmental time course with riboprobes complementary to ovo ( B ) , opsin ( D ) , and tyrosinase ( E ) , S5–S8 . Anterior: top ( S6–S8 ) . D: dorsal . Purple arrowheads: developing eye tissue . Blue arrowheads: trail cells ( eye progenitors ) . Scale bars: 100 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 21052 . 031 In contrast to the synchronous , oriented blastomere cleavage patterns of Spiralian embryos ( Lambert , 2010 ) , blastomeres in freshwater planarian embryos undergo dispersed cleavage among yolk cells: they divide asynchronously and are not in direct contact with one another ( Bardeen , 1902; Cardona et al . , 2005; Hallez , 1887; Ijima , 1884; Le Moigne , 1963; Metschnikoff , 1883; Vara et al . , 2008 ) . During sphere formation , some blastomeres differentiate into temporary embryonic cell types that provide form and function to the embryo , including the primitive ectoderm , temporary embryonic pharynx and primitive gut ( Cardona et al . , 2005; Hallez , 1887; Le Moigne , 1963; Sánchez Alvarado , 2003 ) . Temporary embryonic tissues are not thought to contribute to the juvenile body plan; they are thought to degenerate as the definitive organs form and morphogenesis proceeds ( Cardona et al . , 2005; Le Moigne , 1963; Vara et al . , 2008 ) . In contrast , undifferentiated blastomeres remaining after sphere formation are thought to give rise to all definitive tissues found in juvenile worms ( Hallez , 1887; Hyman , 1951; Le Moigne , 1963; Sánchez Alvarado , 2003; Stevens , 1904 ) . Several attributes of ectolecithal development challenge efforts to accurately stage live embryos including internal fertilization , dispersed cleavage , lack of morphological landmarks , and inherent variability in embryo size and developmental timing . Extant staging series for Polycelis nigra ( Le Moigne , 1963 ) , Schmidtea polychroa ( Cardona et al . , 2005; Martín-Durán et al . , 2010 ) and Girardia tigrina ( Vara et al . , 2008 ) rely upon gross morphological criteria gleaned from live animals and fixed , histologically or antibody-stained specimens . As this is the first systematic characterization of Smed embryogenesis , we established an accurate and objective staging method based on unique gene expression signatures , cohorts of enriched transcripts identified through single embryo RNA-Seq , associated with chronological age and embryo morphology ( Figure 1B–I , Figure 1—source data 1 ) . When appropriate , efforts were made to integrate and adapt extant staging schema from other planarian species . Smed embryos gestate for approximately two weeks at 20°C prior to hatching . We generated total RNA replicates from single Smed embryos for seven chronologically and/or morphologically distinct stages ( S ) , S2-–S8 ( Figure 1B–I ) ; S1 samples ( zygotes , Figure 1—figure supplement 1 ) were not queried by RNA-SSeq . Yolk ( Y ) replicates were prepared from egg capsules lacking developing embryos at 8 days post capsule deposition . In addition , single animal replicates were prepared from C4 and virgin sexually mature adults ( SX ) . RNA-Seq libraries were analyzed for four biological replicates ( i . e . , four individuals ) per stage ( Materials and methods ) . Identification of appropriate reference ( s ) and normalization methods was challenging due to vast differences in sample composition and complexity among different stages . However , clustering of replicates and discrete separation of stages was seen in a multidimensional scaling plot ( Figure 1J ) . Replicates for a given stage generally showed strong correlation among themselves despite not having controlled for differences in genetic background or embryo size ( Figure 1K ) . Notably the S2 and S3 replicates , generated from embryos undergoing sphere formation and nascent spheres , respectively , were intermingled in expression space , showed the greatest variability and few significant differences in gene expression ( Figure 1J–K , Figure 1—source data 1 , Figure 1—figure supplement 2 ) . Discrete stages were retained due to apparent differences in embryo morphology ( Figure 1C–D ) . Similarities among S2 and S3 samples may be due to maternal RNA contribution or to difficulty detecting labile S2 embryos , such that only well-developed S2 embryos were prospected by RNA-Seq . Two approaches were used to identify differentially expressed transcripts: pairwise comparisons of adjacent stages ( Figure 1—source data 1 , Figure 1—figure supplement 2 ) and comparisons of each stage relative to a mixed stage reference generated by averaging the read counts for the remaining replicates ( Y , S2–S8 ) ( Materials and methods , Figure 1—source data 1 , Figure 1—figure supplement 3 ) . The goal of the pairwise comparisons was to identify transcripts with the starkest changes in expression in either direction , without constraints on transcript behavior at other points during embryogenesis . In contrast , the mixed stage reference comparisons maximized the likelihood of identifying transcripts exhibiting stage-specific expression , and only upregulated transcripts were analyzed . Stringent criteria were applied in both scenarios for flagging differentially expressed transcripts , including thresholds based on the Benjamini-Hochberg adjusted p-value , fold-change , normalized RPKM expression level for time points , and identification of at least one open reading frame in the transcript ( Figure 1—source data 1 ) . More upregulated transcripts were identified in the mixed stage reference than in pairwise comparisons , perhaps due to increased sequencing depth of the averaged reference samples , which may enable more reliable detection of lowly expressed transcripts . Furthermore , identification of S2-enriched transcripts suggested that the whole embryo sequencing approach was sensitive enough to detect transcripts expressed in rare cell populations ( i . e . , blastomeres and differentiated cells making up the embryo proper ) . Non-redundant lists of upregulated transcripts , resulting from the union of the pairwise and mixed reference comparisons , served as molecular fingerprints for each time point for downstream analyses , including hierarchical clustering and GO analysis ( Figure 1—source data 1 ) . The molecular staging resource ( Figure 1—source data 1 ) incorporates representative images ( Figure 1C–I , Figure 1—figure supplement 1 ) , gene expression signatures ( Figure 1—figure supplements 4–10 , Figure 1—source data 2–8 ) and written summaries of key developmental events ( Figure 1—source data 1 ) , defining S1–S8 . An expression atlas describing temporary embryonic tissue types and development of the definitive organ systems is also provided ( Figure 1—source data 9 , Figure 1—figure supplements 11–19 ) . The molecular staging resource and expression atlas are also available and searchable online ( https://planosphere . stowers . org ) . The molecular staging series identified S2- and S3-enriched transcripts ( Figure 1C–D , Figure 1—source data 2–3 ) , including elongation factor 1a-like-1 ( EF1a-like-1 ) , expressed in all known embryonic cell populations in nascent spherical embryos ( Figure 2A–B ) . During S2 , primitive ectoderm cells are the first to differentiate ( Hallez , 1887; Ijima , 1884; Le Moigne , 1963; Metschnikoff , 1883 ) . EF1a-like-1 was expressed in primitive ectoderm cells ( Figure 2A–B , Video 1 ) , which flatten , elaborate numerous cytoplasmic processes , and form a single-cell layer bounding the sphere ( Figure 1—source data 9 , Figure 1—figure supplement 11 ) . EF1a-like-1 expression was also detected in the temporary embryonic pharynx ( Figure 2A–B , Video 1 ) an innervated pump containing neurons , radial muscle and associated epithelial cells ( Figure 1—source data 9 , Figure 1—figure supplement 12 ) , and in the primitive endoderm ( Figure 2A ) , which consists of an inner gut cavity and phagocytic cells associated with the temporary embryonic pharynx ( Figure 1—source data 9 , Figure 1—figure supplement 13 ) . A population of undifferentiated blastomeres and yolk cells remain in the embryonic wall , the parenchymal space between the primitive ectoderm and endoderm , in nascent S3 spheres ( Figure 1D ) ( Hyman , 1951; Sánchez Alvarado , 2003 ) . EF1a-like-1 was expressed in undifferentiated blastomeres , but not yolk cells , in S3 embryos ( Figure 2A–B , Video 1 ) . 10 . 7554/eLife . 21052 . 032Figure 2 . Blastomere anarchy drives Smed embryogenesis . ( A–B ) Architectural features of S2 and S3 embryos . ( A ) Expression of the pan embryonic cell marker EF1a-like-1 ( blue ) in S2 ( left ) and S3 ( right ) embryos . Cyan arrowheads: primitive ectoderm cells . Cyan arrows: undifferentiated blastomeres . Red arrowhead: temporary embryonic pharynx . Red arrows: primitive gut cells . ( B ) S3 embryo stained with EF1a-like-1 riboprobes ( red ) and sytox green nuclear counterstain ( green ) . Cyan arrowheads: primitive ectoderm cells . Cyan arrows: undifferentiated blastomeres . Yellow arrowhead: temporary embryonic pharynx . ( C ) Confocal Z-slice of an ovary from a sexually mature Smed hermaphrodite stained with piwi-1 riboprobes ( green ) and DAPI ( blue ) . Yellow arrows: oocytes . ( D ) Dispersed cleavage . S2 embryo stained with piwi-1 riboprobes ( red , blastomeres ) and antibodies raised against the mitotic epitope H3S10p ( green ) . Nuclei stained with DAPI ( blue ) . Yellow arrow: dividing blastomere . ( E ) piwi-1 is expressed in undifferentiated blastomeres of S3 embryos . S3 embryo costained with riboprobes complementary to piwi-1 ( red ) and EF1a-like-1 ( green ) . 100% piwi-1+ blastomeres coexpressed EF1a-like-1 . n = 159 cells scored , n = 5 S3 embryos . Cyan arrows: undifferentiated blastomeres . Yellow arrowhead: temporary embryonic pharynx . Red arrows: fiduciary beads used for SPIM reconstruction . ( F ) piwi-1+ cells are located in the embryonic wall . Paraffin-embedded cross-section of a S3 sphere stained with piwi-1 riboprobes ( blue ) and eosin ( pink ) . Cyan arrows: piwi-1+ cells . GC: yolk-filled gut cavity . Inset: magnified view of a piwi-1+ cell . Scale: 25 µm . ( G ) Left: Average RPKM per embryo for piwi-1 ( S2–S8 ) . Right: WISH developmental time course with piwi-1 riboprobes ( blue ) ( S2–S8 ) . O , oral hemisphere; V , ventral . ( A–G ) Scale: 100 µm . Left: Observed distribution of piwi-1+ cells in S3–S4 embryos ( blue bars ) relative to the oral-aboral axis ( 0–3 . 14 radians ) . Maximum likelihood analysis best described distribution by the function ( ( 1-exp ( -θ/θ’ ) ) *sin ( θ ) , blue line ) . The optimal calculated θ’ was 0 . 45 ± 0 . 045 radians , based on simulations with comparably sized data sets , and was several orders of magnitude more likely to explain the observed distribution than the theoretical normal distribution , sin ( θ ) , ( θ’ = 0 ) , red line . S3: n = 32 embryos , n = 1 , 746 piwi-1+ cells scored . S4: n = 8 embryos , 2 , 665 piwi-1+ cells scored . Right: observed piwi-1+ cell distributions for individual S3 ( top ) and S4 ( bottom ) embryos . ( C–G ) piwi-1+ cells are detected throughout embryogenesis . ( H ) piwi-1+ cell positions are not stereotyped in S3–S4 embryos . DOI: http://dx . doi . org/10 . 7554/eLife . 21052 . 03210 . 7554/eLife . 21052 . 033Video 1 . S3 embryo architecture . SPIM reconstructed S3 embryo costained with EF1a-like-1 ( red ) and sytox green nuclear counterstain . EF1a-like-1 is a pan-embryonic cell marker that stains primitive ectoderm cells , the temporary embryonic pharynx and undifferentiated blastomeres in the embryonic wall . EF1a-like-1 staining is absent from yolk cells in the embryonic wall and gut cavity . DOI: http://dx . doi . org/10 . 7554/eLife . 21052 . 033 piwi-1+ cells were present throughout embryogenesis . piwi-1 expression was detected in oocytes ( Figure 2C ) , suggesting that zygotes contain piwi-1 mRNA . Zygote-derived blastomeres undergoing dispersed cleavage among yolk cells during S2 also expressed piwi-1 ( Figure 2D ) . Costaining with piwi-1 riboprobes and antibodies raised against the G2-M phase mitotic marker H3S10p showed that piwi-1+ blastomeres divide asynchronously during S2 ( Figure 2D ) . As spheres form during S2 , some blastomeres downregulate piwi-1 expression and differentiate into temporary embryonic cell types . piwi-1 expression was restricted to , and expressed throughout , the undifferentiated blastomere population in S3 embryos , as demonstrated by double fluorescent WISH with EF1a-like-1 and piwi-1 riboprobes ( Figure 2E , Video 2 ) . piwi-1+ blastomeres were always located in the embryonic wall ( Figure 2F ) . piwi-1 expression was never detected in the primitive ectoderm , temporary embryonic pharynx or primitive gut ( Figure 2E , G ) . During S3–S5 , piwi-1+ cell number clearly increased , effectively blanketing the sphere ( Figure 2G ) . As definitive gut development proceeded during S6–S8 , piwi-1+ cells occupied the parenchyma between the developing gut branches ( Figure 2G ) . Notably , piwi-1+ cells were not detected in the definitive pharynx , and the compartment receded from the anterior margin as head structures developed ( Figure 2G ) . By S8 , the spatial distribution of piwi-1+ cells was indistinguishable from that of the adult neoblast compartment ( Figure 2G ) . 10 . 7554/eLife . 21052 . 034Video 2 . piwi-1 is expressed in all undifferentiated blastomeres of S3 embryos . SPIM reconstructed S3 embryo costained with piwi-1 ( red ) and EF1a-like-1 ( green ) . piwi-1 is expressed in all undifferentiated blastomeres in the embryonic wall ( piwi-1+ , EF1a-like-1+ cells ) . piwi-1 is not expressed in differentiated tissues marked by EF1a-like-1 alone , including the primitive ectoderm and temporary embryonic pharynx ( green ) . Several fluorescent beads used for three-dimensional reconstruction are visible ( red ) . DOI: http://dx . doi . org/10 . 7554/eLife . 21052 . 034 Analysis of piwi-1+ cell positions in S3–S4 embryos revealed that undifferentiated blastomeres were spatially disordered , or ‘anarchic , ’ as described for other freshwater flatworms ( Bardeen , 1902; Cardona et al . , 2005; Hallez , 1887; Ijima , 1884; Le Moigne , 1963; Metschnikoff , 1883; Vara et al . , 2008 ) . The observed distribution of piwi-1+ cells in S3–S4 embryos with respect to the temporary embryonic pharynx ( Figure 2H , blue line ) was nearly identical to that of a random distribution ( Figure 2H , red line ) . Fewer piwi-1+ cells were located adjacent to the oral pole than was predicted for a theoretical normal distribution , perhaps due to spatial constraints imposed by the temporary embryonic pharynx and associated primitive gut cells ( Figure 2H ) . We used a simple dampening term and maximum likelihood estimation to account for the deviation between the observed and theoretical normal distributions ( Figure 2H ) . However , the distribution of piwi-1+ cells in the embryonic wall was not stereotyped; it varied greatly across individuals ( Figure 2H ) . Strikingly , cell cycle activity was restricted to piwi-1+ cells at all developmental stages assayed , and all piwi-1+ cells in early embryos were cycling . Expression of the cell cycle regulators PCNA and RRM2-2 closely mimicked that of piwi-1 during embryogenesis , both with respect to the spatial distribution of positive cells and trends observed in the RNA-Seq data ( Figures 2G and 3A–B ) . Double fluorescent WISH on S3–S5 embryos revealed that PCNA and RRM2-2 were expressed exclusively in piwi-1+ blastomeres , all of which were cycling ( Figure 3C–D , Videos 3–4 ) . S3–S5 embryos costained with piwi-1 and H3S10p antibodies confirmed that mitotic activity was restricted to piwi-1+ blastomeres ( Figure 3E , Video 5 ) . Consistent with observations made in S2 embryos ( Figure 2D ) , cell divisions were asynchronous during S3–S5 . The mitotic index for S3–S5 piwi-1+ blastomeres was stable , with no statistically significant difference in the calculated division rate ( Figure 3F ) . Analysis of mitotic ( piwi-1+ , H3S10p+ ) cell positions along the oral-aboral axis in S3–S4 embryos did not reveal regional biases in mitotic activity across samples ( Figure 3G ) . 10 . 7554/eLife . 21052 . 035Figure 3 . Cell cycle activity is restricted to the piwi-1+ compartment . ( A–B ) Left: Colorimetric WISH depicting expression of PCNA ( A ) or RRM2-2 ( B ) during stages S2–S8 . Right: Average RPKM values per embryo for PCNA ( A ) or RRM2-2 ( B ) in Y ( yolk ) and S2–S8 . V , ventral . Scale: 100 µm . ( C–D ) S3 ( top ) , S4 ( middle ) and S5 ( bottom ) embryos costained with piwi-1 ( red ) and PCNA ( green [C] ) or RRM2-2 ( green [D] ) riboprobes . The percentage of piwi-1+ cells coexpressing the indicated cell cycle marker ( red ) and the percentage of PCNA+ or RRM2-2+ cells coexpressing piwi-1 ( green ) appear in the lower left corner of merged images . Scale bars: 100 µm . ( C ) S3: n = 273 cells , n = 6 embryos . S4: n = 1 , 267 cells , n = 4 embryos . S5: n = 734 cells , n = 3 embryos . ( D ) S3: n = 130 cells , n = 4 embryos . S4: n = 1 , 295 cells , n = 5 embryos . S5: n = 350 cells , n = 3 embryos . ( E ) Mitotic activity is restricted to the piwi-1+ cell compartment in S3–S5 embryos . Left: S4 embryo costained with piwi-1 and the embryonic pharynx marker LYAG-like ( both in green ) and antibodies against the mitotic epitope H3S10p ( red ) . White arrows: dividing blastomeres . White arrowhead: temporary embryonic pharynx . Scale bar: 100 µm . Right: Bar graph depicting the percentage of mitotic cells scored that expressed piwi-1 in S3–S5 embryos . ( F ) The mitotic index for the piwi-1+ cell compartment did not vary significantly during S3–S5 . Average percentage of piwi-1+ cells in mitosis during S3–S5 . Error bars represent the standard deviation of the mean . Observed distribution of mitotic ( piwi-1+ , H3S10p+ ) cells in S3-–S4 embryos ( blue bars ) along the oral-aboral axis ( 0–3 . 14 radians ) . Using the function derived with maximum likelihood estimation for the piwi-1+ cell distribution , ( 1-exp ( -θ/θ’ ) ) *sin ( θ ) ( blue line ) , and simulations using equivalent sample sizes , the optimal θ’ was calculated to be 0 . 58 ± 0 . 33 , and was 50-fold more likely to explain the observed trend than a simple normal distribution , sin ( θ ) , where θ’=0 ( red line ) . S3: n = 82 mitotic cells , n = 18 embryos . S4: n = 110 mitotic cells , n = 8 embryos . ( G ) Mitotic cell positions are not stereotyped in early embryos . DOI: http://dx . doi . org/10 . 7554/eLife . 21052 . 03510 . 7554/eLife . 21052 . 036Video 3 . Cell cycle activity is restricted to piwi-1 blastomeres , and all blastomeres are cycling . SPIM reconstructed S4 embryo costained with piwi-1 ( red ) and PCNA ( green ) . PCNA expression is restricted to piwi-1+ blastomeres , and all piwi-1+ cells co-express PCNA . DOI: http://dx . doi . org/10 . 7554/eLife . 21052 . 03610 . 7554/eLife . 21052 . 037Video 4 . Cell cycle activity is restricted to piwi-1 blastomeres , and all blastomeres are cycling . SPIM reconstructed S4 embryo costained with piwi-1 ( red ) and RRM2-2 ( green ) . RRM2-2 expression is restricted to piwi-1+ blastomeres , and all piwi-1+ cells co-express RRM2-2 . Several fluorescent beads used for three-dimensional reconstruction are visible ( red ) . DOI: http://dx . doi . org/10 . 7554/eLife . 21052 . 03710 . 7554/eLife . 21052 . 038Video 5 . Mitotic activity is restricted to piwi-1+ blastomeres , which cycle asynchronously . SPIM reconstructed S4 embryo costained with piwi-1 and LYAG-like ( both in green ) and H3S10p antibodies ( red ) . LYAG-like marks the temporary embryonic pharynx and is not expressed in piwi-1+ blastomeres . Several examples of piwi-1+ , H3S10p+ cells are evident . DOI: http://dx . doi . org/10 . 7554/eLife . 21052 . 038 Several striking parallels may be drawn regarding cell cycle behavior of the piwi-1+ population during embryogenesis and adulthood . First , cell cycle activity is largely restricted to this compartment in both contexts , with exception of the male and female germline in sexually mature adults ( Baguñà , 1976; Newmark and Sánchez Alvarado , 2000; Reddien et al . , 2005; Wagner et al . , 2011; Wang et al . , 2007 ) . Second , cycling cells and mitotic figures do not display obvious positional biases within the parenchyma in either S3–S5 embryos or adults during homeostasis ( Newmark and Sánchez Alvarado , 2000 ) . Little substantiating evidence exists in support of a quiescent piwi-1+ cell population during embryogenesis or adulthood ( Newmark and Sánchez Alvarado , 2000 ) , and differences in cell cycle length were not observed between neoblast subclasses in adult asexual animals ( van Wolfswinkel et al . , 2014 ) . Finally , the staggering net increase in piwi-1+ blastomeres during S3–S5 suggests a capacity for self-renewal , a property possessed by neoblasts . Numerous neoblast-enriched transcripts , identified through whole asexual animal irradiation studies and cell sorting ( Eisenhoffer et al . , 2008; Labbé et al . , 2012; Rossi et al . , 2007; Solana et al . , 2012; Wagner et al . , 2012; Wurtzel et al . , 2015 ) , have been vetted for co-expression with piwi-1+ and ascribed function ( s ) in neoblast proliferation , maintenance or cell fate commitment . Some neoblast-enriched transcripts , including those encoding nuage components and cell cycle regulators , are expressed in all neoblasts , whereas others are predominantly expressed in subpopulation ( s ) of cells that may be primed to adopt differentiated fates ( van Wolfswinkel et al . , 2014 ) . To address similarities in the gene expression profiles of the piwi-1+ population during embryogenesis and adulthood , the expression trends of 242 adult asexual neoblast-enriched transcripts were examined using the molecular staging series data . Neoblast-enriched transcript membership was determined by sequences downregulated in whole animals following lethal irradiation across three independent experiments ( Duncan et al . , 2015; Wagner et al . , 2012 ) ( Chen and Sánchez Alvarado , personal communication ) . Strikingly , most adult asexual neoblast-enriched transcripts were expressed throughout embryogenesis ( Figure 4A , Figure 4—source data 1 ) . 74% ( n = 180 ) of the neoblast-enriched transcripts had average RPKM values per embryo ≥1 . 0 in S2 embryos and 52% ( n = 128 ) transcripts had five-fold or greater expression levels in S2 embryos versus yolk , raising the possibility that other adult stem cell genes were expressed in blastomeres . Consistent with this idea , 41% of the adult asexual neoblast-enriched transcripts ( n = 99 ) were present in the molecular expression signature ( s ) for S2–S5 embryos ( Figure 1—source data 2–5 , Figure 4—source data 1 ) . Expression of neoblast-enriched transcripts usually peaked during S4 or S5 , prior to construction of the definitive organ systems , and diminished thereafter ( Figure 4A , Figure 4—source data 1 ) . The apparent decrease in expression after S5 was likely attributable to drastic changes in the complexity of the single embryo RNA samples during organogenesis , and was similarly observed for piwi-1 and the cell cycle regulators PCNA and RRM2-2 ( Figures 2G and 3A–B , Figure 4—source data 1 ) . 10 . 7554/eLife . 21052 . 039Figure 4 . Many adult neoblast markers are similarly expressed throughout the piwi-1+ compartment during embryogenesis . ( A ) Many transcripts with adult asexual neoblast-enriched expression are expressed throughout embryogenesis . Hierarchical clustering of 242 adult asexual neoblast-enriched transcripts during embryonic development using normalized mixed stage reference comparison data . Left: Heat map . Colored bars ( left ) denote clusters . Right: Normalized average RPKM values per embryo , plotted as a function of developmental time , for Clusters 1–8 . Y , yolk; S2–S8 , Stages 2–8 . ( B–E ) Colorimetric WISH depicting expression of piwi-2 ( B ) , piwi-3 ( C ) , tud-1 ( D ) and bruli-1 ( E ) during embryogenesis ( blue ) ( S2–S8 ) . V , ventral . Black arrowheads: temporary embryonic pharynx . Red arrowheads: definitive pharynx . Scale bars: 100 µm . ( F–I ) Many markers of the adult asexual neoblast compartment are also expressed in piwi-1+ blastomeres . Fluorescent WISH on S4 embryos with riboprobes against piwi-1 ( red ) and piwi-2 ( F ) , piwi-3 ( G ) , tud-1 ( H ) or bruli-1 ( I ) ( green ) . Percentage of piwi-1+ cells coexpressing the indicated marker ( red ) and the percentage of the indicated adult asexual neoblast marker coexpressing piwi-1 ( green ) appears in the lower left corner of merged images . Scale bars: 100 µm . ( F ) n = 435 cells , n = 9 S3–S4 embryos . ( G ) n = 535 cells , n = 5 S3–S4 embryos . ( H ) n = 1 , 867 cells , n = 8 S3–S5 embryos . ( I ) n = 1 , 353 cells , n = 3 S4 embryos . DOI: http://dx . doi . org/10 . 7554/eLife . 21052 . 03910 . 7554/eLife . 21052 . 040Figure 4—source data 1 . Hierarchical clustering analysis for 242 adult asexual neoblast-enriched transcripts . Clustering was performed using normalized expression data from the mixed stage reference comparison . Normalized expression profiles , cluster membership , average RPKM values across embryogenesis ( Y–S8 ) and for C4 and SX adults , as well as best BLASTx hits ( E < 0 . 001 ) versus the NR , Swiss-Prot , C . elegans , D . melanogaster , D . rerio , X . tropicalis , M . musculus and H . sapiens RefSeq databases are provided . DOI: http://dx . doi . org/10 . 7554/eLife . 21052 . 04010 . 7554/eLife . 21052 . 041Figure 4—figure supplement 1 . Adult asexual neoblast-enriched markers are coexpressed in piwi-1+ cells during embryogenesis . ( A , C ) Colorimetric WISH depicting expression of SoxP-1 ( A ) and SoxP-2 ( C ) during embryogenesis ( S2–S8 ) . V , ventral . Scale bars: 100 µm . ( B ) Average RPKM per embryo for SoxP-1 ( blue ) and SoxP-2 ( red ) . Y , yolk; S2–S8 , Stages S2–S8 . ( D–G ) Fluorescent WISH costaining on S3–S5 embryos using riboprobes against piwi-1 ( red ) and piwi-2 ( D ) , piwi-3 ( E ) , tud-1 ( F ) , or Y12 antibodies ( S4–S5 only ) ( G ) ( green ) . Scale bars: 100 µm . ( D–E ) S3 and S5 costained samples . S4 samples and quantification are depicted in the Figure 4F–I legend . ( G ) 98% piwi-1+ cells costained with Y12 antibodies . 96% Y12-positive cells costained with piwi-1 , n = 913 cells , n = 7 S4–S5 embryos . DOI: http://dx . doi . org/10 . 7554/eLife . 21052 . 041 Hierarchical clustering of the 242 adult asexual neoblast-enriched transcripts revealed correlated expression of genes associated with DNA replication ( e . g . , the replication licensing factors MCM2 and MCM5 ) , DNA repair ( e . g . , fancd2-like , msh2 and msh6 ) and cell cycle progression ( e . g . , cyclin D-like , cyclin-B1 and cyclin-B2 ) during embryogenesis , and the expression trends for these genes mimicked those of PCNA and piwi-1 ( Figure 4A [Cluster 1] , Figure 4—source data 1 ) . Notably , transcripts encoding the nuage-associated factors piwi-2 and piwi-3 , the RNA-binding protein bruli-1 , and the transcription factors SoxP-1 and junL1-1 , all genes previously implicated in neoblast maintenance or function ( Guo et al . , 2006; Palakodeti et al . , 2008; Reddien et al . , 2005; Wagner et al . , 2012 ) , were also coregulated with piwi-1 during embryogenesis ( Figure 4A [Cluster 3] , Figure 4—source data 1 ) . Consistent with the expression trends detected by RNA-Seq and the previously reported expression pattern for Schmidtea polychroa ( Spol ) tud-1 ( Solana et al . , 2009 ) , piwi-2 , piwi-3 , tud-1 , bruli-1 , Sox-P1 and Sox-P2 were expressed in cells with similar morphology and distribution to the piwi-1+ population during embryogenesis ( Figure 4B–E , Figure 4—figure supplement 1A–C ) . Double fluorescent WISH revealed coincident expression of piwi-1 and the adult stem cell markers piwi-2 , piwi-3 , tud-1 and bruli-1 in S3–S5 embryos ( Figure 4F–I , Figure 4—figure supplement 1D–F , Videos 6–9 ) . Y12 antibodies , which label chromatoid bodies in adult neoblasts ( Rouhana et al . , 2012 ) , stained piwi-1+ blastomeres during S4-–S5 ( Figure 4—figure supplement 1G , Video 10 ) . Taken together , these findings suggest that many adult asexual neoblast markers , including genes implicated in DNA replication and repair , cell cycle control , chromatin remodeling and/or modification , genome surveillance and pluripotency , are also likely to be expressed throughout the piwi-1+ population during embryogenesis . Moreover , shared elements of the blastomere and neoblast expression signatures , including DNA replication and repair pathway , cell cycle , nuage and RNA processing genes , are prominent features of an evolutionarily conserved gene expression signature for metazoan primordial stem cells ( Alié et al . , 2015 ) . 10 . 7554/eLife . 21052 . 042Video 6 . piwi-1+ blastomeres co-express the adult asexual neoblast-enriched gene piwi-2 . SPIM reconstructed S4 embryo costained with piwi-1 ( red ) and piwi-2 ( green ) . piwi-1+ blastomeres co-express the nuage factor piwi-2 , and virtually all piwi-2+ cells co-express piwi-1 . Several fluorescent beads used for three-dimensional reconstruction are visible ( green ) . DOI: http://dx . doi . org/10 . 7554/eLife . 21052 . 04210 . 7554/eLife . 21052 . 043Video 7 . piwi-1+ blastomeres co-express the adult asexual neoblast-enriched gene piwi-3 . SPIM reconstructed S4 embryo costained with piwi-1 ( red ) and piwi-3 ( green ) . piwi-1+ blastomeres co-express the nuage factor piwi-3 , and virtually all piwi-3+ cells co-express piwi-1 . DOI: http://dx . doi . org/10 . 7554/eLife . 21052 . 04310 . 7554/eLife . 21052 . 044Video 8 . piwi-1+ blastomeres co-express the adult asexual neoblast-enriched gene tud-1 . SPIM reconstructed S4 embryo costained with piwi-1 ( red ) and tud-1 ( green ) . piwi-1+ blastomeres co-express the nuage factor tud-1 , and virtually all tud-1+ cells co-express piwi-1 . DOI: http://dx . doi . org/10 . 7554/eLife . 21052 . 04410 . 7554/eLife . 21052 . 045Video 9 . piwi-1+ blastomeres co-express the adult asexual neoblast-enriched gene bruli-1 . SPIM reconstructed S4 embryo costained with piwi-1 ( red ) and bruli-1 ( green ) . piwi-1+ blastomeres co-express the stem cell maintenance gene bruli-1 , and virtually all bruli-1+ cells co-express piwi-1 . Several fluorescent beads used for three-dimensional reconstruction are visible ( red ) . DOI: http://dx . doi . org/10 . 7554/eLife . 21052 . 04510 . 7554/eLife . 21052 . 046Video 10 . piwi-1+ blastomeres possess chromatoid bodies . SPIM reconstructed S4 embryo costained with piwi-1 ( red ) and Y12 antibodies ( green ) . Y12 antibody staining was restricted to , and present throughout , the piwi-1+ blastomere population . DOI: http://dx . doi . org/10 . 7554/eLife . 21052 . 046 Hierarchical clustering of S2-–S4-enriched transcripts using scaled RPKM values identified 1 , 048 sequences in Clusters 5 , 6 and 8 that were downregulated by S5 and remained lowly expressed through S8; these sequences are referred to as early-embryo-enriched ( EEE ) transcripts ( Figure 5A , Figure 5—source data 1 ) . EEE transcripts were likely expressed in blastomeres and/or temporary embryonic tissues , as 98% of the sequences had average expression values at least five-fold greater in S2 embryos than in Y ( Figure 5—source data 1 ) . Most EEE transcripts were expressed at low levels in intact adults regardless of biotype: average RPKM values less than 1 . 0 were recorded for 65% and 59% of the EEE transcripts in C4 or SX , respectively ( Figure 5—source data 1 ) . 10 . 7554/eLife . 21052 . 047Figure 5 . Early-embryo-enriched transcripts are downregulated as organogenesis begins . ( A ) Hierarchical clustering of S2–S4-enriched transcripts ( n = 1 , 756 ) using scaled RPKM data . Left: Heat map . Y , yolk . Colored bars ( left ) denote Clusters 5 , 6 and 8 containing early-embryo-enriched ( EEE ) transcripts . Cluster f5 sequences ( blue , n = 413 ) were expressed at roughly equivalent levels during S2 and S3 , with 66% ( n = 275 ) transcripts showing five-fold or greater declines in average expression values between S3 and S5 . Cluster 6 sequences ( red , n = 523 ) exhibited maximal expression during S2 , and average expression levels declined more than five-fold between S2 and S4 for 81% ( n = 426 ) of these transcripts . Cluster 8 sequences ( green , n = 112 ) showed peak expression during S4 , with 52% ( n = 60 ) of the transcripts showing five-fold or greater declines in average expression values by S5 . Right: Normalized expression trends for EEE transcripts in Clusters 5 ( blue ) , 6 ( red ) and 8 ( green ) plotted as a function of developmental time . Median 50% of transcripts based on expression maxima are plotted . ( B–E ) Colorimetric WISH depicting expression of the EEE transcripts tct-like ( B ) , BTF3-like ( C ) , DDX5-like ( D ) and eIF4a-like ( E ) ( blue ) in S2–S8 embryos and C4 asexual adults . Black arrowheads: temporary embryonic pharynx . Red arrowheads: definitive pharynx . O , oral; V , ventral . Scale bars: 100 µm . ( F–I ) EEE transcripts were expressed throughout the piwi-1+ compartment in S3–S4 embryos . Fluorescent double WISH with riboprobes against piwi-1 ( red ) and the EEE transcripts tct-like ( F ) , BTF3-like ( G ) , DDX5-like ( H ) and eIF4a-like ( I ) ( green ) in S4 embryos . Percentage piwi-1+ cells coexpressing the indicated EEE marker ( red ) and percentage EEE+ cells coexpressing piwi-1 ( green ) appear in the lower left corner of merged images . ( F ) n = 895 piwi-1+ cells , n = 905 tct-like+ cells , n = 7 S3–S4 embryos . ( G ) n = 692 piwi-1+ cells , n = 728 BTF3+ cells , n = 6 S3–S4 embryos . ( H ) n = 676 piwi-1+ cells , n = 681 DDX5-like+ cells , n = 5 S3–S4 embryos . ( I ) n = 312 piwi1+ cells , n = 332 eIF4a+ cells , n = 4 S3–S4 embryos . DOI: http://dx . doi . org/10 . 7554/eLife . 21052 . 04710 . 7554/eLife . 21052 . 048Figure 5—source data 1 . Hierarchical clustering of S2–S4-enriched transcripts across embryogenesis . S2–S4 transcripts ( n = 1 , 756 ) were clustered using scaled RPKM values across embryogenesis ( Y , S2–S8 ) . Cluster membership and average RPKM values ( Y–S8 , C4 and SX ) are provided . Early-embryo-enriched ( EEE ) transcripts ( n = 1 , 048 ) , which were downregulated by S5 and were lowly expressed through S8 , comprise Clusters 5 , 6 and 8 . Separate tabs for EEE transcript Clusters 5 , 6 and 8 include annotation based on best BLASTx hits ( E < 0 . 001 ) versus the NR , Swiss-Prot , C . elegans , D . melanogaster , D . rerio , X . tropicalis , M . musculus and H . sapiens RefSeq databases . DOI: http://dx . doi . org/10 . 7554/eLife . 21052 . 04810 . 7554/eLife . 21052 . 049Figure 5—source data 2 . Validation of transcript expression trends using the Nanostring nCounter platform . Normalized Nanostring data tab: Normalized expression counts for positive ERCC spike-in controls ( POS A–F ) , negative controls ( NEG A–H ) , housekeeping genes , the blastomere and mitotic cell markers H2B and piwi-1 , the differentiating progenitor marker prog-1 , and 107 early-embryo-enriched ( EEE ) transcripts . Clusters tab: hierarchical clustering results for 107 EEE transcripts using normalized count data . Raw data tab: raw expression count data for positive ERCC spike-in controls ( POS A–F ) , negative controls ( NEG A–H ) , housekeeping genes , the blastomere and mitotic cell markers H2B and piwi-1 , the differentiating progenitor marker prog-1 , and 107 EEE transcripts . Nanostring probe sequences tab: target sequence region for capture and reporter probes . DOI: http://dx . doi . org/10 . 7554/eLife . 21052 . 04910 . 7554/eLife . 21052 . 050Figure 5—source data 3 . EEE transcript expression patterns detected by colorimetric WISH . Colorimetric whole mount in situ hybridization screen results detail expression patterns across embryos and C4 intact adults for select EEE transcripts . Numbers in parentheses indicate the number of embryos scored with expression in a given tissue ( numerator ) versus the total number of embryos scored ( denominator ) . pT4P-EEE transcript plasmid insert and cloning primer sequences are also provided . DOI: http://dx . doi . org/10 . 7554/eLife . 21052 . 05010 . 7554/eLife . 21052 . 051Figure 5—figure supplement 1 . Validation of early-embryo-enriched transcript expression trends using the Nanostring nCounter platform . ( A ) Count sums for positive controls ( left ) , housekeeping genes ( middle ) and experimental probes ( right ) for four biological replicates per sample: yolk ( Y ) , S2–S8 embryos , C4 asexual adults ( C4 ) and virgin sexually mature adults ( SX ) . Note the low read counts for the experimental S2 samples . ( B ) Log-scale plot of count sums for positive controls ( POSITIVE A–F ) and negative controls ( NEGATIVE A–H ) across all 40 samples assayed . Positive control probes detect exogenous ERCC control sequences added to the reactions at known concentrations . Negative control probes are not homologous to any known Smed sequence and measure background signal . ( C ) Raw counts for the housekeeping genes ACT-B , RPL-27 , G6PD , LUC7L3 , clathrin-1 , hgp-1 , rpl13a , ubiquilin-1 and zfp207-1 across all 40 experimental samples . ACT-B showed significant variation in expression across samples . ( D ) Normalization factors calculated by taking the geometric mean of the positive control sequences POSITIVE A–F ( left ) or the suite of housekeeping genes ( right ) across sample replicates . ( E ) Normalized expression trends for piwi-1 , H2B , and prog-1 during embryogenesis and adulthood ( Y , S2–S8 , SX , C4 ) mirror those observed in the single embryo RNA-Seq time course . ( F ) Heat map ( left ) generated by hierarchical clustering using normalized count data for 108 EEE transcripts across developmental time ( Y , S2–S8 , SX , C4 ) . Line graphs display expression trends for Clusters 1 ( blue ) , 2 ( red ) , and 4 ( green ) , which corroborate early-embryo-enriched expression observed in the RNA-Seq time course . Transcripts in Clusters 5–8 ( n = 9 ) exhibited low expression throughout the time course and/or exhibited expression trends different from those observed using RNA-Seq . DOI: http://dx . doi . org/10 . 7554/eLife . 21052 . 051 EEE transcript expression trends during embryonic development and adulthood were validated using the Nanostring nCounter platform ( Geiss et al . , 2008 ) . Total RNA replicates from single S2–S8 embryos , Y , C4 and SX adults were queried for expression of 108 EEE transcripts using a custom probe set ( Materials and methods , Figure 5—source data 2 ) . Curiously , S2 replicates had the lowest summed read counts across the experiment ( Figure 5—figure supplement 1A ) , and expression of piwi-1 , the cell cycle marker H2B , and EEE transcripts was not detected during S2 ( Figure 5 – Figure 1—figure supplement 1E–F , Figure 5—source data 2 ) , perhaps due to lack of template amplification in the nCounter assays . Apart from the S2 samples , expression trends for piwi-1 , H2B , the pro-differentiation factor prog-1 and more than 90% of the EEE transcripts mirrored those observed in the RNA-Seq time course ( Figure 5—figure supplement 1E–F ) . Most EEE transcripts queried on the nCounter platform showed peak expression during S3 and/or S4 , downregulation during S5 , and low to undetectable expression in late stage embryos , C4 or SX adults ( Figure 5—figure supplement 1F ) . Hierarchical clustering using normalized Nanostring count data identified three cohorts of EEE transcripts that hadcomposition and decay kinetics similar to those observed in the RNA-Seq time course; 65% of the EEE transcripts assayed co-clustered in both the RNA-Seqand nCounter analyses ( Figure 5—figure supplement 1F ) . To determine which cell types express EEE transcripts , and to examine spatiotemporal changes in EEE transcript expression , colorimetric WISH was performed on S2–S8 embryos and intact C4 adults ( Figure 5—source data 3 ) . Some EEE transcripts were expressed exclusively in differentiated temporary embryonic tissues . VAL-like , MPEG1-like-1 , MPEG1-like-2 and netrin-like were solely expressed in the temporary embryonic pharynx until S6 , whereas gelsolin-like and 4XLIM-like were expressed in both the primitive ectoderm and temporary embryonic pharynx during S3–S4 ( Figure 5—source data 3 , Figure 1—figure supplements 12C–G and 11D–E ) . Expression of these EEE transcripts is likely under zygotic control , occurring during or after the cell fate decisions to downregulate piwi-1 , exit the cell cycle and differentiate . Most EEE transcripts queried by WISH ( n = 15 , 75% assayed ) were expressed in both undifferentiated blastomeres and temporary embryonic tissue ( s ) during S3–S4 ( Figure 5B–E , Figure 5—source data 3 ) . Some of these transcripts may be maternally deposited , albeit we cannot ascertain the relative contribution ( s ) of maternal and zygotic expression from our RNA-Seq data . Expression of these EEE transcripts diminished greatly by S5 , with EEE transcripts that had been expressed at moderate or low levels becoming undetectable; specific expression of robustly expressed transcripts sometimes persisted until S6 ( Figure 5B–E , Figure 5—source data 3 ) . Consistent with the RNA-Seq and Nanostring nCounter results , EEE transcript expression was not detected by colorimetric WISH in S7 or S8 embryos or in C4 adults ( Figure 5B–E , Figure 5—source data 3 ) . Fluorescent double WISH performed with riboprobes complementary to piwi-1 and the EEE transcripts tct-like , BTF3-like , DDX5-like and eIF4a-like revealed coincident expression throughout the S4 blastomere compartment ( Figure 5F–I ) . Intriguingly , EEE transcript expression often decayed quicker in differentiated cells than in undifferentiated blastomeres , raising the possibility that regulation of EEE transcription and/or transcript stability may vary by cell type . Robust expression of many EEE transcripts was detected in blastomeres during S3–S4 , EEE transcript expression in temporary embryonic tissues was present during S3 and drastically diminished by S4 ( Figure 5B–E , Figure 5—source data 3 ) . EEE transcripts expressed throughout the undifferentiated piwi-1+ blastomere population in S3–S4 embryos are downregulated as definitive organogenesis begins during S5 . These transcripts likely represent a key temporal shift in the expression profile of piwi-1+ cells during embryogenesis . Moreover , EEE transcript expression provides a molecular metric to distinguish piwi-1+ blastomeres from adult neoblasts . Many developmental regulators implicated in lineage commitment and differentiation were expressed at low levels in S2–S4 embryos , and were upregulated dramatically as definitive organogenesis began during S5 . Key regulators of cell fate specification for many tissues , including the epidermis ( p53 and zfp-1 ) ( Pearson and Sánchez Alvarado , 2010; van Wolfswinkel et al . , 2014; Wagner et al . , 2012 ) , nervous system ( coe , sim , pax3/7 like , lhx1/5–1 and pitx ) ( Cowles et al . , 2013; Currie and Pearson , 2013; März et al . , 2013; Scimone et al . , 2014 ) , excretory system ( pou2/3 , six1/2–2 , eya , sal1 and osr ) ( Scimone et al . , 2011 ) , photoreceptor neurons ( eya , six-1/2 , otxA and soxB ) , pigment cup cells ( eya , six-1/2 , sp6-9 and dlx ) ( Lapan and Reddien , 2011 , 2012 ) and primordial germ cells ( nos ) ( Wang et al . , 2007 ) were among the S5- and/or S6-enriched transcripts ( Figure 1—source data 5–6 ) . Additional validated or putative drivers of cell fate determination in muscle ( myoD ) ( Cowles et al . , 2013; Scimone et al . , 2014 ) , the gastrovascular system ( prox-1 and foxA1 ) ( Adler et al . , 2014; Scimone et al . , 2014; van Wolfswinkel et al . , 2014 ) , nervous system ( lhx2/9 , six3-1 , nkx6-like , otxB-like , otxA , pax6a and pax6b ) ( Pineda et al . , 2002; Scimone et al . , 2014 ) and eyes ( ovo ) ( Lapan and Reddien , 2012 ) showed statistically significant upregulation during S5 and/or S6 , albeit with adjusted p-values and fold-changes above the stringent thresholds set for inclusion in the S5–S6-enriched transcript lists . GO analysis on S5-enriched transcripts showed statistically significant enrichment of terms associated with patterning and cell fate specification , transcriptional regulation , and development of organ systems including the epidermis , central and peripheral nervous system , muscle , digestive and excretory systems ( Figure 1—source data 5 ) . Taken together , these observations suggest that formation of many definitive organ systems begins during S5 , a supposition bolstered by WISH developmental time course data depicting expression patterns for numerous progenitor and cell type-specific markers during embryogenesis ( Figure 1—source data 9 , Figure 1—figure supplements 13–19 ) . Adult asexual knockdown phenotypes for many developmental TFs upregulated during S5 and S6 suggest that these genes are required for lineage specification , tissue maintenance , and production of new tissue during regeneration , with correspondence between affected tissues and site ( s ) of expression . Heterogeneous expression of TFs in neoblasts and post-mitotic progenitors informed the hypothesis that the neoblast population contains pluripotent stem cells as well as cycling , lineage-primed progenitors ( Reddien , 2013 ) . Single cell sequencing ( SCS ) studies suggest that the zeta ( ζ ) and gamma ( γ ) neoblast subclasses are epidermal and gut progenitors , respectively ( van Wolfswinkel et al . , 2014; Wurtzel et al . , 2015 ) , while the sigma ( σ ) neoblast subclass likely contains both pluripotent stem cells and progenitors for other lineages , including neural subtypes , protonephridia and primordial germ cells ( van Wolfswinkel et al . , 2014 ) . In practice , coexpression of pan-neoblast markers ( e . g . , piwi-1 ) and developmental TFs is used for neoblast subclass identification . Smed embryos are wholly reliant on cycling piwi-1+ cells for creation of new tissues , and heterogeneous expression of key developmental TFs within piwi-1+ blastomeres is predicted to generate the diverse array of progenitors required for organogenesis . While only a small fraction of the piwi-1+ compartment is predicted to be double positive for any given lineage marker , the entire population of lineage-positive cells is predicted to be positive for piwi-1 at its inception . As development proceeds , the fraction of lineage-positive cells coexpressing piwi-1 will decrease as cells downregulate expression of piwi-1 and differentiate further . If parallels with neoblasts hold true , then piwi-1+ blastomeres should self-renew and give rise to differentiating progeny during S5 . To assay for cells exiting the piwi-1+ compartment during S5 , embryos were costained with piwi-1 riboprobes and PIWI-1 antibodies . In adults , piwi-1 mRNA is restricted to the neoblast population , whereas PIWI-1 protein perdures in early post-mitotic progeny ( Guo et al . , 2006; Scimone et al . , 2011; Wagner et al . , 2011 ) . Indeed , recent work suggests that mechanisms exist to sequester piwi-1 mRNA and chromatoid bodies within one daughter cell during neoblast division , producing one cell that maintains neoblast identity and one cell that differentiates ( Lei et al . , 2016 ) . Virtually all S5 piwi-1+ blastomeres also contained PIWI-1 protein ( Figure 6—figure supplement 1 ) . Rare cells positive for PIWI-1 protein but in which piwi-1 mRNA was undetectable were observed in S5 embryos , suggesting that some of the division progeny exited the piwi-1+ blastomere population ( Figure 6—figure supplement 1 ) . To address whether lineages required for organogenesis arise within piwi-1+ blastomeres during S5 , expression of four evolutionarily conserved TFs implicated in tissue differentiation across three germ layers was examined singly and in combination with piwi-1 . p53 and pax6a , regulators of epidermal and neural fates , respectively , were proxies for ectodermal derivatives . Populations expressing myoD , a master regulator of muscle fate , were considered mesodermal derivatives , and populations expressing gata456a , a regulator of gut development , represented endodermal derivatives . While expression of these TFs was occasionally detected in a small number of cells during S4 , robust expression of p53 ( Figure 1—figure supplement 15B ) , gata456a ( Figure 1—figure supplement 13F ) , myoD ( Figure 1—figure supplement 17A ) and pax6a ( Figure 1—figure supplement 16B ) manifest in scattered parenchymal cells during S5 . Indeed , fluorescent double WISH with piwi-1 and p53 , gata456a , myoD or pax6a identified prospective epidermal , gut , muscle and neural progenitor populations that coexpressed piwi-1 and the developmental TFs in S5 embryos , as well as single positive populations for all of the markers assayed ( Figure 6A–D , Videos 11–14 ) . 10 . 7554/eLife . 21052 . 052Figure 6 . Adult lineages arise within the piwi-1+ blastomere population as organogenesis begins . ( A–D ) Developmental transcription factors implicated in tissue specific differentiation programs are expressed in subpopulations of piwi-1+ cells during S5 . Fluorescent WISH with piwi-1 ( red ) and p53 ( A ) , gata456a ( B ) , myoD ( C ) and pax6a ( D ) ( green ) riboprobes on S5 embryos . Embryos in ( B-D ) were costained with VAL-like , a temporary embryonic pharynx specific marker ( also in red ) . Right: Venn diagrams depict percentages of cells that were single or double positive for piwi-1 and the indicated TFs . Scale bars: 100 µm . ( E–G ) Hierarchical clustering of zeta ( ζ , E ) , gamma ( γ , F ) and sigma ( σ , G ) neoblast subclass-enriched transcripts during embryogenesis ( Y and S2–S8 ) , and in asexual ( C4 ) and virgin sexual ( SX ) adults . DOI: http://dx . doi . org/10 . 7554/eLife . 21052 . 05210 . 7554/eLife . 21052 . 053Figure 6—source data 1 . Behavior of ζ , γ and σ adult asexual neoblast subclass-enriched transcripts during embryogenesis . Hierarchical clustering of Zeta ( ζ ) , Gamma ( γ ) and Sigma ( σ ) subclass transcripts was performed using normalized RPKM data from the single embryo RNA-Seq developmental time course . Average RPKM values for Y , S2–S8 embryos and for C4 and SX adults are included . Transcripts with enriched at one or more embryonic stages are flagged . DOI: http://dx . doi . org/10 . 7554/eLife . 21052 . 05310 . 7554/eLife . 21052 . 054Figure 6—figure supplement 1 . PIWI-1 protein may perdure in cells committed to differentiation . ( A ) Fluorescent WISH costaining on S5 embryos using riboprobes against piwi-1 ( red ) and antibodies against PIWI-1 protein ( green ) . Cyan arrows: single positive cells for which PIWI-1 protein , but not piwi-1 mRNA , was detected . Scale bar: 100 µm . 99 ± 0 . 4% piwi-1+ cells were double positive for PIWI-1 protein , n = 4 , 152 piwi-1+ cells scored . 96 ± 3 . 1% PIWI-1 positive cells were double positive for piwi-1 mRNA , n = 4 , 296 PIWI-1+ cells scored . n = 4 S5 embryos . DOI: http://dx . doi . org/10 . 7554/eLife . 21052 . 05410 . 7554/eLife . 21052 . 055Video 11 . Definitive epidermal progenitors arise in the piwi-1+ blastomere population during S5 . SPIM -reconstructed S5 embryo costained with piwi-1 ( red ) and p53 ( green ) . Definitive epidermal progenitors , coexpressing piwi-1 and p53 , are dispersed in the embryonic wall . As definitive epidermal progenitors differentiate , they are predicted to downregulate piwi-1 and to retain expression of p53 . DOI: http://dx . doi . org/10 . 7554/eLife . 21052 . 05510 . 7554/eLife . 21052 . 056Video 12 . Definitive gut progenitors arise in the piwi-1+ blastomere population during S5 . SPIM reconstructed S5 embryo costained with piwi-1 and VAL-like ( both in red ) and gata456a ( green ) . Definitive gut progenitors , coexpressing piwi-1 and gata456a , are dispersed in the embryonic wall . As definitive gut progenitors differentiate , they are predicted to downregulate piwi-1 and to retain expression of gata456a . VAL-like is expressed the temporary embryonic pharynx and is not detected in piwi-1+ blastomeres . DOI: http://dx . doi . org/10 . 7554/eLife . 21052 . 05610 . 7554/eLife . 21052 . 057Video 13 . Muscle progenitors arise in the piwi-1+ blastomere population during S5 . SPIM reconstructed S5 embryo costained with piwi-1 and VAL-like ( both in red ) and myoD ( green ) . Muscle progenitors , coexpressing piwi-1 and myoD , are dispersed in the embryonic wall . As muscle progenitors differentiate , they are predicted to downregulate piwi-1 and to retain expression of myoD . VAL-like is expressed the temporary embryonic pharynx and is not detected in piwi-1+ blastomeres . DOI: http://dx . doi . org/10 . 7554/eLife . 21052 . 05710 . 7554/eLife . 21052 . 058Video 14 . Neural progenitors arise in the piwi-1+ blastomere population during S5 . SPIM reconstructed S5 embryo costained with piwi-1 and VAL-like ( both in red ) and pax6a ( green ) . Neural progenitors , coexpressing piwi-1 and pax6a , are dispersed in the embryonic wall . As neural progenitors differentiate , they are predicted to downregulate piwi-1 and to retain expression of pax6a . VAL-like is expressed the temporary embryonic pharynx and is not detected in piwi-1+ blastomeres . DOI: http://dx . doi . org/10 . 7554/eLife . 21052 . 058 The advent of developmental TF expression within piwi-1+ blastomeres during S5 , coupled with downregulation of EEE transcripts , may signal the emergence of molecularly distinct subpopulations akin to the neoblast subclasses . Whole embryo expression trends for EEE , σ , γ and ζ class transcripts suggest that large scale , cell-intrinsic shifts in gene expression occur within blastomeres between S4 and S5 . Moreover , the developmental output of piwi-1+ blastomeres , as described by molecular fate mapping , diversifies greatly during S5 and S6 . The expression signatures of the ζ and γ neoblast subclasses emerge during S5 . ζ neoblasts require p53 and zfp-1 activities for production of post-mitotic mesenchymal progenitors , which simultaneously differentiate , migrate and ultimately integrate into the epidermis during homeostasis ( van Wolfswinkel et al . , 2014 ) . p53 , zfp-1 , tcf15 , sox-P3 , fgfr1 , egr-1 , six6* , gata123* ( van Wolfswinkel et al . , 2014 ) and 15 additional transcripts used as classifiers of ζ subgroup identity in SCS experiments ( Wurtzel et al . , 2015 ) displayed statistically significant upregulation in whole embryos during S5 or later developmental stages ( Figure 6E , Figure 6—source data 1 , Figure 1—figure supplement 15A , Figure 1—source data 5–8 ) . Furthermore , the ζ neoblast transcripts soxP3 , egr-1 , fgfr1 and prog-1 clustered together in the analysis of adult asexual neoblast-enriched transcripts , indicating that they displayed similar expression profiles during embryogenesis ( Figure 4A , Figure 4—source data 1 ) . Transcripts specifically expressed in post-mitotic epidermal progenitors downstream of ζ neoblasts , including prog-1 , AGAT-1 and zpuf-6 , were enriched and first detected by WISH during S5 ( Figure 1—figure supplement 15C , E–G , Figure 1—source data 5 ) . γ neoblasts are identified by enriched expression of gata456a , hnf4 , prox-1 and nkx2 . 2* ( van Wolfswinkel et al . , 2014; Wagner et al . , 2011 ) . prox-1 and nkx2 . 2 were expressed at low levels in S2–S4 embryos and showed statistically significant upregulation during S5 , and 15 additional γ neoblast-enriched transcripts identified in SCS experiments were enriched during S5 or later in development ( Figure 6F , Figure 6—source data 1 , Figure 1—figure supplement 13E , Figure 1—source data 5–8 ) . The molecular staging series detected expression of gata456a and hnf4 in yolk and early embryos ( S2–S4 ) ( Figure 6F ) . However , gata456a and hnf4 expression were solely detected in the embryo proper by WISH , first in the developing temporary embryonic pharynx during S2 and later in parenchymal cells during S5 ( Figure 1—figure supplement 13F–G ) . Charting the emergence of the σ neoblast subclass is hampered by limitations of the subclass designation: σ neoblasts are presumed to be an amalgamation of several progenitor populations and pluripotent stem cells . Several σ -class genes , notably SoxP-1 and SoxP-2 , were expressed in parenchymal cells throughout embryogenesis , similar to piwi-1 ( Figure 4A , Figure 4—source data 1 , Figure 4—figure supplement 1A–C , Figure 6—source data 1 ) . SoxP-1 is required for neoblast maintenance ( Wagner et al . , 2012 ) , suggesting that σ-class transcripts with expression profiles to SoxP-1 and SoxP-2 during embryogenesis may also be expressed in pluripotent neoblasts . In contrast , lineage-primed progenitor factions within the σ subclass probably arise during S5 . Genes with functions in tissue-specific differentiation programs , including soxB1 ( Lapan and Reddien , 2012; Monjo and Romero , 2015 ) , pou2-3 ( Scimone et al . , 2011 ) , nos ( Wang et al . , 2007 ) , pitx ( Currie and Pearson , 2013; März et al . , 2013 ) , sim ( Cowles et al . , 2013 ) , and smad6/7–1 ( González-Sastre et al . , 2012 ) were expressed at low levels in S2–S4 embryos and were upregulated during S5 ( Figure 6G , Figure 6—source data 1 , Figure 1—source data 5 ) . The gene expression signature of the adult neoblast compartment is an emergent property of the piwi-1+ population during embryogenesis . First , EEE transcripts are uniquely associated with the expression signature ( s ) of undifferentiated piwi-1+ blastomeres in early embryos . Second , adult neoblast subclasses arise as lineages are born within piwi-1+ blastomeres during S5 . Molecular heterogeneity within the neoblast compartment is largely attributed to the diverse array of lineage-dedicated progenitors within the population , a hypothesis supported by our finding that subclass marker expression was dramatically upregulated as organogenesis began . Progenitor subpopulations required for organ formation during embryogenesis persist into adulthood , where steady-state output from different lineages maintains tissue homeostasis . At present , we cannot distinguish whether lineages perpetually re-emerge due to asymmetric division of pluripotent stem cells , or whether progenitor populations established during embryogenesis are maintained by self-renewal . These observations beg the question: do piwi-1+ cells behave similarly to neoblasts throughout embryogenesis ? Or alternatively , is neoblast activity an acquired feature that emerges in tandem with the adult molecular expression signature ? Neoblasts are completely and irreversibly eliminated following treatment with 6 , 000 Rads ( Reddien et al . , 2005; Wagner et al . , 2011 ) , causing irradiation sickness and ultimately death . Transplantation of wildtype adult tissue grafts or cell suspensions into lethally irradiated adult hosts that are devoid of stem cells results in engraftment and expansion of donor-derived piwi-1+ cells , production of differentiated progeny , reconstitution of the neoblast compartment and rescue from lethality ( Baguñà J and Auladell , 1989; Guedelhoefer and Sánchez Alvarado , 2012; van Wolfswinkel et al . , 2014; Wagner et al . , 2011 ) . piwi-1+ cells that form pluripotent , expanding colonies following sublethal irradiation or transplantation into a lethally irradiated host are called clonogenic neoblasts ( cNeoblasts ) ( Wagner et al . , 2011 ) . cNeoblasts are predicted to have a widespread distribution in the parenchyma , and this population contains within it the most primitive stem cells . At present , the operational definition for cNeoblast exists apart from the gene expression signatures for the neoblast subclasses; cNeoblasts are likely contained within , but may not be exclusive to , the σ class . To assess whether Smed embryos harbor piwi-1+ cells with cNeoblast activity , heterochronic , heterotopic transplantations were performed . S4 , S5 , S6 , S7 and S8 embryonic cell suspensions were injected into the tail parenchyma of lethally irradiated sexual adult hosts at 3 days post-irradiation ( dpi ) ( Figure 7A , Materials and Methods ) . To determine whether comparable numbers of piwi-1+ cells were introduced per host for the developmental stages assayed , transplanted hosts were fixed at 1 hr post-transplant ( hpt ) and stained with piwi-1 riboprobes . More than 85% of S4–S8 transplants fixed at 1 hpt contained piwi-1+ cell ( s ) in the tail parenchyma , suggesting that the cell injection technique was robust and reliable ( Figure 7B ) . S5 , S6 , S7 and S8 transplants contained comparable numbers of piwi-1+ cells per host at 1 hpt , while significantly fewer piwi-1+ cells were introduced per S4 embryonic cell transplant ( Figure 7C , F ) . 10 . 7554/eLife . 21052 . 059Figure 7 . Embryonic cells acquire the ability to engraft , persist and proliferate in an adult microenvironment as organogenesis proceeds . ( A ) Schematic depicting the workflow for heterochronic transplantation experiments . S4 , S5 , S6 , S7 or S8 embryonic cell suspensions were injected into the tails of lethally irradiated sexual adult hosts at 3 days post-irradiation ( dpi ) . Cohorts of transplanted animals were fixed at 1 hr and 5 days post-transplantation ( 1 hpt and 5 dpt , respectively ) for staining with piwi-1 riboprobes and H3S10p antibodies . Lethally irradiated , uninjected host controls were fixed and stained at 5 dpt . ( B ) Percentage of transplanted animals fixed at 1 hpt ( blue bars ) or 5 dpt ( red bars ) containing one or more donor-derived piwi-1+ cell ( s ) . X-axis: stage ( S ) of donor cells . ( C ) Number of donor-derived piwi-1+ cell ( s ) per transplant at 1 hpt and 5 dpt . Each point represents one transplanted animal . Mean ± standard deviation ( black bars ) are shown . Statistical tests were performed using a generalized linear model , assuming that the counts followed a Poisson distribution . S4 transplants contained significantly fewer piwi-1+ cells at 1 hpt than S5 , S6 , S7 or S8 transplants ( Tukey post-hoc comparisons , S4 vs S5 , S4 vs S6 and S4 vs S7 , S4 vs S8: p<0 . 001 ) . Group differences in the number of piwi-1+ cells at 1 hpt for S5 and S6 transplants were not statistically significant ( p=0 . 21 ) . Significantly fewer S4 and S5 donor-derived piwi-1+ cells persisted at 5 dpt than were observed for later stages ( Tukey post-hoc comparisons: S4 vs S5 , S4 vs S6 , S4 vs S7 and S4 vs S8: p<0 . 001 . S5 vs S6 , S5 vs S7 , S5 vs S8: p<0 . 001 ) . ( D ) Percentage of transplants with mitotic piwi-1+ cell ( s ) at 5 dpt ( green bars ) . X-axis: Donor cell stage . ( E ) Mitotic index for donor-derived piwi-1+ cells at 5 dpt . Stage-specific differences were not observed for S4–S8 embryonic cell populations using a generalized linear model , assuming counts followed a Poisson distribution and the number of piwi-1+ cells as a covariate . ( B–E ) Numbers of transplants scored: S4: n = 36 ( 1 hpt ) , n = 43 ( 5 dpt ) , four independent experiments . S5: n = 15 ( 1 hpt ) , n = 16 ( 5 dpt ) , two independent experiments . S6: n = 31 ( 1 hpt ) , n = 29 ( 5 dpt ) , four independent experiments . S7: n = 31 ( 1 hpt ) , n = 30 ( 5 dpt ) , four independent experiments . S8: n = 19 ( 1 hpt ) , n = 20 ( 5 dpt ) , three independent experiments . ( F ) Confocal maximal projections of S4 , S5 , S6 , S7 and S8 embryonic cell transplants fixed at 1 hpt and 5 dpt . Animals were stained with piwi-1 riboprobes ( green ) , antibodies against the mitotic marker H3S10p ( red , 5 dpt only ) and DAPI nuclear counterstain ( blue ) . S6 , S7 and S8 insets: mitotic piwi-1+ cells . Red arrows indicate mitotic cells magnified in insets . Yellow arrows: mitotic piwi-1+ cells . Scale bar ( inset ) : 20 µm . Scale bar ( panel ) : 100 µm . ( B–C ) S4–S8 embryonic piwi-1+ cells were reliably introduced into hosts . S6–S8 embryonic piwi-1+ cells persisted in an adult microenvironment . ( D-E ) S6–S8 embryonic piwi-1+ cells proliferated in an adult microenvironment . DOI: http://dx . doi . org/10 . 7554/eLife . 21052 . 059 To assay whether embryonic piwi-1+ cells persisted and proliferated in an adult microenvironment , cohorts of transplanted animals and lethally irradiated uninjected hosts were fixed at 5 days post-transplantation ( dpt ) and stained with piwi-1 riboprobes and H3S10p antibodies ( Figure 7B–F ) . Persistent piwi-1+ cells from S6 , S7 and S8 embryos were observed in the vast majority of transplants scored at 5 dpt ( Figure 7B ) , and no statistically significant difference among stages was detected in the mean number of piwi-1+ cells present per host ( Figure 7C ) . Moreover , most S6 , S7 and S8 embryonic donor cell transplants contained dividing piwi-1+ cell ( s ) at 5 dpt ( Figure 7D–F ) , and no statistically significant difference among stages was detected in the mitotic index of donor-derived piwi-1+ cells per host at 5 dpt ( Figure 7E ) . In contrast , S4 and S5 embryonic piwi-1+ cells were significantly less likely to persist in adult hosts at 5 dpt than were S6 , S7 or S8 embryonic piwi-1+ cells ( Figure 7B–C ) . Persistent S4 derived piwi-1+ cells were rarely observed at 5 dpt , and cell division was not observed ( Figure 7D–E ) . As expected , piwi-1+ cells were never observed in uninjected lethally irradiated hosts ( n = 82 hosts scored at 3 dpi ) . Fewer S5 embryonic cell transplants contained persistent piwi-1+ cells at 5 dpt ( Figure 7B ) , and the number of S5 derived piwi-1+ cells per host at 5 dpt was significantly lower than for later stages ( Figure 7C ) . Similarly , the fraction of S5 embryonic cell transplants containing mitotic piwi-1+ cell ( s ) was reduced relative to S6 and later stages ( Figure 7D ) . The reduced persistence of S4 and S5 derived piwi-1+ cells in an adult microenvironment was probably not attributable to technical variability or to the absolute number of embryonic piwi-1+ cells introduced per host , since comparable numbers of S5 , S6 , S7 and S8 piwi-1+ cells were introduced per transplant ( Figure 7C ) . Stage-specific , cell autonomous factors likely underlie the starkly different responses of S5 and S6 embryonic cells following transplantation into the adult parenchyma . These results suggest that S4 and S5 piwi-1+ blastomeres are functionally distinct from piwi-1+ cells present at S6 and later stages . piwi-1+ blastomeres acquire competency to engraft and respond appropriately to adult environmental cues as development proceeds during S5 . To assess whether embryos undergoing organogenesis harbor cells capable of rescuing and reconstituting lethally irradiated adult hosts , S5 , S6 , S7 and S8 embryonic cell suspensions were injected into lethally irradiated sexual adult hosts at 1 dpi ( Figure 8A , Materials and methods ) . Consistent with previous results , persistent piwi-1+ cells from S6 , S7 and S8 embryos were observed in the vast majority of samples scored at 5 dpt ( Figure 8B ) , and no statistically significant difference was detected in either the mean number or the mitotic index of piwi-1+ donor-derived cells per host at 5 dpt ( Figure 8C–D ) . Likewise , donor-derived piwi-1+ cells from S5 embryos were far less likely to persist and divide in adult hosts ( Figure 8B–D ) . As expected , piwi-1+ cells were never observed in uninjected lethally irradiated hosts ( n = 81 hosts scored at 6 dpi ) . 10 . 7554/eLife . 21052 . 060Figure 8 . Embryos undergoing organogenesis contain cNeoblasts . ( A ) Schematic for heterochronic transplantation experiments . S5 , S6 , S7 or S8 embryonic cell suspensions were injected into the tail parenchyma of lethally irradiated sexual adult hosts at 1 day post irradiation ( dpi ) . Cohorts of transplanted animals and uninjected host controls were fixed at 5 days post-transplantation ( dpt ) for staining with piwi-1 riboprobes and H3S10p antibodies . The remaining animals were monitored for 70 dpt for survival and rescue . ( B ) Percentage of transplants with persistent , donor-derived piwi-1+ cell ( s ) ( blue ) or donor-derived mitotic ( piwi-1+ , H3S10p+ ) cell ( s ) ( red ) at 5 dpt . X-axis: Embryonic donor cell stage . ( C ) Number of piwi-1+ cells per transplanted host at 5 dpt for S5–S8 embryonic cell transplants . Each point represents one transplanted animal . Means ± standard deviation ( SD ) are shown ( black bars ) . Statistically significant differences in the number of persistent piwi-1+ cells per transplant at 5 dpt were observed using a generalized linear model , assuming that count data followed a Poisson distribution . S5 transplants contained fewer persistent piwi-1+ cells than S6 or S7 transplants ( Tukey post-hoc comparisons , S5 vs S6: p<0 . 0001 , S5 vs S7: p<0 . 0001 , S5 vs S8: p<0 . 0001 ) . ( D ) Mitotic index for donor-derived piwi-1+ cells at 5 dpt for S5–S8 embryonic cell transplants . Each point represents one transplanted animal . Means ± standard deviation ( SD ) are shown ( black bars ) . Statistically significant differences in the piwi-1+ cell mitotic index were observed using a generalized linear model with piwi-1+ cell counts as a covariate , assuming that count data followed a Poisson distribution . S5 transplants contained significantly fewer cycling cells than S6 , S7 or S8 transplants ( Tukey post-hoc comparisons , S5 vs S6: p<0 . 01 , S5 vs S7: p<0 . 01 , S5 vs S8: p<0 . 001 ) . ( E ) Confocal maximal projections for S5 , S6 , S7 and S8 embryonic cell transplants fixed at 5 dpt and stained with piwi-1 riboprobes ( green ) , H3S10p antibodies ( red ) and DAPI ( blue ) . S6 , S7 and S8 insets show mitotic piwi-1+ cells . Red arrows indicate mitotic cells magnified in the insets . Yellow arrows: mitotic piwi-1+ cells . Scale bar ( inset ) : 20 µm . Scale bar ( panel ) : 100 µm . ( B–E ) Numbers of transplants scored in four independent experiments: S5 n = 22; S6 n = 24; S7 n = 21; S8 n = 27 in ( C ) , n = 21 in ( D ) . ( F ) Survival curves for S5 , S6 , S7 and S8 embryonic cell transplants and uninjected 6 , 000-Rad-irradiated host controls as a function of time ( days ) post-transplant . ( G ) Live images of regenerating S6 , S7 and S8 rescue animals . Left: Tail fragment after self-amputation of head and trunk tissue . Middle: Tail fragment with unpigmented anterior blastema ( yellow arrowheads ) . Right: Animal with new head tissue and developing eyes ( yellow arrows ) and a regenerated pharynx ( yellow asterisk ) . Animals from different experiments are shown in the S7 panels; the same animals are shown in the S6 and S8 panels . Dorsal view . Anterior: top . Scale: 100 µm . ( F–G ) Numbers of transplants scored in four independent experiments: host controls n = 89; S5 n = 105; S6 n = 90; S7 n = 92; S8 n = 85 . Rescue animals were obtained in two experiments for S6 and S7 transplants , and four experiments for S8 transplants . ( B–E ) S6 , S7 and S8 embryonic donor cells persist and divide in the adult parenchyma . ( F–G ) S6 , S7 and S8 embryonic cells can rescue lethally irradiated adult hosts . DOI: http://dx . doi . org/10 . 7554/eLife . 21052 . 06010 . 7554/eLife . 21052 . 061Figure 8—figure supplement 1 . Progression of irradiation-induced phenotypes , rescue or death for heterochronic transplantation assays . ( A–E ) Stacked bar graphs depicting the percentage of S5 ( A ) , S6 ( B ) , S7 ( C ) and S8 ( D ) transplanted animals or uninjected , 6 , 000-Rad-irradiated hosts ( E ) that displayed no visible phenotype ( blue ) , head regression ( red ) , head and tail regression ( yellow ) , head regression and pharynx lesions ( green ) , head regression and ventral curling ( purple ) , death ( gray ) or rescue ( the development of an anterior blastema and ensuing regeneration of the host animal ) ( orange ) . Bulk cell transplantation was performed as described in Figure 8A and the Materials and methods . Transplanted animals and uninjected controls were monitored for 70 days post-irradiation ( dpi ) . S5 transplants: n = 105 animals . S6 transplants: n = 90 animals . S7 transplants: n = 92 animals . S8 transplants: n = 85 animals . Uninjected sexual adult hosts: n = 89 animals . Results were tallied from four independent experiments for each embryonic stage assayed . DOI: http://dx . doi . org/10 . 7554/eLife . 21052 . 061 Survival of S5–S8 transplants and host controls was monitored for 70 dpt ( i . e . , 71 dpi ) . Worms were scored for irradiation-induced phenotypes , including head and tail regression , lesion formation , ventral curling and death . Rescue , anterior blastema formation and subsequent regeneration of an entire individual from injected tail fragment , was also scored ( Guedelhoefer and Sánchez Alvarado , 2012; Wagner et al . , 2011 ) ( Figure 8F–G , Figure 8—figure supplement 1 ) . Irradiation-induced phenotypes manifest in control and transplanted individuals alike between 7–14 dpt ( i . e . , 8–15 dpi ) ( Figure 8—figure supplement 1A–E ) , and none of the controls survived long-term ( Figure 8F , Figure 8—figure supplement 1E ) . Remarkably S6 , S7 and S8 embryonic donor cells were capable of rescuing lethally irradiated adult hosts ( Figure 8F–G ) . Rescued animals underwent complete head regression , self-amputation posterior to the pharynx , and tail fragments that formed anterior blastemas which promoted regeneration of individuals containing two visible eyes , a central pharynx and triclad gut ( Figure 8F–G , Figure 8—figure supplement 1B–D ) . None of the S5 embryonic cell recipients mounted a rescue response ( Figure 8F , Figure 8—figure supplement 1A ) . The rescue assay results suggest that S6 , S7 and S8 embryos harbor cNeoblasts , cells that are capable of self-renewing and producing the diverse array of cell types required for whole animal regeneration . Taken together , stage-dependent molecular and functional distinctions exist among piwi-1+ blastomeres before and after organogenesis begins . S4 piwi-1+ cells , which express EEE and adult asexual neoblast enriched transcripts , are largely incapable of persisting and dividing in an adult microenvironment ( Figure 7B–C ) . During S5 , as dramatic shifts in gene expression occur , piwi-1+ cells become competent to persist and proliferate in the adult parenchyma ( Figures 7B–C and 8B–D ) . We propose that cNeoblasts arise during S5 . Heterochronic transplantation experiments revealed that S6 , S7 and S8 embryos possess cells that behaved similarly to adult cNeoblasts: they consistently engrafted into adult hosts , proliferated and were ultimately capable of mounting a rescue response ( Figure 8B–G ) . Acquisition of cNeoblast activity during embryogenesis correlates temporally with large-scale changes in gene expression observed at the outset of organogenesis , suggesting that pluripotent stem cells and lineage-primed progenitors first emerge during S5 ( Figure 9 ) . The ontogeny of the adult neoblast compartment can therefore be traced back to the piwi-1+ zygote , which gives rise to anarchic , cycling piwi-1+ blastomeres , some of which persist in the S3–S4 embryonic wall and establish pluripotent neoblasts and progenitor subpopulations during S5 . The remarkable developmental plasticity of adult planarians is likely due to the singular ability of neoblasts to perpetuate and redeploy embryonic developmental programs . 10 . 7554/eLife . 21052 . 062Figure 9 . Ontogeny of the adult neoblast compartment . Asynchronously cycling piwi-1+ cells fuel embryogenesis , giving rise to all temporary and definitive tissues . During S2 , some piwi-1+ blastomeres ( purple cells ) exit the cell cycle and differentiate into temporary embryonic tissues ( primitive ectoderm , temporary embryonic pharynx and primitive endoderm ) . The remaining piwi-1+ blastomeres , located in the embryonic wall ( purple cells , S3-S4 ) , continue to divide and express both EEE transcripts ( turquoise arrow ) and adult asexual neoblast enriched transcripts ( red arrow ) . As organogenesis begins during S5 , EEE transcripts are downregulated throughout the compartment ( purple cells transition into red ) . Concomitantly , progenitor subpopulations required for definitive organ formation are specified via the heterogeneous expression of developmental transcription factors within the piwi-1+ population ( colored cells denote different progenitor subpopulations ) . Adult pluripotent neoblasts , themselves a lineage , are established during S5 ( red cells ) . Embryonic donor cells harvested during or after S6 function similarly to adult neoblasts ( cNeoblast activity , gray arrow ) . Pluripotent and lineage-primed neoblasts established during embryogenesis are maintained throughout the lifetime of the animal . Neoblasts are required for tissue maintenance during homeostasis and the formation of new tissue during regeneration . DOI: http://dx . doi . org/10 . 7554/eLife . 21052 . 062
To identify similarities and key differences between embryogenesis and regeneration , careful consideration must be given to distinctions in context , chronology , scope and type of regeneration ( homeostatic or facultative ) , and to the ontogeny of effector cell type ( s ) . Which aspects of embryonic development are recapitulated during regeneration , and which are context-specific ? What factors influence the competency and extent of regenerative responses , and do these factors have embryonic origins ? The workhorses of modern developmental biology , C . elegans , D . melanogaster , D . rerio , and M . musculus have limited , if any , regenerative potential during adulthood , precluding or severely limiting comparative inquiries . In contrast , Smed adults exhibit robust homeostatic and facultative regenerative potential , both neoblast-dependent phenomena that have largely been studied using a clonal , asexually reproducing strain . Neither descriptive nor functional studies of Smed embryogenesis have been reported . We generated two foundational resources: a molecular staging series for Smed embryogenesis and an expression atlas describing temporary and definitive organ development . To investigate the embryonic origins of regeneration , we provide an ontogeny for the adult neoblast compartment . The molecular staging series facilitated the identification of embryonic predecessors of adult neoblast lineages and the developmental transition when neoblast specification occurs . Adult neoblast lineages , including pluripotent stem cells and lineage-primed progenitors , are established as definitive organogenesis begins . Neoblast lineages , required for organ formation during embryogenesis , persist into adulthood , where they are redeployed for homeostatic maintenance of all differentiated tissues , including the germline , and formation of missing tissues during regeneration . Moving forward , investigation of neoblast dynamics , particularly the regulation of self-renewal and commitment to differentiation , will be a central , unifying feature of comparative studies on organ formation , maintenance , repair and replacement . Unraveling the molecular mechanism of neoblast specification during embryogenesis will provide insight into the identity of adult pluripotent stem cells indispensible for homeostatic and facultative regeneration in planarians . Intensive efforts to define the term neoblast , and to characterize the presumed plurality of cell types within the asexual adult neoblast population , have generated numerous , unreconciled molecular and functional criteria . At present , gene expression signatures for neoblast subpopulations cannot be correlated with functional distinctions ( should they exist ) in self-renewal , pluripotency or cell fate restriction . It is not known whether hierarchical relationships exist between pluripotent cells and progenitors , whether cycling progenitors can dedifferentiate and/or interconvert , how stable subpopulations are over time , or when and how progenitors make an irreversible commitment to differentiation . Furthermore , preferred usage of Smed C4 asexual animals to classify and examine relationships among subpopulations precludes comprehensive examination of the lineage repertoire . The expression of pluripotency factors is probably necessary , but not sufficient , for the assumption of neoblast fate . We report that many genes known to regulate adult asexual neoblast self-renewal and maintenance were expressed throughout embryogenesis , and showed that several of these genes were expressed throughout S3–S5 piwi-1+ blastomeres . Functional differences between blastomeres and neoblasts were observed in heterochronic , heterotopic transplantation experiments . S4 and S5 embryonic donor cells were far less likely to persist and proliferate in stem cell deficient adult hosts than were embryonic donor cells harvested during S6 or later . Moreover , S5 embryonic donor cells did not rescue lethally irradiated hosts , whereas S6–S8 embryonic donor cells functioned like adult neoblasts and were capable of rescue . These results suggest that neoblast specification occurs during S5 , and highlight the importance of cell-intrinsic changes in the blastomere to neoblast transition . Differences in transplanted cell behavior correlated with large-scale changes in gene expression as definitive organogenesis began . EEE transcripts , uniquely expressed in blastomeres , are attractive candidates for repressors of neoblast fate , and/or effectors of molecular changes in blastomeres that are necessary for acquisition of neoblast fate . It is not known whether EEE transcript downregulation in blastomeres is necessary for neoblast specification , or whether expression of EEE transcripts and developmental TFs required for lineage specification are mutually exclusive . Further studies will determine whether and how maternal factors influence the early stages of Smed embryogenesis . Which EEE transcripts are maternally deposited , and which mechanisms effect maternal transcript degradation ? When do wave ( s ) of zygotic genome activation occur , and how does zygotic genome activation in blastomeres relate to cell differentiation and neoblast specification ? Additional factors , such as changes in chromatin state , may also correlate with or play a causative role in the blastomere to neoblast transition . Ectolecithal development and dispersed cleavage pose unique developmental challenges , making Smed embryogenesis a novel paradigm for regulative development . Herein , the collective pluripotency of an anarchic , cycling piwi-1+ population generates the diversity of cell types required for the development of all temporary and definitive organ systems in these bilaterally symmetric , triploblastic animals . We showed that piwi-1+ blastomeres are spatially disordered within the embryonic wall during S3–S4 , and that dispersed epidermal , gut , muscle and neural progenitors arise within the blastomere population during S5 . Further investigation is needed to understand how signals from differentiated tissues impact cell fate decisions within the piwi-1+ blastomere compartment and effect progenitor cell migration , communication and interactions necessary to form organ rudiments . Smed embryogenesis provides a unique vantage point from which to investigate the origin , anatomical composition and signaling logic underlying the neoblast niche . During sphere formation , differentiating blastomeres must interact and self-assemble temporary embryonic tissues , providing structure to the embryo and establishing a microenvironment that promotes maintenance and expansion of piwi-1+ blastomeres . In turn , blastomeres are likely to effect changes that are conducive to the establishment and/or maintenance of the neoblast population . Neoblast lineages are predicted to actively maintain their niche during the lifetime of the animal . Although the adult gastrovascular system has long been suspected of providing trophic support signals to neoblasts ( Forsthoefel et al . , 2012 ) , the molecular mechanisms underlying this phenomenon remain elusive . Expansion of piwi-1+ blastomeres correlates with that of an ill-defined embryonic gut population ( Figure 1—figure supplement 13A–D ) . Investigating gut development and gut communication with the piwi-1+ population during embryogenesis may uncover key regulators of neoblast specification or regulators of neoblast dynamics that may similarly impact stem cell behavior during adulthood . We report that many transcripts implicated in lineage commitment and classification of neoblast subclasses wereexpressed at low levels in early embryos and were dramatically upregulated as organogenesis began . This observation is consistent with the hypothesis that neoblast heterogeneity is due to the presence of different subpopulations of cycling , lineage-primed progenitors within the compartment ( Reddien , 2013 ) . It also suggests that organ formation during embryogenesis probably utilizes many of the same genetic regulatory networks and transition states elucidated during adult homeostasis and regeneration . Development of techniques to interrogate gene function during embryogenesis will enable us to identify master regulators of organogenesis for different tissues , and to address similarities and differences in their modes of action during embryogenesis and adulthood . Embryogenesis may also provide a vantage point for the identification of upstream activators for these developmental TFs , helping to address how diverse , dispersed patterns of gene activation arise in the blastomere and neoblast compartments . The key distinction between embryogenesis and regeneration is the de novo formation of tissues in the former , and the presence of preexisting structures and signaling environments in the latter . Studies can now be performed to assess how tissues that are hypothesized to be instructive for regeneration are initially established during embryogenesis , and how the formation of these tissues relates to the acquisition of regenerative potential during development . For example , planarian body wall muscle is hypothesized to be required for the re-specification of axial identities during regeneration ( Witchley et al . , 2013 ) . At present , technical limitations preclude tissue-specific knockdown experiments that would address requirements for polarity genes in muscle during regeneration . However , we can now address when and how the definitive axes are established during Smed embryogenesis , including the identification of tissues and signals that initially polarize embryos prior to the development of body wall muscle . Examining axis formation during embryogenesis , and comparing the process across different chronological and developmental contexts , may uncover roles for additional tissues and/or novel polarity regulators . Furthermore , we can address which developmental milestones and gene products are required to establish a state of regeneration competency in the embryo . Sustained effort and continued investment in the adaptation and development of new technologies for the molecular interrogation of Smed embryogenesis will facilitate discoveries that may challenge long-held assumptions about developmental processes , including cell fate specification , pattern formation and adult stem cell regulation . Moreover , utilizing Smed for comparative studies of embryogenesis and regeneration presents an unprecedented opportunity for formal examination of the embryonic origins of regenerative potential .
Sexually reproducing S . mediterranea ( Smed ) stocks were descendants of animals collected in Sardinia by Dr . Maria Pala in 1999 . Animals from the clonally derived sexual strain S2F1L3F2 ( Wagner et al . , 2011 ) and from the asexual clonal strain CIW-4 ( C4 ) ( Newmark and Sánchez Alvarado , 2002 ) were propagated via successive rounds of amputation and regeneration . Animals were maintained in 1x Montjuic water at 20°C in the dark and fed homogenized beef liver as previously described in Cebrià and Newmark ( 2005 ) . Cultures subjected to intensive cutting and/or feeding regimens were supplemented with 100 µg/mL gentamicin sulfate ( Gemini Bioproducts , #400–100P ) . Egg capsules were collected daily from outbred cohorts of sexually mature adults cultured at low density ( 6–8 animals per 400 mL culture ) , and were stored in dated Petri dishes at 20°C in constant darkness until use . The collection date was considered 1 day post-egg capsule deposition . To maintain optimal fertility levels , sexually mature animals used for egg capsule collections were replaced every 3–4 months with either juveniles ( 6–8 weeks post hatching ) or adult regenerates ( 6–8 weeks post cut ) . Live embryos were dissected out of egg capsules in 1x Holfreter’s buffer ( 3 . 5 g/L NaCl; 0 . 2 g/L NaHCO3; 0 . 05 g/L KCl; 0 . 2 g/L MgSO4; 0 . 1 g/L CaCl2; 1 . 0 g/L dextrose , pH 7 . 0–7 . 5 ) for S2–S7 egg capsules , or 1x Montjuic water for S8 hatchlings . Yolk ( Y ) samples were obtained from 8 d egg capsules that contained neither spherical nor elongating embryos . Single embryos were imaged on a Leica M205 FA dissecting microscope , transferred into microfuge tubes containing 200 µl TRIzol reagent ( Thermo Fisher , item #15596–018 ) , homogenized by pipetting , and stored at −80°C . Single animal samples of intact C4 adults and virgin , sexually mature adults were homogenized in 1 . 0 mL TRIzol using an IKA Ultra Turrax T 25 Basic tissue disruptor prior to storage at −80°C . Total RNA extraction was performed in 1 . 0 mL TRIzol per sample according to the manufacturer’s protocol , following the recommendations for working with small amounts of tissue . Pellets were resuspended in 25 µl nuclease free water , and 5 µl aliquots were reserved for quality control testing . Total RNA concentration and integrity were determined using Agilent Bioanalyzer 2100 Expert Total RNA Nano or Pico chips ( Agilent Technologies , items # 5067–1511 and 5067–1513 ) . Total RNA samples were prepared for ten biological replicates per time point , and total RNA quality and yield were considered along with embryo size and morphology when selecting samples for library construction . PolyA-selected , single-stranded RNA-Seq libraries were prepared for four biological replicates per stage using the Illumina TruSeq RNA Sample V2 kit ( item # RS-122–2001 and RS-122–2002 ) , starting with 500 ng total RNA per sample ( C4 , virgin sexual adult [SX] , Y , S4 , S5 , S6 , S7 , S8 ) , or 100 ng total RNA per sample ( S2 , S3 ) . Library concentrations and insert sizes were determined using Agilent Bioanalyzer DNA 1000 chips ( Agilent Technologies , item # 5067–1504 ) , and diluted , pooled samples were reanalyzed with Agilent Bioanalyzer 2100 DNA High Sensitivity chips ( Agilent Technologies , item # 5067–4626 ) . Nine barcoded samples , one replicate per time point ( S2–S8 , Y , C4 ) were pooled and sequenced per flow cell lane . Single end , 50 bp reads were acquired on an Illumina Hi-Seq 2000 sequencer . Illumina Primary Analysis version RTA 1 . 13 . 48 . 0 and Secondary Analysis version CASAVA-1 . 8 . 2 were run to demultiplex reads and generate FASTQ files . Barcoded SX replicates were pooled and run on a single flow cell lane of a HiSeq 2500 , and Illumina Primary Analysis version RTA 1 . 17 . 21 . 3 and Secondary Analysis version CASAVA-1 . 8 . 2 were used . The RNA-Seq data have been deposited in the GEO database under the accession number GSE82280 . Sequencing reads were mapped to the smed_20140614 reference transcriptome ( n = 36 , 035 transcripts ) ( Tu et al . , 2015 ) , which contains sequencing data from de novo Trinity assemblies from the C4 and sexual biotypes , staged embryo collections , sorted cycling ( X1 ) cells , and previously published sources ( Adler et al . [2014] , [Böser et al . [2013]; the Dresden transcript collection at PlanMine [http://planmine . mpi-cbg . de] ) . Transcripts were consolidated and reduced to a unique set using the CD-HIT program ( Fu et al . , 2012 ) . smed_20140614 sequences may be downloaded from http://smedgd . stowers . org . Reads were mapped using the Bowtie algorithm , Version 1 . 0 . 0 ( Langmead et al . , 2009 ) , allowing for two mismatches and up to five multi-matches ( --best --strata -v 2 m 5 ) . Read counts for transcripts were tabulated from SAM files using a custom script . Of 36 , 035 transcripts , 32 , 000 accumulated ≥1 CPM across all 40 samples . Samples were each sequenced to an average depth of 19 million reads , and exhibited an average map rate of 89% to the transcriptome . RPKM ( Reads Per Kilobase per Million ) values were scaled using TMM normalization ( Robinson et al . , 2010 ) in edgeR to account for read depth across samples . In addition , 16s ribosomal RNA transcripts ( SMED30032887 ) , which soak up a significant but variable fraction of reads per sample , were removed prior to calculating RPKM values . Differential gene expression was evaluated using the edgeR library ( Robinson et al . , 2010 ) , and adjusted p-values were calculated as described in Hochberg ( 1995 ) . Pairwise comparisons were performed between adjacent time points using edgeR: Y vs S2 , S2 vs S3 , S3 vs S4 , S4 vs S5 , S5 vs S6 , S6 vs S7 and S7 vs S8 . Mapped data were filtered to remove transcripts with less than a sum of 1 CPM across all 32 samples , resulting in 30 , 766 transcripts . The maximum read sum across samples for omitted transcripts was 14 . In addition , transcripts for the 28S ( SMED30027845 ) , 18S ( SMED30032663 ) and 16S ( SMED30032887 ) ribosomal subunits were removed . Differentially expressed genes were identified in mixed stage reference comparisons using the GLM approach in edgeR to contrast each treatment group ( i . e . , developmental stage ) to the average of the remaining groups ( Y , S2–S8 ) . Non-redundant lists of enriched transcripts from the pairwise and mixed stage reference comparisons , for S2 through S8 , were subject to Euclidean distance clustering using scaled RPKM data in edgeR ( Figure 1C–I , Figure 1—figure supplements 4–10 , Figure 1—source data 2–8 ) . Gene Ontology ( GO ) terms ( Gene Ontology Consortium , 2015 ) were assigned to smed_20140614 transcripts on the basis of homologous PFAM domains ( Finn et al . , 2014 ) and significant Swiss-Prot hits ( E-value ≤ 0 . 001 ) , ( UniProt Consortium , 2015 ) . GO term enrichment queries were performed using the R software package topGO , version 2 . 20 . 0 ( Alexa and Rahnenfuhrer , 2010 ) . GO analysis was performed on the non-redundant lists of enriched transcripts for S2–S8 ( Figure 1—source data 1 ) . Categories containing similar and/or related Biological Process ( BP ) GO ids enriched at one or more time point ( s ) were generated manually ( Figure 1—source data 1 ) . Enriched BP GO ids selected for categorization had Benjamini-Hochberg corrected p-values ≤1e-10 ( Benjamini and Hochberg , 1995 ) , and must have been associated with ≥1% of the enriched transcripts for the developmental stage ( s ) in question . BP GO ids were only assigned to one category . BP GO ids that did not describe a cell and/or tissue type present in Smed ( e . g . , heart , lung , neural crest ) were omitted . Using these categories as a guide , lists of enriched BP GO ids and non-redundant lists of associated transcripts were generated for S2–S8 ( Figure 1—source data 2–8 ) . Transcripts may appear in more than one BP GO id category , just as transcripts may be associated with more than one GO term . The neoblast enriched transcript list ( n = 242 ) emerged from the downregulated sequences in whole animals at 24 and/or 48 hr post-lethal irradiation in three independent experiments ( Duncan et al . , 2015; Wagner et al . , 2012 ) ( Chen and Sánchez Alvarado , personal communication ) ( Figure 4A , Figure 4—source data 1 ) . Euclidean distance clustering was performed using the mixed stage reference comparison data . Constructs for riboprobe synthesis were constructed using the pPR-T4P ( J . Rink ) cloning strategy described in Adler et al . ( 2014 ) , with the exception that PCR inserts were amplified using mixed stage embryo cDNA ( S2–S8 ) as a template . Primers used for cloning EEE transcripts and insert sequences appear in Figure 5—source data 3 . Colorimetric and fluorescent WISH was performed as described by King and Newmark ( 2013 ) and Pearson et al . ( 2009 ) , with the following modifications: Immunostaining was performed after fluorescent WISH development with rabbit polyclonal antibodies against H3S10p ( 1:1000; Millipore # 06–570 ) , mouse monoclonal antibodies against Smith Antigen ( Y12 ) ( 1:200 , ThermoFisher Scientific , PIMA190490 ) , or mouse monoclonal antibodies against Smed PIWI-1 ( 1:1000 , a generous gift from J . Rink ) . H3S10p antibodies were detected using preabsorbed Alexa-conjugated secondary antibodies ( 1:1000 , Abcam , ab150086 , ab150069 , ab150071 ) , while anti-Y12 and anti-PIWI staining was visualized with tyramide development using Goat anti-mouse IgG F ( ab' ) 2 HRP ( 1:1000 , Jackson Immunoresearch #115-036-072 ) . Nuclear staining was performed with DAPI ( 1:5000 , 1 mg/mL stock solution , ThermoFisher Scientific , D1306 ) or with Sytox Green ( 1:5000 , 5 mM stock solution , ThermoFisher Scientific , S7020 ) . S2-–S8 colorimetric and fluorescent WISH samples that were to be imaged using light sheet microscopy were mounted as described in the Microscopy section , whereas others were mounted in 80% glycerol supplemented with 2 . 5% DABCO . C4 and sexual adult samples were mounted in Scale A2 mounting media ( Hama et al . , 2011 ) . Embryos ( S2–S8 ) were fixed for 4 hr at room temperature in 4% formaldehyde in 1x PBS , followed by 3 × 10 min washes in 1x PBS and gradual dehydration in 30% , 50% , 70% , 80% , 95% and 100% ethanol . The samples were soaked for 30 min in 5% glycerol diluted in 100% ethanol , cleared in xylene for 10 min , then soaked in two changes of Clear-rite 3 ( Richard-Allan Scientific ) for a total of 25 min . Paraffin infiltration proceeded with 2 × 45 min incubations , and embedded embryos underwent serial sectioning ( 5 µm thickness ) . Paraffin was removed prior to staining by heating slides at 60°C for 20 min , then performing 3 × 2 min washes in xylene , 3 × 1 min washes in 100% ethanol , 3 × 1 min washes in 80% ethanol before rinsing in tap water . Hematoxylin and eosin staining was performed using the ST Infinity H and E Staining System ( Leica Biosystems ) in a Leica Autostainer . Slides were incubated for 30 s in Hemalast , then for 2 min in hematoxylin , and were rinsed for 2 min in tap water . Next , slides were incubated for 45 s in differentiator and for 1 min in bluing agent , with each step followed by a 1 min tap water rinse and a 1 min incubation in 80% ethanol . Slides were stained with eosin for 30 s , dehydrated 3 × 1 min in 100% ethanol and cleared in 3 × 1 min incubations in xylene . A Leica M205 FA stereomicroscope was used to capture images of live animals and colorimetric WISH samples . A Leica DM600B upright microscope was used to capture images of histological sections . A Zeiss LSM-510-VIS confocal and a customized light sheet microscope were used to capture Z-stacks for fluorescent WISH samples . Fixed , stained Smed embryos were mounted in 1% low melt agarose in 1x PBS along with fluorescent conjugated beads required for image registration and reconstruction ( FluoSpheres Polystyrene Microspheres , 1 . 0 µm , red fluorescent [580/605] , Invitrogen/Molecular Probes , F13083; FluoSpheres Carboxylate Modified Microspheres , 0 . 1 µm , yellow-green fluorescent [505/515] , Invitrogen/Molecular Probes , F8803 ) . 1 µM fluorescent bead stock solutions were diluted 1:10 , 000–1: 360 , 000 , depending on the size of the embryo and the magnification of the detection objective used . Samples were placed in an imaging chamber within a Single Plane Illumination Microscopy ( SPIM ) system described in Nakajima et al . ( 2013 ) . S3–S5 embryos were imaged using either a 10x Plan Apochromat or a 5x Plan NeoFluar objective . Z-stacks were taken every 45˚ around the surface of the samples using a rotating stage , producing eight stacks of images per embryo . Multiview data sets were reconstructed using Fiji SPIM plugins for data registration and fusion ( Preibisch et al . , 2010 ) . Reconstructed data sets were viewed in the Imaris software package , where they were cropped and masked to remove beads . Cell positions and the embryonic pharynx were marked manually using the 3D Spot Finder function , and the three-dimensional coordinates for marked cells were exported into excel for analysis of cell positions . Colocalization was determined manually on S3–S4 whole embryos , or on crop3D sections ( 100 µm X 200 µm X 100 µm ) of S5 embryos . Three-dimensional coordinates for piwi-1+ cells or mitotic cells ( piwi-1+ , H3S10p+ ) , the embryonic pharynx , and the embryo center were exported from SPIM reconstructions of S3 and S4 embryos . Spot positions collected in IMARIS were shifted and rotated in MATLAB to a coordinate system where the embryonic center was at the origin and the embryonic pharynx on the z-axis . Relative theta distribution likelihoods of the form ( 1-exp ( -θ/θ’ ) ) *sin ( θ ) were calculated by assuming a θ’ and calculating the likelihood that such a distribution would produce the observed data by multiplying together the individual probabilities . The uncertainty in the best fit θ’’ was found by simulating several datasets , each with a number of cells equal to those observed in the actual data and having a dampening term of θ’’ . For each dataset , the most likely θ’ was found , the standard deviation of which was taken to be the error in θ’’ . Total RNA samples from single embryos and adults ( S2–S8 , Y , C4 and SX; four biological replicates per sample ) were prepared as described for single-animal RNA-Seq . 100 ng total RNA per sample was assayed on the Nanostring nCounter platform ( Geiss et al . , 2008 ) using a custom-made probe set . Reporter and capture probe sequences can be found in Figure 5—source data 2 . Housekeeping control genes were reported in Wenemoser et al . ( 2012 ) . Nanostring data was normalized using the NanoStringNorm library from Bioconductor ( Waggott et al . , 2012 ) , using the sum of the positive controls and the sum of the housekeeping genes as independent normalization factors , with the mean of the negative controls used to estimate background . Raw and normalized nanostring data can be found in Figure 5—source data 2 . Host animals ( 5–7 mm in length , ≥7 day starved ) were selected from the clonally derived Smed sexual strain S2F1L3F2 ( Wagner et al . , 2011 ) and were cultured in 1x Montjuic water supplemented with 100 µg/mL gentamicin sulfate . The host neoblast population was ablated by exposure to 6 , 000 Rads on a GammaCell 40 Exactor irradiator; cohorts of unirradiated animals were reserved to verify complete elimination of the neoblast population by WISH with riboprobes against piwi-1 . Whole-embryo cell suspensions were created for S4 , S5 , S6 , S7 or S8 by mechanically disrupting embryos in chilled , freshly made 1x Holfreter’s solution +5% heat-inactivated fetal bovine serum ( Sigma Aldrich , F4135 ) via repeated pipetting . Ten embryos were disrupted per cell suspension for S5 , S6 , S7 and S8 , whereas 20–35 embryos were disrupted per S4 cell suspension . S4 , S5 and S6 embryos were typically ‘eviscerated’ prior to pipetting by poking with an insect pin and gently squeezing out ingested yolk from the gut . Cell suspension volumes were adjusted to 1 . 0 mL before samples were filtered through 20 µm ( all stages ) and 10 µm ( S5–S8 ) cell strainers ( Partec CellTrics , Sysmex , 04-0042-2315 [20 µm] , 04-0042-2314 [10 µm] ) into low-retention microfuge tubes . Cells were pelleted by centrifugation at 310 rcf for 5 min , were resuspended in a final volume of ~10 µL ( S5–S8 ) or ~5 µl ( S4 ) , and were kept on ice during transplantation . Embryo cell suspensions were loaded by mouth pipetting into borosilicate glass needles ( Sutter Instrument Co . , #B100-75-15 ) pulled using a flaming/brown micropipette puller ( Sutter Instrument Co . , Model P-97 ) and were injected using an Eppendorf FemtoJet at 1 . 0–1 . 5 psi , as described by Wagner et al . ( 2011 ) . Hosts were immobilized on a cold peltier plate , ventral side up , and cells were injected into the tail stripe ( i . e . , the medial , post-pharyngeal parenchymal space between the two posterior branches of the intestine ) . Hosts were injected at 1 day post-irradiation ( dpi ) for rescue experiments and at 3 dpi for short-term experiments . Transplanted animals and uninjected , 6 , 000-Rad-irradiated hosts for rescue experiments were maintained individually in 3 cm petri dishes at 20°C in the dark , with water exchanges and visual inspection of animals performed every 2–3 days . Transplanted animals slated for fixation were reared in 10 cm petri dishes , with 10 or fewer animals per dish , with water exchanges every 2–3 days . Original data underlying this manuscript can be accessed from the Stowers Original Data Repository at http://www . stowers . org/research/publications/libpb-1086 | Flatworms are masters of regeneration . If virtually any piece of a flatworm is cut off , a new fully functional individual will grow from it within two weeks . This is no simple task since flatworms contain a wide variety of organ systems , including a brain , nervous system , eyes , kidneys , gut , muscle and skin . Flatworms owe their regenerative abilities to adult stem cells called neoblasts . Like embryonic stem cells , neoblasts can replicate themselves and they can develop into any type of cell found in an adult worm . In contrast , adult stem cells in fruit flies , zebrafish , mice and humans can only produce the type of cells found in the organ or tissue they live in . Now , Davies et al . have tracked how and when neoblasts develop in embryos of the flatworm species Schmidtea mediterranea by documenting the distinct gene expression signatures in flatworm embryos at various stages of development . An atlas of the genes that are expressed in various embryonic tissues and in major organs as they begin to develop was also created . These tools , and the results of cell transplantation experiments , revealed that neoblasts emerge from embryonic stem cells as the major organs start to form . As the emerging neoblasts start to express the same combination of genes as adult neoblasts , they also begin to behave just like these cells . The populations of neoblasts remain present throughout the life of the flatworm , helping to maintain , repair and regenerate tissues . In the future , work that builds on the results presented here by Davies et al . will help researchers to understand more about how stem cells are maintained and regulated . By learning more about the genetic differences between neoblasts and human adult stem cells scientists may be able to explain why humans and other mammals have a limited ability to regenerate . This information could potentially help to develop treatments that stimulate regeneration in patients with degenerative diseases or traumatic injuries . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"stem",
"cells",
"and",
"regenerative",
"medicine"
] | 2017 | Embryonic origin of adult stem cells required for tissue homeostasis and regeneration |
The evolutionary origins of the hypoxia-sensitive cells that trigger amniote respiratory reflexes – carotid body glomus cells , and ‘pulmonary neuroendocrine cells’ ( PNECs ) - are obscure . Homology has been proposed between glomus cells , which are neural crest-derived , and the hypoxia-sensitive ‘neuroepithelial cells’ ( NECs ) of fish gills , whose embryonic origin is unknown . NECs have also been likened to PNECs , which differentiate in situ within lung airway epithelia . Using genetic lineage-tracing and neural crest-deficient mutants in zebrafish , and physical fate-mapping in frog and lamprey , we find that NECs are not neural crest-derived , but endoderm-derived , like PNECs , whose endodermal origin we confirm . We discover neural crest-derived catecholaminergic cells associated with zebrafish pharyngeal arch blood vessels , and propose a new model for amniote hypoxia-sensitive cell evolution: endoderm-derived NECs were retained as PNECs , while the carotid body evolved via the aggregation of neural crest-derived catecholaminergic ( chromaffin ) cells already associated with blood vessels in anamniote pharyngeal arches .
During hypoxia in vertebrates , respiratory reflexes such as hyperventilation are triggered by neurotransmitter release from hypoxia-sensitive serotonergic cells associated with pharyngeal arch arteries , as well as in the lungs and/or gills ( reviewed by López-Barneo et al . , 2016; Cutz et al . , 2013; Jonz et al . , 2016 ) . In amniotes , these are the ‘glomus cells’ of the carotid body , located at the bifurcation of the common carotid artery ( reviewed by Nurse , 2014; López-Barneo et al . , 2016 ) , and the ‘pulmonary neuroendocrine cells’ ( PNECs ) of lung airway epithelia , found either as solitary , flask-shaped cells , or collected into ‘neuroepithelial bodies’ ( NEBs ) , preferentially located at airway branch points ( reviewed by Cutz et al . , 2013 ) . Glomus cells respond to hypoxia in arterial blood by releasing stored neurotransmitters including acetylcholine , ATP , the catecholamine dopamine , and serotonin ( the latter two most likely acting as autocrine/paracrine neuromodulators ) ( reviewed by Nurse and Piskuric , 2013; Nurse , 2014 ) . These excite afferent terminals of the carotid sinus nerve ( a branch of the glossopharyngeal nerve , arising from neurons in the petrosal ganglion ) in mammals ( reviewed by Nurse and Piskuric , 2013; Nurse , 2014 ) , and of the vagal nerve ( arising from neurons in the nodose ganglion ) in birds ( Kameda , 2002 ) . The afferent nerves relay signals to the nucleus of the solitary tract within the hindbrain , to elicit respiratory reflex responses such as hyperventilation ( reviewed by Teppema and Dahan , 2010 ) . PNECs respond to hypoxia by releasing stored serotonin and various neuropeptides onto vagal afferents , and are thought to act as hypoxia-sensitive airway sensors ( reviewed by Cutz et al . , 2013; also see Branchfield et al . , 2016 ) . PNECs also provide an important stem-cell niche for regenerating the airway epithelium after injury ( Reynolds et al . , 2000; Guha et al . , 2012; Song et al . , 2012 ) and were recently shown to be the predominant cells of origin for small cell lung cancer ( Park et al . , 2011; Sutherland et al . , 2011; Song et al . , 2012 ) . The evolution of glomus cells was critical for the transition from aquatic life , where externally facing hypoxia-sensors are essential for monitoring the variable oxygen levels in water , to fully terrestrial life , where reflex responses to variations in internal oxygen levels are more important , given the stability of oxygen levels in air ( Burleson and Milsom , 2003; Milsom and Burleson , 2007 ) . However , the evolutionary history of glomus cells - and , indeed , PNECs - is uncertain . One commonly suggested hypothesis ( e . g . , Milsom and Burleson , 2007; Hempleman and Warburton , 2013; Jonz et al . , 2016 ) is that glomus cells are homologous to the chemosensory ‘neuroepithelial cells’ ( NECs ) of fish gills . These were originally identified within the primary epithelium of the gills in various teleosts and a shark , as innervated cells ( isolated or clustered ) containing dense-cored vesicles; formaldehyde-induced fluorescence revealed the presence of biogenic amines , identified as serotonin , while electron microscopy following exposure of trout to acute hypoxia revealed fewer vesicles , which appeared degranulated ( Dunel-Erb et al . , 1982 ) . In vitro patch-clamp studies on NECs isolated from zebrafish and catfish gills ( identified by neutral red , a vital dye that stains monoamine-containing cells including serotonergic PNECs; Youngson et al . , 1993 ) , confirmed that teleost gill NECs are hypoxia-sensitive ( Jonz et al . , 2004; Burleson et al . , 2006 ) . The conventional marker for teleost gill NECs is serotonin: although non-serotonergic gill NECs were identified in adult zebrafish by immunoreactivity for synaptic vesicle glycoprotein 2 , these may represent immature NECs ( Jonz and Nurse , 2003; Jonz et al . , 2004 ) . NECs are found near efferent gill arteries and on the basal lamina of the gill epithelia , facing the flow of water , hence can detect hypoxia and other stimuli in either blood or external water ( reviewed by Jonz et al . , 2016 ) . Like glomus cells ( reviewed by Nurse , 2014; López-Barneo et al . , 2016 ) , zebrafish gill NECs also respond to acid hypercapnia ( increased CO2/H+ ) ( López-Barneo et al . , 1988; Buckler , 1997; Jonz et al . , 2004; Qin et al . , 2010 ) . The putative shared evolutionary ancestry of glomus cells and gill NECs ( e . g . Milsom and Burleson , 2007; Hempleman and Warburton , 2013; Jonz et al . , 2016 ) is supported by many similarities: both are associated with pharyngeal arch arteries ( the carotid body develops in association with the third pharyngeal arch artery , which will form the carotid artery ) , provided with afferent innervation by glossopharyngeal and/or vagal nerves , and have background ( 'leak' ) K+ currents that are inhibited by hypoxia , resulting in membrane depolarization , activation of voltage-gated Ca2+ channels , and neurotransmitter release ( López-Barneo et al . , 1988; Buckler , 1997; Jonz et al . , 2004; Qin et al . , 2010 ) . If glomus cells and gill NECs evolved from the same ancestral cell population , they should share a common embryonic origin . Glomus cells are neural crest-derived , as demonstrated in birds by quail-chick neural fold grafts ( Le Douarin et al . , 1972; Pearse et al . , 1973 ) , and in mouse by Wnt1-Cre genetic lineage-tracing ( Pardal et al . , 2007 ) , but the embryonic origin of NECs is unknown . Lack of immunoreactivity for the HNK1 antibody , which labels migrating neural crest cells in many but not all vertebrates , has been reported for NECs ( Porteus et al . , 2013 , 2014 ) . However , the carbohydrate epitope recognized by the HNK1 antibody ( Voshol et al . , 1996 ) is borne by multiple glycoproteins and glycolipids , and gene or antigen expression in itself cannot indicate lineage . An alternative to the hypothesis that gill NECs and glomus cells evolved from a common ancestral cell population is that NECs share ancestry with PNECs , to which they were originally likened ( Dunel-Erb et al . , 1982 ) . Hypoxia-sensing by PNECs , as in NECs and glomus cells , involves inhibition of a K+ current by hypoxia ( reviewed by Cutz et al . , 2013; Nurse , 2014; López-Barneo et al . , 2016; Jonz et al . , 2016 ) . In contrast to the neural crest origin of glomus cells ( Le Douarin et al . , 1972; Pearse et al . , 1973; Pardal et al . , 2007 ) , PNECs have an intrinsic pulmonary epithelial origin: the first experimental support for this was provided by a tritiated thymidine labeling study of hamster lung development ( Hoyt et al . , 1990 ) , later confirmed by mouse genetic lineage-tracing studies using Id2-CreERT2 , Shh-Cre , Nkx2 . 1-Cre , and Sox9-Cre driver lines that showed a common origin for all lung airway epithelial cell types , including PNECs ( Rawlins et al . , 2009; Song et al . , 2012; Kuo and Krasnow , 2015 ) . Here , we demonstrate that neural crest cells do not contribute to gill NECs: instead , these are endoderm-derived . This refutes the hypothesis that glomus cells and gill NECs evolved from a common ancestral cell population , and instead supports an evolutionary relationship between NECs and PNECs , whose endodermal origin we confirm in mouse . We also show that the transcription factor Phox2b , which is required for glomus cell development ( Dauger et al . , 2003 ) , is not expressed by gill NECs ( or by PNECs ) , arguing against the possibility of cell-type homology between glomus cells and gill NECs via activation of the same genetic network . Finally , we report the discovery of neural crest-derived chromaffin ( catecholaminergic ) cells associated with blood vessels in the pharyngeal arches of juvenile zebrafish , which we speculate could share an evolutionary ancestry with glomus cells . Given these results , we propose a new model for the evolution of hypoxia-sensitive cells during the transition to terrestrial life .
In developing zebrafish , gill NECs were previously identified as serotonin ( 5-HT ) -immunoreactive cells in gill filaments from 5-dpf , which are innervated by 7-dpf ( Jonz and Nurse , 2005 ) . We investigated any neural crest contribution to gill NECs via genetic lineage-tracing , using a collection of transgenic zebrafish lines with different cis-regulatory sequences driving Cre and subsequent lineage reporter expression in neural crest-derived cells , as well as via neural crest-deficient zebrafish embryos . Lineage-tracing using different Cre driver and reporter lines enabled us to control for false negatives potentially arising from variable promoter activity in , or incomplete labeling of , some neural crest cells in either neural crest driver or reporter lines . In larval and juvenile zebrafish , we identified NECs in the gill filaments by serotonin immunoreactivity , as previously reported ( Jonz and Nurse , 2005 ) ; we also found similar innervated serotonergic cells scattered in the orobranchial epithelium ( putative NECs ) . NECs in the gill filaments , and putative NECs in the orobranchial epithelium , were unlabeled in larvae/metamorphic juveniles with sox10 cis-regulatory sequences driving Cre and resulting lineage reporter expression [Tg ( -28 . 5sox10:cre ) ;Tg ( ef1a:loxP-DsRed-loxP-EGFP ) ( Kague et al . , 2012 ) ; Tg ( -4 . 9sox10:creERT2 ) ;Tg ( βactin:loxP-SuperStop-loxP-DsRed ) ( Mongera et al . , 2013 ) ] , even when nearby neural crest derivatives such as branchial arch cartilages/mesenchyme and/or gill pillar cells ( Mongera et al . , 2013 ) were reporter-positive ( Figure 1a–f'; 183 serotonergic cells located near such reporter-positive cells [≥6 per fish] were counted across 11 larvae/metamorphic juveniles: n = 8 for −28 . 5sox10; n = 3 for −4 . 9sox10 ) . Gill NECs , and putative NECs in the orobranchial epithelium , were also unlabeled in Tg ( crestin:creERT2 ) ;Tg ( -3 . 5ubi:loxP-GFP-loxP-mCherry ) ( Mosimann et al . , 2011; Kaufman et al . , 2016 ) larvae/metamorphic juveniles , even when nearby neural crest-derived cells in branchial arch cartilages and/or gill filament mesenchyme were mCherry-positive ( Figure 1g-h'; 166 serotonergic cells located near such mCherry-positive cells [≥6 per fish] were counted across 10 larvae/metamorphic juveniles ) . We ruled out the possibility that lack of lineage reporter expression in NECs was a false-negative result arising from inactivity in NECs of the promoters driving the Cre-switchable reporter cassettes , by confirming that NECs expressed the native unrecombined reporter in both transgenic lines [Tg ( -28 . 5sox10:cre ) ;Tg ( ef1a:loxP-DsRed-loxP-EGFP ) and Tg ( crestin:creERT2 ) ;Tg ( -3 . 5ubi:loxP-GFP-loxP-mCherry ) ] ( Figure 1—figure supplement 1 ) . 10 . 7554/eLife . 21231 . 003Figure 1 . Zebrafish NECs are not neural crest-derived: genetic lineage-tracing data . ( a ) Schematic of a 7–8 dpf zebrafish; dotted line indicates section plane in b-f’ . ( b ) Hematoxylin and eosin staining at 7-dpf reveals gill filaments branching from branchial arch cartilages , and the orobranchial cavity . Dashed box indicates approximate region in c , g . ( c–d’ ) In 7-dpf Tg ( -28 . 5sox10:cre ) ;Tg ( ef1a:loxP-DsRed-loxP-EGFP ) zebrafish , GFP labels neural crest-derived branchial arch cartilage and mesenchyme , but not NECs in the gill filaments ( identified by immunoreactivity for serotonin , 5-HT; arrowheads ) . ( e–f’ ) Horizontal section through the gills of a 20-dpf Tg ( -4 . 9sox10:creERT2 ) ;Tg ( βactin:loxP-SuperStop-loxP-DsRed ) zebrafish . DsRed ( brown precipitate ) labels neural crest-derived gill pillar cells , but not NECs ( arrowheads; inverted fluorescent image overlaid on bright-field image ) . ( g–h’ ) In 8-dpf Tg ( crestin:creERT2 ) ;Tg ( -3 . 5ubi:loxP-GFP-loxP-mCherry ) zebrafish , mCherry labels gill pillar cells but not NECs ( arrowheads ) . 5-HT , serotonin; Bac , branchial arch cartilage; Gf , gill filament; La , lamellae; Nt , neural tube; Obc , orobranchial cavity . Scale-bars: 50 μm in b , c , e , g; 25 μm in d , f , h . DOI: http://dx . doi . org/10 . 7554/eLife . 21231 . 00310 . 7554/eLife . 21231 . 004Figure 1—figure supplement 1 . The promoters driving the Cre-switchable reporter cassettes are active in NECs . In 7-dpf Tg ( -28 . 5sox10:cre ) ;Tg ( ef1a:loxP-DsRed-loxP-EGFP ) ( a–a’’’ ) and 25-dpf Tg ( crestin:creERT2 ) ;Tg ( -3 . 5ubi:loxP-GFP-loxP-mCherry ) ( b–b’’’ ) zebrafish , NECs in the gill filaments ( identified by immunoreactivity for serotonin , 5-HT; arrowheads ) express the un-switched reporter gene . 5-HT , serotonin; Bac , branchial arch cartilage; Gf , gill filament . Scale-bar: 25 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 21231 . 004 Finally , we analyzed tfap2amob;foxd3mos zebrafish , which lack neural crest derivatives ( Barrallo-Gimeno et al . , 2004; Montero-Balaguer et al . , 2006; Wang et al . , 2011 ) . At 7-dpf , tfap2amob;foxd3mos mutants lacked pigment cells and lower jaw structures ( n = 8; Figure 2a ) . In the absence of the neural crest-derived pharyngeal endoskeleton , pharyngeal arches and gills are hard to recognize , but putative NECs , visualized here as serotonergic cells associated with HNK1 epitope-immunoreactive neurites ( Metcalfe et al . , 1990 ) , were still present in the orobranchial epithelium ( Figure 2b , c ) ( and occasionally could also be identified ventral to the orobranchial cavity , where the pharyngeal arches would be located; Figure 2d , e ) . We counted all putative NECs in the orobranchial region of three tfap2amob;foxd3mos larvae ( 294 putative NECs counted in total ) and three wild-type siblings ( 244 putative NECs counted in total ) : there was no change in mean number ( mean/embryo ± s . d . : 81 . 3 ± 24 . 0 for wild-type larvae , n = 3; 98 . 0 ± 49 . 7 for tfap2amob;foxd3mos larvae , n = 3; p=0 . 63 , unpaired two-tailed Student’s t-test ) ( Figure 2f–i ) . 10 . 7554/eLife . 21231 . 005Figure 2 . Zebrafish NECs are not neural crest-derived: analysis of neural crest-deficient zebrafish mutants . ( a ) 7-dpf tfap2amob;foxd3mos zebrafish lack all neural crest derivatives , including melanophores and jaw skeleton ( arrowhead ) . Dotted lines: section planes in b-e . ( b , c ) At 7-dpf , tfap2amob;foxd3mos orobranchial epithelium retains innervated serotonergic ( 5-HT+ ) cells ( putative NECs ) , identified by cytoplasmic serotonin surrounded by a ring of HNK1 epitope-immunoreactive neurites . ( d , e ) At 7-dpf , putative NECs ( arrowheads ) persist in tfap2amob;foxd3mos zebrafish in the region ventral to the orobranchial cavity where the pharyngeal arches would be located in wild-type fish . ( NB The hypothalamus [Hyp] extends caudally beneath the midbrain and rostral hindbrain , and often separates from the overlying brain on sections , as here . ) ( f ) The mean number per 7-dpf larva of putative NECs in the orobranchial epithelium does not differ between tfap2amob;foxd3mos ( 98 . 0 ± 49 . 7 s . d . ; n = 3 ) and wild-type zebrafish ( 81 . 3 ± 24 . 0 s . d . ; n = 3 ) ( p=0 . 63 , unpaired two-tailed Student’s t-test ) . All such cells in the orobranchial epithelium were counted for each embryo . Error bars indicate s . d . ( g–i ) Wild-type sibling at 7-dpf . Dotted line: section plane in h-i . Putative NECs are present in the orobranchial epithelium ( arrowheads; dashed box in i , magnified without DAPI in top right corner ) . 5-HT , serotonin; Gf , gill filament; Hyp , hypothalamus; Nt , neural tube; Obc , orobranchial cavity . Scale-bars: 50 μm in b , d , h; 25 μm in c , e , i . DOI: http://dx . doi . org/10 . 7554/eLife . 21231 . 00510 . 7554/eLife . 21231 . 006Figure 2—figure supplement 1 . Putative NECs in the skin of embryonic zebrafish are not neural crest-derived . ( a–b’ ) Three-dimensional rendering of the eye of a 3-dpf Tg ( crestin:creERT2 ) ;Tg ( -3 . 5ubi:loxP-GFP-loxP-mCherry ) embryo , immunostained in whole-mount for serotonin and mCherry . Serotonergic cells are scattered in the epidermis over the eye , as reported ( Coccimiglio and Jonz , 2012 ) , but none is mCherry-positive , i . e . , neural crest-derived ( arrowheads ) . For associated z-stack movies , see Videos 1 and 2 . Scale-bars: 100 μm in a; 50 μm in b . DOI: http://dx . doi . org/10 . 7554/eLife . 21231 . 006 Putative NECs have also been identified as serotonergic cells in the skin of embryonic zebrafish ( Jonz and Nurse , 2006; Coccimiglio and Jonz , 2012 ) and of adult mangrove killifish , which respire through the skin as well as the gills ( Regan et al . , 2011 ) . Although they have not been shown directly ( e . g . , by patch-clamp experiments ) to be hypoxia-sensitive , the putative NECs in killifish increase in area in response to hypoxia ( Regan et al . , 2011 ) , while in zebrafish , hypoxia decreased or delayed , and hyperoxia accelerated , the normal decline in number of these cells seen with increasing age ( Coccimiglio and Jonz , 2012 ) . In zebrafish , these serotonergic cells are most abundant at 3-dpf , and most evident scattered in the skin over the eyes , yolk-sac and tail ( Coccimiglio and Jonz , 2012 ) . Whole-mount immunostaining of Tg ( crestin:creERT2 ) ;Tg ( -3 . 5ubi:loxP-GFP-loxP-mCherry ) embryos at 3-dpf for serotonin and mCherry revealed the expected pattern of scattered serotonergic cells in the epidermis , but none was mCherry-positive ( i . e . , neural crest-derived ) ( Figure 2—figure supplement 1 shows a sample three-dimensional rendering for the eye; for the associated z-stack movies , see Videos 1 and 2 ) . We quantified this in the eye , where the corneal endothelium is neural crest-derived , as previously reported for chicken ( Noden , 1978; Johnston et al . , 1979 ) , mouse ( with a minor mesodermal contribution also noted; Gage et al . , 2005 ) and Xenopus ( Hu et al . , 2013 ) . At 3-dpf , the zebrafish cornea comprises an outer corneal epithelium , a thin acellular collagenous stroma , and a monolayer of flattened corneal endothelial cells ( Soules and Link , 2005; Zhao et al . , 2006; Akhtar et al . , 2008 ) . Across 13 embryos , we counted 328 serotonergic cells in the epidermis over the eye , all of which were mCherry-negative , and 219 nearby mCherry-positive ( i . e . , neural crest-derived ) corneal endothelial cells . Thus , the putative NECs in the skin of embryonic zebrafish are not neural crest-derived . 10 . 7554/eLife . 21231 . 007Video 1 . Putative NECs in the skin of embryonic zebrafish are not neural crest-derived . Z-stack movie showing a whole-mount view of the eye region of a 3-dpf Tg ( crestin:creERT2 ) ;Tg ( -3 . 5ubi:loxP-GFP-loxP-mCherry ) embryo , immunostained for serotonin and mCherry . Serotonergic cells are scattered in the epidermis over the eye , as reported ( Coccimiglio and Jonz , 2012 ) , but none is mCherry-positive , i . e . , neural crest-derived . Video 2 shows a higher-power movie . A three-dimensional rendering for the eye is shown in Figure 2—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 21231 . 00710 . 7554/eLife . 21231 . 008Video 2 . Putative NECs in the skin of embryonic zebrafish are not neural crest-derived . Z-stack movie at higher powerthan Video 1 , showing a whole-mount view of the eye region of a 3-dpfTg ( crestin:creERT2 ) ;Tg ( -3 . 5ubi:loxP-GFP-loxP-mCherry ) embryo , immunostained for serotonin and mCherry . Serotonergic cells ( arrowheads ) are scattered in the epidermis over the eye , as reported ( Coccimiglio and Jonz , 2012 ) , but none is mCherry-positive , i . e . , neural crest-derived . A 3-dimensional rendering for the eye is shown in Figure 2—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 21231 . 008 Taken together , these results show that in zebrafish ( a ray-finned fish ) , the neural crest does not contribute to gill NECs ( the previously proposed homologs of glomus cells ) , or to putative NECs in the orobranchial epithelium and embryonic epidermis . Innervated serotonergic cells in the internal gills of Xenopus tadpoles are the proposed homologs of teleost gill NECs ( Saltys et al . , 2006 ) . We detected scattered serotonergic cells in gill filaments from stage 43 ( Figure 3a , b ) , and in the orobranchial epithelium from stage 41 ( Figure 3a , c ) . To determine if these putative NECs are neural crest-derived , we unilaterally grafted neural folds from GFP-labeled donors to unlabeled hosts ( Figure 3d ) . GFP-labeled neural crest cells migrated away from the neural tube and contributed to branchial arch cartilages and mesenchyme , but putative NECs in the gill filaments and orobranchial epithelium were GFP-negative ( n = 14; Figure 3e–m’ ) . 10 . 7554/eLife . 21231 . 009Figure 3 . Putative Xenopus NECs are not neural crest-derived . ( a ) Schematic of a stage 45 tadpole ( Nieuwkoop and Faber , 1967 ) . Dotted lines: section planes in b , f–g’ ( magenta , transverse ) and in c , i-j’ , l-m’ ( green , oblique ) . ( b ) Serotonergic cells in gill filament epithelium ( putative NECs ) are first detected at stage 43 ( inset shows higher-power view ) . ( c ) Similar serotonergic cells are scattered in the orobranchial epithelium from stage 41 ( inset shows higher-power view ) . ( d ) Schematic ( modified from Nieuwkoop and Faber , 1967 ) showing neural crest labelling: GFP-donors , created by injecting cyto-GFP mRNA into one cell at the two-cell stage , were grown to stages 13–17 , and neural folds grafted unilaterally to wild-type hosts . For grafted embryos grown to stage 53 , donors were transgenic CMV-GFP embryos . ( e ) At stage 22 , GFP labels neural crest cells migrating towards the branchial arches . ( f–g’ ) At stage 43 , GFP labels branchial arch cartilage and surrounding mesenchyme , but not putative NECs ( arrowheads ) in the orobranchial epithelium . ( h–j’ ) At stage 45 , GFP-positive neural crest cells are visible in the branchial arches ( whole-mount and section from different embryos ) . GFP labels branchial arch cartilage and mesenchyme , but not putative NECs ( arrowheads ) in the gill filaments . ( k–m’ ) At stage 53 ( transgenic CMV-GFP donors; whole-mount and section from different embryos ) , GFP labels branchial arch cartilage and mesenchyme , but not putative NECs ( arrowheads ) in the gill filaments . 5-HT , serotonin; Bac , branchial arch cartilage; Gf , gill filament; Nt , neural tube; Obc , orobranchial cavity; Oe , olfactory epithelium . Scale-bar: 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 21231 . 009 In the sea lamprey , the gills are arranged in pairs within the orobranchial cavity , supported by an interbranchial septum ( Figure 4a–c ) . Serotonergic cells in lamprey gills are proposed to correspond to the gill NECs of jawed fishes ( Barreiro-Iglesias et al . , 2009 ) . We detected putative NECs from embryonic day ( E ) 18 . 5 , in clusters on the medial edges of the gills , intimately associated with HNK1 epitope-immunoreactive neurites ( Figure 4d ) . Scattered serotonergic cells were also found in the epithelium lining the roof and floor of the orobranchial cavity . To determine any neural crest cell contribution , the vital lipophilic dye DiI was injected into E5 vagal neural folds and the embryos followed to E19 . 0 ( Figure 4e–n ) . DiI-labeled neural crest cells migrated into the branchial arches and contributed to the branchial arch basket , as expected ( McCauley and Bronner-Fraser , 2003 ) , but putative NECs in the gills and orobranchial epithelia , despite being near DiI-labeled cells , were unlabeled ( n = 19; Figure 4i , j , n; Figure 4—figure supplement 1 ) . 10 . 7554/eLife . 21231 . 010Figure 4 . Putative lamprey NECs are not neural crest-derived . ( a ) Schematic larval lamprey section ( modified from Barreiro-Iglesias et al . , 2009 ) showing gill pairs in the orobranchial cavity , supported by an interbranchial septum , and a single gill at higher power . ( b ) Schematic Piavis-stage 17 ( E19 ) lamprey ( modified from Tahara , 1988 ) . Dotted line shows section plane in c , d , i , j , n . ( c ) Hematoxylin and eosin staining at E19 shows internal gills as ‘stalks’ within the orobranchial cavity . ( d ) Putative NECs ( serotonergic cells associated with HNK1 epitope-immunoreactive neurites ) are first visible at E18 . 5 in the medial gill epithelium . ( e ) Schematic ( modified from Tahara , 1988 ) showing neural crest labelling by DiI injection at E5 ( Piavis stages 11–12 ) . ( f–n ) Two different embryos ( f–j , k–n ) , showing DiI-labeled neural crest cells migrating ventrally ( arrowheads , g , l ) into the branchial arches , contributing to branchial arch basket and gill supporting cells ( arrowheads , j , n ) , but not serotonergic cells ( putative NECs ) in the orobranchial epithelium ( i , j ) or gills ( n ) . 5-HT , serotonin; Ba , branchial arch; Bab , branchial arch basket; En , endostyle; Gs , gill seam; Is , interbranchial septum; Nc , notochord; Nt , neural tube; Obc , orobranchial cavity . Scale-bar: 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 21231 . 01010 . 7554/eLife . 21231 . 011Figure 4—figure supplement 1 . Confirmation of successful targeting of pharyngeal arch-destined neural crest cells in all lamprey embryos analyzed for neural crest contribution to putative NECs . For orientation , the first three panels are repeated from Figure 4 . ( a ) Schematic larval lamprey section ( modified from Barreiro-Iglesias et al . , 2009 ) showing gill pairs in the orobranchial cavity , supported byan interbranchial septum . ( b ) Schematic Piavis-stage 17 ( E19 ) lamprey ( modified from Tahara , 1988 ) . Dotted line: section plane . ( c ) Hematoxylin and eosin staining shows internal gills as ‘stalks’ within the orobranchial cavity . ( d–v ) Each row shows one of the 19 lamprey embryos with DiI-labeled neural crest derivatives near serotonergic putative NECs at the final analysis . The first panel shows the site of DiI injection at one day post-injection ( dpi ) ( missing for embryo Pm19 ) ; the second and third panels show DiI-labeled neural crest cells in the branchial arches in right and/or left-side views at 9-dpi; the fourth panel shows a view of the internal gills in section at 13-dpi ( for section plane , see dotted line in panel a ) , showing DiI contribution to gill support cells , but not to the putative NECs ( serotonergic cells ) in the gill epithelium . Asterisks indicate the embryos from which data are shown in Figure 4 ( Pm12 and Pm17 ) . 5-HT , serotonin; En , endostyle; Gs , gill seam; Is , interbranchial septum; Nc , notochord; Obc , orobranchial cavity . DOI: http://dx . doi . org/10 . 7554/eLife . 21231 . 011 These results show that putative NECs in the internal gills and orobranchial epithelium of a frog ( i . e . , a lobe-finned tetrapod ) and the sea lamprey ( a jawless fish ) are not neural crest-derived . Overall , our data show that the neural crest does not contribute to gill NECs in zebrafish , or to their presumed homologues in Xenopus and lamprey gills ( or to the putative NECs identified in the orobranchial epithelium of all three species ) . Hence , glomus cells and gill NECs cannot have evolved from the same ancestral cell population . In zebrafish , Xenopus and the little skate ( a cartilaginous fish ) , vital dye fate-mapping experiments have shown that the gills and orobranchial cavity are lined with an epithelium derived mostly from endoderm ( Warga and Nüsslein-Volhard , 1999; Chalmers and Slack , 2000; Gillis and Tidswell , 2017 ) , suggesting endoderm as an alternative origin for NECs . Indeed , in one of the first descriptions of fish gill NECs ( Dunel-Erb et al . , 1982 ) , they were likened to the PNECs of amniotes , which share a common embryonic origin with other airway epithelial cells in rodents ( Hoyt et al . , 1990; Rawlins et al . , 2009; Song et al . , 2012; Kuo and Krasnow , 2015 ) . We demonstrated the endodermal origin of PNECs in the mouse by lineage-tracing using the Sox172A-iCre driver line ( in which all endoderm-derived lineages , as well as vascular endothelial cells and the hematopoietic system , express Cre; Engert et al . , 2009 ) crossed to the R26RlacZ or R26RtdTomato reporter lines ( Soriano , 1999; Madisen et al . , 2010 ) ( Figure 5a–e’’’ ) . We also confirmed the recent exclusion by Wnt1-Cre lineage-tracing ( Danielian et al . , 1998 ) of a neural crest contribution to mouse PNECs ( Kuo and Krasnow , 2015 ) ( Figure 5—figure supplement 1a–b’ ) . Similarly , we saw no neural crest contribution to PNECs in the chicken lung after labeling the premigratory neural crest by neural fold grafting from GFP-transgenic donor embryos ( McGrew et al . , 2008 ) ( Figure 5—figure supplement 1c–e’ ) . 10 . 7554/eLife . 21231 . 012Figure 5 . NECs are endoderm-derived , like PNECs . ( a–d ) In Sox172A-iCre/+;R26R/+ mice , all endoderm-derived lineages , as well as vascular endothelial cells and the hematopoietic system , constitutively express β-galactosidase ( Engert et al . , 2009 ) . Serial sections of an E19 . 5 Sox172A-iCre/+;R26R/+ mouse lung show that X-gal labels PNECs ( a , b; black arrowheads ) , whose identity is confirmed by serotonin expression ( c , d; white arrowheads ) . The serotonin-positive cells are clearly all in the epithelium , which is entirely X-gal-positive , although there is some variation in staining level from cell to cell . ( e–e’’’ ) A high-power view of a cluster of Ascl1-expressing PNECs in a section of an E16 . 5 Sox172A-iCre/+;R26RtdTomato mouse lung , in which endoderm-derived lineages express tdTomato . Only the occasional PNEC is serotonin-positive at this stage . The Ascl1-expressing PNECs are tdTomato-positive , i . e . , endoderm-derived . ( f–i’’ ) An endodermal contribution to putative NECs in Xenopus was investigated by performing focal DiI injections into the anterior endoderm at stage 14 ( f ) , as described in Chalmers and Slack ( 2000 ) . At stage 45 , DiI labels the endoderm lining the orobranchial cavity ( g , h ) , and serotonergic cells ( putative NECs , arrowheads ) in the orobranchial epithelium ( i–i’’ ) . ( j–l ) In Tg ( sox17:creERT2;cmlc2:DsRed ) ;Tg ( -3 . 5ubi:loxP-GFP-loxP-mCherry ) zebrafish , the endoderm is labeled with mCherry and ( m–n’’ ) NECs in the gill filaments are mCherry-positive ( arrowheads ) . 5-HT , serotonin; A , anterior; a , airway; Bac , branchial arch cartilage; Gf , gill filament; Obc , orobranchial cavity; P , posterior; tdTom , tdTomato . Scale-bars: 50 μm in a , c , g , k , m; 25 μm in b , d , h , i , l , n; 20 μm in e . DOI: http://dx . doi . org/10 . 7554/eLife . 21231 . 01210 . 7554/eLife . 21231 . 013Figure 5—figure supplement 1 . The neural crest does not contribute to amniote PNECs . ( a–b’ ) Transverse sections through the lungs of Wnt1-Cre;R26R-YFP mouse embryos , in which neural crest cells are permanently labeled with YFP ( Danielian et al . , 1998; Srinivas et al . , 2001 ) , at E14 . 5 ( a , a’ ) and E18 . 5 ( b , b’ ) . PNECs ( Ascl1 [Mash1]-positive cells in the airway epithelium; Ito et al . , 2000 ) are unlabeled , whether solitary or clustered ( white dotted line in b , b’ ) , although nearby neural crest-derived cells in the subjacent mesenchyme ( yellow arrowheads ) are YFP-positive , including putative Schwann cells on a nerve innervating the PNECs ( yellow arrowheads in b , b’ ) and an intrinsic pulmonary ganglion ( white arrow in b , b’ ) , as expected ( Freem et al . , 2010 ) . ( In a , a’ , the two fainter , out-of-focus green spots within the epithelium are background artefacts from the anti-GFP immunostaining . ) ( c ) The vagal neural crest was labeled in the chicken using GFP-transgenic to wild-type neural tube grafts at E1 . 5 ( schematic modified from Le Douarin , 2004 ) . ( d–e’ ) Transverse sections through the lungs of grafted embryos at E14 . 5 ( d , d’ ) and E16 . 5 ( e , e’ ) . PNECs ( serotonergic cells in the lung airway epithelium ) are unlabeled ( white arrowheads ) , while putative Schwann cells ( elongated cells , yellow arrowheads ) and a nearby intrinsic pulmonary ganglion ( white arrow ) are GFP-positive , as expected ( Burns and Delalande , 2005 ) . 5-HT , serotonin; a , airway; Ot , otic vesicle; s , somite . Scale-bars: 10 μm in a; 50 μm in b , d , e . DOI: http://dx . doi . org/10 . 7554/eLife . 21231 . 013 To test the hypothesis that NECs , like PNECs , are endoderm-derived , we first attempted to label the pharyngeal endoderm of Xenopus embryos at stage 14 via focal DiI injections into anterior endoderm ( Chalmers and Slack , 2000 ) ( Figure 5f–i ) . Only three embryos with endoderm-specific DiI labeling survived to stage 45 for analysis . The DiI labeling was very sparse , but in one embryo , 15 serotonergic cells in the orobranchial epithelium ( putative NECs ) were DiI-labeled , supporting an endodermal origin ( Figure 5g–i’’ ) . Since the direct labeling approach in Xenopus proved to be technically challenging , we used genetic lineage-tracing of endodermal sox17 expression in zebrafish , inducing gastrulation-stage Cre expression and recombination in embryos from crosses between a sox17:creERT2 zebrafish driver line [Tg ( sox17:creERT2;cmlc2:DsRed ) ; Joseph J . Lancman , Keith P . Gates , and P . Duc S . Dong , personal communication , March , 2017] and the switchable reporter line Tg ( -3 . 5ubi:loxP-GFP-loxP-mCherry ) ( Mosimann et al . , 2011 ) ( Figure 5j–n’’ ) . At 8-dpf , mCherry expression was seen in both gill NECs and putative NECs in the orobranchial epithelium: 147/331 serotonergic cells counted across six larvae ( ≥36 cells counted per fish ) were mCherry-positive ( Figure 5n–n” ) . [The relatively low labeling efficiency likely results from a lack of optimization of the 4-OHT dose for the Tg ( sox17:creERT2;cmlc2:DsRed ) driver in combination with this particular switchable reporter line ( Mosimann et al . , 2011 ) . ] These data demonstrate an endodermal origin in zebrafish for gill NECs , and also for putative NECs in the orobranchial epithelium . ( Our genetic lineage-tracing data also confirm the endodermal origin of zebrafish gill filament epithelium , previously reported from vital dye fate-mapping experiments; Warga and Nüsslein-Volhard , 1999 . ) Taken together , these results reveal the endodermal origin of gill NECs and putative NECs in the orobranchial epithelium . This supports the shared evolutionary ancestry of NECs with endoderm-derived PNECs , rather than neural crest-derived glomus cells . It is formally possible that , despite their different embryonic origins , glomus cells and NECs could still be homologous cell types through activation of the same genetic network . Although the molecular basis of NEC development has not been investigated , the basic helix-loop-helix transcription factor Ascl1 ( Mash1 ) is required for the formation of both PNECs ( Ito et al . , 2000 ) and glomus cells ( Kameda , 2005 ) . The homeodomain transcription factor Phox2b is also essential for glomus cell development ( Dauger et al . , 2003 ) . However , we were unable to detect Phox2b-positive cells in embryonic zebrafish gills or orobranchial epithelium at 5-dpf ( n = 4 ) or 7-dpf ( n = 3 ) , although Phox2b was expressed by a subset of cells in the hindbrain , as expected ( Coppola et al . , 2012 ) ( Figure 6a–f’ ) . Similarly , we could not detect any Phox2-positive cells in sea lamprey gills or orobranchial epithelia at E16 or E18 ( n = 6 ) , although Phox2 was expressed in the epibranchial ganglia , and in patches of ectoderm and subjacent mesenchyme ventral to the epibranchial ganglia ( Figure 6g–j’ ) , in the same position as the hypobranchial placodes and associated ganglia identified in Xenopus ( Schlosser , 2003 ) . 10 . 7554/eLife . 21231 . 014Figure 6 . Phox2b expression is absent from gill and lung epithelia . ( a–f’ ) In wild-type zebrafish at 5- and 7-dpf , Phox2b is expressed by a subset of cells in the hindbrain ( b , e ) , but not by gill NECs or putative NECs in the orobranchial epithelium ( arrowheads; c–c’ , f–f’ ) . ( g–j’ ) At E16 ( g–h’ ) and E18 ( i–j’ ) in the sea lamprey , Phox2 expression is seen in the neural tube , the epibranchial ( petrosal and nodose ) ganglia ( identified in section by the neuronal marker Elavl3/4 ) , and in patches of ectoderm and subjacent mesenchyme ventral to the epibranchial ganglia ( arrowheads ) . However , Phox2 expression is absent from the gill epithelium , where putative NECS would be located . Dotted lines in panels g and i indicate the section plane in h and j . ( k–l’ ) In a section of an E16 . 5 Sox172A-iCre/+;R26tdTomatomouse lung , Phox2b expression is seen in intrinsic pulmonary ganglia ( arrows ) , but not in Ascl1/serotonin-positive PNECs located in the lung airway epithelium ( dotted lines outline clusters of PNECs ) . ( m , m’ ) In a section of an E13 . 5 chicken lung , Phox2b expression is seen in an intrinsic pulmonary ganglion ( arrow ) , but not in serotonin-positive PNECs scattered in the lung airway epithelium ( arrowheads ) . Insets show higher power views . a , airway; Ba , branchial arch; Bac , branchial arch cartilage; G , gill; Gf , gill filament; N , nodose ganglion; Nt , neural tube; Obc , orobranchial cavity; Oc , oral cavity; P , petrosal ganglion . Scale bars: 50 μm in a , d , h , j , m; 25 μm in b , c , e , f , k , l . DOI: http://dx . doi . org/10 . 7554/eLife . 21231 . 014 We also found that PNECs lack Phox2b expression . In mouse embryos at E16 . 5 , when PNECs can be identified by Ascl1 expression ( serotonin is only expressed in a few PNECs at this stage ) , Phox2b was not seen in PNECs , despite expression in nearby intrinsic pulmonary ganglia ( n = 2; Figure 6k–l’ ) . Similarly , in chicken embryos at E12-E13 . 5 , when scattered PNECs can be identified in the lung epithelium by serotonin immunoreactivity , Phox2b expression was not seen in the lung epithelium either by in situ hybridization or by immunostaining , although it could be detected in nearby intrinsic pulmonary ganglia ( n = 3; Figure 6m , m’ ) . Overall , these data show that NECs ( and PNECs ) do not activate the same genetic network as glomus cells , since they lack expression of a transcription factor , Phox2b , which is essential for glomus cell development ( Dauger et al . , 2003 ) . Hence , NECs and glomus cells cannot be homologous cell types . Since NECs are endoderm-derived , we reasoned that neural crest-derived glomus cells must have evolved independently . Glomus cells are catecholaminergic , like the neural crest-derived chromaffin cells of the adrenal gland , which are also hypoxia-sensitive ( reviewed by López-Barneo et al . , 2016 ) . Intriguingly , in lampreys , catecholaminergic ( chromium salt-staining , i . e . , ‘chromaffin’ ) cells were reported a century ago in association with large blood vessels not only in the trunk , but also as far rostrally as the second branchial arch , ‘in the walls of the segmental veins as these run round the notochord’ ( Giacomini , 1902; Gaskell , 1912 ) . This suggested to us the possibility that glomus cells could have evolved from catecholaminergic cells associated with blood vessels in anamniote pharyngeal arches , if such cells are neural crest-derived . We first used immunostaining for the catecholaminergic marker tyrosine hydroxylase and the neurite-marker acetylated tubulin to confirm the existence of catecholaminergic cells , at least some of which may be innervated , in the walls of the anterior cardinal veins in the gill arches of ammocoete-stage sea lamprey ( Figure 7a–c ) . It would be difficult to test whether the gill arch catecholaminergic cells are neural crest-derived: even if DiI-labeled embryos could be raised to ammocoete stages , the DiI-labeling would likely be very sparse . We reasoned that if present in lamprey , such cells might also be present in the zebrafish . We sectioned metamorphic juveniles and identified similar clusters of catecholaminergic cells in close association with blood vessels in the gill arches ( Figure 7d–e’’’ ) . To our knowledge , this is the first demonstration of the existence of catecholaminergic cells associated with gill arch blood vessels in a jawed anamniote . Importantly , these catecholaminergic cells were mCherry-positive , i . e . , neural crest-derived , in Tg ( crestin:creERT2 ) ;Tg ( -3 . 5ubi:loxP-GFP-loxP-mCherry ) metamorphic juveniles ( Figure 7e–e’’’ ) ( 51 such catecholaminergic cells were mCherry-positive across five juveniles [≥5 counted per fish] ) . This discovery suggests a new , speculative hypothesis for carotid body evolution , namely that it evolved in the amniote lineage via the aggregation of neural crest-derived catecholaminergic cells that were already associated with pharyngeal arch blood vessels in anamniotes . 10 . 7554/eLife . 21231 . 015Figure 7 . Catecholaminergic cells associated with gill arch blood vessels are neural crest-derived in zebrafish . ( a ) Schematic transverse section through ammocoete-stage lamprey gill arch ( modified from Ruppert et al . , 2003 ) . Red box indicates region shown in b . ( b , c ) Tyrosine hydroxylase-positive ( catecholaminergic ) cells are present in the wall of the anterior cardinal vein ( arrowheads ) , closely associated with acetylated tubulin-immunoreactive neurites ( arrow ) . ( d ) Schematic 25-dpf zebrafish . Dotted line indicates transverse section plane through the gill basket in e-e’’’ . ( e–e’’’ ) Tyrosine hydroxylase-positive ( catecholaminergic ) cells ( arrowheads ) seen adjacent to melanocyte-covered gill-filament blood vessels ( dotted lines ) , are neural crest-derived ( mCherry-positive ) in 25-dpf Tg ( crestin:creERT2 ) ;Tg ( -3 . 5ubi:loxP-GFP-loxP-mCherry ) zebrafish . Acv , anterior cardinal vein; Bac , branchial arch cartilage; Bv , blood vessel; Nc , notochord; TH; tyrosine hydroxylase . Scale-bars: 50 μm in b; 25 μm in c , e . DOI: http://dx . doi . org/10 . 7554/eLife . 21231 . 01510 . 7554/eLife . 21231 . 016Figure 8 . Model for the evolution of the hypoxia-sensitive cells involved in amniote respiratory reflexes . ( a ) Schematic ancestral vertebrate with internal gills . Neural crest-derived ( magenta ) chromaffin cells are associated with large blood vessels in the pharyngeal arches , while NECs differentiate within the endoderm-derived ( green ) epithelium lining the gills and orobranchial cavity . ( b ) During the transition to terrestrial life , the glomus cells of the carotid body evolved via the aggregation of neural crest-derived chromaffin cells ( which must also have acquired serotonergic properties ) , while ( c ) NECs persisted as PNECs in lung airway epithelia . Yellow arrows indicate shared ancestry . Bv , blood vessel; NEC , neuroepithelial cell; PNEC , pulmonary neuroendocrine cell . DOI: http://dx . doi . org/10 . 7554/eLife . 21231 . 016
The evolutionary history of the hypoxia-sensitive cells that initiate amniote respiratory reflexes has been obscure , but is critical for our understanding of the transition from aquatic to terrestrial life . This involved a change from aquatic respiration in an environment with low oxygen solubility , to obligate air-breathing in an environment with more stable oxygen levels . It has been proposed that this transition was accompanied by a switch from a dispersed population of externally oriented hypoxia-sensitive cells in the gills that monitored the highly variable external oxygen levels , to one dominant site of hypoxia-sensitive cells that focused on monitoring internal oxygen states ( Burleson and Milsom , 2003; Milsom and Burleson , 2007 ) . Current hypotheses suggest that this change was accompanied by the evolution of the glomus cells of the carotid body from an ancestral population of gill NECs ( e . g . , Milsom and Burleson , 2007; Hempleman and Warburton , 2013; Jonz et al . , 2016 ) . This was entirely plausible , given their common association with pharyngeal arch arteries , afferent innervation by glossopharyngeal and/or vagal nerves , and hypoxia-sensitive K+ currents ( López-Barneo et al . , 1988; Buckler , 1997; Jonz et al . , 2004; Qin et al . , 2010 ) . However , glomus cells are neural crest-derived ( Le Douarin et al . , 1972; Pearse et al . , 1973; Pardal et al . , 2007 ) and the hypothesis is not compatible with our demonstration that neural crest cells do not contribute to gill NECs in zebrafish , or their presumed homologues in Xenopus and lamprey , or to similar innervated serotonergic cells ( putative NECs ) in the orobranchial epithelium of all three anamniote species . In contrast , we found that these cells are endoderm-derived , like PNECs , which differentiate in situ within pulmonary airway epithelia ( Hoyt et al . , 1990; Rawlins et al . , 2009; Song et al . , 2012; Kuo and Krasnow , 2015 ) and whose endodermal origin we demonstrated using Sox17-Cre lineage-tracing in mouse . This suggests that gill NECs , and putative NECs in the orobranchial epithelium if these prove to be hypoxia-sensitive , more likely share a common ancestor with PNECs . Denervation experiments in various fishes ( a shark , as well as some teleosts ) have shown that hypoxia-sensitive and/or CO2-sensitive chemoreceptors involved in ventilation responses are located in the orobranchial cavity , as well as in the gills , although their cellular identity has not been confirmed ( reviewed by Milsom , 2012 ) . Furthermore , putative NECs ( identified by morphology , innervation and serotonin immunoreactivity , although not as yet shown to be hypoxia-sensitive ) have been reported in the epithelium of the air-breathing organs ( where present ) of ray-finned fishes , lobe-finned lungfishes and amphibians ( reviewed by Jonz et al . , 2016; Hsia et al . , 2013 ) . Anamniote air-breathing organs likely evolved from out-pocketings of the caudal orobranchial epithelium after the evolution of gills and NECs ( reviewed by Hsia et al . , 2013 ) . Taken together , this leads us to speculate that hypoxia-sensitive NECs in the epithelia of the gills and orobranchial cavity of ancestral vertebrates were retained in the air-breathing organs of both anamniotes and amniotes . Testing this hypothesis for the evolutionary origin of PNECs must await evidence for the hypoxia-sensitivity of the putative NECs in the orobranchial epithelium of anamniotes . Putative NECs have also been identified via serotonin immunoreactivity in the skin of developing zebrafish ( Jonz and Nurse , 2006; Coccimiglio and Jonz , 2012 ) and of adult mangrove killifish , which respire through the skin as well as the gills ( Regan et al . , 2011 ) . Our genetic lineage-tracing data in zebrafish show that these cells are not neural crest-derived . If these serotonergic cells in the embryonic skin indeed prove to be hypoxia-sensitive NECs , then we suggest that NECs are likely a ‘local epithelial’ rather than uniquely endodermal cell type , i . e . , that they can differentiate within epithelia of either ectodermal or endodermal origin , like taste buds ( Barlow and Northcutt , 1995; Stone et al . , 1995 ) and the ameloblast ( enamel-forming ) layer of teeth ( Soukup et al . , 2008 ) . It remained formally possible that , despite their different embryonic origins , gill NECs and glomus cells could be homologous cell types via activation of the same genetic network . However , we found that the transcription factor Phox2b , which is critical for glomus cell development ( Dauger et al . , 2003 ) , is not expressed by embryonic zebrafish gill NECs , putative NECs in the orobranchial epithelium , or their putative homologues in lamprey embryos . ( We also found that mouse and chicken PNECs lack Phox2b expression . ) Hence , NECs and glomus cells cannot be homologous cell types . Since neural crest-derived glomus cells could not have evolved from endoderm-derived gill NECs , what is their evolutionary history ? Glomus cells are catecholaminergic , and they are strikingly similar to the neural crest-derived chromaffin ( i . e . , catecholaminergic ) cells of the adrenal medulla . For example , in fetal or neonatal mammals , adrenal chromaffin cells release catecholamines in direct response to hypoxia , like glomus cells ( Comline and Silver , 1966; Cheung , 1989 , 1990; Seidler and Slotkin , 1985; Thompson et al . , 1997 ) . This ‘non-neurogenic’ response to hypoxia , in the absence of neural input , facilitates the transition to air-breathing in neonates by stimulating lung fluid absorption and regulating cardiovascular function ( Seidler and Slotkin , 1985; Thompson et al . , 1997 ) . The sensitivity of adrenal chromaffin cells to hypoxia is lost upon postnatal cholinergic innervation of the adrenal gland , although at least some hypoxia-responsive chromaffin cells persist in the adult adrenal medulla ( García-Fernández et al . , 2007; Levitsky and López-Barneo , 2009 ) ( also see López-Barneo et al . , 2016 ) . Furthermore , hypoxia inhibits K+ currents in adrenal chromaffin cells , as in glomus cells ( reviewed by López-Barneo et al . , 2016 ) , and the ‘set point’ of hypoxia sensitivity is controlled in both glomus cells and adrenal chromaffin cells by mutual antagonism between the oxygen-regulated transcription factors hypoxia-inducible factor 1-alpha ( Hif1α/HIF1a ) and hypoxia-inducible factor 2-alpha ( Hif2α/HIF2a ) ( Yuan et al . , 2013 ) . These similarities led us to re-visit century-old reports ( Giacomini , 1902; Gaskell , 1912 ) of chromaffin ( chromium salt-staining , i . e . , catecholaminergic ) cells associated with large branchial arch blood vessels in lamprey . We confirmed the existence of these catecholaminergic cells in ammocoete-stage sea lamprey , and went on to discover catecholaminergic cells associated with pharyngeal arch blood vessels in juvenile zebrafish , whose neural crest origin we demonstrated by genetic lineage-tracing . We speculate that the carotid body may have evolved via the aggregation of such cells , and their subsequent acquisition of serotonergic properties and afferent innervation by glossopharyngeal and/or vagal afferents , such that they became incorporated into the afferent arm of respiratory reflexes . In order to test this hypothesis , it will be necessary to investigate whether these blood vessel-associated catecholaminergic cells in anamniotes secrete catecholamines in response to hypoxia . Glomus cells in the carotid body are enveloped by glial-like sustentacular cells , also neural crest-derived ( Le Douarin et al . , 1972; Pearse et al . , 1973; Pardal et al . , 2007 ) , which have been shown to act as adult stem cells for the production of new glomus cells under hypoxic conditions ( Pardal et al . , 2007 ) . Hypotheses for carotid body evolution must also take these cells into account . The amphibian carotid labyrinth , a maze-like vascular expansion at the bifurcation of the carotid artery , is sensitive to oxygen levels ( Ishii et al . , 1966 ) and considered a carotid body homologue ( reviewed by Kusakabe , 2009 ) . It contains both serotonergic and catecholaminergic cells , innervated by glossopharyngeal and/or vagal nerves ( Reyes et al . , 2014 ) , with both efferent and afferent synapses ( reviewed by Kusakabe , 2009 ) . Furthermore , the glomus cells in the carotid labyrinth are enveloped by fine processes of sustentacular cells ( reviewed by Kusakabe , 2009 ) . In adult amphibians , the carotid labyrinth is hypothesized to replace the NECs of the larval gills as the primary site of the chemosensors responsible for maintaining respiratory homeostasis ( Kusakabe , 2002; Jonz and Nurse , 2006 ) . Direct electrophysiological evidence is lacking for which cells are hypoxia-responsive , however , and their embryonic origin has not been established . Further investigation of the development and physiology of the carotid labyrinth in amphibians , and of the neural crest-derived catecholaminergic cells that we discovered in association with pharyngeal arch blood vessels in zebrafish , should help to test our new hypothesis for carotid body evolution . Given our lineage-tracing data , we present a new model for the evolution of the hypoxia-sensitive cells involved in amniote respiratory reflexes ( Figure 8 ) . We hypothesize that carotid body glomus cells evolved via the aggregation of neural crest-derived catecholaminergic ( chromaffin ) cells that were already associated with blood vessels in anamniote gill arches ( and which must subsequently have acquired serotonergic properties and afferent innervation by glossopharyngeal and/or vagal nerves ) , while NECs differentiating in situ in the endoderm-derived epithelia of the gills and orobranchial cavity were retained as PNECs in lung airway epithelia . This model can be viewed as more parsimonious , since both embryonic lineage ( neural crest versus endoderm ) and function ( physiological versus environmental oxygen monitoring ) are maintained during the proposed evolutionary history of glomus cells and PNECs . Testing the model will require investigation of the physiology and hypoxia-responsiveness of anamniote gill arch blood vessel-associated catecholaminergic cells .
The following zebrafish ( Danio rerio ) lines were used: Tg ( -28 . 5sox10:cre ) ;Tg ( ef1a:loxP-DsRed-loxP-EGFP ) ( Kague et al . , 2012 ) , Tg ( -4 . 9sox10:creERT2 ) ;Tg ( βactin:loxP-SuperStop-loxP-DsRed ) ( Mongera et al . , 2013 ) , Tg ( crestin:creERT2 ) ;Tg ( -3 . 5ubi:loxP-GFP-loxP-mCherry ) ( Mosimann et al . , 2011; Kaufman et al . , 2016 ) , tfap2amob;foxd3mos ( Wang et al . , 2011 ) and Tg ( sox17:creERT2;cmlc2:DsRed ) [created using the sox17 promoter from Mizoguchi et al . ( 2008 ) ; Joseph J . Lancman , Keith P . Gates , and P . Duc S . Dong , personal communication , March , 2017] . Experiments using Tg ( -4 . 9sox10:creERT2 ) ;Tg ( βactin:loxP-SuperStop-loxP-DsRed ) zebrafish were conducted in compliance with the regulations of the Regierungspräsidium Tübingen and the Max Planck Society . Experiments using all other zebrafish lines were conducted according to protocols approved by the Institutional Animal Care and Use Committees in facilities accredited by the Association for Assessment and Accreditation of Laboratory Animal Care International ( AAALAC ) . All zebrafish were fixed overnight at 4°C in 4% paraformaldehyde in phosphate-buffered saline ( PBS ) , except for Tg ( sox17:creERT2;cmlc2:DsRed ) ;Tg ( -3 . 5ubi:loxP-GFP-loxP-mCherry ) zebrafish , which were fixed overnight at 4°C in 4% paraformaldehyde in 0 . 1 M Pipes , 1 mM MgSO4 , 2 mM EGTA , pH 7 ) . To induce Cre activity and recombination in Tg ( -4 . 9sox10:creERT2 ) ;Tg ( βactin:loxP-SuperStop-loxP-DsRed ) zebrafish , embryos were dechorionated at 16 hr post-fertilization ( hpf ) and treated with 5 μM 4-hydroxytamoxifen ( 4-OHT; Sigma-Aldrich , St . Louis , MO ) for 8 hr . As reported in Mongera et al . ( 2013 ) , 4-OHT treatment of this line for 8 hr from 16 hpf is very effective in yielding Cre-induced recombination in the branchial arches , and was used in Mongera et al . ( 2013 ) to demonstrate the neural crest origin of gill pillar cells . To induce Cre activity and recombination in Tg ( crestin:creERT2 ) ;Tg ( -3 . 5ubi:loxP-GFP-loxP-mCherry ) zebrafish , embryos were treated with 20 μM 4-OHT in ethanol at 50% epiboly and again at 24 hpf . To induce Cre activity and recombination in Tg ( sox17:creERT2;cmlc2:DsRed ) ;Tg ( -3 . 5ubi:loxP-GFP-loxP-mCherry ) zebrafish , embryos were treated for 3 hr from 5 hpf with 10 μM 4-OHT . Experiments using Xenopus laevis were conducted in accordance with the UK Animals ( Scientific Procedures ) Act 1986 , with appropriate personal and project licences in place where necessary . Embryos were obtained by in vitro fertilization and initially kept at 14°C in 0 . 1% modified Barth’s saline ( MBS ) . For grafted embryos that would be grown to stage 53 ( after the onset of independent feeding ) , CMV-GFP transgenic embryos ( Marsh-Armstrong et al . , 1999 ) were used as donors and the embryos were grafted and reared at the European Xenopus Resource Centre ( University of Portsmouth , UK ) . For grafted embryos that were to be grown to embryonic stages 41–45 , GFP-positive donor embryos were made by injecting cyto-GFP mRNA into one cell at the two-cell stage , or two cells at the four-cell stage . Briefly , embryos were de-jellied in 2% cysteine and washed several times in 0 . 1% MBS before being transferred and positioned for injection in a mesh-lined Petri dish filled with 4% Ficoll . Injected embryos were allowed to recover in 4% Ficoll for at least 1 hr before being transferred to 0 . 1% MBS . De-jellied GFP-positive embryos and wild-type embryos were allowed to grow to stage 13–17 at 14–18°C . For grafting , embryos were moved to 18 mm Petri dishes lined with plasticine or 1% agarose with depressions and containing a high-salt transplantation solution ( 1x MBSH: 1x MBS , 0 . 7 mM CaCl2 , 0 . 02 mM NaCl , supplemented with 2 mM CaCl2 and 2 . 5 mg/ml gentamycin [Sigma-Aldrich] ) . The region of the neural folds containing premigratory branchial and vagal neural crest ( Sadaghiani and Thiébaud , 1987 ) was removed unilaterally from stage 13–15 wild-type hosts and replaced with GFP-positive tissue from the same region of donor embryos ( Figure 3d ) . The grafted tissue was held in place with a small piece of glass coverslip while embryos recovered in transplantation solution for at least 2 hr , before being moved to 0 . 1% MBS and reared at 18°C . Embryos were overdosed in MS222 ( Sigma-Aldrich ) in PBS before being fixed in 4% paraformaldehyde in PBS overnight at 4°C . Experiments using sea lamprey ( Petromyzon marinus ) were conducted according to protocols approved by the California Institute of Technology Institutional Animal Care and Use Committee . Eggs were collected from adults and fertilized as described ( Nikitina et al . , 2009 ) . Embryos were maintained at 18°C in 0 . 1x or 1x Marc's modified Ringer’s ( MMR ) solution . DiI labeling was performed as described ( Nikitina et al . , 2009 ) with some modifications . Briefly , E5 embryos ( Piavis stages 11–12 ) were manually dechorionated in 0 . 1x MMR , then immobilized and oriented in 18-mm Petri dishes that were lined with a fine mesh . Embryos were pressure-injected into the dorsal neural tube using glass capillary tubes filled with 0 . 5 mg/ml of Cell Tracker-CM-DiI ( Invitrogen , Carlsbad , CA ) diluted in 0 . 3 M sucrose ( from a 5 mg/ml stock diluted in ethanol ) . They were allowed to recover for 24 hr , then individually transferred to an uncoated Petri dish containing 1x MMR and allowed to develop to E19 ( Piavis stage 17 ) . Embryos were periodically checked and imaged throughout , then fixed in 4% paraformaldehyde in PBS for 1 hr at room temperature . The following transgenic mouse lines were used: Wnt1-cre;R26R-YFP ( Danielian et al . , 1998; Srinivas et al . , 2001 ) , Sox172A-iCre;R26R/+ ( Engert et al . , 2009; Soriano , 1999 ) and Sox172A-iCre;R26RtdTomato ( Engert et al . , 2009; Madisen et al . , 2010 ) . Experiments using these mice were conducted in accordance with the UK Animals ( Scientific Procedures ) Act 1986 , with appropriate personal and project licences in place . Embryos were dissected at appropriate stages and fixed at 4°C overnight in 4% paraformaldehyde in PBS . Experiments using chicken ( Gallus gallus domesticus ) embryos were conducted in accordance with the UK Animals ( Scientific Procedures ) Act 1986 , with appropriate personal and project licences in place where necessary . Fertilized wild-type chicken eggs were obtained from Henry Stewart and Co . Ltd . , Norfolk , UK . Fertilized GFP-transgenic chicken eggs ( McGrew et al . , 2008 ) were obtained from the Roslin Institute Transgenic Chicken Facility ( Edinburgh , UK ) , which is funded by the Wellcome Trust and the BBSRC . Fertilized wild-type and GFP-transgenic eggs were incubated in a humidified atmosphere at 38°C for approximately 1 . 5 days to reach 8–11 somites . The neural tube and associated neural folds between the level of somite one and the caudal end of the seventh somite were dissected from a wild-type host and replaced with the equivalent tissue from a GFP-transgenic donor embryo . At E14 , embryos were decapitated and the lungs dissected out and fixed overnight in 4% paraformaldehyde in PBS . At E16 . 5 , embryos were decapitated and fixed overnight in 4% paraformaldehyde in PBS; the lungs were dissected after fixation . The lungs were dehydrated through an ethanol series into 100% ethanol for storage . Stage 14 Xenopus laevis embryos were fixed in place in a plasticine dish filled with 1x MBSH supplemented with 2 mM CaCl2 , and the endoderm was exposed by cutting a flap into the anterior neural plate ( leaving it attached on the anterior side ) with tungsten needles , which was folded back to expose the endoderm . A stock solution of 2 mg/ml Cell Tracker-CM-DiI ( Invitrogen ) in ethanol was diluted 1:10 in 10% sucrose and microinjected into the anterior endoderm ( regions 1 and 5 of Chalmers and Slack , 2000 ) using a glass electrode whose tip was approximately 20 µm in diameter . The endoderm of each embryo was injected at three to five sites . The flap of the neural plate was then folded back in place and pressed down with a small piece of glass coverslip supported on plasticine feet until it healed back in place ( approximately 1–2 hr ) . Embryos were then transferred into 0 . 1 x MBS containing 25 mg/l gentamicin ( Sigma-Aldrich ) , 400 mg/l penicillin ( Sigma-Aldrich ) and 400 mg/l streptomycin sulfate ( Sigma-Aldrich ) . At stage 45 , tadpoles were overdosed in MS222 ( Sigma-Aldrich ) in PBS before being fixed in 4% paraformaldehyde in PBS at 4°C overnight for up to several days , then transferred to PBS . An ammocoete lamprey was euthanized by MS222 ( Sigma-Aldrich ) overdose , fixed in modified Carnoy’s solution ( six volumes ethanol: three volumes 37% formaldehyde: 1 volume glacial acetic acid ) and dehydrated through an ethanol series into 100% ethanol . DiI-labeled lamprey embryos , grafted Xenopus embryos , some Tg ( -28 . 5sox10:cre;ef1a:loxP-DsRed-loxP-EGFP ) , tfap2amob;foxd3mos zebrafish embryos and their wild-type siblings were dehydrated from PBS into 100% methanol and transferred to 100% isopropanol overnight at 4°C . Embryos were transferred to 1:1 isopropanol:chloroform for 1 hr at 4°C and then to 100% chloroform for 2 hr at −20°C . After warming to room temperature , embryos were transferred to 1:1 chloroform:paraffin wax ( Raymond A . Lamb Ltd . , Thermo Fisher Scientific , Waltham , MA ) at 60°C for 30 min , followed by three 30-min incubations and an overnight incubation at 60°C in paraffin wax . Embryos were embedded in plastic molds and sectioned at 6 µm using a rotary microtome . DiI-labeled Xenopus embryos , Tg ( crestin:creERT2 ) ;Tg ( -3 . 5ubi:loxP-GFP-loxP-mCherry ) , Tg ( -4 . 9sox10:creERT2 ) ;Tg ( βactin:loxP-SuperStop-loxP-DsRed ) , some Tg ( -28 . 5sox10:cre ) ;Tg ( ef1a:loxP-DsRed-loxP-EGFP ) and Tg ( sox17:creERT2;cmlc2:DsRed ) ;Tg ( -3 . 5ubi:loxP-GFP-loxP-mCherry ) zebrafish in PBS were sucrose-protected before being embedded in 7 . 5-20% gelatin in plastic molds , flash-frozen in liquid nitrogen and cryosectioned at 6 μm . Mouse embryos and wild-type zebrafish embryos were sucrose-protected before being embedded in O . C . T . compound ( Tissue-Tek , Sakura Finetek , Torrance , CA ) in plastic molds , flash-frozen in isopentane on dry ice and cryosectioned at 10–15 μm . Grafted chicken lungs and the ammocoete lamprey in 100% ethanol were cleared in Histosol ( National Diagnostics , Atlanta , GA ) and incubated in 1:1 Histosol: paraffin wax ( Raymond A . Lamb Ltd . ) for 30 min at 60°C , followed by three 30-min incubations and an overnight incubation in paraffin wax at 60°C . They were then embedded in plastic molds and sectioned at 6–10 µm using a rotary microtome . For immunostaining on paraffin wax sections , slides were de-waxed in Histosol and rehydrated into PBS through a graded ethanol series . Cryosections were allowed to warm to room temperature and washed in PBS . When necessary , gelatin was removed by dipping slides in PBS warmed to 37°C . All anti-serotonin antibodies used required antigen retrieval , which was performed by heating the slides for 30 s in a microwave in 10 mM sodium citrate buffer solution ( pH 6 ) , followed by two washes in PBS . Immunostaining was performed as described ( Nikitina et al . , 2009 ) with slight modifications: slides were incubated overnight at 4°C or at room temperature in primary antibody in blocking solution ( 10% sheep , goat or donkey serum , as appropriate , in PBS with 0 . 1% Triton X-100 ) ; secondary antibodies were incubated at room temperature for 2 hr or overnight at 4°C . For horse-radish peroxidase detection , slides were incubated in 0 . 3 mg/ml diaminobenzidine , 0 . 02% H2O2 , 0 . 05% Triton X-100 in PBS . After immunostaining , sections were counterstained with the nuclear marker DAPI ( 1 ng/ml ) ( Invitrogen ) and mounted in Fluoromount G ( Southern Biotech , Birmingham , AL ) . For whole-mount immunostaining , Tg ( crestin:creERT2 ) ;Tg ( -3 . 5ubi:loxP-GFP-loxP-mCherry ) zebrafish embryos were incubated for 2 hr in blocking buffer ( PBS with 4% bovine serum albumin , 0 . 3% Triton X-100 , 0 . 02% sodium azide ) prior to overnight incubation at 4°C with primary antibodies diluted in blocking buffer . Embryos were washed for 2 hr at room temperature in PBS with 0 . 3% Triton X-100 , then incubated overnight at 4°C in blocking buffer containing secondary antibodies diluted 1:200 and 1 mg/ml DAPI ( Invitrogen ) diluted 1:200 . After washing for 2 hr in PBS with 0 . 3% Triton X-100 , embryos were suspended in 80% glycerol before mounting . Primary antibodies were used against the following antigens: acetylated tubulin [1:250 mouse IgG2b , clone 6-11-B1 , T7451 Sigma-Aldrich; previously used in the sea lamprey , e . g . , Barreiro-Iglesias et al . ( 2008a ) ] , Ascl1 ( Mash1 ) [1:200 mouse IgG1 ( Lo et al . , 1991 ) , kind gift of F . Guillemot , NIMR , London , UK; 1:100 mouse IgG1 , #556604 BD Biosciences , San Jose , CA] , DsRed2 ( 1:100 mouse IgG1 , sc-101526 Santa Cruz Biotechnology , Dallas , TX ) , Elavl3/4 ( HuC/D ) ( 1:500 mouse IgG2b , A-21271 Invitrogen ) , GFP ( 1:500 rabbit , A-6455 Invitrogen; 1:500 mouse IgG1 , #1814460001 Roche , Basel , Switzerland; 1:250 goat , ab6662 Abcam [Cambridge , UK]; 1:150 chicken , ab13970 Abcam ) , HNK-1 carbohydrate epitope ( Abo and Balch , 1981; Voshol et al . , 1996 ) [for zebrafish neurites ( Metcalfe et al . , 1990 ) : 1:100 mouse IgG1 , ZN-12 Developmental Studies Hybridoma Bank; for lamprey neurites ( Barreiro-Iglesias et al . , 2008b ) : 1:50 mouse IgM , 3H5 Developmental Studies Hybridoma Bank] , mCherry ( 1:250 mouse IgG1 , #632543 Clontech Takara Bio USA Inc . , Mountain View , CA; 1:200 goat , orb11618 Biorbyt , Cambridge , UK ) , serotonin ( 5-hydroxytryptamine , 5-HT ) [1:100 ( whole-mount ) or 1:250 ( sections ) rabbit , S5545 Sigma-Aldrich , previously used in zebrafish , e . g . , Kuscha et al . ( 2012 ) , bullfrog ( Reyes et al . , 2014 ) and Arctic lamprey ( Suzuki et al . , 2015 ) ; 1:100 rat , MAB352 Merck Millipore , Temecula , CA , previously used in zebrafish ( Sundvik et al . , 2013 ) ; 1:250 goat , ab66047 Abcam] , Phox2b [1:500 rabbit , kind gift of Jean-François Brunet , Institut de Biologie de l'École Normale Supérieure , Paris , France; previously used in zebrafish ( Coppola et al . , 2012 ) ; Tubb3 ( neuronal β-III tubulin ) ( 1:500 mouse IgG2a , clone TUJ1 , MMS-435P Covance BioLegend , San Diego , CA ) , and tyrosine hydroxylase [1:250 , rabbit , AB152 Merck Millipore; previously used in zebrafish , e . g . , Yamamoto et al . ( 2011 ) , and sea lamprey , e . g . , Barreiro-Iglesias et al . ( 2008a ) . ( The Developmental Studies Hybridoma Bank was developed under the auspices of the NICHD and is maintained by the University of Iowa , Department of Biological Sciences , Iowa City . ) Appropriately matched AlexaFluor or horse-radish peroxidase-conjugated secondary antibodies were obtained from Molecular Probes/Invitrogen . The lamprey ( P . marinus ) Phox2 clone ( Häming et al . , 2011 ) was a kind gift of Marianne Bronner ( Caltech , Pasadena , CA , USA ) . Whole-mount in situ hybridization on lamprey embryos was performed as described ( Nikitina et al . , 2009 ) . After whole-mount in situ hybridization , embryos were incubated in PBS with 5% sucrose for 4 hr at room temperature , followed by incubation overnight at 4°C in 15% sucrose in PBS . Embryos were transferred into pre-warmed 7 . 5% gelatin in 15% sucrose in PBS and incubated for 1–4 hr at 37°C , then oriented and embedded in molds , frozen by immersion in a dry ice-isopentane solution for 30 s , and cryosectioned at 12–16 µm . Gelatin was removed from the slides by a 5-min incubation in PBS pre-warmed to 37°C . The chicken Phox2b clone ( Stanke et al . , 1999 ) was a kind gift of Jean-François Brunet ( Institut de Biologie de l'École Normale Supérieure , Paris , France ) . For in situ hybridization on paraffin wax sections , slides were de-waxed in Histosol ( National Diagnostics ) and rehydrated into diethylpyrocarbonate ( DEPC ) -treated ( Sigma-Aldrich ) PBS through a graded ethanol series . In situ hybridization was performed on sections as described ( Miller et al . , 2017 ) . For X-gal staining on cryosections of mouse tissue , the following staining solution was added to slides prior to incubation at 37°C for 2 hr: 5 mM K3Fe ( CN ) 6 , 5 mM K4Fe ( CN ) 6 , 2 . 7 mM MgCl2 in PBS , supplemented with 75 mg/ml X-gal in dimethyl sulfoxide ( DMSO ) . For Alcian blue plus hematoxylin and eosin staining on paraffin sections , slides were de-waxed in Histosol and rehydrated into water through a graded ethanol series . Slides were rinsed in 3% acetic acid before staining in 2 mg/ml Alcian blue ( Searle Diagnostic , High Wycombe , UK ) in 3% acetic acid for at least 30 min . Slides were then rinsed in water and treated with 0 . 3% NaHCO3 , followed by another rinse in running water and staining in Mayer’s hematoxylin ( Sigma-Aldrich ) for 10 min . Slides were stained in 1% aqueous eosin Y solution ( BDH ) for 8 min , then washed again in running water before dehydration through an ethanol series into 100% ethanol . After washing in Histosol , slides were mounted with DPX ( BDH ) . Whole-mount images were taken using a Leica MZFLIII microscope ( Leica Microsystems , Wetzlar , Germany ) fitted with a QImaging MicroPublisher 5 . 0 RTV camera and QCapture Pro 6 . 0 software ( QImaging , Surrey , BC , Canada ) ; a Zeiss AxioSkop2 microscope fitted with a Zeiss AxioCam HRc camera and Zeiss AxioVision Rel . 4 . 8 software ( Carl Zeiss , Oberkochen , Germany ) ; an Olympus MVX10 microscope ( Olympus Corporation , Tokyo , Japan ) fitted with a Zeiss AxioCam HRc camera and Zeiss AxioVision Rel . 4 . 8 software; an Olympus 1 × 71 inverted microscope fitted with a Hamamatsu ORCA-R2 monochrome camera and HCImage software ( Hamamatsu Photonics , Hamamatsu , Japan ) ; a Zeiss LSM 710 confocal microscope with Zeiss ZEN software; and a Zeiss 710 confocal microscope with Zeiss LSM Image Browser ( version 4 . 2 . 0 . 121 ) software , which was used to create three-dimensional images and stack movies . Images of sections were taken using a Zeiss AxioSkop 2 MOT microscope fitted with a QImaging Retiga 2000R camera , a Qimaging RGB pancake and QCapture Pro 6 . 0 software; a Zeiss Scope . A1 microscope fitted with a Zeiss AxioCam MRm camera and Zeiss ZEN 2012 ( blue edition ) software; and a Zeiss LSM 780 confocal microscope with Zeiss ZEN 2011 ( black edition ) software . All images were further processed in Photoshop CS4 ( Adobe Systems Inc . , San Jose , CA ) and/or ImageJ 1 . 50i software ( NIH , Bethesda , MD ) . Data analysis and statistical tests were performed using Microsoft Excel and GraphPad Prism 7 software ( GraphPad Software , Inc . , La Jolla , CA ) . Data sets were tested for normality using the Shapiro-Wilk test ( alpha = 0 . 05 ) and for equality of variance using an F test ( p=0 . 38 ) , and compared using an unpaired two-tailed Student’s t-test . Data are presented as mean ± standard deviation ( s . d . ) . | The carotid bodies are small glands found in either side of our neck , near the carotid artery . When the level of oxygen in our blood drops , specialized cells in the carotid bodies signal to the brain to increase our heart rate and make us breathe more rapidly and deeply . As a result , more oxygen is delivered to our cells . Fish have similar oxygen-sensitive cells in their gills , known as neuroepithelial cells , that detect changes in the oxygen levels in the surrounding water and their blood . It has been suggested that after our vertebrate ( back-boned animal ) ancestors moved onto land , the neuroepithelial cells in their gills eventually evolved to form the carotid bodies . Knowing whether this is true would allow researchers to better understand how our ancestors were able to adapt to an obligate air-breathing lifestyle on land . If the carotid body did evolve from ancestral neuroepithelial cells , we would expect that they would both develop from the same kind of cells in the embryo . Carotid body cells develop from a group of cells called neural crest cells , which give rise to many tissues , including nerve cells . Hockman et al . have now investigated whether neuroepithelial cells also develop from neural crest cells . Hockman et al . labelled the neural crest cells in the embryos of zebrafish , frogs and lampreys using techniques such as injecting the cells with fluorescent dye or genetically modifying the cells to make fluorescent proteins . Unexpectedly , the neuroepithelial cells that developed in the gills of these embryos did not contain these fluorescent labels , meaning that they did not develop from the neural crest cells . The patterns of gene activity found in the developing neuroepithelial cells were also different from those in the carotid body . Further investigation revealed that neuroepithelial cells develop from the lining of the mouth and gills and may be related to a similar population of oxygen-sensitive cells found in the lungs . Overall , it appears that the carotid body did not evolve from ancestral neuroepithelial cells . However , Hockman et al . did find some cells near blood vessels in the gills of zebrafish that had developed from neural crest cells . Equivalent cells in our ancestors could therefore be the cells that evolved into carotid bodies . A first test of this theory will be to determine whether or not these cells are oxygen-sensitive . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology"
] | 2017 | Evolution of the hypoxia-sensitive cells involved in amniote respiratory reflexes |
Communication between pre- and postsynaptic cells promotes the initial organization of synaptic specializations , but subsequent synaptic stabilization requires transcriptional regulation . Here we show that fibroblast growth factor 22 ( FGF22 ) , a target-derived presynaptic organizer in the mouse hippocampus , induces the expression of insulin-like growth factor 2 ( IGF2 ) for the stabilization of presynaptic terminals . FGF22 is released from CA3 pyramidal neurons and organizes the differentiation of excitatory nerve terminals formed onto them . Local application of FGF22 on the axons of dentate granule cells ( DGCs ) , which are presynaptic to CA3 pyramidal neurons , induces IGF2 in the DGCs . IGF2 , in turn , localizes to DGC presynaptic terminals and stabilizes them in an activity-dependent manner . IGF2 application rescues presynaptic defects of Fgf22-/- cultures . IGF2 is dispensable for the initial presynaptic differentiation , but is required for the following presynaptic stabilization both in vitro and in vivo . These results reveal a novel feedback signal that is critical for the activity-dependent stabilization of presynaptic terminals in the mammalian hippocampus .
Synapses are the sites of neuronal communication in the brain . Proper synapse formation is critical for appropriate brain function; aberrant synaptic connectivity may result in various neurological and psychiatric disorders , such as autism , Fragile X syndrome , epilepsy , and schizophrenia ( Banerjee et al . , 2014; Casillas-Espinosa et al . , 2012; Lisman , 2012; Pfeiffer and Huber , 2009 ) . Synapse formation begins with target recognition by axons , which is followed by synaptic differentiation at the contact sites . Synaptic differentiation is regulated by signals that are exchanged between pre- and postsynaptic sites . Various target-derived presynaptic organizers , such as fibroblast growth factors ( FGFs ) , WNTs , neurotrophins , neuroligins , Ephs/ephrins , SynCAMs , netrin-G ligands ( NGLs ) , and signal regulatory proteins ( SIRPs ) are shown to promote local differentiation of presynaptic terminals ( Darabid et al . , 2014; Fox and Umemori , 2006; Henriquez et al . , 2011; Johnson-Venkatesh and Umemori , 2010; Regehr et al . , 2009; Salinas , 2012; Shen and Scheiffele , 2010; Siddiqui and Craig , 2011; Toth et al . , 2013; Zweifel et al . , 2005 ) . Initial synapses thereafter maturate , resulting in a more stable , functional , and finely tuned neural network ( Goda and Davis , 2003; Waites et al . , 2005; West and Greenberg , 2011 ) . Presynaptic stabilization has been shown to require gene expression . At the Drosophila larval neuromuscular junction ( NMJ ) , a retrograde signal initiated by glass bottom boat ( Gbb ) , the Drosophila homologue of bone morphogenic protein ( BMP ) , controls presynaptic growth and stabilization through transcriptional regulation in motor neurons . During this process , Gbb and its receptor are internalized and transported from the nerve terminal to the cell body as a retrograde signal . This signal then activates a transcription factor , Mothers against decapentaplegic ( Mad ) ( Aberle et al . , 2002; Marqués et al . , 2002; McCabe et al . , 2003 ) . Activated Mad regulates transcription of genes including Trio and dfmr1 ( fly homolog of FMR1 ) . Trio and dFMR1 play critical roles in modulating actin cytoskeletal dynamics and stabilizing microtubules in the presynaptic motor neurons , leading to presynaptic growth and stabilization ( Ball et al . , 2010; Nahm et al . , 2013 ) . In the mammalian brain , changes in gene expression , as a consequence of axon–dendrite contacts , are also likely to contribute to presynaptic stabilization . For example , expression of genes encoding vesicle proteins increases soon after synaptogenesis begins , and neurons synthesize different isoforms of vesicle proteins before and after their axons contact targets ( Campagna et al . , 1997; Lou and Bixby , 1995; Plunkett et al . , 1998; Sanes and Lichtman , 1999 ) . In addition , synaptic stabilization is influenced by neural activity ( Ackermann et al . , 2015; Chia et al . , 2013; Dalva et al . , 2007; Lichtman and Colman , 2000; Ruthazer and Cline , 2004; Waites et al . , 2005 ) . However , it is not known whether and how target-derived molecules control gene transcription in the presynaptic neurons for the stabilization of presynaptic terminals , and whether such a pathway is regulated by neural activity . We have previously found that FGF22 serves as a target-derived presynaptic organizer in the mouse hippocampus , a key brain region associated with learning , memory , emotional processing , and social behavior . FGF22 is released from CA3 pyramidal neurons and promotes local differentiation of the excitatory presynaptic terminals formed onto them ( Terauchi et al . , 2010; 2015 ) . FGF22-dependent presynaptic differentiation requires two FGF receptors ( FGFRs ) , FGFR2b and FGFR1b , in dentate granule cells ( DGCs ) , the major presynaptic neurons for CA3 pyramidal neurons , and the downstream signaling molecules FGFR substrate 2 ( FRS2 ) and PI-3 kinase ( Dabrowski et al . , 2015 ) . Signals mediated by FRS2 and PI-3 kinase are known to regulate gene expression . Therefore , we hypothesized that FGF22 signaling eventually regulates gene expression and that those FGF22-induced molecules , in turn , contribute to the stabilization of presynaptic terminals . Here , we identified FGF22 target genes in the presynaptic DGCs and asked whether the target genes contribute to presynaptic stabilization . We find that i ) target-derived FGF22 signaling induces the expression of the insulin-like growth factor 2 ( Igf2 ) gene in DGCs , ii ) IGF2 then localizes to presynaptic terminals of DGCs and stabilizes them , iii ) the transportation of IGF2 to the presynaptic terminal is activity-dependent , and iv ) IGF2 is not required for the initial presynaptic differentiation , but is required for subsequent presynaptic stabilization both in vitro and in vivo . Thus , FGF22 is a target-derived molecule not only organizing local , initial presynaptic differentiation , but also regulating IGF2 expression in the presynaptic neurons . IGF2 , in turn , contributes to presynaptic stabilization in an activity-dependent manner . Our results reveal a novel feedback signal that is critical for the activity-dependent stabilization of presynaptic terminals in the mammalian brain .
FGF signals are involved in the development of many organs via regulation of gene expression ( Chen et al . , 2012; Mazzoni et al . , 2013 ) . We hypothesized that in the hippocampus , the excitatory presynaptic organizer FGF22 would activate gene expression in the presynaptic neurons for the stabilization of presynaptic terminals . We focused on genes expressed in DGCs , because they provide a major excitatory input to CA3 pyramidal neurons , and their presynaptic differentiation is dependent on FGF22–FGFR signaling ( Dabrowski et al . , 2015 ) . In the hippocampus , synapse formation starts in the first postnatal week and finishes by postnatal day 28 ( P28 ) ( Danglot et al . , 2006; Steward and Falk , 1991 ) . Presynaptic defects in Fgf22-/- mice begin to appear as early as P8 and are evident at P14 ( Terauchi et al . , 2010 ) . To identify FGF22-regulated genes , we dissected P14 DGCs and compared gene expression profiles between wild-type ( WT ) and Fgf22-/- mice . Microarray analysis revealed several genes that are downregulated in Fgf22-/- DGCs relative to controls ( Table 1 ) . One of the most significantly downregulated genes was Igf2 . This down-regulation was confirmed by RT-PCR ( Figure 1—figure supplement 1 ) , qPCR ( Igf2 was decreased to 53 . 1 ± 12 . 2% in Fgf22-/- mice relative to WT mice ) , and in situ hybridization . In situ hybridization experiments showed that at P14 , and not at P7 , expression of Igf2 mRNA was decreased in DGCs ( Figure 1A–D ) . It was not decreased in other hippocampal cells such as CA1 and CA3 pyramidal neurons of Fgf22-/- mice ( Figure 1A–D ) . Interestingly , Igf2 expression was clearly decreased in the inner layer of Fgf22-/- DGCs , where relatively immature DGCs are located ( Figure 1C and D ) ( Aguilar-Arredondo et al . , 2015 ) . 10 . 7554/eLife . 12151 . 003Table 1 . List of genes that are significantly downregulated in DGCs of Fgf22-/- mice at P14 . Genes with Diff Score < -33 ( p-value <0 . 001 ) relative to WT are shown in the table . DOI: http://dx . doi . org/10 . 7554/eLife . 12151 . 003SymbolDefinitionDiff ScoreSlc6a13Solute carrier family 6 ( Neurotransmitter Transporter ) , member 13-56 . 042Col6a1Collagen , type VI , alpha 1-52 . 859PrelpProline arginine-rich end leucine-rich repeat-50 . 844Igf2Insulin-like growth factor 2-42 . 987Mrc2Mannose receptor , C Type 2-41 . 904Gys3Glycogen synthase 3 , brain-41 . 834Aebp1Adipocyte enhancer-binding protein 1-40 . 069Lrrtm3Leucine rich repeat transmembrane neuronal 3-35 . 357Zfp365Zinc finger protein 365-34 . 604Col1a1Collagen , type I , alpha 1-33 . 44410 . 7554/eLife . 12151 . 004Figure 1 . IGF2 expression is decreased in young DGCs in Fgf22-/- mice during the stage of synapse stabilization . ( A–D ) In situ hybridization for Igf2 mRNA . ( A ) At P7 ( initial stage of synaptic differentiation ) , Igf2 mRNA is similarly expressed in the hippocampus of WT and Fgf22-/- mice . Higher magnification views of the boxed areas are shown in ( B ) . ( C ) At P14 ( around the time of synaptic stabilization ) , Igf2 mRNA is decreased in Fgf22-/- mice in the inner molecular layer of DGCs . Higher magnification views of the boxed areas are shown in ( D ) . Observations are from 3–5 animals per age and strain . ( E–H ) P7 and P14 hippocampal sections from WT and Fgf22-/- mice were immunostained for IGF2 and for either calretinin ( CR; young DGCs ) or calbindin ( CB; mature DGCs ) . The illustration shows the pictured area ( boxed ) . ( E ) IGF2 expression in DGCs at P7 . Quantification of IGF2 immunoreactivity in CR and CB layers is shown in ( F ) . There is no significant difference in the IGF2 intensity in either layer of DGCs at P7 . ( G ) IGF2 expression in DGCs at P14 . Quantification of IGF2 intensity in CR and CB layers is shown in ( H ) . In P14 Fgf22-/- mice , IGF2 is significantly decreased in the CR-positive layer , but not in CB-positive layer of DGCs . Error bars are s . e . m . Data are from 77–160 fields from 3–5 animals . Significant difference from control at ***p<0 . 0001 by Student's t-test . Scale bars , ( A and C ) 500 μm , ( B and D ) 50 μm , ( E and G ) 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 12151 . 00410 . 7554/eLife . 12151 . 005Figure 1—figure supplement 1 . RT-PCR showing decreased Igf2 mRNA expression in DGCs of P14 Fgf22-/- mice . Representative picture of RT-PCR detecting Igf2 mRNA . β-actin is used as a control . Total mRNA was extracted from DGCs of P14 WT and Fgf22-/- mice . The following primers were used for RT-PCR: IGF2-forward: 5’-TCTCATCTCTTTGGCCTTCGCCTT-3’ , IGF2-reverse: 5’-GTCCGAACAGACAAACTGAAGCGT-3’ ( amplifying 106 bp ) ; β-actin forward: 5’-GTGGGCCGCTCTAGGCACCAA-3’ , β-actin reverse: 5’-CTCTTTGATGTCACGCACGATTTC-3’ ( amplifying 472 bp ) . Expression of Igf2 mRNA is decreased in Fgf22-/- mice compared to that in WT mice . DOI: http://dx . doi . org/10 . 7554/eLife . 12151 . 00510 . 7554/eLife . 12151 . 006Figure 1—figure supplement 2 . Validation of the anti-IGF2 antibody . The anti-IGF2 antibody ( Santa Cruz , sc-5622 ) was validated by immunostaining of brain sections prepared from P14 WT and Igf2-/- mice . ( A ) Representative images of IGF2 immunostaining in dentate granule cells ( DGCs ) . The illustration shows the pictured area ( boxed ) . IGF2 , green; DAPI , blue . ( B ) Representative images of IGF2 immunostaining in the choroid plexus , which highly expresses IGF2 . The anti-IGF2 antibody did not stain DGCs or the choroid plexus of Igf2-/- mice . Observations are from 3 animals per strain . Scale bars are 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 12151 . 006 We then asked whether IGF2 expression was decreased in a specific developmental stage of DGCs in Fgf22-/- mice . Using specific markers , DGCs can be classified into several developmental subsets: from newborn to mature , DGCs are positive for Ki67 ( dividing DGCs ) , doublecortin ( immature ) , calretinin ( young ) , and calbindin ( mature ) ( Abrous et al . , 2005 ) . Ki67- , doublecortin- , and calretinin-positive DGCs populate the inner granule cell layer , and calbindin-positive DGCs the outer . We found that IGF2 protein expression was decreased in calretinin-positive , but not in calbindin-positive DGCs in Fgf22-/- mice relative to WT mice at P14 ( Figure 1G and H; the specificity of the anti-IGF2 antibody was verified using tissues from Igf2-/- mice: Figure 1—figure supplement 2 ) . Consistent with the in situ results , no changes were observed at P7 ( Figure 1E and F ) . These results indicate that the lack of FGF22 impairs IGF2 expression in young , developing DGCs during the stage of synapse stabilization . FGF22 is highly expressed by CA3 pyramidal neurons in the hippocampus ( Terauchi et al . , 2010 ) . We next examined whether FGF22 derived from CA3 pyramidal neurons is responsible for IGF2 expression in DGCs . For this , we inactivated FGF22 preferentially in CA3 pyramidal neurons using Fgf22flox/flox mice ( Fgf22f/f; EUCOMM ) crossed with Grik4-Cre mice ( Figure 2A ) . Grik4-Cre mice express Cre in 100% of CA3 pyramidal neurons and 10% of DGCs in the hippocampus ( Nakazawa et al . , 2002 ) . We found that at P14 , IGF2 expression in the inner layer , but not in the outer layer , of DGCs of Fgf22f/f::Grik4-Cre mice was significantly decreased relative to that of control littermates ( controls include wild type and Fgf22f/f mice; we did not observe any significant differences in IGF2 staining between wild type and Fgf22f/f mice ) ( Figure 2B–D ) . The level of decrease in IGF2 expression in the inner layer of Fgf22f/f::Grik4-Cre mice was similar to that in Fgf22-/- mice . These results suggest that CA3-derived FGF22 regulates the expression of IGF2 in young DGCs in vivo . 10 . 7554/eLife . 12151 . 007Figure 2 . IGF2 expression in young DGCs is decreased in CA3-selective Fgf22-knockout mice . ( A ) Schematic of CA3-selective Fgf22 deletion: Fgf22flox/flox ( Fgf22f/f ) mice were crossed with mice carrying Grik4-promoter-driven Cre ( Grik4-Cre ) . ( B–D ) IGF2 staining in the DGCs of P14 Fgf22f/f::Grik4-Cre mice and control littermates ( WT and Fgf22f/f mice; we did not observe any significant differences in IGF2 staining between WT and Fgf22f/f mice ) . ( B ) Illustration showing the pictured area ( boxed ) . ( C ) Representative pictures of IGF2 immunostaining in DGCs . The dashed line indicates the border between the inner and outer layers of DGCs . ( D ) Quantification of IGF2 immunoreactivity in the inner and outer DGC layers . CA3-selective inactivation of FGF22 results in a significant decrease in the IGF2 expression in the inner DGC layer , but not in the outer DGC layer . Error bars are s . e . m . Data are from 20–25 fields ( D , inner DGC layer ) and from 40–50 fields ( D , outer DGC layer ) from 4–5 animals . Significant difference from control at ***p<0 . 0001 by Student's t-test . Scale bar , 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 12151 . 007 We next examined whether IGF2 expression in DGCs is increased by FGF22 treatment in culture . DGCs were identified with a marker , Prox1 ( Iwano et al . , 2012 ) . In our hippocampal culture , initial formation of glutamatergic synapses starts from ~3 days in vitro ( 3DIV ) , followed by activity-dependent synapse maturation from ~8DIV to ~12DIV ( Terauchi et al . , 2010; Toth et al . , 2013 ) . When cultured hippocampal cells were treated with FGF22 at 1DIV , IGF2 expression in the soma of DGCs was significantly increased at 7DIV ( Figure 3A and B ) . We next asked whether only specific developmental subsets of DGCs increase IGF2 expression in response to FGF22 treatment . We found that FGF22-dependent increase in IGF2 was observed in calretinin-positive DGCs , but not in calbindin-positive DGCs ( Figure 3C and E ) . In contrast , IGF2 expression did not increase in non-DGCs , such as CA3 pyramidal neurons , which were identified by immunostaining with Py-antibody ( Figure 3D and E ) . These results demonstrate that FGF22 signaling regulates IGF2 expression in young , calretinin-positive DGCs . Note that at 7DIV , a majority of DGCs in culture were calretinin-positive ( 63 . 65 ± 1 . 40%; Figure 3—figure supplement 1 ) , indicating that our results with DGCs ( identified as Prox1-positive cells ) mostly reflect calretinin-positive DGCs . 10 . 7554/eLife . 12151 . 008Figure 3 . Bath and axonal application of FGF22 increases IGF2 expression in young DGCs . ( A–E ) Cultured hippocampal neurons were treated with FGF22 at 1DIV , and fixed and stained at 7DIV . ( A ) Bath application of FGF22 increases IGF2 expression in DGCs ( Prox1-positive ) . ( B ) Quantification of IGF2 immunoreactivity in the cell bodies of DGCs , normalized to untreated condition . ( C ) Bath application of FGF22 increases IGF2 expression in CR-positive DGCs , but not in CB-positive DGCs . ( D ) Bath application of FGF22 does not affect IGF2 expression in CA3 pyramidal neurons ( Py-positive ) . ( E ) Quantification of IGF2 immunoreactivity in the cell bodies of CR- or CB-positive DGCs and CA3 pyramidal neurons . Data are normalized to the intensity from untreated cells . ( F–I ) Hippocampal neurons were plated onto the somal compartment of microfluidic chambers and cultured . FGF22 was applied into the axonal compartment at 2DIV , and cells were fixed and stained at 8DIV . ( F ) Schematic illustration of the microfluidic chamber . Representative pictures in ( G ) and ( H ) are taken from the boxed areas . ( G–I ) Axonal treatment of FGF22 increases IGF2 in the cell body of DGCs . ( G ) Lower magnification views of Prox1 positive DGCs . ( H ) Higher magnification views from the boxed areas in ( G ) . Quantification of IGF2 immunoreactivity in the cell bodies of DGCs is shown in ( I ) . Error bars are s . e . m . Data are from ( B ) 292–260 cells from 3 independent experiments , ( E ) 25–45 cells from 4 to 5 independent experiments , and ( I ) 67–78 cells from 3–4 independent experiments . Significant difference from control at ***p<0 . 0001 by Student's t-test . Scale bars , ( A , C , D and H ) 10 μm , ( F ) 100 μm , ( G ) 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 12151 . 00810 . 7554/eLife . 12151 . 009Figure 3—figure supplement 1 . A majority of DGCs in culture are calretinin-positive . Representative images of cultured hippocampal neurons stained for calretinin and Prox1 . Cultured hippocampal neurons were fixed and stained at 7DIV . A majority of DGCs ( Prox1-positive ) in culture are calretinin-positive ( 63 . 65 ± 1 . 40% ) . Filled and empty arrowheads indicate calretinin-positive and negative DGCs , respectively . Data are from 9 fields from 3 independent experiments . Scale bar is 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 12151 . 009 FGF22 is a target-derived presynaptic organizer that acts on axons . Hence , we next investigated whether local treatment with FGF22 at DGC axons is sufficient to increase expression of IGF2 in DGCs . Cultured hippocampal neurons were divided into axonal and somal compartments using an in vitro microfluidic culture system ( Figure 3F ) . Axons of cultured hippocampal cells appeared in the axonal compartment by 2DIV . Local application of FGF22 at 2DIV to the axonal compartment resulted in an increase in IGF2 level in the cell bodies of Prox1-positive DGCs at 8DIV ( Figure 3G–I ) , indicating that retrograde FGF22 signaling from axon terminals increases IGF2 expression in the soma of DGCs . Application of FGF22 increases IGF2 expression in the DGC soma . We next asked where the induced IGF2 localizes in the DGCs . To address this question , we transfected EGFP-tagged IGF2 ( IGF2-EGFP ) and analyzed its localization in cultured DGCs . IGF2-EGFP showed a punctate pattern in neurofilament-positive axons of DGCs , while it was dim and diffuse throughout MAP2-positive dendrites of these neurons ( Figure 4A ) . IGF2-EGFP puncta in the axons of cultured DGCs were colocalized with cotransfected synaptophysin-mCherry ( 83 . 1 ± 1 . 0% of IGF2-EGFP puncta were colocalized with synaptophysin-mCherry; Figure 4B ) , indicating that IGF2 localizes to presynaptic terminals . We then asked whether IGF2 , which is a secreted protein , is localized on the surface of presynaptic terminals . We stained IGF2-EGFP transfected neurons with the anti-GFP antibody without a detergent followed by Alexa Fluor 647 secondary antibody ( Figure 4C ) . We found that 41 . 2 ± 0 . 9% of IGF2-EGFP was localized on the cell surface . Surface IGF2-EGFP was always colocalized with synaptophysin-mCherry ( Figure 4C ) , suggesting that IGF2 is secreted and tethered on the surface of presynaptic terminals . Next , we examined the localization of IGF2 receptors . IGF2R , the major receptor for IGF2 ( Fernandez and Torres-Aleman , 2012 ) , showed a punctate pattern in the axons of DGCs ( Figure 4—figure supplement 1A ) and was localized at presynaptic terminals ( 84 . 32 ± 0 . 90% of IGF2R puncta were colocalized with VGLUT1 puncta; Figure 4—figure supplement 1B ) . These results are consistent with the notion that IGF2 is secreted from the presynaptic terminal and binds to IGF2R , which is also localized at the presynaptic terminal . 10 . 7554/eLife . 12151 . 010Figure 4 . IGF2 localizes to presynaptic terminals of DGCs . Cultured hippocampal neurons were transfected with the IGF2-EGFP plasmid at 3DIV , and fixed and stained at 10DIV . ( A ) IGF2-EGFP showed a punctate pattern of localization in neurofilament positive axons , while a diffuse pattern in MAP2 positive dendrites of DGCs . Observations were from at least 10 transfected Prox1 positive DGCs from 2 independent experiments . ( B ) The IGF2-EGFP plasmid was co-transfected with synaptophysin-mCherry ( Sphy-mCherry ) plasmid . Most of IGF2-EGFP puncta ( 83 . 1 ± 1 . 0% of total IGF2-EGFP puncta; data are from 53 cells from 3 independent experiments ) co-localizes with synaptophysin-mCherry , a presynaptic terminal marker . ( C ) ~40% of IGF2-EGFP is localized on the surface of presynaptic terminals . At 10DIV , cells were stained with the anti-GFP antibody without a detergent , followed by Alexa Fluor 647 ( shown in blue in the images ) . Surface IGF2-EGFP is always colocalized with synaptophysin-mCherry . Data are from 15 cells from 3 independent experiments . Scale bars , 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 12151 . 01010 . 7554/eLife . 12151 . 011Figure 4—figure supplement 1 . IGF2R , the major receptor for IGF2 , is localized at presynaptic terminals of DGCs . Cultured hippocampal neurons were fixed and stained at 10DIV . ( A ) IGF2R showed a punctate pattern in axons ( neurofilament positive ) and a diffuse pattern in dendrites ( MAP2-positive ) of DGCs ( Prox1-positive ) . ( B ) Most of IGF2R puncta in DGC axons were colocalized with VGLUT1 puncta ( 84 . 32 ± 0 . 90% ) . Observation and data are from 8–12 cells from 2 independent experiments . Scale bar is 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 12151 . 011 We next determined whether IGF2 expressed in DGCs promotes presynaptic development . To detect presynaptic development , we cotransfected IGF2 with synaptophysin-YFP . Overexpression of IGF2 increased the density and the size of synaptophysin-YFP puncta in DGCs compared to those in control cultures ( Figure 5A ) , without apparently altering the morphology of DGCs ( Figure 5—figure supplement 1 ) . No effect of IGF2 overexpression was found in synaptophysin-YFP puncta in Prox1-negative non-DGCs ( Figure 5B ) . These results indicate that IGF2 promotes presynaptic development specifically in DGCs . 10 . 7554/eLife . 12151 . 012Figure 5 . Overexpression of IGF2 in DGCs promotes their presynaptic development . Cultured hippocampal neurons were transfected with the plasmid expressing IGF2 together with the synaptophysin-YFP plasmid at 3DIV , and fixed and stained at 10DIV . ( A ) Clustering of synaptophysin-YFP is increased in IGF2-overexpressed DGCs ( Prox1-positive ) compared to control DGCs . The graphs show quantification of the number and size of synaptophysin-YFP puncta in control and IGF2-overexpressed DGCs . ( B ) Overexpression of IGF2 does not alter clustering of synaptophysin-YFP in Prox1-negative hippocampal neurons . The graph shows quantification of the number and size of synaptophysin-YFP puncta in Prox1 negative neurons with or without IGF2 overexpression . Error bars are s . e . m . Data are from ( A ) 36–44 cells from 5–7 independent experiments , ( B ) 11–13 cells from 2–3 independent experiments . Significant difference from control at ***p<0 . 0001 by Student's t-test . Scale bars , 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 12151 . 01210 . 7554/eLife . 12151 . 013Figure 5—figure supplement 1 . Overexpression of IGF2 does not appear to alter the morphology of DGCs . Cultured hippocampal neurons were transfected with the IGF2 expression plasmid and EGFP-N1 plasmid at 3DIV , and fixed and stained at 10DIV . ( A ) Morphology of calretinin and Prox1 double-positive young DGCs , with or without IGF2 overexpression . ( B ) Morphology of calbindin- and Prox1-double positive mature DGCs , with or without IGF2 overexpression . Overexpression of IGF2 does not appear to affect the morphology of either calretinin- or calbindin-positive DGCs . Scale bars , 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 12151 . 013 Synapse formation can be separated into two stages: the initial synaptic differentiation stage and the synapse maturation stage . Initial synaptic differentiation is usually regarded as an activity-independent step , while synaptic maturation , including synaptic growth , elimination , and stabilization , is influenced by neural activity . Our previous report identified that activity-dependent refinement of DGC–CA3 connections begins at ~P15 ( Yasuda et al . , 2011 ) , which is around when we observed decreased IGF2 expression in Fgf22-/- mice ( Figure 1 ) . Thus , we next asked whether IGF2 expression , IGF2 localization , and/or IGF2 function for synaptogenesis require neural activity . With FGF22 treatment , IGF2 expression increased in the soma of cultured DGCs ( Figure 3A , Figure 6A and B ) . Blockade of neural activity with tetrodotoxin ( TTX ) did not disturb the ability of FGF22 to increase IGF2 expression ( Figure 6A and B ) . We then examined the effect of TTX on presynaptic localization of IGF2 . We found that TTX treatment reduced the clustering and synaptic localization of IGF2 ( Figure 4 , Figure 6C and D ) . Finally , we assessed the requirement of neural activity in IGF2 function to induce presynaptic development . IGF2 overexpression increased synaptophysin-YFP accumulation in cultured DGCs ( Figure 5A , Figure 6E and F ) . This increase was completely blocked by TTX treatment ( Figure 6E and F ) . These results indicate that neural activity is not required for FGF22 to induce IGF2 expression , but is necessary for IGF2 to localize to and organize development of presynaptic terminals . To confirm that the activity of presynaptic neurons is critical for the localization and function of IGF2 , we suppressed intrinsic neuronal excitability of DGCs by sparsely transfecting the inwardly rectifying potassium channel , Kir2 . 1 , in culture ( Johnson-Venkatesh et al , 2015 ) . Similarly to the results with TTX , Kir2 . 1 expression in DGCs decreased the synaptic localization of IGF2 ( Figure 6G and H ) and completely blocked the synaptogenic function of IGF2 ( Figure 6I and J ) . These results indicate that intrinsic neuronal excitability of DGCs is required for the presynaptic localization of IGF2 and its synaptogenic function . 10 . 7554/eLife . 12151 . 014Figure 6 . Neural activity is necessary for synaptic localization and synaptogenic effects of IGF2 , but not for FGF22-dependent IGF2 expression . ( A and B ) Cultured hippocampal neurons were treated with FGF22 at 1DIV , and fixed and stained at 7DIV , as in Figure 3A . At 1DIV and 5DIV , TTX was added in the media to block neuronal activity . Bath application of FGF22 increases IGF2 expression in DGCs without ( left panels ) as well as with activity blockade ( right panels ) . Quantification of IGF2 immunoreactivity in the cell bodies of DGCs is shown in ( B ) . ( C–F ) TTX treatment impairs synaptic localization of IGF2 and its synaptogenic function . ( C and D ) Cultured hippocampal neurons were transfected with the IGF2-EGFP plasmid together with synaptophysin-mCherry ( Sphy-mCherry ) plasmid at 3DIV , and fixed and stained at 10DIV , as in Figure 4B . At 3DIV and 7DIV , TTX was added in the media to block global neuronal activity . Quantification of the number and size of IGF2-EGFP puncta , and percentage of IGF2-EGFP puncta that colocalized with synaptophysin-mCherry puncta are shown in ( D ) . TTX treatment decreases IGF2-EGFP clustering and synaptic localization . ( E and F ) Cultured hippocampal neurons were transfected with the plasmid expressing IGF2 together with the synaptophysin-YFP plasmid at 3DIV , and fixed and stained at 10DIV , as in Figure 5 . TTX was added in the media at 3DIV and 7DIV . IGF2 overexpression increases clustering of synaptophysin-YFP in Prox1-positive DGCs without TTX ( upper panels ) , but not with TTX ( lower panels ) . ( F ) Quantification of the number and size of synaptophysin-YFP in DGCs with or without IGF2 overexpression in the presence or absence of TTX . ( G–J ) Suppression of intrinsic neuronal excitability impairs synaptic localization of IGF2 and its synaptogenic function . ( G and H ) Cultured hippocampal neurons were transfected with the plasmids expressing Kir2 . 1 , IGF2-EGFP , and synaptophysin-mCherry ( Sphy-mCherry ) at 3DIV . Cells were fixed and stained at 10DIV . Percentage of IGF2-EGFP puncta that colocalized with synaptophysin-mCherry puncta in Prox1-positive DGCs is shown in ( H ) . Kir2 . 1 expression decreases synaptic localization of IGF2-EGFP in DGCs . ( I and J ) Cultured hippocampal neurons were transfected with the plasmids expressing Kir2 . 1 , IGF2 , and synaptophysin-YFP at 3DIV . Cells were fixed and stained at 10DIV . ( J ) Quantification of the number and size of synaptophysin-YFP in DGCs with or without IGF2 overexpression in the presence or absence of Kir2 . 1 . IGF2 overexpression increases clustering of synaptophysin-YFP in Prox1-positive DGCs , but the increase is suppressed by Kir2 . 1 expression . Error bars are s . e . m . Data are from ( B ) 35–292 cells from 3 independent experiments , ( D ) 20–32 cells from 3–4 independent experiments , ( F ) 22–24 cells from 3 independent experiments , ( H ) 7 cells from 3 independent experiments , ( J ) 20 cells from 4 independent experiments . Significant difference from control at ***p<0 . 0001 by two-way ANOVA followed by Tukey's multiple comparison test ( B , F , and J ) and by Student's t-test ( D and H ) . n . s . : no statistical significant difference . Scale bars , 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 12151 . 014 We next asked whether IGF2 acts downstream of FGF22 signaling in the regulation of excitatory presynaptic development . Our previous analysis showed that in CA3 of Fgf22-/- mice , synaptic connections are made , but synaptic vesicles fail to appropriately accumulate to the presynaptic terminals ( Terauchi et al . , 2010 ) . In Fgf22-/- cultures , the accumulation of glutamatergic synaptic vesicles , as assessed by the number and size of VGLUT1 ( vesicular glutamate transporter 1 ) puncta , onto dendrites of CA3 pyramidal neurons is specifically impaired ( Terauchi et al . , 2010; also see Figure 7A ) . We examined whether application of IGF2 could rescue the defects in synaptic vesicle accumulation in Fgf22-/- cultures . In WT cultures , bath application of IGF2 increased the clustering of glutamatergic synaptic vesicles on the dendrites of CA3 pyramidal neurons at 13DIV ( Figure 7A ) ; consistent with this result , IGF2 overexpression increased synaptophysin-YFP puncta in DGCs ( Figure 5A , Figure 6E and F ) . In Fgf22-/- cultures , IGF2 application rescued the defects in VGLUT1 accumulation: the restored number and size of VGLUT1 puncta were comparable to those seen in IGF2-treated WT cultures ( Figure 7A ) . IGF2 treatment , as well as deficiency of FGF22 , did not alter the clustering of VGAT ( vesicular GABA transporter; a marker of GABAergic synaptic vesicles ) on the dendrite of CA3 pyramidal neurons , indicating that IGF2 and FGF22 are not involved in inhibitory presynaptic development on CA3 pyramidal neurons ( Figure 7B ) . In addition , IGF2 treatment and loss of FGF22 did not change the accumulation of postsynaptic markers , PSD95 ( glutamatergic ) and gephyrin ( GABAergic ) , associated with CA3 pyramidal neurons ( Figure 7C and D ) . Altogether , IGF2 specifically rescues excitatory presynaptic defects in Fgf22-/- neurons , suggesting that IGF2 is a mediator of FGF22 signaling to promote excitatory presynaptic development on the dendrites of CA3 pyramidal neurons . 10 . 7554/eLife . 12151 . 015Figure 7 . IGF2 treatment rescues the impairment of glutamatergic presynaptic development in Fgf22-/- cultures . Hippocampal neurons from WT and Fgf22-/- mice were cultured with or without IGF2 treatment . IGF2 was applied into culture media at 1DIV , and cells were fixed and stained at 13DIV . ( A ) VGLUT1 clustering on the dendrites of CA3 pyramidal neurons ( immunolabeled with Py antibody ) . IGF2 treatment increases VGLUT1 clustering in WT culture and rescues defects in VGLUT1 clustering in Fgf22-/- cultures . ( B–D ) Application of IGF2 does not change VGAT clustering ( B ) , PSD95 clustering ( C ) , and gephyrin clustering ( D ) on the dendrites of CA3 pyramidal neurons in WT and Fgf22-/- cultures . The density ( number/mm ) and size of VGLUT1 , VGAT , PSD95 and gephyrin puncta on CA3 pyramidal neurons were analyzed and shown in the graphs . Error bars are s . e . m . Data are from 12–39 cells from 3–7 independent experiments . Significant difference from control at *p<0 . 05 , **p<0 . 01 and ***p<0 . 001 by two-way ANOVA followed by Tukey's multiple comparison test . n . s . : no statistical significant difference . Scale bars , 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 12151 . 015 We then asked whether IGF2 is necessary for FGF22-dependent excitatory presynaptic development . Defects in excitatory synapse formation are observed in the CA3 region of the hippocampus in Fgf22-/- mice as early as P8 , an early stage of synapse formation ( Terauchi et al . , 2010 ) . Meanwhile , decreased IGF2 expression was observed in Fgf22-/- mice at P14 , but not at P7 ( Figure 1 ) , and the effects of IGF2 on presynaptic development is activity-dependent ( Figure 6C–F ) . These results raise the possibility that IGF2 contributes to FGF22-dependent presynaptic development in a stage-specific manner: IGF2 is not required during the initial stage but is required during the following stage of synapse formation , i . e . , activity-dependent synapse stabilization . To test this possibility , we employed an shRNA knockdown approach ( Figure 8—figure supplement 1 ) to silence the expression of IGF2 in DGCs from 1DIV by using two independent shRNA plasmids . Knockdown of IGF2 did not apparently alter the morphology of DGCs ( Figure 8—figure supplement 2 ) . We assessed synaptic vesicle ( synaptophysin-YFP ) accumulation in response to FGF22 in the axons of IGF2-knockdown DGCs ( from 1DIV ) during the initial presynaptic differentiation ( 6DIV ) and subsequent presynaptic stabilization ( 12DIV ) stages . At 6DIV , we did not observe any differences between IGF2-knockdown and control DGCs: the basal levels of synaptophysin-YFP accumulation were not different , and IGF2-knockdown DGCs still responded to FGF22 treatment to increase the accumulation of synaptophysin-YFP ( Figure 8A and B ) . On the other hand , at 12DIV , the number and size of synaptophysin-YFP puncta were significantly decreased in IGF2-knockdown DGCs compared to those in controls ( Figure 8C and D ) . In addition , IGF2-knockdown DGCs no longer show FGF22-dependent increases in the accumulation of synaptophysin-YFP ( Figure 8C and D ) . These results suggest that IGF2 is not necessary for the initial presynaptic differentiation induced by FGF22 , but is required for a later stage of presynaptic development . To confirm the role of IGF2 in the late stage of presynaptic development , we knocked down IGF2 from 6DIV and assessed presynaptic development at 12DIV . IGF2 knockdown from 6DIV decreased the number and size of presynaptic terminals ( as assessed by synaptophysin-YFP accumulation ) and blocked the synaptogenic effects of FGF22 at 12DIV ( Figure 8E and F ) , suggesting that IGF2 is indeed critical for the later stage of presynaptic stabilization . 10 . 7554/eLife . 12151 . 016Figure 8 . IGF2 is dispensable for FGF22-dependent initial presynaptic differentiation , but is required for subsequent presynaptic stabilization . ( A-D ) Cultured hippocampal neurons were transfected with the plasmid expressing synaptophysin-YFP together with the plasmid expressing either control-shRNA , IGF2-shRNA#1 , or IGF2-shRNA#2 at 1DIV . FGF22 was applied into culture media after the transfection at 1DIV . Cells were fixed , stained , and clustering of synaptophysin-YFP in DGCs was analyzed at 6DIV or 12DIV . ( A and B ) Clustering of synaptophysin-YFP in DGCs at 6DIV . Quantification of the number and size of synaptophysin-YFP clusters in DGCs are shown in the graph ( B ) . FGF22 treatment increases the number and size of synaptophysin-YFP puncta in both control and IGF2-knockdown DGCs . ( C and D ) Clustering of synaptophysin-YFP in DGCs at 12DIV . Quantification of the number and size of synaptophysin-YFP clusters are shown in the graph ( D ) . Without IGF2 , the effects of FGF22 on synaptophysin-YFP clustering disappear . ( E and F ) Cultured hippocampal neurons were treated with FGF22 at 1DIV . At 6DIV , neurons were transfected with the plasmid expressing synaptophysin-YFP together with the plasmid expressing either control-shRNA , IGF2-shRNA#1 , or IGF2-shRNA#2 . Cells were fixed and stained at 12DIV . Quantification of the number and size of synaptophysin-YFP clusters are shown in the graph ( F ) . IGF2-knockdown at a late stage of synapse development ( from 6DIV ) decreases synaptophysin-YFP clustering and blocks the synaptogenic effects of FGF22 in DGCs . Error bars are s . e . m . Data are from 12–62 cells from 5–8 independent experiments . Significant difference from control at *p<0 . 05 , **p<0 . 01 and ***p<0 . 001 by two-way ANOVA followed by Tukey’s multiple comparison test . n . s . : no statistical significant difference . Scale bars , 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 12151 . 01610 . 7554/eLife . 12151 . 017Figure 8—figure supplement 1 . Efficiency of shRNA-mediated IGF2 knockdown . Cultured hippocampal neurons were transfected with the plasmids expressing IGF2-EGFP and ( A ) control-shRNA , ( B ) IGF2-shRNA#1 , and ( C ) IGF2-shRNA#2 at 3DIV . Neurons were fixed at 6DIV . Total GFP intensities in the IGF2-shRNA transfected cells ( normalized to the intensity in the control-shRNA transfected cells ) were analyzed and shown in ( D ) . Both IGF2-shRNAs effectively inhibited the expression of IGF2 ( to < 25% of controls ) . Scale bars , 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 12151 . 01710 . 7554/eLife . 12151 . 018Figure 8—figure supplement 2 . Effects of shRNA-mediated IGF2 knockdown on DGC morphology . Cultured hippocampal neurons were transfected with the plasmid expressing either control-shRNA , IGF2-shRNA#1 , or IGF2-shRNA#2 at 1DIV . FGF22 was applied into culture media after the transfection at 1DIV . Cells were fixed and stained at 12DIV for morphological observations . Morphology of transfected DGCs ( labeled with co-expressing RFP ) does not appear affected by shRNA-mediated knockdown of IGF2 or FGF22 treatment . Scale bar , 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 12151 . 018 Finally , to investigate whether and at which developmental stages IGF2 is necessary for excitatory presynaptic development in vivo , we examined clustering of excitatory synaptic vesicles in CA3 of Igf2-/- mice from P8 to P29 . In the hippocampus , synapse development from P0 to ~P14 is not apparently influenced by neural activity ( "initial synapse differentiation" ) , but that from ~P14 to ~P28 is regulated by activity , where activity-dependent synapse maturation ( e . g . , Toth et al . , 2013 ) and activity-dependent synapse elimination ( Yasuda et al . , 2011 ) take place . We found that at P8 , clustering of VGLUT1 in CA3 of Igf2-/- mice was similar to that of WT mice ( Figure 9A–D ) . At P14 , Igf2-/- mice still had a similar number of VGLUT1 puncta , but their size was smaller in the CA3 stratum lucidum ( SL ) region , where the axons of DGCs form excitatory synapses with CA3 pyramidal neurons . At P21 and P29 , Igf2-/- mice showed a significant decrease in the number and size of VGLUT1 puncta in the SL layer ( Figure 9A–C ) . The targeting of DGC axons to CA3 appeared normal in Igf2-/- mice ( Figure 9—figure supplement 1 ) . No defects were observed in the CA3 stratum radiatum ( SR ) region , where CA3 to CA3 synapses are located , throughout the time periods we examined ( Figure 9A and D ) . These results suggest that IGF2 is critical for a later stage ( after P14 ) of DGC–CA3 synapse development , but not for CA3–CA3 synapse development . 10 . 7554/eLife . 12151 . 019Figure 9 . Glutamatergic presynaptic stabilization is impaired at the DGC–CA3 synapses in Igf2-/- mice . Hippocampal sections from WT and Igf2-/- mice at P8 , P14 , P21 , and P29 were immunostained for VGLUT1 . The illustration shows the pictured area ( boxed ) . ( A ) Representative pictures from CA3 regions . SL: stratum lucidum ( DGC–CA3 synapses ) ; SR: stratum radiatum ( CA3–CA3 synapses ) . Density and size of VGLUT1 puncta are quantified and shown in ( B: SL ) and ( D: SR ) . ( C ) Time course of VGLUT1 clustering in the SL layer . In the SL layer , Igf2-/- mice show no presynaptic defects at P8 , but start to show mild defects at P14 , and exhibit significant defects at P21 and P29 . No significant defects were observed in the SR layer . Error bars are s . e . m . Data are from ( B and C ) 23–42 fields from 3–5 independent experiments , ( D ) 13–34 fields from 3–5 independent experiments . Significant difference from control at *p<0 . 05 and ***p<0 . 0001 by Student's t-test . Scale bar , 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 12151 . 01910 . 7554/eLife . 12151 . 020Figure 9—figure supplement 1 . The targeting of DGC axons to CA3 appears normal in Igf2-/- mice . Hippocampal sections from WT and Igf2-/- mice at P21 were immunostained for calbindin , which labels DGC axons ( mossy fibers ) . No apparent defects were found in the targeting of mossy fibers to the CA3 region . Scale bar , 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 12151 . 020 To further understand the role of IGF2 in synaptic development , we analyzed ultrastructure of excitatory ( asymmetric ) synapses in the SL and SR layers of CA3 in Igf2-/- mice at P28–P29 . Both in the SL and SR layers , the number and size of postsynaptic densities were similar between WT and Igf2-/- mice ( Figure 10 ) . In the presynaptic terminals in the SL layer of Igf2-/- mice , there were fewer synaptic vesicles , less clustering of synaptic vesicles , fewer docked vesicles , and smaller synaptic vesicles relative to WT mice ( Figure 10A and B ) . In contrast , no structural defects were found in the SR layer of Igf2-/- mice ( Figure 10C and D ) . Together with the immunostaining results ( Figure 9 ) , our results suggest that in Igf2-/- mice , synaptic vesicles were not stabilized/maintained in the presynaptic terminals specifically at the DGC–CA3 synapses . 10 . 7554/eLife . 12151 . 021Figure 10 . Electron microscopic analysis shows defects in excitatory presynaptic terminals selectively in the SL layer of Igf2-/- mice . Electron microscopic analysis of asymmetric ( excitatory ) synapses in WT and Igf2-/- mice ( P28–29 ) . ( A and B ) Asymmetric synapses in the CA3 SL layer . ( A ) Two representative images of asymmetric synapses in WT and Igf2-/- mice . ( B ) Quantification of synaptic vesicles ( SVs ) and postsynaptic densities ( PSDs ) . Number of SVs within 400 nm from the active zone ( total SV ) , % SVs within 150 nm from active zone , number of docked vesicles per synapse , size of SVs , number of PSDs in 100 nm2 , and length of PSDs are shown . ( C and D ) Asymmetric synapses in the CA3 SR layer . ( C ) Representative images of asymmetric synapses in WT and Igf2-/- mice . ( D ) Quantification of SVs and PSDs . Igf2-/- mice show a loss of SVs in the SL layer , but not in the SR layer , without apparent changes in PSDs . Error bars indicate s . e . m . Data are from 14–50 synapses from 12–13 fields from 2 mice per strain . Significant differences from WT mice at ***p<0 . 0001 by Student's t-test . Scale bars , 100 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 12151 . 021 To address the physiological consequences of IGF2 deficiency in synaptic function , we recorded synaptic current from the CA3 region of the hippocampus from adult WT and Igf2-/- mice . The frequency but not the amplitude of miniature excitatory postsynaptic currents ( mEPSCs ) was significantly decreased in CA3 pyramidal neurons of adult Igf2-/- mice ( Figure 11A and B ) , suggesting that loss of IGF2 has a prolonged impact on excitatory presynaptic function , without significantly affecting postsynaptic function . To examine whether synaptic defects are specific to DGC–CA3 connections , we recorded evoked field excitatory postsynaptic potentials ( fEPSPs ) at DGC–CA3 or CA3–CA3 synapses ( Figure 11C ) . fEPSP responses were verified as DGC–CA3 or CA3–CA3 based on the sensitivity to DCG-IV , the group 2 mGluR agonist that selectively blocks DGC–CA3 responses ( red traces in Figure 11C; Nicoll and Schmitz , 2005 ) . The fEPSP slope was significantly smaller in Igf2-/- than WT mice for DGC–CA3 responses ( Figure 11C ) , but not for CA3–CA3 responses . Paired-pulse facilitation at DGC–CA3 , but not CA3–CA3 , synapses was decreased in Igf2-/- mice relative to WT mice ( Figure 11D ) . Taken together , these results suggest that excitatory synaptic transmission at DGC–CA3 synapses is specifically impaired in Igf2-/- mice due to the loss of synaptic vesicles from the presynaptic terminals . 10 . 7554/eLife . 12151 . 022Figure 11 . Excitatory synaptic transmission at DGC–CA3 synapses , but not CA3–CA3 synapses , is specifically impaired in Igf2-/- mice . ( A and B ) Whole-cell recordings of mEPSCs from CA3 pyramidal neurons of adult WT and Igf2-/- hippocampal slices . ( A ) Representative traces of mEPSCs . ( B ) Quantification of the frequency and amplitude of mEPSCs . The frequency , but not amplitude , of mEPSCs is specifically decreased in CA3 of Igf2-/- mice . 31–32 cells from 5 mice per genotype . *p<0 . 05 by Student’s t-test . ( C ) fEPSP responses evoked in the CA3 region of the hippocampus . Responses were characterized as either DGC–CA3 or CA3–CA3 based upon sensitivity to DCG-IV treatment . Black traces: original responses , red traces: after 10 min of DCG-IV treatment . Graphs show the quantification of the maximum fEPSP slope . At DGC–CA3 synapses , the maximum elicited response is significantly smaller in Igf2-/- mice than in WT mice ( n = 8 and 8; **p<0 . 01 by Student’s t-test ) . At CA3–CA3 synapses , the maximum elicited response is not different between WT and Igf2-/- mice ( n = 14 and 24; p = 0 . 71 by Student’s t-test ) . ( D ) Paired pulse facilitation ( PPF ) . PPF at the DGC–CA3 synapses , and not CA3–CA3 synapses , is significantly decreased in Igf2-/- mice ( ***p<0 . 0001 by Two-way ANOVA followed by a Tukey test ) . Example traces demonstrate responses with a 50 ms inter-stimulus interval . DOI: http://dx . doi . org/10 . 7554/eLife . 12151 . 022
In Drosophila motor neurons , target-derived Gbb signaling induces the expression of Trio in the presynaptic neurons; Trio is then transported back to the nerve terminal and stabilizes cytoskeletal structures of nerve terminals ( Ball et al . , 2010 ) . In contrast , in the mammalian brain , such feedback pathways for presynaptic stabilization had not been identified . We focused on FGF signaling , because gene expression is one of the most significant outcomes of FGF signaling ( Dorey and Amaya , 2010; Partanen , 2007 ) . Thus , FGF-dependent gene expression is poised to have important effects on neuronal network development . Here , we have shown that the FGF22–IGF2 signaling serves as a feedback pathway important for the stabilization of DGC presynaptic terminals in the mammalian hippocampus . An important next question is how IGF2 stabilizes presynaptic terminals . IGF2 , which often acts as an autocrine factor ( Pollak et al . , 2004 ) , mediates its functions mainly via IGF2 receptor ( IGF2R ) as well as via IGF1 receptor ( IGF1R ) ( Fernandez and Torres-Aleman , 2012 ) . In neurons , both receptors are used to mediate IGF2 signal and participate in various physiological functions ( Agis-Balboa et al . , 2011; Chen et al . , 2011; Schmeisser et al . , 2012 ) . We showed that IGF2 is secreted and tethered on the surface of presynaptic terminals of DGCs ( Figure 4C ) . IGF2R ( Figure 4—figure supplement1 ) and IGF1R ( Gazit et al . , 2016 ) are also localized at presynaptic terminals . These results suggest that IGF2 is secreted from the presynaptic terminal and binds to IGF2R/IGF1R , which is also localized at the presynaptic terminal , so that IGF2 acts locally in an autocrine manner for its effects . Yet , we cannot exclude the possibility that IGF2 may also act as a global presynaptic organizer if IGF2 is released from the presynaptic terminal . IGF1R is a receptor tyrosine kinase , which may regulate local translation , and IGF2R is known to signal through G proteins , which may ultimately affect calcium homeostasis . Calcium homeostasis and local translation of synaptic proteins , as well as regulation of cytoskeletal structures like Trio may all contribute to IGF2-dependent presynaptic stabilization . Neural activity is involved in synaptic stabilization . It has been proposed that active synapses are stabilized and inactive ones are destabilized ( Lichtman and Colman , 2000; Ruthazer and Cline , 2004 ) . Various forms of activity are involved: in most cases , synaptic transmission and synaptic competition are considered critical for synapse refinement ( Lichtman and Colman , 2000; Waites et al . , 2005 ) . In addition , intrinsic activity also plays critical roles in synapse development ( Johnson-Venkatesh et al . , 2015 ) . However , the molecular mechanisms by which intrinsic activity contributes to synapse stabilization during development are largely unknown . We found that IGF2's function in synaptic stabilization requires intrinsic neuronal activity ( Figure 6 ) . IGF2 is not transported to presynaptic terminals when intrinsic neuronal excitability of DGCs is suppressed . Thus , our results reveal that IGF2 is a mediator of intrinsic activity-dependent presynaptic stabilization . Since IGF2's effects are for the stabilization of synaptic vesicles but not for the initial recruitment of synaptic vesicles ( which is carried out by FGF22 ) , it is reasonable that TTX/Kir2 . 1 did not affect synaptic vesicle transportation ( Figure 6 ) . How does intrinsic activity regulate IGF2 localization ? Neural activity is known to control motor function of KIF proteins and regulate intracellular transport of synaptic components , such as mitochondria and AMPA receptors ( Hoerndli et al . , 2015; Saxton and Hollenbeck , 2012 ) . Neural activity also modulates phosphorylation of the C terminal domain of KIF3A , which affects loading of N-cadherin containing cargos and contributes to the maintenance of homeostatic synaptic plasticity ( Ichinose et al . , 2015 ) . Thus , neural activity may regulate the IGF2 transport complex through phosphorylation of motor proteins . It will be interesting to investigate the mechanisms of IGF2 transport to presynaptic terminals and how neural activity regulates the transport . In addition , neural activity may regulate the secretion of IGF2 . Neural activity controls exocytotic secretion of cytoplasmic vesicles in neurons: in olfactory bulb neurons , K+-induced depolarization activates the exocytotic Ca2+-sensor , synaptotagmin-10 , to induce secretion of IGF1 ( Cao et al . , 2011 ) . Thus , in addition to transportation , secretion of IGF2 at presynaptic sites might also be controlled by neural activity . If this is the case , one may speculate that activity blockade would increase the intracellular IGF2 levels . However , we did not observe an increase in the intracellular IGF2 levels , probably because neural activity is also critical for IGF2 transportation ( Figure 6 ) . It is also possible that various activity-dependent genes might influence IGF2/IGF2R localization , which is an interesting future study . Our Kir2 . 1 experiments , in which Kir2 . 1 was sparsely transfected so that we can ignore effects from postsynaptic neurons , suggest that changes in postsynaptic activity do not play critical roles in IGF2 localization and function . In addition to presynaptic stabilization , signals propagated from target-derived molecules may influence further cell-wide development of the presynaptic neurons . In the mammalian brain , signaling from target-derived neurotrophins has been well characterized . For example , nerve growth factor ( NGF ) signaling regulates gene expression , including TrkA , p75 , Bdnf , and Ntf5 , and controls cell survival and death of own and neighboring neurons ( Ascano et al . , 2012; Deppmann et al . , 2008; Singh et al . , 2008 ) as well as dendritic development ( Sharma et al . , 2010 ) . Since IGF2 has been shown to exhibit broad functions in the brain , IGF2 may not only contribute to presynaptic stabilization , but also various aspects of cell-wide development , including neurogenesis , neurite growth , and spine maturation . For neurogenesis , IGF2 is implicated in maintenance and expansion of neural stem cells ( Ziegler et al . , 2012; 2014 ) and proliferation of neuronal progenitor cells ( Burns and Hassan , 2001; Lehtinen et al . , 2011 ) in the developing brain . IGF2 also contributes to adult neurogenesis in the subgranular zone of the hippocampus ( Bracko et al . , 2012; Ouchi et al . , 2013 ) . For neuronal development , IGF2 leads to nerve sprouting ( Caroni and Grandes , 1990 ) , neurite outgrowth ( Jeong et al . , 2013 ) , as well as spine maturation in cultured hippocampal neurons ( Schmeisser et al . , 2012 ) . Thus , IGF2 may serve as a general regulator of the development of presynaptic neurons ( DGCs ) downstream of FGF22 . Since IGF2 is still expressed in calbindin-positive DGCs in Fgf22-/- mice ( Figure 1 ) , IGF2 can be expressed in an FGF22-independent manner as well . Indeed , IκB , which is not utilized by FGF22 signaling , has been shown to induce expression of IGF2 for spine maturation in mature hippocampal neurons ( Schmeisser et al . , 2012 ) . The expression of the mouse Igf2 gene is regulated by three alternative promoters ( Sasaki et al . , 1992 ) . Thus , different signals seem to be used to regulate distinct phases of neuronal development . The presynaptic effects of IGF2 are stage and cell-type specific . FGF22-dependent IGF2 expression was observed in young , calretinin-positive DGCs , but not in mature , calbindin-positive DGCs ( Figures 1–3 ) . Specific responsiveness of calretinin-positive DGCs to FGF22 appears to be linked to the developmental stage . Calretinin-positive DGCs elongate axons to CA3 and contact CA3 pyramidal neurons to form synapses ( Aguilar-Arredondo et al . , 2015; Li et al . , 2009; Ming and Song , 2005; Yasuda et al . , 2011 ) . Thus , around that stage , DGCs may become more responsive to FGF22 . Our results are consistent with the idea that IGF2 is induced by CA3-derived FGF22 when DGC axons contact with their target dendrites . The data from CA3-selective Fgf22 knockout mice ( Figure 2 ) and the local FGF22 application experiments , in which IGF2 was induced by axonal application of FGF22 ( Figure 3F–I ) , further support this idea . Our results also suggest that IGF2 is not critical for initial synaptic differentiation , because no defects were found in the absence of IGF2 at 6DIV in vitro ( Figure 8A , B ) or P8 in vivo ( Figure 9 ) . However , it is important for the later stages of synapse development: after 6DIV in vitro ( Figure 8E , F ) and P14 in vivo ( Figure 9 ) , when neural activity influences synapse formation ( Toth et al . , 2013; Yasuda et al . , 2011 ) . As IGF2 effects are activity dependent ( Figure 6 ) , we propose that IGF2 , induced by target-derived FGF22 , is mainly important for the stages of activity-dependent synapse stabilization . Interestingly , induction of IGF2 by FGF22 is specific to DGCs . CA3 pyramidal neurons , which release FGF22 , receive excitatory synaptic inputs from collateral/associational CA3 pyramidal neurons and stellate cells in the entorhinal cortex , in addition to inputs from DGCs ( Urban et al . , 2001 ) . However , CA3 neurons do not respond to FGF22 to induce IGF2 ( Figure 3 ) . Since Fgf22-/- mice show defects in excitatory synapse formation both in the SL ( where DGC to CA3 synapses are located ) as well as SR ( where CA3 to CA3 synapses are located ) ( Terauchi et al . , 2010 ) , the lack of IGF2 induction in CA3 neurons indicates that IGF2 is a unique target of FGF22 specifically in differentiating DGCs . It is possible that CA3 neurons express genes other than IGF2 in an FGF22-dependent manner for presynaptic stabilization . If so , different types of presynaptic neurons might process FGF22 signaling through their own transcription regulation to promote neuron type-specific presynaptic stabilization . It is worth noting that many synapse organizing molecules show input specificity . FGF22 affects synapse formation in the SR and SL layers of CA3 , but not the SLM ( stratum lacunosum moleculare ) layer ( Terauchi et al . , 2010 ) . Neuroligins in cerebellar Purkinje cells have a role in the formation of climbing fiber synapses , but not parallel-fiber synapses ( Zhang et al . , 2015 ) . Cadherin-9 is involved in the formation and differentiation of DGC–CA3 synapses , but not CA3–CA3 synapses ( Williams et al . , 2011 ) . Neuroligin2 regulates inhibitory synaptic function in a pathway-specific manner ( Gibson et al . , 2009 ) . Thus , it appears that there are input specific synaptic organizers that cooperate to establish precise networks in the brain . Our work on IGF2 , which is specific to DGC–CA3 synapses , would add an example of how our brain utilizes specific molecules to emerge synapse specificity . It will be also interesting to examine whether stabilization of inhibitory synapses is regulated by a similar feedback pathway . In CA3 , FGF7 acts as a target-derived presynaptic organizing molecule for inhibitory synapses ( Terauchi et al . , 2010 ) . Our preliminary screen suggests that FGF22 and FGF7 may regulate distinct molecules in excitatory ( DGCs ) vs . inhibitory neurons for synapse stabilization , because Igf2 was not identified as a target gene of FGF7 . IGF2 is implicated in behavioral phenotypes such as fear extinction ( Agis-Balboa et al . , 2011; Agis-Balboa and Fischer , 2014 ) , depression ( Luo et al . , 2015 ) , and memory consolidation and enhancement in rodents ( Chen et al . , 2011; Pascual-Lucas et al . , 2014 ) . Administration of IGF2 rescues spine formation and excitatory synaptic function in the hippocampus of a mouse model of Alzheimer’s disease ( Pascual-Lucas et al . , 2014 ) . In addition , Igf2-/- mice are resistant to acquiring epileptiform events in response to kainate administration ( Dikkes et al . , 2007 ) . Interestingly , Fgf22-/- mice show seizure resistant phenotype ( Terauchi et al . , 2010 ) and depression-like behavior ( Williams et al . , 2016 ) . Thus , defects in synapse stabilization through the FGF22–IGF2 pathway may be involved in diseases like epilepsy and depression . Our work may help develop new treatment strategies for such neuropsychiatric disorders .
Fgf22-/- mice were described previously ( Terauchi et al . , 2010 ) . The strain was maintained on the C57/BL6 background . Fgf22flox/flox mice ( Fgf22tm1a ( EUCOMM ) Hmgu ) were from EUCOMM . Grik4-Cre mice were from Jackson ( Nakazawa et al . , 2002 ) . Igf2-/- mice were described previously ( Lehtinen et al . , 2011 ) . Both males and females of these knockout and littermate control mice were used in our study . C57/BL6 or ICR mice ( Jackson Laboratory and Charles River Laboratories ) were used to prepare cultures . All animal care and use was in accordance with the institutional guidelines and approved by the Institutional Animal Care and Use Committees at Boston Children’s Hospital and University of Michigan . DGCs were dissected from P14 WT and Fgf22-/- mice ( n=4 per genotype ) . RNA was prepared with the RNeasy kit ( Qiagen , Germantown , MD ) , and its quality was verified with Agilent Bioanalyzer using PICO chips . Microarray was performed on Illumina Bead station 500 with mouse-6 expression beadchip . Igf2 was identified as one of the most down-regulated genes in Fgf22-/- DGCs ( Diff Score = -43; see Table 1 ) . In situ hybridization was performed as described ( Schaeren-Wiemers and Gerfin-Moser , 1993 ) . Digoxigenin-labeled cRNA probes were generated by in vitro transcription using DIG RNA labeling mix ( Roche , Switzerland ) . The probe for Igf2 was generated from the 3' untranslated region of the mouse Igf2 cDNA ( Open Biosystems , Lafayette , CO ) . In situ images were taken with a Nikon Coolpix 990 distal camera ( Nikon , Japan ) attached to an Olympus BX61 upright microscope ( Olympus , Japan ) under bright-field optics with 4x , 10x and 20x objective lenses . The expression plasmid for IGF2 was generated by subcloning the full-length mouse Igf2 cDNA into the NheI/XhoI sites of APtag-5 ( GenHunter , Nashville , TN ) . The expression plasmid for IGF2-EGFP was generated by subcloning the full-length mouse Igf2 cDNA ( minus the stop codon ) into the NheI/HindIII sites of pEGFP-N1 ( Clontech , Mountain View , CA ) in frame . The expression plasmid for Kir2 . 1 was generated by subcloning the full-length Kir2 . 1 cDNA into SmaI/XhoI sites of pCMV-SPORT6 . The IGF2 specific shRNA expression plasmids were constructed using synthetic oligonucleotides , which were cloned into the BamHI/HindIII sites of the HuSH shRNA vector ( pRFP-C-RS , OriGene , Rockville , MD ) . The following shRNA targeting sequences were used: IGF2-shRNA#1 , CGGACCGCGGCTTCTACTTCAGCAGGCCT and IGF2-shRNA#2 , GTTGGTGCTTCTCATCTCTTTGGCCTTCG . The efficiency of shRNA-mediated knockdown of IGF2 was confirmed by co-transfecting each IGF2 shRNA plasmid with the IGF2-EGFP plasmid into cultured hippocampal neurons and measuring the total EGFP fluorescence intensity relative to control shRNA transfected neurons ( Figure 8—figure supplement 1 ) . The synaptophysin-YFP plasmid was described previously ( Terauchi et al . , 2010; Toth et al . , 2013; Umemori et al . , 2004 ) . The synaptophysin-mCherry plasmid was a kind gift from M . Sutton ( University of Michigan ) . Hippocampi were dissected from P0 mice , and hippocampal cells were dissociated in a solution containing 0 . 5% trypsin and 0 . 02% DNase I as described previously ( Terauchi et al . , 2010 ) . 3–5 x 104 hippocampal cells were plated on a poly-D-lysine coated glass coverslip ( diameter 12 mm , No . 1 , Carolina Biological , Burlington NC ) and cultured in neurobasal media supplemented with B27 ( Invitrogen , Waltham , MA ) . Transfection was performed using the CalPhos Mammalian transfection kit ( Clontech ) . Cultured cells were transfected at 1–3DIV with 1 . 5–2 . 2 μg of plasmid DNA per coverslip . Recombinant FGF22 and IGF2 ( both from R&D systems , Minneapolis , MN ) were applied at 2 nM ( FGF22 ) or 1 . 35 nM ( IGF2 ) into culture medium at 1–3DIV . For neuronal activity blockade , TTX was applied at 1 μM into culture medium every fourth day stating at the time of transfection or factor application . The chambers fabricated in polydimethylsiloxane ( Xona Microfluidics SND450 , Xona Microfluidics , Temecula , CA ) were placed on a poly-D-lysine coated glass coverslip ( 25 x 25 mm , No . 1 , Carolina Biological ) by physical contact . The chamber consists of two microfluidic compartments , somal and axonal sides , which are connected via microgrooves with a high fluidic resistance . Hippocampal cells ( 22 , 500 cells ) were plated in the somal side compartment . Fluidic isolation of the axonal side compartment was established by applying 180 μl of media in the somal side and 110 μl in the axonal side . At 2DIV , 2 . 5 nM of FGF22 was applied into the axonal compartment . Cells were stained at 8DIV . Mouse brains were perfused with 4% paraformaldehyde ( PFA ) in PBS followed by further fixation in 4% PFA in PBS overnight . Sagittal and coronal sections were prepared on a cryostat ( 16–20 μm thick ) , and processed for staining . For IGF2 staining , sections were treated with methanol for 5 min at -20°C . For IGF2 staining together with calbindin or calretinin staining , sections were treated with acetone for 2 min at -20°C . Cultures were fixed with 3 or 4% PFA for 10 min at 37°C or methanol for 2–5 min at -20°C and stained as described previously ( Terauchi et al . , 2010 ) . For immunostaining for IGF2 , cultures were fixed with acetone for 2–3 min at -20°C . Dilutions and sources of antibodies are: monoclonal anti-calbindin ( 1:200; Sigma-Aldrich , St . Louis , MO; C9848 ) , goat anti-calbindin ( 1:500; Frontier Institute , Japan ) , monoclonal anti-calretinin ( 1:200; Millipore , Billerica , MA; MAB1568 ) , rabbit anti-calretinin ( 1:500; Synaptic Systems , Germany; 214102 ) , monoclonal anti-Prox1 ( 1:1500; Millipore; MAB5652 ) , rabbit anti-Prox1 ( 1:500; Millipore; AB5475 ) , anti-VGLUT1 ( 1:5000; Millipore; AB5905 ) , anti-VGAT ( 1:1500; Synaptic Systems; 131003 ) , anti-PSD95 ( 1:700; Affinity Bioreagents , Golden , CO; MA-045 and 1:250; NeuroMab , Davis , CA; 75–028 ) , anti-gephyrin ( 1:150; Synaptic Systems; 147021 ) , anti-MAP2 ( 1:3000; Sigma-Aldrich; M4403 ) , anti-neurofilament ( 1:1000; Covance , Princeton , NJ; SMI-312 ) , rabbit anti-GFP ( 1:1000; Millipore; AB16901 ) , chicken anti-GFP ( 1:2500; Aves Labs , Tigard , OR; GFP-1020 ) , anti-DsRed ( for staining of mCherry to enhance the fluorescence signal , 1:500; Clontech; 632496 ) , anti-IGF2 ( 1:50 for brain section staining and 1:70 for cultured cell staining; Santa Cruz , Dallas , TX; sc-5622 ) , anti-IGF2R ( 1:100; Santa Cruz; sc-25462 ) , and antibody Py ( 1:50; a kind gift from M . Webb and P . L . Woodhams ) . When cells were co-stained with anti-Prox1 monoclonal antibody with another mouse IgG1 antibody , Zenon Alexa Fluor 568 Mouse IgG1 Labeling Kit ( Invitrogen ) was used to label anti-Prox1 monoclonal antibody . P28–P29 WT and Igf2-/- mice were perfused transcardially with fixative ( 2% PFA and 2 . 5% glutaraldehyde in 0 . 1 M cacodylate buffer pH 7 . 4 ) , and their brains were postfixed for overnight . Hippocampi were removed , cross-sections ( 250 μm thick ) were prepared , and small pieces ( about 0 . 6 mm x 0 . 6 mm x 0 . 25 mm ) of the SL and SR layers of the CA3 region were dissected . Dissected pieces were postfixed overnight , washed with 0 . 1 M cacodylate buffer pH 7 . 4 , and treated with 2% osmium tetroxide in 0 . 1 M cacodylate buffer or 1 . 5% potassium ferrocyanide , 2% osmium tetroxide in 0 . 1 M cacodylate buffer for 1–5 hr . The samples were then rinsed with water , stained en bloc with 3% uranyl acetate for 1 hr , dehydrated in graded alcohols and propylene oxide , and embedded in TAAB 812 Resin ( Canemco-Marivac , Canada ) . Blocks were kept for 48 hr at 60°C to complete polymerization . Both semi- and ultra-thin sections ( 10 and 70 nm ) were prepared with Diatome Histo and Diatome Ultra 45° diamond knives , respectively , on Leica UC7 ultramicrotome ( Leica , Germany ) , and observed with Tecnai G2 Spirit BioTWIN Transmission Electron Microscope . The digital images were captured with AMT 2k CCD camera system operated with AMT software ( Advanced Microscopy Techniques Corp . , Woburn , MA ) . For mEPSC recordings: Acute hippocampal slices were prepared from 5–7 month old mice . Mice were decapitated and the brains were removed , and 300 µm sections were cut using a Leica VT1000S vibratome . Sections were cut in an ice cold solution containing ( in mM ) : 206 sucrose , 2 . 8 KCl , 2 MgSO4 , 1 MgCl2 , 1 . 25 NaH2PO4 , 1 CaCl2 , 10 glucose , 26 NaHCO3 , and 0 . 4 ascorbic acid . Then , sections were incubated in an NMDG-HEPES recovery solution , containing ( in mM ) : 92 NMDG , 92 HCl , 2 . 5 KCl , 10 MgSO4 , 0 . 5 CaCl2 , 1 . 2 NaH2PO4 , 20 HEPES , 30 NaHCO3 , 25 glucose , 5 sodium ascorbate , 2 thiourea , and 3 sodium pyruvate , for 15 min at 34°C before putting the slices into artificial cerebral spinal fluid ( aCSF ) for 1 hr at room temperature . aCSF contained ( in mM ) : 124 NaCl , 2 . 8 KCl , 2 MgSO4 , 1 . 25 NaH2PO4 , 2 CaCl2 , 10 glucose , 26 NaHCO3 , and 0 . 4 ascorbic acid . All solutions were continuously bubbled with 95% O2/5% CO2 . Neurons were visualized using a customized Scientifica/Olympus microscope . Data were obtained with a Multiclamp 700B amplifier ( Axon Instruments , Union City , CA ) , digitized with Digidata 1440A ( Axon Instruments ) and collected with Clampex 10 . 0 ( Axon Instruments ) . Whole-cell patch-clamp recordings were conducted with 4–6 MΩ pipette containing ( in mM ) 135 K-MeSO4 , 7 NaCl , 10 HEPES , 4 Mg-ATP , 0 . 3 Li-GTP , and 7 phosphocreatine . Cells were held at -70 mV . aCSF was supplemented during recording with 500 nM tetrodotoxin and 50 µM picrotoxin and warmed to 32°C . mEPSCs were analyzed using Minianalysis ( Synaptosoft , Decatur , GA ) . For fEPSP recordings: mice ( 2–3 months old ) were decapitated and the hippocampal lobules cut in the same solution as above . Transverse slices ( 400 µm ) of the hippocampus were then cut using a tissue chopper ( Stoelting , Kiel , WI ) . After slicing , sections were incubated in an NMDG-HEPES recovery solution for 15 min at 34°C before putting the slices into aCSF for 1 hr . Slices were then incubated in aCSF at room temperature for at least 1 hr before recording . Then , slices were transferred to a recording chamber , maintained at 32°C and continuously perfused at 1–2 ml/min with oxygenated aCSF . Recording electrodes were pulled from borosilicate capillary glass and filled with 1 M NaCl , 25 mM HEPES ( 1 . 5 mm o . d . ; Sutter Instruments , Novato , CA ) . The recording pipette was placed in the in the SL or SR layers of the CA3 region of the hippocampus . Recordings were made with a MultiClamp 700B amplifier , collected using Clampex 10 . 3 , and analyzed using Clampfit 10 . 3 . fEPSPs were evoked using cluster electrodes ( FHC ) placed in the SL or SR layers of the CA3 region of the hippocampus . Current between 0 . 1–1 mA for 0 . 1 s was used to elicit a response . Maximum responses were then used for paired-pulse facilitation experiments . 1 µM DGC-IV was then added to the aCSF and the experiments were repeated after 10 min of perfusion with drug . Fluorescent images were taken on epi-fluorescence microscopes ( Olympus BX61 and BX63 ) or confocal microscopes ( Olympus FV1000 and Carl Zeiss LSM700 , Germany ) . With epi-fluorescence microscopes , 12-bit images at a 1 , 376 x 1 , 032 ( Olympus BX61 ) or 1 , 376 x 1 , 038 ( Olympus BX63 ) pixel resolution were acquired with 40x and 20x objective lenses using an F-View II CCD camera ( Soft Imaging System , Germany ) or an XM10 Monochrome camera ( Olympus ) . With confocal microscopes , 12-bit ( FV1000 ) or 8-bit ( LSM700 ) images at a 1 , 024 x 1 , 024 pixel resolution were obtained using 20x , 40x and 63x objective lenses with a 1 . 0 or 1 . 5x zoom . Images were acquired as a z-stack ( 17–20 optical sections , 0 . 4 μm step size ) . Images in the same set of experiments were acquired with the identical acquisition settings regarding the exposure time , laser power , detector gain , or amplifier offset . The intensity of stained signals and the size and density of stained puncta were quantified and analyzed using MetaMorph software . For images of hippocampal sections , the staining intensity in the lateral ventricle was calculated as the background signal and subtracted from each image . For images of cultured neurons stained for synaptic proteins , the staining intensity of the dendritic shaft in control cultures was calculated as the background signal and subtracted from each image . For images of cultured neurons stained for anti-IGF2 antibody , the lowest staining intensity in the culture was calculated as the background signal and subtracted from each image . For colocalization analyses , objects were considered to colocalize if more than 25% of the object was overlapped with the other object . The statistical tests performed were two-tailed Student’s t-test or Two-way ANOVA , as indicated in the figure legend . Two-way ANOVA was followed by Tukey's post hoc test . All data are expressed as mean ± s . e . m . No statistical methods were used to pre-determine sample sizes , but our sample sizes were similar to those reported in previous publications in the field ( Murata and Constantine-Paton , 2013; Sharma et al . , 2010; Terauchi et al . , 2010; Toth et al . , 2013 ) . No data points were excluded from any experiments . | Nerve cells in the developing brain must organize themselves into complex networks by forming appropriate connections with one another . These connections are known as synapses , and they assemble via two critical stages . First , a new synapse forms , and then it stabilizes . This first stage is a localized event that involves the contact site between the two nerve cells , while the stabilization of a synapse requires the expression of genes in a nerve cell’s nucleus . Furthermore , only active synapses may be stabilized . Many synapses form in a region of the brain called the hippocampus , which plays a key role in learning and memory . A protein called fibroblast growth factor 22 ( or FGF22 for short ) helps synapses to initially form within the hippocampus . However , much less is known about the signals that regulate the stabilization of synapses and the genes that are involved . It is also not clear if these genes might be controlled by FGF22 signaling . To address these questions , Terauchi et al . searched the mouse hippocampus for genes with expression that depended on FGF22 signaling . One gene in particular , which encodes a protein called insulin-like growth factor 2 ( IGF2 ) , was much less expressed in mice that lack FGF22 compared to normal mice . Further experiments revealed that only active nerve cells transport IGF2 to synapses , and that IGF2 helps to stabilize these structures . By contrast , IGF2 is not required for synapse to initially form . This indicates that FGF22 controls both the formation and stabilization of synapses , and that it controls the first stage directly , and the second stage indirectly via its effects on IGF2 expression . Terauchi et al . also showed that FGF22-IGF2 signaling is not involved in the stabilization of all synapses in the mouse hippocampus . Instead , synapses between different types of nerve cell appear to use distinct signals for synapse formation and stabilization . A key topic for future studies will be to understand these specific signals and how they cooperate in the brain to establish precise networks of nerve cells . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology",
"neuroscience"
] | 2016 | Retrograde fibroblast growth factor 22 (FGF22) signaling regulates insulin-like growth factor 2 (IGF2) expression for activity-dependent synapse stabilization in the mammalian brain |
Transcription is tightly regulated to maintain energy homeostasis during periods of feeding or fasting , but the molecular factors that control these alternating gene programs are incompletely understood . Here , we find that the B cell lymphoma 6 ( BCL6 ) repressor is enriched in the fed state and converges genome-wide with PPARα to potently suppress the induction of fasting transcription . Deletion of hepatocyte Bcl6 enhances lipid catabolism and ameliorates high-fat-diet-induced steatosis . In Ppara-null mice , hepatocyte Bcl6 ablation restores enhancer activity at PPARα-dependent genes and overcomes defective fasting-induced fatty acid oxidation and lipid accumulation . Together , these findings identify BCL6 as a negative regulator of oxidative metabolism and reveal that alternating recruitment of repressive and activating transcription factors to shared cis-regulatory regions dictates hepatic lipid handling .
The classical studies of Jacob and Monod on the bacterial lac operon established a central paradigm for transcriptional repression to direct metabolic responses and sustain life in an environment of discontinuous food supply ( Jacob and Monod , 1961; Payankaulam et al . , 2010 ) . In metazoans , nutrient-elicited transcription likewise coordinates the feeding to fasting transition of metabolism , yet a gap remains in our knowledge of the participating factors and their genomic coordination . In the fed state , sterol and carbohydrate regulatory element-binding proteins ( SREBP and ChREBP ) direct lipogenesis and glycolysis ( Abdul-Wahed et al . , 2017; Horton et al . , 2002 ) . Conversely , fasting disinhibits forkhead box transcription factors ( FOXOs ) and activates glucocorticoid receptor ( GR ) and cAMP response element binding protein ( CREB ) to promote gluconeogenesis ( Rui , 2014 ) . Extended fasting further stimulates peroxisome proliferator-activated receptor alpha ( PPARα ) to induce fatty acid oxidation , ketogenesis , and the fasting hormone FGF21 ( Badman et al . , 2007; Inagaki et al . , 2007; Kersten et al . , 1999; Leone et al . , 1999 ) . Despite progress revealing these various transcriptional activators , their dynamic genome-wide regulation and the influence of additional factors , particularly repressors , on the feeding to fasting transition remains poorly understood ( Goldstein and Hager , 2015 ) . Recently , fasting-regulated enhancers were mapped using H3K27 acetylation ChIP- and DNase I hypersensitivity sequencing and footprinting , which inferred the presence of unknown repressors at regions enriched with STAT motifs ( Goldstein et al . , 2017 ) . Our focus turned to B-cell lymphoma 6 ( BCL6 ) , a key immune cell repressor with affinity for STAT-like DNA recognition sequences ( Dent et al . , 1998; Dent et al . , 1997; Zhang et al . , 2012 ) . BCL6 is a member of the ZBTB family of C2H2-type zinc finger proteins and represses transcription through a variety of interactions with corepressors including SMRT , NCoR , BCoR , CtBP , MTA3/NuRD , and HDACs ( Basso and Dalla-Favera , 2012 ) . Although well-recognized for critical roles in B-cell and T-cell development and lymphomagenesis , BCL6 is also broadly expressed outside of the immune system where its functions are largely unknown . In this work , using genome-wide DNA binding and transcriptomic analyses as well as hepatocyte-specific gene targeting , we reveal an unexpected role for BCL6 as a potent antagonist of PPARα-directed gene regulation . We find that BCL6 and PPARα bind independently at thousands of shared regulatory regions in sub-nucleosomal proximity , often at multiple locations along the same gene . Genes harboring these BCL6-PPARα regulatory modules constitute over 50% of fasting-responsive transcripts and exhibit particularly dynamic expression . Moreover , we find that ablation of hepatocyte Bcl6 increases lipid oxidation , prevents high-fat-diet-induced steatosis , and reverses fasting-related defects in Ppara-/- mice including aberrant enhancer activity , transcription , ketosis , and lipid accumulation . These restorations in Ppara-/- mice devoid of liver Bcl6 were linked to loss of HDAC3-containing BCL6 repressive complexes and enhanced recruitment of PPARδ to BCL6-PPAR shared enhancers . Together , these findings establish BCL6 as a critical repressor of oxidative metabolism .
To establish the genomic sites for BCL6 regulation , we used ChIP-seq to map its genome-wide set of cis-acting targets ( cistrome ) in liver . Under fed conditions , we identified over fifteen thousand high confidence BCL6 binding sites from three biological replicates . Ontologies for nearby genes were dominated by lipid and ketone metabolism , PPAR signaling , and functions in peroxisomes and mitochondria ( Figure 1A ) . Additionally , motif analysis of BCL6 binding sites compared to random whole genome sequences revealed striking enrichment of response elements not only for BCL6 but also for lipid-activated PPAR nuclear hormone receptors ( Figure 1B ) ( Evans et al . , 2004 ) , the pioneer factor FOXA1 , the enhancer remodeler C/EBP ( Grøntved et al . , 2013 ) , and the developmental and lipid regulatory factors HNF4 ( Hayhurst et al . , 2001; Li et al . , 2000 ) and HNF6 ( Clotman et al . , 2005; Zhang et al . , 2016 ) . Highly similar BCL6 peak calling , gene ontology and motif analysis was obtained using either wild-type liver input chromatin or BCL6 ChIP-seq from livers of hepatocyte-specific Bcl6 knockouts ( Bcl6LKO mice ) as background controls for enrichment ( Figure 1—figure supplement 1A and B ) indicating that the liver BCL6 cistrome reflected binding events specific to hepatocytes . Based on motif predictions , we pursued the possibility of genomic convergence between BCL6 and PPARs . Direct quantification of TF consensus sites near BCL6 binding sites further reflected enrichment of motifs for PPARs , its heterodimeric partner RXR , and to a lesser extent FXR , whereas motifs for other abundant liver transcription factors such as LXR were absent ( Figure 1—figure supplement 1B ) . Ppara is the dominantly expressed PPAR subtype in liver ( Figure 1C ) . In line with PPARα’s critical role to regulate the adaptive response to fasting , its RNA and protein levels increase with overnight food deprivation ( Figure 1C–E ) ( Kersten et al . , 1999 ) . In contrast , Bcl6 mRNA and corresponding protein diminish sharply from the fed to the fasted state ( Figure 1D and E ) . Accordingly , BCL6 occupancy was diminished at the majority of its binding sites and its cistrome was reduced by 39% , whereas PPARα recruitment was enhanced and its cistrome was expanded by 36% with fasting ( Figure 1F , top and middle panels , and Figure 1—figure supplement 2A , left panel ) . In addition , fasting resulted in a redistribution of binding sites for each factor . Direct comparison of the combined fed and fasted ChIP-seq peaks for BCL6 and PPARα revealed 13 , 608 overlapping binding regions ( <200 bp between peak centers ) between these factors , representing 77% and 41% of the BCL6 and PPARα cistromes , respectively ( Figure 1F , bottom panel ) . Of these overlapping peaks , the vast majority ( >96% ) demonstrated a distance of <100 bp between peak centers ( Figure 1—figure supplement 2B ) . Over 95% of these overlapping sites occurred outside of promoter regions in intragenic and intergenic locations ( Figure 1—figure supplement 2C ) . BCL6-PPARα co-occurring peaks represented the strongest binding events for each factor , indicating they likely represent true DNA interactions as opposed to non-specific events at open chromatin regions ( Figure 1G ) ( Landt et al . , 2012 ) . At these shared sites , binding by BCL6 decreased while PPARα increased upon fasting ( Figure 1H and Figure 1—figure supplement 2A , right panel ) , which was evident at several PPARα target genes , such as Acot4/3 and Por ( Figure 1—figure supplement 2D and E ) and confirmed by ChIP qPCR ( Figure 1—figure supplement 2F ) . Thus , extensive cistromic overlap and reciprocal genome-wide binding suggested BCL6 and PPARα may control a common regulatory program . Next , we assessed whether BCL6 and PPARs compete or collaborate for DNA binding . Using livers from Ppara-/- and wild-type control mice , we found that ablation of Ppara had no impact on BCL6 enrichment at BCL6-PPARα binding sites ( Figure 2A , left panel and Figure 1—figure supplement 2D ) . Likewise , liver-specific deletion of Bcl6 did not alter PPARα binding ( Figure 2A , right panel and Figure 1—figure supplement 2E ) . Ppard is expressed at relatively low levels in liver ( Figure 1C ) , but it was previously reported that unliganded PPARδ binds and sequesters BCL6 , releasing it in the presence of PPARδ ligands ( Lee et al . , 2003 ) . Thus , to test whether a protein complex between PPARδ and BCL6 could account for BCL6-PPAR genomic co-localization , we characterized BCL6 binding in the presence or absence of hepatocyte PPARδ using mice harboring floxed alleles of Ppard and Albumin-Cre ( PpardLKO mice ) . The livers of PpardLKO animals exhibited 96% decreased levels of Ppard mRNA with no significant change in Bcl6 ( Figure 2—figure supplement 1A ) , yet in comparison to wild type control livers , BCL6 binding was unaltered across the BCL6 cistrome and at its subset of BCL6-PPARα shared sites ( Figure 2B and C ) . Thus , these findings did not support a model in which BCL6 binds to PPARs on chromatin . Additionally , we mapped the liver PPARδ cistrome using an isotype-specific antibody . 8 , 194 PPARδ-binding sites were identified collectively in fed and fasted livers , 85% of which overlapped with the more extensive PPARα cistrome of 33 , 379 sites ( Figure 2D ) . Overall , PPARδ binding was diminished by half upon fasting ( Figure 2—figure supplement 1B ) , but this reduction was only evident at sites shared with PPARα such as the Acot4/3 and Ehhadh loci ( Figure 2C ) , suggesting that PPARα and PPARδ compete for binding at common response elements ( Figure 2—figure supplement 1C ) . While PPARδ and BCL6 co-localized at only 87 genomic sites without PPARα , we detected 8 , 975 BCL6-PPARα peaks which were not bound by PPARδ ( Figure 2D ) . Gene ontology analysis revealed that BCL6-PPARα-PPARδ shared or BCL6-PPARα exclusive peaks annotate predominantly to genes controlling the metabolism of lipids and lipoproteins , fatty acids , triacylglycerol , ketone bodies , PPAR signaling , and biological oxidations ( Figure 2E ) . Collectively , these results provided further evidence that extensive BCL6 genome-wide colocalization with PPARα and , to a more limited degree , with PPARδ occurs due to independent , yet proximate DNA-binding events along genes controlling lipid metabolism . To better understand how BCL6 modulates gene expression in liver , we first identified the BCL6-regulated transcriptome . We generated mice with hepatocyte-specific Bcl6 deletion ( Bcl6LKO ) by crossing animals with floxed alleles of Bcl6 to mice expressing Cre under control of the albumin enhancer/promoter . Bcl6LKO mice exhibited 75% reduced Bcl6 mRNA and over 90% diminished protein levels in the liver ( Figure 3—figure supplement 1A and B ) . In ad lib fed Bcl6LKO mice , RNA-seq revealed 721 upregulated genes , while only 362 were downregulated by more than two fold compared to controls ( Figure 3A ) . These findings indicated that liver BCL6 predominantly functions as a repressor of transcription , which was particularly apparent at genes with strongly bound BCL6-binding sites ( Figure 3B ) . BCL6 is known to control transcription in immune cells through interactions with many different cofactors ( Barish et al . , 2012; Basso and Dalla-Favera , 2012; Hatzi et al . , 2013 ) . To test whether BCL6 regulates transcription through similar interactions in liver , we used ChIP-seq to characterize SMRT , NCoR , and HDAC3 binding in ad lib fed Bcl6fl/fl and Bcl6LKO mice . In Bcl6fl/fl animals , we found extensive cistrome overlap between BCL6 and all three corepressors ( 6 , 643 common sites ) , although 21% of BCL6 sites were unique ( Figure 3C ) . SMRT and HDAC3 exhibited very few independent binding regions , with only ~2% unique for either cofactor . NCoR exhibited the most extensive cistrome , and 45% of its sites did not overlap with BCL6 , HDAC3 , or SMRT . In line with their known biochemical interactions , SMRT , NCoR , and HDAC3 peaks were enriched in motifs for nuclear receptors ( ERR , PPAR , RXR ) as well as FOX and HNF transcription factors when compared to whole genome DNA as background ( Figure 3—figure supplement 2A ) ( Perissi et al . , 2010 ) . For each corepressor , we further analyzed peaks shared with BCL6 ( peak centers colocalizing within 200 bp ) and non-overlapping ( unique ) cofactor sites ( Figure 3—figure supplement 2A ) . When compared against DNA sequences from unique peaks , shared peaks were overrepresented with motifs for BCL6 , STAT , and FOX transcription factors , as well as CUX2 and HNF6 . In contrast , when tested against the DNA sequences of BCL6-shared peaks , unique SMRT , NCoR , and HDAC3 sites were enriched in motifs for ETS and ELK transcription factors . Next , we quantified SMRT , NCoR , and HDAC3 occupancy at BCL6-binding sites that colocalized with corepressor peaks in control versus Bcl6LKO livers ( Figure 3D and Figure 3—figure supplement 2B ) . For each corepressor , binding at BCL6 sites was significantly reduced in Bcl6LKO livers . Moreover , loss of these complexes was inversely correlated to histone 3 lysine 27 acetylation ( H3K27ac ) , a marker for enhancer activity ( Creyghton et al . , 2010; Wang et al . , 2008 ) , which was significantly elevated along BCL6-SMRT/NCoR-HDAC3 sites in Bcl6LKO livers . Together , these findings revealed a role for BCL6 to recruit a subset of liver SMRT/NCoR-HDAC3 complexes and repress associated regulatory regions . Gene ontology analysis of differentially expressed transcripts in the livers of Bcl6LKO animals revealed lipid metabolism , oxidation , and PPAR signaling as top scoring terms ( Figure 4A ) . This regulatory signature and the extensive genomic intersection between BCL6 and PPARα prompted us to determine whether BCL6 could likewise control fasting-induced gene expression . Livers from mice restricted from food overnight exhibited 162 genes upregulated and 174 genes downregulated by at least 2-fold using RNA-seq ( Figure 4B ) , and fasting regulated a common set of gene expression pathways with Bcl6 ablation ( Figure 4A ) . Notably , over 40% of robustly regulated fasting genes ( 135/336 ) were controlled by BCL6 ( Figure 4C , top panel ) and for the vast majority , Bcl6 ablation mimicked the impact of fasting on transcription ( Figure 4C , bottom panel and Figure 4D ) . Unsupervised clustering analyses of liver gene expression revealed that patterns in Bcl6LKO mice , irrespective of nutrition status , more closely resembled profiles from fasting than fed control mice ( Figure 4D and Figure 4—figure supplement 1A ) . Genes co-regulated by fasting and Bcl6 deletion are enriched in ontologies for lipid and ketone body metabolism as well as PPARα signaling ( Figure 4—figure supplement 1B ) . For example , visualization of ChIP-seq and RNA-seq tracks demonstrated that PPARα and BCL6 reciprocally occupy regions along the Acot4/3 and Vnn1 genes , whose expression was strongly induced by either fasting or Bcl6 ablation ( Figure 4E ) . Quantitative PCR further confirmed dozens of liver genes that were similarly upregulated by fasting or Bcl6 ablation , including many involved in mitochondrial and peroxisomal β-oxidation ( Abcd1/2 , Acadvl , Acnat2 , Acot2 , Acot3/4 , Ehhadh , Hadh , Idh2 , Ucp2 ) , microsomal ω-hydroxylation ( Aldh3a2 , Cyp4a31 ) , ketogenesis ( Acss3 , Bdh1 , Fgf21 , Hmgcl ) , and lipid metabolism ( Abhd2 , Acot1 , Cd36 ) ( Figure 4—figure supplement 1C ) . Together , these results suggested that loss of Bcl6 mimics the fasting-induced transcriptional program controlling liver lipid metabolism . We next examined the extent to which BCL6 and PPARα cis-regulatory sites alone or in combination control fasting transcription . Hypergeometric testing revealed a 1 . 1-fold enrichment ( p-value 3 . 8e-14 ) for BCL6-PPARα peaks relative to the entirety of PPARα genome-wide peaks along all genes differentially regulated ( p-value<0 . 05 ) by fasting . Over 50% of these fasting genes contained co-occurring BCL6-PPARα-binding sites ( Figure 4—figure supplement 2A ) , with a median of two co-occurring sites per gene ( Figure 4—figure supplement 2B ) . By contrast , just 24% or 1 . 4% of fasting genes contained PPARα-only or BCL6-only sites , respectively , and these occurred with a median of just one regulatory region per gene . In addition , fasting-regulated genes with BCL6-PPARα regulatory elements exhibited significantly greater ranges of expression than those with PPARα peaks that lack this heterotypic module ( Figure 4—figure supplement 2C ) , and their ontology was particularly enriched for functions in lipid regulation and oxidative metabolism ( Figure 4—figure supplement 2D ) . In summary , over half of fasting-regulated genes are controlled by BCL6-PPARα-binding sites , and this gene subset is particularly dynamic in transcription . PPARα is critical for the fasting induction of genes mediating peroxisomal and mitochondrial fatty acid β-oxidation as well as microsomal ω-hydroxylation ( Contreras et al . , 2013; Gao et al . , 2015; Hardwick et al . , 2009; Hashimoto et al . , 2000; Kersten et al . , 1999; Leone et al . , 1999; Montagner et al . , 2016 ) . To determine whether loss of the BCL6 repressor in liver compensates for transcriptional defects in Ppara-/- mice , we generated animals with combined whole body deletion of Ppara and liver-specific ablation of Bcl6 ( Ppara-/-;Bcl6LKO mice ) . RNA-seq revealed that loss of Bcl6 rescued 209 of 795 dysregulated genes in fasted Ppara-/- mice compared to fasted controls ( Figure 5A ) . Among genes normally upregulated with fasting , Bcl6 deletion restored expression of genes involved in monocarboxylic acid and lipoprotein metabolism; ketone body synthesis; AMPK and PPAR signaling; and peroxisomes ( Figure 5B , top panel ) . By contrast , genes normally downregulated upon fasting and rescued in Ppara-/-;Bcl6LKO mice represented pathways mostly unrelated to lipid metabolism ( Figure 5B , bottom panel ) . Restoration of Ppara-/- defective fasting transcription in Ppara-/-;Bcl6LKO mice was confirmed by qPCR at genes involved in β-oxidation ( Acot2/3/4 , Idh2 ) , ω-hydroxylation ( Aldh3a2 , Cyp4a31 ) , ketone body synthesis ( Acss3 , Fgf21 , Hmgcl , Hmgcs2 ) , and lipid metabolism ( Abhd2 , Cd36 , Vldlr ) ( Figure 5—figure supplement 1 ) . Thus , loss of Bcl6 restores expression at a subset of PPARα-directed genes controlling lipid metabolism . Opposing regulation between PPARα and BCL6 was also observed at the level of chromatin modification . We profiled histone H3K27ac in overnight fasted Bcl6fl/fl control , Ppara-/- , and Ppara-/-;Bcl6LKO mice using ChIP-seq ( Figure 5C , left panel ) . Fasting-induced genes with impaired expression in Ppara-/- mice demonstrated low H3K27ac signal around BCL6-PPARα-binding sites in Ppara-/- compared to control mice . By contrast , in livers of Ppara-/-;Bcl6LKO animals , H3K27ac is reestablished or even enhanced at these sites ( Figure 5C , top left panel; and Figure 5—figure supplement 2A ) . Reciprocal H3K27ac patterns were found at impaired fasting-repressed genes in Ppara-/- and Ppara-/-;Bcl6LKO animals ( Figure 5C , bottom left panel ) . This pattern in H3K27ac at BCL6-PPARα sites occurred only at fasting impaired genes that were rescued in Ppara-/-;Bcl6LKO mice ( Figure 5—figure supplement 2B ) . Thus , BCL6 de-repression restores aberrant liver cis-regulatory activity in Ppara-/- mice along fasting responsive genes . Next , we sought to further understand how ablation of hepatocyte Bcl6 could rescue fasting expression defects in Ppara-/- mice . The reestablishment of acetylation at fasting enhancers with BCL6-PPARα sites pointed to a shift in the balance of transcription factor complexes with histone deacetylase ( HDAC ) and acetyltransferase ( HAT ) activities at these co-regulated regions . Since hepatocyte BCL6 binds to SMRT/NCoR-HDAC3 at a subset of its binding sites ( Figure 3C and D ) , we specifically examined HDAC3 occupancy at BCL6-PPARα peaks along rescued fasting genes . In the absence of Bcl6 , HDAC3 was substantially diminished at these BCL6-PPARα sites ( Figure 5D ) . Additionally , we tested whether BCL6 could influence other PPAR isotypes , which can be associated with CBP/p300 HAT complexes that acetylate H3K27 ( Jin et al . , 2011 ) . Using qPCR , we found that Ppard levels were significantly increased in fasted Ppara-/-;Bcl6LKO compared to Ppara-/- mice , while Pparg levels were unchanged ( Figure 5E ) . Moreover , BCL6 ChIP-sequencing revealed that BCL6 binds multiple intronic sites along the Ppard gene ( Figure 5—figure supplement 2C ) , suggesting that it directly represses Ppard expression . Consistent with their enhanced Ppard levels , we observed increased PPARδ binding near rescued genes in Ppara-/-;Bcl6LKO compared to Ppara-/- mice ( Figure 5C , right panel ) , particularly at upregulated fasting genes ( Figure 5F ) , including Acot2 , ( Figure 5G ) , Hmgcs2 , Aldh3a2 ( Figure 5—figure supplement 2D ) , and others ( Figure 5—figure supplement 2E ) . Together , these observations identified that loss of BCL6 directly relieves repression and potentiates PPARδ-mediated transactivation to restore fasting liver gene expression in Ppara-/-;Bcl6LKO mice . Next , we determined the functional impact of the BCL6 regulatory program on hepatic regulation and lipid processing in vivo . Ad libitum fed Bcl6LKO mice exhibited higher circulating ketone bodies compared to Bcl6fl/fl mice ( Figure 6A ) . This difference persisted after a 24 hr fast . Additionally , mice lacking hepatic Bcl6 have higher rates of complete fatty acid oxidation as measured by oxidation of 14C-palmitate in liver homogenates ( Figure 6B ) . In contrast , analysis of fatty acid uptake , triglyceride secretion , and hepatic lipogenesis based on in vivo deuterium incorporation revealed no other differences in lipid metabolism between Bcl6LKO mice and controls ( Figure 6C–E ) . To test a broader role for BCL6 in lipid processing , we assessed hepatic triglyceride content after feeding mice high-fat diet ( HFD ) for 19 weeks . Bcl6LKO mice were profoundly protected from developing steatosis , as demonstrated by oil red O staining and more than a 50% reduction in hepatic triglyceride content compared to Bcl6fl/fl controls , despite similar increases in body weight ( Figure 6F–H ) . Accompanying these reductions in hepatic lipid accumulation , HFD-exposed Bcl6LKO mice exhibited significantly lower levels of fasting serum glucose ( Figure 6—figure supplement 1A ) and a non-significant reduction in insulin ( Figure 6—figure supplement 1B ) . Moreover , when challenged with a shorter term 5-week HFD , Bcl6LKO mice exhibited a trend towards enhanced insulin responsiveness , as measured by levels of phosphorylated AKT following acute administration of exogenous insulin ( Figure 6—figure supplement 1C ) . These combined observations demonstrate that mice lacking hepatic Bcl6 have heightened capacity to catabolize lipids via β-oxidation and subsequent ketogenesis or TCA cycling , as well as improved glucose homeostasis when challenged with high-fat diet . Ppara-/- mice exhibit fasting hypoketonemia and impaired fatty acid oxidation leading to steatosis ( Gao et al . , 2015; Hashimoto et al . , 2000; Kersten et al . , 1999; Leone et al . , 1999; Montagner et al . , 2016 ) . After 48 hr of fasting , Ppara-/- mice developed centrilobular macrosteatosis ( Figure 6I ) , as previously reported ( Hashimoto et al . , 2000 ) . Remarkably , Ppara-/-;Bcl6LKO animals were strongly protected from hepatic triglyceride accumulation based upon histological analysis with oil red O staining and demonstrated 23% reduced triglyceride accumulation compared to Ppara-/- mice ( Figure 6I and J ) . Compared to fasted Ppara-/- mice , Ppara-/-;Bcl6LKO mice also had higher rates of 14C-palmitate oxidation in liver homogenates , exhibited in both completely oxidized 14CO2 and incompletely oxidized 14C-acid soluble intermediates ( Figure 6K ) . In line with their reduced lipid accrual , Ppara-/-;Bcl6LKO mice also revealed higher ketone body levels ( Figure 6L ) , suggesting that ablation of Bcl6 can de-repress ketone body synthesis even in the absence of Ppara . Overall , these results established that loss of liver Bcl6 rescues metabolic defects of Ppara deficiency .
Dynamic transcriptional programming is necessary to sustain life in environments of varying access to food , and the liver is central to orchestrate systemic metabolism in response to such changes . However , our understanding of the epigenomic programs that underpin feeding and fasting metabolism is limited and dominated by studies of hormonally-cued transcriptional activators . Our work has identified BCL6 as a potent repressor of lipid catabolism , both in the context of fasting and dietary lipid overload . On a genome-wide scale , BCL6 converges with PPARα at over 13 , 000 regulatory regions on which BCL6 binding is enriched with feeding , whereas PPARα is induced by fasting . This dynamic BCL6-PPARα cis-regulatory module annotates to over 1 , 400 fasting-responsive genes . Moreover , Bcl6 ablation mimics the fasting transcriptional response , and a myriad of defects in Ppara-/- mice are partially rescued by concomitant deletion of hepatocyte Bcl6 , ranging from defective fasting enhancer activity and gene expression to impaired fatty acid oxidation , hypoketonemia , and susceptibility to steatosis . Together , these findings evidence a powerful role for BCL6 to epigenomically oppose PPARα and to suppress fatty acid oxidation . Previously , BCL6 functions outside of hematopoietic cells were poorly defined . In liver , prior analysis supported a role for BCL6 in competing with STAT5 and modulating responses to growth hormone and drug metabolism ( Chikada et al . , 2018; Zhang et al . , 2012 ) . Further , a study of whole-body knockout mice posited a role for BCL6 in systemic metabolism , but it was confounded by analysis limited to animals with severe and frequently fatal inflammatory disease ( LaPensee et al . , 2014 ) . Original characterization of Bcl6-/- mice demonstrated variable degrees of growth retardation and ill health within three weeks of life , with half dying before 5 weeks of age ( Dent et al . , 1997 ) . Over 80% of Bcl6-/- mice exhibit myocarditis and over 70% have pulmonary vasculitis with elevated levels of IL-4 , –5 , and −13 , cytokines known to directly impact liver metabolism ( Ricardo-Gonzalez et al . , 2010; Stanya et al . , 2013 ) . Thus , metabolic phenotyping of whole body Bcl6 knockout mice was uninformative ( LaPensee et al . , 2014 ) , and the role for BCL6 in cell-intrinsic hepatic lipid metabolism was previously unknown . Using genetic , genomic , and isotopic analyses we reveal that loss of Bcl6 in hepatocytes causes cell-autonomous enhancement of fatty acid oxidation without a direct impact on lipid synthesis . Physical interactions between BCL6 and various cofactors have been well documented and in immune cells mediate distinct functional roles ( Huang et al . , 2014; Huang et al . , 2013 ) . Among these interaction partners are SMRT and NCoR , which bind to the BCL6 N-terminal BTB domain and function as scaffolds to recruit HDAC3 and other corepressive machinery ( Perissi et al . , 2010 ) . In liver , we found nearly 80% overlap between BCL6 and the cistromes of SMRT , NCoR , and HDAC3 . Moreover , loss of BCL6 was associated with significantly diminished occupancy of these cofactors at BCL6-bound regulatory regions ( Figure 3 ) . However , thousands of SMRT , HDAC3 , and particularly NCoR binding peaks were independent of BCL6 . Furthermore , even at regulatory regions where BCL6 and these coregulators colocalize , persistent ChIP-seq signals for SMRT , NCoR , and HDAC3 are often observed in the genetic absence of Bcl6 . These findings indicate that SMRT , NCoR , and HDAC3 may frequently engage multiple transcription factor complexes within a single regulatory region . Given their extensive interactions , there is tremendous complexity in deciphering corepressor roles in metabolic regulation . Indeed , knockouts and various knockin mutants of NCoR , SMRT , and HDAC3 have demonstrated hepatic steatosis phenotypes ( Knutson et al . , 2008; Mottis et al . , 2013; Shimizu et al . , 2015; Sun et al . , 2012 ) , in contrast to the lipid overload-protected phenotype observed here with hepatocyte Bcl6 ablation . The cofactor requirements for BCL6-mediated control in the liver and extent to which SMRT/NCoR-HDAC3 are responsible for its potent repression of lipid catabolism warrant further investigation . The clustering of transcription factors at regulatory regions has been proposed as a flexible mechanism to control diverse gene expression patterns during development and in response to environmental stimulus ( Arnone and Davidson , 1997; Smith et al . , 2013 ) . Motif enrichment indicated a relationship between the BCL6 repressor and the PPAR subfamily of lipid-activated nuclear receptors . PPARα is the predominant PPAR isotype in liver , while PPARδ is expressed at lower levels but was reported to physically interact with BCL6 ( Lee et al . , 2003 ) . However , we find that Ppara and Ppard are each genetically dispensable for chromatin recruitment of BCL6 . Thus , BCL6 opposition to PPARα occurs via proximate binding at independent cis-regulatory elements , a mechanism distinct from FXR , which has been reported to counter PPARα transcriptional outputs through competition for DR1-binding sites ( Lee et al . , 2014 ) . The regulatory interaction between BCL6 and PPARα is also unique from other integrative regulators of hepatic lipid metabolism such as HNF6 and REV-ERBα , which cooperatively repress transcription via tethering ( Zhang et al . , 2016 ) . In addition to PPARs , it is possible that other transcriptional activators predicted to converge with BCL6 including HNF6 and HNF4 , FOXA1 , and C/EBP ( Figure 1B ) , collaborate with BCL6 and PPARα in hepatic lipid regulation ( Hayhurst et al . , 2001; Zhang et al . , 2016 ) . The BCL6-PPARα regulatory module is remarkable for its widespread occurrence along genes controlling lipid catabolism . We speculate that BCL6-PPAR elements in fasting enhancers endow them with variably repressive or activating regulatory potential , contributing to highly dynamic gene expression across the feeding to fasting transition . In a related manner , genetic ablation of Bcl6 de-represses these regulatory regions and compensates for loss of PPARα transactivity in Ppara-/-;Bcl6LKO mice . Remarkably , this occurs both directly , via loss of active repression at BCL6-PPAR elements , and indirectly by upregulating Ppard to enhance transactivity at BCL6-PPAR sites . Thus , we find that liver metabolic shifts are not simply directed by inducible transactivating factors . Rather , ‘active repression’ ( Hanna-Rose and Hansen , 1996 ) by BCL6 and its dynamic modulation play key additional roles in toggling between the fed and fasted state and determining hepatic lipid accumulation . Since inhibitors of BCL6 have been developed to target BCL6 and selective interactions with its corepressors ( Cardenas et al . , 2016; Lu et al . , 2018 ) , these findings also raise the possibility that BCL6 de-repression could represent a future therapeutic strategy for non-alcoholic fatty liver disease .
Bcl6fl/fl mice were generated through the UC Davis Mouse Biology Program by engineering loxP sites between exons 5 and 6 of the mouse Bcl6 locus . Cre-mediated deletion creates a frameshift mutation , resulting in a protein of 138 amino acids ( compared to 708 amino acids in wild-type BCL6 ) lacking exons 5–10 and the zinc finger DNA binding domain . Ppara-/- ( Stock #008154 ) and Ppardfl/fl ( Stock #005897 ) mice were obtained from Jackson Laboratories . Bcl6fl/fl and Ppardfl/fl mice were crossed with Albumin-Cre animals ( Jackson Laboratories , Stock #003574 ) to generate Bcl6fl/fl; Albumin-Cre ( Bcl6LKO ) and Ppardfl/fl; Albumin-Cre ( PpardLKO ) mice , respectively . Mice were maintained on a 14:10 light: dark ( LD ) cycle with free access to water . Unless otherwise specified , ‘fed’ refers to ad libitum feeding with standard chow and ‘fasted’ refers to a 16–18 hr overnight fast . High-fat diet containing 45% of kcal from fat was obtained from Research Diets , Inc ( Stock #D12451 ) . All animal care and use procedures were conducted in accordance with regulations of the Institutional Animal Care and Use Committee at Northwestern University . Chromatin immunoprecipitation ( ChIP ) was performed as previously described ( Barish et al . , 2010 ) . ChIP samples were prepared in biological triplicate ( three animals per condition ) , unless otherwise specified . Mouse livers were harvested , rinsed in PBS , and crosslinked at room temperature for 30 min in 2 mM disuccinimidyl glutarate and then for 10 min in 1% formaldehyde . After quenching with 125 mM glycine , crosslinked material was rinsed twice with cold PBS and frozen at −80°C until further processing . Crosslinked material was lysed in buffer containing 0 . 75M NaCl , 1% Triton X , 0 . 5 mM Tris , 0 . 05 mM EDTA , and 0 . 5% NP-40 . Isolated nuclei were then sheared in buffer containing 1% SDS , 10 mM EDTA , and 50 mM Tris for six cycles ( 30 s on , 30 s off ) using a Diagenode Bioruptor to shear chromatin into 200–1000 bp fragments . Protein-DNA complexes were incubated overnight with antibody against BCL6 ( custom polyclonal to mouse BCL6 ) , PPARα ( Santa Cruz ) , PPARδ ( custom polyclonal to mouse PPARδ ) ( Fan et al . , 2017 ) , SMRT ( custom polyclonal to mouse SMRT ) ( Barish et al . , 2012 ) , NCoR ( custom polyclonal to mouse NCoR ) ( Barish et al . , 2012 ) , HDAC3 ( Santa Cruz ) or H3K27ac ( Active Motif ) . Antibody complexes were precipitated with IgG paramagnetic beads ( ThermoFisher ) for ChIP-seq or Protein A agarose beads ( Millipore ) for ChIP followed by qPCR . DNA was decrosslinked and purified using MinElute PCR purification columns ( Qiagen ) . ChIP DNA was either assessed via qPCR and expressed as percent recovery of input chromatin or further processed into libraries for ChIP-seq . See Supplementary file 1 for ChIP qPCR primers . Sequencing libraries were generated from ChIP DNA using KAPA DNA Library Preparation kits ( Kapa Biosystems ) according to manufacturer’s instructions . Libraries were assessed by Bioanalyzer ( Agilent ) and qPCR-based quantification ( Kapa Biosystems ) and sequenced on an Illumina NextSeq 500 instrument using 75 bp single-end reads . Raw sequence reads were aligned to a reference genome ( mm10 ) using Bowtie version 1 . 1 . 1 ( Langmead et al . , 2009 ) using ‘-m 1’ and ‘--best’ parameters to ensure reporting of uniquely mapped reads . Tag directories were generated using ‘makeTagDirectory’ using the -tbp 1 option to limit the number of reads starting at the same position to 1 . ChIP-seq peaks were identified and analyzed using HOMER ( Heinz et al . , 2010 ) . ChIP-seq peaks were identified in HOMER using the ‘getDifferentialPeaksReplicates . pl’ command , specifying ‘-style factor’ to generate a high confidence set of peaks across triplicate samples . This command generates a peak list in three steps: first , it pools target tag directories to perform an initial peak identification against input; second , it quantifies raw reads of each target and input tag directory at the initial putative peaks; third , it calls DESeq2 to calculate enrichment values for each peak using the individual raw counts and returns only those peaks that pass two fold enrichment and FDR < 0 . 05 . Peaks were annotated to nearest genes using ‘annotatePeaks . pl . ’ To characterize enriched motifs near BCL6-binding sites , we used HOMER’s ‘findMotifsGenome . pl’ command to scan 50 bp windows surrounding BCL6 peaks , including the -mask option compared to random whole genome sequences . Motif densities were then quantified using HOMER’s ‘annotatePeaks . pl’ using known motifs for PPARE , RXR , LXRE , and FXR; the BCL6 motif displayed in the density plots was identified with HOMER’s de novo motif discovery tool using 200 bp scanning windows surrounding BCL6 peaks . Motif finding near SMRT , NCoR , and HDAC3 peaks used a 200 bp scanning window; the top 20 motifs by p-value were included in the heatmap . Enriched motifs were identified in all peaks for each factor compared to random whole genome sequences . Peak sets were then each compared to BCL6 to identify enriched motifs at shared sites against DNA sequences from unique sites . Conversely , for each cofactor , enriched motifs were identified at unique sites compared against DNA sequences from peaks shared with BCL6 . To generate the tag density scatter plots and histograms , tags were quantified using HOMER’s ‘annotatePeaks . pl’ command , with either ‘-size 400’ option or ‘-size 2000 -hist 25’ options , for scatter plots and histograms , respectively . HOMER’s ‘mergePeaks’ was used to compare different peak sets , defining overlapping peaks as those with a maximum distance between peak centers of 200 bp . BigWig browser tracks were generated using HOMER’s ‘makeMultiWigHub . pl’ program and then viewed on the UCSC Genome Browser ( Kent WJ et al . , 2002 ) . Gene ontologies for ChIP-seq data were generated using GREAT ( McLean et al . , 2010 ) by annotating ChIP-seq peaks to the single nearest gene . To calculate distance between BCL6 and PPARα peaks , the distance between each BCL6 peak and the nearest PPARα peak was calculated using HOMER’s ‘annotatePeaks . pl’ command and the ‘-pdist’ option . Distances < 200 bp were plotted in a histogram where each bin represents increasing increments of 10 bp . The four-way Venn was generated using Intervene ( Khan and Mathelier , 2017 ) . Liver samples ( <30 mg ) were stored in 1 mL of RNAlater Stablization Solution ( Ambion ) at −80° immediately following harvest . To isolate total RNA , tissues were homogenized in 1 mL buffer RLT ( Qiagen ) using the Mo Bio Powerlyzer . RNA was isolated and purified using RNeasy columns according to the manufacturer’s protocol ( Qiagen ) . RNA quality was assessed using a Bioanalyzer ( Agilent ) to ensure a RIN score greater than 7 . 0 . Sequencing libraries were constructed from purified RNA using the KAPA Stranded RNA-seq Kit with RiboErase ( HMR ) according to the manufacturer’s instructions . Libraries were quantified using both a Bioanalyzer ( Agilent ) and qPCR-based quantification ( Kapa Biosystems ) and sequenced on an Illumina NextSeq 500 instrument using 75 bp single-end reads . RNA raw sequence reads were aligned to a reference genome ( mm10 ) using STAR version 2 . 4 . 0 hr ( Dobin et al . , 2013 ) . Aligned reads included only unique mappers and those with fewer than four mismatches . Gene expression at exons was quantified using HOMER ( Heinz et al . , 2010 ) . Differentially expressed RNAs were then normalized and identified using DESeq2 version 1 . 14 . 1 ( Love et al . , 2014 ) with an adjusted FDR < 0 . 05 . Direct comparisons were made between Bcl6fl/fl fed and fasted animals , as well as between fed Bcl6fl/fl and Bcl6LKO animals to generate lists of differentially expressed genes . To compare the BCL6 cistrome and transcriptome , BCL6 ChIP-seq peaks were annotated to the nearest transcription start site using ‘annotatePeaks . pl’ in HOMER and then grouped based on the liver gene expression change of the annotated gene in Bcl6LKO mice compared to Bcl6fl/fl controls . Peaks were grouped as ‘repressed , ’ ‘activated , ’ or ‘unchanged’ BCL6 peaks if gene expression was higher , lower , or unaffected in Bcl6LKO , respectively . To determine ‘rescued’ gene expression in Ppara-/-;Bcl6LKO animals , impaired gene expression in Ppara-/- animals was first defined . To do this , we first identified genes that were significantly changed ( adjusted p-value<0 . 05 ) with fasting in Bcl6fl/fl control animals . Then , among genes normally upregulated with fasting , we identified genes that were significantly less expressed ( adjusted p-value<0 . 05 ) in fasted Ppara-/- compared to fasted control mice . Similarly , we identified genes normally downregulated with fasting that were significantly more expressed in fasted Ppara-/- compared to control fasted mice . Collectively , these significantly different up- and downregulated genes represent dysregulated genes in fasted Ppara-/- mice . To then determine rescued gene expression in Ppara-/-;Bcl6LKO animals , we identified dysregulated Ppara-/- genes that demonstrated no significant difference in gene expression between control fasted mice and Ppara-/-;Bcl6LKO fasted mice , indicating a restoration of fasting gene expression to control levels . We also identified partially rescued gene expression . Upregulated fasting genes with significantly lower expression in fasting Ppara-/- animals were considered partially rescued if fasting Ppara-/-;Bcl6LKO gene expression was significantly higher than Ppara-/- but also still significantly lower than control gene expression . Similarly , downregulated fasting genes with significantly higher expression in fasting Ppara-/- animals were considered partially rescued if fasting Ppara-/-;Bcl6LKO gene expression was significantly lower than Ppara-/- but also still significantly higher than control gene expression . Gene ontology analysis was performed on differential genes with an adjusted p-value<0 . 05 and a log fold change greater than one using Metascape ( Tripathi et al . , 2015 ) . Enrichment analysis included terms from Reactome Gene Sets , GO Biological Processes and KEGG Pathways . RPKM values were generated using HOMER ( Heinz et al . , 2010 ) and displayed as heatmaps using Morpheus ( Gould , 2019 ) . The distance matrix heatmap was generated using the ‘dist’ function in R version 3 . 4 . 3 ( R Development Core Team , 2017 ) and plotted with the heatmap . 2 function and the ‘RColorBrewer’ package ( Neuwirth , 2014 ) . To generate the heatmap of H3K27ac and PPARδ enrichment at BCL6-PPARα peaks near rescued genes , relevant BCL6-PPARα-bound regions were identified by annotating peaks to the nearest transcription start site using HOMER . H3K27ac ChIP-seq enrichments in fasted control , Ppara-/- and Ppara-/-;Bcl6LKO samples were then quantified at these peaks using ‘annotatePeaks . pl’ and the ‘-size 6000 -hist 25 -ghist’ options; PPARδ ChIP-seq enrichment was quantified using ‘-size 3000 . ’ The signal averages across biological replicates were plotted as a heatmap using Morpheus . All UCSC genome browser tracks represent combined tag directories across replicates . Genes differentially expressed with fasting were classified based on their association with regulatory regions . First , shared BCL6-PPARα ChIP-seq peaks were annotated to the single nearest gene using HOMER . Using these annotations , genes differentially expressed with fasting ( adjusted p-value<0 . 05 ) were then grouped based on presence or absence of a nearby shared BCL6-PPARα annotated peak . Of those differential fasting genes without an annotated shared BCL6-PPARα peak , genes were further grouped based on presence of annotated BCL6 unique and PPARα unique ChIP-seq peaks . Some genes had both BCL6 unique and PPARα unique peaks , but these were non-overlapping . Other genes had neither BCL6- nor PPARα-annotated peaks nearby . The frequency of BCL6-PPARα , PPARα only , or BCL6 only peaks per fasting gene was also calculated . Frozen liver tissues were homogenized in Trizol ( Ambion ) using a Mo Bio Powerlyzer . Chloroform was added at 200 μL to 1 mL homogenates in Trizol . The clear aqueous phase was extracted after centrifugation . RNA was then isolated with a RNeasy kit ( Qiagen ) , according to manufacturer’s protocol . cDNA was synthesized with 600–1000 ng of RNA using the iScript cDNA Synthesis Kit ( BioRad ) . Gene expression was then assessed via qPCR using iTaq Universal SYBR Green Supermix ( BioRad ) . Gene expression was normalized to the housekeeping gene , 36b4 . See Supplementary file 2 for primer sequences . For the hematoxylin and eosin ( H & E ) staining , liver tissues were fixed in 10% formalin overnight and then moved to 70% EtOH . Fixed tissues were paraffin embedded , cut , and stained by the Northwestern University Research Histology and Phenotyping Laboratory which is supported by NCI P30-CA060553 awarded to the Robert H . Lurie Comprehensive Cancer Center . For oil red O staining , liver samples frozen in OCT were cut to 5–7 μm with a Leica cryostat , mounted onto slides , stained with oil red O , and counterstained with hematoxylin . We measured serum triglycerides ( Infinity Thermo Fisher ) and ketone bodies ( Cayman Chemical ) using commercial kits . To measure tissue triglycerides , we extracted lipids using a modified version of the Folch Method ( Folch et al . , 1957 ) . In brief , tissues were homogenized in 1 mL of methanol using the Mo Bio Powerlyzer . Homogenates were transferred to glass tubes and incubated several hours in 1:2 methanol:chloroform after briefly vortexing . 0 . 9% NaCl was added to homogenates overnight to separate the chloroform lipid-containing layer from the methanol layer . The next day , the methanol and any floating tissue was aspirated . The remaining chloroform layer was dried under nitrogen gas . Lipid was resuspended in 2-propanol and quantified using the Infinity Thermo Fisher triglyceride kit . Quantified lipid was normalized to tissue weight . Serum insulin was measured via ELISA ( Crystal Chem ) and serum glucose was measured using a colorimetric assay ( BioVision ) . We fasted mice for 4 hr and then injected mice intraperitoneally with a 7 . 5% poloxamer solution in PBS at a dose of 1 mg/g body weight . Tail-vein blood samples were collected over time using capillary Microvettes ( Sarstedt ) . The rate of hepatic lipogenesis was determined via incorporation of 2H into newly made TG-bound fatty acids , as described elsewhere ( Bederman et al . , 2006 ) . Briefly , mice were injected i . p . with 0 . 7 mL of 2H-labeled saline ( 9 g of NaCl in 1 L of 99 . 9% 2H2O ) . For the next 24 hr , mice were maintained on 6% 2H-labeled drinking water and then harvested . Terminal serum and liver tissue samples were collected and flash frozen . Sample processing and GC/MS analysis was performed as described previously ( Bederman et al . , 2012 ) . We determined rates of fatty acid oxidation in liver homogenates by measuring oxidation of 14C palmitate ( Hirschey and Verdin , 2010 ) . Briefly , tissue was dounce homogenized in sucrose/Tris/EDTA buffer and incubated for 30 or 60 min in a reaction mixture containing 0 . 4 uCi 14C palmitate . After reacting with the labeled palmitate , mixtures were transferred to tubes containing 1M perchloric acid with Whatman paper discs soaked in 1M NaOH in the lids . Scintillation counting was used to measure 14C in the acid-soluble fraction and in disc-trapped CO2 , representing partially and fully oxidized radiolabeled palmitate , respectively . Fatty acid oxidation rates were then expressed as amount of substrate oxidized per tissue weight per minute . Mice were injected with BODIPY-C16 ( Life Technologies ) to assess lipid uptake , as described elsewhere ( Wilson et al . , 2016 ) . BODIPY-C16 was resuspended in dimethylsulfoxide at 10 mM . Then , a 0 . 1 μg/μL working stock was made in 0 . 25% fatty-acid-free BSA ( Sigma-Aldrich ) solution in PBS . Mice were fasted for 4 hr and then injected intraperitoneally with BODIPY-C16 at 0 . 5 μg/g of body weight . After 5 hr , tissues were collected and flash frozen . 80–120 mg of liver tissues were dounce homogenized in RIPA buffer . 25 μL volumes of cleared tissue homogenates were diluted 1:4 in PBS and analyzed using a fluorescent plate reader ( Ex 485 nm , Em 515 nm ) . Saline-injected mouse liver homogenates were used to control for background fluorescence . Tissue fluorescence was normalized to tissue weight . Mice were placed on 5 weeks of high-fat diet . After a 5 hr fast , mice were injected intraperitoneally with 1 U/kg recombinant insulin . Ten minutes later , mice were sacrificed and their tissues were harvested . Frozen liver tissues were dounce homogenized in RIPA buffer . After incubating on ice for 10 min , homogenates were centrifuged at full speed for 15 min at 4°; supernatant was then collected and stored at −80° . Protein was quantified with a BCA assay ( Thermo Scientific ) and 2 μg/μL lysates were boiled for 5 min in 5x loading buffer . Denatured protein lysates were loaded in precast polyacrylamide gels ( BioRad ) and transferred to PVDF membranes ( BioRad ) . Membranes were blocked with 5% milk in PBST and probed with primary antibodies for BCL6 ( Santa Cruz , D-8 ) at 1:200 , PPARα ( Santa Cruz , H-98 ) 1:500 , pAKT ( Cell Signaling , 4060 ) 1:1000 , panAKT ( Cell Signaling , 4691 ) 1:1000 or β-actin ( Sigma , A1978 ) 1:1000 overnight at 4° . Secondary antibodies were added for 1 hr at room temperature ( Jackson ImmunoResearch ) . Protein was then visualized using ECL ( ThermoScientific ) . MemCode Reversible Stain was used to visualize total protein ( Thermo Fisher Scientific ) . Protein densitometry was quantified using ImageJ 1 . 51 s ( Schneider et al . , 2012 ) . All RNA-seq and ChIP-seq data are deposited in GEO SuperSeries accession #GSE118789 . | Obesity has nearly tripled worldwide since the 1970s . A major health concern related to obesity is that excess fat can spill into organs such as the liver . This can lead to fatty liver disease or even liver cancer . Therefore , it is important to fully understand the mechanisms that lead to fat accumulation in the liver in order to develop new treatments . Our bodies are designed to even out the highs and lows of an unpredictable diet by storing and releasing calories . When we are well-fed , liver cells switch on genes involved in making fat . When we have not eaten for a while , they switch them off and turn on genes involved in burning fat . Each switch involves thousands of genes , controlled by proteins called transcription factors . Some work as activators , turning genes on , whilst others work as repressors , turning genes off . For example , the transcription factor PPAR alpha is a well-known activator that helps to regulate fat burning . However , we know much less about the repressors that stop cells burning fat when there is plenty of food available . To find out more , Sommars et al . studied the repressor BCL6 in mouse liver cells . The results revealed that BCL6 interacts with hundreds of the same genes as PPAR alpha . When the mice were eating , BCL6 turns off the genes involved in fat burning , but when they were starved PPAR alpha activated those genes . However , when BCL6 was experimentally removed , many fat-burning genes were permanently switched on . So , even when mice were fed a high-fat diet , they burned off fat in their livers . Understanding the role of genetic switches like PPAR alpha and BCL6 is crucial for understanding how and why our bodies store energy . This could help us to create treatments that enhance the liver's ability to burn excess fat . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression"
] | 2019 | Dynamic repression by BCL6 controls the genome-wide liver response to fasting and steatosis |
Polo-like kinases ( PLK ) are eukaryotic regulators of cell cycle progression , mitosis and cytokinesis; PLK4 is a master regulator of centriole duplication . Here , we demonstrate that the SCL/TAL1 interrupting locus ( STIL ) protein interacts via its coiled-coil region ( STIL-CC ) with PLK4 in vivo . STIL-CC is the first identified interaction partner of Polo-box 3 ( PB3 ) of PLK4 and also uses a secondary interaction site in the PLK4 L1 region . Structure determination of free PLK4-PB3 and its STIL-CC complex via NMR and crystallography reveals a novel mode of Polo-box–peptide interaction mimicking coiled-coil formation . In vivo analysis of structure-guided STIL mutants reveals distinct binding modes to PLK4-PB3 and L1 , as well as interplay of STIL oligomerization with PLK4 binding . We suggest that the STIL-CC/PLK4 interaction mediates PLK4 activation as well as stabilization of centriolar PLK4 and plays a key role in centriole duplication .
Centrosomes are the major organizing centers for the microtubule network in animal cells and facilitate many microtubule-dependent cellular processes throughout the cell cycle ( reviewed in Bettencourt-Dias and Glover , 2007; Bornens , 2012; Gonczy , 2012 ) . During interphase , centrosomes contribute to cell shape , motility and polarity; in mitosis , they form the poles of the mitotic spindle and direct chromosome segregation . The core components of centrosomes , the centrioles , also function as basal bodies for the assembly of cilia and flagella . Mutations of centrosomal proteins have been associated with a variety of human diseases , notably ciliopathies , microcephalies and dwarfisms ( Nigg and Raff , 2009; Chavali et al . , 2014 ) . Furthermore , numerical and/or structural centrosome abnormalities have been implicated in carcinogenesis ( Nigg , 2002; Basto et al . , 2008; Ganem et al . , 2009; Nigg and Raff , 2009 ) . Centrosomes duplicate during S-phase by formation of a procentriole , the daughter centriole , orthogonally arranged to each pre-existing mother centriole ( Firat-Karalar and Stearns , 2014; Sluder , 2014 ) . The duplication process relies on a set of proteins conserved from Caenorhabditis elegans and Drosophila melanogaster to humans . The human core components are: the serine/threonine Polo-like kinase PLK4 ( ZYG-1 in C . elegans ) , the centrosomal protein Cep192 ( SPD-2 in C . elegans ) , the spindle assembly 6 homolog ( C . elegans ) protein SAS-6 , the SCL/TAL1 interrupting locus protein STIL ( SAS-5 in C . elegans ) and the centrosome protein CPAP ( SAS-4 in C . elegans ) ( Strnad and Gonczy , 2008; Azimzadeh and Marshall , 2010; Carvalho-Santos et al . , 2011; Nigg and Stearns , 2011; Brito et al . , 2012; Gonczy , 2012 ) . Overexpression of either , PLK4 , SAS-6 or STIL , causes formation of multiple daughter centrioles around a single mother centriole ( Habedanck et al . , 2005; Kleylein-Sohn et al . , 2007; Strnad et al . , 2007; Tang et al . , 2011; Arquint et al . , 2012; Vulprecht et al . , 2012 ) , while depletion of any one of these proteins blocks centriole formation ( Bettencourt-Dias et al . , 2005; Habedanck et al . , 2005; Leidel et al . , 2005; Tang et al . , 2011; Arquint et al . , 2012; Vulprecht et al . , 2012 ) . Thus , PLK4 , SAS-6 and STIL constitute key centriole duplication factors , the activity and levels of which need to be tightly controlled to maintain the correct centriole number ( Strnad et al . , 2007; Guderian et al . , 2010; Holland et al . , 2010; Arquint and Nigg , 2014 ) . In addition to these core components , other proteins are also essential for centriole duplication in human cells , this includes notably Cep152 ( Cizmecioglu et al . , 2010; Dzhindzhev et al . , 2010; Hatch et al . , 2010 ) , which cooperates with Cep192 in PLK4 recruitment ( Kim et al . , 2013; Sonnen et al . , 2013; Park et al . , 2014 ) . The early phase of centriole biogenesis is marked by the assembly of the cartwheel structure that serves as a scaffold for deposition of centriolar microtubules and confers the characteristic ninefold symmetry to the centriole ( Nakazawa et al . , 2007; Gonczy , 2012; Hirono , 2014; Winey and O'Toole , 2014 ) . SAS-6 has been shown to self-assemble into cartwheel-like structures in vitro , indicating that it is a central component of the cartwheel ( Kitagawa et al . , 2011; van Breugel et al . , 2011; Guichard et al . , 2013; van Breugel et al . , 2014 ) . In human cells , SAS-6 , STIL and PLK4 localize to the cartwheel region , suggesting a functional interaction of these proteins in cartwheel assembly ( Strnad et al . , 2007; Arquint et al . , 2012; Sonnen et al . , 2012; Fong et al . , 2014 ) . Such an interaction is supported by recent evidence demonstrating that PLK4 regulates complex formation between STIL and SAS-6 via phosphorylation of STIL ( Dzhindzhev et al . , 2014; Ohta et al . , 2014; Kratz et al . , 2015 ) . This process depends on two highly conserved regions of STIL: a short coiled-coil ( CC ) motif ( STIL-CC , residues 720–751 ) and the STAN ( STIL/Ana2 ) domain ( residues 1061–1147 ) ( Stevens et al . , 2010; Dzhindzhev et al . , 2014 ) . This recent progress focuses attention on a detailed mechanistic understanding of the interaction between STIL and PLK4 , and this in turn requires definitive structural information . PLK4 belongs to the PLK family , which in vertebrates comprises four functional paralogues , PLK1-4 . PLKs are characterized by an N-terminal Ser/Thr- kinase domain followed by a C-terminal region containing two or three Polo-box folds ( PB ) , which regulate substrate binding , kinase activity , and localization ( reviewed in Lowery et al . , 2005; Archambault and Glover , 2009; Zitouni et al . , 2014 ) . Among the PLKs , PLK1 is the best studied; it comprises two Polo-boxes , PB1 and PB2 , that form a Polo-box domain ( PBD ) , through intramolecular heterodimerization . The PLK1-PBD generally binds to target proteins after their phosphorylation on Ser/Thr- sites within a PBD-docking motif ( Cheng et al . , 2003; Elia et al . , 2003a , 2003b; Yun et al . , 2009; Xu et al . , 2013 ) ; however , in the context of the Drosophila microtubule-associated protein Map205 phospho-independent binding has also been described ( Archambault et al . , 2008 ) . PLK4 is unique among the PLKs as it contains three -rather than two- Polo-boxes ( PB1-3 ) ( Slevin et al . , 2012 ) . The first two Polo-boxes of PLK4 , PB1 and PB2 ( formerly referred to as cryptic Polo-box [CPB] ) , are sufficient for centriole localization of PLK4 ( Habedanck et al . , 2005; Slevin et al . , 2012 ) . Isolated PLK4-PB3 can also localize to centrioles , but with less efficiency ( Leung et al . , 2002; Slevin et al . , 2012 ) . In contrast to PLK1-PBD , PLK4-PB1/2 as well as PB3 have been described to form intermolecular homodimers and to bind their targets in a different , phospho-independent manner ( Leung et al . , 2002; Slevin et al . , 2012; Kim et al . , 2013; Park et al . , 2014; Shimanovskaya et al . , 2014 ) . Recent work has established a crucial role for the binding of acidic regions in Cep192 and Cep152 to basic residues in PLK4-PB1/2 ( Kim et al . , 2013; Sonnen et al . , 2013; Park et al . , 2014 ) . However , no interactions of PLK4-PB3 with binding partners have been resolved so far . Moreover , the relevance of the reported domain-swapped structure of murine PB3 ( Leung et al . , 2002 ) for in vivo interactions remains unclear . Here , we identify STIL as a direct interaction partner and substrate of PLK4 and confirm that the STIL-CC region is essential for STIL function in centriole duplication . Most importantly , we determined the solution structure of the human PLK4-PB3 and a crystal structure of the PLK4-PB3/STIL-CC complex and use structure-based mutagenesis of STIL to demonstrate an essential role of STIL-CC for PLK4 binding and the regulation of centriole biogenesis in vivo . Specifically , we show that STIL-CC interacts with two regions within PLK4: it targets not only the L1 region but also is the first identified binding partner of the unique PLK4-PB3 . We further show that STIL-CC binding is implicated in the stabilization of centriolar PLK4 and its concomitant activation . Collectively , our results contribute to a detailed structural and mechanistic understanding of a crucial initial step of centriole biogenesis .
To identify centrosomal binding partners of the PLK4 Polo-box motifs , we performed an S-peptide pulldown experiment coupled to mass spectrometry analysis . We generated a U2OS Flp-In T-REx cell line that allowed for inducible expression of an S-peptide-EGFP-tagged PLK4 fragment ( residues 570–970 ) comprising the three Polo-boxes PB1-3 . We identified a set of centrosomal proteins including the two well-known PLK4-PB1/2 binding partners Cep152 and Cep192 ( 16 and four identified peptides , respectively ) ( Cizmecioglu et al . , 2010; Dzhindzhev et al . , 2010; Hatch et al . , 2010; Kim et al . , 2013; Sonnen et al . , 2013 ) . In addition , the key centriole duplication factor STIL co-purified with the PLK4 fragment ( one identified peptide ) ( Firat-Karalar et al . , 2014 ) . This prompted us to further analyze the functional and structural interaction between PLK4 and STIL . As 3D-SIM imaging of U2OS cells revealed extensive co-localization of STIL and PLK4 at the proximal end of daughter centrioles ( Figure 1A ) , we asked whether the two proteins depend on each other for recruitment to this site . Upon depletion of PLK4 , localization of STIL to centrioles was drastically reduced ( 1 . 7 ± 2 . 3% residual intensity compared to untreated cells , Figure 1B ) , suggesting that PLK4 is essential for STIL centriolar targeting and/or maintenance . On the other hand , PLK4 localization to centrioles was not abrogated in STIL depleted cells . On the contrary , centriolar PLK4 levels were strongly elevated and PLK4 localized in a ring- , rather than a spot-like pattern to the outer wall of centrioles ( Ohta et al . , 2014 ) ( Figure 1C ) . Western blot analysis confirmed significant elevation of PLK4 levels in STIL depleted cells ( Figure 1—figure supplement 1A–D ) . This increase in PLK4 levels was comparable to that observed after depletion of βTrCP , which is known to interfere with PLK4 degradation ( Cunha-Ferreira et al . , 2009; Rogers et al . , 2009; Guderian et al . , 2010; Holland et al . , 2010 ) ( Figure 1—figure supplement 1 ) . These data suggest that PLK4 degradation is strongly reduced in the absence of STIL , which then results in its accumulation around centrioles . In further support of a functional interaction between PLK4 and STIL , we also observed that STIL was phosphorylated by a recombinant GST-PLK41-430 fusion protein in vitro ( Dzhindzhev et al . , 2014; Ohta et al . , 2014; Kratz et al . , 2015 ) ( Figure 1—figure supplement 2A ) . 10 . 7554/eLife . 07888 . 003Figure 1 . STIL is an interaction partner of PLK4 . ( A ) U2OS cells were fixed and stained with the indicated antibodies for 3D-SIM imaging . A representative 3D-SIM image is shown , demonstrating the co-localization of PLK4 and STIL at the daughter centriole . Top panel: centrin ( purple ) , PLK4 ( red ) , STIL ( green ) . Scale bar: 0 . 5 µm . Bottom panel: magnified view of the centrosome ( overlay image ) . The rectangles illustrate the orientation of the mother ( M ) and daughter ( D ) centrioles . ( B ) Immunostaining of STIL localization in U2OS cells depleted of endogenous PLK4 . Cells were transfected for 72 hr with control or PLK4 siRNA oligonucleotides and stained with the indicated antibodies . DAPI is shown in blue . Scale bars: 1 µm . ( C ) Immunostaining of PLK4 localization in U2OS cells depleted of endogenous STIL ( control and siSTIL ) . ‘Material and methods’ as in ( B ) . In ( B ) and ( C ) , only prophase cells harboring 1–2 centrioles were analyzed ( indicative of successful PLK4 or STIL depletion , respectively ) . ( D ) Western blots showing the interaction of myc-tagged PLK4 and FLAG-tagged STIL in HEK293T cells . Cells were transfected with the indicated plasmids for 36 hr , followed by lysis and immunoprecipitation using anti-myc or anti-FLAG antibodies . Antibodies used for Western blot detection are indicated . ( E ) Western blot showing the interaction of myc-PLK4 with endogenous STIL . Myc-PLK4 expression was induced by addition of tetracycline to U2OS T-REx cells stably harboring the myc-PLK4 transgene ( ± Tet , 24 hr ) . Cells were processed for anti-myc co-immunoprecipitations and Western blot analysis using the indicated antibodies . DOI: http://dx . doi . org/10 . 7554/eLife . 07888 . 00310 . 7554/eLife . 07888 . 004Figure 1—figure supplement 1 . Plk4 levels are elevated in STIL depleted cells . ( A ) U2OS cells were transfected with control ( siGL2 ) , three different STIL ( siSTIL1-3 ) , βTrCP and Plk4 siRNA oligonucleotides for 72 hr , lysed and subjected to Western blot analysis using the indicated antibodies . ( B ) Graph representing Plk4 band intensities measured in Western blots from two independent experiments as described in ( A ) . Error bars represent SEM . ( C ) HeLa S3 cells were transfected with control ( siGL2 ) , three different STIL ( siSTIL1-3 ) , βTrCP and Plk4 siRNA oligonucleotides for 72 hr , lysed and subjected to Western blot analysis using the indicated antibodies . ( D ) Graph representing Plk4 band intensities measured in Western blots from two independent experiments as described in ( C ) . Error bars represent SEM . ( E ) U2OS cells were transfected with control ( siGL2 ) , three different STIL ( siSTIL1-3 ) , βTrCP and Plk4 siRNA oligonucleotides for 72 hr , fixed , stained with indicated antibodies and analysed in immunofluorescence microscopy . Representative images are shown . Scale bar: 1 µm . ( F ) Graph representing Plk4 levels measured in the centrosomal region of images as described in ( E ) . 15 cells were measured for each condition in three independent experiments , error bars denote SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 07888 . 00410 . 7554/eLife . 07888 . 005Figure 1—figure supplement 2 . STIL is a phosphorylation target of PLK4 and PLK4-ND . ( A ) FLAG-STIL is phosphorylated by GST-PLK41–430 in vitro . HEK293T cells were transfected with FLAG-STIL for 48 hr . After cell lysis , FLAG-STIL was purified using anti-FLAG antibodies and subjected to an in vitro kinase assay with recombinant GST-PLK41–430 or , as a control , with a kinase-inactive version of GST-PLK41–430 ( D154A ) . On the left , the Western blot probed with anti-FLAG antibody shows the amounts of FLAG-STIL used for the reactions . The autoradiograph ( right side ) shows the phosphorylation of FLAG-STIL by GST-PLK41–430 ( upper band ) as well as the autophosphorylation of GST-PLK41–430 ( lower band ) . ( B ) HEK293T cells were transfected with the indicated plasmids and cell extracts were subjected to anti-myc co-immunoprecipitations followed by Western blot analysis . PLK4-ND—the non-degradable PLK4 mutant used throughout this study ( S285A and T289A [Guderian et al . , 2010] ) —exhibits enhanced stabilization and thus facilitates visualization of STIL binding ( higher amounts of FLAG-STIL were detected in the precipitate of myc-PLK4-ND compared to wild-type myc-PLK4 ) . Note the upshift of the STIL band upon co-expression with PLK4-ND , indicating that FLAG-STIL is phosphorylated by myc-PLK4-ND . DOI: http://dx . doi . org/10 . 7554/eLife . 07888 . 005 To confirm the interaction of STIL with PLK4 , we transfected HEK293T cells with myc- and FLAG-tagged versions of both proteins and performed co-immunoprecipitation experiments . As expected , we found STIL and PLK4 to be present in the immunoprecipitates of the respective interaction partner ( Figure 1D ) . Moreover , we detected endogenous STIL along with Cep152 in an immunoprecipitate of myc-PLK4 , which had been isolated from a U2OS T-REx cell line ( Figure 1E ) . Thus , STIL and PLK4 form a stable complex in vivo . To map the region of STIL required for binding to PLK4 , we cloned truncated versions of the STIL protein: an N-terminal ( STIL N-ter . , residues 1–440 ) , middle ( STIL-MD , residues 441–880 ) and C-terminal ( STIL C-ter . , residues 881–1287 ) part , and subjected these fragments to co-immunoprecipitation experiments with myc-tagged PLK4-ND ( Figure 2A , B ) . The PLK4-ND point mutant exhibits enhanced stabilization ( Guderian et al . , 2010 ) and thus facilitates the visualization of STIL binding ( Figure 1—figure supplement 2B ) . We found that the N- and C-terminus of STIL did not bind PLK4-ND , whereas the middle part displayed efficient PLK4 binding . Accordingly , two STIL truncations containing the middle part but lacking either the N- or C-terminus ( STIL-ΔN , residues 441–1287; STIL-ΔC , residues 1–880 ) , strongly bound to PLK4-ND ( Figure 2A , B ) . 10 . 7554/eLife . 07888 . 006Figure 2 . The STIL-CC motif binds to PLK4 . ( A ) Schematic illustration of STIL constructs used to map the PLK4-binding region in STIL . On the right , the relative strengths of the interactions as determined by co-immunoprecipitation experiments are indicated ( + , strong; ± , weak; - , not detected ) . ( B–E ) Western blot analysis of co-immunoprecipitation experiments from HEK293T cells co-expressing STIL fragments or STIL-ΔCC/ΔSTAN mutants and myc-PLK4-ND . Cells were transfected for 36 hr with the indicated plasmids and whole cell lysates were used for co-immunoprecipitation experiments with anti-myc , anti-FLAG or anti-EGFP antibodies . Antibodies for Western blot detection are indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 07888 . 006 The middle region of STIL contains a predicted CC motif ( residues 720–751 ) ( Stevens et al . , 2010 ) . To test the involvement of STIL-CC in PLK4 binding , we further truncated the STIL-MD and analyzed the interaction with PLK4-ND ( Figure 2A , C ) . As long as the CC motif was intact , immunoprecipitation of PLK4-ND was not affected . However , truncating or removing the CC motif severely disrupted PLK4 binding , indicating that PLK4 interacts with the STIL-CC region . This finding was further confirmed by the observation that an EGFP-tagged version of STIL-CC efficiently pulled down PLK4 ( Figure 2D ) , suggesting that STIL-CC alone is sufficient to bind PLK4 . Accordingly , a mutant of STIL lacking the CC motif ( STIL-ΔCC ) did not interact with PLK4 , whereas removal of another conserved region in STIL , the STAN domain ( STIL-ΔSTAN ) , had no impact on the interaction with the kinase ( Figure 2A , E ) . Therefore , the CC domain is both necessary and sufficient for STIL binding to PLK4 . Having established the importance of the CC motif for the PLK4/STIL interaction , we next tested the requirement of this motif for STIL functionality in centriole reduplication . Therefore , we transiently overexpressed STIL-ΔCC in U2OS cells and monitored centriole numbers using immunofluorescence microscopy ( Figure 3A , B ) . Overexpression of wild-type STIL ( STIL-WT ) caused centriole amplification in 45% of transfected cells , and roughly 20% of cells displayed ‘flower-like’ staining , the near-simultaneous formation of several daughter centrioles around one mother centriole ( Figure 3A , B ) . In cells overexpressing STIL-ΔCC only background levels of centriole amplification could be detected ( Figure 3B ) . Moreover , we observed a similar reduction in centriole amplification with the STIL-∆STAN mutant , in line with the requirement for the STAN domain in centriole duplication ( Vulprecht et al . , 2012 ) . Importantly , deletion of the STAN domain had only little impact on centriolar association of STIL , whereas removal of the CC motif strongly impaired the localization of STIL to centrioles ( Ohta et al . , 2014 ) ( Figure 3C ) . Thus , together with the observation that PLK4 depletion leads to loss of STIL from centrioles ( Figure 1B ) , these results indicate that PLK4 directly recruits STIL to the site of centriole formation . We next asked whether STIL overexpression has an impact on the localization of PLK4 to centrioles ( Figure 3C ) . To this end , we overexpressed EGFP tagged STIL in U2OS cells and stained for endogenous PLK4 ( Figure 3D–G ) . Under these conditions , STIL-WT triggered the near-simultaneous formation of several daughter centrioles and PLK4 formed a ring around preexisting centrioles , suggesting that STIL stabilizes centriolar PLK4 ( Figure 3D ) . Overexpression of STIL-ΔCC had no effect on either centriole amplification or PLK4 localization ( Figure 3E ) and , most importantly , overexpression of STIL-ΔSTAN stabilized PLK4 at centrioles , even though this mutant did not cause formation of extra centrioles ( Figure 3F ) . Therefore , we conclude that STIL can stabilize PLK4 at centrioles in a manner that is independent of centriole formation . 10 . 7554/eLife . 07888 . 007Figure 3 . The STIL-CC motif is essential for centriole duplication . ( A ) Immunofluorescence microscopy of U2OS cells transfected with STIL-WT , STIL-ΔCC or STIL-ΔSTAN for 48 hr . Cells were fixed and stained with the indicated antibodies . Scale bar denotes 1 µm . ( B ) Quantification of centriole numbers in U2OS cells after overexpression of the indicated STIL plasmids ( 3 experiments , a total of 300 cells were analyzed for each condition ) . Error bars denote SD . ( C ) Scatter plot to illustrate STIL signal intensity at centrosomes , after overexpression of STIL-WT , STIL-ΔCC or STIL-ΔSTAN ( 20 centrosomes were analyzed for each condition ) . ( D–F ) 3D-SIM images of U2OS cells that have been transfected with EGFP-tagged STIL-WT , STIL-ΔCC and STIL-ΔSTAN and stained with the indicated antibodies . ( G ) Scatter plot to illustrate measured PLK4 signal intensities at centrosomes , after overexpression of STIL-WT/ΔCC or ΔSTAN ( 60 centrosomes were analyzed for each condition ) . Scale bar denotes 1 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 07888 . 00710 . 7554/eLife . 07888 . 008Figure 3—figure supplement 1 . The STIL-CC domain is essential for STIL oligomerization . ( A ) Schematic illustration of STIL constructs used for the co-immunoprecipitation experiments shown in ( B–D ) . On the right , the relative strengths of the interactions are indicated ( + , strong; ± , weak; - , not detected ) . ( B–D ) Western blot analysis of co-immunoprecipitation experiments to map the region required for STIL self-association . HEK293T cells were transfected with the indicated plasmids to co-express the corresponding STIL constructs for 24–36 hr . Subsequently , cells were lysed and co-immunoprecipitations were performed with anti-myc or anti-FLAG antibodies . DOI: http://dx . doi . org/10 . 7554/eLife . 07888 . 008 As STIL has been shown to self-associate ( Tang et al . , 2011 ) , we also tested a possible involvement of the STIL-CC motif in self-interaction . We found that STIL-MD , and , more precisely , the CC motif , is indeed strictly required for STIL self-association , whereas the STAN domain is not ( Figure 3—figure supplement 1 ) . We conclude that the CC motif is critical for the function of STIL in centriole duplication through its role in PLK4 binding , STIL self-interaction , and STIL centriolar recruitment . To determine which regions of PLK4 are involved in STIL binding , we generated a series of EGFP- and FLAG-tagged PLK4 fragments comprising either the N-terminal PLK4 part ( residues 1–570 ) , including the kinase domain ( 1–271 ) and the linker region L1 ( 265–570 ) , or the C-terminal region containing the Polo-boxes ( PB1/2 and PB3 , residues 570–970; PB1/2 , residues 570–820; L2-PB3 , residues 814–970; PB3 , residues 880–970 ) ( Figure 4A ) . Co-expression of the EGFP-tagged PLK4 fragments with FLAG-tagged full-length STIL , followed by anti-EGFP co-immunoprecipitation , revealed that both , the N-terminal PLK4 fragment spanning 1–570 or all C-terminal fragments harboring PB3 were sufficient for the interaction with full-length STIL ( Figure 4B ) . Moreover , the STIL-CC motif alone was sufficient to bind both the N-terminus ( residues 1–570 ) and PB3 ( residues 880–970 ) of PLK4 ( Figure 4C ) . Within the N-terminal PLK4 part , the linker region ( 265–570 ) participated in the interaction , whereas the kinase domain itself ( 1–271 ) did not ( Figure 4D ) . Attempts to further narrow down the interaction region in L1 were unsuccessful , indicating a non-linear folded binding region . To quantify the PLK4-PB3/STIL-CC interaction we determined the binding affinity of PLK4-PB3 and STIL-CC using isothermal titration calorimetry . For this purpose , PB3 ( residues 884–970 ) was recombinantly expressed in Escherichia coli and purified , and a synthetic peptide corresponding to STIL-CC was subsequently titrated into a solution of PLK4-PB3 . The integrated raw data are well fitted by a one-site binding model demonstrating a direct interaction between PB3 and STIL-CC with a dissociation constant Kd of 280 ± 60 nM and an equimolar stoichiometry ( Figure 4E ) . 10 . 7554/eLife . 07888 . 009Figure 4 . PB3 of PLK4 directly interacts with STIL-CC . ( A ) Schematic illustration of PLK4 fragments used to map the STIL-CC binding site . Kinase domain ( KD ) , grey; PB1 , yellow; PB2 , orange; PB3 , blue . The relative strengths of the interactions are indicated ( + , strong; - , not detected ) . ( B–D ) Western blots of co-immunoprecipitation experiments using HEK293T cells co-transfected with plasmids expressing PLK4 fragments and FLAG-STIL ( B ) or EGFP-STIL-CC ( C and D ) . Antibodies used for Western blot detection are indicated . ( E ) Isothermal titration calorimetry of STIL-CC into a solution of PLK4-PB3 . Left panel: Direct measurement of the Gibbs energy associated with STIL-CC binding to PLK4-PB3 . Right panel: integrated and fitted raw data using a one-site binding model . DOI: http://dx . doi . org/10 . 7554/eLife . 07888 . 009 In summary , we find that the STIL-CC motif directly interacts with PLK4-PB3 and that binding occurs with nanomolar affinity . Moreover , we show that STIL-CC additionally interacts with the N-terminus of PLK4 , but not the PB1/2 domain . Importantly , our findings thus identify a novel binding mode for PLK4 , since the previously known PLK4 binding partners Cep152 and Cep192 interact exclusively with the PB1/2 domain ( Cizmecioglu et al . , 2010; Dzhindzhev et al . , 2010; Hatch et al . , 2010; Kim et al . , 2013; Sonnen et al . , 2013 ) . As STIL is the first protein known to bind PLK4-PB3 , we next characterized the structural basis of the PLK4-PB3/STIL-CC interaction . NMR diffusion experiments confirmed that both free and STIL-bound PLK4-PB3 are monomeric in solution ( Supplementary file 1 ) . The structure of human PLK4-PB3 ( residues 884–970 ) was determined by solution NMR spectroscopy ( Figure 5A; Table 1 ) . No crystals diffracting to high resolution were obtained for this construct , despite extensive screening of conditions . In contrast , plate-like crystals of the PLK4-PB3/STIL-CC complex diffracting to 2 . 6 Å resolution were obtained using seeding . The structure was solved by molecular replacement , and refined to Rwork/free of 0 . 22/0 . 25 , respectively ( Table 1 ) with one monomeric PLK4-PB3/STIL-CC per asymmetric unit ( Figure 5B ) . The NMR structure of PLK4-PB3 in solution and the crystal structure of its STIL-CC complex display a conserved overall fold ( Figure 5C ) comprising a six-stranded antiparallel β-sheet ( β1–β6 ) and a C-terminal α-helix ( α1 ) , which packs against the β-sheet and contacts residues from all six strands . The α-helical STIL-CC is bound in a hydrophobic cleft formed by both the β-sheet and the α1 helix of PLK4-PB3 . 10 . 7554/eLife . 07888 . 010Figure 5 . PB3 adopts a canonical Polo-box fold . ( A ) Ensemble of 20 NMR conformers with the lowest target function of free PLK4-PB3 ( dark blue ) ( B ) the X-ray structure of the PLK4-PB3/STIL-CC complex ( light blue/green ) and ( C ) comparison of the free PLK4-PB3 ( dark blue ) to the PLK4-PB3 ( light blue ) in complex with STIL-CC by structural superposition ( STIL-CC not shown for clarity ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07888 . 01010 . 7554/eLife . 07888 . 011Figure 5—figure supplement 1 . Binding of STIL-CC to PLK4-PB3 . ( A ) Overlay of 2D [15N , 1H]-TROSY spectra of 0 . 8 mM [U-15N]-PLK4-PB3 ( dark blue ) and 0 . 6 mM [U-15N]-PLK4-PB3 in complex with unlabeled STIL-CC ( light blue ) recorded at 20°C . The sequence-specific resonance assignments for PLK4-PB3 are indicated . The assignments for PLK4-PB3 in complex with STIL-CC are not shown for clarity . Corresponding backbone amide resonances with chemical shift difference larger than two s . d . ( see below ) between apo and holo PLK4-PB3 are connected by red arrows . ( B ) Chemical shift perturbation of PLK4-PB3 amide moieties upon STIL–CC binding . The magnitude of one s . d . ( 0 . 12 p . p . m . ) is indicated by an orange line and the magnitude of two s . d . ( 0 . 24 p . p . m . ) is indicated by a red line . Asterisks indicate residues that are not assigned in both forms . ( C ) Chemical shift perturbation of PLK4-PB3 upon STIL binding , as mapped on the crystal structure of PLK4-PB3/STIL-CC in ribbon representation . Residues with ( Δδ ( HN ) < 0 . 12 p . p . m . ) , ( 0 . 12 p . p . m . < Δδ ( HN ) < 0 . 24 p . p . m . ) , and ( 0 . 24 p . p . m . < Δδ ( HN ) ) are shown in grey , orange and red , respectively . Unassigned residues are shaded in light grey . The STIL helix is shown green . DOI: http://dx . doi . org/10 . 7554/eLife . 07888 . 01110 . 7554/eLife . 07888 . 012Figure 5—figure supplement 2 . Backbone dynamics of PLK4-PB3 . Measurements of the 15N{1H}-NOE , which is sensitive to local dynamics on the ps–ns time scale . Backbone dynamics of PLK4-PB3 ( dark blue , top ) and PLK4-PB3 as part of the PLK4-PB3/STIL–CC complex ( light blue , center ) and their difference ( black , bottom ) . Asterisks indicate residues that are unassigned in at least one of two forms . DOI: http://dx . doi . org/10 . 7554/eLife . 07888 . 01210 . 7554/eLife . 07888 . 013Figure 5—figure supplement 3 . Secondary structure elements of PLK4-PB3 in solution . Secondary backbone 13C-chemical shifts of PLK4-PB3 ( dark blue ) and PLK4-PB3 as part of the PLK4-PB3/STIL–CC complex ( light blue ) and their differences ( black ) , plotted against the amino acid residue number of PLK4-PB3 . Asterisks indicate unassigned residues . Secondary structural elements as identified in the PLK4-PB3/STIL-CC crystal structure are shown at the top . STIL–CC binding increases the helicity at residues 954 and 955 in helix α1 , as indicated by an increase of secondary 13C chemical shifts . DOI: http://dx . doi . org/10 . 7554/eLife . 07888 . 01310 . 7554/eLife . 07888 . 014Figure 5—figure supplement 4 . PB3 adopts a canonical Polo-box fold . ( A ) Comparison of the fold of PLK4-PB3 ( lightblue ) to the folds of PB1 ( 1 . 7 Å rmsd , 72 Cα ) and PB2 ( 1 . 3 Å rmsd , 79 Cα ) of PLK1 ( PDB accession code: 1Q4O , left ) ( Cheng et al . , 2003 ) and of PB1 ( 2 . 5 Å rmsd , 66 Cα ) and PB2 ( 1 . 4 Å rmsd , 68 Cα ) of PLK4 ( PDB accession code: 4N9J , right ) ( Park et al . , 2014 ) by structural superposition ( Krissinel and Henrick , 2004 ) . For the alignments the crystal structure of PLK4-PB3 was used . ( B ) Crystal structure of the domain-swapped dimer of the PLK4-PB3 murine ortholog SAK in cartoon representation ( Leung et al . , 2002 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07888 . 01410 . 7554/eLife . 07888 . 015Table 1 . NMR and X-ray data collection and refinement statisticsDOI: http://dx . doi . org/10 . 7554/eLife . 07888 . 015PLK4-PB3NMR distance and dihedral constraints Distance constraints Total NOE399 Intra-residue49 Inter-residue350 Sequential ( |i − j| = 1 ) 109 Medium-range ( |i − j| < 4 ) 100 Long-range ( |i − j| > 5 ) 141 Hydrogen bonds16 Total dihedral angle restraints122 ϕ61 ψ61Structure statistics Violations ( mean and s . d . ) Distance constraints ( Å ) 0 . 0158 ± 0 . 0028 Dihedral angle constraints ( ° ) 1 . 686 ± 0 . 147 Max . dihedral angle violation ( ° ) 8 . 40 Max . distance constraint violation ( Å ) 0 . 278 Deviations from idealized geometry Bond lengths ( Å ) 0 . 004 ± 0 . 000 Bond angles ( ° ) 0 . 483 ± 0 . 017 Impropers ( ° ) 1 . 391 ± 0 . 102 Average pairwise r . m . s . deviation* ( Å ) Heavy†1 . 60 ± 0 . 21 Backbone†1 . 09 ± 0 . 25PLK4-PB3/STIL-CCX-ray data collection Space groupC2221 Cell dimensions a , b , c ( Å ) 86 . 6 , 136 . 3 , 33 . 4 α , β , γ ( ° ) 90 . 0 , 90 . 0 , 90 . 0 Resolution ( Å ) 68 . 1–2 . 60 ( 2 . 60–2 . 76 ) Rmerge0 . 15 ( 2 . 37 ) CC1/2 outer shell0 . 63 I/σI10 . 3 ( 0 . 9 ) Completeness ( % ) 99 . 7 ( 99 ) Redundancy6 . 4 ( 6 . 6 ) Refinement Resolution ( Å ) 20 . 80–2 . 60 No . reflections6409 Rwork/Rfree0 . 22/0 . 25 No . atoms Protein1829 Water15 B-factors Protein89 . 7 Water78 . 6 R . m . s . deviations Bond lengths ( Å ) 0 . 01 Bond angles ( ° ) 1 . 15*Pairwise r . m . s . deviation was calculated among 20 refined structures . †Statistics applied for ordered regions ( residues 890–961 ) . To further characterize differences in structure and dynamics of PLK4-PB3 seen upon STIL-CC binding in aqueous solution , free and STIL-CC-bound PLK4-PB3 were subjected to 2D [15N , 1H]-TROSY experiments to reveal chemical shift perturbations and to 15N-{1H}NOE measurements to characterize backbone dynamics on the ps- to ns-timescale ( Figure 5—figure supplements 1 , 2 ) . Chemical shift perturbations are observed throughout most of the PB3 backbone and comprise direct STIL-CC interactions and perpetuated structural changes throughout PLK4-PB3 . The most significant changes locate however around residues C954 and L955 on helix α1 , where a slight kink is formed in apo PLK4-PB3 . Notably , the secondary chemical shifts of these residues are substantially increased upon binding STIL-CC , suggesting a stabilization of helical conformation ( Figure 5—figure supplement 3 ) . STIL-CC binding leads to an increase in averaged heteronuclear NOEs for β-strands and the α1 helix , suggesting a general reduction of fast backbone motions of PLK4-PB3 upon STIL-CC binding . Overall , the structural data reveal two core differences in PLK4-PB3 induced by STIL-CC binding: first , strand β1 is N-terminally extended by three residues to the range 888–893 , resulting in a shortening of the unstructured N-terminal region in the STIL-CC complex . Second , helix α1 slightly changes its orientation and is stabilized by STIL-CC binding ( Figure 5C ) . The structure of the human PLK4-PB3 resembles the canonical structures of related Polo-boxes ( Figure 5—figure supplement 4A ) . It aligns structurally well with both Polo-boxes of the PLK1-PBD ( Cheng et al . , 2003; Elia et al . , 2003b ) . PLK1-PB2 is a close structural homologue ( 1 . 3 Å rmsd , 79 Cα ) , with minor differences only in the linker to the α-helix . PLK1-PB1 is slightly more divergent , in that the C-terminal end of the α1-helix is bent towards the region , where STIL-CC is bound in PLK4-PB3 . PLK4-PB3 also aligns with its two companion Polo-boxes of the PLK4-PB1/PB2 domain ( Slevin et al . , 2012; Park et al . , 2014 ) . Structural divergence to PLK4-PB1 occurs in the β-hairpin region between β3-β4 , which gives PLK4-PB1 its unique winged structure , as well as in the linker between β5-β6 . A major distinction between PLK4-PB3 and PLK4-PB2 is the length of helix α1 , which extends beyond the β-sheet in PLK4-PB2 . PB3 of human PLK4 is closely related to those of its murine ortholog SAK ( 97% sequence identity [Sievers et al . , 2011] ) . Based on the high degree of sequence identity one would expect a highly similar structure for this protein . However , the structure of the human PLK4-PB3 determined here diverges drastically from the structure of murine SAK-PB3 , that was crystallized as a domain swapped dimer ( Leung et al . , 2002 ) ( Figure 5—figure supplement 4B ) . In SAK-PB3 , the β-sheet is formed by strands β2 , β3 and β4 from one monomer and strands β5 and β6 from the second ( numbering according to human PLK4-PB3 ) and the α1 helix is shortened compared to human PLK4-PB3 . The sequence region corresponding to the C-terminal half of this helix is swapped between monomers in SAK-PB3 and transformed into a β-strand , which occupies the position of strand β1 in human PLK4-PB3 . The best explanation for this divergence might be the existence of an equilibrium between a monomeric and a domain-swapped form of SAK-PB3 , the latter of which may be a lowly populated species that is not occurring in the full-length protein in vivo . Either the chemical conditions of crystallization shifted this equilibrium towards the non-native domain-swapped form or crystallization occurred selectively for the domain-swapped state . Nevertheless , our consistent results from solution and crystal structure determination strongly indicate that the canonical Polo-box fold is a relevant physiological state of the PLK4-PB3 domain . PLK4-PB3 and STIL-CC interact along the entire STIL-helix and form a substantial interface of 934 Å2 buried surface area ( Krissinel and Henrick , 2007 ) . Two regions on PLK4-PB3 contribute to the predominantly hydrophobic binding interface: First , the surface of the PLK4-PB3 β-sheet around residues V907 , L917 , V919 , I926 and Y928 ( Figure 6—figure supplement 1 ) , and second the α1 helix ( I948 , L952 , L955 and L959 ) and the linker ( L944 ) leading into it ( Figure 6A ) . Key interacting residues on STIL-CC are leucine and isoleucine residues ( L733 , L736 , I740 , L743 , L744 ) pointing towards the hydrophobic surface of PLK4-PB3 . Additional interactions are provided by backbone–backbone hydrogen bonds between PB3G922-STILQ739 and PB3K943-STILM750 ( Figure 6—figure supplement 1 ) , respectively . The orientation of the two helices and their hydrophobic interactions is mainly mediated by leucine residues and resembles a leucine zipper interaction , consistent with the predicted CC propensity of STIL-CC ( Stevens et al . , 2010 ) . 10 . 7554/eLife . 07888 . 016Figure 6 . Analysis of structure-based STIL-CC mutants . ( A ) Close-up view of the STIL-CC/PLK4-PB3 binding interface . Key contributing residues to the hydrophobic interaction between the PLK4-PB3 ( light blue ) α-helix and the STIL-CC ( green ) α-helix are indicated . ( B ) Multiple sequence alignment ( ClustalW ) of the STIL-CC domain . Residues that directly participate in the PLK4-PB3 interaction are marked on top . Hydropathy index values ( according to Kyte-Doolittle ) for each amino acid are depicted in red . CC probability values ( according to MARCOIL prediction ) for each amino acid are depicted in black below the alignment . The position of each amino acid in the predicted heptad repeat ( labelled a-g , whereas a and d are hydrophobic positions ) is shown in green on top of the alignment . ( C , D ) Control , WT or mutant versions ( M1-7 ) of HA-S-EGFP-tagged STIL-CC were co-transfected with either PLK4-L2-PB3 ( 814–970 ) ( C ) or N-terminal PLK4 ( 1–570 ) ( D ) in HEK293T cells . EGFP-immunoprecipitations were performed and analysed by Western blotting with the indicated antibodies . ( E , F ) To assess centriole amplification , EGFP-tagged WT and mutants of full-length STIL ( M1-7 ) were overexpressed in U2OS cells ( 48 hr ) . EGFP was used as control . ( E ) Quantification of transfected cells with more than 4 centrioles ( n = 3 , 50 cells each ) . ( F ) Immunofluorescence images of U2OS cells after overexpression of EGFP-tagged STIL-WT or mutants M1-7 ( 48 hr ) . Centrioles were visualized by staining with antibodies against CP110 and γ-Tubulin ( gTub ) was used as marker for centrosomes . Scale bar: 1 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 07888 . 01610 . 7554/eLife . 07888 . 017Figure 6—figure supplement 1 . STIL–CC binding to PLK4-PB3 mimics coiled-coil interactions . Close-up views of the STIL-CC/PLK4-PB3 binding interface . Key contributing residues to the interaction between PLK4-PB3 ( light blue ) and STIL-CC ( green ) are indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 07888 . 01710 . 7554/eLife . 07888 . 018Figure 6—figure supplement 2 . STIL–CC binding to PLK4-PB3 resembles an intramolecular interaction of PB2 and Pc in PLK1 . ( A ) Schematic representation of previously observed substrate peptide binding modes in Polo-boxes ( Cheng et al . , 2003; Elia et al . , 2003b; Xu et al . , 2013; Park et al . , 2014 ) . ( B ) Left: Structural superposition of PLK4-PB3 ( light blue ) onto PB2 ( orange ) in the intact PLK1-PB1/2 structure ( Elia et al . , 2003b ) . The bound STIL-CC peptide ( green ) occupies the same position on PLK4-PB3 as the Polo-cap helix ( PLK1-Pc , yellow ) , which binds intramolecularly in the PLK1-PB1/2 structure . Right: Schematic representation of the relative orientation of PLK1-PB1 and PLK1-PB2 and the position of the PLK1-Pc . DOI: http://dx . doi . org/10 . 7554/eLife . 07888 . 018 Polo-box domains are crucial mediators of the interaction of Polo-like kinases with their targets and have been demonstrated to interact with irregular substrate peptides and phosphopeptides . PLK1 , for example , binds phosphopeptides containing a consensus Ser-[pSer/pThr]-[Pro/X] motif ( Elia et al . , 2003a ) through a cleft within its PBD ( comprising PB1 and PB2 ) ( Cheng et al . , 2003; Elia et al . , 2003b; Sledz et al . , 2011 ) and a neighboring binding site on PB1 is used for phospho-independent recognition of a Map205 peptide ( Xu et al . , 2013 ) ( Figure 6—figure supplement 2A ) . The PLK4-PB1/2 domain has recently been shown to bind either two Cep192- or one Cep152-derived peptide in a mutually exclusive manner using large interaction interfaces extending along PB1 and PB2 ( Park et al . , 2014 ) . PLK4-PB3 reveals a novel binding mode by interacting with the helical STIL-CC region in a leucine-zipper-style via the α1 helix with further hydrophobic contacts to the central β-sheet . Remarkably , this external target interaction of PLK4-PB3 closely resembles an intramolecular interaction observed in PLK1: there , the internal Polo-cap ( Pc ) , an N-terminal extension of PLK1-PB1 which comprises an α-helix and a linker to PB1 , directly binds to PLK1-PB2 and thereby determines the relative orientation of PB1 and PB2 ( Elia et al . , 2003b ) . The 17-residue α-helix of the Pc is shorter compared to the 31-residue STIL-CC , but forms a leucine-zipper with the α1-helix of PLK1-PB2 , very much like STIL-CC with the α1-helix of PB3 in PLK4 ( Figure 6—figure supplement 2B ) . Guided by the structure of the PLK4-PB3/STIL-CC complex , we designed seven mutants of STIL to functionally evaluate the role of the PLK4-PB3 interaction with STIL ( mutants M1 to M7 ) : in the first mutant ( M1 ) L733 and L743 were replaced by two large residues ( L733Y , L743Y ) . In the second mutant ( M2 ) as well as in the mutants M4-7 we replaced hydrophobic residues ( at positions a and d of the CC heptads ) with charged amino acids . In mutant M3 we exchanged Q737E and Q739K in order to interfere with hydrogen bond formation ( STILQ739-PB3K943 ) ( Figure 6B ) . To test the impact of these mutations on the binding affinity to PLK4 , we co-transfected HA-S-EGFP-tagged mutants of STIL-CC with either FLAG-tagged PLK4-L2-PB3 ( residues 814–970 ) or N-terminal PLK4 ( residues 1–570 ) . The corresponding EGFP-immunoprecipitation experiments revealed that all mutants except for M3 completely lost binding to PB3 ( Figure 6C ) . Thus , the hydrophobic residues L733 and L743 as well as L736 and I740 are indeed essential for the interaction . Disruption of a hydrogen bond at position Q739 in M3 severely diminished the interaction with PB3 but still allowed for residual binding . Interestingly , all mutants maintained binding to the N-terminal PLK4 fragment spanning residues 1–570 ( Figure 6D ) . This indicates that STIL-CC binds to the two PLK4 regions in different ways and hence may associate with the two regions simultaneously . In parallel to the above binding studies , we also analysed the ability of the STIL mutants to cause centriole amplification . We overexpressed EGFP-tagged full-length WT or mutant STIL proteins in U2OS cells and scored for cells with more than four centrioles . STIL-WT produced centriole amplification in 54% of cells , half of them displaying a flower-like centriole arrangement . In contrast , the mutants M1-2 and M4-7 caused centriole amplification in only 11–14% of cells , comparable to centriole amplification in EGFP-transfected control cells ( Figure 6E , F ) . Interestingly , in the case of M3 , 30% of cells produced amplified centrioles ( 7% with a flower-like centriole arrangement ) ( Figure 6E , F ) , in line with the observed residual binding capacity to PB3 ( Figure 6C ) .
Polo-like kinases are a family of kinases with key regulatory roles in cell cycle progression , mitosis , cytokinesis and centriole duplication , with all human genes thought to have arisen by gene duplication from an ancestral PLK1-like gene ( Zitouni et al . , 2014 ) . PLK4 , the master regulator of centriole duplication , is the most distant member of the family . It is distinguished from all other PLKs by the presence of a third C-terminally located Polo-box , PB3 , in addition to the two central and closely linked Polo-boxes PB1 and PB2 ( Zitouni et al . , 2014 ) . PB1 and PB2 provide a dimerization platform to regulate PLK4 trans-autophosphorylation ( Klebba et al . , 2015 ) and mediate recognition of crucial interaction partners ( Slevin et al . , 2012; Park et al . , 2014; Zitouni et al . , 2014 ) . However , no interaction partner of PLK4-PB3 had previously been identified . Here , we demonstrate that monomeric PB3 of human PLK4 adopts a canonical Polo-box-fold and directly interacts with nanomolar affinity with the central STIL-CC region , STIL-CC . The binding mode of the PLK4-PB3/STIL-CC interaction is completely different from all previously described target interactions of Polo-boxes , but resembles an intramolecular interaction of PB2 with the Pc-helix in PLK1 . We show that the interaction of STIL with PLK4 in vivo regulates centriole biogenesis , confirming and extending recent reports that PLK4-mediated STIL phosphorylation is crucial for SAS-6 recruitment and for triggering centriole duplication ( Dzhindzhev et al . , 2014; Ohta et al . , 2014; Kratz et al . , 2015 ) . We further demonstrate that the STIL-CC region is necessary and sufficient to mediate the STIL-PLK4 interaction and map the binding of STIL-CC on PLK4 to two distinct domains , first to PB3 , and second , to the L1 linker between the kinase and PB1/2 domain . Previous studies indicated that either the N-terminal kinase and L1 ( Kratz et al . , 2015 ) or , alternatively , the PB1/2 domain ( Ohta et al . , 2014 ) of PLK4 are required for STIL interaction , but had failed to reveal an involvement of PB3 . However our in vivo , quantitative biophysical and structural data firmly establish a PLK4-PB3/STIL-CC interaction . The binding of STIL-CC to PB3 resembles a zipper-type CC interaction based on an amphiphilic helix in PB3 . However , the L1 linker region is predicted to not contain any Polo-box folds or other regions favouring CC interactions . This suggests different interaction modes between STIL-CC and PB3 or L1 , respectively , and raises the exciting prospect that STIL-CC could regulate internal interactions between PB3 and the L1 linker region . As further discussed below , this may constitute an important mechanism for regulating PLK4 activity . Furthermore , our data clearly demonstrate a dual role for the STIL-CC region: first , it is an interaction partner for PLK4-PB3 in its monomeric form , and , second , it is involved in STIL-self interactions , presumably via tetramerization ( Cottee et al . , 2015 ) . Mutations in the CC motif clearly abolish the interaction with PLK4-PB3 . However , the same hydrophobic residues of STIL-CC are also predicted to be essential for STIL self-interaction and it is thus likely that these two processes are coupled . Therefore , the inability of STIL-CC mutations to support centriole amplification can be due to either compromised PLK4-PB3 binding , lack of STIL self association or due to failure in both processes . Furthermore , PB3-binding to STIL-CC is expected to affect the self-association of STIL with potential consequence for protein–protein interactions downstream of STIL . Recent studies have revealed a regulatory role for PB3 in the activation of SAK/PLK4 in Drosophila ( Klebba et al . , 2015 ) . Specifically , it was suggested that PB3 , autophosphorylation of L1 , and potentially further yet unidentified protein partners are important for relief of kinase auto-inhibition after SAK/PLK4 homodimerization . Our data on human PLK4 suggest to attribute a key role in such regulatory mechanism to STIL: first , STIL is the only identified interaction partner of PB3 with a direct influence on PB3 structure and dynamics; second , through its CC domain STIL is also able to interact with purported regulatory regions within L1 of human PLK4 . These features make STIL a prime candidate for the role of an external factor regulating relief of PLK4 auto-inhibition in time and space ( Figure 7 , upper panel ) . 10 . 7554/eLife . 07888 . 019Figure 7 . STIL binding to PLK4 regulates centriole duplication . Hypothetical mechanism for STIL-mediated PLK4 activation: ( 1 ) PLK4 is bound to the mother centriole . It is intrinsically inactive , likely due to an autoinhibition by linker L1 . ( 2a ) STIL binds to PLK4 that has been recruited to centrioles through interactions with CEP192 and/or Cep152 . ( 2b ) STIL binding relieves the autoinhibition of PLK4 , thus activating PLK4 . ( 3 ) Activated PLK4 phosphorylates STIL in the STAN motif , which induces SAS-6 recruitment and daughter centriole biogenesis ( 4 ) . Activated PLK4 also phosphorylates neighboring PLK4s in the degradation motif , triggering their degradation . At the site of cartwheel formation , the STIL-bound PLK4 is protected against degradation . DOI: http://dx . doi . org/10 . 7554/eLife . 07888 . 019 Although some mechanistic aspects of centriole biogenesis are likely to differ between species ( Kim et al . , 2014 ) it is interesting to consider our structural data on the interaction between PLK4 and STIL-CC in light of recent insights into the first steps of centriole formation . Specifically , our data supports a model where STIL binds to PLK4 that has been recruited to centrioles through interactions with CEP192 and/or Cep152 ( Figure 7 , lower panel ) , either around the centriolar ring or in a localized dot-like pattern . Additional contributions to stable localization of STIL to centrioles may arise from its STAN-motif or C-terminal region , as a STIL C-terminal truncation ( amino acids 1–1060 ) that contains the CC domain , interfered with correct localization to the centrioles ( Vulprecht et al . , 2012; Arquint and Nigg , 2014 ) . Similarly , the isolated CC domain of Ana2 is not sufficient for centriolar targeting in Drosophila embryos ( Cottee et al . , 2015 ) . Therefore , further interactions , presumably between the STAN-motif or the C-terminal region of STIL and SAS-6 , are likely required to stably integrate STIL into the centriolar cartwheel structure . Once in a complex with PLK4 , STIL is phosphorylated , triggering the recruitment of SAS-6 in preparation for cartwheel formation ( Dzhindzhev et al . , 2014; Ohta et al . , 2014; Moyer et al . , 2015 ) . In parallel , phosphorylation may affect STIL oligomerization and association with the walls of centrioles . In view of most recent studies on PLK4 activation , our data additionally indicate that STIL-CC relieves the auto-inhibition of PLK4 ( Klebba et al . , 2015; Moyer et al . , 2015 ) . The resulting PLK4 trans-autophosphorylation is predicted to cause recruitment of βTrCP-SCF , followed by ubiquitin-dependent proteasomal degradation of activated PLK4 ( Guderian et al . , 2010; Holland et al . , 2010 ) . In line with this idea , depletion of STIL leads to remarkable increase of PLK4 levels ( Figure 1C and Figure 1—figure supplement 1 ) , suggesting that PLK4 is accumulating in an inactive and hence stabilized conformation in the absence of STIL . Our data also indicate that STIL protects activated PLK4 from degradation ( Figure 3D–G ) and ( Ohta et al . , 2014 ) ) , possibly through binding to the PLK4-L1 linker region , which might shield the phosphorylated degradation motif ( DSGHAT , residues 284–289 ) from recognition by βTrCP-SCF . Overall , these mechanisms provide a possible explanation for the concentration of PLK4 at the site of centriole formation ( Figure 7 , lower panel ) . In summary , we have identified and structurally characterized an interaction between STIL and PLK4 , two key centriole duplication factors . We show that the interaction is mediated via the STIL-CC domain and is crucial for centriole biogenesis . Importantly , STIL-CC is the first bona fide interaction partner of PLK4-PB3 . The interaction of STIL-CC with PLK4-PB3 and a second region within the PLK4 L1 linker likely results in PLK4 activation and STIL phosphorylation by PLK4 . These novel insights into the interaction and crosstalk of two key factors in centriole biogenesis provide a new perspective for further work on the critical step of condensing a PLK4 ring to a spot in localizing the initial position of daughter centriole growth . | Centrioles are structures that organize the molecular scaffolding inside cells , which is important for a cell's shape and activity , as well as the segregation of duplicated chromosomes during cell division . Centrioles also form part of the base of the antenna-like structures called cilia , which project out from the cell's surface and allow cells to sense chemicals and touch or even to move . A cell that is not dividing contains a pair of centrioles . In dividing cells , the two centrioles duplicate once per cycle of division and a new centriole forms next to each of the existing ones . It is essential that centrioles duplicate only once , because extra copies can lead to problems that may cause birth defects and cancer . Centrioles require two proteins , called PLK4 and STIL , in order to duplicate . An excess of either of these proteins results in extra centrioles . On the other hand , if these are missing , duplication cannot take place . PLK4 belongs to a large family of enzymes called kinases . A kinase attaches a phosphate group to other proteins , which can either activate or deactivate the other protein . PLK4 can add phosphate groups onto STIL , but it is not known precisely how these two proteins interact with each other . Arquint , Gabryjonczyk , Imseng , Böhm et al . have analyzed this interaction in human cells and found that PLK4 and STIL bind directly to one another . Part of the STIL protein adopts a so-called ‘coiled-coil’ structure in which twisted lengths of protein wrap around each other like a piece of string . The coiled-coil interacts with two different parts of PLK4 . Following on from these observations , the three-dimensional structure of PLK-4 bound to STIL was visualized using X-ray crystallography and nuclear magnetic resonance . These techniques revealed that the coiled-coil region of STIL forms an elongated structure and PLK-4 interacts along its entire length . Arquint , Gabryjonczyk , Imseng , Böhm et al . then analyzed whether PLK4 and STIL need one another in order to get recruited to centrioles . When PLK4 was depleted in cells , STIL was lost from centrioles , suggesting that PLK4 directly recruits STIL . However , contrary to expectations , when STIL levels were reduced , PLK4 accumulated at centrioles . This suggests that STIL maintains appropriate levels of PLK4 via stimulation of its kinase activity . Further work is needed to precisely understand how PLK4 and STIL interact with other proteins that act downstream to lead to the formation of new centrioles in a highly controlled manner . | [
"Abstract",
"Introduction",
"Results",
"Discussion"
] | [
"cell",
"biology",
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2015 | STIL binding to Polo-box 3 of PLK4 regulates centriole duplication |
Digital signaling enhances robustness of cellular decisions in noisy environments , but it is unclear how digital systems transmit temporal information about a stimulus . To understand how temporal input information is encoded and decoded by the NF-κB system , we studied transcription factor dynamics and gene regulation under dose- and duration-modulated inflammatory inputs . Mathematical modeling predicted and microfluidic single-cell experiments confirmed that integral of the stimulus ( or area , concentration × duration ) controls the fraction of cells that activate NF-κB in the population . However , stimulus temporal profile determined NF-κB dynamics , cell-to-cell variability , and gene expression phenotype . A sustained , weak stimulation lead to heterogeneous activation and delayed timing that is transmitted to gene expression . In contrast , a transient , strong stimulus with the same area caused rapid and uniform dynamics . These results show that digital NF-κB signaling enables multidimensional control of cellular phenotype via input profile , allowing parallel and independent control of single-cell activation probability and population heterogeneity .
Cells must make decisions in noisy environments and have to decrease the chance of an errant response . One way cells can reduce sensitivity to noise is through digital or switch-like activation , such that only sufficiently strong signal exceeds an internal threshold and initiates a response . Switch-like activation occurs through diverse mechanisms ( Shah and Sarkar , 2011 ) . For example , observations in Xenopus oocytes showed that the MAPK pathway converted graded progesterone input to digital output in p42 MAPK that determined oocyte maturation ( Petty et al . , 1998 ) . Subsequently , similar observations were seen for the JNK pathway ( Bagowski and Ferrell , 2001 ) . The scaffolding protein Spe5 was found to mediate digital MAPK activation of mating in yeast ( Malleshaiah et al . , 2010 ) . More recently , it was found that inflammasome signaling leads to all-or-none caspase1 activation that mediates apoptosis ( Liu et al . , 2013 ) . Both amplitude ( dose ) and duration of input signals provide information that regulates cellular decisions . The duration of Epidermal Growth Factor ( EGF ) stimulation modulates ERK dynamics and controls differentiation ( Santos et al . , 2007; von Kriegsheim et al . , 2009; Ahmed et al . , 2014 ) . Glucose sensing in plants showed that cells have gene regulatory network mechanisms to allow similar responses to a short , intense or sustained , moderate stimulus ( Fu et al . , 2014 ) . Lymphocytes must precisely measure both antigen affinity and frequency to decide differentiation and proliferation ( Iezzi et al . , 1998; Gottschalk et al . , 2012; Miskov-Zivanov et al . , 2013 ) . Although digital pathway activation allows robust cellular decision across a wide range of systems , it is not clear how digital signaling impacts processing of dose and duration information . NF-κB is a critical regulator of phenotype in immunity and disease ( Hayden and Ghosh , 2008 ) and responds digitally to Tumor Necrosis Factor ( TNF ) stimulation ( Tay et al . , 2010; Turner et al . , 2010 ) . NF-κB activation occurs for a multitude of cell stress and inflammatory signals that converge on the IKK ( IκB Kinase ) signaling hub , which induces degradation of the cytoplasmic inhibitor IκB and liberates NF-κB to enter the nucleus and regulate gene expression ( Hayden and Ghosh , 2008 ) . Multi-layered negative and positive feedback lead to complex pathway dynamics including oscillations ( Hoffmann et al . , 2002; Nelson et al . , 2004; Tay et al . , 2010; Kellogg and Tay , 2015 ) . Although it is not fully resolved how NF-κB coordinates gene and phenotype regulation , it is known that dynamic NF-κB activation is involved in input–output specificity and information transmission ( Werner et al . , 2005; Ashall et al . , 2009; Behar and Hoffmann , 2013; Selimkhanov et al . , 2014 ) . The core IκB-NF-κB regulatory module is well-studied and appears largely consistent across multiple stimulation contexts ( Hoffmann et al . , 2002; Nelson et al . , 2004; Tay et al . , 2010; Hughey et al . , 2014 ) ; however , the role of module upstream of IKK activation including receptor-ligand binding and adaptor protein assembly in input-encoding remains unclear . To probe how diverse IKK-upstream signaling architectures impact NF-κB processing of pathogen- and host-associated inflammatory inputs , we used microfluidic cell culture to precisely modulate dose and duration of LPS and TNF stimuli and measured NF-κB dynamics using live cell imaging ( Figure 1 ) ( Junkin and Tay , 2014; Kellogg et al . , 2014 ) . We found that lipopolysaccharide ( LPS ) induces NF-κB activation in a digital way where cells respond in an all-or-none fashion , but in a distinct manner from TNF , with greater ultrasensitivity and pronounced input-dependent activation delay . Computational modeling predicted and experiments confirmed that LPS integral over the stimulus or ‘area’ ( concentration × duration ) controls the percentage of cells that activate in the population . Importantly , dynamics of NF-κB activation depend on input temporal profile , so that a long duration , low-dose ( LL ) signal induces delayed , heterogeneous activation timing in the population while a short duration , strong amplitude ( SS ) signal with the same area causes rapid activation without cell-to-cell timing variability ( Figure 1 ) . These results reveal a function for digital signaling beyond simple noise filtering: digital activation controls fate along a two dimensional space by allowing an input signal to independently control the population response ( percentage of responding cells ) and single-cell response ( transcription factor dynamics and gene expression phenotype ) though modulation of signal area and shape . 10 . 7554/eLife . 08931 . 003Figure 1 . How does input profile determine digital signaling response ? Since the amplitude and time profile of input signals depends on biological context , such as distance to an infection site or pathogen loading , we use microfluidics to manipulate dose ( A ) and duration ( B ) of LPS and TNF input signals , which induces digital activation of NF-κB . ( C ) Switch-like digital NF-κB responses are analyzed in terms of fraction of cells that activate in the population and heterogeneity in the dynamic responses in activating cells . DOI: http://dx . doi . org/10 . 7554/eLife . 08931 . 003
To initially evaluate the behavior of the LPS/NF-κB pathway , we stimulated 3T3 NF-κB reporter cells ( Lee et al . , 2009; Tay et al . , 2010 ) with different concentrations of LPS in a microfluidic system ( Gómez-Sjöberg et al . , 2007; Kellogg et al . , 2014 ) and performed time-lapse live microscopy to record NF-κB nucleus-cytoplasm translocation over time ( Figure 2A ) . Each experimental condition is measured in duplicate chambers on the chip . We found that LPS-exposed cells activated NF-κB in an ultrasensitive , digital fashion . The population consisted of cells either responding or ignoring the LPS input , with the percentage of responding cells in the population scaling with LPS concentration , from 5% at 0 . 25 ng/ml to 100% at 500 ng/ml ( Figure 2B , D ) . Amplitude is highly variable across doses . Median amplitude increases gradually with dose though this change is statistically less significant than change in response time ( Figure 2—figure supplement 2 ) . NF-κB dynamics in activating cells showed small oscillations beyond the first peak . When ligand is flowed continuously through the chamber to replace ligand loss due to cellular internalization , oscillations sustain for the duration of the stimulus ( Figure 2—figure supplement 1A ) . Under low intensity LPS stimulation , most cells did not respond ( Figure 2C , D ) . This was a similar effect as previously observed under TNF ( Tay et al . , 2010; Turner et al . , 2010 ) . While both LPS and TNF are digital in preserving first peak area , the TNF-induced NF-κB initial peak becomes flatter and wider with increasing response time but unchanging onset time for decreasing input dose , while LPS experiences greater dose-dependent onset delay and timing variability and maintains a consistent peak shape ( Figure 2C and Figure 2—figure supplement 1B ) . The dose-response curve for LPS was steeper than that for TNF ( fitted to Hill dynamics reveals Hill coefficients of 2 . 3 and 1 . 5 , respectively ) ( Figure 2C , Figure 2—figure supplement 1C , Figure 5—figure supplement 1B ) . These results indicate that the LPS pathway activates in a switch-like manner , with increasing fraction of cells in the population responding as dose is increased , but with distinct activation dynamics compared to TNF input . 10 . 7554/eLife . 08931 . 004Figure 2 . Digital , time-delayed NF-κB activation under varied LPS dose stimulation . ( A ) Cells process pathogen signal dose and duration to dynamically activate NF-κB , which induces gene expression and coordinates the innate immune response . We test the role of pathogen load by varying LPS concentration from 0 . 25 ng/ml to 500 ng/ml using microfluidic cell culture . ( B ) Time series images of NF-κB activation following LPS treatment . Top row: high LPS dose causes nearly 100% of cells to respond synchronously . Bottom row: at low LPS concentration , less than 5% of cells respond and initiate NF-κB activation with variable , delayed timing . The cells respond digitally , with nearly all cytoplasmic NF-κB moving into the nucleus . The response amplitude ( indicated by peak intensity of nuclear p65-dsRed fluorescence ) depends on the initial NF-κB abundance in the nucleus and exhibits high variability across doses . ( C ) Trajectories of NF-κB activation ( intensity of nuclear p65-dsRed ) tracked in single cells over time for LPS doses ranging from 500 to 0 . 25 ng/ml . As the LPS dose decreases , response timing becomes delayed and variable , and the percent of responding cells in the population drops . ( D ) Across the LPS doses tested: top panel , dose-response curve of the fraction of active cells ( plotted is the mean of two duplicate cell chambers in the chip for each condition ) , middle panel , the intensity of nuclear NF-κB at the peak of the response , and lower panel , time until the peak of the response . Peak nuclear NF-κB amplitude is highly variable across doses . The dose response shows a sharp drop in fraction of active cells between 1 and 5 ng/ml concentration , indicating that the activation threshold is within this range for most cells . With lower dose , the response time increases in both median duration and variability . In middle and lower panels , data points and error bars represent median and interquartile range , respectively . ( E ) NF-κB dependent gene expression dynamics under varied LPS concentrations ( blue: 500 ng/ml , green: 100 ng/ml , red: 50 ng/ml ) . With lower LPS concentration , several genes show delayed induction . TNF dose-modulated expression of the same genes can be found in Figure 3 of Tay et al . ( 2010 ) and Figure 2A of Pękalski et al . ( 2013 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08931 . 00410 . 7554/eLife . 08931 . 005Figure 2—figure supplement 1 . Digital , time-delayed NF-κB activation under continuous LPS stimulation . ( A ) Sustained NF-κB oscillations for continuously perfused LPS achieving constant LPS concentration . Bolded blue line: example cell . ( B ) Analysis of the LPS induced NF-κB nuclear localization peak . Lower LPS doses induce a pronounced onset delay , but the peak shape is mostly conserved across different doses . Data points and error bars are median and interquartile range , respectively . ( C ) Cartoon illustrating activation curves for LPS vs TNF . ( D ) Cartoon illustrating first peak time profile for LPS vs TNF . LPS leads to greater dose-dependent delay than TNF and a more conserved peak shape . ( E ) Area under the NF-κB localization curve show little change across all doses tested . Data are plotted as median and interquartile range . DOI: http://dx . doi . org/10 . 7554/eLife . 08931 . 00510 . 7554/eLife . 08931 . 006Figure 2—figure supplement 2 . Statistical analysis for NF-κB peak amplitude and timing measurements under LPS dose modulation ( corresponding to Figure 2D ) . Each table contains p-values of two-sample T-test for each dose combination . Red denotes statistical significance ( p < 0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08931 . 006 We next analyzed dynamics of the NF-κB response in those cells that activate . Notably , there were differences in the timing of the response for high versus low dose . High-dose long duration input caused a rapid response with the response peak occurring at approximately 35 min after stimulation . In contrast , low-dose long duration ( LL ) input led to a pronounced statistically significant delay in the response ( Figure 2C , D and Figure 2—figure supplement 2 ) . At lowest doses , the median delay until the peak of the response exceeded 80 min with heterogeneity in the response timing between cells ( Figure 2C , D ) . We next asked whether this delayed response impacted LPS and NF-κB-mediated gene expression . We explored how the increase in delay for 500 ng/ml LPS versus 50 ng/ml LPS impacted gene regulation . Notably , gene expression of early and intermediate genes exhibited a dose-dependent delay ( Figure 2E ) . The extent and magnitude of the dose-dependent delay and heterogeneity differs from TNF stimulation of the same cell type ( Tay et al . , 2010 ) . While decreasing TNF dose altered the response slope , LPS response maintained a stereotypical peak shape that shifts later in time with lower dose ( Figure 2E ) . Delayed gene expression observed under LPS stimulation contrasts with TNF-α input that does not induce delayed induction of these genes ( Tay et al . , 2010; Pękalski et al . , 2013 ) . For early genes IκBα and A20 , gene expression peak is shifted from 30 min to 1 hr after stimulation . IκBϵ expression shifts from maximum expression at 1 hr–2 hr and from 30 min to 2 hr for TNF mRNA under LPS input . Intermediate genes Ccl2 and Icam shift expression peaks from 1 hr to 2 hr and from 1 hr to 3 hr ( Figure 2E ) , respectively . Late genes Ccl5 and Casp4 do not reflect the delayed NF-κB activation due to slower induction kinetics . Both NF-κB dynamics in microfluidics and mRNA responses in tissue culture may be affected by autocrine signaling loops ( Pękalski et al . , 2013 ) . Overall , these results indicate that LPS induces digital NF-κB activation with an input dose-dependent delay that carries through to gene expression dynamics . To study how various pathway components upstream of IKK influence input information transfer to NF-κB , we developed a model of LPS-induced NF-κB switch activation . LPS activates NF-κB by TLR4 engagement via CD14 , leading to TLR4 dimerization . TLR4 dimers recruit MyD88 , IRAK2/4 , and other adaptor proteins leading to clustering and higher order assembly of Myddosome and TRAF6 lattice structures , which cooperatively activates IKK ( Yin et al . , 2009; Lin et al . , 2010; Zanoni et al . , 2011 ) . Following IKK activation , nuclear NF-κB induces expression of IκBα , which negatively regulates NF-κB and IKK , respectively ( Hayden and Ghosh , 2008 ) . Experimental IκBα expression kinetics were similar for LPS and TNF , despite induction delay under LPS ( Figure 2E ) . Multiple efforts have modeled NF-κB pathway dynamics under TNF stimulation ( Hoffmann et al . , 2002; Lipniacki et al . , 2007; Ashall et al . , 2009; Paszek et al . , 2010; Tay et al . , 2010; Pękalski et al . , 2013 ) . Extrinsic noise including variation in receptor-level and pathway components contributes to cell-to-cell heterogeneity in cell sensitivity and response dynamics ( Snijder et al . , 2009; Tay et al . , 2010 ) . We based our mathematical model on the core IKK-NF-κB regulatory module ( Tay et al . , 2010 ) , which has been extensively validated experimentally . To extend the NF-κB core model for LPS , we added species for LPS , TLR4 , and TRAF6 ( Appendix 1 ) ( Figure 3A ) . To introduce variability in the model , we allowed fluctuation in the number of TLR4 receptor molecules between cells . TLR4 is expressed at relatively low level compared to CD14 and furthermore varies significantly between cells in the population ( Zanoni et al . , 2011 ) . To account for cooperative activation due to Myddosome assembly and TRAF6 lattice formation , we model IKK phosphorylation by TRAF6 using Hill kinetics . The model reproduced the observed LPS induced NF-κB dynamics in single cells for different LPS doses ( Figure 3B–D and Figure 3—figure supplement 1 ) , though showed more dose-dependent first peak amplitude variation than observed in experiments . The distinct feature of the proposed LPS model is the Myddosome formation leading to cooperative activation of IKK ( coopertivity coefficient = 4 , note this value is distinct from the slope of the active cell fraction response curve ) , which simultaneously assures delay in activation observed experimentally for low doses , and steeper response curve than in the case of TNF stimulation ( Figure 3 and Figure 2—figure supplement 1 ) . 10 . 7554/eLife . 08931 . 007Figure 3 . Model scheme and simulation of LPS dose modulation . ( A ) The scheme of the model . LPS binds TLR4 leading to TRAF6 activation , which cooperatively activates IKK . Active IKK induces IκB degradation , which allows NF-κB to enter the nucleus and upregulate expression of IκB and A20 . New IκB sequesters NF-κB in the cytoplasm and A20 inhibits upstream pathway activation by IKK and TRAF6 . ( B ) Simulated versus experimental LPS dose response . ( C ) NF-κB peak intensities ( expressed as proportion of total NF-κB molecules in the nucleus ) . ( D ) NF-κB response time as function of LPS dose . ( E ) Sample simulated curves of nuclear NF-κB fractions under LPS treatment . In box and whisker plots ( C , D ) , the central red line is the median , the edges of the box are the 25th and 75th percentiles , and the whiskers extend to the most extreme data points . DOI: http://dx . doi . org/10 . 7554/eLife . 08931 . 00710 . 7554/eLife . 08931 . 008Figure 3—figure supplement 1 . Simulated NF-κB trajectories for various doses of LPS treatment . ( A–H ) LPS concentration decreases over time due to cellular internalization in sealed microfluidic chambers , leading to damped oscillations . ( Note: y-scale changes between plots . ) DOI: http://dx . doi . org/10 . 7554/eLife . 08931 . 00810 . 7554/eLife . 08931 . 009Figure 3—figure supplement 2 . Modelling predictions for the LPS pathway . ( A ) Extrinsic noise is generated by selecting the number of TLR4 molecule for each cell from a lognormal distribution . Fraction of active cells was analyzed for increased TLR4 variability ( parameter Sigma ) . With higher TLR4 number variability , the change in fraction of active cells becomes more gradual with changing LPS concentration . ( B ) Simulation response delay under varied LPS dose with and without clustering . Clustering mediating cooperative IKK activation is required to reproduce the experimentally observed response delay with decreasing LPS dose . Central point and error bars represent median and interquartile range , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 08931 . 009 Cellular environments like infected tissue encode information in both amplitude and duration of input signals ( Gottschalk et al . , 2012; Fu et al . , 2014 ) . To understand how pathogen input signal duration and input integral or area ( concentration × duration ) impact digital NF-κB signaling , we performed a simulated screen across a large range of LPS concentration and duration combinations using our model . We first observed that just as LPS concentration modulates fraction of active cells so does duration . Simulations keeping concentration high and changing input duration on a short , sub-minute timescale altered the percent of activating cells ( Figure 4 ) . Nearly , all cells respond for durations exceeding 1 min at 500 ng/ml . Notably , in contrast to changing concentration under constant long duration , which introduces timing delay and heterogeneity ( Figures 2C , 3D ) , changing sub-minute duration of a high-amplitude signal in simulation controlled fraction of active cells while maintaining uniformly timed , rapid NF-κB responses ( Figure 4B , D and Figure 4—figure supplement 1 ) . 10 . 7554/eLife . 08931 . 010Figure 4 . Model simulation predicts that stimulus duration controls fraction of activating cells and response timing variability . ( A ) Simulated fractions of activated cells under increasing duration of 500 ng/ml LPS pulse . ( B ) Sample simulated curves of NF-κB under 5- to 40-s duration LPS ( 500 ng/ml ) pulse . ( C ) Distributions of nuclear NF-κB amplitude and ( D ) response times of activating cells under various durations of LPS treatment . In box and whisker plots ( C , D ) , the central red line is the median , the edges of the box are the 25th and 75th percentiles , and the whiskers extend to the most extreme data points . DOI: http://dx . doi . org/10 . 7554/eLife . 08931 . 01010 . 7554/eLife . 08931 . 011Figure 4—figure supplement 1 . ( A–F ) Simulated NF-κB single-cell trajectories with randomly sampled numbers of TLR4 and NF-κB for 1- to 60-s durations of LPS exposure . The concentration of LPS is 500 ng/ml . Color: activated cells . Gray: inactivated cells . ( Note: y-scale changes between plots . ) DOI: http://dx . doi . org/10 . 7554/eLife . 08931 . 011 We experimentally provided pulsed LPS at 500 ng/ml for sub-minute durations using microfluidic cell culture ( Figure 5A ) . In agreement with simulation predictions , we found that precisely controlled stimulus duration regulated the activation of cells in the population in a strongly all-or-none manner , with 1- to 40-s duration LPS exposure ( 500 ng/ml ) activating ∼3–88% of the population ( Figure 5B and Supplementary Videos 1 , 2 ) . Short duration ( i . e . , 1 s ) stimulation , mimicking very brief exposure to bacteria , activated a small percentage of cells in the population . Moreover , under short duration , strong amplitude ( SS ) input , responses were fast and uniform ( Figure 5B , top row ) in contrast to low , long ( LL ) stimulation that led to delayed , variable responses ( Figure 2C ) . For example , 3–5% activation occurred for both a 1-s short pulse at 500 ng/ml LPS ( SS signal ) and a 0 . 25 ng/ml constant input signal ( LL signal ) ( Figure 5B , Figure 2D ) . However , modulating duration of the SS signal from 1 s to 40 s ( activating 3 . 3% and 87 . 5% the population , respectively ) changes median response timing by less than 2 min ( Figure 5B , C ) . Statistical analysis indicates no significant difference in response time under varied pulse durations ( Figure 5—figure supplement 2 ) . In contrast , modulating concentration of the LL signal from 0 . 25 to 500 ng/ml ( activating 4 . 9% and 98% of the population , respectively ) changes the response time more than 35 min ( reduced from 80 to 43 min ) . Moreover , while the variability in the response time scales with dose under LL stimulation , timing variability remains low under duration-modulated SS stimulation ( Figures 4D , 5C ) . From an immunological perspective , this experiment indicates that brief but high pathogen load leads to uniform and strong NF-κB response in the population , while chronic low-grade pathogen exposure leads to population variability and delay . 10 . 7554/eLife . 08931 . 012Figure 5 . Short duration LPS pulse stimulation modulates responding cell fraction and fast , uniformly timed response . ( A ) LPS duration is manipulated using microfluidic cell culture in the range of 1–40 s . Dose is held constant at 500 ng/ml . ( B ) Single-cell NF-κB trajectories for 1- to 40-s duration LPS pulse stimulation . Short pulse LPS reduces variation in timing in the start of NF-κB activation . ( C ) Top panel: fraction of active cells as a function of LPS pulse duration . ( Plotted points are the mean of two duplicate chambers in chip . ) Middle panel: NF-κB nuclear response intensity as a function of LPS pulse duration . Lower panel: time of the NF-κB response peak as a function of LPS pulse duration . Timing variability is dramatically reduced under short-pulsed stimulation ( see Figure 2D for comparison to constant stimulation ) . In middle and lower panels , data points and error bars represent median and interquartile range , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 08931 . 01210 . 7554/eLife . 08931 . 013Figure 5—figure supplement 1 . NF-κB dynamics under pulsed stimulation with TNF at 10 ng/ml concentration . ( A ) Single-cell NF-κB trajectories for TNF pulse durations of 1 , 5 , 10 , 20 , 40 , and 60 s . The number and percent of activated cells is indicated in the plot . ( B ) Comparison of active NF-κB cell fraction ( top ) , response amplitude ( middle ) , and response time ( bottom ) under LPS vs TNF pulses . The fraction-active curve is less steep under TNF than LPS . Fraction of active cells is plotted as mean for two duplicate cell chambers for each condition . In middle and lower panels , data points and error bars represent median and interquartile range , respectively . LPS data are duplicated from main Figure 5C for comparison . DOI: http://dx . doi . org/10 . 7554/eLife . 08931 . 01310 . 7554/eLife . 08931 . 014Figure 5—figure supplement 2 . Statistical analysis for NF-κB peak amplitude and timing measurements under LPS duration modulation ( corresponding to Figure 5B ) . Each table contains p-values of two-sample T-test comparing each duration combination . Red denotes statistical significance ( p < 0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08931 . 01410 . 7554/eLife . 08931 . 015Figure 5—figure supplement 3 . Statistical analysis for NF-κB peak amplitude and timing measurements under TNF ( 10 ng/ml ) duration modulation ( corresponding to Figure 5—figure supplement 1 ) . Each table contains p-values of two-sample T-tests comparing each duration combination . Overall , amplitude is affected to a greater extent than response time under changing TNF duration . Red denotes statistical significance ( p < 0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08931 . 01510 . 7554/eLife . 08931 . 016Video 1 . Digital NF-κB response unders 1 second duration LPS exposure ( 500 ng/ml ) . This video shows fibroblast cells expressing NF-κB p65-dsRed responding in a digital fashion to brief ( 1-s duration ) stimulus in a microfluidic chamber . Only 3–4% of cells show a response . DOI: http://dx . doi . org/10 . 7554/eLife . 08931 . 01610 . 7554/eLife . 08931 . 017Video 2 . Digital NF-κB response under 10 second duration LPS exposure ( 500 ng/ml ) . This video shows fibroblast cells expressing NF-κB p65-dsRed responding in a digital fashion to 20-s duration stimulus in a microfluidic chamber , activating more than half ( ∼55% ) of cells in the population . DOI: http://dx . doi . org/10 . 7554/eLife . 08931 . 017 We next changed the stimulus type to TNF instead of LPS . In vivo , TNF is secreted from immune cells that come in contact with pathogenic signals like LPS . Again , we observed the phenomena that the fraction of active cells changed while the response timing did not ( Figure 5—figure supplement 1 ) , though amplitude is more significantly affected ( Figure 5—figure supplement 3 ) . Together , these results indicated that SS input achieves control over the fraction of cells activating without affecting the dynamics in the response . Therefore , duration sensing allows control of percentage of cells that produce a response without affecting response timing or heterogeneity . This contrasts to amplitude ( concentration ) sensing , where response dynamics in activating cells differs for high versus low amplitude . Since NF-κB dynamics influence gene expression , duration modulation to control percent population activation is therefore a strategy to achieve more homogeneous gene expression and phenotype outcomes between cells . We sought to fully characterize the relationship between signal amplitude and duration in NF-κB switch activation . Since modulating either amplitude or duration was able to change the percentage of activating cells , we hypothesized that the fraction of activation may depend on the integral of the input ( concentration × duration ) . Indeed , mathematical analytical analysis suggested that percent activating cells should scale with the input area ( Appendix 1 ) . To validate our mathematical analysis and clarify how digital activation integrates stimulus dose and duration , we performed simulations . Each simulation series fixed the LPS stimulus dose and varied duration from 1 to 500 s . The output of these simulations as a function of stimulus duration shows multiple dose-response curves that do not coincide , indicating that duration is not the only predictor of switching probability or fraction of active cells ( Figure 6A ) . However , when instead plotted as a function of stimulus area ( concentration × duration ) , all simulation series closely coincide , indicating that stimulus area clearly determines the percentage of cells that activate in the population ( Figure 6A ) . 10 . 7554/eLife . 08931 . 018Figure 6 . Simulations demonstrating an ‘Area Rule’ , that is , the relationship between LPS stimulus area and fraction of active cells . ( A ) The simulated fractions of activated cells for pulsed inputs of LPS with various doses and durations . Each fraction is estimated by 500 independent simulations . When the points are plotted as a function of stimulus area ( rather than duration ) , all points fall on the same curve , indicating that stimulus area tightly controls the fraction of active cells . ( B ) The minimal duration for certain fractions of activation as a function of dose . The minimal duration is determined by searching for the first tested time point where the estimated fractions of activation are above the threshold . The doses for which the threshold level cannot be achieved are not shown in the figure . Blue: 10% activation , green: 50% activation , red: 90% activation . ( C ) Further verification of the relationship between stimulus area and active cell fraction using square wave input profiles . Equal area input was generated using either a single pulse ( top left ) , square wave with 10-s period ( lower left ) , or square wave with 20-s period ( top right ) . Regardless of input shape , all simulated points fall on the same curve when plotted as a function of stimulus area . For square wave inputs , one input begins high ( blue ) while another input ( green ) begins low . Note that the curves intersect for durations 10 s , 20 s , 30 s ( or 20 s , 40 s , 0 s , … for the input with 20-s period ) when area under the two signals is the same . DOI: http://dx . doi . org/10 . 7554/eLife . 08931 . 01810 . 7554/eLife . 08931 . 019Figure 6—figure supplement 1 . Stimulus area simulation using a TNF model ( Tay et al . , 2010 ) revealed that the ‘Area Rule’ holds also for TNF . ( A ) Simulated fractions of activated cells for pulsed inputs of TNF with various doses and durations . ( B ) The minimal duration for certain fractions of activation as a function of dose . ( C ) Verification of the relationship between stimulus area and active cell fraction using square wave input profiles . Equal area input was generated using either a single pulse ( top left ) , square wave with 10-s period ( lower left ) , or square wave with 20-s period ( top right ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08931 . 019 To illustrate further the relationship between stimulus area and percentage of active cells , we plotted for each simulation dose the minimum duration needed to achieve 10% , 50% , and 90% activation ( Figure 6B ) . This analysis revealed a reciprocal relationship between dose and duration in NF-κB switch activation ( high dose requires less duration to achieve activation and vice versa ) . Simulations therefore supported analytical derivation of an ‘Area Rule’ in which concentration × duration determines the percentage of cells that activate in the population for a given stimulus . Importantly , for concentrations that achieve less that 100% activation , increasing the duration infinitely will not further increase the active fraction ( Figure 6B ) . Once duration is sufficiently long to activate the maximal potential cell fraction for a given dose , further increases in area by lengthening duration do not further increase percentage of active cells , indicating a limitation in the Area Rule . We next simulated whether the Area Rule holds for fluctuating signals . When we compared a constant input signal to square wave input signals , with one square wave that ‘starts high’ and another that ‘starts low’ , simulation revealed an equal percentage of activating cells when the two opposing square waves have equal area ( i . e . , the duration is a multiple of the square wave period ) ( Figure 6C ) . Further , the fraction of active cells matched that for a constant input signal with the same area ( Figure 6C ) . Performing identical simulations using a model of TNF-induced NF-κB activation ( Tay et al . , 2010 ) , we found that the Area Rule held also for the TNF network ( Figure 6—figure supplement 1 ) . We found that in both the LPS amplitude-modulated and duration-modulated microfluidic experiments , stimulus area is an accurate predictor of fraction of active cells in the population ( Figure 7A ) , as predicted by the model simulations . To experimentally verify that the integral over time of the input signal determines the fraction of activating cells , we performed microfluidic experiments varying temporal profile while maintaining the same integrated area . We observed 500 ng/ml LPS pulsed for 10 s activated approximately half the population ( Figure 5C ) . Therefore , we tested two additional input profiles having the same area ( 5000 ng ml−1*s ) : 50 ng/ml for 100 s and 100 ng/ml for 50 s . In agreement with model prediction , each of these conditions also activates a similar fraction of the population ( 51% and 54% , respectively ) ( Figure 7C ) . Together , experimental findings , simulations , and mathematical analysis demonstrate how cells integrate amplitude and duration of input signals in switch-like pathway activation . Stimulus integral ( or area ) determines the effective ‘probability’ that a given cell activates NF-κB ( based on the percentage that activate in the population ) . These results indicate that the pathogen load ( i . e . , LPS dose ) and duration of exposure ( i . e . , LPS pulse duration ) are integrated by NF-κB system and together determine the population response . 10 . 7554/eLife . 08931 . 020Figure 7 . Stimulus area determines NF-κB population response . ( A ) Stimulus area determines fraction of active cells . The experimentally tested dose and duration inputs fall on the same hill-like activation curve when plotted as a function of stimulus area , as predicted by model simulations , indicating that total integrated ligand concentration ( stimulus area ) controls the probability of cell activation . These results show that pathogen load ( i . e . , LPS dose ) and duration of exposure ( i . e . , LPS pulse duration ) are integrated by NF-κB system and together determine the population response . ( B ) Response time discriminates between sustained , low intensity ( blue ) and transient , high intensity ( red ) stimulus . Data points and error bars represent median and interquartile range , respectively . ( C ) Experimental verification that stimulus integral over time determines the fraction of active cells . A pulse of either LPS 50 ng/ml for 100-s duration or 100 ng/ml for 50-s duration generated approximately the same responding cell percentage as a , 51% and 54% for the two inputs , respectively . ( D ) Input profile controls digital responses along two axes: integral over stimulus ( area ) controls the fraction of activated cells in the population . Input temporal profile ( shape ) controls dynamic heterogeneity in responding cells . In ( B ) , data points and error bars represent median and interquartile range . DOI: http://dx . doi . org/10 . 7554/eLife . 08931 . 02010 . 7554/eLife . 08931 . 021Figure 7—figure supplement 1 . LPS response heterogeneity under pulsed and continuous input . ( A ) Comparison of cell-to-cell dynamic heterogeneity for different fractions of active cells for either constant or pulsed input ( Figures 2 and 5 ) . Pulsed ( duration-modulated ) LPS input ( red ) achieves lower response time variability than constant ( dose-modulated ) input for the same fraction of active cells . ( B ) Unlike response delay , response intensity does not provide sufficient information to distinguish between pulsed LPS and constant LPS signals . Data points and error bars represent median and interquartile range , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 08931 . 021 We showed that modulating stimulus amplitude altered response dynamics by changing the amount of activation delay . In contrast , modulating stimulus duration did not affect activation delay but changed the population variability . These findings indicate tradeoffs in dose versus duration sensing . Duration sensing allows for controlling only the population response while not affecting the single-cell response , so that it is possible to achieve homogeneous dynamics and uniform phenotype in a desired proportion of cells . In contrast , it may be useful to transmit information that can instruct different dynamics and phenotype , which is achieved by modulating dose . While dose information is transmitted through the NF-κB digital response , duration information is lost at the single-cell level . However , transmitting information using only dose modulation necessarily changes the percentage of cells in the population that respond . In physiological settings , it may be desirable to transmit information without affecting population response , that is , for a signal to affect response dynamics in activating cells without impacting the proportion that activate . To achieve this requires modulating both dose and duration to maintain input area , leading to a shift in input temporal profile from a SS to a LL signal . Cells distinguish an SS versus LL signal profile based on NF-κB and gene expression dynamics . We show that an intense , brief ( SS ) signal induces distinct dynamics than a weak , sustained ( LL ) signal , but the percentage of cells responding is the same in both cases ( Figure 2C , Figure 5B ) . Response timing and intensity are dynamic features that provide information for discriminating input temporal profile . Indeed , plotting response delay as a function of stimulus area shows that SS and LL signals can be distinguished on the basis of response delay ( Figure 7B ) . Cells can discern the category of the signal ( whether SS or LL ) for a given input area based on whether the response time falls above or below a separation line ( Figure 7B ) . Modulating input amplitude associated with higher timing variability than input duration modulation for controlling the fraction of active cells ( Figure 7—figure supplement 1A ) . Because response amplitude exhibits high variability between cells , amplitude alone does not provide sufficient information to discriminate an SS versus LL signal ( Figure 7—figure supplement 1B ) . Physiological cues are in fact commonly transmitted by changing from an SS to an LL input profile ( Iezzi et al . , 1998; Fu et al . , 2014 ) . Biological systems therefore appear to take advantage of the unique ability of digital signaling to separate control of population and single-cell dynamics , by modulating input area to determine the proportion of cells that activate in the population and input shape to instruct to determine phenotype outcomes in the activating subset of cells ( Figure 7 ) .
This study asked how stimulus amplitude and duration determine NF-κB digital activation . Modeling and experiments showed that NF-κB activation is achieved by integrating the input: stimulus integral or area ( concentration × duration ) controlled the percentage of cells that activated for both a ‘foreign’ pathogen signal LPS and a ‘self’ immune signal TNF ( population response ) . However , switch dynamics and gene expression phenotype varied depending on the input dose ( single-cell response ) , with rapid homogeneous responses at high dose and delayed heterogeneous responses at low dose . Dynamics of transcription factor activation determine the timing and specificity of gene expression and phenotype responses ( Werner et al . , 2005; Kobayashi et al . , 2009; Purvis et al . , 2012 ) . Therefore , intercellular signaling systems may achieve distinct phenotype outcomes by controlling the input shape or temporal profile ( whether SS or LL ) , while input area determines percentage of cells that respond ( Figure 7B ) . Greater heterogeneity with decreasing dose and decreased heterogeneity under short duration input is measured by coefficient of variation ( Figure 7—figure supplement 1 ) . In lymphocyte signaling , T- and B-cells' cell fate depends on both antigen quality ( affinity ) and quantity ( amount of presented antigen ) . Antigen quality is encoded in the duration of receptor-antigen contact , with characteristic interaction times on the order of seconds ( Altan-Bonnet and Germain , 2005; Gottschalk et al . , 2012; Miskov-Zivanov et al . , 2013 ) . T- and B-cell receptor binding with antigen-MHC triggers digital activation and cell fate control via NF-κB ( Kingeter et al . , 2010; Oh and Ghosh , 2013; Gerondakis et al . , 2014; Shinohara et al . , 2014 ) . A reciprocal relationship is observed between antigen quality and quantity in lymphocyte activation: Higher antigen affinity requires lower dose of antigen to trigger T-cell proliferation , and inversely , lower affinity requires higher dose ( Gottschalk et al . , 2012 ) . Moreover , an intense , transient compared to a weak , sustained signal induces positive versus negative selection of naive thymic T cells ( Iezzi et al . , 1998 ) and T helper cell differentiation into alternatively CD4 or CD8 status ( Adachi and Iwata , 2002 ) . Therefore , analogous to our findings , while a combination of antigen dose and contact duration determines the probability of activation , input profile determined by relationship between antigen quality and quantity decides the phenotypic outcome of lymphocyte activation . We show that switch-like signaling enables parallel and independent control over response probability and response dynamics: while stimulus area ( concentration × duration or antigen quantity × quality ) regulates the percentage of cells that respond , the stimulus temporal profile or shape ( for example , whether short-strong or low-long or antigen quality/quantity ratio ) determines the response timing and gene expression phenotype in responding cells . Dose and duration sensing may be beneficial in different contexts . Dose information is encoded in the delay timing and heterogeneity of NF-κB response . On the other hand , modulating duration on the sub-minute timescale does not regulate response dynamics . Indeed , achieving control of percentage active in a population without introducing heterogeneity requires modulating duration of a high-dose input ( Figure 4 ) . It was shown that signaling dynamics mediates transfer of input dose information ( Selimkhanov et al . , 2014 ) . We find that while dose information is transmitted through dynamics of NF-κB activation , on short ( minute ) time scales duration , information is lost in the single-cell response but retained in the population response ( fraction of activated cells ) . Between the innate immune signals TNF and LPS , we found that LPS exhibits greater ultrasensitivity ( a steeper stimulus-response curve ) and more pronounced activation delay than TNF . Both of these features are explained by higher coopertivity in IKK activation for LPS than for TNF ( Figure 3—figure supplement 2B ) . Distinct higher order adapter protein architectures may activate IKK with different effective coopertivities ( Kazmierczak and Lipniacki , 2010 ) . We note that the LPS-signaling response in macrophages may differ from that in 3T3 cells , including effects due to stronger auto and paracrine TNF signaling . While TNF signaling activates formation of a filamentous amyloid complex involving RIP1 and RIP3 kinases , LPS signaling is mediated through helical assembly of the Myddosome complex ( Lin et al . , 2010 ) , which interfaces with a TRAF6 lattice structure to activate IKK ( Yin et al . , 2009 ) . Heterogeneity in switching threshold between cells may arise from cell-to-cell expression differences in signalosome components such as RIP1/3 , MyD88 , IRAK2/4 , and TRAF6 , leading to altered kinetics of signalosome assembly and IKK activation . Because the IKK hub mediates NF-κB responses for a multitude of input types and coordinates cross-talk with other signaling pathways , understanding how different signalosome architectures induce specific responses paves the way to interventions directed at switch-like signaling to modulate population and individual cell dynamics towards therapeutic outcomes ( Negro et al . , 2008; Behar et al . , 2013 ) . In this study , we have shown that the switch-like character of NF-κB activation enables orthogonal control over two critical aspects of the response—probability of activation ( fraction of active cells ) and the heterogeneity of response—through the integral and temporal profile of the input . Secretion of signaling molecules often occurs in discrete or quantized way in the form of secretory bursts , and particularly in the case of short range paracrine signaling , cells may produce brief but intense secretion to achieve , for example , low probability but high predictability responses ( non-heterogeneous dynamics ) . Overall , these results expand the repertoire of functions for digital signaling beyond increasing robustness to also facilitate multidimensional phenotype control based on temporal information in input signals .
We used p65-knockout 3T3 fibroblasts ( courtesy Markus Covert ) modified using lentiviral vectors to express p65-DsRed under its endogenous promoter along with an H2B-GFP nuclear reporter , as described previously ( Lee et al . , 2009 ) . The cell line was clonally derived to express at p65-DsRed at lowest detectable level to preserve near endogenous expression . Automated microfluidic cell culture was performed as previously described ( Gómez-Sjöberg et al . , 2007; Tay et al . , 2010; Kellogg et al . , 2014 ) . Briefly , microfluidic chambers were fibronectin treated and seeded with cells at approximately 200 cells/chamber . Cells were allowed to grow for 1 day with periodic media replenishment until 80% confluence . To stimulate cells , media equilibrated to 5% CO2 and containing the desired LPS amount was delivered to chambers , leading to a step increase in LPS concentration . All LPS doses were tested in parallel in a single chip . To produce LPS and TNF pulses , chambers were washed with media after incubation with ligand for the desired duration . Stimulations were applied in duplicate chambers on the chip . Following stimulation , chambers were sealed and imaged at 5- to 6-min intervals . DsRed and GFP channels were acquired using a Leica Microsystems ( Wetzlar , Germany ) DMI6000B widefield microscope at 20× magnification with a Retiga-SRV CCD camera ( QImaging - Surrey , BC , Canada ) using Leica L5 and Y3 filters to acquire GFP and DsRED signals , respectively , and a Leica EL6000 mercury metal halide light source . One or two images were acquired per chamber and stitched if required using ImageJ ( Pairwise stitching plugin ) . CellProfiler software ( www . cellprofiler . org ) and custom Matlab software was used to automatically track cells and quantify NF-κB translocation , and automated results were manually compared with images to ensure accuracy prior to further analysis . Mitotic cells were excluded from analysis . NF-κB activation was quantified as mean nuclear fluorescence intensity normalized by mean cytoplasm intensity . Area of the first peak was integrated after baseline correction from the time of LPS stimulation to the first minimum for each cell using Matlab function trapz . For peak analysis , data were smoothed ( Matlab function smooth ) followed by peak detection ( Matlab function mspeaks ) to extract NF-κB peak properties ( intensity , area , delay ) with manual verification using a custom interface in Matlab . Statistical analysis of NF-κB peak amplitude and timing data was performed by unpaired two-sample T-test ( Matlab function ttest2 ) . Cells were seeded at 10 , 000 cells/well in a 96-well plate and left to attach overnight before stimulation with LPS ( 1–500 ng/ml ) . Cells were stimulated with LPS and then lysed , the RNA reverse-transcribed and cDNA pre-amplified ( specific target amplification with the set of 24 primers ) using the One-Step RT-PCR kit from Invitrogen ( San Diego , CA , United States ) . Quantitative PCR with technical duplicates was carried out on 48 . 48 dynamic arrays from Fluidigm according to manufacturer instructions , and expression was normalized to GAPDH . | Cells have communication systems called signaling pathways that enable them to detect and respond to changes in their surrounding environment . For example , in humans and other animals , a signaling pathway called NF-κB signaling is part of the immune system and regulates the inflammation that is caused by damage to cells , or by an invading microbe . Several signal molecules , including a protein called TNF—which is released by cells during an immune response—activate NF-κB signaling . However , the levels of TNF in the environment around a cell may fluctuate randomly even when there is no immune response . Therefore , the NF-κB pathway needs to be able to tell the difference between this ‘noise’ and a large increase in TNF associated with an immune response . To get around this problem , many signaling pathways are activated in a switch-like manner so that only a strong signal that exceeds a particular threshold will lead to a response . These so called ‘digital’ responses help cells to filter out noise caused by random fluctuations in the amount of a signal molecule . NF-κB signaling responds to TNF in a digital manner , but it is not clear how information about the length of the signal can influence the degree to which NF-κB signaling is activated . Kellogg et al . used a combination of mathematical modeling and microscopy techniques to study the activation of NF-κB signaling in mouse cells . The study shows that a molecule called LPS—which is produced by microbes known as bacteria—can also switch on the signaling pathway in a digital manner , but in a different way to TNF . In a population of cells , the fraction that activate NF-κB signaling in response to LPS or another signal is determined by the level of the signal ( also known as its ‘concentration’ ) multiplied by the signal's duration . This is known as the signal's ‘area’ . On the other hand , the way that these cells respond to the activation of NF-κB signaling depends on the nature of the activity produced by the signal pathway . For example , a short but strong burst of LPS signal leads to rapid and uniform responses in the cells . A weaker but longer lasting signaling activity leads to slower , more varied responses in cells . These findings reveal that such switch-like , digital responses do more than just filter out noisy signals . They can also integrate information about the timing and intensity of the signal to independently control different aspects of cell responses . The next challenge will be to extend this understanding to more complex scenarios , such as when signals contain several types of molecules at the same time . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"computational",
"and",
"systems",
"biology",
"immunology",
"and",
"inflammation"
] | 2015 | Digital signaling decouples activation probability and population heterogeneity |
The Eph receptor tyrosine kinase ( RTK ) family is the largest subfamily of RTKs playing critical roles in many developmental processes such as tissue patterning , neurogenesis and neuronal circuit formation , angiogenesis , etc . How the 14 Eph proteins , via their highly similar cytoplasmic domains , can transmit diverse and sometimes opposite cellular signals upon engaging ephrins is a major unresolved question . Here , we systematically investigated the bindings of each SAM domain of Eph receptors to the SAM domains from SHIP2 and Odin , and uncover a highly specific SAM–SAM interaction-mediated cytoplasmic Eph-effector binding pattern . Comparative X-ray crystallographic studies of several SAM–SAM heterodimer complexes , together with biochemical and cell biology experiments , not only revealed the exquisite specificity code governing Eph/effector interactions but also allowed us to identify SAMD5 as a new Eph binding partner . Finally , these Eph/effector SAM heterodimer structures can explain many Eph SAM mutations identified in patients suffering from cancers and other diseases .
The Eph ( erythropoietin-producing hepatocyte ) transmembrane receptor tyrosine kinase superfamily , with its first member identified30 years ago ( Hirai et al . , 1987 ) , contains 14 members in mammals and is the largest among all receptor tyrosine kinase families ( Lemmon and Schlessinger , 2010; Manning et al . , 2002; Murai and Pasquale , 2003 ) . Chiefly based on their engaging ephrin ligands , Eph receptors are classified into the EphA and EphB subfamilies , each with nine and five members in mammals , respectively ( Eph Nomenclature Committee , 1997; Gale et al . , 1996; Murai and Pasquale , 2003 ) . Owing to broad expressions in essentially all tissues and at every life stage , Ephrin-Eph signaling regulates many cellular processes both during development and in developed animals such as stem cell maintenance and differentiations , tissue morphogenesis , and tissue–tissue boundary formation ( Batlle and Wilkinson , 2012; Genander and Frisén , 2010; Jülich et al . , 2009; McMillen and Holley , 2015; Munarini et al . , 2002; Park et al . , 2011; Poliakov et al . , 2004 ) . Not surprisingly , mutations of ephrins and Ephs are known to cause many forms of diseases including cancers and brain disorders ( Boyd et al . , 2014; Chen et al . , 2008; Hahn et al . , 2012; Kania and Klein , 2016; Merlos-Suárez and Batlle , 2008; Pasquale , 2008; Zhuang et al . , 2012 ) . Ephrin ligand binding-mediated inter-cellular signaling is the classic mode of Eph receptor signaling ( also known as ephrin-Eph ‘forward’ signaling ) , which is responsible for the majority of cellular functions characterized for the ephrin-Eph signaling ( Pitulescu and Adams , 2010; Taylor et al . , 2017; Yokoyama et al . , 2001 ) . Presumably , the versatile forward ephrin-Eph signals are transmitted by the cytoplasmic portion of Eph receptors . However , the cytoplasmic portion of all 14 Eph receptors are highly similar , each containing a membrane-juxtaposing kinase domain , a protein-binding SAM domain immediately followed by a short carboxyl tail PDZ domain-binding motif ( PBM ) ( Figure 1A ) . The cytoplasmic portions of Eph receptors are often presumed to function similarly . However , multiple Eph receptors are typically co-expressed in one tissue . Paradoxically , it has been observed that two different Eph receptors on the same cell type can respond to a single ephrin ligand but elicit opposite cellular responses . For example , ephrin-A5 binds to EphA2 and EphA4 with similar affinity but induces cell adhesion or cell collapse , respectively ( Cooper et al . , 2008; Zhou et al . , 2007 ) , although different multimerization modes of EphA2 and EphA4 extracellular domains induced by ephrin-A5 binding can also contribute to the opposite cell spreading phenotype ( Seiradake et al . , 2013 ) . Therefore , the cytoplasmic domains of Eph receptors must be able to engage different intracellular effectors in response to ephrin ligands . How specific Eph receptor cytoplasmic domain-mediated signaling might occur has been a major unresolved question in the ephrin-Eph signaling . We reasoned that the SAM domain of each Eph receptor is likely to play a role in specifying their cytoplasmic effector engagements for the following two reasons . First , SAM domain is a well-known protein–protein interaction module ( Qiao and Bowie , 2005 ) , and the SAM domain of EphA2 is known to bind to SAM domain from SHIP2 ( SH2 domain-containing Inositol 5'-Phosphatase 2 , aka INPPL1 for INositol Polyphosphate Phosphatase-Like protein 1 ) and Odin ( aka Anks1a ) ( Kim et al . , 2010; Lee et al . , 2012; Leone et al . , 2009; Mercurio et al . , 2012; Zhuang et al . , 2007 ) , though the binding properties of the SAM domains from other Eph receptors are largely unknown . Second , although the PBM sequences of Eph receptors are somewhat different , the short PBM-mediated target bindings are rather promiscuous ( Ye and Zhang , 2013 ) and thus unlikely to be fully responsible for the very diverse Eph intracellular signaling events . In this study , we systematically characterized and compared the bindings of the SAM domain from every Eph receptor to the SAM domains from SHIP2 and Odin . This characterization revealed a highly specific Eph SAM and effector SAM-binding pattern . We then elucidated the mechanistic basis governing such specific Eph SAM and effector SAM binding by solving several pairs of the SAM-SAM heterodimer complexes structures . Such comparative structural analysis , together with biochemical , bioinformatics and cell biology studies , revealed an exquisitely specific effector binding code mediated by the Eph SAM domains , which helps to answer the major question on the ephrin-Eph forward signaling specificity . Additionally , our study also provides mechanistic explanations to numerous disease-causing mutations identified in the SAM domains of Eph receptors , and allows us to discover SAMD5 as a new intracellular effector of Eph receptors .
SHIP2 is a mammalian inositol polyphosphate 5-phosphatases and is the only member in the family that contains a C-terminal SAM domain ( Figure 1A ) . It has been reported that SHIP2 was a binding partner of EphA2 through SAM–SAM interaction ( Lee et al . , 2012; Leone et al . , 2009 ) . We first confirmed this interaction . Both Eph SAM and SHIP2 SAM alone behaved as homogeneous monomers in solution as indicated by the analytical gel filtration analysis ( Figure 1B ) . While the 1:1 mixture of EphA2 SAM and SHIP2 SAM was eluted at a smaller volume than the individual proteins , suggesting the formation of a hetero SAM-SAM complex ( Figure 1B ) . ITC ( Isothermal Titration Calorimetry ) experiment revealed that EphA2 SAM bound to SHIP2 SAM with a dissociation constant ( Kd ) of ~2 . 22 μM at a 1:1 stoichiometry ( Figure 1C ) . We then measured the binding affinities of SHIP2 SAM with the SAM domain from other members of Eph receptors by ITC . We found that only the SAM domains of EphA1/EphA2/EphA6 specifically bound to SHIP2 SAM , and the rest of Eph SAM domains displayed no detectable binding to SHIP2 SAM ( Figure 1D ) . Odin was reported to be another binding partner of EphA2 ( Kim et al . , 2010; Mercurio et al . , 2012 ) , and the interaction had been implicated in affecting the stability of EphA2 by modulating its ubiquitination process ( Kim et al . , 2010 ) . We also verified the interactions between Odin and Eph receptors SAM domains ( Figure 1E and F ) . The results showed that , similar to SHIP2 SAM , Odin SAM1 only selectively bound to the SAM domains from EphA1/EphA2/EphA6 ( Figure 1D ) . The quantitative and systematic binding results shown in Figure 1D indicated that the interactions between Eph SAMs and their downstream effectors were highly specific . To elucidate the molecular mechanism governing the specificity of the bindings of SHIP2 SAM and Odin SAM1 to Eph receptors , we determined the crystal structure of the SHIP2 SAM-EphA2 SAM complex and the Odin SAM1-EphA6 SAM complex at the 1 . 5 Å and 1 . 3 Å resolutions , respectively ( Table 1 ) . The crystals diffracting at very high resolutions of both complexes were facilitated by fusing the Eph SAM domain to the C-terminal tail of the SHIP2 SAM or Odin SAM1 with a flexible linker ( 14 residues ‘SSGENLYFQSGSSG’ for the SHIP2 SAM-EphA2 SAM complex; 17 residues ‘PSGSSGENLYFQSGSSG’ for the Odin SAM1-EphA6 SAM complex ) . The covalent linkage did not appear to affect the overall structure of the complexes , as the linkers in both complexes are sufficiently long and flexible ( Figure 2—figure supplement 1 ) . The two complex structures adopt an essentially identical End-Helix/Mid-Loop binding mode , in which positively charged residues from the N-terminal end of α5 ( End-Helix ) of EphA2/EphA6 bind to negatively charged residues from the loop connecting α2-α4 ( Mid-Loop ) from SHIP2/Odin , forming the well-known tail-to-head SAM domain heterodimer ( Figure 2A–E ) ( Qiao and Bowie , 2005 ) . The structures determined in this study were also consistent with an earlier NMR-derived EphA2 SAM/SHIP2 SAM heterodimer ( Lee et al . , 2012; Leone et al . , 2009 ) . However , it should be noticed that some of the critical features mediating the specific SAM–SAM interaction revealed in our study were not revealed in the NMR structures ( Figure 2 and Figure 2—figure supplement 2 ) , likely due to insufficient distance restraints of the NMR experiments . To avoid redundancy , we will not describe the detailed binding interactions of the two complexes here , except that we further validated some of such charge–charge interactions using site substitution approach ( Figure 2H ) . The structures of the complexes also revealed that the surfaces of the EphA2/EphA6 SAM Mid-Loop do not complement with the surfaces of the End-Helix of their own or with those of the SHIP2/Odin SAM domains ( Figure 2—figure supplement 3 ) , explaining that these four SAM domains neither form homo-oligomers nor polymerize into hetero-oligomers ( Knight et al . , 2011; Qiao and Bowie , 2005; Stapleton et al . , 1999; Thanos et al . , 1999 ) . A noticeable feature in both complexes is that the backbone methylene of a Gly residue at the beginning of α5 from Eph SAM ( Gly954A2/Gly1104A6 ) is in close contact with an aromatic residue from SHIP2/Odin SAM ( Trp1221SHIP2/Phe738Odin ) ( Figure 2B and D and Figure 2—figure supplement 4 ) . As such , replacing Gly at the beginning of α5 of Eph SAM with any other amino acid residues will introduce steric hindrance in preventing their binding to SHIP2/Odin SAM . It is further noted that Gly is highly preferred at the beginning of α5 among Eph SAM domains ( 12 out of the 14 members are Gly; see Figure 2—figure supplement 5A ) . The very high-resolution crystal structures of the two complexes also allowed us to identify a unique interaction feature that is critical for the exquisite specific interaction between the EphA2/A6 SAM domain and SHIP2/Odin SAM domains . Taking the EphA2/SHIP2 complex as the example , a special cation-π interaction between the Arg958A2 and Phe1226SHIP2 was uncovered by the high-resolution crystal structure ( Figure 2F ) . The guanidinium group of Arg958A2 forms hydrogen bond network with Asp1222SHIP2 and His955A2 on one side and with the backbone carbonyl oxygen of Ile917A2 . As such the guanidinium plane of Arg958A2 ( and its delocalized π-system ) is in the same plane with the π-system of the planer peptide bond between Ile917A2 and Lys918A2 , forming energetically favorable π-π stacking with the benzene ring of Phe1226SHIP2 ( Figure 2F ) ( Ma and Dougherty , 1997 ) . The exactly same interaction pattern occurs for the EphA6/Odin complex ( Figure 2G ) . It is predicted that alteration of any of the interactions in the above-described π-π stacking will perturb the binding of EphA2/6 to SHIP2 or Odin . The most subtle substitution of Arg is probably by the positively charged Lys . Consistent with the binding pattern , the corresponding residue of Arg958 in EphA2 is also found in EphA1 and A6 , which could interact with SHIP2 and Odin . Whereas the rest of Eph SAMs contain a Lys in this position except EphA10 , which is an Ala ( Figure 2—figure supplement 5 ) . Based on the structures shown in Figure 2F , replacing Arg958A2 with Lys would eliminate the planar π-system formed by Arg958A2 and the Ile917A2-Lys918A2 peptide bond and thus seriously weaken the binding , even though the positive charge at the site is retained . Totally consistent with this structural analysis , substitution of Arg958A2 with Lys completely eliminated the bindings of EphA2 SAM to SHIP2 SAM or Odin SAM ( Figure 2H and I ) . Correspondingly , substitution of Phe1226 of SHIP2 SAM with non-aromatic hydrophobic residues ( e . g . Leu or Ala ) also totally eliminated the binding between SHIP2 SAM and EphA2 SAM ( Figure 2H and J ) . Additionally , replacing Asp1222 in SHIP2 SAM with Ala also eliminated the binding between SHIP2 and EphA2 , highlighting the importance of the hydrogen bond between Asp1222SHIP2 and Arg958A2 in stabilizing the guanidinium group of Arg958 ( Figure 2H ) . We mined COSMIC cancer somatic mutation database ( Forbes et al . , 2017; http://cancer . sanger . ac . uk/cosmic ) and found that an Arg957 ( corresponding to mouse Arg958 ) to Cys somatic mutation of EphA2 has been detected in ovary carcinoma patients . As expected , substitution of Arg with Cys also completely eliminated EphA2 SAM’s binding to the SHIP2 or Odin ( Figure 2H and I ) . The above structural and biochemical analysis highlights the critical role and the exquisite selectivity of Arg958 in EphA2 ( or the corresponding Arg in EphA1/EphA6 ) in terms of bindings to the effectors such as SHIP2 and Odin . We pushed this concept further by testing whether we might be able to convert a non SHIP2/Odin binding Eph SAM domain into a binding one by simply substituting the Lys at the position corresponding Arg958A2 to Arg ( i . e . a single residue ‘gain-of-function’ mutation ) . We chose EphA5 SAM domain to test this hypothesis , as it shares ~50% sequence identity to EphA2 SAM but has no or minimal binding to SHIP2 or Odin ( Figure 2H ) . Satisfyingly , substituting Lys857 ( corresponding Arg958A2 ) with Arg converted EphA5 SAM into a SHIP2 binding SAM domain , though the binding was still relatively weak . The same substitution also enhanced EphA5 SAM’s binding to Odin SAM1 by more than 100-fold ( Figure 2H , bottom ) . Taken together , the above structural and biochemical analysis highlighted that the SAM–SAM domain interactions between Eph receptors and their cytoplasmic effectors are highly specific . We next tested the cellular function of the specific SAM domain-mediated effector association of the EphA2 . DU145 cells , which have a low expression level of endogenous EphA2 receptors , were used to study its ligand-induced inhibition of cell spreading ( Barquilla et al . , 2016; Lee et al . , 2012; Miao et al . , 2000; Shi et al . , 2017 ) . To focus our studies on the forward signal pathway of EphA2 , we chose to engage the EphA2 receptor by adding soluble ephrinA1-Fc chimera to the cell culture media . To minimize potential trans signaling from neighboring cells , all experiments were performed in low-density cultures ( at about 30% confluency or below ) . Cells were infected with the lenti-virus containing the 3 × Flag tagged EphA2 FL WT , EphA2-R958K , EphA2 with the SAM domain deleted ( EphA2 delSAM ) , and EphA2 with its SAM substituted by EphA5 SAM ( EphA2-SAMA5 chimera ) . Cells infected with the 3 × Flag tagged GFP were the negative controls . As expected and consistent with the previous reports ( Barquilla et al . , 2016; Shi et al . , 2017 ) , in response to ephrinA1-Fc stimulation , cells expressing the wild-type full-length EphA2 receptor dramatically retracted and became rounded , whereas the negative control cells did not respond to ephrinA1 ( Figure 3A and quantified in Figure 3B ) . Using this specific EphA2-mediated cell rounding assay , the repulsion phenotype of the cells expressing EphA2 delSAM , EphA2 R958K or EphA2-SAMA5 chimera ( each expressed at a similar level with that of EphA2 WT; Figure 3C ) were evaluated . None of them displayed ephrinA1-induced cell rounding ( Figure 3A and B ) , indicating that the highly specific EphA2 SAM domain-mediated effector association is required for the cell spreading function of EphA2 . We next performed cell spreading assay by transfecting SHIP2-SAM to cells infected with lenti-virus expressing EphA2 or the GFP control . The data showed that overexpressing SHIP2-SAM effectively reversed the cell collapse phenotype caused by ephrinA1 activation of EphA2 , whereas over-expressing SHIP2-SAM did not lead to any observable changes in the GFP control ( Figure 3—figure supplement 1 ) , indicating that this blocking effect by SHIP2-SAM is EphA2-dependent . Arg957 ( corresponding to Arg958 of mouse EphA2 SAM ) , which is found to be mutated to Cys in patients with ovary carcinoma , is located in the SAM-SAM binding interface . We also tested the impact of this mutation as well as the R958K mutation using HEK293T cells , another cell line widely used for cell spreading assay ( Lawrenson et al . , 2002; Yamazaki et al . , 2009 ) . Similar to the R958K mutation in the DU145 cells assay , we also observed that HEK293T cells expressing either of the EphA2 R958C or R958K mutants lost the capacity to undergo ephrinA1-induced retraction , whereas cells expressing EphA2 WT showed effective ephrinA1-induced retractions ( Figure 3—figure supplement 2 ) . The above crystal structures of EphA2-SHIP2 and EphA6-Odin revealed the importance of the negatively charged residues within the α2-α3 loop and at the beginning of the α4 helix of SAM domain in binding to Eph SAM ( Figure 2 and Figure 4A ) . We tried to look for new binding partners for Eph SAMs . By inspecting each of the 163 SAM domains in 127 mouse SAM domain containing proteins ( SMART’s nrdb database: http://smart . embl-heidelberg . de/ ) , we found several potential candidates that may interact with Eph SAM . AIDA1 , also named as Ankyrin repeat and sterile alpha motif domain-containing protein 1B ( Anks1b ) , is a paralog of Odin ( Anks1a ) and thus its SAM1 domain is expected to bind to EphA2/A6 SAM with the same mode and specificity as Odin SAM1 does . Another promising candidate is SAMD5 , whose SAM domain is similar to SHIP2 SAM ( Figure 4A ) and highly expressed in breast cancer cells and mainly cytoplasmic ( Lo et al . , 2015 ) . We purified SAMD5 SAM and measured its binding to the SAM domain of every Eph receptor . Indeed , SAMD5 SAM was found to be a strong binder of Eph SAM domains such as EphA5 and others ( Figure 4B–D ) . Surprisingly , although the amino acid sequence of SAMD5 SAM can be aligned well ( the key binding residues in particular ) with the SAM domain sequences of SHIP2 and Odin ( Figure 4A , top ) , SAMD5 bound to EphA5-8 and EphB1-4 SAM domains with strong affinities but only weakly to EphA1/2 SAM domains ( Figure 4D ) . This biochemical data further substantiated that the SAM domains of Eph receptors can provide diverse binding specificities to their effectors . To delineate the mechanism governing the Eph/SAMD5 interaction , we solved the EphA5/SAMD5 SAM–SAM complex crystal structure at the resolution of 1 . 9 Å ( Table 1 ) . Similar to that of EphA2/SHIP2 or EphA6/Odin , EphA5 SAM also uses its End-Helix to interact with the Mid-Loop SAMD5 SAM ( Figure 5A and B ) . However , their detailed binding mechanisms are quite different . The most prominent difference is that the EphA5/SAMD5 SAM-SAM does not contain the π-π stacking observed in the EphA2/SHIP2 and EphA6/Odin complexes ( Figure 5C ) . Instead , the EphA5/SAMD5 SAM-SAM is essentially completely mediated by charge–charge and hydrogen bonding interactions ( Figure 5C ) . Perturbation of these interactions invariably weakened the interaction ( Figure 5E ) . Similar to what we observed in the EphA2/SHIP2 and EphA6/Odin structures , Gly853A5 at the beginning of α5 in EphA5 SAM is also critical for its binding to SAMD5 SAM . The backbone amine of Gly853A5 forms a strong hydrogen bond with Tyr27SAMD5 sidechain hydroxyl group ( Figure 5C ) . This hydrogen bond brings the α2-α3 loop of SAMD5 very close to α5 of EphA5 ( Figure 2—figure supplement 4C ) , and the site has no room to accommodate any residues with side chains , further highlighting the importance of a Gly at this position in the Eph SAM domains ( Figure 2—figure supplement 5A ) . To our surprise , substitution of Tyr27SAMD5 with Phe completely eliminated SAMD5’s binding to EphA5 ( Figure 5E ) , indicating the critical role of the Gly853A5-Tyr27SAMD5 hydrogen binding in addition to their steric role . Our crystal structure of the EphA5/SAMD5 complex can explain why SAMD5 does not bind to EphA1-4 , EphA10 , and EphB6 . For the SAM domains from EphA3/EphA4/EphA10/EphB6 , each has one or more of the key residues missing ( e . g . EphA10 and EphB6 are missing Gly at the beginning of α5 , and EphA3/4 SAMs are missing positively charged Lys at the end of α2 or in the middle of α5; Figure 2—figure supplement 5 ) . For the EphA1/EphA2 SAM domains , although all the key residues are present , it is noticed that a Pro residue is at the position corresponding to Val852A5 ( Figure 2—figure supplement 5 ) . The sidechain pyrrolidine ring of Pro would introduce steric hindrance with Asp40 , and thus may perturb SAMD5 from binding to EphA1/EphA2 SAM domains ( Figure 5D ) , resulting in low affinities ( Figure 4D ) . Supporting this analysis , replacing Val825 of EphA5 SAM with a Pro led to about 10-fold affinity decrease in its binding to SAMD5 . Additionally , substitution Pro953 in EphA2 with a smaller residue Ala increased its SAMDs SAM binding by more than 10 fold ( Figure 5E ) . Finally , EphA6 SAM could bind to both SAMD5 and SHIP2/Odin SAM domains ( Figure 1D and 4D ) , since it satisfies the binding criteria of both types . Eph receptors mediate cell-cell contact signaling and are vital for cell adhesion and migration ( Kania and Klein , 2016 ) . It is not surprising that mutations of Eph receptors can cause various human diseases including cancers ( Barquilla and Pasquale , 2015; Boyd et al . , 2014; Gaitanos et al . , 2015; Genander and Frisén , 2010; Kania and Klein , 2016; Pasquale , 2005; 2008; 2010 ) . We surveyed the COSMIC cancer somatic mutation database ( Forbes et al . , 2017; http://cancer . sanger . ac . uk/cosmic; a collection of somatic mutations found in cancer patients mainly from large-scale genome sequencing studies ) and found that numerous mutations occur in almost every subtype of Eph genes except for EphB6 . A significant number of these mutations fell into the SAM domain regions ( Figure 6A ) . The biochemical and structural information provided in this study allowed us to test the impact of these mutations found in cancer patients in terms of binding to the cytoplasmic effectors including SHIP2 and SAMD5 . For the practical workload reason , we chose to investigate the mutation sites that are at the positions critical for the SAM/SAM interactions analyzed in our structural studies shown in Figure 2 and 5 , and are shaded in orange in Figure 6A . We purified each of these mutant SAM domains of Eph proteins , and measured their individual bindings to SHIP2 SAM or SAMD5 SAM by ITC-based assays . The results are summarized in Figure 6B and C . As expected , EphA1 R966C , EphA2 R957C , and EphA6 R1014Q mutations ( the cation-π forming Arg at α5 helix ) totally abolished their binding to SHIP2 ( Figure 6B ) . Similar deleterious effect can also be seen in EphA1 α1-α2 loop mutation R926G ( Figure 6B ) . EphA5 G1014S , EphA7 G972V , and EphB3 G974D mutations disrupted their binding to SAMD5 ( Figure 6C ) , further supporting our structural finding that the sidechain-less Gly at the beginning of α5 is critical ( Figure 2—figure supplement 4 ) . Mutations on other critical sites such as EphA5 K1018M , EphA6 R1014Q , EphA8 R943H , and EphB1 K926M , A961T all weakened or even disrupted their binding to SAMD5 ( Figure 6C ) . Taken together , the above analysis of the impact of the Eph SAM mutations found in cancer patients illustrated the tremendous values of the structural and biochemical studies of Eph SAM-mediated target interactions presented in the current study .
Since the discovery of the Eph receptors three decades ago , numerous efforts have been invested in elucidating ligand-induced signaling mechanisms of this classical family of RTKs in many biological processes . However , in contrast to the explosive amount of information on the functions uncovered , our understandings of the forward signaling mechanisms of the receptors are relatively poor . It is particularly unclear how Eph receptors may engage their specific cytoplasmic effectors to transmit biological signals using their highly similar cytoplasmic domains . At present , very little knowledge is available on what downstream targets of Eph receptors are and whether each Eph receptor engages different cytoplasmic effectors upon ligand binding , and this has seriously hampered our understanding on the action mechanisms of this family of RTKs in both physiological and patho-physiological conditions . In this study , we performed systematic biochemical and structural studies of the bindings of three SAM domains , two from previously identified Eph-binding proteins ( SHIP2 and Odin ) and one from a new Eph-binding protein discovered in this study ( SAMD5 ) , to the SAM domains from every Eph receptor . Our study reveals that the Eph SAM domains have exquisitely specific target SAM domain-binding properties , although their overall SAM-SAM heterodimer formation modes are very similar . The binding affinities for all the interactions reported here are in the submicromolar to micromolar range . In living cells , their binding avidities are likely to be further enhanced due to the following two reasons . First , Eph receptors are known to multimerize upon ligand activation , mainly via their extracellular domains ( Janes et al . , 2012; Seiradake et al . , 2013 ) . The multimerization of the extracellular domains can bring a number of Eph cytoplasmic tails in close proximity . Second , several studies have revealed that effectors like SHIP2 can be recruited to the cell membrane via SH2 domain mediated interactions to other cell surface receptors ( Pesesse et al . , 2001; Wang et al . , 2004 ) . These two mechanisms can increase the local concentrations of Ephs and their downstream targets , and thus enhance their binding avidities . This result provides a partial answer in rationalizing diverse functions of Eph receptors that can be induced by the same or similar ephrin ligand ( s ) . It is anticipated that there exist additional SAM domain-containing proteins capable of binding to Eph receptors . Since there are 110 SAM domain-containing proteins in addition to the 14 Eph receptors and three Eph-binding proteins studied here , one approach to identify new Eph SAM-binding proteins is to survey bindings of all 14 Eph SAM domains to the rest of the SAM domains in the human proteome using purified recombinant proteins by protein array-based methods . We demonstrated that , unlike many other previously characterized polymer forming SAM domains , every Eph receptor SAM domain adopts stable monomers in solution ( Harada et al . , 2008; Knight et al . , 2011; Stapleton et al . , 1999; Thanos et al . , 1999; Wang et al . , 2016 ) . Additionally , no SAM–SAM hetero-complexes between Eph SAM domains could form ( data not shown ) . The structures solved in this study revealed that the SAM domains from EphA2 , EphA5 , and EphA6 all bind to its effector SAM domain using their respective End-Helix . It is noticed that every Eph SAM domain also contains an extremely conserved Mid-Loop ( see EphA2/EphA5/EphA6 SAM sequence alignments in Figure 2—figure supplement 5B–D for an example ) . It is compelling to speculate that at least some Eph SAM domains may also use their Mid-Loops to bind to the End-Helix of their effector SAM domains . This hypothesis is further supported by the observation that many residues in the Mid-Loop regions of Eph SAM domains are found to be mutated in cancer patients ( Figure 6A ) . The structures of Eph SAM domains in complex with the SAM domains of several different effectors presented in this study are very useful in interpreting numerous mutations/variants found in the Eph SAM domains in patients with different diseases ( e . g . cancers analyzed in Figure 6 ) . It is expected that not all mutations occur in the SAM domains will alter functions of Eph receptors . The structures of the SAM–SAM complexes presented in this work , coupled with amino acid sequence-based analysis , readily predict the following three categories of mutations that will likely alter functions of Eph receptors via changing their SAM domain structures and effector binding . First , mutations of residues in the folding core or residues playing other critical structural roles ( e . g . residues highlighted in red in Figure 6A ) may impair the overall structure and thus effector binding of Eph SAM domains . We tested a few of such mutations ( e . g . EphA2 W913C and D944N ) , and found that these SAM mutants invariably expressed as inclusion bodies ( data not shown ) . Second , mutations occur in the End-Helix region of Eph SAM domains . This category of Eph SAM domain mutants often has defects or even total impairments in binding to their effectors ( Figure 6B and C ) , and thus are expected to have impaired downstream signaling . Third , mutations occur in the Mid-Loop region of the Eph SAM domains may also impair their binding to effector SAM domain as we mentioned above , although this prediction will need to be verified . We propose that the Eph SAM mutations found in patients that fall into the above three categories will have higher chance to be disease-relevant mutations and thus are given higher priority to be investigated . Certainly , it cannot be ruled out that mutations of the Eph SAM domains that are outside the three categories may also impair functions of Eph receptors , perhaps via still unknown SAM domain-mediated target bindings .
DNA encoding EphA2 SAM ( NP_034269 . 2 , residues 901–977 ) , SHIP2 SAM ( NP_034697 . 2 , residues 1201–1257 ) , Odin SAM1 ( NP_852078 . 1 , residues 712–776 ) and SAMD5 SAM ( NP_796245 . 2 , residues 1–66 ) were amplified from Mus musculus cDNA libraries as the template and individually cloned into a modified pET vector ( Liu et al . , 2011 ) . All mutants were created using the standard two-step PCR methods . The fusion constructs of SHIP2 SAM/EphA2 SAM and Odin SAM1/EphA6 SAM each contained a TEV protease recognition site ‘ENLYFQ’ flanked by several Gly-Ser repeats as the flexible linkers . Recombinant proteins with N-terminal His6-tag were expressed E . coli BL21 ( DE3 ) strain . Expressed proteins were purified by the Ni2+-NTA Sepharose 6 Fast Flow beads ( GE Healthcare , China ) affinity chromatography followed by a Superdex-200 prep grade size-exclusion chromatography ( GE Healthcare ) in a buffer containing 50 mM Tris , pH 7 . 5 , 100 mM NaCl , 1 mM DTT , and 1 mM EDTA . The N-terminal His-tag of each protein was cleaved by 3C protease and removed by another step of size-exclusion chromatography . Analytical gel-filtration chromatography was performed on an AKTA system ( GE Healthcare ) using the Superose 12 10/300 GL column . Protein samples ( each with 100 µL at 50 µM ) were injected into the column pre-equilibrated with a buffer containing 50 mM Tris , pH 7 . 5 , 100 mM NaCl , 1 mM DTT , and 1 mM EDTA . Isothermal titration calorimetry experiments were carried out on the MicroCal ITC200 calorimeter ( Malvern , UK ) at 25°C . The concentration of the injected samples in the syringe was 500 μM , and the concentration of the samples in the cell was fixed at 50 μM . The sample in the syringe was sequentially injected into the sample cell with a time interval of 150 s ( 0 . 5 μL for the first injection and 2 μL each for the following 19 injections ) . The titration data were analyzed by the Origin 7 . 0 software and fitted with the one-site binding model . The SHIP2 SAM-EphA2 SAM fusion protein ( with a 14-residue linker ‘SSGENLYFQSGSSG’ ) , the Odin SAM-EphA6 SAM fusion protein ( with a 17-residue linker ‘PSGSSGENLYFQSGSSG’ ) , and a 1:1 mixture of SAMD5 SAM and EphA5 SAM were concentrated to ~10–20 mg/mL for crystallization . Crystals were obtained at 16°C by the sitting-drop vapor diffusion against 80 μL well solution using 48-well format crystallization plates . SHIP2 SAM-EphA2 SAM complex crystals were grown in a buffer containing 0 . 2 M Succinic acid pH 7 . 0 , 20% w/v Polyethylene glycol 3350 . Odin SAM1-EphA6 complex crystals were grown in a buffer containing 0 . 1 M HEPES sodium pH 7 . 5 , 1 . 5 M lithium sulfate monohydrate . SAMD5 SAM-EphA5 SAM complex crystals were grown in a buffer containing 0 . 2 M ammonium acetate , 0 . 1 M BIS-TRIS pH 5 . 5 , 25% w/v Polyethylene glycol 3350 . Crystals were soaked in the crystallization solution containing additional 5% ( for SHIP2/EphA2 and SAMD5/EphA5 ) or 20% ( for Odin/EphA6 ) v/v glycerol for cryo-protection . Diffraction data were collected at the Shanghai Synchrotron Radiation Facility BL17U1 at 100 K . Data were processed and scaled using HKL2000 ( Otwinowski and Minor , 1997 ) . Structures were all solved by molecular replacement with the EphA2 SAM domain ( PDB: 3KKA ) and the SHIP2 SAM domain ( PDB: 2K4P ) or the Odin first SAM domain ( PDB: 2LMR ) structures as the searching models using PHASER ( McCoy et al . , 2007 ) . Further manual model buildings and refinements were completed iteratively using Coot ( Emsley et al . , 2010 ) and PHENIX ( Adams et al . , 2010 ) or Refmac5 ( Murshudov et al . , 2011 ) . The final models were validated by MolProbity ( Chen et al . , 2010 ) . The final refinement statistics are summarized in Table 1 . The structure figures were prepared by PyMOL ( http://www . pymol . org ) . The structure factors and the coordinates of the structures reported in this work have been deposited to PDB under the accession codes of 5ZRX , 5ZRY and 5ZRZ for the EphA2/SHIP2 , EphA6/Odin and EphA5/SAMD5 complex structures , respectively . GST-tagged EphA2 , EphA5 , SHIP2 , SAMD5 SAM domains and their mutant proteins or GST alone were incubated with HEK293T cell lysates expressing the GFP-tagged target proteins for 1 hr at 4°C . The mixture was then loaded onto 20 μL Glutathione Sepharose 4B beads ( GE Healthcare ) in PBS buffer for 0 . 5 hr at 4°C . After washing twice , the proteins captured by the beads were eluted by boiling with SDS-PAGE loading buffer , resolved by SDS-PAGE and detected by an anti-GFP antibody using western blotting . Both DU145 and HEK293T cells were cultured in Dulbecco’s Modified Eagle Medium ( DMEM ) supplemented with 10% fetal bovine serum ( FBS ) , and 1% of penicillin-streptomycin at 37°C with 5% CO2 . HEK293T cells were transfected with GFP-tagged full-length wild type EphA2 , EphA2-R958K , EphA2-R958C and GFP alone using ViaFect Transfection Reagent ( Promega , Madison , WI ) as per manufacturer’s protocol . The full-length wild type EphA2 , EphA2-R958K , EphA2 delSAM , EphA2-SAMA5 chimera and GFP control , each with a C-terminal 3 × Flag tag , were individually cloned into the pLVX plasmid for commercial viral packaging ( Shanghai Taitool Bioscience Co . Ltd , China ) . All infectious lenti-viruses were produced in HEK293T cells by co-transfected with the VSV-G glycoprotein expressing plasmid pMD2 . G and pCMVΔR8 . 91 packaging construct . The lenti-viruses were used to infect DU145 cells and selected with 1 μg/mL puromycin for ten days . For experiments involving SHIP2-SAM , DU145 cells infected with lenti-virus expressing GFP control or full-length wild type EphA2 were further transfected with Myc-tagged SHIP2-SAM using ViaFect Transfection Reagent ( Promega ) . The expression levels of proteins were verified by immunoblotting . These cells were not individually authenticated and not found to be on the list of commonly misidentified cell lines ( International Cell Line Authentication Committee ) . Cells were tested negative for mycoplasma contamination by cytoplasmic DAPI staining . Cells were preserved to passage into a chamber ( ibid μ-Slide VI0 . 4 , Germany ) , starved for 4 hr in serum-free medium , and then stimulated with ephrinA1-Fc ( R and D Systems , 1 μg/mL in PBS ) for 30 min at 37°C . The cells were then fixed for 15 min with 4% paraformaldehyde in PBS and stained with rhodamine-conjugated phalloidin for 30 min . Cell morphologies were analyzed with a Zeiss Confocal microscope ( LSM710 ) or an Olympus fluorescence microscope ( BX61 ) equipped with a digital camera with a 20 × objective lens . The cell areas were measured using the ImageJ software ( https://imagej . nih . gov/ij/ ) . Two-way ANOVA with multiple comparisons test was used to compare cell areas among different experimental groups , and presented with the GraphPad Prism software . | As an animal’s body develops , its cells need to find their way to the right place to form its tissues and organs . On top of this , nerve cells need to set up connections as they grow . A family of receptors called Eph receptors help to make this happen . They sit across cell membranes , waiting for signals from molecules called ephrins . Once activated , these receptors interact with other proteins inside the cell . There are 14 different Eph receptors , but the parts inside the cell are similar , with three domains arranged in a set order . Next to the membrane , there is a tyrosine kinase domain , an enzyme that can add a phosphate group to a protein . Then , there is a SAM domain , which interacts with other proteins . Finally , there is a PDZ domain binding motif , which anchors the receptor to the cell's internal skeleton . The similarity between the internal portions of the Eph receptors suggests that they should work in the same way . But , different receptors on the same cell , responding to the same external signal , can have opposite effects . Here , Wang et al . tested each of the 14 SAM domains to find out how this happens . SAM domains on Eph receptors interact with SAM domains on other proteins , including SHIP2 and Odin . Analysis of the interactions revealed specific patterns for each receptor . Even though SAM domains are similar in shape , their exact amino acids – the basic building blocks of proteins – differ at particular positions . This changes the way they interact , allowing them to bind to different partners . Wang et al . then used a technique called X-ray crystallography to reveal the three-dimensional structures of SHIP2 bound to EphA2 and Odin bound to EphA6 , to see how the proteins interact in fine detail . It turns out that a piece of each Eph receptor called the “end helix” binds to a “mid-loop” structure in SHIP2 or Odin . Crucial amino acids in each ensure that these interactions are specific . Changing these critical positions prevented the proteins coming together or allowed them to bind to a completely different partner . The structures revealed the importance of negatively charged amino acids within the mid-loop of the Eph binding partners . Using this information , Wang et al . predicted and confirmed a brand-new interaction between EphA5 and one of the 127 SAM-containing proteins found in mice , a protein called SAMD5 . Understanding the impact of protein structure on Eph receptors could aid research into human disease . Lastly , an analysis of a database containing genetic changes found in cancer patients revealed that many of the mutations occur inside SAM domains . Pinpointing the positions that affect Eph receptor binding could point the way to future treatments . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2018 | Specific Eph receptor-cytoplasmic effector signaling mediated by SAM–SAM domain interactions |
Porcine reproductive and respiratory syndrome virus ( PRRSV ) and transmissible gastroenteritis virus ( TGEV ) are two highly infectious and lethal viruses causing major economic losses to pig production . Here , we report generation of double-gene-knockout ( DKO ) pigs harboring edited knockout alleles for known receptor proteins CD163 and pAPN and show that DKO pigs are completely resistant to genotype 2 PRRSV and TGEV . We found no differences in meat-production or reproductive-performance traits between wild-type and DKO pigs , but detected increased iron in DKO muscle . Additional infection challenge experiments showed that DKO pigs exhibited decreased susceptibility to porcine deltacoronavirus ( PDCoV ) , thus offering unprecedented in vivo evidence of pAPN as one of PDCoV receptors . Beyond showing that multiple gene edits can be combined in a livestock animal to achieve simultaneous resistance to two major viruses , our study introduces a valuable model for investigating infection mechanisms of porcine pathogenic viruses that exploit pAPN or CD163 for entry .
Porcine reproductive and respiratory syndrome ( PRRS ) is a highly infectious viral disease characterized by reproductive disorders including premature birth , late abortion , stillbirth , weak and mummy fetuses , and respiratory dysfunction in piglets and in growing pigs ( Wensvoort et al . , 1991 ) . Since its discovery in the United States in 1987 , PRRS has rapidly spread worldwide , with frequent outbreaks causing large economic losses ( Holtkamp et al . , 2013 ) . Three surface receptors on porcine alveolar macrophages ( PAMs ) have been shown to function in PRRSV invasion in vivo: heparin sulphate ( HS ) , sialoadhesin ( Sn ) , and CD163 ( Calvert et al . , 2007; Crocker and Gordon , 1986; Jusa et al . , 1997 ) . Multiple studies have reported that CD163 is an essential receptor for PRRSV infection , with scavenger receptor cysteine-rich domain 5 ( SRCR5 ) serving as the core domain for virus recognition ( Calvert et al . , 2007; Van Gorp et al . , 2010; Patton et al . , 2009 ) . Gene editing technology has been emerging as an important approach of livestock animal and plant germplasm improvement . The technology makes possible for precise modification of more than one gene simultaneously , which is particularly desirable for obtaining important economic traits that are controlled by multiple genes . In 2016 , Prather's group was the first to use CRISPR/Cas9 technology to generate SRCR5 domain-targeted CD163 knockout pigs . They demonstrated that a CD163 knockout line was completely resistant to genotype 2 PRRSV infection ( Whitworth et al . , 2016 ) . Subsequently , several laboratories have generated anti-PRRSV pigs targeting CD163 . For example , the CD163 SRCR5 domain was replaced with human CD163-Like SRCR8 domain to generate PRRSV genotype 1 resistance ( Wells et al . , 2017 ) . Wei et al . , 2018 reported homozygous gene-edited Large White pigs with a 50 bp deletion in exon 7 of the CD163 gene ( Wei et al . , 2018 ) that are fully resistant to genotype 2 PRRSV . There are also examples of deletion of the SRCR5 domain seeking resistance to both PRRSV genotypes ( Burkard et al . , 2017 ) , or introducing a premature termination in the CD163 SRCR5 domain to generate HP-PRRSV ( highly pathogenic PRRSV ) -resistant Duroc pigs ( Yang et al . , 2018 ) . Deleting the SRCR5 LBP region has also been reported to generate a PRRSV genotype 2 resistant pigs ( Guo et al . , 2019 ) . All these studies demonstrate that PRRSV-resistant pig breeds can be generated by editing the CD163 gene , enabling alleviation of the severity of PRRSV . In addition to PRRSV , transmissible gastroenteritis virus ( TGEV ) , an acute high-contact infectious virus , is known to frequently occur to co-infect with other porcine diarrhea-associated viruses such as porcine epidemic diarrhea virus ( PEDV ) , porcine rotavirus ( PoRV ) ( Zhang et al . , 2013 ) . TGEV is globally distributed and causes tremendous economic losses in pork production ( Gerdts and Zakhartchouk , 2017 ) . Characterized by vomiting , severe diarrhea , and dehydration , the mortality rate of TGEV-infected piglets under the age of 14 days approaches 100% . TGEV is a single-stranded , positive-sense RNA coronavirus which targets pig intestinal epithelium for infection ( Brierley et al . , 1989; Wesley and Lager , 2003 ) . Studies have shown that the pAPN protein acts as a receptor in mediating TGEV infection . The viral glycoproteins bind to pAPN receptors on the surface of small intestinal epithelial cells and mediate membrane fusion , thus resulting in the virus entering into epithelial cells ( Delmas et al . , 1992; Hansen et al . , 1998 ) . Inhibition or direct knockout of pAPN in small intestinal epithelial cells can mitigate TGEV infection ( Ji et al . , 2018; Zhu et al . , 2018 ) . pAPN knockout pigs are resistant to TGEV ( Luo et al . , 2019; Whitworth et al . , 2019 ) . PDCoV is a highly virulent porcine coronavirus discovered in 2012 that causes watery diarrhea and vomiting in sows and piglets , with piglet mortality rates of 30% to 40% ( Wang et al . , 2014; Woo et al . , 2012 ) . There is controversy about whether or not pAPN is a functional receptor for PDCoV . Wang et al . , 2018 showed that pAPN functions as a receptor to promote PDCoV entry into cells ( Wang et al . , 2018 ) , while Zhu et al . , 2018 confirmed its involvement but showed that pAPN was an unnecessary important functional receptor for PDCoV infection ( Zhu et al . , 2018 ) . Li et al . , 2018 suggested that PDCoV infection may require a co-receptor , in addition to pAPN ( Li et al . , 2018 ) . Using cells isolated from pAPN knockout pigs , however , Stoian et al . , 2020 showed that these pig cells were still susceptible to PDCoV infection in vitro . It was suggested that pAPN may be one of the receptors for PDCoV , and an unknown receptor or factor may compensate for pAPN function in the absence of pAPN ( Stoian et al . , 2020 ) . However , whether pAPN knockout pigs may be resistant to PDCoV infection in vivo remains unknown . Although gene-edited CD163 knockout ( PRRV resistant ) pigs and pAPN knockout ( TGEV resistant ) pigs have been previously generated , respectively , pigs that are resistant to the infection of both viruses are lacking . Our objectives in the present study were ( 1 ) to knockout CD163 and pAPN simultaneously using a gene editing approach; ( 2 ) to verify if the resultant DKO pigs are simultaneously resistant to infection by PRRSV and TGEV; ( 3 ) to use the DKO pigs as an in vivo experimental model to test for potential pAPN-mediated resistance to PDCoV infection . We report successfully generated gene-edited Large White pigs with both CD163 and pAPN gene knockouts using CRISPR/Cas9 and somatic cell nuclear transfer ( SCNT ) . Through viral challenge experiments , we found that these DKO pigs exhibit complete resistance to genotype 2 PRRSV and TGEV , and exhibit decreased susceptibility to PDCoV infection . In addition , with the exception of meat color score and iron content , no differences in the production performance , reproductive performance , or pork nutrient content were observed between DKO pigs and WT pigs . Thus , in addition to demonstrating that our DKO pigs are robustly resistant to both PRRSV and TGEV without suffering deleterious effects for production performance , our study also provides insights into ongoing controversy about the pAPN protein as a potential receptor for PDCoV infection of pigs .
In order to generate CD163 and pAPN DKO cloned pigs , we constructed sgRNA delivery plasmids targeting these genes , and selected successful DKO pig fetal fibroblasts ( PEFs ) as nuclear transfer donors ( Figure 1A ) . For CD163 , the SRCR5 domain-binding site for PRRSV in exon 7 ( Van Gorp et al . , 2010; Ma et al . , 2017 ) was selected as the sgRNA recognition site . To inactivate the pAPN protein , a sgRNA target site in exon two immediately downstream of the ATG start codon was selected ( Figure 1B ) . Successful DKO colonies were cultured as donor cells for SCNT ( Supplementary file 1 ) . The cloned pigs generated in this experiment were obtained via both primary and secondary clonings . For primary cloning , the selected DKO cells are used as donors for nuclear transplantation . For secondary cloning , the ear-derived fibroblasts of the primary cloned pigs are re-cloned , which rapidly provided a large number of high-quality DKO donor cells , thus improving cloning efficiency and resulting in many genotypically identical pigs . In our primary cloning , a total of 3780 reconstructed embryos were transplanted into 11 surrogate sows , of which two were pregnant and gave birth to eight live piglets . Of these piglets , four survived after weaning ( Figure 1C and Supplementary file 2 ) . We determined the CD163 and pAPN genotypes of the four surviving piglets using PCR and Sanger sequencing . The genotypes of the three piglets ( #1143 , #1144 , and #1145 ) matched that of cell colony #25 , which had an 8 bp deletion on both copies of CD163 near the target site , and a copy of pAPN carrying a 5 bp deletion on one copy and a 26 bp deletion on the other , both resulting in frameshift mutations or premature termination after the target site ( Figure 1D ) . In order to generate more DKO pigs for viral challenge experiments , we collected ear tissue samples from three piglets ( #1143 , #1144 , and #1145 ) and isolated ear-derived fibroblasts . A total of 2270 reconstructed embryos generated from ear-derived fibroblasts of #1145 were transplanted into nine surrogates . Four sows successful gave birth to a total of 20 live piglets , among which 12 survived post-weaning ( Supplementary file 2 ) . The genotypes of these 12 piglets matched that of #1145 , and the three DKO primary clones used for subsequent experiments . We used flow cytometry and western blotting for CD163 , immunohistochemistry ( IHC ) and western blotting for pAPN , and confirmed that expression of both proteins was undetectable in DKO pigs but detectable in WT pigs of the same age and breed ( Figure 1E–G ) . We designed multiple pairs of amplification primers for the pX330 vector backbone to confirm that no random integration of pX330 vector fragments were in cloned pigs ( Figure 1—figure supplement 1 ) . We also tested for off-target modifications in DKO pigs using 10 potential off-target sites for each of the two sgRNAs and found no alteration in any of these 20 predicted sites in the cloned pigs ( Supplementary file 3 ) . This data demonstrates that clones of Sus scrofa line with multiple gene-edited can be generated through primary and secondary cloning with high efficiency and no off-target detected . For testing of PRRSV resistance in PAMs derived from DKO pigs , we selected the highly pathogenic genotype 2 PRRSV strain WUH3 to challenge DKO and WT PAMs at a multiplicity of infection ( MOI ) of 0 . 1 . qRT-PCR and western blot analyses were used to assess PRRSV proliferation in PAMs . At 12 hr post-infection ( hpi ) , DKO PAMs carried a significantly lower PRRSV load compared with WT PAMs , and no viral RNA or PRRSV-N protein was detected thereafter in DKO PAMs ( Figure 2A and B ) . The low level of PRRSV RNA that was initially detectable in the DKO line at 12 hpi is likely attributable to the adsorbed PRRSV independent of the existence of CD163 , as CD163 is thought to be primarily responsible for the uncoating and viral RNA release processes of PRRSV infection ( Chen et al . , 2019; Van Gorp et al . , 2008 ) . We next sought to examine if DKO pigs are resistant to PRRSV in vivo . Four 45-day-old DKO pigs and six WT control pigs of the same age were challenged with the PRRSV strain WUH3 . Nasal intubation drip ( 2 mL: 106 TCID50/mL ) and intramuscular injection ( 2 mL: 106 TCID50/mL ) were used to infect both experimental groups . The phenotypic data of body temperature , feed intake , respiration , defecation , and mental condition were recorded daily after infection . As shown in Figure 2C , while fever ( over 40°C ) began at 1 day post-infection ( dpi ) and persisted throughout the remainder of the experimental period in the WT group , the body temperature of the DKO pigs stayed normal throughout the 14 days of the post-viral challenge observation period . Scoring for other clinical symptoms of PRRSV at 1 dpi showed that WT pigs exhibited decreased appetite , shortness of breath , cough , malaise , drowsiness , and difficulty walking , whereas the DKO group displayed no abnormalities except for a brief cough and diarrhea in two pigs at 4 dpi and 9 dpi , respectively ( Figure 2D ) . The body weight of the DKO pigs increased , throughout the 14 day post viral challenge observation period: the detected body weights of the WT pigs were all lower than DKO pigs after 0 dpi ( Figure 2E ) . Of the six challenged WT pigs , one was slaughtered at 10 dpi to harvest PAMs , and the five remaining WT pigs died within 11 dpi . In sharp contrast , all four pigs in the DKO group remained healthy , and survived for the entire duration of the 14-day experiment ( Figure 2F ) . Among the dead and slaughtered WT pigs , the lungs were swollen , with severe bleeding , and obvious lesions , while the lung tissues of dissected DKO pigs did not exhibit lesions or any other distinct symptoms associated with PRRSV ( Figure 2G ) . Hematoxylin and eosin ( H and E ) staining showed thickening of the alveolar walls and infiltration of a large number of inflammatory cells in the pulmonary interstitium of the WT pig lungs , while no pathological changes were found in the lung tissue of DKO pigs ( Figure 2H , upper panel ) . Examination of PRRSV antigens in lung tissue via IHC , it was revealed that the viral antigens were present in the lungs of the WT group , but not that of the DKO pigs ( Figure 2H , lower panel ) . Moreover , we measured the PRRSV viral load in the serum of both groups at 0 , 3 , 7 , 10 , and 14 dpi and found that in the WT group , the PRRSV load increased rapidly and significantly by 7 dpi , reaching its maximum at 7 dpi . In agreement with other experiments showing viral resistance , the PRRSV viremia in the DKO group remained negative throughout the challenge ( Figure 2I ) . We also tested the PRRSV viral load in PAMs , lung tissues , and tonsil tissues of the two groups of pigs after viral challenge . Whereas a high titer of PRRSV was detected in all tissues examined in the WT group , PRRSV was almost undetectable in DKO pigs ( Figure 2J ) . From 3 dpi , the amount of PRRSV-specific ELISA antibodies in the serum of WT pigs increased significantly , and antibody levels were positive ( S/P≥0 . 4 ) at 7 and 10 dpi , while such antibodies in DKO pigs remained consistently negative ( S/P<0 . 4 ) ( Figure 2K ) . Taken together , these results provide compelling in vitro and in vivo evidence that the DKO pigs are resistant to PRRSV infection . Following characterization of PRRSV resistance , we next sought to determine if double knockout of CD163 and pAPN also conferred resistance against TGEV . Four 45-day-old DKO pigs and six WT control pigs of the same age and breed were fed under the same conditions and infected with TGEV . A total of 10 mL of TGEV ( 7 × 105 TCID50/mL ) were orally administered to each pig in two doses ( day 0 and day 1 , 5 mL/day ) . At 3 dpi , one DKO pig and one WT pig were slaughtered to collect intestinal tissues for pathological examination , and the remaining pigs were housed under regular husbandry conditions until slaughter , and tissues were sampled at 14 dpi . Body temperature was recorded daily beginning at Day 0 , prior to inoculation , and piglet weighing and blood sampling for serum separation were conducted at 0 , 7 , and 14 dpi . During the viral challenge period , no abnormalities were observed among the pigs , with the exception of two WT pigs that had diarrhea . There was no significant difference in weight gain between the two groups ( data not shown ) . Detection of TGEV-specific neutralizing antibodies in serum showed no neutralizing antibodies in the DKO pigs throughout the experiment , while two of the WT pigs were positive for neutralizing antibodies at 7 dpi , and all WT pigs were positive by 14 dpi ( Figure 3A ) . All slaughtered pigs from both WT and DKO groups ( sampled at 3 dpi and 14 dpi ) were dissected to examine potential lesions in small intestine tissues . For the DKO group , no lesions were found in the small intestine samples collected at either 3 dpi or 14 dpi ( Figure 3B ) . In marked contrast , WT group tissues collected at 3 dpi demonstrated a thin and yellowing small intestine wall , with hemorrhages typical of TGEV clinical symptoms , and by 14 dpi there were notable duodenum , jejunum , and ileum hemorrhages , accompanied by intestinal wall thinning and enlarged mesenteric lymph nodes ( Figure 3B ) . Pathological examination of small intestine tissue sections revealed pathological changes , including necrosis and shedding of intestinal mucosal epithelial cells , intestinal villi fusion , plasma cells accumulating in the lamina propria , and infiltration of eosinophils in the duodenum , jejunum , and ileum of WT pigs at 3 dpi and 14 dpi , while the same small intestine tissues in DKO pigs appeared healthy ( Figure 3C ) . We also analyzed the ratio of intestinal villus height ( VH ) to the crypt depth ( CD ) . The smaller the ratio , the more severe the intestinal villi atrophy . We found that compared with the mock group , the three intestinal segments of the WT group had significant intestinal villous atrophy , and the intestinal villi of these intestinal segments in the DKO group did not show atrophy; that is , the degree of intestinal villous atrophy in the three intestine segments in the WT group was significantly higher than that in the DKO group ( Figure 3D ) . These results consistently demonstrate that our CD163/pAPN DKO pigs exhibit strong resistance to TGEV infection . PDCoV is a highly pathogenic virus that has recently been shown to cause diarrhea in newborn piglets , although the functional receptors for PDCoV have not yet been confirmed ( Li et al . , 2018; Stoian et al . , 2020; Zhu et al . , 2018 ) . Whether pAPN functions as a receptor or co-receptor in PDCoV infection of pigs remains controversial . To test the hypothesis that pAPN may functionally mediate PDCoV infection , we tested the susceptibility of our DKO pigs to this virus . Two 45-day-old DKO pigs and four WT pigs of the same age and breed were challenged with PDCoV . A total of 16 mL of PDCoV ( 2 . 5 × 108 TCID50/mL ) was orally administered to each pig in two doses ( Day 0 and Day 1 , 8 mL/day ) . During the 14 days of PDCoV challenge study , both the DKO and WT pigs appeared normal , with no distinct differences in body temperature or weight ( data not shown ) . Blood was collected at 0 , 7 , and 14 dpi to assay for levels of virus-specific antibodies . At 7 and 14 dpi , WT pigs were all antibody-positive , while the DKO pigs were all antibody-negative at 7 dpi , but carried antibody levels comparable to that of the WT group by 14 dpi ( Figure 4A ) . This suggests that the double-gene knockout led to a delayed onset of humoral immunity in pigs , possibly due to delayed-immune system exposure to the virus . All pigs were slaughtered at 14 dpi , and the small intestine tissues were collected to evaluate disease severity . It was found that the intestinal wall of the WT had become thinner , with watery fluid in the small intestine , and mesenteric hyperemia , none of which was observed in the small intestine of the DKO pigs ( Figure 4B ) . Pathological examination of small intestine tissue sections revealed significant lesions in the small intestine tissues of both of the WT and DKO groups , which included intestinal villi fusion , infiltration of lymphocytes in the intestinal mucosa , with many lesions in the intrinsic membrane in the duodenum and jejunum tissues . In the ileum , there were signs of necrosis and shedding of intestinal mucosal intraepithelial cells and naked lamina propria . The extent of lesions in the WT pigs was more severe than that of the DKO pigs ( Figure 4C ) . We also detected the ratio of intestinal villus height to the crypt depth , and found that compared with the mock group , the three intestinal segments of both of the WT group and the DKO group had intestinal villous atrophy , but the degree of villous atrophy in the ileal tissue in the DKO group was lower than that of the WT group ( Figure 4D ) . In addition , we tested the resistance of PAMs derived from DKO pigs to PDCoV . DKO and WT PAMs were infected with PDCoV , and indirect immunofluorescence assays ( IFA ) , tissue culture infectious dose 50 ( TCID50 ) assays , qRT-PCR , and western blot analyses to assess PDCoV proliferation in PAMs all indicated that DKO PAMs exhibit significantly decreased susceptibility of PDCoV infection compared to WT PAMs ( Figure 4—figure supplement 1 ) . These data suggest that although the DKO line is still susceptible to PDCoV infection , the viral invasion and damage to the small intestines was partially inhibited compared to that of the WT line . We next evaluated the growth and performance indices of DKO pigs . Three 11-month-old DKO Large White boars and three WT Large White boars of the same age were selected for slaughter testing . The live weight at slaughter , carcass weight and length , dressing percentage , ham percentage , lean rate , loin eye area , average backfat thickness , muscle pH , marbling , and drip loss were determined . As shown in Table 1 , with the exception of meat coloring score , DKO pigs showed no difference in comparison with WT pigs for these indices . In addition , there was no significant difference in birth weight or in the average daily gain between WT and DKO pigs ( Supplementary file 8 ) . Most notably , the meat color score in the DKO pigs ( 4 . 667 ± 0 . 1667 N=3 ) was significantly higher than that of WT pigs ( 3 . 833 ± 0 . 1667 N=3 ) , although both were within the normal range of 2 to 5 according to the guideline of ‘rules for performance testing of breeding pigs’ document published by the Ministry of Agriculture and Rural Affairs of PR China ( NY/T 821–2004 ) ( Table 1 and Figure 5A–B ) . Since the CD163 protein is known to play a role in the degradation of haemoglobin-haptoglobin ( Hb-Hp ) , and considering that Fe is an important component of haemoglobin , we reasoned that the increased meat color score ( redness ) may be due to the decreased Hb metabolism as a consequence of CD163 knockout , and subsequently mild accumulation of Fe containing Hb in the meat . To test this hypothesis , the meat Fe level was analyzed , it was found that the concentration of Fe was significantly higher in DKO pigs compared to WT pigs ( Figure 5C ) . We also tested the serum haptoglobin ( Hp ) content and found that the Hp content in DKO pigs was significantly higher than that of WT control pigs ( Figure 5D ) . Evaluation of the nutritional components of pork such as total protein , total fat , ash , moisture , specific minerals , and amino acid content was also performed . As shown in Table 2 and Supplementary file 4 , no differences in these indices were observed between the two groups . In order to test the reproductive performance of the DKO boars , semen from DKO male pigs ( n = 3 ) and that of WT pigs ( n = 4 ) of the same age and breed were analyzed . It was revealed that the concentration , motility , and velocity distribution of the sperm from DKO boars did not differ from WT boars ( Table 3 ) . Furthermore , there was no difference in the litter size between the two genotypes: DKO litters were 10 . 67 ± 1 . 202 ( N = 3 , litter size from 9 to 13 ) and the WT litters were 12 . 05 ± 0 . 6496 ( N = 22 , litter size from 7 to 17 ) . In addition , these three DKO pigs did not show any growth abnormalities or disease phenomena during the 11-month rearing process , and no abnormalities were observed in the main tissues and organs after slaughter ( data not shown ) . Taken together , with the exception of slight meat coloring score increase , these results show that the simultaneous , editing-based disruption of the CD163 and pAPN loci , does not affect the normal growth and reproductive performance of the resultant DKO pigs .
Conventional breeding for complex traits using molecular marker-assisted selection is a lengthy process , requiring multiple rounds of crosses and backcrosses to introgress each individual gene . CRISPR/Cas9 gene editing not only allows bypassing of this long process , but also provides a possible means to obtain multiple beneficial genotypes in a single generation while also avoiding gene penetration from donor species , thus maintaining the desirable qualities of the original species . Zhou et al . , 2015 first used CRISPR/Cas9 in combination with SCNT to generate knockout of PARK2 and PINK1 genes , whose dysfunction are known to contribute to the early onset of Parkinson’s disease in humans ( Zhou et al . , 2015 ) . Huang et al . , 2017 got the pig model with metabolic disorder successfully by editing apolipoprotein E and low density lipoprotein receptor genes simultaneously ( Huang et al . , 2017 ) . Our study is the first report on how multiple gene edits can be combined in livestock animal to offer simultaneous resistance to two major viral infection . Similar to the previous reports above , double knockout efficiency using CRISPR/Cas9-mediated dual gene editing method without any drug or flow cytometry screening was high in our study , reaching 6 . 30% ( 17 DKO cell colonies out of 270 cell colonies ) . In this experiment , we quickly generated a large number of DKO pigs by re-cloning . We found that the re-cloning efficiency ( 0 . 9% , 20/2270 ) was much higher than the primary cloning efficiency ( 0 . 2% , 8/3780 ) . A possible reason for this elevated efficiency could be that the monoclonal cells used for the primary cloning must be cultured in vitro for a long time , which has been reported to inhibit cloning efficiency ( Li et al . , 2003; Magnani et al . , 2008; Mastromonaco et al . , 2006 ) . The donor cells used in re-cloning were ear-derived fibroblasts isolated directly from DKO pigs , eliminating the requirement for a long-term , in vitro screening process . Our findings support the notion that the efficiency of this approach is not gene specific , and may be applicable to the knockout of other genes that allow improving disease resistance or animal production . In 2007 , Calvert et al . first discovered that CD163 functions as a PRRSV receptor protein during PAMs infection , which has since been confirmed by several studies ( Calvert et al . , 2007; Van Gorp et al . , 2008; Guo et al . , 2014; Patton et al . , 2009 ) . Structural studies of CD163 revealed that the SRCR5 domain corresponding to CD163 exon seven is necessary to mediate PRRSV infection ( Van Gorp et al . , 2010 ) . In recent years , several groups have successfully generated PRRSV-resistant gene-edited pigs by targeting exon 7 of the pig CD163 gene ( Burkard et al . , 2017; Guo et al . , 2019; Whitworth et al . , 2016; Yang et al . , 2018 ) . In the present study , we used a single sgRNA targeting exon 7 of CD163 , generated an 8 bp double-stranded deletion that terminated protein translation near the target site . Our finding on the complete resistance to PRRSV genotype 2 in our knockout line is consistent with those previous reports . CD163 is known to play a role in promoting the clearance of plasma free haemoglobin ( Kristiansen et al . , 2001 ) . Our finding that the DKO pigs have higher meat Fe content and have elevated serum Hp levels is consistent with this idea , and may explain the observed darker red color in our DKO meat . Interestingly , and consistent to our finding , Wells et al . , 2017 also reported that the serum Hp levels are elevated in CD163 knockout pigs ( Wells et al . , 2017 ) . Despite the slight color score increase , no abnormal growth or reproductive performance was observed in our DKO pigs , and the meat color of both DKO pigs and WT pigs were within the normal range . Production performance evaluations and identification of pork nutritional components showed that our DKO pigs were indistinguishable from that of the WT pigs in growth rate and reproductive performances , except for the meat color score and iron content . However , the number of DKO pigs tested by us is still small , and the production performance of DKO pigs still needs to be verified in large populations in the future . APN is known to be a receptor for many coronaviruses , and studies have shown that separate domains function in virus recognition vs . hydrolase catalytic activity ( Reguera et al . , 2012 ) . Two research groups have recently demonstrated that pAPN knockout pigs block TGEV but not PEDV infection ( Luo et al . , 2019; Whitworth et al . , 2019 ) . Our data showing that pAPN knockout can completely prevent TGEV virus infection are consistent with these recently published findings . In addition to TGEV and PRRSV , we also determined if pAPN deletion conferred protection against PDCoV . APN is a receptor for multiple coronaviruses and is abundantly expressed on small intestinal epithelial cells , which has led to the speculation that pAPN may also be a receptor for PDCoV . Wang et al . , 2018 and Li et al . , 2018 proposed that pAPN functions as a receptor in mediating PDCoV infection ( Li et al . , 2018; Wang et al . , 2018 ) . However , another study found that knockout of pAPN in IPI-2I cells inhibited but did not completely block PDCoV infection , suggesting that pAPN was not essential for viral recognition ( Zhu et al . , 2018 ) . Taken together , these studies suggest that pAPN may be involved in PDCoV infection , but PDCoV may also be able to enter cell through other pathway ( s ) . Our results on the delayed PDCoV-specific neutralizing antibodies production , and a reduced extent of gross and histopathological lesions on small intestine in DKO pigs compared to WT pigs are consistent with this previous suggestion that pAPN may play a role but is not the only path for PDCoV cell entry . Interestingly , a recent study showed that PAMs , but not lung fibroblast-like cells , from pAPN knockout pigs showed resistance to PDCoV infection ( Stoian et al . , 2020 ) , a finding consistent with our in vitro experiments showing that DKO PAMs exhibit decreased susceptibility to PDCoV infection . In addition , pAPN knockout pigs are susceptible to PDCoV when virus levels were detected using qRT-PCR , and virus neutralization activity was measured , although the extent of tissue lesions between the KO and WT groups was not compared ( Stoian et al . , 2020 ) . Our findings are in line with this study reporting that pAPN knockout pigs are still susceptible to PDCoV . However , as reflected by the delay in neutralizing antibody response , and much lighter intestine damage in the DKO pigs , the susceptibility of the pAPN knockout group to the virus is reduced compared that of the WT pigs , indicating the potential role of pAPN in mediating PDCoV infection . Additionally , the effect of CD163 knockout in the delayed adaptive immune response cannot be ignored . Despite the important role of CD163 in innate immunity , an inhibiting effect of soluble CD163 on the adaptive immune system has also been reported ( Frings et al . , 2002; O'Connell et al . , 2017 ) . It is thus possible that the delayed adaptive immune response we observed in PDCoV-infected DKO pigs may be associated with CD163 knockout-induced immunosuppression . In summary , the DKO pigs generated in this study are simultaneously resistant to PRRSV and TEGV , and exhibit decreased susceptibility to PDCoV , while maintaining the same growth and reproductive production traits when compared to WT animals . These pigs may offer breeding starting points for disease-resistant pig colony generation and will be a valuable model to help deepen our understanding of the role and mechanisms of these receptor proteins in the infection mechanisms of multiple viruses .
For the CD163 gene , the sgRNA was designed to target exon 7 , and for the pAPN gene , the sgRNA was designed to target exon 2 . The sequences of the two sgRNAs are as follows: GGAAACCCAGGCTGGTTGGAGGG ( CD163-sgRNA ) and GCATCCTCCTCGGCGTGGCGG ( pAPN-sgRNA ) . The PAM is indicated in bold font . The two sgRNA sequences were cloned into the pX330 vector ( Addgene plasmid # 42230 ) and named pX330-CD163 and pX330-pAPN , respectively . Two plasmids were extracted ( TIANGEN , DP117 ) in large quantities and used to transfect the fetal fibroblasts of Large White pigs . LLC-PK1 cells and ST cells were obtained from American Type Culture Collection ( LLC-PK1 cells: ATCC CL-101; ST cells: ATCC CRL-1746 ) . Authentication of the cell lines was performed by STR profiling and had a negative mycoplasma contamination testing status . The fetuses of Large White pigs at 35-day-old were used to isolate PEFs , which were then cultured in DMEM medium containing 20% FBS . When the cells grew to 80% confluence , approximately 106 cells were transfected with pX330-CD163 ( 2 . 5 ug ) and pX330-pAPN ( 2 . 5 ug ) plasmids . A Lonza 2B nuclear transfection system was used for transfection with Nucleofector program T-016 . The entire transfection process was performed according to the kit instructions ( Lonza , VPI-1002 ) . Cells were cultured for 48 hr after transfection and then seeded into 10 cm dishes at a density of 150 cells/dish . The culture medium was changed every 3 days , and cells were cultured for 10 days to form single-cell colonies . Single-cell colonies were transferred to 48-well plates for expansion culture . When cells in the 48-well plates reached confluence , 1/3 of the cells were taken for genotype identification , and the remaining cells continued to expand . Cells with genotypes identified as double-gene mutations were cultured and frozen for SCNT . The oocytes for SCNT were derived from a nearby slaughterhouse , and the nuclear donor cells were the DKO fibroblasts . The nuclear transfer donor cells were transferred into enucleated oocytes , and the reconstructed embryos were activated and cultured to develop into blastocysts . We then selected well-developed recombinant embryo clones to be surgically transferred into the oviduct of recipient gilts on the day after estrus was observed . After the embryo transfer , the technicians observed the estrus of the sow , and regularly checked the pregnancy by B-ultrasound . The CD163 and pAPN genotypes of colonies and piglets born after nuclear transfer were detected by PCR and Sanger sequencing . One third of the cells in the 48-well plate and the ear tissue were used to extract the genomic DNA . The primer pairs CD163-F/CD163-R and pAPN-F/pAPN-R were used to amplify the sequences near the sgRNA target sites in the CD163 and pAPN genes , respectively . The primer sequences are shown in Supplementary file 5 . The PCR products were genotyped by Sanger sequencing . Potential off-target sites were predicted using an online software: CRISPOR ( http://crispor . tefor . net/ ) . We identified the 10 potential off-target sites for each of the two sgRNAs . Twenty pairs of primers were designed to amplify the potential off-target sites from the genomic DNA isolated from the 3 DKO pigs ( 1143# , 1144# , 1145# ) . Sanger sequencing was performed to determine whether any mutations occurred . The primer sequences are shown in Supplementary file 7 . The total protein extracted from lung tissue , liver tissue , and spleen tissue of non-challenged WT pigs and DKO pigs was used to detect CD163 , and protein extracted from duodenal , jejunal , and ileal tissues were used to detect pAPN expression . Whole cell lysates of PRRSV-infected PAMs and PDCoV-infected PAMs were used to quantify the expression levels of PRRSV nucleocapsid ( N ) protein and PDCoV nucleocapsid ( N ) protein , respectively . The protein samples were separated by 8% or 12% SDS-PAGE and transferred to a polyvinylidene fluoride membrane ( Millipore ) . The membrane was blocked with 5% skim milk for 2 hr , and then incubated with primary antibody at 4°C overnight and secondary antibody at room temperature for 2 hr . Chemiluminescent signals were developed with SuperSignal West Pico PLUS Chemiluninescent Substrate ( Thermos Scientific ) and captured with a Tanon-520 ( Tanon ) . CD163 rabbit polyclonal antibody ( 16646–1-AP; Proteintech ) was used to detect porcine CD163 . APN polyclonal antibody ( A5662; ABclonal ) was used to detect pAPN , anti-PRRSV-N antibody ( made in our laboratory ) was used to detect PRRSV-N protein , anti-PDCoV-N antibody ( made in our laboratory ) was used to detect PDCoV-N protein , GAPDH rabbit antibody ( 3683; Cell Signaling ) or β-actin rabbit antibody ( AC026; Abclonal Technology ) was used to stain GAPDH or β-actin as a loading control . HRP-conjugated affinipure goat anti-rabbit IgG ( H+L ) ( SA00001-2; Proteintech ) and HRP-conjugated affinipure goat anti-mouse IgG ( H+L ) ( SA00001-1; Proteintech ) were used as the secondary antibody . PAMs were isolated from DKO piglets and WT piglets . The lungs were obtained from the euthanized piglets . The lung surfaces were rinsed with PBS , and PAMs were subsequently obtained by bronchoalveolar lavage with PRMI-1640 medium ( Gibco , USA ) . The collected lavage solution was dispensed into a 50 mL centrifuge tube , centrifuged at 300 g for 10 min , and the supernatant was discarded . PAMs were washed again with PRMI-1640 medium and then frozen in cryopreservation solution containing 90% FBS and 10% DMSO . For further in vitro infection experiments , PAMs were cultured in RPMI-1640 medium with 10% FBS and 1 × antibiotic antimycotic ( 15240062; Invitrogen ) at 37°C/5% CO2 , and then infected with a highly pathogenic PRRSV ( HP-RRSV ) strain WUH3 ( GenBank accession number HM853673 ) ( Li et al . , 2009 ) at a dose of MOI = 0 . 1 and PDCoV strain CHN-HN-2014 ( GenBank accession number KT336560 ) ( Dong et al . , 2016 ) at a dose of MOI = 10 . The production of progeny PRRSV was evaluated through western blot , and qRT-PCR assays , and the production of progeny PDCoV was evaluted through IFA , TCID50 , qRT-PCR and western blot assays . PAMs were fixed in 4% formaldehyde for 15 min at room temperature . The cells were then blocked with 2% BSA overnight at 4°C and incubated with mouse anti-pig CD163 mAbs ( MCA2311PE; Bio-Rad ) at 37°C for 1 hr in the dark . After washing with PBS three times , PAMs were resuspended in PBS and immediately analyzed using a BD FACSVerse flow cytometer ( BD Biosciences , CA ) and FlowJo software ( TreeStar , CA ) . All WT pigs used in the infection experiment were born from natural breeding , and they were matched by age and breed with the DKO pigs . The four DKO and six WT pigs used for PRRSV WUH3 viral challenge were both about 45 days old . Viral inoculation was conducted by nasal intubation drip ( 2 mL: 106 TCID50/mL ) and intramuscular injection ( 2 mL: 106 TCID50/mL ) . During the 14 days of PRRSV challenge , piglet rectal temperature and clinical symptoms data ( feeding , breathing , defecation , mental state ) were collected every morning . At the same time , piglet survival rate was recorded , blood was collected , and the piglets were weighed regularly . If any pigs died during the course of the PRRSV challenge , pictures were immediately taken and samples were collected . All surviving pigs were slaughtered at 14 days post-infection ( dpi ) and lung tissue was examined for disease symptoms . Four DKO pigs and six WT pigs were used for TGEV challenge . Pigs were inoculated with a total of 10 mL of TGEV strain WH-1 ( GenBank accession number HQ462571 ) ( An et al . , 2014 ) ( 7 × 105 TCID50/mL ) that were orally administered to each pig in two doses ( day 0 and day 1 , 5 mL/day ) . For the PDCoV challenge , two DKO pigs and four WT pigs were orally administered a total of 16 mL of PDCoV strain CHN-HN-2014 ( 2 . 5 × 108 TCID50/mL ) divided into two doses delivered on day 0 and day 1 ( 8 mL/day ) . For both the TGEV and PDCoV groups , the rectal temperature of the pigs was measured daily for the full 14 day experiment and the diarrhea of the piglets was observed . Blood was collected and the piglets were weighed regularly . In the TGEV group , a DKO pig and a WT pig were slaughtered on day 3 , and the remaining pigs in the TGEV group and all pigs in the PDCoV group were slaughtered at 14 dpi . After slaughter , the pigs were dissected to observe the gross lesions in small intestine tissue , to collect small intestine tissue samples , and to detect any pathological changes by H and E staining . Meanwhile , during these 14 days , 4 WT pigs and 2 DKO pigs were reared under the same conditions without any virus infection , and these pigs were used as the Mock group . Lung tissues of pigs in the PRRSV challenge group , and duodenum , jejunum , and ileum tissues in the TGEV and PDCoV groups were collected . The tissues were fixed in 4% paraformaldehyde fixative , dehydrated , embedded , and cut into 3 ~ 8 μm-thick sections . For histopathology , the sections were stained by H and E . For IHC , tissue sections were stained with antibodies specific to the corresponding protein antigens . Tissue sections were then observed and photographed with a fluorescence microscope . The antibodies used to detect pAPN protein were purchased from Abcam ( ab108310 ) ; the antibody used to detect PRRSV-N protein was made by our laboratory . The blood tissues of three experimental groups of pigs were collected at different times after viral challenge and the sera were separated . For the PRRSV group , the sera from all samples were subjected to PRRS antibody detection by commercially available enzyme-linked immunosorbent assay ( ELISA ) kit ( IDEXX , ME ) . The antibody level was determined to be negative or positive according to the S/P value . If S/P<0 . 4 , the antibody is negative , and if S/P≥0 . 4 , the antibody is positive . In order to detect TGEV-specific and PDCoV-specific antibody levels in serum , we used a serum neutralization test ( SNT ) . Briefly , sera were heat inactivated by 30 min of incubation in a 56°C water bath . Then serial 2-fold dilutions of serum samples in four replicates were mixed with 200 TCID50 of TGEV strain WH-1 in a 1:1 ration . After incubation , 100 μl of the mixture was added into ST cells ( a swine testicular cell line permissive of TGEV infection; ATCC CRL-1746 ) at a confluence of ~90% , seeded in 96-well cell culture plates . Appropriate serum , virus ( 200 TCID50 , 20 TCID50 , 2 TCID50 , and 0 . 2 TCID50 ) , and cell controls were included in this test . For about 72 hr after incubation , the cells were monitored for TGEV-specific cytopathic effects . Neutralization titers were calculated as the reciprocal of the highest dilution resulting in complete neutralization . Similarly , sera were diluted mixed with 200 TCID50 of PDCoV strain CHN-HN-2014 . In contrast , PDCoV titers were assessed using LLC-PK1 cells ( a porcine kidney cell line permissive of PDCoV infection; ATCC CL-101 ) that were washed twice with Dulbecco’s Modified Eagle’s Medium ( DMEM ) ( Invitrogen , CA ) , and supplemented with 7 . 5 μg/mL trypsin ( Gibco , USA ) prior to and after 1 hr incubation with these mixtures . Cells were then cultured in DMEM supplemented with 7 . 5 μg/mL trypsin for approximately 72 hr , and the neutralization titers of sera from PDCoV group were calculated . To quantify the copies of PRRSV and PDCoV in the infected experimental group , we extracted PRRSV RNA from PAMs , serum , lung tissue , and tonsil tissue from both the challenge and the mock-inoculated group , and extracted PDCoV RNA from DKO PAMs and WT PAMs after infected with PDCoV . RNA extraction was performed using TRIzol reagent ( Omega Bio-Tek ) . The RNA was reverse transcribed into cDNA according to the instructions for a Transcriptor First Strand cDNA Synthesis Kit ( Roche ) . The cDNA was then amplified with SYBR green real-time PCR master mix ( Applied Biosystems ) in an ABI 7500 real-time PCR system ( Applied Biosystems ) . RNA copy numbers were calculated from a standard curve drawn from positive standards at different dilutions . The primers used for qRT-PCR are listed as follows: 5’-GCAATTGTGTCTGTCGTC-3’ and 5’-CTTATCCTCCCTGAATCTGAC-3’ for PRRSV; 5’-GCCCTCGGTGGTTCTATCTT-3’ and 5’-TCCTTAGCTTGCCCCAAATA-3’ for PDCoV . DKO PAMs and WT PAMs in 24-well cell culture plates were infected or mock-infected with PDCoV at a multiplicity of infection ( MOI ) of 10 . At 24 hpi , cells were fixed with 4% paraformaldehyde for 15 min and permeabilized with methanol for 10 min at room temperature . The cells were then blocked with bovine serum albumin ( 5% ) diluted in phosphate-buffered saline ( PBS ) for 1 hr , and incubated with a PDCoV-N-protein-specific monoclonal antibody for 1 hr and an Alexa Fluor 488-conjugated donkey anti-mouse IgG for 1 hr . The cell nuclei were counterstained with 4’ , 6-diamidino-2-phenylindole ( DAPI ) for 15 min at room temperature . After three washes with PBS , the stained cells were observed with an inverted fluorescence microscope ( Olympus IX73 , Japan ) . PDCoV-infected PAMs were frozen and thawed repeatedly to completely release viruses . Next , LLC-PK1 cells ( a pig kidney cell line known to be highly permissive to PDCoV infection ) were seeded in 96-well plates and were infected with 10-fold serial dilutions of virus samples in eight replicates . At 72 hpi , PDCoV titers were calculated based on cytopathic effects and expressed as the TCID50 value per milliliter , using the Reed–Muench method . The amount of Hp in serum was measured using an enzyme-linked immunosorbent assay ( ELISA ) kit ( 6250–40 , Alpha Diagnostic ) specific to pig Hp , as previously described ( Yang et al . , 2018 ) . Assays were performed in triplicate for each sample . The quality and performance of pigs related to slaughter were determined by a third-party testing center ( The national breeding swine quality supervision and testing center ( Chongqing ) , Ministry of Agriculture and Rural Affairs of China ) . All testing followed the guidelines stipulated in the ‘rules for performance testing of breeding pigs’ document published by the Ministry of Agriculture and Rural Affairs of PR China ( NY/T 822–2004 ) . Briefly , DKO pigs and control WT pigs were weighed before slaughter , euthanized after fasting for 24 hr , and hairs , heads , hoofs , and internal organs were removed after carcass dissection . The weight of carcass , length of carcass , loin eye areas , thickness of skin , and backfat thickness of carcass were all measured . Ham , skin , bone , lean , and fat were dissected from the left side of the carcass and their individual weights were determined . To evaluate meat quality , we measured muscle pH , meat color score , intramuscular fat , marbling , and drip loss of longissimus dorsi . For analysis of pork nutrition , total protein , total fat , ash , moisture , amino acid , and individual minerals , amino acids were analyzed for the longissimus dorsi . The nutritional content of the pork was tested by the Beijing Institute of Nutritional Sources . Semen from DKO pigs and WT control pigs were collected and returned to the laboratory in a 17°C incubator for testing their quality . The detection system was Hamilton-Thorne Research IVOS II computer-assisted sperm analyzer to measure the concentration , motility , and velocity distribution of the sperm . All data are presented as the mean ± standard error of mean ( SEM ) . Data from each of the two groups of pigs were compared with an unpaired t-test when a normal distribution was not obtained . The significance levels were set at 0 . 05 , 0 . 01 , and 0 . 001 , as indicated by * , ** , *** , respectively . The data was analyzed with GraphPad Prism 6 . 0 . 0 for Windows ( GraphPad Software , La Jolla , California ) . | Pig epidemics are the biggest threat to the pork industry . In 2019 alone , hundreds of billions of dollars worldwide were lost due to various pig diseases , many of them caused by viruses . The porcine reproductive and respiratory virus ( PRRS virus for short ) , for instance , leads to reproductive disorders such as stillbirths and premature labor . Two coronaviruses – the transmissible gastroenteritis virus ( or TGEV ) and the porcine delta coronavirus – cause deadly diarrhea and could potentially cross over into humans . Unfortunately , there are still no safe and effective methods to prevent or control these pig illnesses , but growing disease-resistant pigs could reduce both financial and animal losses . Traditionally , breeding pigs to have a particular trait is a slow process that can take many years . But with gene editing technology , it is possible to change or remove specific genes in a single generation of animals . When viruses infect a host , they use certain proteins on the surface of the host’s cells to find their inside: the PRRS virus relies a protein called CD163 , and TGEV uses pAPN . Xu , Zhou , Mu et al . used gene editing technology to delete the genes that encode the CD163 and pAPN proteins in pigs . When the animals were infected with PRRS virus or TGEV , the non-edited pigs got sick but the gene-edited animals remained healthy . Unexpectedly , pigs without CD163 and pAPN also coped better with porcine delta coronavirus infections , suggesting that CD163 and pAPN may also help this coronavirus infect cells . Finally , the gene-edited pigs reproduced and produced meat as well as the control pigs . These experiments show that gene editing can be a powerful technology for producing animals with desirable traits . The gene-edited pigs also provide new knowledge about how porcine viruses infect pigs , and may offer a starting point to breed disease-resistant animals on a larger scale . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"microbiology",
"and",
"infectious",
"disease"
] | 2020 | CD163 and pAPN double-knockout pigs are resistant to PRRSV and TGEV and exhibit decreased susceptibility to PDCoV while maintaining normal production performance |
The Firmicutes are a phylum of bacteria that dominate numerous polymicrobial habitats of importance to human health and industry . Although these communities are often densely colonized , a broadly distributed contact-dependent mechanism of interbacterial antagonism utilized by Firmicutes has not been elucidated . Here we show that proteins belonging to the LXG polymorphic toxin family present in Streptococcus intermedius mediate cell contact- and Esx secretion pathway-dependent growth inhibition of diverse Firmicute species . The structure of one such toxin revealed a previously unobserved protein fold that we demonstrate directs the degradation of a uniquely bacterial molecule required for cell wall biosynthesis , lipid II . Consistent with our functional data linking LXG toxins to interbacterial interactions in S . intermedius , we show that LXG genes are prevalent in the human gut microbiome , a polymicrobial community dominated by Firmicutes . We speculate that interbacterial antagonism mediated by LXG toxins plays a critical role in shaping Firmicute-rich bacterial communities .
Bacteria in polymicrobial environments must persist in the face of frequent physical encounters with competing organisms . Studies have revealed Gram-negative bacterial species contend with this threat by utilizing pathways that mediate antagonism toward contacting bacterial cells ( Konovalova and Søgaard-Andersen , 2011 ) . For instance , Proteobacteria widely employ contact-dependent inhibition ( CDI ) to intoxicate competitor cells that share a high degree of phylogenetic relatedness ( Hayes et al . , 2014 ) . Additionally , both Proteobacteria and bacteria belonging to the divergent phylum Bacteroidetes deliver toxins to competitor Gram-negative cells in an indiscriminate fashion through the type VI secretion system ( T6SS ) ( Russell et al . , 2014a , 2014b ) . Although toxin delivery by CDI and the T6SS is mechanistically distinct , cells harboring either pathway share the feature of prohibiting self-intoxication with immunity proteins that selectively inactivate cognate toxins through direct binding . Few mechanisms that mediate direct antagonism between Gram-positive bacteria have been identified . In Bacillus subtilis , Sec-exported proteins belonging to the YD-repeat family have been shown to potently inhibit the growth of contacting cells belonging to the same strain ( Koskiniemi et al . , 2013 ) ; however , to our knowledge , a pathway that mediates interspecies antagonism between Gram-positive bacteria has not been identified . Given that Gram-positive and Gram-negative bacteria inhabit many of the same densely populated polymicrobial environments ( e . g . the human gut ) , it stands to reason that the former should also possess mechanisms for more indiscriminate targeting of competing cells . Contact-dependent toxin translocation between bacteria is primarily achieved using specialized secretion systems . Gram-negative export machineries of secretion types IV , V , and VI have each been implicated in this process ( Aoki et al . , 2005; Hood et al . , 2010; Souza et al . , 2015 ) . A specialized secretion system widely distributed among Gram-positive bacteria is the Esx pathway ( also referred to as type VII secretion ) ( Abdallah et al . , 2007 ) . This pathway was first identified in Mycobacterium tuberculosis , where it plays a critical role in virulence ( Stanley et al . , 2003 ) . Indeed , attenuation of the vaccine strain M . bovis BCG can be attributed to a deletion inactivating ESX-1 secretion system present in virulence strains ( Lewis et al . , 2003; Pym et al . , 2003 ) . Subsequent genomic studies revealed that the Esx pathway is widely distributed in Actinobacteria , and that a divergent form is present in Firmicutes ( Gey Van Pittius et al . , 2001; Pallen , 2002 ) . Though they share little genetic similarity , all Esx pathways studied to-date utilize a characteristic FtsK-like AAA+ ATPase referred to as EssC ( or EccC ) to catalyze the export of one or more substrates belonging to the WXG100 protein family ( Ates et al . , 2016 ) . Proteins in this family , including ESAT-6 ( EsxA ) and CFP10 ( EsxB ) from M . tuberculosis , heterodimerize in order to transit the secretion machinery . The presence of the Esx secretion system in environmental bacteria as well as commensal and pathogenic bacteria that specialize in colonizing non-sterile sites of their hosts , suggests that the pathway may be functionally pliable . Supporting this notion , ESX-3 of M . tuberculosis is required for mycobactin siderophore-based iron acquisition and the ESX-1 and ESX-4 systems of M . smegmatis are linked to DNA transfer ( Gray et al . , 2016; Siegrist et al . , 2009 ) . In Firmicutes , a Staphylococcus aureus Esx-exported DNase toxin termed EssD ( or EsaD ) has been linked to virulence and contact-independent intraspecies antibacterial activity ( Cao et al . , 2016; Ohr et al . , 2017 ) . Aravind and colleagues have noted that Esx secretion system genes are often linked to genes encoding polymorphic toxins belonging to the LXG protein family ( Zhang et al . , 2012 ) . Analogous to characteristic antimicrobial polymorphic toxins of Gram-negative bacteria , the LXG proteins consist of a conserved N-terminal domain ( LXG ) , a middle domain of variable length , and a C-terminal variable toxin domain . The LXG domain is predicted to adopt a structure resembling WXG100 proteins , thus leading to speculation that these proteins are Esx secretion system substrates ( Zhang et al . , 2011 ) . Despite the association between LXG proteins and the Esx secretion system , to-date there are no experimental data linking them functionally . However , an intriguing study performed by Hayes and colleagues demonstrated antibacterial properties of B . subtilis LXG RNase toxins via heterologous expression in E . coli ( Holberger et al . , 2012 ) . This growth inhibition was alleviated by co-expression of immunity determinants encoded adjacent to cognate LXG genes . We show here that LXG proteins transit the Esx secretion system of Streptococcus intermedius ( Si ) and function as antibacterial toxins that mediate contact-dependent interspecies antagonism .
We initiated our investigation into the function of LXG proteins by characterizing the diversity and distribution of genes encoding these proteins across all sequenced genomes from Firmicutes . As noted previously , the C-terminal domains in the LXG family members we identified are highly divergent , exhibiting a wide range of predicted activities ( Figure 1a ) ( Zhang et al . , 2012 ) . LXG protein-encoding genes are prevalent and broadly distributed in the classes Clostridiales , Bacillales and Lactobacillales ( Figure 1A ) . Notably , a significant proportion of organisms in these taxa are specifically adapted to the mammalian gut environment . Indeed , we find that LXG genes derived from reference genomes of many of these gut-adapted bacteria are abundant in metagenomic datasets from human gut microbiome samples ( Figure 1A and Figure 1—figure supplement 1 ) . An LXG toxin that is predicted to possess ADP-ribosyltransferase activity – previously linked to interbacterial antagonism in Gram-negative organisms – was particularly abundant in a subset of human gut metagenomes ( Zhang et al . , 2012 ) . Close homologs of this gene are found in Ruminococcus , a dominant taxa in the human gut microbiome , potentially explaining the frequency of this gene ( Wu et al . , 2011 ) . 10 . 7554/eLife . 26938 . 003Figure 1 . The LXG protein family contains diverse toxins that are broadly distributed in Firmicutes and found in the human gut microbiome . ( A ) Dendogram depicts LXG-containing genera within Firmicutes , clustered by class and order . Circle size indicates the number of sequenced genomes searched within each genus and circle color represents percentage of those found to contain at least one LXG protein . For classes or orders in which no LXG domain-containing proteins were found , the number of genera evaluated is indicated in parentheses; those consisting of Gram-negative organisms are boxed with dashed lines . Grey boxes contain predicted domain structures for representative divergent LXG proteins . Depicted are LXG-domains ( pink ) , spacer regions ( light grey ) and C-terminal polymorphic toxin domains ( NADase , purple; non-specific nuclease , orange; AHH family nuclease , green; ADP-ribosyltransferase , blue; lipid II phosphatase based on orthology to TelC ( defined biochemically herein ) , yellow; EndoU family nuclease , brown; unknown activity , dark grey ) . ( B ) Heatmap depicting the relative abundance ( using logarithmic scale ) of selected LXG genes detected in the Integrated Gene Catalog ( IGC ) . A complete heatmap is provided in Figure 1—figure supplement 1 . Columns represent individual human gut metagenomes from the IGC database and rows correspond to LXG genes . Grey lines link representative LXG toxins in ( A ) to their corresponding ( ≥95% identity ) IGC group in ( B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 26938 . 00310 . 7554/eLife . 26938 . 004Figure 1—figure supplement 1 . Complete list of LXG genes found in human gut metagenomes . Heatmap depicting the relative abundance ( using logarithmic scale ) of LXG genes detected in the Integrated Gene Catalog ( IGC ) . Columns represent individual human gut metagenomes from the IGC database and rows correspond to LXG genes . Row and column ordering was determined by hierarchical clustering using Euclidian distance and complete linkage . DOI: http://dx . doi . org/10 . 7554/eLife . 26938 . 004 We next sought to determine whether LXG proteins are secreted via the Esx pathway . The toxin domain of several of the LXG proteins we identified shares homology and predicted catalytic residues with M . tuberculosis TNT , an NAD+-degrading ( NADase ) enzyme ( Figure 2—figure supplement 1A ) ( Sun et al . , 2015 ) . Si , a genetically tractable human commensal and opportunistic pathogen , is among the bacteria we identified that harbor a gene predicted to encode an NADase LXG protein ( Claridge et al . , 2001 ) ; we named this protein TelB ( Toxin exported by Esx with LXG domain B ) . Attempts to clone the C-terminal toxin domain of TelB ( TelBtox ) were initially unsuccessful , suggesting the protein exhibits a high degree of toxicity . Guided by the TNT structure , we circumvented this by assembling an attenuated variant ( H661A ) that was tolerated under non-induced conditions ( TelBtox* ) ( Figure 2—figure supplement 1A ) ( Sun et al . , 2015 ) . Induced expression of TelBtox* inhibited E . coli growth and reduced cellular NAD+ levels ( Figure 2A , Figure 2—figure supplement 1B ) . The extent of NAD+ depletion mirrored that catalyzed by expression of a previously characterized interbacterial NADase toxin , Tse6 , and importantly , intracellular NAD+ levels were unaffected by an unrelated bacteriostatic toxin , Tse2 ( Hood et al . , 2010; Whitney et al . , 2015 ) . Furthermore , substitution of a second predicted catalytic residue of TelB ( R626A ) , abrogated toxicity of TelBtox* and significantly restored NAD+ levels ( Figure 2—figure supplement 1B–C ) . 10 . 7554/eLife . 26938 . 005Figure 2 . LXG-domain proteins of S . intermedius are secreted by the Esx-pathway . ( A ) NAD+ levels in E . coli cells expressing a non-NAD+ -degrading toxin ( Tse2 ) , the toxin domain of a known NADase ( Tse6tox ) , an inducibly toxic variant of the C-terminal toxin domain of TelB ( TelBtox* ) , a variant of TelBtox* with significantly reduced toxicity ( TelBtox*R626A ) and TelBtox* co-expressed with its cognate immunity protein TipB . Cellular NAD+ levels were assayed 60 min after induction of protein expression and were normalized to untreated cells . Mean values ( n = 3 ) ± SD are plotted . Asterisks indicate statistically significant differences in NAD+ levels compared to vector control ( p<0 . 05 ) . ( B ) NAD+ consumption by culture supernatants from the indicated Si strains . Fluorescent images of supernatant droplets supplemented with 2 mM NAD+ for 3 hr; brightness is proportional to NAD+ concentration and was quantified using densitometry . Mean values ± SD ( n = 3 ) are plotted . Asterisks indicate statistically significant differences in NAD+ turnover compared to wild-type SiB196 ( p<0 . 05 ) . ( C ) Regions of the SiB196 genome encoding Esx-exported substrates . Genes are colored according to functions encoded ( secreted Esx structural components , orange; secreted LXG toxins , dark purple; immunity determinants , light purple; WXG100-like proteins , green; other , grey ) . ( D ) Western blot analysis of TelC secretion in supernatant ( Sup ) and cell fractions of wild-type or essC-inactivated SiB196 . DOI: http://dx . doi . org/10 . 7554/eLife . 26938 . 00510 . 7554/eLife . 26938 . 006Figure 2—figure supplement 1 . TelB resembles NADase toxins and inhibits the growth of bacteria . ( A ) Structural model of TelBtox based on TNT toxin from M . tuberculosis . Conserved residues implicated in NAD+ binding are indicated . ( B ) Viability of E . coli cells grown on solid media harboring inducible plasmids expressing TelBtox* , TelBtox*R626A or an empty vector control . Mean c . f . u . values ± SD ( n = 3 ) are plotted . Asterisk indicates a statistically significant difference in E . coli viability relative to vector control ( p<0 . 05 ) ( C ) Growth in liquid media of E . coli cells expressing plasmids shown in ( B ) . Protein expression was induced at the indicated time ( arrow ) . Error bars indicate ± SD ( n = 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 26938 . 006 Determination of the biochemical activity of TelB provided a means to test our hypothesis that LXG proteins are substrates of the ESX secretion pathway . Using an assay that exploits fluorescent derivatives of NAD+ that form under strongly alkaline conditions , we found that concentrated cell-free supernatant of an Si strain containing telB ( SiB196 ) possesses elevated levels of NADase activity relative to that of a strain lacking telB ( Si27335 ) ( Figure 2B ) ( Johnson and Morrison , 1970; Olson et al . , 2013; Whiley and Beighton , 1991 ) . Furthermore , the NADase activity present in the supernatant of SiB196 was abolished by telB inactivation . Export of Esx substrates relies on EssC , a translocase with ATPase activity ( Burts et al . , 2005; Rosenberg et al . , 2015 ) . Inactivation of essC also abolished NADase activity in the supernatant of SiB196 , suggesting that TelB utilizes the Esx pathway for export . The genome of SiB196 encodes two additional LXG proteins , which we named TelA and TelC ( Figure 2C ) . To determine if these proteins are also secreted in an Esx-dependent fashion , we collected cell-free supernatants from stationary phase cultures of wild-type and essC-deficient SiB196 . Extensive dialysis was used to reduce contamination from medium-derived peptides and the remaining extracellular proteins were precipitated and identified using semi-quantitative mass spectrometry ( Liu et al . , 2004 ) . This technique revealed that each of the LXG proteins predicted by the Si genome is exported in an Esx-dependent manner ( Table 1 ) . Western blot analysis of TelC secretion by wild-type and the essC-lacking mutant further validated Esx-dependent export ( Figure 2D ) . Together , these data indicate that LXG proteins are substrates of the Esx secretion system . 10 . 7554/eLife . 26938 . 007Table 1 . The Esx-dependent extracellular proteome of S . intermedius B196 . DOI: http://dx . doi . org/10 . 7554/eLife . 26938 . 007Locus tagWild-typeΔessCRelative abundance ( Wild-type/ΔessC ) Esx functionNameSIR_0169*19 . 67†0Not detected in ΔessCLXG protein‡TelASIR_017614 . 670Not detected in ΔessCStructural componentEsaASIR_148912 . 000Not detected in ΔessCLXG proteinTelCSIR_15169 . 330Not detected in ΔessC-Trigger FactorSIR_01795 . 330Not detected in ΔessCLXG proteinTelBSIR_0166140 . 0017 . 488 . 01Structural componentEsxASIR_027315 . 332 . 286 . 73--SIR_162615 . 002 . 286 . 58-GroELSIR_083212 . 338 . 361 . 48-EnolaseSIR_190449 . 0037 . 241 . 32-Putative serine proteaseSIR_138226 . 0019 . 761 . 32-Fructose-bisphosphate aldolaseSIR_064821 . 6717 . 481 . 24-50S ribosomal protein L7/L12SIR_021247 . 0039 . 521 . 19-Elongation Factor GSIR_00818 . 677 . 601 . 14-Putative outer membrane proteinSIR_167616 . 3314 . 441 . 13-phosphoglycerate kinaseSIR_152312 . 6712 . 920 . 98-DnaKSIR_115410 . 3310 . 640 . 97-Putative bacteriocin accessory proteinSIR_102763 . 0067 . 640 . 93-Elongation Factor TuSIR_145514 . 0015 . 960 . 88--SIR_075813 . 0015 . 200 . 86--SIR_13879 . 3311 . 400 . 82-Putative extracellular solute-binding proteinSIR_049212 . 3315 . 200 . 81-Putative adhesion proteinSIR_103317 . 6724 . 320 . 73--SIR_135914 . 0019 . 760 . 71-Penicillin-binding protein 3SIR_001112 . 3317 . 480 . 71-Beta-lactamase class ASIR_15468 . 3312 . 160 . 69--SIR_0040101 . 67160 . 360 . 63-Putative stress proteinSIR_160811 . 0018 . 240 . 60-Putative endopeptidase OSIR_15497 . 3312 . 160 . 60--SIR_167579 . 00132 . 240 . 60-Putative cell-surface antigen I/IISIR_141811 . 3321 . 280 . 53-Putative transcriptional regulator LytRSIR_008011 . 0021 . 280 . 52--SIR_102528 . 3363 . 840 . 44-LysozymeSIR_011310 . 6724 . 320 . 44--SIR_02978 . 3324 . 320 . 34--*Rows highlighted in green correspond to proteins linked to the Esx pathway . †Values correspond to average SC ( spectral counts ) of triplicate biological replicates for each strain . ‡Functional link of LXG proteins to Esx secretion pathway defined in the study . The export of LXG proteins by the Esx pathway motivated us to investigate their capacity for mediating interbacterial antagonism . The C-terminal domains of TelA ( TelAtox ) and TelC ( TelCtox ) bear no homology to characterized proteins , so we first examined the ability of these domains to exhibit toxicity in bacteria . TelAtox and TelBtox* inhibited growth when expressed in the cytoplasm of E . coli , whereas TelCtox did not exhibit toxicity in this cellular compartment ( Figure 3A ) . Given the capacity of some interbacterial toxins to act on extracellular structures , we assessed the viability of Si cells expressing TelCtox targeted to the sec translocon . In contrast to TelCtox production , overexpression of a derivative bearing a signal peptide directing extracellular expression ( ss-TelCtox ) exhibited significant toxicity ( Figure 3B ) . 10 . 7554/eLife . 26938 . 008Figure 3 . S . intermedius LXG proteins inhibit bacterial growth and mediate contact-dependent interbacterial antagonism . ( A ) Viability of E . coli cells grown on solid media harboring inducible plasmids expressing the C-terminal toxin domains of the three identified SiB196 LXG proteins or an empty vector control . ( B ) SiB196 colonies recovered after transformation with equal concentrations of constitutive expression plasmids carrying genes encoding the indicated proteins . ss-TelCtox is targeted to the sec translocon through the addition of the secretion signal sequence from S . pneumoniae LysM ( SP_0107 ) . Error bars represent ± SD ( n = 3 ) . Asterisk indicates a statistically significant difference in Si transformation efficiency relative to TelCtox ( p<0 . 05 ) . ( C ) Viability of E . coli cells grown on solid media harboring inducible plasmids co-expressing the indicated proteins . Empty vector controls are indicated by a dash . Mean c . f . u . values ± SD ( n = 3 ) are plotted . Asterisks indicate statistically significant differences in E . coli viability relative to vector control ( p<0 . 05 ) ( D ) Intra-species growth competition experiments between the indicated bacterial strains . Competing strains were mixed and incubated in liquid medium or on solid medium for 30 hr and both initial and final populations of each strain were enumerated by plating on selective media . The competitive index was determined by comparing final and initial ratios of the two strains . Asterisks indicate outcomes statistically different between liquid and solid medium ( n = 3 , p<0 . 05 ) . ( E ) Intra-species growth competition experiments performed as in ( D ) except for the presence of a filter that inhibits cell-cell contact . No contact , filter placed between indicated donor and susceptible recipient ( ∆telB ∆tipB ) strains; Contact , donor and susceptible recipient strains mixed on same side of filter . Asterisks indicate statistically different outcomes ( n = 3 , p<0 . 05 ) . Note that recipient cell populations have an Esx-independent fitness advantage in these experiments by virtue of their relative proximity to the growth substrate . ( F ) Inter-species growth competition experiments performed on solid or in liquid ( E . faecalis ) medium between Si wild-type and ∆essC donor strains and the indicated recipient organisms . Si23775 lacks tipA and tipB and is therefore potentially susceptible to TelA and TelB delivered by SiB196 . Asterisks indicate outcomes where the competitive index of wild-type was significantly higher than an ∆essC donor strain ( n = 3 , p<0 . 05 ) . Genetic complementation of the mutant phenotypes presented in this figure was confounded by inherent plasmid fitness costs irrespective of the inserted sequence . As an alternative , we performed whole genome sequencing on strains ∆essC , ∆telB , ∆telC , ∆telB ∆tipB , and ∆telC ∆tipC , which confirmed the respective desired mutation as the only genetic difference between these strains . Sequences of these strains have been deposited to the NCBI Sequence Read Archive ( BioProject ID: PRJNA388094 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 26938 . 00810 . 7554/eLife . 26938 . 009Figure 3—figure supplement 1 . TelC directly interacts with its cognate immunity protein TipC . ( A ) Bacterial two-hybrid assay for interaction between TelC and TipC . Adenylate cyclase subunit T25 fusions ( TipC and Zip control protein ) and T18 fusions ( TelC and fragments thereof ) were coexpressed in the indicated combinations . Successful interaction results in production of blue pigment . ( B ) ITC analysis indicates TelCtox and mature TipC ( TipCΔSS ) interact with nanomolar affinity . The top panel displays the heats of injection , whereas the bottom panel shows the normalized integration data as a function of the syringe and cell concentrations . DOI: http://dx . doi . org/10 . 7554/eLife . 26938 . 00910 . 7554/eLife . 26938 . 010Figure 3—figure supplement 2 . TelC levels elevated by high cell density or addition of purified protein fail to yield cellular intoxication in liquid media . ( A ) Intra-species growth competition experiments between the indicated Si B196 strains . Competing strains were mixed at a ratio of 40:1 [donor:recipient] and concentrated to OD600nm = 20 . Competition outcomes were determined after 24 hr by enumerating c . f . u . on selective media . ( B ) Coomassie stained gel of purified TelC-his6 . ( C ) Growth in liquid media of SiB196ΔtelC ΔtipC cells incubated with buffer ( Ctrl ) or 0 . 1 mg/mL TelC-his6 . Error bars indicate ± SD ( n = 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 26938 . 010 We next evaluated whether the Tel proteins , like the substrates of interbacterial toxin delivery systems in Gram-negative bacteria , are inactivated by genetically linked specialized cognate immunity determinants . By co-expressing candidate open reading frames located downstream of each tel gene , we identified a cognate tip ( tel immunity protein ) for each toxin ( Figure 3B–C and Figure 3—figure supplement 1 ) . We then sought to inactivate each of these factors to generate SiB196 strains sensitive to each of the Tel proteins . In SiB196 , telA tipA loci are located immediately upstream of conserved esx genes ( Figure 2C ) . We were unable to generate non-polar telA tipA-inactivated strains , and thus focused our efforts on the other two tel tip loci . We reasoned that if LXG toxins target non-self cells , this process would occur either through diffusion or by facilitated transfer , the latter of which would likely require cell contact . Since we detect TelA-C secretion in liquid medium , we began our attempts to observe intercellular intoxication with wild-type and toxin-sensitive target cell co-culture . These efforts yielded no evidence of target cell killing or growth inhibition , including when co-incubations were performed at cell densities higher than that achievable through growth ( Figure 3D , Figure 3—figure supplement 2A ) . The application of concentrated supernatants or purified TelC ( to a final concentration of 0 . 1 mg/mL ) to sensitive strains also did not produce evidence of toxicity ( Figure 3—figure supplement 2B–C ) . This result is perhaps not surprising given the barrier presented by the Gram-positive cell wall ( Forster and Marquis , 2012 ) . Next , we tested conditions that enforce cell contact . In each of these experiments , donor and recipient strains were grown in pure culture before they were mixed at defined ratios and cultured on a solid surface for 30 hr to promote cell-cell interactions . We observed significant growth inhibition of TelB- or TelC-susceptible strains co-cultured with wild-type , but not when co-cultured with strains lacking telB or telC , respectively ( Figure 3D ) . A strain bearing inactivated essC was also unable to intoxicate a sensitive recipient . In competition experiments performed in parallel wherein the bacterial mixtures were grown in liquid culture , TelB and TelC-susceptible strains competed equally with wild type , suggesting that Esx-mediated intoxication requires prolonged cell contact . To further probe this requirement , we conducted related experiments in which wild-type donor cells were segregated from sensitive recipients by a semi-permeable ( 0 . 2 μm pore size ) membrane ( Figure 3E ) . This physical separation blocked intoxication , which taken together with the results of our liquid co-culture experiments and our finding that purified TelC is not bactericidal , strongly suggests that the mechanism of Esx-dependent intercellular LXG protein delivery requires immediate cell-cell contact . In Gram-negative bacteria , some antagonistic cell contact-dependent pathways display narrow target range , whereas others act between species , or even between phyla ( Hayes et al . , 2014; Russell et al . , 2014a ) . To begin to determine the target range of Esx-based LXG protein delivery , we measured its contribution to SiB196 fitness in interbacterial competition experiments with a panel of Gram-positive and -negative bacteria . The Esx pathway conferred fitness to SiB196 in competition with Si23775 , S . pyogenes , and Enterococcus faecalis , an organism from a closely related genus ( Figure 3F ) . On the contrary , the pathway did not measurably affect the competitiveness of SiB196 against Gram-negative species belonging to the phyla Proteobacteria ( E . coli , Burkholderia thailandensis , Pseudomonas aeruginosa ) or Bacteroidetes ( Bacteriodes fragilis ) . These results demonstrate that the Esx pathway can act between species and suggest that its target range may be limited to Gram-positive bacteria . The Esx pathway is best known for its role in mediating pathogen-host cell interactions ( Abdallah et al . , 2007 ) . Given this precedence , we considered the possibility that the antibacterial activity we observed may not be relevant physiologically . TelB degrades NAD+ , a molecule essential for all cellular life , and therefore this toxin is not definitive in this regard . We next turned our attention to TelC , which elicits toxicity from outside of the bacterial cell ( Figure 3B ) . This protein contains a conserved aspartate-rich motif that we hypothesized constitutes its enzymatic active site ( Figure 4—figure supplement 1A ) . To gain further insight into TelC function , we determined the crystal structure of TelCtox to 2 . 0 Å resolution ( Table 2 ) . The structure of TelCtox represents a new fold; it is comprised of distinct and largely α-helical N- and C-terminal lobes ( Figure 4A ) . The single β element of TelCtox is a hairpin that protrudes from the N-terminal lobe . Although TelCtox does not share significant similarity to previously determined structures , we located its putative active site within a shallow groove that separates the N- and C-terminal lobes . This region contains a calcium ion bound to several residues that comprise the conserved aspartate-rich motif . Site-specific mutagenesis of these residues abrogated TelC-based toxicity ( Figure 4B , C , Figure 4—figure supplement 1B ) . 10 . 7554/eLife . 26938 . 011Figure 4 . TelC is a calcium-dependent lipid II phosphatase . ( A ) Space-filling representation of the 2 . 0 Å resolution TelCtox X-ray crystal structure . Protein lobes ( red and blue ) , active site cleft ( white ) and Ca2+ ( green ) are indicated . ( B ) TelCtox structure rotated as indicated relative to ( A ) with transparent surface revealing secondary structure . ( C ) Magnification of the TelC active site showing Ca2+ coordination by conserved aspartate residues and water molecules . ( D ) Viability of S . aureus cells harboring inducible plasmids expressing the indicated proteins or a vector control . ss-TelCtox is targeted for secretion through the addition of the signal sequence encoded by the 5’ end of the hla gene from S . aureus . Mean c . f . u . values ± SD ( n = 3 ) are plotted . Asterisk indicates a statistically significant difference in S . aureus viability relative to vector control ( p<0 . 05 ) ( E ) Representative micrographs of S . aureus expressing ss-TelCtox or a vector control . Frames were acquired eight and 12 hr after spotting cells on inducing growth media . ( F ) Thin-layer chromotography ( TLC ) analysis of reaction products from incubation of synthetic Lys-type lipid II with buffer ( Ctrl ) , TelCtox , or TelCtox and its cognate immunity protein TipC . ( G ) Partial HPLC chromatograms of radiolabeled peptidoglycan ( PG ) fragments released upon incubation of Lys-type lipid II with the indicated purified proteins . Schematics depict PG fragment structures ( pentapeptide , orange; N-acetylmuramic acid , dark green; N-acetylglucosamine , light green; phosphate , black ) . Known fragment patterns generated by PBP1B + LpoB and colicin M serve as controls . ( H ) TLC analysis of reaction products generated from incubation of buffer ( Ctrl ) , TelCtox or TelCtox and TipC with undecaprenyl phosphate ( C55–P ) ( left ) or undecaprenyl pyrophosphate ( C55–PP ) ( right ) . DOI: http://dx . doi . org/10 . 7554/eLife . 26938 . 01110 . 7554/eLife . 26938 . 012Figure 4—figure supplement 1 . TelC contains an aspartate-rich motif required for toxicity . ( A ) Sequence logo representation of the aspartate-rich motif found among TelC orthologous proteins . Numbers indicate the amino acid positions in TelC from SiB196 . ( B ) Image of SiB196 colonies recovered from transformation with plasmids expressing the indicated TelCtox variants . DOI: http://dx . doi . org/10 . 7554/eLife . 26938 . 01210 . 7554/eLife . 26938 . 013Figure 4—figure supplement 2 . TelC does not degrade intact Gram-positive sacculi . ( A and B ) HPLC analysis of muropeptides generated by incubation of TelCtox , cellosyl muramidase or buffer with either S . aureus peptidoglycan sacculi ( A ) or lysostaphin endopeptidase treated ( non-cross-linked ) peptidoglycan sacculi ( B ) . ( C and D ) HPLC analysis of S . aureus cell wall extracts from cells expressing ss-TelCtox or a vector control . Prior to chromatographic separation , cell walls were treated with either lysostaphin endopeptidase ( C ) or cellosyl muramidase ( D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 26938 . 01310 . 7554/eLife . 26938 . 014Figure 4—figure supplement 3 . TelC degrades lipid II , contributes to interbacterial antagonism and is not toxic to yeast cells . ( A ) TLC analysis of reaction products from incubation of synthetic Lys-type lipid II with buffer ( Ctrl ) or TelC . ( B ) Inter-species growth competition experiments performed on solid medium between the indicated Si donor strains and E . faecalis . Asterisks indicate statistically significant differences in competitive indices ( n = 3 , p<0 . 05 ) . ( C ) Growth of Saccharomyces cerevisiae upon expression of native TelCtox , or a derivative in which an added signal sequence targets the protein to the yeast secretory pathway ( ss-TelCtox ) . Yeast strains carrying the empty vector or a toxic protein ( Ctrl ) are included for comparison . ( D ) Western blot analysis of TelCtox and ss-TelCtox in Saccharomyces cerevisiae . Black arrow denotes proteolytically processed ss-TelCtox . DOI: http://dx . doi . org/10 . 7554/eLife . 26938 . 01410 . 7554/eLife . 26938 . 015Table 2 . X-ray data collection and refinement statistics . DOI: http://dx . doi . org/10 . 7554/eLife . 26938 . 015TelC202-CT ( Semet ) Data CollectionWavelength ( Å ) 0 . 979Space groupC2221Cell dimensionsa , b , c ( Å ) 127 . 4 , 132 . 7 , 58 . 3α , β , γ ( ° ) 90 . 0 , 90 . 0 , 90 . 0Resolution ( Å ) 49 . 20–1 . 98 ( 2 . 03–1 . 98 ) *Total observations891817Unique observations34824Rpim ( % ) 6 . 6 ( 138 . 5 ) I/σI11 . 4 ( 0 . 8 ) Completeness ( % ) 100 . 0 ( 99 . 9 ) Redundancy25 . 6 ( 23 . 4 ) RefinementRwork / Rfree ( % ) 22 . 4/24 . 6Average B-factors ( Å2 ) 53 . 8No . atoms Protein2539 Ligands3 Water145Rms deviations Bond lengths ( Å ) 0 . 008 Bond angles ( ° ) 0 . 884Ramachandran plot ( % ) Total favored96 . 9 Total allowed99 . 7Coordinate error ( Å ) 0 . 28PDB code5UKH*Values in parentheses correspond to the highest resolution shell . We next assessed the morphology of cells undergoing intoxication by TelCtox . Due to the potent toxicity of TelCtox in Si , we employed an inducible expression system in S . aureus as an alternative . S . aureus cells expressing extracellularly-targeted TelCtox exhibited significantly reduced viability ( Figure 4D ) , and when examined microscopically , displayed a cessation of cell growth followed by lysis that was not observed in control cells ( Figure 4E , Videos 1–2 ) . Despite eliciting effects consistent with cell wall peptidoglycan disruption , isolated cell walls treated with TelCtox and peptidoglycan recovered from cells undergoing TelC-based intoxication showed no evidence of enzymatic digestion ( Figure 4—figure supplement 2A–D ) . These data prompted us to consider that TelC corrupts peptidoglycan biosynthesis , which could also lead to the lytic phenotype observed ( Harkness and Braun , 1989 ) . 10 . 7554/eLife . 26938 . 016Video 1 . Time-lapse series of S . aureus USA300 pEPSA5 growth . Cells were imaged every 10 min . DOI: http://dx . doi . org/10 . 7554/eLife . 26938 . 01610 . 7554/eLife . 26938 . 017Video 2 . Time-lapse series of S . aureus USA300 pEPSA5::ss-telCtox growth . Cells were imaged every 10 min . DOI: http://dx . doi . org/10 . 7554/eLife . 26938 . 017 The immediate precursor of peptidoglycan is lipid II , which consists of the oligopeptide disaccharide repeat unit linked via pyrophosphate to a lipid carrier ( Vollmer and Bertsche , 2008 ) . Likely due to its distinctive and conserved structure , lipid II is the target of diverse antibacterial molecules ( Breukink and de Kruijff , 2006; Oppedijk et al . , 2016 ) . To test activity against lipid II , we incubated the molecule with purified TelCtox . Analysis of the reaction products showed that TelCtox cleaves lipid II – severing the molecule at the phosphoester linkage to undecaprenyl ( Figure 4F–G , Figure 4—figure supplement 3A ) . Reaction products were confirmed by mass spectrometry and inclusion of TipC inhibited their formation . Consumption of lipid II for peptidoglycan assembly generates undecaprenyl pyrophosphate ( UPP ) , which is converted to undecaprenyl phosphate ( UP ) , and transported inside the cell . The UP molecule then reenters peptidoglycan biosynthesis or is utilized as a carrier for another essential cell wall constituent , wall teichoic acid ( WTA ) . Our experiments showed that TelCtox is capable of hydrolyzing cleaved undecaprenyl derivatives but displays a strict requirement for the pyrophosphate group ( Figure 4H ) , indicating the potential for TelC to simultaneously disrupt two critical Gram-positive cell wall polymers . Consistent with its ability to inhibit a conserved step in peptidoglycan biosynthesis , TelC exhibited toxicity towards diverse Gram-positive species including Si ( Figure 2B ) , S . aureus ( Figure 4D ) and E . faecalis ( Figure 4—figure supplement 3B ) . These data do not explain our observation that cytoplasmic TelC is non-toxic , as the substrates we defined are present in this compartment . The substrates may be inaccessible or TelC could be inactive in the cytoplasm . It is worth noting that TelC contains a calcium ion bound at the interface of its N- and C-terminal lobes . Many secreted proteins that bind calcium utilize the abundance of the free ion in the milieu to catalyze folding . Taken together , our biochemical and phenotypic data strongly suggest that TelC is a toxin directed specifically against bacteria . While we cannot rule out that TelC may have other targets , we find that its expression in the cytoplasm or secretory pathway of yeast does not impact the viability of this model eukaryotic cell ( Figure 4—figure supplement 3C–D ) . The majority of Esx substrates identified to-date belong to the WXG100 protein family . These proteins typically display secretion co-dependency and are essential for apparatus function . M . tuberculosis ESX-1 exports two WXG100 proteins , ESAT-6 and CFP10 , and the removal of either inhibits the export of other substrates ( Ates et al . , 2016; Renshaw et al . , 2002 ) . LXG proteins do not belong to the WXG100 family; thus , we sought to determine how the Tel proteins influence Esx function in Si . Using Western blot analysis to measure TelC secretion and extracellular NADase activity as a proxy for TelB secretion , we found that telB- and telC-inactivated strains of Si retain the capacity to secrete TelC and TelB , respectively ( Figure 5A–B ) . These data indicate that TelB and TelC are not required for core apparatus function and do not display secretion co-dependency . 10 . 7554/eLife . 26938 . 018Figure 5 . LXG domain proteins are independently secreted and require interaction with cognate WXG100-like partners for export . ( A ) NAD+ consumption assay of culture supernatants of the indicated SiB196 strains . Mean densitometry values ± SD ( n = 3 ) are plotted . Asterisk indicates statistically significant difference in NAD+ turnover compared to wild-type SiB196 ( p<0 . 05 ) . ( B ) Western blot analysis of TelC secretion in supernatant ( Sup ) and cell fractions . ( C ) Western blot and coomassie stain analysis of CoIP assays of TelC-his6 co-expressed with either WxgB-V or WxgC-V proteins . ( D ) Bacterial two-hybrid assay for interaction between Tel and WXG100-like proteins . Adenylate cyclase subunit T25 fusions ( WXG100-like proteins ) and T18 fusions ( Tel proteins and fragments thereof ) were co-expressed in the indicated combinations . Bait-prey interaction results in blue color production . ( E ) Model depicting Esx-dependent cell-cell delivery of LXG toxins between bacteria . The schematic shows an Si donor cell containing cognate TelA-C ( light shades ) and WxgA-C ( dark shades ) pairs intoxicating a susceptible recipient cell . Molecular targets of LXG toxins identified in this study are depicted in the recipient cell . DOI: http://dx . doi . org/10 . 7554/eLife . 26938 . 01810 . 7554/eLife . 26938 . 019Figure 5—figure supplement 1 . Domain architecture of the Tel proteins . The boundaries for the LXG and toxin domains for each protein are based on the protein-protein interaction data and bacterial toxicity assays , respectively , described in this work . DOI: http://dx . doi . org/10 . 7554/eLife . 26938 . 019 Interestingly , we noted genes encoding WXG100-like proteins upstream of telA-C ( wxgA-C ) ( Figure 2C ) ; however , these proteins were not identified in the extracellular proteome of Si ( Table 1 ) . Given the propensity for Esx substrates to function as heterodimers , we hypothesized that the Tel proteins specifically interact with cognate Wxg partners . In support of this , we found that WxgC , but not WxgB co-purified with TelC ( Figure 5C ) . Moreover , using bacterial two-hybrid assays , we determined that this interaction is mediated by the LXG domain of TelC ( Figure 5D ) . To investigate the generality of these findings , we next examined all pairwise interactions between the three Wxg proteins and the LXG domains of the three Tel proteins ( TelA-CLXG ) ( Figure 5—figure supplement 1 ) . We found that WxgA-C interact specifically with the LXG domain of their cognate toxins ( Figure 5D ) . The functional relevance of the LXG–WXG100 interaction was tested by examining substrate secretion in a strain lacking wxgC . We found that wxgC inactivation abrogates TelC secretion , but not that of TelB ( Figure 5A–B ) . In summary , these data suggest that cognate Tel–Wxg interaction facilitates secretion through the Esx pathway of Si ( Figure 5E ) .
We present multiple lines of evidence that Esx-mediated delivery of LXG toxins serves as a physiological mechanism for interbacterial antagonism between Gram-positive bacteria . Our results suggest that like the T6S pathway of Gram-negative bacteria , the Esx system may mediate antagonism against diverse targets , ranging from related strains to species belonging to other genera ( Schwarz et al . , 2010 ) . This feature of Esx secretion , in conjunction with the frequency by which we detect LXG genes in human gut metagenomes , suggests that the system could have significant ramifications for the composition of human-associated polymicrobial communities . Bacteria harboring LXG toxin genes are also components or pathogenic invaders of polymicrobial communities important in agriculture and food processing . For instance , LXG toxins may assist Listeria in colonizing fermented food communities dominated by Lactobacillus and Lactococcus ( Farber and Peterkin , 1991 ) . Of note , the latter genera also possess LXG toxins , which may augment their known antimicrobial properties . Our findings thus provide insights into the forces influencing the formation of diverse communities relevant to human health and industry . Palmer and colleagues recently reported that the Esx system of Staphylococcus aureus exports EssD , a nuclease capable of inhibiting the growth of target bacteria in co-culture ( Cao et al . , 2016 ) . The relationship between these findings and those we report herein is currently unclear . S . aureus EssD does not possess an LXG domain and was reported to be active against susceptible bacteria during co-incubation in liquid media , a condition we found not conducive to LXG toxin delivery ( Figure 3D ) . It is evident that the Esx pathway is functionally pliable ( Burts et al . , 2005; Conrad et al . , 2017; Gray et al . , 2016; Gröschel et al . , 2016; Manzanillo et al . , 2012; Siegrist et al . , 2009 ) ; therefore , it is conceivable that it targets toxins to bacteria through multiple mechanisms . The capacity of EssD to act against bacteria in liquid media could be the result of its over-expression from a plasmid , although we found that the exogenous administration of quantities of TelC far exceeding those likely achievable physiologically had no impact on sensitive recipient cells ( Figure 3—figure supplement 2C ) . A later study of EssD function found no evidence of interbacterial targeting and instead reported that its nuclease activity affects IL-12 accumulation in infected mice ( Ohr et al . , 2017 ) . Our data suggest that , like a subset of substrates of the Esx systems of M . tuberculosis , LXG family members require hetero-dimerization with specific WXG100-like partners to be secreted ( Ates et al . , 2016 ) . Hetero-dimerization is thought to facilitate secretion of these substrates due to the requirement for a bipartite secretion signal consisting of a YxxxD/E motif in the C-terminus of one partner in proximity to the WXG motif present in the turn between helices in the second protein ( Champion et al . , 2006; Daleke et al . , 2012a; Poulsen et al . , 2014; Sysoeva et al . , 2014 ) . While the canonical secretion signals found in other Esx substrates appear to be lacking in the LXG proteins and their interaction partners , structure prediction algorithms suggest they adopt similar helical hairpin structures , which could facilitate formation of an alternative form of the bipartite signal . Unlike previously characterized Esx substrates , we found that the LXG proteins are not co-dependent for secretion , and we failed to detect secretion of their WXG100-like interaction partners . This suggests that WxgA-C could function analogously to the EspG proteins of M . tuberculosis , which serve as intracellular chaperones facilitating delivery of specific substrates to the secretion machinery ( Daleke et al . , 2012b; Ekiert and Cox , 2014 ) . Alternatively , Wxg–Lxg complexes could be secreted as heterodimers , but for technical reasons the Wxg member was undetected in our experiments . The paradigm of Lxg-Wxg interaction likely extends beyond S . intermedius , as we observe that LXG proteins from other species are commonly encoded within the same operon as Wxg homologs . Our study leaves open the question of how Esx-exported LXG proteins reach their targets . In the case of TelC , the target resides on the extracellular face of the plasma membrane , and in the case of TelA and TelB , they are cytoplasmic . Crossing the thick Gram-positive cell wall is the first hurdle that must be overcome to deliver of each of these toxins . The size of LXG toxins exceeds that of molecules capable of free diffusion across the peptidoglycan sacculus ( Forster and Marquis , 2012 ) . Donor cell-derived cell wall hydrolytic enzymes may facilitate entry or the LXG proteins could exploit cell surface proteins present on recipient cells . Whether the entry of LXG toxins is directly coordinated by the Esx pathway is not known; our experiments do not rule-out that the requirement for donor-recipient cell contact reflects a step subsequent to secretion by the Esx pathway . Once beyond the sacculus , TelA and TelB must translocate across the plasma membrane . Our study has identified roles for the N- and C-terminal domains of LXG proteins; however , the function of the region between these two domains remains undefined and may participate in entry . Intriguingly , the central domains of TelA and TelB are each over 150 residues , whereas the LXG and toxin domains of TelC , which does not require access to the cytoplasm , appear to directly fuse ( Figure 5—figure supplement 1 ) . Based on the entry mechanisms employed by other interbacterial toxins , this central domain – or another part of the protein – could facilitate direct translocation , proteolytic release of the toxin domain , interaction with a recipient membrane protein , or a combination of these activities ( Kleanthous , 2010; Willett et al . , 2015 ) . We discovered that TelC , a protein lacking characterized homologs , adopts a previously unobserved fold and catalyzes degradation of the cell wall precursor molecule lipid II . This molecule is the target of the food preservative nisin , as well as the last-line antibiotic vancomycin , which is used to treat a variety of Gram-positive infections ( Ng and Chan , 2016 ) . Lipid II is also the target of the recently discovered antibiotic teixobactin , synthesized by the soil bacterium Eleftheria terrae ( Ling et al . , 2015 ) . A particularly interesting property of this potential therapeutic is the low rate at which resistance is evolved . The apparent challenge of structurally modifying lipid II in order to subvert antimicrobials may explain why interbacterial toxins targeting this molecule have evolved independently in Gram-negative ( colicin M ) and -positive ( TelC ) bacteria ( El Ghachi et al . , 2006 ) . We anticipate that biochemical characterization of additional LXG toxins of unknown function will reveal further Gram-positive cell vulnerabilities that could likewise be exploited in the design of new antibiotics .
S . intermedius strains used in this study were derived from the sequenced strains ATCC 27335 and B196 ( Supplementary file 1 ) . S . intermedius strains were grown at 37°C in the presence of 5% CO2 in Todd Hewitt broth ( THYB ) or agar ( THYA ) supplemented with 0 . 5% yeast extract . When needed , media contained spectinomycin ( 75 μg/mL ) or kanamycin ( 250 μg/mL ) . S . aureus USA300 derived strains were grown at 37°C in tryptic soy broth ( TSB ) or agar ( TSA ) supplemented with chloramphenicol ( 10 μg/mL ) and xylose ( 2% w/v ) when needed . E . faecalis OG1RF and S . pyogenes 5005 were grown at 37°C on Brain Heart Infusion ( BHI ) media . P . aeruginosa PAO1 and B . thailandensis E264 were grown at 37°C on THYA . B . fragilis NCTC9343 was grown anaerobically at 37°C on Brain Heart Infusion-supplemented ( BHIS ) media . E . coli strains used in this study included DH5α for plasmid maintenance , BL21 for protein expression and toxicity assays and MG1655 for competition experiments . E . coli strains were grown on LB medium supplemented with 150 μg/mL carbenicillin , 50 μg/mL kanamycin , 200 μg/mL trimethoprim , 75 μg/mL spectinomycin , 200 μM IPTG or 0 . 1% ( w/v ) rhamnose as needed . For co-culture experiments with S . intermedius strains , E . coli , B . thailandensis , P . aeruginosa , S . aureus , E . faecalis , S . pyogenes were grown on THYA . BHIS agar supplemented with sheep’s blood was used when B . fragilis was grown in co-culture with S . intermedius . S . cerevisiae BY4742 was grown on Synthetic Complete -uracil ( SC-ura ) medium at 30°C . S . intermedius mutants were generated by replacing the gene to be deleted with a cassette conferring resistance to spectinomycin ( derived from pDL277 ) or kanamycin ( derived from pBAV1K-T5 ) , as previously described ( Tomoyasu et al . , 2010 ) . Briefly , the antibiotic resistance cassette was cloned between ~800 bp of sequence homologous to the regions flanking the gene to be deleted . The DNA fragment containing the cassette and flanking sequences was then linearized by restriction digest , gel purified , and ~250 ng of the purified fragment was added to 2 mL of log-phase culture pre-treated for two hours with competence peptide ( 200 ng/ml ) to stimulate natural transformation . Cultures were further grown for four hours before plating on the appropriate antibiotic . All deletions were confirmed by PCR . All DNA manipulation procedures followed standard molecular biology protocols . Primers were synthesized and purified by Integrated DNA Technologies ( IDT ) . Phusion polymerase , restriction enzymes and T4 DNA ligase were obtained from New England Biolabs ( NEB ) . DNA sequencing was performed by Genewiz Incorporated . A comprehensive list of all clade names in the Firmicutes phylum was obtained from the List of Prokaryotic names with Standing in Nomenclature ( http://www . bacterio . net/; updated 2017-02-02 ) , a database that compiles comprehensive journal citations for every characterized prokaryotic species ( Euzéby , 1997 ) . This list was then compared with results obtained from a manually curated Jackhmmer search and LXG-containing Firmicutes were tabulated at the order , family , and genus levels ( Finn et al . , 2015; Mitchell et al . , 2015 ) . These results were binned into three categories based on the number of sequenced species and then further differentiated by the number of LXG-positive species within each genus . For species belonging to orders containing no predicted LXG encoding genes , the number of genera examined was tabulated and included in the dendogram . The 240 nucleotide tags from the toxin domains were mapped using blastn to the Integrated Gene Catalog ( Li et al . , 2014 ) – a large dataset of previously identified microbiome genes and their abundances in several extensive microbiome studies ( including HMP [Human Microbiome Project Consortium , 2012] , MetaHiT [Qin et al . , 2010] , and a T2D Chinese cohort [Qin et al . , 2012] ) . Genes to which at least one tag was mapped with >95% identity and >50% overlap were labeled as LXG genes . This set of LXG genes was further manually curated to filter out genes that lack the LXG targeting domain . In analyzing the relative abundance of the LXG genes across samples , relative abundances < 10−7 were assumed to represent noise and were set to 0 . LXG genes that were not present above this threshold in any sample and samples with no LXG genes were excluded from the analysis . Measurement of cellular NAD+ levels was performed as reported previously ( Whitney et al . , 2015 ) . Briefly , E . coli strains harboring expression plasmids for Tse2 , Tse6tox , TelBtox* , TelBtoxR626A , TelBtox*–TipB and a vector control were grown in LB media at 37°C to mid-log phase prior to induction of protein expression with 0 . 1% ( w/v ) rhamnose . 1 hr post-induction , cultures were diluted to OD600 = 0 . 5 and 500 μL of cells were harvested by microcentrifugation . Cells were then lysed in 0 . 2 M NaOH , 1% ( w/v ) cetyltrimethylammonium bromide ( CTAB ) followed by treatment with 0 . 4 M HCl at 60°C for 15 min . After neutralization with 0 . 5 M Tris base , samples were then mixed with an equal volume of NAD/NADH-Glo Detection Reagent ( Promega ) prepared immediately before use as per the instructions of the manufacturer . Luciferin bioluminescence was measured continuously using a Synergy H1 plate reader . The slope of the luciferin signal from the linear range of the assay was used to determine relative NAD+ concentration compared to a vector control strain . S . intermedius strains were grown to late-log phase before cells were removed by centrifugation at 3000 g for 15 min . Residual particulates were removed by vacuum filtration through a 0 . 2 um membrane and the resulting supernatants were concentrated 100-fold by spin filtration ( 30 kDa MWCO ) . NADase assays were carried out by mixing 50 μL of concentrated supernatant with 50 μL of PBS containing 2 mM NAD+ followed by incubation at room temperature for 2 hr . Reactions were terminated by the addition of 50 μL of 6M NaOH and incubated in the dark at room temperature for 15 min . Samples were analyzed by UV light at a wavelength of 254 nm and imaged using a FluorChemQ ( ProteinSimple ) . Relative NAD+ consumption was determined using densitometry analysis of each of the indicated strain supernatants using the ImageJ software program ( https://imagej . nih . gov/ij/ ) . To assess TelA and TelB toxicity in bacteria , stationary phase cultures of E . coli BL21 pLysS harboring the appropriate plasmids were diluted 106 and each 10-fold dilution was spotted onto 3% LB agar plates containing the appropriate antibiotics . 0 . 1% ( w/v ) L-rhamnose and 100 μM IPTG were added to the media to induce expression of toxin and immunity genes , respectively . For TelB , plasmids containing the wild-type toxin domain ( under non-inducing conditions ) were not tolerated . To circumvent this , SOE pcr was used to assemble a variant ( H661A ) that was tolerated under non-induced conditions . Based on the similarity of TelBtox to M . tuberculosis TNT toxin , this mutation likely reduces the binding affinity of TelB to NAD+ ( Sun et al . , 2015 ) . To generate a TelB variant that exhibited significantly reduced toxicity under inducing conditions , a second mutation ( R626A ) was introduced in the toxin domain of TelB . For examination of TelC toxicity in S . intermedius , the gene fragment encoding TelCtox was fused to the constitutive P96 promoter followed by a start codon and cloned into pDL277 ( Lo Sapio et al . , 2012 ) . For extracellular targeting of TelCtox in S . intermedius , the gene fragment encoding the sec-secretion signal ( residues 1–30 ) of S . pneumoniae LysM ( SP_0107 ) was fused to the 5’ end of telCtox , each of the telCtox site-specific variants and the telCtox–tipC bicistron . 500 ng of each plasmid was transformed in S . intermedius B196 and toxicity was assessed by counting the number of transformants . For examination of TelC toxicity S . aureus , the gene fragment encoding TelCtox was cloned into the xylose-inducible expression vector pEPSA5 . For extracellular targeting , the gene fragment encoding the sec-secretion signal for hla was fused to the 5’ end of telCtox and telCtoxD401A . TelC-based toxicity was assessed in the same manner as was done for the above E . coli toxicity experiments except that xylose ( 2% w/v ) was included in the media to induce protein expression . Detailed plasmid information can be found in Supplementary file 2 . S . aureus USA300 pEPSA5::ss-telC202-CT and S . aureus USA300 pEPSA5 were resuspended in TSB and 1–2 μL of each suspension was spotted onto an 1% ( w/v ) agarose pad containing typtic soy medium supplemented with 2% ( w/v ) xylose and sealed . Microscopy data were acquired using NIS Elements ( Nikon ) acquisition software on a Nikon Ti-E inverted microscope with a 60× oil objective , automated focusing ( Perfect Focus System , Nikon ) , a xenon light source ( Sutter Instruments ) , and a CCD camera ( Clara series , Andor ) . Time-lapse sequences were acquired at 10 min intervals over 12 hr at room temperature . Movie files included are representative of three biological replicates for each experiment . 200 mL cultures of S . intermedius B196 wild-type and ΔessC strains were grown to stationary phase in THYB before being pelleted by centrifugation at 2500 × g for 20 min at 4°C . Supernatant fractions containing secreted proteins were collected and spun at 2500 × g for an additional 20 min at 4°C and subsequently filtered through a 0 . 2 μm pore size membrane to remove residual cells and cell debris . Protease inhibitors ( 1 mM AEBSF , 10 mM leupeptin , and 1 mM pepstatin ) were added to the filtered supernatants prior to dialysis in 4L of PBS using 10 kDa molecular weight cut off tubing at 4°C . After four dialysis buffer changes , the retained proteins were TCA precipitated , pelleted , washed in acetone , dried and resuspended in 1 mL of 100 mM ammonium bicarbonate containing 8 M urea . The denatured protein mixture was then desalted over a PD10 column prior to reduction , alkylation and trypsin digestion as described previously ( Eshraghi et al . , 2016 ) . The resulting tryptic peptides were desalted and purified using C18 spin columns ( Pierce ) following the protocol of the manufacturer before being vacuum dried and resuspended in 10 µL of acetonitrile/H2O/formic acid ( 5/94 . 9/0 . 1 , v/v/v ) for LC-MS/MS analysis . Peptides were analyzed by LC-MS/MS using a Dionex UltiMate 3000 Rapid Separation nanoLC and a linear ion trap – Orbitrap hybrid mass spectrometer ( ThermoFisher Scientific ) . Peptide samples were loaded onto the trap column , which was 150 µm x 3 cm in-house packed with 3 µm C18 beads , at flow rate of 5 µL/min for 5 min using a loading buffer of acetonitrile/H2O/formic acid ( 5/94 . 9/0 . 1 , v/v/v ) . The analytical column was a 75 µm x 10 . 5 cm PicoChip column packed with 1 . 9 µm C18 beads ( New Objectives ) . The flow rate was kept at 300 nL/min . Solvent A was 0 . 1% formic acid in water and Solvent B was 0 . 1% formic acid in acetonitrile . The peptide was separated on a 90 min analytical gradient from 5% acetonitrile/0 . 1% formic acid to 40% acetonitrile/0 . 1% formic acid . The mass spectrometer was operated in data-dependent mode . The source voltage was 2 . 10 kV and the capillary temperature was 275°C . MS1 scans were acquired from 400 to 2000 m/z at 60 , 000 resolving power and automatic gain control ( AGC ) set to 1 × 106 . The top ten most abundant precursor ions in each MS1 scan were selected for fragmentation . Precursors were selected with an isolation width of 1 Da and fragmented by collision-induced dissociation ( CID ) at 35% normalized collision energy in the ion trap . Previously selected ions were dynamically excluded from re-selection for 60 s . The MS2 AGC was set to 3 × 105 . Proteins were identified from the MS raw files using Mascot search engine ( Matrix Science ) . MS/MS spectra were searched against the UniprotKB database of S . intermedius B196 ( UniProt and UniProt Consortium , 2015 ) . All searches included carbamidomethyl cysteine as a fixed modification and oxidized Met , deamidated Asn and Gln , acetylated N-terminus as variable modifications . Three missed tryptic cleavages were allowed . The MS1 precursor mass tolerance was set to 10 ppm and the MS2 tolerance was set to 0 . 6 Da . A 1% false discovery rate cutoff was applied at the peptide level . Only proteins with a minimum of two unique peptides above the cutoff were considered for further study . MS/MS spectral counts were extracted by Scaffold 4 ( Proteome Software Inc . ) and used for statistical analysis of differential expression . Three biological replicates were performed and proteins identified in all three wild-type replicates were included in further analysis . After replicate averaging , low abundance proteins ( less than five spectral counts in wild-type ) were excluded from the final dataset . Overnight cultures of S . intermedius strains were used to inoculate 2 ml of THYB at a ratio of 1:200 . Cultures were grown statically at 37°C , 5% CO2 to mid-log phase , and cell and supernatant fractions were prepared as described previously ( Hood et al . , 2010 ) . Full-length TelC protein was expressed and purified as described below ( see protein expression and purification ) except that PBS buffer was used instead of Tris-HCl for all stages of purification . Ten milligrams of purified TelC protein was sent to GenScript for polyclonal antisera production . Western blot analyses of protein samples were performed using rabbit α-TelC ( diluted 1:2000 ) or rabbit α-VSV-G ( diluted 1:5000 , Sigma ) and detected with α-rabbit horseradish peroxidase-conjugated secondary antibodies ( diluted 1:5000 , Sigma ) . Western blots were developed using chemiluminescent substrate ( SuperSignal West Pico Substrate , Thermo Scientific ) and imaged with a FluorChemQ ( ProteinSimple ) . For intraspecific competition experiments donor and recipient strains were diluted in THYB to a starting OD600 of 0 . 5 and 0 . 05 , respectively . Cell suspensions were then mixed together in a 1:1 ratio and 10 μL of the mixture was spotted on THYA and grown at 37°C , 5% CO2 for 30 hr . The starting ratio of each competition was determined by enumerating donor and recipient c . f . u . Competitions were harvested by excising the agar surrounding the spot of cell growth followed by resuspension of cells in 0 . 5 mL of THYB . The final donor and recipient ratio was determined by enumerating c . f . u . For all intraspecific experiments , counts of donor and recipient c . f . u . were obtained by dilution plating on THYA containing appropriate antibiotics . To facilitate c . f . u . enumeration of wild-type S . intermedius B196 , a spectinomycin resistance cassette was inserted into the intergenic region between SIR_0114 and SIR_0115 . For interspecies competition experiments , donor and recipient strains were diluted in THYB to a starting OD600 of 0 . 75 and 0 . 00075 , respectively . Cell suspensions were then mixed together in a 1:1 ratio and 10 μL of the mixture was spotted on THYA and grown at 37°C , 5% CO2 for 30 hr . The starting ratio of each competition was determined by enumerating donor and recipient c . f . u . Competitions were harvested by excising the agar surrounding the spot of cell growth followed by resuspension of cells in 0 . 5 mL of THYB . The final donor and recipient ratio was determined by enumerating c . f . u . Counts of donor and recipient c . f . u . were obtained by dilution plating on THYA containing appropriate antibiotics ( S . intermedius ) , BHI under standard atmospheric conditions ( E . coli , E . faecalis and S . pyogenes ) , LB under standard atmospheric conditions ( P . aeruginosa and B . thailandensis ) or BHIS supplemented with 60 μg/mL gentamicin under anaerobic conditions ( B . fragilis ) . Statistically significance was assessed for bacterial competition experiments through pairwise t-tests of competitive index values ( n = 3 for each condition ) . Stationary phase overnight cultures of E . coli BL21 pETDuet-1::telC , E . coli BL21 pETDuet-1::telC202-CT ( encoding TelCtox ) and E . coli BL21 pETDuet-1::tipCΔss were used to inoculate 4L of 2 x YT broth and cultures were grown to mid-log phase in a shaking incubator at 37°C . Upon reaching an OD600 of approximately 0 . 6 , protein expression was induced by the addition of 1 mM IPTG followed by incubation at 18°C for 16 hr . Cells were harvested by centrifugation at 6000 g for 15 min , followed by resuspension in 35 mL of buffer A ( 50 mM Tris-HCl pH 8 . 0 , 300 mM NaCl , 10 mM imidazole ) . Resuspended cells were then ruptured by sonication ( 3 pulses , 50 s each ) and cellular debris was removed by centrifugation at 30 , 000 g for 45 min . Cleared cell lysates were then purified by nickel affinity chromatography using 2 mL of Ni-NTA agarose resin loaded onto a gravity flow column . Lysate was loaded onto the column and unbound proteins were removed using 50 mL of buffer A . Bound proteins were then eluted using 50 mM Tris-HCl pH 8 . 0 , 300 mM NaCl , 400 mM imidazole . The purity of each protein sample was assessed by SDS-PAGE followed by Coomassie Brilliant Blue staining . All protein samples were dialyzed into 20 mM Tris-HCl , 150 mM NaCl . Selenomethionine-incorporated TelC202-CT was obtained by growing E . coli BL21 pETDuet-1::telC202-CT in SelenoMethionine Medium Complete ( Molecular Dimensions ) using the expression conditions described above . Cell lysis and nickel affinity purification were also performed as described above except that all buffers contained 1 mM tris ( 2-carboxyethyl ) phosphine . Purified selenomethionine-incorporated TelC202-CT was concentrated to 12 mg/mL by spin filtration ( 10 kDa cutoff , Millipore ) and screened against commercially available crystallization screens ( MCSG screens 1–4 , Microlytic ) . Diffraction quality crystals appeared after 4 days in a solution containing 0 . 1 M Sodium Acetate pH 4 . 6 , 0 . 1 M CaCl2 , 30% PEG400 . X-ray diffraction data were collected using beamline 5 . 0 . 2 at the Advanced Light Source ( ALS ) . A single dataset ( 720 images , 1 . 0° Δφ oscillation , 1 . 0 s exposure ) was collected on an ADSC Q315r CCD detector with a 200 mm crystal-to-detector distance . Data were indexed and integrated using XDS ( Kabsch , 2010 ) and scaled using AIMLESS ( Evans and Murshudov , 2013 ) ( table S2 ) . The structure of TelC202-CT was solved by Se-SAD using the AutoSol wizard in the Phenix GUI ( Adams et al . , 2010 ) . Model building was performed using the AutoBuild wizard in the Phenix GUI . The electron density allowed for near-complete building of the model except for N-terminal residues 202–211 , two C-terminal residues and an internal segment spanning residues 417–434 . Minor model adjustments were made manually in COOT between iterative rounds of refinement , which was carried out using Phenix . refine ( Afonine et al . , 2012; Emsley et al . , 2010 ) . The progress of the refinement was monitored by the reduction of Rwork and Rfree ( Table 2 ) . Purified TelCtox was dialyzed against 20 mM sodium acetate pH 4 . 6 , 150 mM NaCl , 10 mM CaCl2 . Cross-linked peptidoglycan sacculi and lysostaphin endopeptidase pre-treated ( non-cross-linked ) sacculi from S . aureus were then incubated with 5 μM TelCtox , 2 . 5 μg of cellosyl muramidase or buffer at 37°C for 18 hr . Digests were then boiled for 5 min at 100°C and precipitated protein was removed by centrifugation . The resulting muropeptides were reduced by the addition of sodium borohydride and analyzed by HPLC as described previously ( de Jonge et al . , 1992 ) . For the analysis of cell walls isolated from TelC-intoxicated cells , 1L of S . aureus USA300 pEPSA5::ss-telCtox and S . aureus USA300 pEPSA5::ss-telCtoxD401A cells were grown to mid-log phase prior to induction of protein expression by the addition of 2% ( w/v ) xylose . 90 min post-induction , cultures were rapidly cooled in an ice-water bath and cells were harvested by centrifugation . After removal of supernatants , cell pellets were resuspended in 40 mL of ice-cold 50 mM Tris-HCl pH 7 . 0 and subsequently added dropwise to 120 mL boiling solutions of 5% SDS . PG was isolated as described ( de Jonge et al . , 1992 ) and digested with either cellosyl muramidase or lysostaphin endopeptidase and cellosyl , reduced with sodium borohydride and analyzed by HPLC as described above . Purified TelCtox and TelCtox–TipCΔss complex were dialyzed against 20 mM sodium acetate pH 4 . 6 , 150 mM NaCl , 10 mM CaCl2 . C14-labelled Lys-type lipid II was solubilized in 5 μL of Triton X-100 before being added to 95 μL of reaction buffer containing 15 mM HEPES pH 7 . 5 , 0 . 4 mM CaCl2 ( excluded from the PBP1B-LpoB reaction ) , 150 mM NaCl , 0 . 023% Triton X-100 and either PBP1B–LpoB complex , TelCtox , TelCtox–TipCΔss complex or Colicin M followed by incubation for 1 hr at 37°C . The reaction with PBP1B-LpoB was boiled and reduced with sodium borohydride . All the reactions were quenched by the addition of 1% ( v/v ) phosphoric acid and analyzed by HPLC as described ( Bertsche et al . , 2005 ) . Three biological replicates were performed for each reaction . The lipid II degradation products of TelCtox digestion were confirmed by mass spectrometry . Lipid II was kindly provided by Ute Bertsche and was generated as described previously ( Bertsche et al . , 2005 ) . For thin-layer chromatography ( TLC ) analysis of Lys-type lipid II degradation by TelC , TelCtox or TelCtox–TipCΔss , 40 μM lipid II was solubilized in 30 mM HEPES/KOH pH 7 . 5 , 150 mM KCl and 0 . 1% Triton X-100 before adding either 2 μM TelCtox , 2 μM TelCtox–TipCΔss complex or protein buffer , followed by incubation for 90 min at 37°C . Samples were extracted with n-butanol/pyridine acetate ( 2:1 ) pH 4 . 2 and resolved on silica gel ( HPTLC silica gel 60 , Millipore ) in chloroform/methanol/ammonia/water ( 88:48:1:10 ) . For the undecaprenyl phosphate reactions 100 μM undecaprenyl phosphate ( Larodan ) was solubilized in 20 mM HEPES/KOH pH 7 . 5 , 150 mM KCl , 1 mM CaCl2 and 0 . 1% Triton X-100 before adding 2 μM TelCtox ( final concentration ) , 2 μM TelCtox–TipCΔss or protein buffer , followed by incubation for 5 hr at 25°C and 90 min at 37°C . Samples were extracted and separated by TLC as indicated above . For the undecaprenyl pyrophosphate synthesis reactions coupled to the degradation by TelCtox0 . 04 mM Farnesyl pyrophosphate and 0 . 4 mM isopentenyl pyrophosphate were solubilized in 20 mM HEPES/KOH pH 7 . 5 , 50 mM KCl , 0 . 5 mM MgCl2 , 1 mM CaCl2 , 0 . 1% Triton X-100 and incubated with 10 μM UppS and 2 μM TelCtox ( final concentrations ) or protein buffer for 5 hr at 25°C and 90 min at 37°C . Samples were extracted and separated by TLC as indicated above . To target TelC to the yeast secretory pathway , telCtox was fused to the gene fragment encoding the leader peptide of Kluyveromyces lactis killer toxin ( Baldari et al . , 1987 ) , generating ss-telCtox . S . cerevisiae was transformed with pCM190 containing telCtox , ss-telCtox , a known toxin of yeast or empty vector and grown o/n SC-ura +1 ug/mL doxycycline . Cultures were resuspended to OD600 = 1 . 5 with water and serially diluted 5-fold onto SC-ura agar . Plates were incubated at 30°C for 2 days before being imaged using a Pentax WG-3 digital camera . Images presented are representative of three independent replicate experiments . Proteolytic processing of the leader peptide of ss-TelCtox was confirmed by western blot . Solutions of 25 μM TelC202-CT and 250 μM TipCΔss were degassed prior to experimentation . ITC measurements were performed with a VP-ITC microcalorimeter ( MicroCal Inc . , Northampton , MA ) . Titrations consisted of 25 10 μL injections with 180 s intervals between each injection . The ITC data were analyzed using the Origin software package ( version 5 . 0 , MicroCal , Inc . ) and fit using a single-site binding model . E . coli BTH101 cells were co-transformed with plasmids encoding the T18 and T25 fragments of Bordetella pertussis adenylate cyclase fused to the proteins of interest . Stationary phase cells were then plated on LB agar containing 40 mg/mL X-gal , 0 . 5 mM IPTG , 50 mg/mL kanamycin and 150 mg/mL carbenicillin and grown for 24 hr at 30°C . Plates were imaged using a Pentax WG-3 digital camera . Images representative of at least three independent replicate experiments are presented . E . coli BL21 ( DE3 ) pLysS cells were co-transformed with plasmids encoding TelC-his6 and WxgB-V or TelC-his6 and WxgC-V . Cells were grown to an OD600 of 0 . 6 prior to induction of protein expression with 0 . 5 mM IPTG for 6 hr at 30°C . Cultures were harvested by centrifugation and cell pellets were resuspended in Buffer A prior to lysis by sonication . Clarified lysates were then incubated with Ni-NTA resin and incubated at 4°C with rotation for 90 min . Ni-NTA resin was then washed four times with Buffer A followed by elution of bound proteins with Buffer B . After the addition of Laemmli sample buffer , proteins were separated by SDS-PAGE using an 8–16% gradient TGX Stain-Free gel ( Bio-Rad ) . TelC-his6 was visualized by UV activation the trihalo compound present in Stain-Free gels whereas WxgB-V and WxgC-V were detected by western blotting . | Most bacteria live in densely colonized environments , such as the human gut , in which they must constantly compete with other microbes for space and nutrients . As a result , bacteria have evolved a wide array of strategies to directly fight their neighbors . For example , some bacteria release antimicrobial compounds into their surroundings , while others ‘inject’ protein toxins directly into adjacent cells . Bacteria can be classified into two groups known as Gram-positive and Gram-negative . Previous studies found that Gram-negative bacteria inject toxins into neighboring cells , but no comparable toxins in Gram-positive bacteria had been identified . Before a bacterium can inject molecules into an adjacent cell , it needs to move the toxins from its interior to the cell surface . It had been suggested that a transport system in Gram-positive bacteria called the Esx pathway may export toxins known as LXG proteins . However , it was not clear whether these proteins help Gram-positive bacteria to compete against other bacteria . Whitney et al . studied the LXG proteins in Gram-positive bacteria known as Firmicutes . The experiments reveal that Firmicutes found in the human gut possess LXG genes . A Firmicute known as Streptococcus intermedius produces three LXG proteins that are all toxic to bacteria . To avoid being harmed by its own LXG proteins , S . intermedius also produces matching antidote proteins . Further experiments show that LXG proteins are exported out of S . intermedius cells and into adjacent competitor bacteria by the Esx pathway . Examining one of these LXG proteins in more detail showed that it can degrade a molecule that bacteria need to make their cell wall . Together , these findings suggest that LXG proteins may influence the species living in many important microbial communities , including the human gut . Changes in the communities of gut microbes have been linked with many diseases . Therefore , understanding more about how the LXG proteins work may help us to develop ways to manipulate these communities to improve human health . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"microbiology",
"and",
"infectious",
"disease"
] | 2017 | A broadly distributed toxin family mediates contact-dependent antagonism between gram-positive bacteria |
Alternative splicing of mRNA precursors represents a key gene expression regulatory step and permits the generation of distinct protein products with diverse functions . In a genome-scale expression screen for inducers of the epithelial-to-mesenchymal transition ( EMT ) , we found a striking enrichment of RNA-binding proteins . We validated that QKI and RBFOX1 were necessary and sufficient to induce an intermediate mesenchymal cell state and increased tumorigenicity . Using RNA-seq and eCLIP analysis , we found that QKI and RBFOX1 coordinately regulated the splicing and function of the actin-binding protein FLNB , which plays a causal role in the regulation of EMT . Specifically , the skipping of FLNB exon 30 induced EMT by releasing the FOXC1 transcription factor . Moreover , skipping of FLNB exon 30 is strongly associated with EMT gene signatures in basal-like breast cancer patient samples . These observations identify a specific dysregulation of splicing , which regulates tumor cell plasticity and is frequently observed in human cancer .
Alternative splicing ( AS ) of mRNA precursors is a fundamental biological process that provides a reversible mechanism to modulate the expression of related but distinct proteins in response to internal and external stimuli ( Chen and Manley , 2009 ) . Regulation of alternative splicing occurs at several levels including the expression and the targeting of specific RNA-binding proteins ( RBPs ) . Dysregulation of alternative splicing plays a direct role in a variety of human diseases including cancer ( David and Manley , 2010 ) . During cancer initiation and progression , the epithelial-to-mesenchymal transition ( EMT ) triggers the dissociation and migration of carcinoma cells from primary to distant sites ( Ye and Weinberg , 2015 ) . We previously demonstrated that the EMT is also tightly linked to a stem-like cell state in breast cancer , as overexpression of EMT transcription factors induces the expression of tumor-initiating cell markers and increases the ability of cells to form mammospheres , a property associated with mammary epithelial stem cells ( Chaffer et al . , 2013; Mani et al . , 2008 ) . In addition , the EMT has been implicated in several other cancer-related phenotypes , for example , in cancers that acquired resistance either to the EGFR inhibitor gefitinib or to the HER2 receptor inhibitor trastuzumab ( Boulbes et al . , 2015; Sequist et al . , 2011 ) . EMT also involves a dramatic reorganization of the actin cytoskeleton and concomitant formation of membrane protrusions to gain migratory and invasive properties ( Yilmaz and Christofori , 2009 ) . The dynamic change in the actin cytoskeleton , a prerequisite for cell motility and cancer cell invasion , is a highly controlled equilibrium of local assembly and disassembly of actin filaments ( Yilmaz and Christofori , 2009 ) . The filamin family proteins crosslink actin filaments and are also translocated into the nucleus to regulate the transcriptional activity of the androgen receptor and the FOXC1 transcription factor ( Bedolla et al . , 2009; Berry et al . , 2005; Loy et al . , 2003; Zhou et al . , 2010 ) . The three members of this family ( FLNA , FLNB and FLNC ) are involved in both development and normal tissue homeostasis through regulating diverse processes including cell locomotion and integrin signaling ( Zhou et al . , 2010 ) , and mutations in the FLNB gene cause a broad range of skeletal dysplasias ( Daniel et al . , 2012 ) . Alternative splicing has been previously associated with EMT . Mesenchymal cancer cells show distinct alternative splicing patterns in comparison with their epithelial counterparts ( Braeutigam et al . , 2014; Shapiro et al . , 2011; Venables et al . , 2013 ) . While ESRP1 and ESRP2 are epithelial state-inducing RBPs that govern splicing patterns for the epithelial cell state ( Shapiro et al . , 2011; Warzecha et al . , 2010; Warzecha et al . , 2009; Yang et al . , 2016 ) , less is known about the identity and functional significance of RBPs that can promote the mesenchymal cell state . QKI and RBFOX2 have been shown to be responsible for alternative splicing events that occur during EMT , such as exon skipping in KIF13A and CTTN ( Braeutigam et al . , 2014; Venables et al . , 2013; Yang et al . , 2016 ) and in circular RNA formation ( Conn et al . , 2015 ) . Nevertheless , it remains unclear whether the upregulation of any specific RBPs is sufficient or required for the induction of mesenchymal state transitions or is merely one of many downstream manifestations of the EMT . Furthermore , although many splicing changes occur during EMT , only a small number of specific splicing events are known to functionally contribute to EMT including changes in the splicing of CD44 , FGFR2 and Exo70 ( Brown et al . , 2011; Lu et al . , 2013; Warzecha et al . , 2009 ) . Here , we have undertaken a comprehensive approach to identify genes that regulate the EMT in breast cancer and found that genes whose protein products participate in AS regulate the transition to mesenchymal- and stem-like cell states .
In prior work , we described a genetically defined , experimental model of breast cancer , derived from introducing vectors expressing the telomerase catalytic subunit , the SV40 large-T and small-t antigens , and an H-Ras oncoprotein into human mammary epithelial cells ( HMLER cells ) ( Elenbaas et al . , 2001 ) . Subsequent work demonstrated that the CD44 cell surface antigen is a surrogate marker for the EMT cell state change in this model ( Chaffer et al . , 2011; Chaffer et al . , 2013 ) . Thus , we separated the CD44-high and -low populations of HMLER cells by fluorescence-activated cell sorting ( FACS ) and confirmed that the CD44-low cells displayed epithelial properties , as measured by levels of EMT marker expression ( Figure 1—figure supplement 1A ) . The highly purified CD44-low cell population remained in the epithelial cell state for at least 4 weeks in the experimental conditions . In contrast , the CD44-high HMLER cells showed elevated expression of mesenchymal markers and a greater propensity to form mammospheres , an in vitro surrogate assay for the stemness of mammary epithelial cells ( Figure 1—figure supplement 1B , C ) . To study inducers of the EMT and stem-like cell state , we performed a genome scale open-reading frame ( ORF ) screen to identify genes that convert the HMLER cells from the CD44-low state to the CD44-high state . Each ORF in the human ORFeome library collection 8 . 1 ( Yang et al . , 2011 ) was tagged with a unique 24-nucleotide barcode and introduced into FACS purified CD44-low HMLER cells by lentiviral-mediated gene transfer . Following 7 days in culture , we purified the newly arising CD44-high HMLER cells by FACS and identified ORFs enriched in these cells by massively parallel sequencing ( Figure 1A ) . We found that the consistency between the biological replicates of the screen was high ( Figure 1—figure supplement 2A ) . SNAI1 , a well-characterized EMT-inducing transcription factor ( EMT-TF ) ( Nieto et al . , 2016 ) , scored as the top hit in the screen , as did BCL6 , JMJD6 and FOS , which have previously been shown to play key roles in regulating EMT ( Aprelikova et al . , 2016; Eger et al . , 2000; Yu et al . , 2015 ) ( Figure 1B , Figure 1—figure supplement 2B and Figure 1—source data 1 ) ; these findings indicated that this screen was robust . We used a cut-off of three standard deviations ( S . D . ) above the mean and analyzed the top-scoring candidates to identify protein complexes or pathways enriched for regulators of EMT . Sixty-eight ORFs met this criterion ( Figure 1B and Figure 1—figure supplement 2B ) , including transcription factors , RNA splicing factors , kinases and phosphatases , epigenetic regulators , and genes involved in the regulation of spermatogenesis , apoptosis and the metabolic processing of cellular amides ( Figure 1C ) . Other EMT transcription factors did not meet the 3 S . D . cutoff possibly due to mutations in the ORF constructs or low ORF representation in the library . Using the GeNets analysis tool ( apps . broadinstitute . org/genets ) , we found three gene networks centered around QKI ( Quaking , an RNA-binding protein ) , SRPK2 ( a kinase involved in RNA splicing ) and PPP1CC ( a phosphatase ) ( Figure 1D ) . When we used the top candidates to interrogate gene ontology , we found that the ‘regulation of mRNA metabolic process’ and ‘regulation of mRNA splicing , via spliceosome’ scored as the top terms ( Figure 1E ) and that ‘RNA metabolic process’ was one of the top gene sets enriched by gene set enrichment analysis ( GSEA ) ( Figure 1F ) ( Reich et al . , 2006 ) . Of note , ‘Regulation of mRNA metabolic process’ is a parent GO term for RNA processing and RNA splicing . Several RNA-binding proteins have been previously associated with EMT . For example , ESRP1 and 2 promote an epithelial phenotype , while QKI and RBFOX2 ( a homolog of RBFOX1 that scored in the screen ) regulate a number of EMT-associated splicing events ( Braeutigam et al . , 2014; Venables et al . , 2013; Yang et al . , 2016 ) . Of note , although RBFOX2 has been shown to play a role in EMT ( Braeutigam et al . , 2014; Venables et al . , 2013 ) , the RBFOX2 clone present in the ORFeome collection 8 . 1 library harbors three mutations ( a 396–449 deletion , a 752–763 deletion and a C to T substitution at 1007 ) , which likely explained why this ORF did not score in the screen . However , whether the expression of any RBPs is functionally sufficient or required to induce a mesenchymal cell state remains unclear . Since we found a striking enrichment of RBPs in this screen , we focused on the top candidates implicated in pre-mRNA splicing to understand their possible role in regulating the EMT and stem-like cell states in breast cancer pathogenesis . In the ORF expression screen , we identified eight candidate RBPs ( QKI , RBFOX1 , MBNL1 , MBNL2 , CELF4 , SFPQ , SRSF9 and HNRNPUL1 ) that scored when tested individually . We systematically tested these genes in five assays to examine EMT-associated phenotypes or marker expression to find the RBPs that meet the following criteria: ( 1 ) Expression of the RBP promotes an increase in the CD44-high population ( Figure 2—figure supplement 1A ) ; ( 2 ) Expression of the RBP upregulates the expression of a panel of mesenchymal markers examined by both quantitative PCR ( Figure 2A and Figure 2—figure supplement 1B ) and immunoblotting ( Figure 2B ) ; ( 3 ) Expression of the RBP induces mammosphere formation when cells are grown in suspension , a characteristic of the stem-like and mesenchymal cell properties ( Figure 2C and Figure 2—figure supplement 1C ) ( Chaffer et al . , 2013; Mani et al . , 2008 ) ; ( 4 ) Endogenous expression of the RBP is upregulated upon overexpression of an EMT-inducing transcription factor , SNAI1 or ZEB1 ( Figure 2—figure supplement 2A–D ) ; ( 5 ) Expression of the RBP promotes tumor formation in vivo , a feature associated with stem-like cells ( Figure 2D andFigure 2—figure supplement 2; Figure 2—figure supplement 2E ) ( Chaffer et al . , 2013; Mani et al . , 2008 ) . Together , we discovered that the expression of QKI ( NCBI Reference: NM_006775 . 2 , also known as QKI-5 ) and RBFOX1 ( NCBI Reference: NM_145893 . 2 , also known as RBFOX1 beta ) strongly induced the mesenchymal and stem-like phenotypes in all the experiments tested , while MBNL1 , MBNL2 and CELF4 scored in some assays . We also found that SRSF9 , SFPQ and HNRNPUL1 are unlikely to initiate a mesenchymal and stem-like cell state ( Figure 2J ) . The CD44-high cells induced by QKI , RBFOX1 or SNAI1 shared a similar elongated and spindle shape cell morphology ( Figure 2—figure supplement 3A ) . In addition , QKI and RBFOX1 overexpression also significantly increased the CD44-high cell populations in two additional breast cancer cell lines ( MCF7 and ZR75-1 ) ( Figure 2—figure supplement 3B ) . We thus proceeded to focus on the role of QKI and RBFOX1 in EMT . Of note , overexpression of QKI and RBFOX1 reduced cell proliferation by 40% to 45% as would be expected if the cells undergo an EMT ( Figure 2—figure supplement 3C ) ( Tsai et al . , 2012; Vega et al . , 2004 ) . In addition , the expression of QKI , RBFOX1 and other RBPs failed to decrease the expression of epithelial markers ( Figure 2A , B and Figure 2—figure supplement 1B ) , suggesting that the cell state triggered by expression of the RBPs involves elevated expression of mesenchymal markers with retention of pre-existing epithelial marker expression . This spectrum of marker expression is reminiscent of an intermediate EMT state that is implicated in development and tumor progression ( Bierie et al . , 2017; George et al . , 2017; Nieto et al . , 2016; Schmidt et al . , 2015 ) . To determine whether expression of QKI or RBFOX1 was also required for the induction of an EMT program , we silenced endogenous QKI or RBFOX1 by short hairpin RNA ( shRNA ) -mediated suppression and by CRISPR/Cas9-mediated knockout . First , we expressed the SNAI1 EMT-TF to induce EMT and then depleted QKI , RBFOX1 or other candidate RBPs with shRNAs ( shSNAI1 as a positive control ) ( Figure 2E and Figure 2—figure supplement 3D ) . shRNA-mediated suppression of QKI and RBFOX1 led to a significant reduction in the CD44-high cell population ( Figure 2E ) , suggesting that the expression of QKI and RBFOX1 was partially required for the induction of the CD44-high cell population after SNAI1 overexpression . To eliminate the potential off-target effects of the shRNAs , we used CRISPR/Cas9 to target the QKI and RBFOX1 genes and found that the ablation of QKI and RBFOX1 also significantly suppressed the induction of CD44-high cells ( Figure 2F ) , the expression of mesenchymal markers ( Figure 2G , H ) and the formation of mammospheres ( Figure 2I ) after SNAI1 overexpression . Thus , these loss-of-function studies revealed that QKI and RBFOX1 are partially required for induction of the EMT . We next examined whether the expression of QKI and RBFOX1 also correlated with mesenchymal features in murine or human tumor samples . We discovered that the expression of QKI was highly upregulated in mesenchymal breast tumor patient samples available from the Cancer Genome Atlas ( TCGA ) ( Ciriello et al . , 2015 ) ( Charafe-Jauffret et al . , 2006 ) ( Figure 2—figure supplement 4A and the Materials and methods section for data analysis ) . The lack of a significant change in expression of RBFOX1 suggested that QKI instead may play a major role in driving the alternative splicing patterns in these samples . In addition , both Qk and Rbfox1 are highly associated with the activation of an EMT program in a murine mammary tumor model ( Figure 2—figure supplement 4B–D and the Materials and methods section ) ( Goel et al . , 2016 ) . Collectively , although QKI has been previously associated with AS changes occurring during EMT , our observations demonstrate that overexpression of QKI or RBFOX1 suffices to promote an intermediate mesenchymal and stem-like cell state and are also necessary for the SNAI1-induced EMT . Further , the expression of endogenous QKI and RBFOX1 were also induced by EMT-TFs such as SNAI1 or ZEB1 ( Figure 2J ) . These results extend prior observations implicating these RNA binding proteins in EMT and confirm that our screen identified key regulators of EMT . Although QKI and RBFOX2 ( a homolog of RBFOX1 ) have been shown to regulate AS events during EMT ( Braeutigam et al . , 2014; Venables et al . , 2013; Yang et al . , 2016 ) , it remains unclear whether QKI and RBFOX1 alter splicing of genes directly involved in EMT or if the expression of these RNA binding proteins merely correlate with the mesenchymal cell state . To dissect the mechanism by which QKI and RBFOX1 induce the intermediate mesenchymal and stem-like cell states , we overexpressed each of these or a control protein ( hcRED or EGFP ) in HME cells and used RNA-sequencing to assess changes in transcriptional programs . We subsequently used replicate multivariate analysis of transcript splicing ( rMATS ) to individually quantify and analyze differences in AS events in HME cells expressing either the hcRED or EGFP control proteins versus QKI , RBFOX1 or SNAI1 ( Shen et al . , 2014 ) . Indeed , HME cells that expressed QKI or RBFOX1 exhibited a > 5 fold increase in the number of alternatively spliced events compared to control cells that expressed hcRED ( Figure 3—figure supplement 1A and Figure 3—source datas 1 and 2 ) . Among all detected types of splicing events , the majority of splicing changes after overexpression of QKI or RBFOX1 occurred in skipped exons ( Figure 3A ) . We next used pre-ranked GSEA to analyze the pathways that are regulated by QKI or RBFOX1 and found that their downstream splicing targets were enriched in gene modules involved in cell motility/cytoskeleton organization , stem cell fate determination , oncogenic signaling and epigenetic targets ( Figure 3—figure supplement 1B , C ) . We then individually validated the top alternatively spliced genes regulated by both QKI and RBFOX1 with the hypothesis that shared targets were more likely to be involved in EMT . We focused on the genes with alternatively skipped exons , as it is the most prevalent type of AS in higher eukaryotes ( Keren et al . , 2010 ) . We confirmed that , consistent with the RNA-seq results , pre-mRNAs of specific exons in the genes involved in cell motility/cytoskeleton organization , FLNB , SLK , NUMB , CA12 , ESYT2 and ATP5C1 showed substantially greater skipping in cells expressing QKI and RBFOX1 , as compared to control cells expressing hcRED or EGFP ( Figure 3B ) . Interestingly , the same AS pattern for these genes was also observed in mesenchymal HME cells overexpressing SNAI1 , indicating that the AS events observed in SNAI1-expressing cells are likely to be due to the activity of QKI and RBFOX1 . From our RNA-sequencing analysis , we found that many AS events , and in particular , skipped exons , were regulated by both QKI and RBFOX1 ( Figure 4A , Figure 4—figure supplement 1A and Figure 4—source data 1 ) . To identify direct targets of QKI and RBFOX1 , we performed enhanced UV crosslinking and immunoprecipitation followed by sequencing ( eCLIP-seq ) in HME cells ( Figure 4—figure supplement 1B ) ( Van Nostrand et al . , 2016 ) . QKI-binding sites were located predominantly in introns , while the majority of RBFOX1-binding sites were found both in introns and 3’UTRs , and consistent with prior studies . We also recovered the known QKI ( ACUAAC ) and RBFOX1 ( UGCAUG ) binding motifs ( Figure 4B , C ) . Interestingly , we found that QKI-binding sites were also highly enriched for the RBFOX-binding motif , UGCAUG ( Figure 4B , C ) and overall , there was a substantial degree of overlap between QKI and RBFOX1 eCLIP-binding peaks ( p<0 . 001 , Figure 4D and Figure 4—source data 2 ) . When examining the 183 exon skipping events that we found to be regulated by both QKI and RBFOX1 , we detected binding sites for both QKI and RBFOX1 for 36 events , with peaks overlapping the exon itself or positioned in the flanking introns ( Figure 4E ) . Since the QKI and RBFOX1 proteins have previously been shown to physically associate with one another ( Lim et al . , 2006 ) , we then tested whether these two proteins were also interacting in HME cells . When we isolated endogenous QKI complexes by immunoprecipitation , we detected a robust interaction with RBFOX1 protein that did not require the presence of RNA ( Figure 4—figure supplement 1C ) . Thus , these observations demonstrate that QKI and RBFOX1 interact in human mammary epithelial cells and suggest that they concurrently bind to and regulate the AS of common downstream targets . To identify transcripts whose AS is likely to play a functional role in promoting EMT , we assessed which QKI and RBFOX1-regulated AS events were also associated with an EMT gene signature across a panel of breast cancer cell lines from the Cancer Cell Line Encyclopedia ( CCLE ) ( Barretina et al . , 2012 ) . We examined the AS events in breast cancer cell lines that were ranked by their EMT gene signature score ( Byers et al . , 2013 ) , using the Information Coefficient ( IC ) , an information-theoretic measure ( Kim et al . , 2016 ) , and an empirical permutation test for statistical significance of the top hits ( Barretina et al . , 2012 ) . Among all the common targets of QKI and RBFOX1 , we found that CD44 ( IC:0 . 857 , p value:<6 . 59e-07 ) and FLNB ( IC:0 . 848 , p value:<6 . 59e-07 ) scored as the top two genes that most strongly associated with the EMT signature in breast cancer cell lines ( Figure 4F ) . CD44 and FLNB were also among the top genes regulated by both QKI and RBFOX1 in HME cells ( Figure 4G ) . Prior work has demonstrated that AS of CD44 to produce the standard shorter isoform promoting EMT , and that CD44 splicing is not only a marker of the EMT state but also contributes to EMT ( Brown et al . , 2011 ) . However , the functional importance of FLNB in EMT has not yet been characterized . Exon 30 of FLNB is skipped when QKI and RBFOX1 are overexpressed ( Figure 3B ) , and we found that both QKI and RBFOX1 were strongly bound to the intron flanking this exon ( Figure 4H , QKI peak p value = 2 . 2e-16; RBFOX1 peaks p value = 3 . 0e-7 and 1 . 8e-9 ) . Although RBFOX1-binding downstream of an exon typically results in splicing enhancement ( Conboy , 2017 ) , we found that binding of RBFOX1 downstream of FLNB exon 30 instead results in splicing repression . Together these observations support the view that QKI and RBFOX1 coordinately regulate the AS of genes associated with EMT . Based on gene expression analysis , prior studies stratified breast cancer cell lines into basal B , basal A and luminal clusters , among which , the basal B subtype expresses mesenchymal markers and displays a high degree of stem-like cell features ( Kao et al . , 2009; Neve et al . , 2006 ) . To identify the alternative transcripts that correlated with the basal B subtype of breast cancer , we analyzed alternatively spliced events in breast cancer cell lines included in the CCLE ( Barretina et al . , 2012 ) . We found several targets of QKI and RBFOX1 , including FLNB , SLK , USO1 , ENAH , ESYT2 , NUMB and ARHGEF1 , to be among the most differentially spliced genes in basal B breast cancer cell lines ( Figure 5—figure supplement 1A ) . Strikingly , we observed a bimodal distribution for the AS of FLNB ( Figure 5A ) , in which the shorter mesenchymal FLNB isoform corresponding to a lower exon 30 'Percent Spliced In' ( PSI ) value , occurred overwhelmingly in basal B cell lines , while the longer epithelial FLNB isoform existed predominantly in luminal and basal A cell lines . We validated this finding in two basal B ( BT549 and MDAMB231 ) and two luminal ( ZR75-1 and MCF7 ) cell lines by RT-PCR ( Figure 5A ) . We further found that there is a strong association between the AS of FLNB exon 30 and EMT gene expression features in breast cancer cell lines ( Figure 5—figure supplement 1B ) . When we examined all non-hematopoietic cancer cell lines in the CCLE , we found that the degree of FLNB exon 30 splicing correlated significantly with a ZEB1 target signature , an epithelial differentiation signature , two metastasis signatures and a mammary stem cell signature ( Figure 5B and Figure 5—figure supplement 1C ) . These observations further confirmed that the AS of FLNB exon 30 strongly associates with EMT and a stem-like cell state . In addition , the strong association between FLNB splicing and EMT features suggest that FLNB exon 30 splicing may serve as a biomarker for residence of cancer cells in a mesenchymal state . Since mesenchymal and stem-like cell features are enriched in basal-like breast cancer , we examined whether the splicing of FLNB and the expression of QKI or RBFOX1 were associated with the basal-like subtype in TCGA Breast Invasive Carcinoma ( BRCA ) samples . We observed lower expression of the longer FLNB isoform with exon 30 included and higher expression of the shorter FLNB isoform in samples classified as the basal subtype , consistent with the notion that FLNB splicing plays a role in regulating the mesenchymal and stem-like cell state ( Figure 5—figure supplement 1D ) . Similarly , we discovered elevated expression of QKI ( NM_006775 , also called QKI-5 ) in basal-like breast cancers relative to other subtypes of breast cancers ( Figure 5—figure supplement 1E ) . FLNB is a member of the Filamin family of actin-binding proteins ( FLNA , B and C ) . Prior work has implicated the role of filamins in actin crosslinking , focal adhesion kinase and integrin signaling , and regulating transcriptional activity ( Feng and Walsh , 2004; van der Flier et al . , 2002; Xu et al . , 2010; Zhou et al . , 2010 ) . Filamins share an N-terminal actin-binding domain , two hinge regions , and 24 filamin-type immunoglobulin-like ( FLN ) domains that are involved in the formation of tail-to-tail dimers ( Feng and Walsh , 2004; van der Flier et al . , 2002 ) . Exon 30 of FLNB encodes the first hinge ( H1 ) domain , which governs filamin protein flexibility and calpain cleavage sensitivity ( Figure 6A ) ( Feng and Walsh , 2004; Xu et al . , 2010 ) . The skipping of exon 30 results in loss of the H1 domain from the full-length protein without altering the remainder of the protein . Hereafter , we refer to the longer isoform of FLNB ( including exon 30 ) as FLNB-L , and to the shorter isoform ( which lacks exon 30 ) as FLNB-ΔH1 . When we tested whether the splicing of FLNB differed between the CD44-high and CD44-low cell populations ( Figure 6B ) , we found that FLNB exon 30 skipping occurred exclusively in the CD44-high mesenchymal and stem-like cell population . To investigate the function of FLNB in regulating EMT , we suppressed FLNB expression using siRNAs targeting the FLNB 3’UTR region in HMLE cells , in which the FLNB-L isoform represents the majority of FLNB protein . We found that suppression of FLNB-L upregulated the expression of mesenchymal markers , VIM and FN1 , indicating that FLNB-L plays a negative role in regulating EMT ( Figure 6C ) . To dissect the respective role of FLNB-L and FLNB-ΔH1 isoforms , we rescued the suppression of endogenous FLNB by ectopically expressing each isoform of FLNB ( Figure 6D ) . Depletion of FLNB promoted the expression of mesenchymal markers . We found that FLNB-L reduced the expression of mesenchymal markers , FN1 and VIM . In contrast , the expression of FLNB-ΔH1 did not decrease the mesenchymal marker expression . When the two isoforms of FLNB were expressed in a mesenchymal cell line , MDA-MB-231 , we also found that FLNB-L overexpression suppressed mesenchymal marker expression , strongly suggesting that FLNB-L inhibits the EMT ( Figure 6—figure supplement 1A ) . Interestingly , when we expressed each isoform in HMLE cells in the presence of the endogenous FLNB-L , FLNB-ΔH1 ectopic expression elevated mesenchymal markers while the expression of FLNB-L did not significantly alter expression of the same set of markers ( Figure 6—figure supplement 1B , C ) . Since filamin proteins dimerize ( Berry et al . , 2005; Pudas et al . , 2005; Stossel et al . , 2001 ) , the effects of FLNB-ΔH1 proteins likely represent interactions with the endogenous FLNB-L , which blocks the suppressive effect mediated by FLNB-L . As before , we did not observe robust changes in the expression of pre-existing epithelial markers in these experiments , reminiscent of our previous observation that QKI or RBFOX1 induces an intermediate mesenchymal state with retention of epithelial markers and acquisition of mesenchymal ones ( Figure 2A , B and Figure 2—figure supplement 1B ) . Together , these results support the view that the skipping of the exon 30 of FLNB switches the function of the FLNB from suppressing to promoting the EMT . To manipulate the ratio of the two FLNB isoforms and dissect the function of the FLNB exon 30 skipping , we modified the genomic locus of the intron-exon junction using CRISPR/Cas9-mediated genome editing to skew the isoform ratio of the endogenous FLNB transcripts . We designed several sgRNAs that target the junction of intron 29 and exon 30 ( sgFLNB-SK2 and SK4 ) . Remarkably , we found that disrupting this junction in the genomic locus was effective in causing skipping of the endogenous exon 30 of FLNB ( Figure 6E , F ) . In line with our previous observations , FLNB exon 30 skipping induced by this approach also increased the expression of mesenchymal markers ( Figure 6G ) . We also discovered that FLNB exon 30 skipping induced a modest but significant increase in the CD44-high cell population and in the number of mammospheres under low attachment growth conditions ( Figure 6H , I ) . In addition , we used two sets of siRNAs that targeted either exon 30 , or the junction between exon 29 and exon 31 when exon 30 is skipped . The siRNAs that target the exon 29–31 junction selectively disrupt formation of the FLNB-ΔH1 isoform , since the FLNB-L isoform lacks the siRNA target sequences . This approach effectively altered the ratio of FLNB-L and FLNB-ΔH1 ( Figure 6—figure supplement 1D , E ) and revealed that an elevated ratio of the FLNB-ΔH1 isoform over the FLNB-L isoform and a reduction of FLNB protein levels significantly increased the level of mesenchymal markers , consistent with the effect of ectopically expressing FLNB-ΔH1 or CRISPR/Cas9-mediated editing of the splice junction ( Figure 6C–I ) . Together , these observations demonstrate that the skipping of FLNB exon 30 contributes to the acquisition of a mesenchymal-like cell state . In addition to their function in the cytoplasm , actin-binding proteins , such as the Filamin family , have been shown to localize to the nucleus and regulate transcription and gene expression ( Bedolla et al . , 2009; Berry et al . , 2005; Olson and Nordheim , 2010; Zheng et al . , 2009; Zhou et al . , 2010 ) . We tested whether the two isoforms of FLNB generated by alternative splicing of exon 30 localized to different subcellular compartments . By comparing HMLE cells expressing a control sgRNA targeting GFP ( sgGFP ) or sgRNAs that induce exon 30 skipping ( sgFLNB-SK2 and SK4 ) ( Figure 6E ) , we found that the FLNB-L isoform localized to both the cytoplasm and the nucleus while FLNB-ΔH1 was preferentially localized to the cytoplasm ( 50% reduction in nuclear localization , Figure 7A ) . We confirmed this change of FLNB localization using immunofluorescence ( Figure 7—figure supplement 1A ) . Thus , alternative splicing of exon 30 changes the nuclear localization of FLNB . Filamin A has been reported to assemble a protein complex with FOXC1 and PBX1 , which inhibits FOXC1 transcriptional activity ( Berry et al . , 2005; Zheng et al . , 2009; Zhou et al . , 2010 ) . To test whether Filamin B also forms a complex with FOXC1 in HMLE cells , we confirmed that FLNB interacts with FOXC1 by co-immunoprecipitation ( Figure 7—figure supplement 1B ) . Furthermore , we found that the interaction among FLNB , FOXC1 and PBX1 was reduced when we induced FLNB exon 30 skipping , largely due to the decreased amount of nuclear FLNB protein ( Figure 7B ) . Based on these observations , we conclude that FLNB nuclear exclusion , mediated by exon 30 skipping , regulates its interaction with FOXC1 . FOXC1 is a transcription factor that induces EMT ( Han et al . , 2017; Huang et al . , 2017; Ou-Yang et al . , 2015; Zhu et al . , 2017 ) . Since the nuclear filamins inhibit FOXC1 activity ( Berry et al . , 2005; Zheng et al . , 2009; Zhou et al . , 2010 ) , we hypothesized that the reduced nuclear localization of FLNB promotes EMT by releasing FOXC1 from FLNB . Specifically , we tested whether the EMT induced by FLNB isoform switching was dependent on FOXC1 . Indeed , we found that FOXC1 depletion by siRNA significantly dampened the upregulation of mesenchymal marker expression ( Figure 7C ) and the formation of mammospheres ( Figure 7D ) mediated by FLNB exon 30 skipping . Furthermore , we found that FOXC1 is also partially required for the upregulation of mesenchymal marker expression ( Figure 7E ) and increase in mammosphere formation ( Figure 7F , Figure 7—figure supplement 1C ) induced by QKI and RBFOX1 expression , which regulate the alternative splicing of FLNB exon 30 . In summary , the skipping of FLNB exon 30 promotes EMT by reducing FLNB nuclear localization and release of the FOXC1 transcription factor .
Splicing is a key step in the regulation of almost all human transcripts . The recent genomic characterization of cancers has revealed recurrent somatic mutations and copy number alterations in RNA splicing factors and RBPs in a significant subset of human tumors ( Dvinge et al . , 2016 ) . Cancer cells harboring aberrant splicing factor expression or mutations in genes encoding splicing factors display unique cancer-specific mis-splicing that may facilitate tumor formation and progression . Although alternative splicing has been associated with EMT previously , in-depth studies are needed to better understand the mesenchymal cell state-specific RBPs and their functional downstream targets . Here , we found that QKI and RBFOX1 regulate the splicing of an exon in the actin-binding protein FLNB to regulate the EMT in breast cancer . This finding suggests that the AS of a single exon may serve as a highly quantifiable surrogate molecular biomarker for the process of EMT in solid tumors . Recent work has shown that cells often progress through a spectrum of intermediate states between the fully epithelial and fully mesenchymal cell phenotypes ( George et al . , 2017; Nieto et al . , 2016 ) . We found that the mesenchymal cell state mediated by the expression of QKI and RBFOX1 exhibited upregulation of mesenchymal markers with continued retention of certain epithelial markers , indicating that this cell state is one that lies in-between the fully epithelial and fully mesenchymal poles of this spectrum . Consistent with prior studies ( Nieto et al . , 2016; Schmidt et al . , 2015 ) , our results suggest that the intermediate/partial mesenchymal cell state displays a high degree of stem cell features and fosters tumor formation in vivo . The control of AS by RNA-binding proteins is highly context dependent ( Fu and Ares , 2014 ) and tissue specific ( Yeo et al . , 2004 ) . QKI has been shown to be a tumor suppressor in brain tumors ( Chen et al . , 2012 ) , colon cancers ( Taube et al . , 2010 ) and prostate cancers ( Zhao et al . , 2014 ) . In contrast , QKI has also been reported to promote tumor formation in both esophageal carcinoma ( He et al . , 2016 ) and glioblastoma ( Wang et al . , 2013; Xi et al . , 2017 ) . Here , we found that QKI promoted tumor formation in human mammary epithelial cells . These distinct observations may be due to the differences in the initial cell states in which the cancer cells may reside . The RBFOX1 ORF that we isolated from the screen encodes an isoform ( NM_145893 . 2 ) that has been previously shown to partially localized to the cytoplasm in neuronal cells ( Lee et al . , 2009 ) . In breast cells , we observed that 38% of this RBFOX1 isoform ( NM_145893 . 2 ) localize in the nucleus to regulate pre-mRNA splicing ( Figure 7—figure supplement 1D ) . Further studies will be needed to determine whether the cytoplasmic fraction of RBFOX1 also plays an additional role in regulating EMT . Interestingly , the overexpression of EMT-related TFs such as SNAl1 and ZEB1 induced the QKI- and RBFOX1-mediated splicing program . Moreover , overexpression of the QKI and RBFOX1 splicing factors themselves promoted a mesenchymal cell state in which SNAI1 and ZEB1 expression were also elevated . These observations indicate that transcriptional and post-transcriptional regulation of EMT complement and regulate one another , suggesting how EMT can be dynamically controlled . Filamins bind to proteins with diverse functions and important roles in multiple cellular process such as the regulation of cell signaling , transcription and organ development ( Zhou et al . , 2010 ) . In this study , we demonstrated that the skipping of exon 30 in FLNB , which encodes a hinge region ( H1 ) , not only serves as a marker for mesenchymal cells but also promotes EMT by releasing the FOXC1 transcription factor from an inhibitory complex . These observations provide a mechanistic explanation of how mesenchymal-specific splicing factors such as QKI and RBFOX1 induce EMT . Collectively , we conclude that QKI and RBFOX proteins play important roles in establishing the mesenchymal and stem-like cell state in breast cancers , which is in part mediated through their mutual regulation of the skipping of FLNB exon 30 . Alternative splicing of pre-mRNAs represents a mechanism for flexibly regulating gene expression by enabling the generation of protein isoforms with distinct or even opposing functions without altering rates of transcription . As such , it offers yet another level of epigenetic control of gene expression . Accordingly , it may provide another means of regulation that tumor cells exploit in order to produce proteins that favor cell survival and cell state changes , such as the EMT programs studied here .
The human mammary epithelial ( HME ) cell line , and its derived cell lines HMLE and HMLER , were grown in Mammary Epithelial Cell Growth Medium ( MEGM , Lonza , #CC-3150 ) . MCF7 , BT549 , MDAMB231 , ZR75-1 cells were grown in Dulbecco’s minimum essential medium ( DMEM ) or Roswell Park Memorial Institute ( RPMI ) medium containing 10% fetal bovine serum and 1% antibiotics , while 293 T cells were grown in DMEM supplemented with 10% fetal bovine serum and 1% antibiotics . All cell lines were obtained from the Cancer Cell Line Encyclopedia directly from original sources and had their identity confirmed by SNP fingerprinting ( Barretina et al . , 2012 ) . All cells were tested for mycoplasma periodically . The ORFs of EGFP , HcRed , CELF4 , MBNL1 , MBNL2 , QKI , RBFOX1 , SFPQ , HNRNPUL1 , SRSF9 , RBM47 , TGFBR2 and SNAI1 were obtained from the Genetic Perturbation Platform at the Broad Institute . Plasmids containing the two isoforms of FLNB ( FLNB-L and FLNB-∆H1 ) were generous gifts from Dr . Arnoud Sonnenberg at the Netherlands Cancer Institute . The ORFeome 8 . 1 library was produced in the Genetic Perturbation Platform at the Broad Institute . There are the 17 , 255 ORF clones in the library , among which , 12952 ORF clones have at least 99% nucleotide and protein match ( 75 . 1% ) . Of those genes 7547 ORF clones are 100% protein matches ( 43 . 7% ) and 6040 are 100% nucleotide matches ( 35 . 0% ) . The construction of the ORF library has been described previously ( Yang et al . , 2011 ) . All protocols with use of animals were approved by DFCI’s Institutional Animal Care and Use Committee ( IACUC ) . For tumor studies , HMLER cells expressing different ORFs were washed and suspended in 50% Matrigel/PBS mix ( Corning Matrigel Basement Membrane Matrix , #354234 ) , then injected subcutaneously in both flanks and in the back of 6-week-old female immunocompromised NCr-nude mice ( Taconic , NCRNU-F , CrTac:NCr-Foxn1nu ) . Two million cells were injected per site . Mice were sacrificed after 15 weeks or when tumors reached a diameter of 2 cm . Cells were trypsinized , suspended in phosphate-buffered saline ( PBS ) with CD44-PE-Cy7 antibody ( Affymetrix # 25-0441-81; 1:500 dilution ) , and stained for 20 min at room temperature; cells were mixed at 5 min intervals , and then washed with PBS to remove excess antibodies . Immediately after , cells were sorted on a BD FACSAria SORP or analyzed on a BD Fortessa , using BD FACSDiva Software ( BD Biosciences , USA ) . Mammosphere cultures were generated as described ( Chaffer et al . , 2013 ) . Briefly , 1000 cells were seeded per well in a 96-well Corning Ultra-Low attachment plate , in replicates of 6 ( Corning , USA; CLS3474 ) . Cells were grown in a serum-free mammary epithelial cell growth medium , supplemented with B27 ( Invitrogen ) , 10 ng/mL EGF , 20 ng/mL bFGF ( BD Biosciences ) and 1% methycellulose . Bovine pituitary extract was excluded . Spheroid numbers were counted between days 8 and 12 microscopically . The genome-scale screen was performed in two biological replicates . HMLER cells in the CD44-low state were pre-sorted using flow cytometry . The purity of CD44-low cells was >99 . 99% for each experiment . HMLER_CD44-low cells were then transduced with the ORF library and cultured for 7 days , with one passage on Day 4 . On Day 7 , CD44-high cells were sorted from more than 200 million transduced HMLER cells , using flow cytometry for each biological replicate . Sorted CD44-high cells ( about 200 thousand cells for each replicate ) and their corresponding unsorted HMLER cells were subjected to genomic DNA extraction using the QIAamp DNA Mini Kit ( Qiagen # 51304 ) . The barcodes corresponding to each ORF were amplified using PCR , and analyzed by next-generation sequencing . Enriched barcodes were analyzed as follows: ( i ) Each sample was normalized to a total of 1 million barcode reads . ( ii ) The number of each barcode after normalization was calculated to its log base two value . The log value of each barcode in the unsorted group was subtracted from the CD44-high group to obtain the log fold-change in the value of each barcode . ( iii ) The averages and standard deviations ( SD ) of the log fold-change values in all samples were determined , and Z scores for each barcode were calculated as follows: Z Barcode X= ( Log value Barcode X - average ) /SD . The Z score was used to evaluate the enrichment of a certain ORF in the CD44-high population , compared with the unsorted population . A higher Z score indicated an enhanced capability for an ORF to promote the conversion of HMLER cells to the CD44-high state . To prepare libraries for RNA sequencing of HME cells that overexpress HcRed , EGFP , QKI , RBFOX1 or SNAI1 , we first extracted total RNA using the RNeasy Mini Kit ( QIAGEN ) . Next , 1 . 5 ug of total RNA was used to generate first strand cDNA using Oligo ( dT ) 12–19 primers ( Invitrogen ) and AffinityScript Multi-Temp Reverse Transcriptase ( Agilent ) . Second strand cDNA was synthesized using the NEBNext mRNA Second Strand Synthesis Module ( NEB ) and washed with AMPure XP beads ( Beckman Coulter ) . Finally , libraries were generated from cDNA using the Nextera XT DNA Sample Preparation Kit ( Illumina ) and Nextera XT Indexes ( Illumina ) . Libraries were pooled and sequenced on the Illumina NextSeq 500 sequencer ( paired-end , 150 bp ) . Image analysis and base calling were done using the standard Illumina pipeline , and then demultiplexed into FASTQ files . Reads were first trimmed using Trimmomatic ( version 0 . 33 ) to remove Nextera adapter sequences down to a uniform length of 100nt ( for compatibility with downstream splicing analysis software ) . Trimmed reads were then aligned to the human genome ( hg19/GRCh37 ) using STAR ( version 2 . 5 . 2b ) ( Dobin et al . , 2013 ) and Gencode V19 gene annotations . Alternative splicing was quantified using rMATS ( version 3 . 2 . 5 ) ( Shen et al . , 2014 ) by comparing each ORF to EGFP , with at least two to three replicates per group . The output from rMATS was further filtered to include only events for which the sum of inclusion counts ( IC ) and skipping counts ( SC ) was greater or equal to 10 for both sets of samples . Alternative splicing quantification across cell lines in CCLE and TCGA breast invasive carcinoma was performed using JuncBASE v . 0 . 8 with default parameters after initial sequence alignment using TopHat v1 . To incorporate potentially novel exons , Cufflinks de novo transcript annotations were included from the CCLE data only . Total RNA was isolated using the RNeasy Mini kit ( Qiagen , 74104 ) according to the manufacturer’s protocol . A cDNA sample , prepared from 1 μg total RNA , was used for quantitative reverse transcription polymerase chain reaction ( RT-PCR ) performed with the High Capacity cDNA Reverse Transcription Kit ( Life Technologies , 4368814 ) or iScript Reverse Transcription Supermix ( BIO-RAD , 1708840 ) . Quantitative PCR ( qPCR ) was done with the Power SYBR Green Master Mix ( Life Technologies; 4368708 ) ; data were collected and analyzed on a Bio-Rad Real-Time PCR Detection System or a Roche LightCycler 480 qPCR instrument . Thermal-cycling parameters for the PCR were as follows: 95°C for 10 min , followed by 45 cycles each of 95°C for 20 s , 60°C for 60 s . The relative quantity of mRNA was normalized against the relative quantity of RPLP0 or GAPDH mRNA in the same sample . Primer sequences in a 5′ to 3′ orientation are shown in Supplementary file 1 . EMT UP and DOWN signatures were derived from previously published datasets based on their pattern of expression relative to the EMT phenotype ( TAUBE_EMT_UP/DN , EMT gene set ( Taube et al . , 2010 ) , GROGER_EMT_UP/DN , EMT gene set ( Gröger et al . , 2012 ) , BYERS_EMT_UP/DN ( Byers et al . , 2013 ) . The EMT signature scores across CCLE were generated by using the ssGSEA algorithm ( Subramanian et al . , 2005 ) . These scores were used to identify the top associated splice targets based on degree of association , an information-theoretic measure Information Coefficient ( IC ) ( Kim et al . , 2016 ) . An empirical permutation test was performed for statistical significance calculations . RNA-sequencing data of TCGA Breast invasive carcinoma ( BRCA ) samples were downloaded from the GDAC portal of the Broad Institute ( http://gdac . broadinstitute . org/ ) . The EMT signature scores across TCGA_BRCA samples were generated by the ssGSEA algorithm based on a previously published EMT gene expression signature ( CHARAFE_EMT_UP and _DOWN combined ) ( Charafe-Jauffret et al . , 2006; Subramanian et al . , 2005 ) . The top 20% of samples ( total n = 1212 , mesenchymal tumor = 242 ) that had the highest EMT scores were counted as mesenchymal tumor samples and the top 20% of samples ( n = 242 ) that had the lowest EMT scores were counted as epithelial tumor samples . The gene expression of EMT markers and RBPs were compared between these mesenchymal and epithelial samples in Figure 3G . In addition , these scores were used to identify the top correlated gene expression based on degree of association by calculating Pearson Correlation Coefficiency ( PCC ) and their p values . Breast cancer subtypes were obtained from a PAM50 gene signature-based TCGA analysis ( Ciriello et al . , 2015 ) and correlated with the expression of specific isoforms . RNA sequencing data for gene expression in primary and recurrent MMTV-HER2 mammary tumors were previously published ( Goel et al . , 2016 ) . In this model , withdrawal of HER2 expression leads to primary mammary tumor regression but is eventually followed by recurrence of HER2-resistant tumors that harbor a mesenchymal phenotype ( Figure 2—figure supplement 1D–F ) . Ten out of 11 such recurrent tumors underwent EMT as shown by the expression of mesenchymal markers and the spindle-like cellular morphology ( Figure 2—figure supplement 1D ) ( Goel et al . , 2016 ) . Strikingly , based on an analysis of the RNA-sequencing results from Goel et al . ( 2016 ) , we found that the expression of Qk ( mouse homolog of human QKI ) and Rbfox1 were significantly upregulated in the recurring mesenchymal mammary tumors relative to their expression in the corresponding , initially formed epithelial tumors ( Figure 2—figure supplement 1D , E ) . The differential gene expression was evaluated by p values calculated by student’s t-test of the normalized expression values between the recurrent tumors and primary tumors . The false discovery rate ( FDR ) values were generated by comparative marker selection analysis in Genepattern ( Reich et al . , 2006 ) . Cell extract preparation and immunoblotting were completed as described ( Li et al . , 2013 ) . All antibodies used for immunoblotting were listed in Supplementary file 1 . For RBFOX1 immunoblotting , we detected a 42 kDa band in HME cell lysate ( predicted size ) . RBFOX1 levels are higher in HMLER cells and we observed a 33 kDa lower band , in addition to the 42 kDa band that corresponds to the predicted size of RBFOX1 ( Figure 2—figure supplement 2C , D ) , which is likely a cleaved form of RBFOX1 . For preparation of whole cell extract , HME or 293 T cells were harvested , and lysed on ice for 30 min with IP buffer containing 50 mM Tris HCl pH 7 . 0 , 150 mM NaCl , 1 mM EDTA/pH 8 , 0 . 5% Na-deoxycholate , 0 . 5% NP-40 and 10% glycerol , with protease inhibitors added before use . The lysates were sonicated with 10 pulses on ice , then centrifuged at 14 , 000 rpm for 5 min , and the supernatants were collected for immunoprecipitation . For preparation of nuclear protein , HMLE nuclear extract was prepared using the Nuclear Complex Co-IP kit ( Active Motif #54001 ) as described in manufacturer’s instructions . Briefly , cell pellets were resuspended in hypotonic buffer to break cell membrane and the nucleus were isolated by centrifugation . The nuclear fraction was further lysed by digestion buffer with enzymatic shearing cocktail before it was further diluted in the IP buffer provided in the kit . For immunoprecipitations , QKI antibody ( Bethyl Laboratories # A310-050A ) or FLNB antibody ( Millipore # AB9276 ) was added at a concentration of 1 ug per 1 mg of cell lysate , and the lysates were incubated for 2 hr at 4°C; protein A/G agarose was then added and lysates were further incubated for 2 hr at 4°C on a rotator . Protein A/G beads were washed four times in cold IP buffer followed by centrifugation . Samples were boiled in SDS loading buffer , and separated on an SDS-PAGE gel , followed by immunoblotting . For QKI immunoprecipitations , to digest the RBP-associated RNAs , cell lysate was incubated with 50 ng/ml of RNase at room temperature for 20 min before antibodies were added . HME , HMLE or HMLER cells were transfected with siRNAs , using Lipofectamine RNAi-MAX . Six hours before the siRNA transfection , cells were split into six-well plates . To prepare transfection complexes , 5 ul of RNAiMAX was mixed into 150 ul of OptiMEM medium in one tube , while 5 ul of 20 mM siRNA was mixed into 150 ul of OptiMEM medium in another tube . Tubes were incubated at room temperature for 20 min before being added to the cells . The cells were harvested 72 hr after transfection . Immunofluorescence procedures have been described previously ( Li et al . , 2013 ) . Briefly , HMLE cells were fixed with cold methanol for 2 min and permeabilized with PBS-1% Triton X-100 for 5 min . Cells were blocked in PBS-donkey serum for 1 hr before being incubated with primary antibody for 2 hr . Alexafluor 488-conjugated donkey anti-rabbit IgG ( Invitrogen # R37118 ) was used as a secondary antibody . DNA was stained with DAPI . Images were acquired with a Nikon inverted microscope . For Phalloidin staining , cells were fixed with Formalin for 8 min and incubated with Phalloidin-Alexafluor 488 for 1 hr at room temperature before DAPI staining and image analysis . RBP-RNA interactions were crosslinked by UV exposure ( 254 nm , 400 mJ/cm2 ) using a Spectrolinker XL-1500 UV crosslinker . eCLIP was then performed as previously described ( Van Nostrand et al . , 2016 ) ( ENCODE protocol v1 . P 20151108 ) with some minor modifications as follows: ( 1 ) immunoprecipitated RNA was 3’ end ligated to a custom RNA adapter ( ‘3 SR_RNA’ ) ; ( 2 ) RNA was released from the nitrocellulose membrane after transfer by treatment with 200 ul of an SDS solution ( 100 mM Tris , pH 7 . 5; 50 mM NaCl; 1 mM EDTA; 0 . 2% SDS ) containing 10 ul of proteinase K ( Life Technologies , AM2546 ) and incubating in an Eppendorf thermomixer ( 60 min at 50°C: 15 s at 1000 r . p . m . , 30 s rest ) , as described in the irCLIP protocol ( Zarnegar et al . , 2016 ) ; ( 3 ) reverse transcription was done with a custom RT primer ( ‘SR_RT’ ) ; ( 4 ) the 3’ end of the cDNA was ligated to a custom DNA adapter ( ‘SR_DNA’ ) ; ( 4 ) amplification of ligated cDNA was done with NEBNext Multiplex Oligos for Illumina ( NEB , E7335S ) ; ( 5 ) PCR amplified libraries were purified twice with AMPure XP beads ( 1 . 0X both times ) and then directly quantified by qPCR and run on a Bioanalyzer High Sensitivity DNA chip , before being pooled and submitted for sequencing on the Illumina NextSeq 500 ( single-end , 75 bp ) . For each RBP ( QKI and RBFOX1 ) , we prepared and sequenced two replicates and a single size-matched input control derived from the first replicate . Sequenced reads were processed as previously described ( Van Nostrand et al . , 2016 ) ( ENCODE pipeline v1 . P 20160215 ) , with some minor modifications . First , the unique molecular index ( UMI ) from the 5' end of each read was extracted using UMI Tools ( parameters: umi_tools extract --bc-pattern=NNNNN ) . Next , adapters were trimmed using cutadapt ( parameters: cutadapt --match-read-wildcards --times 1 -e 0 . 1 -O 1 --quality-cutoff 6 m 18 -a NNNNAGATCGGAAGAGCACACGTCTGAACTCCAGTCAC ) . Trimmed reads were first aligned to a database of human repetitive elements ( RepBase ) using STAR ( v2 . 5 . 2b ) ( parameters: STAR --runMode alignReads --genomeDir/path/to/RepBase --readFilesCommand zcat --outSAMunmapped Within --outFilterMultimapNmax 30 --outFilterMultimapScoreRange 1 --outSAMattributes All --outStd BAM_Unsorted --outSAMtype BAM Unsorted --outFilterType BySJout --outReadsUnmapped Fastx --outFilterScoreMin 10 --outSAMattrRGline ID:foo --alignEndsType EndToEnd ) . Reads not mapping to RepBase were then aligned to the human reference genome ( hg19 with Gencode V19 annotations ) using STAR ( v2 . 5 . 2b ) ( parameters: STAR --runMode alignReads --genomeDir/path/to/hg19 --readFilesIn --outSAMunmapped Within --outFilterMultimapNmax 1 --outFilterMultimapScoreRange 1 --outStd BAM_Unsorted --outSAMattributes All --outSAMtype BAM Unsorted --outFilterType BySJout --outReadsUnmapped Fastx --outFilterScoreMin 10 --outSAMattrRGline ID:foo --alignEndsType EndToEnd ) . PCR duplicates were then removed using UMI Tools ( parameters: umi_tools dedup --method directional-adjacency --spliced-is-unique ) leaving only uniquely mapping reads . Finally , CLIP peaks were called using CLIPper software ( Lovci et al . , 2013 ) and identified peaks were normalized to the appropriate size-matched input control with ‘Peak_input_normalization_wrapper . pl’ ( https://github . com/YeoLab/gscripts ) before being merged between replicates . Binding motifs were identified using MEME-ChIP ( Machanick and Bailey , 2011 ) . All data represent the average of at least three independent experiments , unless otherwise indicated . Significance was calculated by two-tail Student's t-test , using GraphPad software . Differences were considered significant when p was <0 . 05 . All primer and oligo sequences are listed in Supplementary file 1 . Both the RNA-seq data and the CLIP-seq data are deposited at NCBI Gene Expression Omnibus ( accession number GSE98210 ) . | As the human body develops , countless cells change from one state into another . Two important cell states are known as epithelial and mesenchymal . Cells in the epithelial state tend to be tightly connected and form barriers , like skin cells . Mesenchymal state cells are loosely organized , move around more and make up connective tissues . Some cells alternate between these states via an epithelial-to-mesenchymal transition ( EMT for short ) and back again . Without this transition , certain organs would not develop and wounds would not heal . Yet , cancer cells also use this transition to spread to distant sites of the body . Such cancers are often the most aggressive , and therefore the most deadly . The epithelial-to-mesenchymal transition is dynamically regulated in a reversible manner . For example , the genes for some proteins might only be active in the epithelial state and further reinforce this state by turning on other ‘epithelial genes’ . Alternatively , there might be differences in the processing of mRNA molecules – the intermediate molecules between DNA and protein – that result in the production of different proteins in epithelial and mesenchymal cells . Li , Choi et al . wanted to know which of the thousands of human genes can endow epithelial state cells with mesenchymal characteristics . A better understanding of the switch could help to prevent cancers undergoing an epithelial-to-mesenchymal transition . From a large-scale experiment in human breast cancer cells , Li , Choi et al . found that a group of proteins that bind and modify mRNA molecules are important for the epithelial-to-mesenchymal transition . Two proteins in particular promoted the transition , most likely by binding to the mRNA of a third protein called FLNB and removing a small piece of it . FLNB normally works to prevent the epithelial-to-mesenchymal transition , but the smaller protein encoded by the shorter mRNA promoted the transition by turning on ‘mesenchymal genes’ . This switching between different FLNB proteins happens in some of the more aggressive breast cancers , which also contain mesenchymal cells . Finding out which FLNB protein is made in a given cancer may provide an indication of its aggressiveness . Also , looking for drugs that can target the mRNA-binding proteins or FLNB may one day lead to new treatments for some of the most aggressive breast cancers . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology",
"cancer",
"biology"
] | 2018 | An alternative splicing switch in FLNB promotes the mesenchymal cell state in human breast cancer |
Myocardial infarction ( MI ) promotes a range of systemic effects , many of which are unknown . Here , we investigated the alterations associated with MI progression in heart and other metabolically active tissues ( liver , skeletal muscle , and adipose ) in a mouse model of MI ( induced by ligating the left ascending coronary artery ) and sham-operated mice . We performed a genome-wide transcriptomic analysis on tissue samples obtained 6- and 24 hr post MI or sham operation . By generating tissue-specific biological networks , we observed: ( 1 ) dysregulation in multiple biological processes ( including immune system , mitochondrial dysfunction , fatty-acid beta-oxidation , and RNA and protein processing ) across multiple tissues post MI and ( 2 ) tissue-specific dysregulation in biological processes in liver and heart post MI . Finally , we validated our findings in two independent MI cohorts . Overall , our integrative analysis highlighted both common and specific biological responses to MI across a range of metabolically active tissues .
Cardiovascular disease ( CVD ) is the leading cause of death worldwide , accounting for more than 17 million deaths globally in 2016 ( WHO , 2019 ) . Myocardial infarction ( MI ) is one of the most common causes of CVD-related death and is the result of severe coronary artery disease that develops from tapered arteries or chronic blockage of the arteries caused by accumulation of cholesterol or plaque ( atherosclerosis ) . Many behavioral risk factors ( including unhealthy diet , physical inactivity , excessive use of alcohol , and tobacco consumption ) , which are responsible for hypertension , obesity , diabetes , and hyperlipidemia by significantly altering metabolism , are also implicated in MI . These abnormalities are known as the high-risk factors of MI and CVDs in general . Systems biology has been used in many studies to reveal the underlying molecular mechanisms of complex human diseases and to answer important biological questions related to the progression , diagnosis , and treatment of the diseases . The use of systems biology has aided the discovery of new therapeutic approaches in multiple diseases ( Mardinoglu et al . , 2017a; Mardinoglu and Nielsen , 2015; Nielsen , 2017 ) by identifying novel therapeutic agents and repositioning of existing drugs ( Turanli et al . , 2019 ) . Systems biology has also been employed in the identification of novel biomarkers , characterization of patients , and stratification of heterogenous cancer patients ( Benfeitas et al . , 2019; Bidkhori et al . , 2018; Lee et al . , 2016 ) . Specifically , integrated networks ( INs ) ( Lee et al . , 2016 ) and co-expression networks ( CNs ) ( Lee et al . , 2017 ) have been proven to be robust methods for revealing the key driver of metabolic abnormalities , discovering new therapy strategies , as well as gaining systematic understanding of diseases ( Bakhtiarizadeh et al . , 2018; Mukund and Subramaniam , 2017 ) . Previously , multiple studies in individual tissues have been performed and provided new insights into the underlying mechanisms of diseases ( Pedrotty et al . , 2012; Das et al . , 2019; Ounzain et al . , 2015; Williams et al . , 2018 ) . However , the crosstalk between different tissues and their dysregulation has not been examined in MI and other CVD-related complications ( Priest and Tontonoz , 2019 ) . Here , we performed an integrated analysis of heart and other metabolically active tissues ( liver , skeletal muscle and adipose tissue ) using a mouse model of MI . We used several systems biology approaches to obtain a systematic picture of the metabolic alterations that occur after an MI ( Figure 1A ) , and validated our findings in two independent datasets .
To study global biological alterations and systemic whole-body effects associated with MI , we obtained heart , liver , skeletal muscle , and white adipose tissue from mice 6 hr and 24 hr after either an MI ( induced by ligating the left ascending coronary artery ) or a sham operation ( as control ) . Total of 20 mice were used in this study ( five mice in each time and condition combination ) ( Figure 1A ) . We generated transcriptomics data and identified differentially expressed genes ( DEGs ) 6 and 24 hr post MI and sham operation in all tissues , with the most significant differences occurring after 24 hr ( Supplementary file 1 , Figure 1B ) . Principal component analysis ( PCA ) showed a close clustering between the control ( for both time points ) and MI ( 6 hr and 24 hr separately ) samples for heart tissue but clustering by extraction time points ( 6 hr and 24 hr clusters ) for the other tissues ( Figure 1—figure supplement 1 ) . We present the transcriptional changes associated with MI in Supplementary file 1 and the DEGs ( FDR < 5% ) using an UpSet plot ( Lex et al . , 2014 ) in Figure 1C . All tissues showed a more pronounced effect in terms of the number of DEGs 24 hr post MI ( Figure 1C ) . As expected , the most affected tissue was the heart ( 393 DEGs at 6 hr , 3318 DEGs at 24 hr , and 318 DEGs were the same at both time points ) . By contrast , 136 , 641 , and 374 genes were significantly changed in liver , skeletal muscle and adipose tissues 24 hr post MI compared to control , respectively . More than 33% of the DEGs that significantly changed in the other tissues also changed in the heart ( Figure 1C ) . Interestingly , more than 97% of the shared DEGs between heart and skeletal muscle changed in the same direction , with corresponding numbers of 88% and 64% in adipose and liver , respectively . We performed gene-set enrichment analysis ( GSEA ) with KEGG pathways ( Supplementary file 2 , Figure 1D ) and gene ontology ( GO ) biological processes ( BPs ) ( Supplementary file 3 , Figure 2A ) to identify altered biological functions and pathways 24 hr after an MI . Mitochondrial functions ( specifically , mitochondrial translation , respiratory chain and oxidative phosphorylation ) were significantly downregulated in the heart , muscle and adipose tissues but not in the liver . Processes related to oxidative stress were upregulated in the heart and skeletal muscle . Fatty acid beta-oxidation was downregulated in the heart and adipose but upregulated in the liver . Processes and pathways related to immune systems were significantly upregulated in the heart and skeletal muscle but significantly downregulated in liver . Processes associated with protein and RNA processing , ribosome biogenesis and protein targeting endoplasmic reticulum were upregulated in all tissues except liver , whereas protein processing in endoplasmic reticulum and RNA transport pathways were upregulated in all tissues . We also observed that liver was showing opposite trends compared to the other tissues in other important functions , such as fatty acid metabolism and immune response . By checking regulation at the gene level , we observed that only 16 DEGs in liver showed opposite regulation compared to the other tissues , whereas 97 out of the 136 DEGs in liver were not DEGs in any other tissues ( Supplementary file 4 ) . Therefore , the differences we observed in liver were mainly due to different DEGs rather than opposite regulation compared to other tissues . The functional analysis also indicated that several metabolic pathways ( including cholesterol , ascorbate and aldarate , linoleic acid , and sphingolipid metabolism pathways ) and signaling pathways ( including GnRH , FoxO , cAMP and prolactin signaling pathways ) were significantly upregulated in heart 6 hr after an MI ( Supplementary file 2 , Figure 1—figure supplement 2 ) . We also observed significant down regulation of tryptophan metabolism and upregulation of glycosaminoglycan biosynthesis in heart 24 hr after an MI ( Supplementary file 2 , Figure 1—figure supplement 2 ) . Processes related to retinol metabolism were upregulated in heart at both timepoints . Pathways that were previously associated with cardiac hypertrophy and cardiac remodeling ( e . g . JAK-STAT , MAPK , estrogen , and TNF signaling pathways , and ECM-receptor interaction ) were significantly upregulated in heart 6 and 24 hr after an MI ( Figure 1—figure supplement 4 ) . Our analysis also indicated significant metabolic differences in adipose tissue 24 hr after an MI ( Figure 1—figure supplement 3 ) . Fructose and mannose metabolism , glyoxylate and dicarboxylate metabolism , glycolysis/gluconeogenesis , and pentose phosphate pathways , glycine , serine and threonine metabolism and pyrimidine metabolism , as well as endocrine systems ( e . g . insulin signaling pathway and regulation of lipolysis in adipocytes ) were downregulated in adipose tissue . We observed that the PPAR signaling pathway was upregulated , whereas glutathione was downregulated in liver 24 hr post-infarction ( Figure 1—figure supplement 3 ) . We found that sphingolipid metabolism and immune-related pathways were upregulated in skeletal muscle 24 hr post-infarction ( Figure 1—figure supplement 3 ) . To predict the effect of the transcriptional changes on metabolism , we performed reporter metabolite analyses ( Supplementary file 5 ) using the gene-to-metabolites mapping from the Mouse Metabolic Reaction database ( Mardinoglu , 2015 ) ; results in each tissue 24 hr after MI are shown in Figure 2B . In agreement with our analyses above , reporter metabolites related to oxidative phosphorylation , such as ubiquinol , ubiquinone , NADH and NAD+ , were downregulated in all tissues except liver . Moreover , linolenoyl-CoA , acetyl CoA , and several other fatty acyl-CoA-related metabolites were downregulated in heart and adipose tissue but upregulated in liver . We also found that several 5-S-glutathionyl metabolite forms , known to be related to phenylalanine , tyrosine and tryptophan biosynthesis , were downregulated in heart , liver , and skeletal muscle . The same pattern of downregulation was also observed for metabolites related to estrogen metabolism , specifically metabolites related to oestrone and its glutathione conjugate derivative . Moreover , 12-keto-LTB4 and 12-oxo-c-LTB3 , related to leukotriene metabolism , and hepoxilin A3 , an arachidonic acid , were also found to be downregulated in heart , liver , and skeletal muscle . The liver showed the highest alteration in reporter metabolites , which is attributed to its role as one of the most metabolically active tissues . We found that several reporter metabolites related to retinol metabolism , namely retinal , retinol , retinoate , and all-trans-18-hydroxyretinoic acid , were significantly downregulated only in liver tissue . Retinol metabolism has been previously associated with MI ( Lima et al . , 2018; Palace et al . , 1999 ) . The use of co-expression network ( CN ) analyses can assist in elucidating the functional relationships between genes in a specific cell and tissue ( Lee et al . , 2017 ) . Here , we performed CN analysis to reveal the functional relationship between the DEGs by generating tissue-specific CNs and selected highly connected genes ( the top 5% positively correlated genes that fulfilled FDR < 0 . 05 ) ( Table 1 ) . To better define the structure of the networks , we used the Leiden clustering algorithm ( Traag et al . , 2019 ) by maximizing the modularity scores ( Figure 3A–D ) and selected the clusters that include more than 30 genes . Next , we superimposed DEGs 24 hr post-infarction onto the network ( Supplementary file 1 ) and identified the components of the clusters that were affected by an MI . We also used functional analysis with GO BP and KEGG pathways to understand the specific functions associated with each cluster by using the Enrichr algorithm ( FDR < 0 . 05 ) ( Chen et al . , 2013; Kuleshov et al . , 2016 ) . We summarized the GO BP terms with Revigo ( Supplementary file 6; Supek et al . , 2011 ) and checked the average clustering coefficient to define the centrality of each cluster ( Supplementary file 6; Lee et al . , 2017 ) . Among the clusters , we identified the key clusters as those with the highest average clustering coefficient , allowing us to identify sets of genes whose time-dependent coordinated changes showed the strongest relationships . Interestingly , key clusters contained genes with similar functionalities including RNA processing , transports , and RNA metabolic processes in all tissue-specific CNs ( Supplementary file 6 ) . In addition , we found that the majority of the DEGs associated with those clusters were significantly upregulated . These observations strengthen the findings of the functional analysis above ( Figure 2A ) and further highlight how embryonically distinct tissues display similar functional responses to MI , with the most highly connected groups of genes preserved between different tissues ( Supplementary file 6 , Figure 3E ) . We investigated the tissue specificity of each cluster by performing enrichment analysis with data from the Mouse Gene Atlas ( Su et al . , 2004 ) , which involved counting the number of tissue-specific genes . The heart network showed the highest number of tissue-specific genes in cluster Heart-3 ( 302 genes ) . Based on DEG analysis , we found that 522 genes were downregulated and 192 genes were upregulated in the cluster . The enriched GO BP terms in the cluster were mitochondrial transport , protein processing and respiratory chain , cardiac muscle cell action potential , response to muscle stretch , and heart contraction ( Figure 3F ) . We observed that the results of the KEGG pathway enrichment analysis were consistent with those obtained from GO BP analysis ( Supplementary file 6 ) . In the liver network , cluster Liver-2 showed the highest tissue specificity ( 479 genes ) . In this cluster , we found that 15 genes were significantly downregulated and 17 genes were significantly upregulated . Based on GO BP enrichment analysis , the genes in this cluster were associated with cholesterol metabolism and homeostasis , lipid transport , glutathione metabolism , lipoprotein metabolism , and glucose 6-phosphate metabolism ( Supplementary file 6 ) . KEGG enrichment analysis also showed that the genes in the cluster were related to retinol , carbohydrate , lipid , and amino acid metabolism ( Supplementary file 6 ) . The muscle network had two clusters with high tissue specificity: cluster Muscle-4 ( 276 genes ) and Muscle-5 ( 143 genes ) . Muscle-4 showed association with GO BP terms such as mitochondrial transport , protein processing and respiratory chain , response to muscle stretch , and muscle contraction ( Supplementary file 6 ) . In contrast , the KEGG pathway in this cluster showed relation to glycolysis/glucogenesis , propanoate metabolism , glyoxylate and dicarboxylate metabolism , and several signaling pathways ( e . g . oxytocin , glucagon , cGMP-PKG , and HIF-1 ) ( Supplementary file 6 ) . Muscle-5 was enriched in GO BP terms associated with protein dephosphorylation , muscle contraction and intracellular protein transport ( Supplementary file 6 ) . We also found that insulin , MAPK and Wnt signaling pathways were associated to Muscle-5 from the KEGG enrichment analysis ( Supplementary file 6 ) . The adipose tissue network showed tissue specificity in cluster Adipose-2 ( 33 genes ) , which is associated with GO BP processes including mRNA processing , regulation of mitotic cell cycle phase , ribosome biogenesis , and viral processes ( Supplementary file 6 ) . We observed that the results of the KEGG pathway enrichment analysis were consistent with those obtained from GO BP analysis , with additional associations with multiple signaling and regulatory pathways ( Supplementary file 6 ) . To understand the specific behavior of each tissue , we further studied the tissue-specific clusters in the CNs ( Figure 4A ) . Heart specific cluster , Heart-3 , was driven by several central genes including Pln , Pde4b , and Atp2a2 ( related to regulation of cardiac muscle contraction ) and Pdha1 and Vdac1 ( related to mitochondrial functions ) . These genes were also found to be significantly differentially expressed in heart 24 hr post MI ( Supplementary file 1 ) . Genes in the heart-specific cluster were related to multiple other processes/pathways , for example oxytocin signaling pathway , and several metabolic pathways ( glycogen , inositol phosphate , and purine ) ( Supplementary file 6 ) . Mitochondrial dysfunction in the heart leads to disturbance of energy ( ATP ) production ( Kiyuna et al . , 2018; Palaniyandi et al . , 2010 ) and , in the presence of oxygen , to accumulation of reactive oxygen species ( ROS ) , which can cause oxidative stress . Vdac1 , a key gene for regulation of mitochondria function and one of the central genes in the heart-specific cluster ( see above ) , is significantly downregulated in MI ( Camara et al . , 2017 ) . Vdac1 is located in the outer mitochondrial membrane and is involved directly in cardioprotection ( Schwertz et al . , 2007 ) within the cGMP/PKG pathway ( Figure 4—figure supplement 1 ) . In the same pathway , we also observed down-regulation of the reporter metabolite hydrogen peroxide ( Supplementary file 5 ) , a ROS that is related to cardioprotection ( Schwertz et al . , 2007; Yada et al . , 2006 ) . We also observed downregulation of Pdha1 , which is known to have a substantial role in both the HIF-1 signaling pathway and the pyruvate metabolism pathway that converts pyruvate to acetyl-CoA in the mitochondria ( Figure 4—figure supplement 2 ) . Acetyl-CoA is used in the TCA cycle to produce NADH and FADH2 , which are both needed for ATP production and were downregulated in our reporter metabolite analysis of the heart . Our findings are thus consistent with dysfunctional mitochondria and ATP production in the heart in response to an MI . Pdha1 has been also been linked to the heart sensitivity during to ischemic stress , where its deficiency can compromise AMP-activated protein kinase activation ( Sun et al . , 2016 ) . In skeletal muscle and adipose tissue , we found that central genes in their respective tissue-specific clusters related to fatty acid metabolism and lipid metabolism were significantly altered ( Supplementary file 6 , Figure 5 ) . In liver-specific cluster , we found that their central genes were related to fatty-acid beta oxidation ( Cyp4a31 , Cyp4a32 ) and glutathione metabolism ( Gstm3 ) ( Supplementary file 6 , Figure 5A ) . Alterations of fatty acid beta-oxidation and glutathione metabolism have previously been reported in non-alcoholic fatty liver disease , a known risk factor of CVD ( Mardinoglu et al . , 2017b; Alexander et al . , 2019 ) . Moreover , in liver , we also found that retinol metabolism was uniquely related to genes in the liver-specific cluster , mainly driven by four significantly differentially expressed central genes of the clusters , that is Cyp26a1 , Cyp4a31 , Cyp4a32 , and Hsd17b6 ( Supplementary file 6 ) . A previous study showed that mortality from CVD in older individuals was accompanied by impaired liver ability to store retinol ( Lima et al . , 2018 ) . To investigate the metabolic responses to MI in and across tissues in the mice , we constructed a multi-tissue genome-scale metabolic model . The model consisted of five tissue-specific genome scale metabolic models , namely heart , liver , skeletal muscle , adipose , and small intestine . The small intestine model ( for which we do not have transcriptomic data ) was added to include ingestion and conversion of dietary nutrients into chylomicrons , which are directly secreted into blood and transport lipids to other tissues ( Mardinoglu , 2015 ) . The final mouse multi-tissue model included 19 , 859 reactions , 13 , 284 metabolites , 7116 genes , and 41 compartments . We predicted the metabolic fluxes in mice 24 hr after an MI or sham operation by integrating the dietary input , tissue-specific resting energy expenditure and transcriptomics data . The modeling showed that oxygen uptake , carbon dioxide production and the oxidative phosphorylation pathway in heart , adipose and skeletal muscle were decreased in MI mice , in agreement with the downregulation of oxidative phosphorylation we observed in these tissues ( Supplementary file 7 ) . By contrast , liver showed slightly increased oxygen uptake , which might be due to the slightly ( not statistically significant ) upregulated oxidative phosphorylation ( Supplementary file 7 ) . These findings indicate that the changes in oxygen and carbon dioxide fluxes and the oxidative phosphorylation pathway could serve as a positive control for predicting the changes due to MI in the fluxes . Next , we investigated the tissue-specific metabolic flux changes in the same model ( Supplementary file 7 ) . We found that the pentose phosphate pathway was upregulated in heart 24 hr post MI , consistent with upregulated glucose metabolism after an MI . Elevated glycolysis could allow the heart to rapidly generate energy under stress conditions , and the enhanced pentose phosphate pathway could increase the NADPH level , which could help maintain the level of reduced glutathione in heart ( Tran and Wang , 2019 ) . In addition , we observed an increase uptake of alpha-ketoglutarate ( AKG ) of heart 24 hr after MI . It has been reported that supplementation of AKG could prevent heart from ischaemic injury ( Kjellman et al . , 1995 ) , and the increased uptake of AKG we observed after MI might be a natural protective metabolic response to MI . Moreover , we found there is a net lactate metabolic flux coming from liver to heart in the MI group . The influx of lactate has been reported to be positively correlated with the fraction of regional ejection of heart ( Hattori et al . , 1985 ) and this net flux not only agrees well with the previous report but also additionally suggested the source of the lactate . We also found that adipose tissue secreted more ketone bodies , including acetoacetate and butyrate , into plasma; the plasma level of ketone bodies has been reported as a stress marker in acute MI ( Miyamoto et al . , 1999 ) . Notably , relatively small metabolic changes were found in liver and skeletal muscle , which is probably due to the small number of transcriptomic changes in metabolic pathways in these tissues . We validated our observations in heart tissue in two independent cohorts of bulk RNA-seq data from mouse heart ( Supplementary file 8 ) . We filtered both validation cohorts to get and analyzed only 24 hr post-MI data . We found that there were 2169 DEGs from our heart 24 hr post MI data were validated in at least one of the independent cohorts ( 959 DEGs validated in both ) ( Figure 6A ) . We also found that 109 out of the 123 most connected genes in our heart-specific cluster were also significantly differentially expressed in at least one of the independent cohorts ( 81 in both ) . By performing functional analysis of the validation cohorts , we found that ~61% of GO BP and 84% of KEGG pathways identified in our analysis of the heart were also present in at least one of the validation cohorts 24 hr after infarction ( Figure 6B–C ) . In both cohorts , we observed downregulation of mitochondrial functions and fatty acid metabolism processes . We also observed upregulation of processes and pathways related to retinol metabolism and inflammatory response in both validation cohorts . We observed that Flnc , Lgals3 , Prkaca , and Pprc1 showed important role to MI . These genes were 4 of 16 genes that were DEGs in at least three tissues and validated in both validation cohorts ( Supplementary file 9 ) . Flnc , Lgals3 , and Pprc1 were upregulated in heart , skeletal muscle , and adipose , whereas Prkaca was downregulated in these three tissues . We further retrieved their neighbors at each tissue specific CNs , showed their regulations from differential expression results , and performed functional analysis in Supplementary file 9 . Flnc , which encodes filamin-C , was part of heart and skeletal muscle-specific CN cluster ( Figure S4 ) . Its neighbor genes were found to be significantly ( FDR < 0 . 05 ) associated to several functions , including TCA cycle , pyruvate metabolism , glycolysis pathway , and involved in mitochondrial functions . Specifically , they were related to heart-specific processes in heart , VEGF signaling pathway in muscle , carbohydrate metabolism in adipose , and to MAPK signaling pathway and muscle contraction in heart and muscle . Lgals3 ( encodes galectin-3 ) and Prkaca were among the most central genes in central clusters ( Supplementary file 6 ) . The neighbors of Lgals3 were significantly related to cell cycle and protein digestion and absorption pathway in all tissues , and to RNA and mRNA related-processes in muscle and adipose tissue . The neighbors of Prkaca were related to insulin signaling pathway in heart and adipose , and several mitochondrial functions in adipose . Pprc1 was part of most central clusters in heart and adipose tissue CN , and its neighbors were related to ribosomal RNA processing and ribosome biogenesis .
CVD has a complex etiology and is responsible for a range of systemic effects , hindering our understanding of its consequences on different tissues . Here , we took advantage of the technological advances in high-throughput RNA-seq and applied integrative network analyses to comprehensively explore the underlying biological effects of MI . Specifically , we generated RNA-seq data from heart , liver , skeletal muscle , and adipose tissue obtained from mice 6 and 24 hr after an MI or sham operation . We used transcriptomics data analyses ( differential expression , functional analysis , and reporter metabolites analysis ) to determine the systemic effects of the MI across multiple tissues . Moreover , we performed CN analyses to pinpoint important key and tissue-specific clusters in each tissue , and identified the key genes in each cluster . Finally , we used a whole-body modeling approach to identify the crosstalk between tissues and reveal the global metabolic alterations , before finally validating our findings with publicly available independent MI cohorts . Based on our analyses , we observed downregulation of heart-specific functions and upregulation of lipid metabolism and inflammatory response in heart , muscle , and adipose tissue after an MI ( Figure 4B ) . Liver showed a distinct response with respect to the other three tissues , including downregulation of inflammatory response . We observed that fatty acid metabolism was downregulated in heart and adipose tissue , whereas fatty acid beta-oxidation was upregulated and glutathione metabolism was downregulated in liver . We also observed upregulation of oxidative stress in heart and skeletal muscle . We also observed downregulation of mitochondrial functions in heart , muscle , and adipose tissue . Furthermore , we found upregulation of retinol metabolism in heart and downregulation of retinol metabolites in liver ( Figure 4B ) . We hypothesized that downregulation of fatty acid metabolism from adipose tissue was due to exchange of fatty acids with other tissues ( liver and muscle ) ( Figure 4B ) . We also observed the flow of retinol from liver to heart during MI , consistent with previous reports ( Palace et al . , 1999 ) . These MI-associated alterations lead to dysfunctional mitochondria and decreased energy production , especially in heart and skeletal muscle . We also validated our results with publicly available MI datasets generated in separate independent studies . The validation results strengthened our findings on the altered functions/pathways and the important heart-specific genes after an MI . Importantly , our analyses of gene clusters highlighted multiple key genes in the response to MI in different tissues . Specifically , we observed that Flnc , Prkaca , Lgals3 , and Pprc1 showed important responses in heart , skeletal muscle , and adipose tissue . Flnc is involved in actin cytoskeleton organization in heart and skeletal muscle , and previous studies have shown that this gene has critical role in CVD ( Zhou et al . , 2020; Hall et al . , 2020 ) . Similarly , Prkaca , an important metabolic gene , has also been shown to play an important function during CVD ( Diviani et al . , 2011; Turnham and Scott , 2016; Bers , 2008 ) . Lgals3 , related to acute inflammation response , has been studied intensively in recent years as a key gene in CVD , and as a potential CVD therapy target ( Zhong et al . , 2019; Suthahar et al . , 2018 ) . Lastly , Pprc1 , as important regulator of mitochondrial biogenesis , has not been explored for its direct relationship with CVD; however , mitochondrial biogenesis appears to be an important response to CVD ( Ren et al . , 2010; Siasos et al . , 2018; Piantadosi and Suliman , 2012 ) . We recognized several limitations to be noted on this research . First , only transcriptomic data was analyzed in this research , hence the sensitivity might be limited especially for short timepoint , for example 6 hr after MI . Second , we focused our analysis in this research only on protein-coding genes . Third , to explore more about the shift in metabolism due to MI , longer timepoints needs to be explored . This opens new opportunities for future research , including analyzing the non-protein-coding gene signatures and longer timepoints . In summary , we systematically unveiled the deregulation of biological processes and pathways that resulted from MI in heart , liver , muscle , and adipose tissue by integrating transcriptomic data and the use of biological networks . We also identified the key clusters and central genes using generated tissue-specific CNs . In this study , we demonstrated a strategy to utilize multi-tissue transcriptomic data to identify alteration of biological processes and pathways to systemically explore the effect of a disease .
Ten-week-old male C57Bl/6N mice were fasted for 4 hr before induction of myocardial infarction . The mice were then anesthetized with isoflurane , orally intubated , and connected to a small-animal ventilator ( SAR-830 , Geneq , Montreal , Canada ) distributing a mixture of oxygen , air and 2–3% isoflurane . ECG electrodes were placed on the extremities , and cardiac rhythm was monitored during surgery . An incision was made between the 4th and 5th ribs to reveal the upper part of the anterior left ventricle ( LV ) wall and the lower part of the left atrium . Myocardial infarction was induced by ligating the left anterior descending ( LAD ) coronary artery immediately after the bifurcation of the left coronary artery 1 . The efficacy of the procedure was immediately verified by characteristic ECG changes , and akinesis of the LV anterior wall . After verification of the infarction , the lungs were hyperinflated , positive end-expiratory pressure was applied , and the chest was closed . Sham mice were handled identically ( fasted , anesthetized , intubated , and connected to ventilator , and subsequently incised between 4th and 5th ribs ) , but no ligation of the LAD coronary artery was performed ( and thus , no ischemia was induced in these mice ) . The mice received an intraperitoneal injection of 0 . 1 ml buprenorphine to relieve postoperative pain and were allowed to recover spontaneously after stopping isoflurane administration . Mice were killed with an overdose of isoflurane 6 hr or 24 hr after occlusion or sham operation . We collected the left ventricle ( the whole left ventricle containing mainly infarcted tissue ) of the heart , whereas white adipose tissue ( WAT ) was collected from the abdomen and musculus soleus was taken as the muscle tissue . Mouse hearts and biopsies from the liver , muscle and WAT were snap-frozen in liquid nitrogen and stored at −80°C until analysis . All mice studies were approved by the local animal ethics committee and conform to the guidelines from Directive 2010/63/EU of the European Parliament on the protection of animals used for scientific purposes . Echocardiographic examination , using VisualSonics VEVO 2100 system ( VisualSonics Inc , Ontario , Canada ) , which includes an integrated rail system for consistent positioning of the ultrasound probe was performed 6 and 24 hr after an MI to determine the size of the MI . We calculated infarct size based on wall motion score index ( WMSI ) 24 hr after myocardial infarction by a 16-segments model on three short axis images , as 0 for normal , ½ for reduced wall thickening and excursion in a segment and one for no wall thickening and excursion in a segment . WMSI was calculated as the sum of scores divided by the total number of segments . Hair removal gel was applied to isofluorane-anesthetized ( 1 . 2% ) mice chest to minimize resistance to ultrasonic beam transmission . The mice were then placed on a heating pad and extremities were connected to an ECG . A 55 MHz linear transducer ( MS550D ) was used for imaging . An optimal parasternal long axis ( LAX ) cine loop of >1000 frames/s was acquired using the ECG-gated kilohertz visualization technique . Parasternal short axis cine-loops were acquired at 1 , 3 , and 5 mm below the mitral annulus . Infarct size was calculated based on wall motion score index 6 and 24 hr after myocardial infarction by a 16-segments model on LAX and three short axis images view , as 0 for normal , ½ for reduced wall thickening and excursion in a segment and one for no wall thickening and excursion in a segment . The data were evaluated using VevoStrain software system ( VisualSonics Inc , Ontario , Canada ) . Total RNA was isolated from snap-frozen tissues using RNeasy Fibrous Tissue Mini Kit ( Qiagen ) for heart and skeletal muscle , RNeasy Mini Kit ( Qiagen ) for liver , or RNeasy Lipid Tissue Mini Kit ( Qiagen ) for adipose tissue . cDNA was synthesized with the high-capacity cDNA Reverse Transcription Kit ( Applied Biosystems ) and random primers . mRNA expression of genes of interest was analyzed with TaqMan real-time PCR in a ViiA seven system ( Applied Biosystems ) . RNA sequencing library were prepared with Illumina RNA-Seq with Poly-A selections . Subsequently , the libraries were sequenced on NovaSeq6000 ( NovaSeq Control Software 1 . 6 . 0/RNA v3 . 4 . 4 ) with a 2 × 51 setup using ‘NovaSeqXp’ workflow in ‘S1’ mode flow cell . The Bcl was converted to FastQ by bcl2fastq_v2 . 19 . 1 . 403 from CASAVA software suite ( Sanger/phred33/Illumina 1 . 8 + quality scale ) . The raw RNA-sequencing results were processed using Kallisto ( Bray et al . , 2016 ) with index file generated from the Ensembl mouse reference genome ( Release-96 ) ( Zerbino et al . , 2018 ) . The output from Kallisto , both estimated count and TPM ( Trancript per kilobase million ) , were subsequently mapped to gene using the mapping file retrieved from Ensembl BioMart website , by filtering only protein coding genes and transcripts . Genes with mean expression less than 1 TPM in each condition were filtered . For data exploration , we used PCA from sklearn package ( Pedregosa , 2011 ) in Python 3 . 7 and used TPM values as the input . Subsequently , we performed differential gene expression analysis using DESeq2 ( Love et al . , 2014 ) package in R . We utilized the capabilities from DESeq2 to normalize the rounded estimated count data and to correct for confounding factors ( such as time ) . To define a gene as differentially expressed ( DEGs ) , a gene has to fulfill a criterion of FDR < 5% . The results of differential expression analysis were then used for functional analysis . We checked the tissue specificity of the DEGs in each tissue with the data from Mouse Gene Atlas ( Su et al . , 2004 ) . For all the tissue-specific genes , we also checked their human-homolog genes in the human secretome database ( Uhlén et al . , 2019 ) . We performed functional analysis using the R package PIANO ( Väremo et al . , 2013 ) . As the input , we used the fold changes and p-values from the DESeq2 , and also GO BP and KEGG pathways gene-set collections from Enrichr ( Chen et al . , 2013; Kuleshov et al . , 2016 ) , and metabolites from Mouse Metabolic Reaction database ( Mardinoglu , 2015 ) . To define a process or pathway as significant , we used a cut off of FDR < 5% for the distinct direction of PIANO ( both up and down ) . We generated the co-expression network by generating gene-gene Spearman correlation ranks within a tissue type , using spearmanr function from SciPy ( Jones et al . , 2001 ) in Python 3 . 7 . Using the same environment , we performed multiple hypothesis testing using Benjamini-Hochberg method from statsmodels ( Perktold et al . , 2017 ) . Correlation data were filtered with criterion of adjusted p-value<5% . The top 5% of filtered correlation results were then loaded into iGraph module ( Csardi and Nepusz , 2006 ) in Python 3 . 7 as an unweighted network . To find the subnetworks , we employed the Leiden clustering algorithm ( Traag et al . , 2019 ) with ModularityVertexPartition method . Each cluster was analyzed by using Enrichr ( Chen et al . , 2013; Kuleshov et al . , 2016 ) to get the enriched GO BP and KEGG pathways . Criterion FDR < 0 . 05 were used to find the significantly enriched terms . Clusters with less than 30 genes were discarded , to be able to get significant functional analysis results . Since GO BP was relatively sparse , we used Revigo ( Supek et al . , 2011 ) to summarize the GO BP into a higher level . Revigo was further employed to build a GO BP network . Clustering coefficient was calculated based on the average local clustering coefficient function within iGraph . We combined tissue-specific models ( of heart , liver , muscle , adipose and small intestine ) constructed previously ( Mardinoglu , 2015 ) in a multi-tissue model by adding an additional compartment representing the plasma , which allows the exchange of metabolites among different tissues . Blocked reactions that could not carry fluxes ( and the unused metabolites and genes linked to these reactions ) were removed from the models . In addition , the dietary input reactions and constraints were added to the small intestine model to simulate the food intake ( Supplementary file 7 ) . Specifically , we assumed that the mice weighed 30 g and consumed 4 . 5 g chow diet per day ( 15 g/100 g body weight ) based on a previous study ( Kummitha et al . , 2014 ) . We also calculated the tissue-specific resting energy expenditures and set them as mandatory metabolic constraints based on previous studies and resting energy expenditure for other tissues was incorporated by including a mandatory glucose secretion flux out from the system with the lower bound calculated based on ATP ( Supplementary file 7; Kummitha et al . , 2014 ) . To simulate the metabolic flux distribution in the sham-operated mice , we set the lipid droplet accumulation reaction in adipose tissue ( m3_Adipose_LD_pool ) as the objective function as we assume the energy additional to the resting energy expenditure will be mostly stored as fat rather than used by the muscle for physical activities because mice raised in the cages might have very little exercise . Then , we used parsimonious FBA to calculate the flux distribution . To simulate the flux distribution after an MI , we calculated an expected flux fold change of each reaction based on the FDR and expression fold changes of all genes associated with the reaction , and obtains a flux distribution that is closest to this expected flux distribution while satisfying the stoichiometric balance and flux constraints of the model . The mathematical formulation of the method is described as below , minimizeZ=∑i|vi−viexp|s . t . S*v=0lb≤v≤ubwhere S , v , lb , ub represent the stoichiometric matrix , flux distribution , lower bound and upper bound of all reactions , respectively . The viexp represents the expected flux of ith reaction which is calculated as follows , viexp=viref*∏j=1mFCjmwhere n is the number of gene sets that could independently catalyze the corresponding reaction , and FCj represents the expected expression changes of jth gene set which is calculated below , FCj=1-P1*fc1+P11-P2*fc2+⋯+∏k=1m-1Pk1-Pn*fcnwhere m is the number of genes in the ith gene sets , and Pj and fcj respectively represents the FDR and fold change of gene expression with jth smallest fold change in this gene set . In this way , genes with lowest fold change will have a dominating effect within a gene set encoding a protein complex , while the geometric mean of expected fold changes of gene sets encoding different isozymes of this reaction will be used as the final expected flux fold change of this reaction . We validated our findings by performing similar steps of RNA sequencing and functional analysis for the publicly available mouse MI datasets GSE104187 and GSE52313 ( Ounzain et al . , 2015; Williams et al . , 2018 ) . All raw RNA-sequencing data generated from this study can be accessed through accession number GSE153485 . Codes used during the analysis are available on https://github . com/sysmedicine/ArifEtAll_2020_MultiTissueMI ( copy archeived at swh:1:rev:e79df3ef069674c1344c096ef6b011e771cf506b; Arif , 2021 ) . | The human body is like a state-of-the-art car , where each part must work together with all the others . When a car breaks down , most of the time the problem is not isolated to only one part , as it is an interconnected system . Diseases in the human body can also have systemic effects , so it is important to study their implications throughout the body . Most studies of heart attacks focus on the direct impact on the heart and the cardiovascular system . Learning more about how heart attacks affect rest of the body may help scientists identify heart attacks early or create improved treatments . Arif and Klevstig et al . show that heart attacks affect the metabolism throughout the body . In the experiments , mice underwent a procedure that mimics either a heart attack or a fake procedure . Then , Arif and Klevstig et al . compared the activity of genes in the heart , muscle , liver and fat tissue of the two groups of mice 6- and 24-hours after the operations . This revealed disruptions in the immune system , metabolism and the production of proteins . The experiments also showed that changes in the activity of four important genes are key to these changes . This suggests that this pattern of changes could be used as a way to identify heart attacks . The experiments show that heart attacks have important effects throughout the body , especially on metabolism . These discoveries may help scientists learn more about the underlying biological processes and develop new treatments that prevent the harmful systemic effects of heart attacks and boost recovery . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"medicine",
"computational",
"and",
"systems",
"biology"
] | 2021 | Integrative transcriptomic analysis of tissue-specific metabolic crosstalk after myocardial infarction |
Inducible epigenetic changes in eukaryotes are believed to enable rapid adaptation to environmental fluctuations . We have found distinct regions of the Arabidopsis genome that are susceptible to DNA ( de ) methylation in response to hyperosmotic stress . The stress-induced epigenetic changes are associated with conditionally heritable adaptive phenotypic stress responses . However , these stress responses are primarily transmitted to the next generation through the female lineage due to widespread DNA glycosylase activity in the male germline , and extensively reset in the absence of stress . Using the CNI1/ATL31 locus as an example , we demonstrate that epigenetically targeted sequences function as distantly-acting control elements of antisense long non-coding RNAs , which in turn regulate targeted gene expression in response to stress . Collectively , our findings reveal that plants use a highly dynamic maternal ‘short-term stress memory’ with which to respond to adverse external conditions . This transient memory relies on the DNA methylation machinery and associated transcriptional changes to extend the phenotypic plasticity accessible to the immediate offspring .
While genetic variation is the primary source of long-term adaptation and evolution , numerous studies have pointed to induced epigenetic changes to facilitate rapid adaptation to short-term environmental fluctuations ( Franks and Hoffmann , 2012 ) . Because plants are sessile organisms , it has been suggested that they can efficiently integrate environmental signals into a ‘stress memory’ that is transmitted to the immediate progeny . This newly acquired information could allow populations to respond efficiently to repeated exposure to the same stress , a phenomenon known as 'priming' or 'acclimation' ( Crisp et al . , 2016; Conrath et al . , 2006; Sani et al . , 2013; Slaughter et al . , 2012 ) . It has recently been proposed that epigenetic marks could be induced depending on the consistency of the cues that individuals perceive directly from the environment ( Uller et al . , 2015 ) . Yet it remains unclear how often this actually occurs in nature , and whether such adaptive responses can be transmitted over multiple non-stressed generations , a phenomenon termed “transgenerational stress memory” ( Hauser et al . , 2011; Paszkowski and Grossniklaus , 2011 ) . Stress memory in plants is believed to be mostly epigenetic in nature , because priming responses have been associated with changes in chromatin and DNA methylation ( Hauser et al . , 2011; Ito et al . , 2011 ) . Genome-wide studies in plants have shown that environmental stress dynamically modifies the chromatin landscape , creating novel patterns of gene expression , and thereby affecting short-term adaptation to stress ( Sani et al . , 2013 ) . Moreover , heritable traits resulting from environmental stress have been associated with DNA methylation changes in promoter regions ( Bilichak et al . , 2012; Le et al . , 2014 ) , gene-coding regions ( Bilichak et al . , 2012; Jiang et al . , 2014 ) , transgenes ( Lang-Mladek et al . , 2010; Molinier et al . , 2006 ) and transposable elements ( TEs ) ( Boyko et al . , 2010; Le et al . , 2014; Secco et al . , 2015 ) . Although TEs in plants are often mutagenic , they are nonetheless deemed to be potentially beneficial for regulating gene expression in response to a wide range of biotic and abiotic stresses . For instance , pathogen attack in Arabidopsis can cause changes in DNA methylation primarily at TEs and repeats located proximal to genes associated with transcriptional defense responses ( Dowen et al . , 2012 ) , while temperature stress can trigger specific TE activation that , when inserted near genes , can confer stress-mediated transcriptional responses ( Cavrak et al . , 2014; Naito et al . , 2009; Yasuda et al . , 2013 ) . Similarly in rice , phosphate starvation alters methylation at TEs near environmentally induced genes ( Secco et al . , 2015 ) . These findings , which are reminiscent of the domestication of viral DNA for human immunity ( Chuong et al . , 2016 ) , support the view that TEs play pivotal roles in environmental stress-sensing ( Kalendar et al . , 2000 ) . The precise mechanism by which this occurs remains elusive . It is possible that the specific repetitive sequences present in transposons can generate and/or be recognised by stress-induced small non-coding RNAs that direct de novo DNA methylation ( Law and Jacobsen , 2010 ) . Alternatively , DNA methylation could be affected by stress independently of small RNAs through the RDR6-RdDM pathway ( Nuthikattu et al . , 2013 ) or be targeted and demethylated by DNA glycosylases ( Kim and Zilberman , 2014; Zhang and Zhu , 2012 ) , thereby imparting dynamic transcriptional changes at neighbouring genes . This is supported by observations of mutants defective for genes of these epigenetic pathways , and which have impaired biotic and abiotic stress responses ( Boyko et al . , 2010; Ito et al . , 2011; Le et al . , 2014; Luna and Ton , 2012 ) . While the somatic stability of environmentally-induced epigenetic changes is well documented , robust evidence for their sexual transmission to the next generation is rare in both animals and plants ( Heard and Martienssen , 2014; Paszkowski and Grossniklaus , 2011 ) . During the initial stages of sexual reproduction , an active epigenetic reprogramming takes place in plant gametes as a means of silencing transposons ( Gutierrez-Marcos and Dickinson , 2012; Kawashima and Berger , 2014 ) , but it is not known whether this process also affects the transmission of environmentally-induced alterations in DNA methylation . Here we report that repeated hyperosmotic stress induces DNA methylation changes that primarily affect epigenetically labile regions of the Arabidopsis genome , i . e . , regions that are susceptible to changes in methylation status . Some of these changes are transmitted to the offspring , where they affect the transcriptional regulation of a small group of genes associated with enhanced tolerance to environmental stress . In the absence of a renewed stress stimulus , the acquired epigenetic and phenotypic changes are gradually reset in subsequent generations . Further , newly acquired stress tolerance and associated de novo DNA methylation marks are preferentially transmitted through the female germline . Epigenetic inheritance relies on DNA methylation changes at sequences that function as distantly acting control elements of key stress-response regulators , including antisense long non-coding RNAs ( lncRNAs ) . Collectively , our data provide a new mechanistic model for the establishment and transient inheritance of plant stress adaptation .
To evaluate the extent to which stress-induced transgenerational adaptation occurs , we first exposed Arabidopsis plants ( Arabidopsis thaliana accession Col-0 ) to two different hyperosmotic conditions for over five generations ( P0 of G1-G5 , Figure 1A ) ( see Materials and methods for details ) . Plants were subjected to stress only during the vegetative phase and were transferred to normal soil before most flowers formed; thereby reducing the possibility of parental stress exerting a direct effect on gametes ( see Materials and methods for details ) . To uncouple parent and progeny environments , we grew seeds derived from G1-G5 treated and control plants for two additional generations ( P1 and P2 ) without stress ( Figure 1A ) . Germination and survival rates of the progeny were then assessed in three independent experiments for all generations under stress and control conditions . Similar to non-treated plants , the progeny derived from the first generation of stressed plants ( G1 ) did not display significant signs of adaptation . In contrast , the direct progeny ( P1 ) of G2-G5 stressed plants displayed higher germination and survival rates , and more robust vegetative growth on high-salinity ( 150 mM ) medium ( Figure 1B–C , Figure 1—figure supplement 1 and Supplementary file 1 ) , suggesting that hyperosmotic priming requires repetitive exposure to stress . Notably , adaptation was already lost in the second-generation progeny ( P2 ) of G2-G5 stressed plants . Thus , two generations of a stress-free environment were sufficient to revert the stress-induced changes ( Figure 1C and Figure 1—figure supplement 1 ) . Our data thus suggest that recurrent hyperosmotic stress in plants induces intergenerational adaptation , but that this response does not persist in the absence of stress . Several studies have suggested that environmental stress induces genome-wide epigenetic changes that can be transmitted to the offspring ( Boyko et al . , 2010; Jiang et al . , 2014; Luna et al . , 2012; Rasmann et al . , 2012; Slaughter et al . , 2012 ) . For this to occur , such changes must escape the epigenetic reprogramming that takes place during sexual reproduction ( Calarco et al . , 2012; Ibarra et al . , 2012 ) . Therefore , to confirm that the adaptive responses seen in the offspring of stressed parents were due to newly acquired genome-wide epigenetic changes , we grew well-characterized epigenetic mutants that are defective in RNA-directed DNA methylation ( RdDM ) or in the active removal of DNA methylation ( Law and Jacobsen , 2010; Zhang and Zhu , 2012 ) , and exposed them to hyperosmotic stress for two successive generations . We assessed the progeny for enhanced tolerance to high salinity in three independent experiments ( Figure 1D ) . In contrast to progeny of stressed wild-type plants , immediate progeny of stressed nrpd1a ( Herr et al . , 2005 ) cmt3 ( Chan et al . , 2006 ) and ros1/dml2/dml3 ( rdd ) ( Penterman et al . , 2007 ) plants did not show enhanced survival under hyperosmotic stress conditions ( Supplementary file 1 ) . These data imply that transgenerational adaptation to hyperosmotic stress relies in part on the DNA methylation machinery , although these phenotypic data do not reveal how extensive the epigenetic changes are . To determine the primary genomic targets susceptible to epigenetic changes induced by hyperosmosis , we performed whole-genome bisulfite sequencing ( Supplementary file 2 ) . To ensure statistically robust results , we excluded inter-individual epigenetic variation that can arise over the course of several generations ( Becker et al . , 2011; Schmitz et al . , 2011 ) by collecting duplicate samples of leaf tissue from 10 plants each from the G1 , G3 and G5 generations for each treatment ( control , 25 mM- , and 75 mM-NaCl ) ( Figure 1A ) . We sought to compare DNA methylation patterns for the different treatments in non-stressed P1 and P2 progeny derived from control or salt-stressed P0 parents ( Figure 1A ) . Individual cytosines with a significantly altered methylation frequency , termed differentially methylated positions ( DMPs ) , were first identified by pairwise comparisons between two samples . Because the three stressed generations had been grown and treated at different time points , we only compared samples belonging to the same treatment group , thus excluding methylation changes that were due to stochastic fluctuations in growth conditions . Single-site polymorphisms between any two samples were rare , with on average 6 , 866 DMPs ( 40% CG , 15% CHG and 45% CHH ) detected per generation in all pairwise comparisons ( Supplementary file 3 ) . Principal component analysis ( PCA ) and complete linkage clustering of methylation frequencies grouped all stress-treated samples ( P0 ) , separating them from control , P1 , and P2 samples ( Figure 2—figure supplement 1A–B ) . This indicated that hyperosmotic stress had a small but noticeable effect on single methylated cytosines , and that this effect was largely transient . Overall we observed three times more methylation gains than losses in salt-treated P0 plants compared to the control ( Figure 2—figure supplement 1C ) . Because we found considerably fewer DMPs than recently reported for multi-generational hyperosmotic stress treatments ( Jiang et al . , 2014 ) , we re-analysed the published data . We found that only a small fraction of the DNA methylation changes reported by Jiang et al . ( 2014 ) were consistently induced by hyperosmotic stress ( Figure 2—figure supplement 1D ) . The properties of DMPs are distinct from those of differentially methylated regions ( DMRs , i . e . , contiguous stretches of methylation change ) , as DMPs mostly occur at sparsely distributed CG sites within gene bodies , whereas DMRs tend to occur in densely methylated areas of mixed methylation context ( Hagmann et al . , 2015 ) . To identify stress-induced DMRs , we used a statistically robust Hidden Markov Model-based algorithm that supports the confident detection of differential methylation also in CHG and CHH contexts ( Hagmann et al . , 2015 ) . We identified on average 24 , 700 methylated regions ( MRs ) per sample with a median length of 272 bp ( mean: 856 bp ) . To identify generation-specific and treatment-dependent DMRs , we considered samples of the same generation ( G1 , G3 or G5 , Figure 1A ) and treatment regime ( control or salt-treated ) as replicates ( Supplementary file 4 ) . For all three generations , DMRs mapped mainly to TEs and intergenic regions , and were three- to seven-fold over-represented in 2-kb regions upstream of transcription start sites compared with overall methylation in MRs ( Figure 2A ) . No significant differences were found in the average methylation frequencies at MRs in control and stressed samples ( Figure 2—figure supplement 2 ) , or for CG methylation frequencies at DMRs in control and stress-treated samples ( P > 0 . 05 in all generations , unpaired two-tailed Student’s t-test ) ( Figure 2B; Figure 2—figure supplement 3 ) . However , methylation in CHG and CHH contexts at DMRs was significantly altered in stress-treated P0 versus control P0 samples ( P < 0 . 01 in all generations , unpaired two-tailed Student’s t-test ) ( Figure 2B; Figure 2—figure supplement 3 ) . Complete linkage clustering of DMRs grouped salt-treated P0 samples in all three generations ( Figure 2C ) , similar to the clustering and Principal Component Analysis ( PCA ) on single polymorphic sites ( Figure 2—figure supplement 1A–B ) . For G1 , the cluster comprising the non-stressed samples did not have any clear substructure . By contrast , for G3 and G5 , the P1-descendants of salt-treated P0 plants formed a clear sub-group distinct from the control and P2 plants ( Figure 2C ) . These data concur with the adaptive responses we observed specifically in the P1 , but not in the P2 progeny of G3 and G5 salt-treated plants ( Figure 1C; Figure 1—figure supplement 1A–B ) . P2-descendants showed methylation patterns similar to control plants , which correlated with the observed lack of high salinity tolerance . We also analysed published DNA methylation data derived from individual plants subjected to hyperosmotic stress ( Jiang et al . , 2014 ) , focussing on DMRs , and considering individual samples as replicates . We confirmed our finding that the methylation changes in CHG and CHH correlated well with stress treatment , whereas changes in CG methylation did not , indicating that CG methylation patterns occur stochastically in treated and non-treated samples ( Figure 2—figure supplement 1E ) . Thus , hyperosmotic stress directs DNA methylation changes primarily at non-CG sites located in intergenic TE-related sequences , and these epigenetic modifications are associated with an acquired transient adaptation to stress . Although a large fraction of DMRs appeared to arise as a consequence of salt treatment , only a few recurred in G3 and G5 , or overlapped between all three generations ( Figure 2D ) . To determine the significance of these hyperosmotic-stress DMRs ( HS-DMRs ) , we asked whether they were also responsive to other abiotic stresses , such as cold treatment . Comparing our data with a small set of DMRs in cold-stressed Arabidopsis seedlings ( Seymour et al . , 2014 ) , we did not detect any overlap between the two datasets . By contrast , 49% of HS-DMRs in stressed plants ( P0 ) overlapped with or were in close proximity ( <500 bp ) to DMRs that had been reported to arise spontaneously in mutation accumulation ( MA ) lines that had been grown over 30 generations under controlled conditions ( Figure 2D ) ( Becker et al . , 2011; Hagmann et al . , 2015 ) . These spontaneous MA-DMRs are often found in more than one individual , pointing to specific regions of the genome being particularly susceptible to epigenetic reprogramming . We compared pools of ten plants to identify HS-DMRs , while individual plants had been compared to define MA-DMRs . That there is nevertheless substantial overlap indicates that stress-triggered epigenetic reprogramming of specific genomic regions is not entirely random . To determine whether the DMRs found in both the hyperosmotic stressed plants and MA lines differed from those others found in only one of the two populations , we performed complete linkage clustering on both sets . Overlapping and non-overlapping DMRs behaved similarly for G3 and G5 ( but not G1 ) , with stressed samples clustering in one group and control and untreated progeny in another . This indicated that overlapping and non-overlapping DMRs carried a similar hyperosmotic stress signature ( Figure 2E; Figure 2—figure supplement 4 ) , but the two DMR classes differed in their association with adjacent annotated genes ( Figure 2—figure supplement 5 ) . In particular , HS-DMRs that did not overlap with spontaneous MA-DMRs were enriched near genes with functions related to metabolic responses and ion transport ( Figure 2—figure supplement 5 ) . These data thus suggest that exposure to hyperosmotic stress targets discrete , epigenetically labile regions of the genome . In contrast to previous reports suggesting wholesale DNA methylation changes induced by environmental stress ( Boyko et al . , 2010; Dowen et al . , 2012; Jiang et al . , 2014 ) , we noticed that methylation frequencies within HS-DMRs differed between generations . Most HS-DMRs ( 81% ) in G3 and G5 were hypermethylated in the hyperosmotic-treated P0 samples , whereas G1 plants had similar number of hyper- and hypo-methylated DMRs ( Figure 3A ) . Further , in progenies of stress-treated plants , the CHG and CHH methylation changes were gradually lost and reverted to the control states , as seen in the P1 and P2 progeny grown in the absence of stress ( Figure 3A–B; Figure 3—figure supplement 1 ) . Hyper- and hypo-methylated HS-DMRs mapped to different genomic regions , with methylation gain after hyperosmotic stress frequently found within or proximal to TEs , but methylation loss occurring more frequently near genes ( Figure 3C ) . Hyper-methylated HS-DMRs were significantly enriched near Helitrons ( unpaired Student’s t-test; **p<0 . 01 ) ( Figure 3D ) , a TE family known to be targeted by RdDM ( Nuthikattu et al . , 2013 ) , and near genes involved in RNA-directed DNA polymerase and reverse transcription activities ( Figure 2—figure supplement 5 ) . In contrast , hypo-methylated DMRs were found proximal to Copia , HAT , and Line L1 TEs ( Figure 3D ) . These transposon families are targeted by DNA demethylases and are associated with gene expression response to environmental stress ( Le et al . , 2014 ) . By contract , an increase in non-CG hyper-methylation is indicative of RdDM activity ( Law and Jacobsen , 2010 ) . To test whether RdDM was responsible for some of the HS-DMRs , we exposed drm1/drm2 double mutants to hyperosmotic stress over two consecutive generations . In contrast to wild-type , drm1/drm2 plants did not show hyper-methylation in non-CG contexts ( Figure 3—figure supplement 2 ) . However , when we analysed public datasets for known siRNA loci ( Fahlgren et al . , 2010 ) , we could not detect a correlation between methylation status and the presence of active siRNA production ( Figure 3E ) . Because hyperosmotic priming has been associated with discrete changes in the chromatin landscape ( Sani et al . , 2013 ) , we assessed the relationship between induced changes in DNA methylation and chromatin marks . Although hyperosmotic stress and growth conditions used by Sani et al . , 2013 differed from ours , we found that HS-DMRs were enriched for hyperosmosis-altered tri-methylated lysine 27 in histone 3 ( H3K27me3 ) compared to background MRs ( 63% vs . 13% ) ( Table 1 ) . Notably , hypermethylated HS-DMRs ( 38% ) were associated with decreased H3K27me3 , revealing an antagonistic relationship between these two repressive epigenetic marks in discrete genome regions involved in hyperosmotic priming . Collectively , our data show that hyperosmotic stress induces transient DNA methylation and chromatin changes at intergenic elements derived from specific transposon families . Because adaptive stress responses in plants have been proposed to be largely under maternal control ( Agrawal , 2001; Pecinka and Mittelsten Scheid , 2012 ) , we investigated the mode of inheritance of the enhanced tolerance to hyperosmotic stress by reciprocally crossing stressed and unstressed plants . We found that enhanced tolerance to hyperosmotic stress conditions was primarily conferred through the female germline ( Figure 4A ) . As the epigenetic reprogramming of male gametes is mediated to a large extent by DEMETER ( DME ) DNA glycosylase activity ( Calarco et al . , 2012; Ibarra et al . , 2012 ) , we investigated whether hyperosmotic priming could be passed on through the male germline if DME-dependent reprogramming was disrupted . We stressed dme-6 plants for two generations and tested the offspring for tolerance to hyperosmotic stress . The progeny of dme-6 stressed plants were even more tolerant to hyperosmotic stress ( Student’s t-test , p<0 . 001 ) than that of stressed control plants ( Figure 4A ) , thus implicating DME’s involvement in resetting stress-directed methylation marks in the male germline . To confirm this hypothesis , we performed reciprocal crosses between unstressed wild-type and stressed heterozygous dme-6 plants ( Figure 4A ) . The differential responses observed confirmed that DME diminishes the paternal transmission of hyperosmotic priming responses . To define the magnitude of the transmission of newly acquired epigenetic marks through the male germline , we compared the methylomes of sperm cell ( SC ) and vegetative nuclei ( VN ) isolated from control and hyperosmotic-stressed plants ( Figure 4—figure supplement 1 and Supplementary file 2 ) . Unlike in somatic tissue , hyperosmotic stress induced very few methylation changes in SC and VN ( Figure 4B ) such that only three DMRs became hypomethylated upon hyperosmotic treatment . This strikingly contrasted with the13 , 776 DMRs that distinguished SCs from VNs ( SV-DMRs ) ( Figure 4C ) . Most of the SV-DMRs predominantly localized next to TEs and were particularly enriched in adjacent regions of coding sequences ( Figure 4D ) . Significantly , over three quarters of the HS-DMRs ( 76% ) overlapped with SV-DMRs ( Figure 4C ) and half of them with MA-DMRs ( Figure 2D ) . Further , when compared with somatic tissues , CHG methylation in SC was elevated in pericentromeric regions , while methylation in VN was depleted in all sequence contexts and reduced to the central centromeric region ( Figure 4E ) . Because siRNAs produced in the vegetative nuclei can silence transposons in sperm cells in a DME-dependent process ( Ibarra et al . , 2012; Duan et al . , 2016 ) , we assessed the methylation state of HS-DMRs in dme mutants . Most HS-DMRs were hyper-methylated in dme-6 sperm cells but hypo-methylated in dme-6 vegetative nuclei ( Figure 4—figure supplemental 2 ) . However , we observed similar methylation differences at non-DMR coordinates ( Figure 4—figure supplemental 2 ) , indicating that HS-DMRs are only a subset of loci that are under the control of DME . In summary , DNA glycosylase activity in the male germline is pivotal for both the epigenetic silencing of transposons and for the resetting of epigenetic marks induced by environmental stress . As a consequence , the hyperosmotic priming effects are unequally transmitted through the male and female germlines . It has been proposed that abiotic stress can lead to heritable epigenetic changes particularly affecting transgenic repeats ( Lang-Mladek et al . , 2010; Molinier et al . , 2006 ) and TEs ( Bilichak et al . , 2012; Ito et al . , 2011 ) , and possibly the expression of neighbouring genes ( Dowen et al . , 2012; Wang et al . , 2013 ) . In agreement , HS-DMRs were frequently identified in transposon-related sequences and over-represented in regions proximal to protein-coding genes ( Figure 2A ) , suggesting that salt-induced methylation changes might be linked to differential gene expression . We found 123 genes that were flanked by HS-DMRs ( Figure 2D ) of which one third ( 32% ) have been previously shown to be responsive to osmotic stress ( Zeller et al . , 2009 ) ( Figure 5—figure supplement 1 and Supplementary file 5 ) . We identified one HS-DMR overlapping two TEs ( Figure 5A ) that was found upstream of MYB DOMAIN PROTEIN 20 ( MYB20 ) , which encodes a transcription factor involved in abscisic acid ( ABA ) signalling and implicated in stress tolerance ( Cui et al . , 2013 ) . This HS-DMR became hyper-methylated in P0 plants exposed to hyperosmotic stress , which was maintained in the P1 progeny of stressed plants , but then changed to levels similar to that seen in control plants in the P2 progeny . We did not detect MYB20 expression changes in response to high hyperosmotic treatment in plants whose progenitors had not experienced hyperosmotic stress ( Figure 5B ) . However , when P0 plants had experienced such stress , this gene was strongly downregulated in P1 and P2 progeny ( p-value 0 . 006 ) ( Figure 5B ) . Another HS-DMR was located downstream of the CARBON/NITROGEN INSENSITIVE 1 ( CNI1 ) gene ( Figure 5A ) , which encodes a membrane RING-type ubiquitin ligase implicated in metabolic sensing ( Sato et al . , 2009 ) . The CNI1 HS-DMR , which also overlapped a TE , had reduced DNA methylation in the P1 progeny of stressed plants , and remained hypomethylated in P2 progeny . Hyperosmotic stress strongly reduced CNI1 expression in progeny of untreated plants , and to a lesser extent in progeny of plants exposed to hyperosmotic stress ( Figure 5C ) . We also analysed four additional genes with adjacent HS-DMR and found similar modes of epigenetic regulation and inheritance ( Figure 5—figure supplement 2 ) . We then examined whether the stress responsiveness of these genes were altered in epigenetic mutants that were immune to hyperosmotic stress adaptation ( Figure 1D ) . Wild-type and mutant plants were grown in control or hyperosmotic stress conditions for two generations , and gene expression was analysed in P1 progeny exposed to stress . Independent of growth condition or priming , the expression of MYB20 , CNI1 and four other genes with associated HS-DMRs was altered in rdd demethylation mutants , and the transcriptional stress response in P1 progeny of stressed plants was impaired in RdDM mutants ( Figure 5B–C; Figure 5—figure supplement 2 ) . We then assessed the transcriptional response of genes encoding components of the DNA methylation and demethylation pathways and found that many of them were sensitive to hyperosmotic salt treatment ( Figure 5—figure supplement 3 ) . Moreover , analysis of epigenomic data for RdDM and demethylation mutants ( Stroud et al . , 2013 ) revealed that methylation at regions corresponding to HS-DMRs from our data set was severely affected in met1 and rdd mutants ( Fischer’s Exact test p=0 , 005 ) ( Figure 5—figure supplement 4 ) . We focused our attention on the CNI1 HS-DMR due to its unusual location downstream of the transcription unit . We hypothesised that this HS-DMR may act as a long-distance regulatory element . We therefore analysed the effects of two independent mutant alleles , in which T-DNAs were inserted between the CNI1 transcription unit and the downstream HS-DMR , possibly impeding communication between the HS-DMR and the CNI1 locus ( Figure 6A ) . As expected , CNI1 sense transcription in both insertion alleles was misregulated in response to stress ( Figure 6B ) . To demonstrate the functional relevance of the sequence targeted epigenetically after hyperosmotic-stress , we generated a deletion line ( Δcni1-DMR ) using CRISPR/Cas9 genome editing ( Jinek et al . , 2012 ) ( Figure 6A; Figure 6—figure supplement 1 ) . Deletion of the HS-DMR sequence reduced downregulation of sense CNI1 transcripts in response to hyperosmotic stress ( Figure 6C ) , indicating that this HS-DMR acts as a distant regulatory element . Finally , to demonstrate directly that the methylation status of this element determines its activity , we introduced an inverted repeat ( IR ) hairpin ( Matzke et al . , 2002 ) that directs DNA methylation to the CNI1 HS-DMR by RdDM ( Figure 6A ) . McrBC assays and bisulfite sequencing confirmed that methylation in IR hairpin transgenic lines remained even under hyperosmotic stress ( Figure 6—figure supplement 2 ) . Levels of CNI1 sense transcript in response to stress were no longer reduced in these plants ( Figure 6D ) , thus strengthening the argument that the CNI1 HS-DMR acts as an epigenetically sensitive regulatory element . Because stress can trigger the expression of lncRNAs ( Liu et al . , 2012; Matsui et al . , 2008 ) , we analysed published datasets ( Jin et al . , 2013 ) to investigate whether a lncRNA might mediate the effects of the HS-DMR on CNI1 expression . We found that hyperosmotic stress increased expression of a lncRNA that is transcribed in the antisense direction and overlaps with CNI1 , both in control plants ( CNI1-AS1 ) ( Figure 6—figure supplement 3 ) and in the progeny of plants that had gone through our control non-salt conditions ( p-value 0 . 004; Figure 6E ) . By contrast , hyperosmotic stress reduced expression of this lncRNA in progeny of plants exposed to hyperosmotic stress ( p-value 0 . 015; Figure 6E ) . That this was linked to differential methylation was indicated by the hyperosmotic responsiveness of CNI1-AS1 being impaired in progeny of rdd mutants ( Figure 6E ) . In the two insertion alleles and in the deletion line , the stress-mediated transcriptional response was similarly altered ( Figure 6F ) . These data support a model in which the HS-DMR downstream of CNI1 acts as an epigenetic sensor that controls CNI1 expression by modulating the expression of an antisense lncRNA . Our finding that lncRNA expression was insensitive to hyperosmotic stress in IR hairpin lines ( Figure 6G ) adds further weight to this model . To investigate whether there might be a broader role for lncRNAs in mediating the effects of stress induced DMRs , we comparatively analysed all stress-responsive genes adjacent to HS-DMRs and to MA-DMRs . Only the first group was enriched for hyperosmotic-responsive antisense lncRNAs ( Fisher’s Exact test , p=0 . 008 ) ( Supplementary file 6 ) , indicating that HS-DMRs preferentially act as regulatory elements of stress-induced antisense lncRNAs .
The extent and mechanism by which organisms acquire heritable adaptive traits after parental exposure to environmental stress is a central question in genetics and evolution . Unlike animals , plants present a fascinating model to examine this problem because their sessile nature makes it likely that parent and offspring will be exposed to similarly stressful conditions . Here , we have used a systematic approach to assess transgenerational stress adaptation in Arabidopsis ( Figure 1 ) . Our main conclusion is that intergenerational priming responses to hyperosmotic stress are triggered by recurrent exposure to stimuli , but that this response is rapidly lost in the absence of stress . Whether stress memory is less transient in perennial plants , which are rooted in place over consecutive years , needs to be determined . In animals that lack DNA methylation , the primary mechanisms involved in the acquisition and inheritance of new characters directed by the environment rely on histone modifications ( Öst et al . , 2014 ) and small RNAs ( Ashe et al . , 2012; Shirayama et al . , 2012 ) . In mammals such adaptation is usually associated with changes in DNA methylation ( Radford et al . , 2014 ) . In plants , environmentally-directed heritable traits have also been proposed to be associated with whole-scale changes in global DNA methylation ( Boyko et al . , 2010; Dowen et al . , 2012; Jiang et al . , 2014 ) . Bycontrast , using our highly controlled and replicated syste , and a strict method for detecting methylated regions , we found no clear evidence for such indiscriminate changes . Instead , we found that hyperosmotic stress directs DNA methylation changes primarily to discrete genome regions that are rich in TE-related sequences ( Figures 2 and 3 ) . The dynamic DNA methylation changes are of functional consequence because the adaptive response is impaired in mutants defective in DNA methylation pathways ( Figure 1 ) , which are also known to regulate TE activity ( Kim and Zilberman , 2014 ) . Mutants defective in DNA methylation and demethylation pathways displayed similar adaptive stress memory abilities , thus indicating that adaptive memory is controlled by complex processes that may not strictly rely on DNA methylation changes alone ( Crisp et al . , 2016 ) . A small fraction of acquired epigenetic changes is transmitted to the immediate offspring only after recurrent stress exposure , but in the absence of a renewed stimulus these are reset in subsequent generations . This implies that epigenetic stress memory in annual plant species may be transient in nature , however this response may differ in perennial plants that grow over longer periods of time before producing any offspring or in plants that reproduce asexually . Half of the newly acquired methylation changes identified in our stressed plants overlapped with regions privileged for epimutation also found in near-isogenic greenhouse-grown populations ( Figure 2 ) ( Hagmann et al . , 2015 ) . This finding confirms that distinct regions of the plant epigenome are particularly labile , and it suggests that such acquired epimutations could modulate the ability of plants to respond to stress . Importantly , in our system , enhanced stress responses were only passed on after two consecutive generations of stress exposure , but not when plants were exposed to hyperosmotic stress for only one episode , which may be the reason why previous studies often failed to find clear evidence for inheritance of stress-induced epigenetic changes ( Eichten and Springer , 2015; Secco et al . , 2015 ) . One explanation for repeated stress exposure being required for induction of transiently heritable stress resistance may reside in the importance of poised epigenetic states for environmental priming ( Jaskiewicz et al . , 2011; Sani et al . , 2013 ) , which in turn may facilitate the establishment of new epigenetic marks at discrete genomic regions that are sensitive to stress . This view is supported by the substantial overlap ( >30% ) between HS-DMRs established in recurrently stressed generations ( Figure 2 ) and dynamic chromatin occupancy directed by hyperosmosis ( Table 1 ) . Intriguingly , recurrent exposure to hyperosmotic stress did not significantly increase the number of newly acquired epimutations , suggesting that the extent of epigenetic plasticity elicited by the environment is limited to a few key genomic regions that may be under purifying selection ( Hollister and Gaut , 2009 ) . It is likely that certain genomic regions are epigenetically targeted , depending on the type of stress because the DNA methylation changes induced by hyperosmotic stress did not overlap with methylation changes reported for other abiotic stresses ( Seymour et al . , 2014 ) . A direct comparison of the latter study to this study is however difficult to make due to inconsistencies in experimental methodology , as Seymour et al . , 2014 did not investigate methylation changes in subsequent generations grown in the absence of stress . While it is accepted that some epigenetic variation caused by the environment can be transmitted to the immediate offspring of plants and mammals ( Feil and Fraga , 2011; Heard and Martienssen , 2014 ) , the importance of transmission through either the male or female germline has not been previously investigated . Our analyses revealed that hyperosmotic priming responses were transmitted primarily through the maternal germline ( Figure 4 ) . We attribute this difference to differences in the meiotic transmission of newly acquired DNA methylation marks . In support of this hypothesis , most of the stress-associated methylation changes in leaves were largely absent in mature male gametes ( Figure 4 ) , indicating that environmentally directed epigenetic marks are more efficiently reset in male rather than in female gametes . Indeed , not only do male gametes undergo an active reprogramming of DNA methylation at transposon sequences by the DNA glycosylase DME ( Borges et al . , 2012a; Ibarra et al . , 2012 ) , but also at stress-dependent DNA methylation sites ( Figure 4 ) . Moreover , the enhanced tolerance to hyperosmotic stress in progeny of stressed dme-6 plants implicates DNA demethylation both in the active reprogramming of TEs and in the resetting of environmentally-directed epigenetic changes in male gametes . The male transmission of stress-associated adaptive responses is under the strict control of DME ( Figure 4 ) . Although the precise mechanism of DME in resetting HS-DMRs is unknown , it may be linked to the movement of RNA to sperm cells from the surrounding vegetative cells ( Duan et al . , 2016 ) . Why male and female gametes should differ in their ability to reset newly acquired epigenetic marks remains an enigma . One explanation is that female gametes reside on the mother plant where they are eventually fertilised by sperm from pollen that may have travelled a great distance from its male progenitor to produce seed . Given that seed dispersal usually occurs at a short distance from the mother , a solely maternal transmission of the newly acquired epigenetic marks would be an efficient way of retaining only the most relevant parental stress-associated adaptive responses in the progeny . Our study has revealed that environmentally-induced priming responses involve DNA ( de ) methylation pathways that control the extent to which acquired epigenetic states are inherited and maintained in successive generations ( Figure 1 ) . Previous studies have focused on these pathways because they mediate global stress responses ( Zhu , 2009 ) . We now provide compelling evidence that stress can specifically alter methylation at the adjacent sequences of several key stress-response regulators ( Figures 2 , 5 and 6 ) , a process that is mediated by the RdDM and DNA demethylation pathways ( Figure 7 ) . Targeted studies have revealed that some TEs proximal to upstream regions and sensitive to methylation changes directed by hyperosmotic stress have regulatory roles ( Baek et al . , 2011; Xu et al . , 2015 ) and are associated with quantitative traits and adaptive behaviour ( Baxter et al . , 2012; Busoms et al . , 2015 ) . Notably , both DNA methylation and demethylation activities are also tightly regulated by the action of adjacent upstream TEs that act as an epigenetic sensors ( Lei et al . , 2015; Williams et al . , 2015 ) . The epigenetic regulation of HS-DMRs is not always associated with upstream regulatory sequences and sense transcription; instead we found that they are preferentially associated with the transcription of antisense lncRNAs ( Figure 6 ) . Although the functions of these HS-DMR-associated non-coding transcripts are largely unknown , antisense lncRNAs have been implicated in directing chromatin changes ( Ariel et al . , 2014; Heo and Sung , 2011; Swiezewski et al . , 2009 ) , and thereby in influencing transcription , splicing and transcript stability ( Bardou et al . , 2014; Borsani et al . , 2005 ) . In this study , antisense lncRNAs were shown to be regulated by epigenetically labile control elements sensitive to stress ( Figure 7 ) . Methylation changes in these genome regions could modulate transcription factor binding ( Zhong et al . , 2013 ) or chromatin regulatory loops ( Ariel et al . , 2014 ) . Adaptive epigenetic inheritance has been a topic of fascination , but also of scientific controversy ( Lysenko , 1951 ) . The adaptive value of this inheritance over multiple generations must depend on the cost of epigenetic resetting , as well as on the degree and predictability of environmental stress ( Herman et al . , 2014 ) . Because conditions in many natural environments are highly stochastic , an adaptive bet-hedging strategy ( Simons , 2011 ) that is mediated by increased epigenetic variation could be advantageous . Under our controlled stress conditions , the contribution of adaptive epigenetic variation shows parental effects ( Figure 4 ) , which may be favoured over bet-hedging in relatively stable environments ( Kuijper and Johnstone , 2016 ) . Hence , our work provides insights into the importance of epigenetically driven adaptive changes and illustrates the evolutionary significance of epigenetic plasticity in plants .
The wild-type background studied was A . thaliana Col-0 . The multigenerational hyperosmotic stress experiments used the reporter line L5 , which harbours a silenced reporter encoding β-glucuronidase linked to the cauliflower mosaic virus 35S promoter ( 35Spro::GUS ) ( Morel et al . , 2000 ) . To isolate male gametes , we used reporter lines MGH3p::MGH3-eGFP and ACT11p::H2B-mRFP , in which either sperm or vegetative cells are marked by fluorescent protein expression ( Borges et al . , 2012b ) ( Figure 4—figure supplement 1 ) . Mutants cmt3-11 ( Chan et al . , 2006drm1-2 drm2-2 ( Chan et al . , 2006 ) , dme-6 ( Shirzadi et al . , 2011 ) , nrpda1-4 ( Herr et al . , 2005 ) , rdr2-1 ( Xie et al . , 2004 ) , ros1-4 ( Zheng et al . , 2010 ) , and the ros1-3 dml2-1 dml3-1 ( Penterman et al . , 2007 ) triple mutant have been described . Plants were grown at 22°C under long days ( 16 hr light , 8 hr dark; light intensity 120 µmol/sec/m2 ) . Lines carrying T-DNA insertions downstream of CNI1 were obtained from the SALK collection ( cni1-2 , Salk_100221 and cni1-3 , Salk_030235 ) . Seeds from a single founder plant were germinated and grown on MS medium ( control ) for two weeks and transferred to MS medium supplemented with 25 or 75 mM NaCl to induce mild hyperosmotic stress for 4 weeks . Before flower buds were visible , plants were transferred to soil ( generation 1 ) . We sampled 10 individual plants from each treatment , whereby ten-week-old leaf samples and mature seeds were collected separately from each plant . This process was repeated for five successive generations . In each generation , offspring of the salt treated and control plants were grown in non-stress condition for two successive generations to produce P1 and P2 plants ( Figure 1A ) . We modified plasmids previously described ( Fauser et al . , 2014 ) . A pair of guide RNAs was selected using the CRISPR-PLANT tool ( Xie et al . , 2014 ) , the corresponding DNA oligonucleotides ( Integrated DNA Technologies ) were cloned into pEN-Chimera using BbsI and BsmBI to generate plasmids pEN-CNI1 . 1 . Constructs were transferred into pDE-CAS9 plasmid by Gateway cloning ( Invitrogen ) and transformed by floral dipping ( Clough and Bent , 1998 ) . Deletions where identified by PCR ( Supplementary file 7 ) and confirmed by sequencing . A DNA fragment was chemically synthesized ( IDT ) and introduced into the hairpin vector pJawohl-Act2 using Gateway cloning ( Life Technologies ) . Constructs were transformed by floral dipping . Transgenic T2 lines ( T2 ) with DNA hypermethylation at the CNI HS-DMR after exposure to hyperosmotic stress ( 175 mM NaCl ) were identified by CHOP-PCR ( Zhang et al . , 2014 ) after digestion with HpyCH4IV ( NEB ) and PCR amplification ( Supplementary file 7 ) . All phenotypic tests were carried out with six independent replicates . For germination assays , 50 seeds were sown per plate on MS with or without 200 mM NaCl , a concentration of salt we found to be highly selective in the germination of Col-0 seeds . Seeds were scored as having germinated based on radicle emergence 14 days after sowing . For survival assays , 50 seeds were sown on MS or on MS supplemented with 150 mM NaCl , a concentration of salt known to allow germination but affecting vegetative growth in Col-0 . Survival was scored based on presence/absence of green leaves 14 days after sowing . Data are summarized in Supplementary file 1 . Plants were grown on MS medium with or without 100 mM NaCl for 5 weeks . Leaves were collected , weighed , and washed in distilled water . Chlorophyll was extracted by incubating 0 . 02 – 0 . 03 g of ground leaf material in 80% ( v/v ) aqueous acetone at 4°C for 48 hr . Total chlorophyll content ( chlorophyll a and b ) was measured using a spectrophotometer at 663 . 6 nm and 646 . 6 nm absorbance ( Porra , 2002 ) . Plants were grown on MS medium with or without 100 mM NaCl for 5 weeks . Leaves were collected , and washed in distilled water . Plant material was dried at 80°C for 48 h , then weighed . Ions were acid-extracted from dried plant material using 2 ml of concentrated nitric acid and microwave digestion . The digestion program consisted of: 5 min at 100°C , 2 min at 120°C , 5 min on 160°C , 22 min at 180°C , and cooling down to 70°C . After samples had cooled down , the digested samples were diluted with 23 ml distilled water . The sodium ion concentration of the diluted samples was measured using Inductively Coupled Plasma Mass Spectrometry ( ICP-MS ) . MGH3::MGH3-eGFP/ACT11::H2B plants were germinated and grown for 6 weeks on MS medium without NaCl or with 25 mM or 75 mM NaCl before being transferred to soil to induce flowering . Approximately 10 g of flower tissue were collected into 50 ml Falcon tubes , 10 ml of sperm nuclei buffer was added and the pollen suspension was vortexed for 3 min . The pollen suspension was filtered through a Miracloth mesh and centrifuged for 1 min at 3000 rpm; the supernatant was carefully removed from the pollen pellet . For the extraction of nuclei , the pellet was re-suspended in 1 ml sperm nuclei buffer , loaded into 1 . 5 ml tubes containing 100 μl of acid-washed glass beads ( 425–600 μm ) and mixed for 4 min . The crude extract was filtered through a 28 μm micro-filter sieve , leaving the nuclei intact . Vegetative and sperm nuclei were isolated from the crude extract of disrupted pollen using Fluorescence-Activated Cell Sorting ( FACS ) with a MoFlo high-speed cell sorter ( Beckman Coulter , Fort Collins , USA ) ( Borges et al . , 2012b ) ( Figure 4—figure supplement 1 ) . One laser was set to 140 mW at 488 nm for forward scatter ( FSC ) and side scatter ( SSC ) measurements , and for GFP excitation . A second laser was set to 38 mW at 561 nm for RFP excitation . GFP and RFP were detected using 530/40 nm and 630/75 nm bandpass filters . P1 and P2 progeny of the G3 generation were grown on MS medium supplemented with 125 mM NaCl for 2 weeks . Leaves were collected from 50 seedlings and total RNA was extracted using the RNeasy Plant Mini Kit ( Qiagen ) according to the manufacturer’s instructions . RNA was treated with TURBO DNA-free ( Promega , Madison , WI ) . cDNA was synthesized from 1 μg of extracted RNA using the RevertAid First Strand cDNA Synthesis Kit ( Thermo Scientific ) . Quantitative real time PCR analyses were performed on a MyiQ System ( BIO-RAD ) , using oligonucleotide primers designed with Primer3 ( Rozen and Skaletsky , 2000 ) ( Supplementary file 7 ) . PCR fragments were analysed using a dissociation protocol to ensure that each amplicon was a single product . Amplicons were also sequenced to verify the specificity of PCR . The amplification efficiency was calculated from raw data using LingRegPCR ( Ramakers et al . , 2003 ) . All RT-qPCR experiments were performed using five biological replicates , with a final volume of 25 µl containing 5 µl of cDNA template ( diluted beforehand 1:10 ) , 0 . 2 µM of each primer , and 12 . 5 µl of 2×MESA Blue qPCR MasterMix ( Eurogentec Headquarters ) . The following cycling profile was used: 95°C for 10 min , followed by 40 cycles of 95°C for 10 s , 60°C for 15 s , and 72°C for 15 s . The melting curve was determined in the range of 60–95°C , with a temperature increment of 0 . 01°C/sec . Each reaction was run in triplicate ( technical replicates ) . Negative controls included in each run were a reaction without reverse transcriptase and one without template ( 2 μL of nuclease-free water instead of 2 μL of cDNA ) . No signals were observed in the negative controls . Raw Ct data were analysed using GeneEx Pro ( Kubista et al . , 2006 ) . Analysis of expression data was performed according to the ΔΔCT method ( Livak and Schmittgen , 2001 ) using GADPH ( At1g13440 ) , PDF2 ( At1g13320 ) and UBQ5 ( At3g62250 ) for normalization ( Lippold et al . , 2009 ) . To measure CNI1 antisense lncRNA transcripts , 1 μg of total RNA was isolated from seedlings . Reverse transcription used At5g27420_ant , PP2AA3 Reverse and GAPDH Reverse oligonucleotides ( Supplementary file 7 ) in the same reaction with SuperScriptIII Reverse Transcriptase ( Invitrogen ) . qPCR reactions used At5g27420 Forward and At5g27420 Reverse primers ( Supplementary file 7 ) following the same conditions described for the sense reactions . These experiments were performed using six technical replicates for each reaction . Expression data were analysed according to the ΔΔCT method ( Livak and Schmittgen , 2001 ) using GADPH ( At1g13440 ) and PP2AA3 ( At1g13320 ) for normalization . PCR reactions were performed in duplicate and RT-minus controls were included to confirm absence of genomic DNA contamination . For somatic tissue , rosette leaves were pooled from 10 plants for each treatment group . For male gamete analysis , sperm and vegetative nuclei were collected from 100 plants for each treatment group . gDNA was extracted from leaf samples with the DNAeasy Plant Mini Kit ( Qiagen ) , and from sperm and vegetative nuclei with MasterPureDNA Purification Kit ( Epicentre ) . DNA libraries were generated using the Illumina TruSeq Nano kit ( Illumina , CA , USA ) . DNA was sheared to 350 bp . The bisulfite treatment step using the Epitect Plus DNA Bisulfite Conversion Kit ( Qiagen , Hilden , Germany ) was inserted after the adaptor ligation; incubation in the thermal cycler was repeated once before clean-up . After clean-up of the bisulfite conversion reaction , library enrichment was done using Kapa Hifi Uracil+ DNA polymerase ( Kapa Biosystems , MA , USA ) . Libraries were sequenced with 2 x 101 bp paired-end reads on an Illumina HiSeq 2000 instrument , with conventional gDNA libraries in control lanes for base calling calibration . Seven to eight libraries with different indexing adapters were pooled in one lane . For image analysis we used Illumina RTA 1 . 13 . 48 . The procedure followed ( Becker et al . , 2011 ) . In brief , the SHORE pipeline v0 . 9 . 0 ( Ossowski et al . , 2008 ) was used to trim and quality-filter the reads . Reads with more than 5 ( or 2 ) bases in the first 25 ( or 12 ) positions with a base quality score of below 5 were discarded . Reads were trimmed to the right-most occurrence of two adjacent bases with quality values of at least 5 . Trimmed reads shorter than 40 bases were discarded . Reads were then aligned against the Col-0 reference genome sequence using GenomeMapper implemented in SHORE ( Schneeberger et al . , 2009 ) . We used published methods ( Becker et al . , 2011 ) . The number of covered and methylated sites for each sample as well as the false methylation frequencies retrieved from read mappings against the chloroplast sequence can be found in Supplementary file 1 . On average , 40 . 7 million cytosines were covered by at least 3 reads and with a quality score above 25 in more than half of the samples . Of these , 7 . 2 million cytosines were methylated in at least one sample ( Supplementary file 8 ) . For DMP calling , we modified the approach from Becker et al . ( 2011 ) , without removing sites classified as differentially methylated between replicates . We applied Fisher’s Exact test for all pairwise sample comparisons on cytosine sites with a methylation frequency difference to another sample of at least 30% . We used the same P value correction scheme as in Becker et al . ( 2011 ) . We first identified MRs in each sample separately using a Hidden Markov Model ( HMM ) ( Hagmann et al . , 2015 ) . MRs of replicates were merged into a common set of MRs . Whenever different samples were treated as a replicate group ( e . g . control and salt-treated samples ) , their MRs were merged into a common set . Regions that showed statistically significant methylation differences between at least two sets of strains were identified as DMRs ( Hagmann et al . , 2015 ) . In brief , segmentations across the genomes of every sample served to set breakpoints of start and end coordinates of all predicted MRs . Each combination of coordinates in this set defined a segment to perform the test for differential methylation in all pairwise comparisons of the strains , if at least one strain was in a high methylation state throughout this whole segment ( Hagmann et al . , 2015 ) . Per pairwise comparison , between 30 , 000 and 50 , 000 segments were tested ( Hagmann et al . , 2015 ) . For tests within generations , we grouped P0 control , P1 control and P2 control samples as 'non-stressed'; P0 salt-treated samples as 'stressed'; P1 samples derived from salt-treated P0 plants as 'stressed-P1'; and P2 samples derived from salt-treated P0 plants as 'stressed-P2' . Tests for DMRs were then carried out between these four groups . In addition we separately tested without the respective remaining groups for 'non-stressed' vs . 'stressed' , 'stressed' vs . 'stressed P1' , 'stressed' vs . 'stressed P2' , and 'stressed P1' vs 'stressed P2' . This latter step was done to assess the number of DMRs directly identified between two groups , without multiple testing correction for comparisons with and between other groups . DMRs from the MA lines were taken from a previous publication ( Hagmann et al . , 2015 ) . We used the TAIR10 annotation for genes , exons , introns and untranslated regions; transposon annotation was according to Slotte et al . ( 2013 ) . Positions and regions were hierarchically assigned to annotated elements in the order CDS > intron > 5’ UTR > 3’ UTR > transposon > intergenic space . We defined as intergenic positions and regions those that were not annotated either as CDS , intron , UTR or transposon . Each position was assigned to the corresponding element that contained it . DMRs were assigned to annotated elements by basepair , i . e . each position in the DMR was assigned in the above-mentioned order . A DMR can stretch over several annotated elements . We tested the overlap of DMRs with other DMRs or with genes using bedtools ( Quinlan and Hall , 2010 ) , either requesting a direct overlap or an overlap within a window of n bp downstream and upstream of the regions . For overlap between DMRs and TEs , we required either a direct overlap or an overlap within 2 , 000 bp windows downstream and upstream of the DMRs . Overlapping TEs were then sorted into their superfamilies according to TAIR nomenclature . The TE profiles for hypo- and hypermethylated DMRs were compared against the expected values taken from the whole genome TE profile . For each TE superfamily the expected values were calculated as: Se= ( wgs/wgt ) *st , where Se is the expected value for that superfamily in that sample ( hypo or hyper ) , wgs is the number of transposons of that superfamily in the whole genome , wgt is the total number of transposons in the whole genome , and st is the total number of transposons in this sample . For the identification of genes regulated by HS-DMRs we first identified the genes within 2 kb upstream or downstream of DMRs and analysed their expression in shoots or roots exposed to hyperosmotic stress using AtGenExpress ( Kilian et al . , 2007 ) . We used Protein ANalysis THrough Evolutionary Relationships ( PANTHER 9 . 0 ) software ( Mi et al . , 2013 ) to classify significantly enriched Gene Ontology ( GO ) terms associated with/without overlap with MA line DMRs and with hypo- and hypermethylated DMRs . Heatmaps for GO analysis were generated using R version 3 . 0 . 1 ( www . r-project . org ) . Complete linkage clustering was done in R version 3 . 0 . 1 ( www . r-project . org ) using the ‘heatmap . 2’ function of the ‘gplots’ package in combination with the ‘hclust’ function of the ‘fastcluster' package using the complete linkage clustering method . Uncertainty in hierarchical clustering analyses was estimated using the pvclust package in R . Graphs were generated using R version 3 . 0 . 1 ( www . r-project . org ) . Circular display of genomic information in chromosomes was rendered using Circos version 0 . 63 ( Krzywinski et al . , 2009 ) . The DNA and RNA sequencing data have been deposited at the European Nucleotide Archive under accession numbers PRJEB9076 and PRJEB13558 . DNA methylation data and MR coordinates have been uploaded to the epigenome browser of the EPIC Consortium ( https://www . plant-epigenome . org/; https://genomevolution . org/wiki/index . php/EPIC-CoGe ) and can be accessed at http://genomevolution . org/r/939v . | Most plants spend their entire lives in one fixed spot and so must be able to quickly adapt to any changes in their surroundings . For example , high levels of salt in the soil – which can be toxic to cells – triggers stress responses in plants that help them to mitigate any damage . Once the stress has passed , plants are able to retain a memory of it , which allows them to respond more quickly if they face the same stress in future . Furthermore , plants may pass on this ‘stress memory’ to their offspring . It is thought that stress memory is programmed by chemical modifications to DNA known as epigenetic marks . These marks do not alter the genetic information that is encoded by the DNA itself , but they can change the activity of particular genes . Environmental stress leads to changes in the epigenetic marks found on many plant genes , which can be directly passed on from the parent plant to its offspring . However , it was not clear whether the epigenetic marks that programme stress memory can be passed on in this way . Wibowo , Becker et al . investigated how a model plant called Arabidopsis thaliana is able to remember periods of salt stress . The experiments show that high levels of salt can trigger changes in the patterns of epigenetic marks associated with particular regions of DNA . This memory is reinforced by repetitive exposure to similar salt stress and can be passed onto offspring , primarily through the maternal line . However , this stress memory is not fixed in future generations as the epigenetic marks can be reset to their original patterns if plants find themselves growing and reproducing under non-stress conditions . In sum , the findings of Wibowo , Becker et al . show that epigenetic marks allow plants to inherit stress memory on a temporary basis while the stress is present , but to gradually lose the memory if the stress does not return . Future studies will focus on finding out if stress memory in crop plants works in the same way . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"computational",
"and",
"systems",
"biology",
"genetics",
"and",
"genomics"
] | 2016 | Hyperosmotic stress memory in Arabidopsis is mediated by distinct epigenetically labile sites in the genome and is restricted in the male germline by DNA glycosylase activity |
Non-rapid eye movement ( NREM ) sleep , characterized by slow-wave electrophysiological activity , underlies several critical functions , including learning and memory . However , NREM sleep is heterogeneous , varying in duration , depth , and spatially across the cortex . While these NREM sleep features are thought to be largely independently regulated , there is also evidence that they are mechanistically coupled . To investigate how cortical NREM sleep features are controlled , we examined the astrocytic network , comprising a cortex-wide syncytium that influences population-level neuronal activity . We quantified endogenous astrocyte activity in mice over natural sleep and wake , then manipulated specific astrocytic G-protein-coupled receptor ( GPCR ) signaling pathways in vivo . We find that astrocytic Gi- and Gq-coupled GPCR signaling separately control NREM sleep depth and duration , respectively , and that astrocytic signaling causes differential changes in local and remote cortex . These data support a model in which the cortical astrocyte network serves as a hub for regulating distinct NREM sleep features .
Sleep is characterized by distinct electrophysiological features that reflect the rhythmic activity of large populations of neurons . One phase of sleep—non-rapid eye movement ( NREM ) sleep—is critical for several important functions including memory consolidation/destabilization and synaptic homeostasis ( Klinzing et al . , 2019; Genzel et al . , 2014; Kim et al . , 2019; Tononi and Cirelli , 2014; Diekelmann and Born , 2010; Tononi and Cirelli , 2006; Tononi and Cirelli , 2020; Ji and Wilson , 2007 ) . These functions are thought to require slow-wave activity ( SWA ) , the distinct oscillatory pattern of neural activity in the cortex that occurs during NREM sleep and differentiates it from the relatively desynchronized activity during wakefulness and REM sleep . However , neural activity during NREM sleep is not uniform over the course of sleep , but varies in duration and depth ( as measured by SWA intensity ) . Past work has demonstrated that NREM sleep duration and depth can be independently controlled ( Dijk and Beersma , 1989; Patrick and Gilbert , 1896 ) . Indeed , the circuit mechanisms known to underlie sleep depth and duration are largely independent from each other and operate on very different time-scales: sleep duration is mediated by subcortical nuclei that receive direct input from circadian centers and drive sleep/wake transitions through release of neuromodulatory signals ( Holst and Landolt , 2018; Saper and Fuller , 2017; Lee and Dan , 2012 ) . On the other hand , SWA intensity is largely regulated by cortical and thalamocortical circuits ( Chen et al . , 2012; Steriade et al . , 1993; Lemieux et al . , 2015; Volgushev et al . , 2006; Steriade and Timofeev , 2003; Sheroziya and Timofeev , 2014; Amzica and Steriade , 1995; Sanchez-Vives and McCormick , 2000 ) . While these two physiological measures of sleep have been mostly described in non-overlapping mechanistic terms , there is also physiological evidence that sleep depth and duration can be coupled . For example , cortical calcium ( Ca2+ ) signaling can act on a millisecond time-scale to modulate cortical synchrony during SWA while also engaging longer term signaling cascades that regulate the sleep/wake cycle ( Ode et al . , 2017; Tatsuki et al . , 2016 ) . Thus , the extent to which the neural mechanisms underlying sleep depth and duration are linked remains unclear . The cortex—where mammalian sleep is most often measured—is a brain region where neural mechanisms underlying sleep duration and sleep depth coincide: many neuromodulatory nuclei associated with sleep/wake transitions send direct projections to the cortex ( Björklund and Lindvall , 1978; Woolf , 1991; Loughlin et al . , 1986; Panula et al . , 1989 ) , and the cortex plays an instrumental role in generating and propagating SWA during sleep ( Volgushev et al . , 2006; Sanchez-Vives and McCormick , 2000; Niethard et al . , 2018; Stroh et al . , 2013; Luczak et al . , 2007; Massimini et al . , 2004; Krone , 2020; Sanchez-Vives and Mattia , 2014; Lemieux et al . , 2014 ) . Further , cortical SWA intensity can be locally regulated , leading to heterogeneity of SWA across cortex ( Huber et al . , 2004; Funk et al . , 2016; Siclari and Tononi , 2017 ) . However , how the cortex integrates separate regulatory signals to orchestrate activity across sleep and wake is unknown . In untangling sleep mechanisms , both in cortex and throughout the brain , the historical focus has almost exclusively been on neurons and neuronal circuits . Yet astrocytes—the largest class of non-neuronal brain cells—are also situated to play critical roles in sleep regulation within the cortex . Astrocytes tile the cortex , can participate in bidirectional communication with thousands of neurons ( Halassa et al . , 2007; Allen and Barres , 2005; Bushong et al . , 2002; Bazargani and Attwell , 2016 ) , exhibit morphological and transcriptional changes during sleep ( Bellesi et al . , 2015 ) , and regulate SWA under anesthesia ( Szabó et al . , 2017; Poskanzer and Yuste , 2016; Durkee et al . , 2019 ) . Further , multiple canonical astrocytic functions are also associated with sleep/wake regulation , including regulation of extracellular glutamate ( Poskanzer and Yuste , 2016; Poskanzer and Yuste , 2011 ) , extracellular ion dynamics ( Ding et al . , 2016 ) , release of neurotransmitters ( Halassa et al . , 2009; Papouin et al . , 2017; Fellin et al . , 2009 ) , and metabolic regulation ( Petit and Magistretti , 2016; Bellesi et al . , 2018; DiNuzzo and Nedergaard , 2017 ) . Astrocyte physiology is primarily measured via intracellular Ca2+dynamics , which vary widely in size , shape , and location , and can propagate within or even between cells ( Wang et al . , 2019; Khakh and McCarthy , 2015; Shigetomi et al . , 2013; Shigetomi et al . , 2016; Guerra-Gomes et al . , 2017 ) . Because imaging complex astrocyte Ca2+ activity in vivo is relatively new , it remains unknown whether these diverse astrocytic Ca2+ dynamics map onto different circuit functions . However , the potential of astrocytes to influence large populations of cortical neurons across different time-scales is significant ( Stobart et al . , 2018; Lind et al . , 2013 ) . The majority of astrocyte Ca2+ activity is thought to result from upstream activation of G-protein coupled receptors ( GPCRs ) ( Durkee et al . , 2019; Di Castro et al . , 2011; Agulhon et al . , 2008; Kofuji and Araque , 2021 ) . Importantly , many astrocytic GPCRs are activated by neuromodulators , including those associated with sleep/wake regulation , such as norepinephrine , acetylcholine , and histamine . Since GPCRs regulate a diverse array of Ca2+-dependent intracellular signals on many different time-scales ( Grundmann and Kostenis , 2017; Kholodenko et al . , 2010 ) , they are prime candidates for differentially regulating individual features of NREM sleep , such as duration and depth . A downstream target of GPCRs , the inositol triphosphate type two receptor ( IP3R2 ) , has been recently shown to be involved in sleep regulation ( Bojarskaite et al . , 2020 ) . In astrocytes , both Gi- and Gq-coupled GPCRs activate IP3R2s and lead to increases in intracellular Ca2+ ( Durkee et al . , 2019; Mariotti et al . , 2016; Nagai et al . , 2019 ) , while also engaging separate signaling cascades . Despite this , scant attention has been paid to whether the activation of different astrocytic GPCRs , and resulting Ca2+ signals , have differential effects on the surrounding neural circuit . Indeed , GPCR signaling in astrocytes may underlie mechanisms by which astrocytes perform multiple , parallel functions in the neural circuit . Here , we leveraged a recently developed image analysis tool that captures the spatiotemporal complexity of astrocyte Ca2+ dynamics ( Wang et al . , 2019 ) and astrocyte-specific chemogenetics to investigate the mechanisms by which cortical astrocytes both link and independently regulate different features of NREM sleep via GPCR signaling . To do this , we carried out in vivo two-photon ( 2P ) imaging of astrocyte Ca2+ while recording electrophysiological sleep rhythms to examine astrocyte Ca2+ changes across natural sleep and wake . We find that endogenous Ca2+ activity is inversely correlated with SWA and exhibits bidirectional changes prior to sleep-wake transitions . Using chemogenetics to selectively manipulate astrocytic Gi- and Gq-GPCR pathways , we demonstrate that astrocytes actively regulate both NREM sleep duration and depth , via separate GPCR signaling pathways: astrocytic Gi-induced Ca2+ is sufficient to increase SWA ( sleep depth ) , while sleep-wake transitions ( sleep duration ) is dependent on Gq-GPCRs . We demonstrate a role for astrocytes in both local and cortex-wide sleep regulation; manipulating astrocytic Ca2+ in primary visual cortex ( V1 ) alters not only local SWA , but also affects SWA in contralateral frontal cortex ( FC ) . Further , we find that while local changes in SWA arise from greater changes in delta waves , remote SWA effects in FC are due to increases in slow oscillations . Since these two slow waves underlie different functions , our data support the concept that astrocytes exert different effects on neuronal populations depending on both the type of GPCR activated and their localization within cortical circuits . Together , our data support a role for the cortical astrocytic network as a hub for the regulation of sleep depth and duration across cortex .
To study the role of astrocytes in sleep regulation , we conducted 2P imaging of astrocyte Ca2+ dynamics as animals naturally transitioned between sleep and wake states ( Niethard et al . , 2018; Seibt et al . , 2017 ) . To specifically express the Ca2+ indicator GCaMP6f in cortical astrocytes , we injected mice with AAV-GFAP-GCaMP6f 2–4 weeks before experiments ( Figure 1B , left ) . Electrodes were implanted for local field potential ( LFP ) and electromyogram ( EMG ) recordings ( Figure 1B , right ) to assess sleep state . During recording sessions , mice were head-fixed on a horizontal treadmill , and locomotion was recorded ( Figure 1A ) . To control for the effect of circadian rhythm and sleep pressure , all recording sessions took place between ZT 2–5 . Experiments were conducted after mice had been previously habituated to head-fixation to allow natural sleep . To analyze astrocyte Ca2+ activity , we used our recent tool , AQuA ( Wang et al . , 2019 ) , an event-based approach to detect spatiotemporally distinct Ca2+ events without predetermined regions-of-interest ( ROIs ) . This allowed automatic detection of individual astrocyte Ca2+ events , independent of size and shape , across sleep and wake ( Figure 1C , Video 1 ) . To investigate the relationship between in vivo cortical astrocyte activity and NREM sleep , we first quantified the relationship between Ca2+ event rate and SWA ( 0 . 5–4 Hz power ) , a marker of NREM sleep depth . By dividing entire 2–3 hr recordings into two-min bins , we found that Ca2+ event rate and SWA were negatively correlated ( Figure 1D ) , that is when SWA is low , astrocyte Ca2+ event rate is high , and vice versa . This finding suggests astrocytes may play roles regulating SWA , an idea supported by previous studies demonstrating that astrocyte Ca2+ plays a causal role in driving low frequency-dominated cortical states under anesthesia ( Szabó et al . , 2017; Poskanzer and Yuste , 2016; Fellin et al . , 2009 ) . To determine whether the negative correlation between astrocyte Ca2+ and SWA is specific to a particular behavioral state , we analyzed our data by dividing recording periods into sleep , locomotory wake , and stationary wake ( Figure 1E ) . We separated wake by locomotion to quantify Ca2+ dynamics independently from large Ca2+ bursts that occur with locomotion onset ( Wang et al . , 2019; Paukert et al . , 2014; Nimmerjahn et al . , 2009 ) . As predicted by the negative correlation between event rate and SWA ( Figure 1D ) , we found that Ca2+ event rate was highest during locomotory wake , lower during stationary wake , and lowest during sleep ( Figure 1F ) . To confirm , we compared SWA in the three behavioral states and found an inverse relationship of Ca2+ event rate , namely SWA was highest during sleep , lower during stationary wake , and lowest during locomotory wake ( Figure 1G ) . These findings are supported by recent work demonstrating the same pattern of Ca2+ activity across similar behavioral states , using an ROI-based image analysis approach ( Bojarskaite et al . , 2020 ) , confirming that our event-based image analysis can generate comparable results when the same metrics are quantified . Together , these data demonstrate that changes in Ca2+ event frequency co-occur with major changes in behavioral state , consistent with levels of SWA . To explore whether each behavioral state can be characterized by the types of astrocytic Ca2+ events that occur during these states , we first compared the events' size , duration , and amplitude . As predicted by the large , synchronous bursts observed during locomotion , we found locomotory wake Ca2+ events were larger in size and duration than events observed in the other two states . However , when we controlled for locomotion we did not find differences in size , duration , or amplitude of events between sleep and stationary wake when these features were compared individually ( Figure 1—figure supplement 1A ) . However , astrocyte events have many other features beyond size , duration , and amplitude , such as event perimeter or propagation . Because of this spatiotemporal complexity , we next used a dimensionality reduction approach , implementing principal component analysis to explore whether astrocyte Ca2+ events differed among behavioral states . This approach allowed us to incorporate 20 different event features calculated by AQuA . We found that the first three principal components ( PCs ) represented spatial- , temporal- , and amplitude-related features respectively . We then focused on the five PCs that explained the most variance in the imaging data ( Figure 1—figure supplement 1B ) and compared them among the three behavioral states . While the largest differences in each PC were between locomotory wake and the other two states , we also found significant differences between sleep and stationary wake in all five PCs examined ( Figure 1—figure supplement 1C ) . Together , this analysis demonstrates that while no state-specific differences are observed by comparisons of individual event features , there are unique spatial , temporal , and amplitude signatures of sleep-specific astrocyte Ca2+ events when multiple features are incorporated . We next examined the relationship between astrocyte Ca2+ event frequency and SWA within stationary behavioral states and found , similar to Figure 1D , a negative correlation between Ca2+ frequency and SWA ( Figure 1H ) . The strong association found between Ca2+ frequency and SWA during sleep , namely high Ca2+ activity during sleep periods of low SWA and vice versa , is suggestive of a possible role of astrocytic Ca2+ specifically in sleep depth . Lastly , we explored the role of IP3R2 in the relationship between astrocyte Ca2+ activity and SWA since IP3R2s are enriched in astrocytes ( Zhang et al . , 2014 ) , underlie a significant fraction of astrocytic Ca2+ dynamics through Ca2+ release from intracellular stores ( Beck et al . , 2004 ) , and IP3R2 KO mice show a total decrease in SWA during NREM sleep ( Bojarskaite et al . , 2020 ) . To test whether the inverse relationship of astrocyte Ca2+ and SWA is dependent on IP3R2s , we imaged astrocyte Ca2+ dynamics over natural sleep and wake in IP3R2 KO mice ( Petravicz et al . , 2008 ) . Similar to previous work ( Srinivasan et al . , 2015 ) , we noted a reduction , but not complete abolishment , of Ca2+ events in IP3R2 KO mice . In IP3R2 KO mice , Ca2+ event rate and SWA were negatively correlated , but the correlation was decreased compared to controls ( Figure 1I ) , suggesting the astrocyte-SWA relationship is at least partially dependent on the IP3R2 . The change in correlation between control and IP3R2 KO was most dramatic in sleep , implicating IP3R2-dependent astrocytic Ca2+ signaling in the regulation of SWA intensity in the sleep state . Since both Gq- and Gi-GPCR signaling can increase Ca2+ in astrocytes through IP3R2s ( Durkee et al . , 2019; Nagai et al . , 2019 ) and astrocytes express many GPCRs that have been implicated in sleep-wake regulation ( Durkee et al . , 2019; Di Castro et al . , 2011; Agulhon et al . , 2008; Kofuji and Araque , 2021 ) , the relationship between astrocytic Ca2+ and SWA may result from astrocytic sensing of sleep-wake cues through GPCR signaling . To understand how astrocyte Ca2+ activity is related to SWA on a shorter time-scale , we asked whether consistent electrophysiological changes occur in the seconds around the onset of astrocyte Ca2+ events . As earlier , we separated the recordings by sleep , stationary wake , and locomotory wake states ( Figure 2A ) . Although SWA was , by definition , highest during sleep , we also observed significant fluctuation between periods of relative high and low SWA within each behavioral state ( Figure 2B ) . We next calculated Ca2+ event-triggered averages of SWA , separated by behavioral state . Because the majority of locomotory wake Ca2+ events were in bursts tied to locomotion onset , we focused on sleep and stationary wake states . We found a pattern in which Ca2+ events were preceded by decreases in SWA and followed by increases in SWA ( Figure 2C , left ) . This modulation was significantly higher during sleep compared to stationary wake ( Figure 2D ) . Further , this SWA modulation was decreased in IP3R2 KO mice ( Figure 2C , right , Figure 2D ) , indicating partial dependence of this relationship on IP3R2s ( as in Figure 1I ) . This specific pattern of SWA change centered on astrocyte Ca2+ events—low SWA before astrocyte events and higher afterward—suggests an active role of astrocytes in regulating sleep depth . Specifically , we speculate that astrocytes may be associated with a homeostatic process that increases SWA in response to a transient decrease in SWA . Although we cannot determine this from the data shown here , several lines of evidence support this hypothesis: astrocytes exhibit Ca2+ increases in response to many neuromodulators associated with decreased low-frequency power ( Ding et al . , 2013; Khan et al . , 2001; Takata et al . , 2011; Shelton and McCarthy , 2000 ) and cortical astrocytes have the ability to increase low-frequency power ( Szabó et al . , 2017; Poskanzer and Yuste , 2016 ) . If , in fact , astrocyte Ca2+ events are ‘triggered’ by decreases in SWA , we would expect to observe more Ca2+ events when SWA is low , which we indeed found in the correlation analysis above ( Figure 1D , H ) . While many other cell types may also play roles in a SWA homeostatic process , we wondered whether astrocytes may be involved in the consistent increase in SWA that we observe after astrocyte Ca2+ event onsets ( Figure 2C ) . To address this question , we next used chemogenetics to specifically manipulate GPCR pathways that shape astrocyte Ca2+ dynamics . To test whether astrocyte Ca2+ may play a causal role in SWA control , we acutely manipulated cortical astrocyte Ca2+ , since genetic manipulations—such as IP3R2 KO—can lead to compensatory developmental effects . Because IP3R2 can mediate the astrocyte-SWA relationship ( Figures 1I and 2C–D ) , and both Gi- and Gq-GPCR mediated Ca2+ changes in astrocytes are dependent on the IP3R2 pathway ( Durkee et al . , 2019; Mariotti et al . , 2016; Nagai et al . , 2019 ) , we chose to use Designer Receptors Exclusively Activated by Designer Drugs ( DREADDs ) ( Roth , 2016 ) to selectively manipulate GPCR pathways in astrocytes . The inhibitory neurotransmitter GABA has been implicated in cortical synchrony during sleep through the mediation of synchronous DOWN states ( Lemieux et al . , 2015; Sheroziya and Timofeev , 2014; Zucca et al . , 2017 ) and the excitatory neurotransmitter glutamate has been implicated in cortical UP states ( Sanchez-Vives and McCormick , 2000; Poskanzer and Yuste , 2011 ) . Astrocytes respond to both GABA and glutamate via Gi-GPCRs ( via GABAB and mGluR3 receptors in adults ) ( Durkee et al . , 2019; Mariotti et al . , 2016; Nagai et al . , 2019 ) . Thus , we chose the inhibitory human M4 muscarinic receptor DREADD ( hM4Di ) to selectively drive this well described Gi-GPCR pathway in astrocytes ( Figure 3 ) . The same experimental setup as earlier ( Figure 1A ) was used , but mice were co-injected with AAV-GFAP-GCaMP6f and AAV-GFAP-hM4D ( Gi ) -mCherry to express both GCaMP6f and Gi-DREADD specifically in cortical astrocytes ( Figure 3A–B , Figure 3—figure supplement 1 ) . In these experiments , we monitored the effects of I . P . administration of the hM4Di agonist clozapine-N-oxide ( CNO , 1 mg/kg ) on Ca2+ dynamics , SWA , and sleep state . Because of the known sedative effects of CNO , we first verified that CNO itself ( 1 mg/kg , I . P ) did not alter Ca2+ dynamics or sleep features in the absence of DREADD expression . We found no change in Ca2+ dynamics or sleep features between administration of 1 mg/kg CNO and the saline control ( Figure 3—figure supplement 2 ) . While Gi-DREADD has been used in astrocytes in vivo previously , its effects on astrocytic Ca2+ have not yet been established during natural wake and/or sleep . Here , we confirmed that Gi-DREADD activation indeed altered astrocyte Ca2+ , causing an increase in event frequency across the entire 2 hr recording period after CNO administration ( Figure 3C–D , Video 2 ) . This finding is consistent with studies of astrocytic Ca2+ activity in ex vivo slices and in anesthetized mice ( Durkee et al . , 2019; Nagai et al . , 2019; Chai et al . , 2017 ) . Next , we asked whether Gi-induced Ca2+ event increases were sufficient to alter SWA . We found that activation of Gi-DREADDs by CNO significantly increased SWA during sleep compared to a saline injection in the same animal ( Figure 3E ) . In contrast , total time spent in sleep and wake was not affected by Gi-DREADD activation ( Figure 3F , H ) . Thus , although the total duration of sleep did not change , the sleep was characterized by higher SWA , or greater sleep depth . Together , these data demonstrate that regulation of SWA and sleep duration can be separated , and that astrocyte Ca2+ , through Gi-GPCR activation , is sufficient to increase SWA during sleep . We hypothesized that astrocytes were part of a homeostatic mechanism regulating SWA , where in response to decreases in SWA , astrocyte Ca2+ causes an increase in SWA . Here , we artificially increased Ca2+ beyond endogenous levels through Gi-GPCR signaling and found we could drive SWA increases above control levels , consistent with the hypothesis that astrocytes are part of a homeostatic mechanism that regulates SWA . Because we found similar relationships between endogenous Ca2+ dynamics and SWA in sleep and stationary wake ( Figure 1H , Figure 2C–D ) , we next quantified the effect of Gi-GPCR activation on SWA during wake . In contrast to the change in SWA during sleep ( Figure 3E ) , we found no change in SWA during the entire wake state ( Figure 3G ) . Likewise , when calculating SWA only in the stationary wake state , we observed no significant difference in SWA ( Figure 3—figure supplement 1f ) . This negative result suggests that a different mechanism underlies the astrocyte-SWA relationship in wake , and assigns the role of Gi-induced Ca2+ dynamics to regulating SWA specifically during sleep . To investigate this difference , we performed PCA on the Ca2+ data collected after saline or CNO administration . We found that CNO resulted in significantly larger differences in multiple PCs for sleep relative to wake ( Figure 3—figure supplement 3 ) . This selective change in Ca2+ event properties during sleep , but not wake , may explain the sleep-specific effects in SWA . Because the astrocyte-SWA relationship is partly dependent on IP3R2s ( Figures 1I and 2C–D ) , we tested whether the effect of Gi-GPCR activation on SWA was also dependent on IP3R2s by repeating these Gi-DREADD experiments in IP3R2 KO mice . Unlike control mice ( Figure 3C–D ) , CNO administration did not significantly increase astrocyte Ca2+ in IP3R2 KO mice ( Figure 3I ) , demonstrating that Gi-DREADD-induced Ca2+ events rely , at least in part , on IP3R2 . In accordance with the lack of change in Ca2+ in the IP3R2 KO animals , we also observed no significant change in SWA with CNO administration ( Figure 3J ) , indicating that the change in sleep depth we observe ( Figure 3E ) is dependent on IP3R2 . While NREM sleep is broadly characterized by SWA , it has become increasingly clear that there are two main types of slow waves: delta waves and slow oscillations ( Genzel et al . , 2014; Kim et al . , 2019; Steriade et al . , 1993; Steriade and Timofeev , 2003; Siclari et al . , 2014; Dang-Vu et al . , 2008; Bernardi et al . , 2018 ) . These two types of slow waves are characterized by different regulatory mechanisms and are associated with distinct functions in NREM sleep . Delta waves are thought to promote the weakening of memories , while slow oscillations support memory consolidation ( Genzel et al . , 2014; Kim et al . , 2019 ) . In light of our finding that astrocytic Gi-GPCR-induced Ca2+ is sufficient to increase sleep SWA ( Figure 3E ) , we explored whether this increase could be attributed to specific changes in delta waves or slow oscillations . A specific change could point to specific roles of astrocytic Gi-signaling in sleep . For this analysis , we implemented an established approach to distinguish delta waves and slow oscillations by their distinct waveforms ( Kim et al . , 2019 ) . Slow oscillations had larger positive peaks and larger positive-to-negative deflections that occurred within 500 ms ( Figure 4A–B ) . Across recordings , slow oscillations and delta waves were differentiated by their peak and trough amplitudes using k-means clustering ( Figure 4C ) . We first looked at the effect of astrocytic Gi-DREADD activation on the number of identified delta waves and slow oscillations , and found no effect on the rate of delta waves or slow oscillations during sleep ( Figure 4E ) . This negative result was expected by this analysis , because delta waves and slow oscillations were identified using amplitude percentile thresholds ( see Materials and methods ) that were set for each individual recording . However , when quantifying the amplitude of these waveforms , we noted increases in the mean amplitude , particularly for delta waves ( Figure 4F , I ) . Indeed , by plotting peak vs . trough amplitude , we observed a clear change in delta waves after CNO , resulting in higher peak and lower trough amplitudes ( Figure 4G ) . This change was smaller in the slow oscillation waveforms ( Figure 4J ) . Similarly , we quantified the change in total peak – trough amplitude after CNO administration . While we saw a significant increase in size for delta waves ( Figure 4H ) and slow oscillations ( Figure 4K ) compared to saline controls in the same animal , the change in delta waves was significantly higher than that for slow oscillations ( Figure 4L , Figure 4—figure supplement 1 ) . Together , these data demonstrate that astrocyte Ca2+ , through Gi-GPCR signaling , preferentially increases SWA by altering delta wave amplitude . Delta waves are more local than slow oscillations and are thought to be generated within the cortex ( Genzel et al . , 2014; Siclari and Tononi , 2017; Siclari et al . , 2014; Bernardi et al . , 2018; Spoormaker et al . , 2010; Nir et al . , 2011 ) . Given that our Gi-astrocytic manipulation is restricted to a small portion of cortex ( Figure 3—figure supplement 1 ) , the result that astrocytic Gi-DREADD activation affects delta waves more than slow oscillations may indeed be expected . We next wondered whether astrocytes might play a role beyond the regulation of sleep depth , to also influence sleep duration . Data here ( Figure 2 ) suggest that a component of astrocyte signaling may be important for sleep/wake state transitions , which would directly affect sleep duration . To study these transitions , we first examined endogenous cortical astrocyte Ca2+ dynamics in the 30 s leading up to transitions between sleep or wake . We found a pattern in which Ca2+ events consistently increased before the sleep-to-wake transition and decreased before the wake-to-sleep transition ( Figure 5A ) . This is supported by a recent study that demonstrated , using an alternative image analysis technique , that Ca2+ increases preceding sleep-to-wake transitions ( Bojarskaite et al . , 2020 ) . We next divided all sleep and wake periods , regardless of length , into three equal bins ( Figure 5C ) . This allowed us to study how astrocyte Ca2+ dynamics generally change throughout a sleep or wake period . In so doing , we found that Ca2+ event rate increased in the last third of sleep and decreased in the last third of wake ( Figure 5D ) . Since Ca2+ event rate is higher during wake than sleep ( Figure 1F ) and Ca2+ events occur after dips in SWA ( Figure 2C ) , the increase in event rate preceding the transition to wake could reflect a gradual shift in SWA to a wake state . In fact , various ascending brainstem neuromodulatory neurons associated with wakefulness have been shown to increase firing prior to the transition to wake and decrease firing prior to the transition to sleep ( Takahashi et al . , 2006; Eban-Rothschild et al . , 2016; Aston-Jones and Bloom , 1981; Lee et al . , 2005; Trulson and Jacobs , 1979 ) . Astrocytes express receptors and exhibit increased Ca2+ dynamics in response to many of these neuromodulators ( Ding et al . , 2013; Khan et al . , 2001; Takata et al . , 2011; Shelton and McCarthy , 2000 ) . Thus , this change in event rate prior to sleep/wake transitions may be due to neuromodulator-driven GPCR signaling in astrocytes . We reasoned that if the Ca2+ dynamics observed around state transitions were due to astrocytic GPCR signaling , we would expect that these Ca2+ dynamics would be altered in IP3R2 KO mice . As predicted , we found that the changes in event rate preceding transitions were abolished in IP3R2 KO mice ( Figure 5B ) . When quantifying the change in event rate in the last third of sleep and wake for IP3R2 KO mice , we found that IP3R2 KO mice did not exhibit the same increase in event rate in the last third of sleep ( Figure 5D , left ) . However , the change in event rate observed during wake was unchanged in IP3R2 KO mice ( Figure 5D , right ) , suggesting a specific role of IP3R2s in sleep . Since Gi-DREADD activation did not affect sleep duration ( Figure 3F ) , we next tested the hypothesis that Gq-GPCR-mediated Ca2+ signaling in astrocytes regulates sleep/wake transitions . To drive the astrocytic Gq-GPCR pathway and test for a role of astrocyte Ca2+ in mediating sleep/wake transitions , we selectively expressed the human M3 muscarinic receptor DREADD ( hM3Dq ) in astrocytes . We were also motivated by the knowledge that neuromodulatory signals play an important role in mediating sleep and wake transitions ( Holst and Landolt , 2018; Saper and Fuller , 2017; Lee and Dan , 2012; Scammell et al . , 2017 ) , and many of these endogenous signals can act at Gq-GPCRs in astrocytes ( Zhang et al . , 2014; Chai et al . , 2017 ) . We used a similar approach as above ( Figure 3A ) , but here selectively expressed GCaMP6f and the Gq-DREADD in astrocytes ( Figure 6A–B , Figure 6—figure supplement 1 ) . As above , we imaged astrocyte Ca2+ after I . P . CNO administration to confirm the effect of Gq-DREADD activation on Ca2+ activity in vivo . Although astrocytic Gq-DREADD activation in vivo has been performed previously ( Durkee et al . , 2019; MacDonald et al . , 2020; Bonder and McCarthy , 2014; Adamsky et al . , 2018 ) , validation of Gq-DREADD-mediated astrocytic Ca2+ increases has only been performed under anesthesia or ex vivo , in part because several in vivo astrocyte DREADD experiments have been carried out in brain regions that are less accessible than cortex ( MacDonald et al . , 2020; Adamsky et al . , 2018 ) . Thus , the effect of Gq-DREADD activation on Ca2+ in awake mice has not been previously reported . Canonically , Gq-GPCR signaling results in an increase in Ca2+ activity via IP3-dependent release of intracellular Ca2+ ( Petravicz et al . , 2008 ) . However , we were surprised to find that Ca2+ dynamics only increased in the first 5–10 min after I . P . injection of CNO ( 150 . 9% ± 135 . 9 , Video 3 ) . After this initial period of increased Ca2+ events , Ca2+ dynamics were almost completely abolished ( −97 . 3% ± 0 . 79% , Figure 6C–D , Figure 6—figure supplement 1H , Video 3 ) . This ‘silent’ state of Ca2+ dynamics lasted for the rest of the entire recording ( 2–3 hr ) . To test whether this unexpected result was due to CNO concentration ( 1 mg/kg ) , we administered lower doses of CNO . While the initial period of increased Ca2+ dynamics was slightly longer ( 15–20 min ) following administration of a ten-fold lower dose of CNO ( 0 . 1 mg/kg ) , this very low dose still resulted in a strong reduction in Ca2+ events for long time periods ( Figure 6—figure supplement 1F–h ) . The observed inhibition of astrocyte Ca2+ could be due to depletion of intracellular Ca2+ stores and/or interference with store-operated Ca2+ channels ( Sakuragi et al . , 2017 ) . To compare the inhibition of Ca2+ events with changes in fluorescence , we used an ROI-based approach to analyze fluorescence in somas and processes after CNO administration . We found that fluorescence in both somas and processes remained elevated above baseline after 1 mg/kg CNO ( Figure 6—figure supplement 1I ) , which suggests that Ca2+ levels may be clamped at saturating levels . Together , these results indicate that ( 1 ) we cannot assume that Gq-DREADD activation simply increases astrocytic Ca2+ in vivo , and ( 2 ) when feasible , astrocytic experiments using chemogenetics in vivo should be validated individually , particularly for those involving circuit function and animal behavior . To test whether the astrocytic Ca2+ silencing we observed following Gq-DREADD activation could be reproduced with endogenous GPCR signaling , we measured Ca2+ activity in ex vivo cortical slices in response to a cocktail of neuromodulators associated with wakefulness , including norepinephrine , acetylcholine , dopamine , and histamine . We adapted methodology ( Ding et al . , 2016 ) , using half the concentration of each neuromodulator as previously , since each experiment involved two total applications of this ‘wake cocktail’ ( 20 μM norepinephrine , 5 μM acetylcholine , 5 μM dopamine , 2 . 5 μM histamine ) . We also included TTX in the circulating bath to block neuronal firing . As predicted from previous studies reporting astrocytic Ca2+ increases to various neuromodulators ( Ding et al . , 2013; Khan et al . , 2001; Takata et al . , 2011; Shelton and McCarthy , 2000; Pankratov and Lalo , 2015 ) , we observed a dramatic increase in Ca2+ activity in response to the cocktail ( Figure 6—figure supplement 1J ) . However , after this initial increase in Ca2+ , GCaMP fluorescence did not return to baseline levels , but remained high and further Ca2+ events were almost completely absent ( Figure 6—figure supplement 1J ) , similar to in vivo dynamics observed 5–10 min after CNO administration . To test whether this ‘silent’ state altered the astrocytic response to further neuromodulatory input , we bath-applied a second round of the wake cocktail . In contrast to the initial Ca2+ increase , we observed no further increase in astrocyte Ca2+ ( Figure 6—figure supplement 1J ) . We speculate that the mechanism underlying the inability of astrocytes to respond to a second dose of wake cocktail may be similar to that underlying the inhibition of Ca2+ dynamics in vivo in response to circulating CNO . The finding that Gq-DREADD chemogenetics can inhibit an intracellular GPCR signaling pathway in astrocytes makes this is a particularly useful tool for understanding astrocytes' roles in cortical state regulation . To investigate whether astrocytes regulate sleep duration , we focused on the long period of Ca2+ suppression in these experiments . We found that mice spent significantly more time in sleep after CNO administration ( Figure 6E , left ) . Further , in the absence of Gq-GPCR-mediated Ca2+ events , mice made fewer sleep-to-wake transitions ( Figure 6F , left ) and accordingly , we observed fewer sleep bouts of longer duration ( Figure 6—figure supplement 2A ) . This suggests that the IP3R2-dependent increase in event rate prior to sleep-to-wake transitions ( Figure 5 ) is important to transition the cortex to the wake state . The transition data ( Figure 5 ) also showed that endogenous Ca2+ decreases toward the end of wake periods , just prior to wake-to-sleep transitions ( Figure 5 ) . Thus , we wondered whether Ca2+ suppression via Gq-DREADDs would affect wake as well . We observed a decrease in the percent time awake ( Figure 6E , right ) , as predicted by the increase in sleep observed ( Figure 6E , left ) . However , we also observed less frequent transitions out of wake , demonstrating that astrocyte Ca2+ is important for both wake-to-sleep transitions ( Figure 6F , right ) , and sleep-to-wake transitions ( Figure 6F , left ) . As predicted from the decrease in transitions , we also observed fewer wake bouts and wake bouts of longer duration ( Figure 6—figure supplement 2B ) . We hypothesize that decreased astrocytic Ca2+ prior to wake-to-sleep transitions ( Figure 5 ) is important for the transition to sleep , but astrocytes were unable to make this significant decrease due to clamped Ca2+ in these experiments . If Gq-GPCR signaling is an important bidirectional regulator of sleep/wake transitions , we would expect that increases in Gq-GPCR signaling to have the opposite effect from decreased Gq-GPCR Ca2+ signaling . We thus used data from the short , initial period with elevated Ca2+ activity to ask whether this is the case . Because this period is so short ( 5–10 min ) , we were somewhat limited in our analysis . However , of the animals that exhibited some sleep in either the CNO or saline condition ( n = 4 ) , we observed a significant decrease in the percent time sleeping ( Figure 6G ) . This bidirectional change in sleep time strongly supports the hypothesis that Gq-GPCR-mediated Ca2+ plays a critical role in regulating sleep duration . Interestingly , we did not observe a change in the amount of sleep with Gi-GPCR activation ( Figure 3F ) , which similarly increased Ca2+ dynamics . This difference between Gq- and Gi-mediated Ca2+ increases indicates an important functional dissociation between Gq- and Gi-GPCR-mediated Ca2+ activity in astrocytes and highlights the likelihood that other signaling molecules involved in GPCR signaling cascades play roles in regulating sleep-wake transitions . Because we observed a significant increase in sleep depth in response to the Ca2+ increase with Gi-DREADDs ( Figure 3E ) , we also wondered whether Ca2+ suppression via Gq-DREADDs would have an opposing effect . In contrast to manipulation of the Gi-GPCRs , we found that Ca2+ suppression via Gq-GPCR manipulation had no significant effect on SWA during sleep ( Figure 6H ) . This suggests that astrocytic regulation of SWA is specifically dependent on the Gi-GPCR pathway and provides further evidence that astrocytic Gi- and Gq-GPCR signaling regulate separable sleep/wake features . SWA during NREM sleep is considered a widespread phenomenon , involving the synchronization of neurons across the entire cortex . While widespread oscillatory activity has been observed in several animal models ( Lemieux et al . , 2015; Amzica and Steriade , 1995 ) , recent work has also emphasized the existence of more local and asynchronous sleep ( Genzel et al . , 2014; Huber et al . , 2004; Funk et al . , 2016; Siclari and Tononi , 2017; Bernardi et al . , 2018; Nir et al . , 2011 ) . The morphology and interconnectedness of cortical astrocytes and astrocytic networks make them well positioned to mediate neural activity across broad swaths of cortex . Cortical astrocytes are non-overlapping , in all cortical layers , gap junctionally coupled , and contain highly ramified processes that can contact tens of thousands of synapses ( Halassa et al . , 2007 ) . We therefore wondered how they may be involved in both local and remote changes in cortical synchronization in sleep . To address this question , we implanted a second electrode to record EEG in the contralateral frontal cortex ( FC-EEG , Figure 7A ) . This second electrode was far ( both rostral-caudally and medial-laterally ) from the imaging window/LFP electrode in V1 , but still over cortex ( Figure 7A , Figure 3—figure supplement 1 , Figure 6—figure supplement 1 ) . With two recording sites , we first explored endogenous relationships between astrocytic Ca2+ events in V1 and cortical state in contralateral FC in sleep . Using Ca2+ event-triggered averages , we found a similar relationship with V1 Ca2+ events and FC as previously described ( Figure 2C ) : SWA in FC decreased before and increased after V1 Ca2+ event onsets , although the magnitude of this modulation was smaller than that observed locally ( Figure 7C ) . To look at the synchronization between these cortical areas , we examined the coherence between local V1 and remote FC oscillations . We found that the V1-FC coherence ( between 5 and 10 Hz ) was higher immediately following astrocyte Ca2+ events when compared to coherence measured from randomly chosen epochs ( Figure 7D ) . We also found an increase in astrocyte event-locked coherence ( 0–15 Hz ) at the end of sleep periods , in the 15 s prior to sleep-to-wake transitions ( Figure 7E ) . These data provide evidence that astrocytes may be involved in mediating endogenous cortex-wide physiological activity . We next tested whether astrocytes play a causal role in brain-wide SWA during sleep using Gi-DREADD activation in V1 and the FC-EEG . To assess how the spread of astrocytic DREADD expression compared with the location of the two recording electrodes , we performed immunohistochemistry on brain slices across the rostral-caudal axis ( Figure 3—figure supplement 1D ) . As expected , the majority of expression was centered around the V1-LFP electrode where viruses had been injected , while no expression was observed in FC at the site of the EEG electrode ( Figure 3—figure supplement 1E ) . To assess a causal role for astrocyte Ca2+ in brain-wide oscillatory activity , we compared the effect of Gi-DREADD activation of V1 astrocytes at both V1-LFP and FC-EEG electrodes . Here , we found that increasing Ca2+ via Gi-DREADDs in V1 was sufficient to increase SWA in the contralateral frontal cortex , although this increase was smaller than that observed in V1 ( Figure 7F–H ) . We also found that the SWA change was accounted for by a significant increase in slow oscillation amplitude , but not delta wave ( Figure 7I; Figure 7J ) . This is in contrast with the greater delta wave change observed locally ( Figure 4I ) . Moreover , the change in slow oscillation amplitude in FC ( Figure 7J , 6 . 8 ± 2 . 2% ) was similar to the change of the slow oscillation amplitude in V1 ( Figure 4L , 8 . 8 ± 3 . 2% ) , suggesting that slow oscillations generated in V1 travelled to FC . These results indicate that astrocytes can influence cortex-wide dynamics on a large scale via specific changes to the slow oscillation component of SWA .
We found that astrocytic Gi-DREADD activation increases SWA , but not sleep duration , while Gq-DREADD activation altered sleep duration but not SWA . This suggests a separation in the mechanisms underlying sleep depth , measured by SWA , and sleep duration . Previous work has also shown separation in sleep depth and duration by demonstrating that following sleep deprivation , recovery sleep has higher sleep depth but the duration is not significantly changed ( Dijk and Beersma , 1989; Patrick and Gilbert , 1896 ) . On the other hand , Ca2+ has been suggested to mechanistically link the regulation of sleep duration and depth ( Ode et al . , 2017; Tatsuki et al . , 2016 ) . Interestingly , our data supports both these ideas . We found that endogenous astrocyte Ca2+ is modulated in relation to both sleep duration ( Figure 5 ) and sleep depth ( Figure 1 ) . Nevertheless , we were able to affect one without the other by selectively manipulating different GPCR pathways , suggesting these mechanisms are also separable . The two-process model of sleep regulation has attributed the regulation of sleep duration to the interaction of sleep pressure , measured by SWA , and circadian rhythm ( Borbély et al . , 2016; Daan et al . , 1984; Borbely , 2016 ) . Our findings that Gi-DREADD activation increased SWA without affecting sleep duration ( Figure 3E , F ) suggests that SWA does not directly influence sleep duration . However , we cannot discount the two-process model from this data alone , since we did not investigate the effect of circadian rhythm , nor did we directly study sleep homeostasis; our recordings were performed at the same time of day and were limited to 2–3 hr . In fact , astrocyte Ca2+ changes with circadian rhythm in the suprachiasmatic nucleus ( Brancaccio et al . , 2017 ) and increases with sleep need after sleep deprivation ( Ingiosi et al . , 2020 ) . Longer recordings and sleep deprivation interventions to examine astrocytic integration of circadian signals and sleep pressure will be informative . Additionally , our methodology led us to focus on the role of astrocyte GPCR signaling in NREM sleep regulation , although previous work has demonstrated astrocyte Ca2+ changes with REM sleep ( Bojarskaite et al . , 2020; Ingiosi et al . , 2020; Foley et al . , 2017 ) . The further study of astrocytic regulation of REM sleep may reveal interesting differences between regulation of behavioral sleep and cortical state , which is similar between REM sleep and wake . Indeed , our data suggests that Gi-GPCR signaling would be much attenuated during REM sleep , which is characterized by a lack of SWA . Since astrocytes differentially control SWA and sleep/wake transitions , we hypothesize that Gi- and Gq- GPCR activation in astrocytes—while both drive Ca2+ changes—lead to different downstream effects which may elucidate new mechanisms of sleep regulation . In fact , many astrocytic functions associated with sleep may be important , such as extracellular glutamate regulation ( Poskanzer and Yuste , 2016 ) , extracellular ion dynamics ( Ding et al . , 2016 ) , and adenosine release ( Halassa et al . , 2009 ) . While we don’t yet know what downstream astrocytic effects underlie the sleep changes observed here , we have established that the functional astrocytic output is a not a simple consequence of changed Ca2+ levels in the cell , but rather of signaling downstream of either Gi- or Gq-GPCRs . Many new optical sensors , such as those for glutamate ( Marvin et al . , 2013 ) , ATP ( Lobas et al . , 2019; Kitajima et al . , 2020 ) , and adenosine ( Wu , 2020 ) , in combination with astrocyte-specific manipulations , may be useful to link specific GPCR-driven Ca2+ dynamics with relevant astrocyte outputs . The activation of both astrocytic Gi- and Gq-GPCRs increases intracellular Ca2+ ( Durkee et al . , 2019; Mariotti et al . , 2016; Nagai et al . , 2019 ) , and yet we observed a functional dissociation between manipulation of the Gi- and Gq-GPCR pathways . These data underscore the complexity of Ca2+ signals in astrocytes and demonstrate that caution is necessary when attributing astrocytic Ca2+ increases to one specific downstream function . Many different roles have been attributed to astrocytes , suggesting astrocytes have the capacity to perform several functions in parallel . Our findings suggest that , through GPCR signaling , astrocytes interpret Ca2+ dynamics within the cell differently , resulting in different functional outputs . This may be a consequence of the many other signaling molecules downstream of Gi- and Gq-GPCRs , including phospholipase C or protein kinase A . Tools such as AQuA ( Wang et al . , 2019 ) that allow the accurate capture of complex astrocyte Ca2+ signaling are a first step in elucidating how Ca2+ dynamics map to the myriad of functions associated with astrocytes . The next step we took was to extract meaning from these signals by analyzing the multi-dimensional nature of astrocyte Ca2+ signals . We used PCA to reduce the 20 different properties describing each Ca2+ event and revealed differences that were not observed with individual comparisons of event rate , duration , size , or amplitude ( Figure 1—figure supplement 1 , Figure 3—figure supplement 3 ) . This both illustrates the complexity of astrocyte Ca2+ signaling and emphasizes the importance of implementing more robust analysis tools . Understanding how Gi- and Gq-GPCR activation gives rise to different effects on sleep will require further examination of these signaling pathways in astrocytes . First , we will need to identify the specific endogenous ligands during sleep that alter SWA and sleep/wake transitions . Candidates for sleep/wake transitions include neuromodulators , since many Gq-GPCRs for neuromodulators are expressed by astrocytes . In contrast , regulation of SWA has been attributed to both GABA ( Lemieux et al . , 2015; Sheroziya and Timofeev , 2014; Zucca et al . , 2017 ) and glutamate ( Sanchez-Vives and McCormick , 2000; Poskanzer and Yuste , 2011 ) . GABA and glutamate are attractive candidate endogenous ligands because astrocytic GABAergic ( via GABAB ) and glutamatergic ( via mGluR3 in adults ) signaling are both mediated via Gi-GPCRs . Acute , astrocyte-specific knock-out of these receptors will provide important insight into the relevant receptors . Importantly , astrocyte-specific knock-outs will also reveal whether Gi-GPCR signaling is necessary to regulate SWA . While we demonstrated that astrocytic Gi-GPCR signaling is sufficient to alter SWA , other signaling cascades and cell types may also play important roles in SWA regulation . Second , further studies will be required to understand how intracellular signaling cascades for Gi- and Gq-GPCRs differ in astrocytes . Using PCA , we found differences in the effect of Gi-DREADD activation on Ca2+ in sleep versus wake , suggesting the action of the Gi-DREADD may be interacting with endogenous signaling that differs across behavioral states . Both Gi- and Gq-GPCRs increase Ca2+ via IP3R2 ( Durkee et al . , 2019 ) , and we similarly noted a partial dependence on IP3R2 in both Gi- and Gq-GPCR-mediated sleep effects ( Figures 3 and 5 ) . This only partial dependence on IP3R2s in this data could be due to compensation for global IP3R2 absence , but it could also indicate that perhaps other signaling molecules unique to Gq- and Gi-signaling are critical for the sleep features described here . In fact , 1 . 4% of all astrocyte transcripts are regulated by sleep/wake state ( Bellesi et al . , 2015 ) and multiple biochemical assays have already identified various important molecules in sleep/wake regulation ( Suzuki et al . , 2013; Funato et al . , 2016; Mikhail et al . , 2017 ) . Similar molecular studies specifically focused on astrocytic GPCR signaling will be critical to further understand the regulation of sleep duration and depth . While astrocytes have previously been implicated in sleep physiology ( Bellesi et al . , 2015; Ding et al . , 2016; Halassa et al . , 2009; Papouin et al . , 2017; Petit and Magistretti , 2016; Bellesi et al . , 2018; DiNuzzo and Nedergaard , 2017; Bojarskaite et al . , 2020; Ingiosi et al . , 2020; Foley et al . , 2017; Ulv Larsen et al . , 2020; Frank , 2013; Clasadonte et al . , 2017 ) , we present the first example of an acute in vivo astrocytic manipulation that changes natural sleep . Acute manipulation via chemogenetics was advantageous because DREADD activation mimics endogenous signaling pathways known to be important in astrocyte signaling . However , it is still critical to properly validate these tools specifically in astrocytes , especially since they were developed and have been more widely used in neurons . While astrocytic Gq-DREADD activation can increase Ca2+ under anesthesia ( Durkee et al . , 2019; Bonder and McCarthy , 2014 ) , we report for the first time the effect of Gq-DREADD activation on astrocyte Ca2+ in awake mice , both for long time periods ( 2–3 hr ) and with several CNO concentrations . Gq-DREADD activation increased Ca2+ only for a short time after CNO injection , after which we observe a complete suppression of Ca2+ activity for several hours , for all CNO concentrations ( Figure 6C–D , Figure 6—figure supplement 1 ) . This unexpected result points to the importance of validating DREADD responses for each in vivo experiment when feasible . We think that it is most likely that the opposite effects of CNO on Ca2+ activity in Gi-DREADD- and Gq-DREADD-expressing astrocytes reveal important differences in the Gi and Gq signaling pathways . However , these results may also indicate that expression levels of Gi- and Gq-DREADD may be more different than expected based on the observed immunostaining . Another possibility is that differential Ca2+ responses may be caused by differences in CNO action on Gi- and Gq-DREADD receptors . For example , CNO may be less efficacious on Gi-DREADDs in astrocytes . This possibility could be tested by determining whether higher doses of CNO are sufficient to suppress Ca2+ with Gi-DREADD . Two unique slow-waves have been characterized in NREM sleep: delta waves and slow oscillations ( Genzel et al . , 2014; Kim et al . , 2019; Steriade et al . , 1993; Steriade and Timofeev , 2003; Siclari et al . , 2014; Dang-Vu et al . , 2008; Bernardi et al . , 2018 ) . Here , we found that astrocytic Gi-DREADD activation increases SWA by preferentially increasing the amplitude of delta waves in V1 ( Figure 4 ) . This data supports existing literature suggesting delta waves are generated locally within the cortex by the spreading of DOWN states ( Genzel et al . , 2014; Siclari and Tononi , 2017; Siclari et al . , 2014; Bernardi et al . , 2018; Spoormaker et al . , 2010; Nir et al . , 2011 ) . Since our Gi-DREADD manipulation was restricted within the cortex ( Figure 3—figure supplement 1 ) and DOWN states are thought to be generated through GABAergic inhibition ( Chen et al . , 2012; Lemieux et al . , 2015; Sheroziya and Timofeev , 2014; Zucca et al . , 2017 ) , we hypothesize that Gi-GPCR signaling in astrocytes mediates the local synchronization of delta waves via control of inhibition . Additionally , astrocytes may also mediate synchronization of UP states through glutamate ( Sanchez-Vives and McCormick , 2000; Poskanzer and Yuste , 2011 ) . In addition to the increase in delta waves , we observed a smaller , but significant , increase in slow oscillation amplitude that was equal in magnitude to that observed in contralateral FC ( Figure 7 ) . This suggests that cortical astrocytes may have influence over more global , cortex-wide neural activity . We explored this further and found that SWA in contralateral FC was modulated around endogenous Ca2+ events recorded in V1 . Further , we found that coherence between V1 and FC was increased immediately following astrocyte Ca2+ events . Interestingly , coherence was increased in the range of 5–10 Hz , which is higher than expected for slow oscillations , but might also indicate a role for astrocytes in the connectivity across cortex during REM or wake . The synchronization across broad areas of cortex may involve astrocytic gap junctions ( Szabó et al . , 2017; Clasadonte et al . , 2017 ) , which could mediate fast recruitment of neurons in synchronous waves . If this is indeed the mechanism , our findings indicate that GPCR activation regulates gap junction coupling in astrocytes , which can be explicitly tested ( Murphy-Royal et al . , 2020 ) . Slow oscillations are more global than delta waves . Here , the change in slow oscillation amplitude with CNO administration was similar in V1 and FC ( Figure 7J ) . One explanation for this finding is that activation of V1 astrocytes is sufficient to recruit subcortical circuitry , such as thalamocortical circuits , that can underlie brain-wide synchronous events ( Steriade , 2006; Crunelli et al . , 2018 ) . This hypothesis is supported by studies showing subcortical ‘bottom-up’ regulation of slow oscillations ( Steriade et al . , 1993; Siclari et al . , 2014; Bernardi et al . , 2018 ) and a role of astrocytes in mediating communication between different brain areas ( Kol , 2019; Sardinha et al . , 2017 ) , and could be tested by simultaneously recording from thalamus and cortex during astrocyte activation . Recent work indicates that slow oscillations and delta waves have distinct functions in memory during NREM sleep ( Genzel et al . , 2014; Kim et al . , 2019 ) . Since Gi-driven astrocyte Ca2+ preferentially drives changes in delta waves locally , we might expect that astrocytic activity in sleep is more involved in forgetting than memory consolidation . This could be tested by quantifying the effect of Gi-DREADD activation during sleep following a learning paradigm , such as fear conditioning . Further , a light-activated Gi-GPCR ( Siuda et al . , 2015 ) in astrocytes would provide temporal control for selective Gi-DREADD activation specifically during NREM sleep following learning , to further explore how astrocytic effects on sleep impact cortical memory functions .
All procedures were carried out using adult mice ( C57Bl/6 , P50–100 ) in accordance with protocols approved by the University of California , San Francisco Institutional Animal Care and Use Committee ( IACUC ) . All animals were housed in a 12:12 light-dark cycle with food and water provided ad libitum . Male and female mice were used for all experiments . IP3R2 KO mice ( Dr . Katsuhiko Mikoshiba , RIKEN ) carry null alleles for Itpr2 . Following surgery , all animals were singly housed , to protect electrodes , with additional enrichment . Adult mice ( C57Bl/6 , P50–100 ) were administered dexamethasone ( 5 mg/kg , s . c . ) prior to surgery and anesthetized with isoflurane . A custom-made titanium headplate was attached to the skull using C and B Metabond ( Parkell ) , and a 3 mm diameter craniotomy was created over visual cortex . A titanium wire was inserted in V1 lateral to the craniotomy , and a bone screw was inserted in contralateral V1 for reference ( all measurements from bregma , −3 . 5 mm , 1 . 2 mm lateral ) . Two twisted titanium wires were inserted in the nuchal muscles for EMG recordings . In a subset of animals , an additional bone screw for EEG was inserted into FC , contralateral to the craniotomy ( +2 . 7 mm , 1 . 2 mm lateral ) . For endogenous Ca2+ imaging , two 300 nL injections of AAV5-GFaABC1D . cyto-GCaMP6f were made in the brain before placing the cranial window . For Gi-DREADD experiments , two injections of AAV5-GFaABC1D . cyto-GCaMP6f ( 200 nL each , 400 nL total ) and AAV5-GFAP-hM4D ( Gi ) -mCherry ( 100–200 nL each , 200–400 nL total ) were co-injected . For Gq-DREADD experiments , two injections of AAV5-GFaABC1D . cyto-GCaMP6f ( 200–300 nL each , 400–600 nL total ) and AAV5-hM3D ( Gq ) -mCherry ( 200–500 nL each , 400–1000 nL total ) were injected before placing the cranial window . All injections were 0 . 2–0 . 3 mm from the pial surface , −0 . 5 – −3 . 5 mm , 1 . 2–2 . 5 mm lateral at 30–60 nL/min , followed by a 10 min wait for diffusion . Following viral injection , a glass cranial window for chronic imaging was implanted and secured using C and B metabond ( Goldey et al . , 2014 ) . Post-operative care included administration of 0 . 05 mg/kg buprenorphine and 5 mg/kg carpofen . Mice were allowed 10 days to recover , then were habituated to head-fixation on a circular treadmill for 5 days , prior to imaging . For DREADD experiments , mice were habituated for 1–2 days . 2P imaging experiments were carried out on a microscope ( Bruker Ultima IV ) equipped with a Ti:Sa laser ( MaiTai , SpectraPhysics ) . The laser beam was intensity-modulated using a Pockels cell ( Conoptics ) and scanned with linear galvonometers . Images were acquired with a 16x , 0 . 8 N . A . Nikon objective via a photomultiplier tube ( Hamamatsu ) using PrairieView ( Bruker ) software . For GCaMP imaging , 950 nm excitation and a 515/30 emission filter was used . All recordings started at ZT 2 . Mice were head-fixed to a circular treadmill and Ca2+ activity was recorded at ~ 1 . 7 Hz effective frame rate from layer 2/3 of visual cortex with a 512 × 512 pixel resolution at ~ 1 μm/pixel . Locomotion speed was monitored using an optoswitch ( Newark Element 14 ) connected to an Arduino . For LFP and FC-EEG , differential recordings were acquired using the contralateral bone screw as a reference . For EMG , differential recordings were acquired using the two wires implanted in the nuchal muscles . All recordings were amplified ( Warner ) with a gain of 1K , high-pass filtered at 0 . 1 Hz , and low-pass filtered at 10 KHz . Electrophysiology and locomotion recordings were acquired simultaneously with 2P imaging at 1 KHz using PrairieView ( Bruker ) software . Astrocyte Ca2+ image analysis was performed using Astrocyte Quantitative Analysis ( AQuA ) software ( Wang et al . , 2019 ) . Videos were preprocessed by registering images using the ImageJ plugin MOCO ( Dubbs et al . , 2016 ) . Events were detected using AQuA ( in MATLAB ) using the in-vivo-GCaMP-cyto preset . Signal detection threshold was adjusted for each video after manually checking for accurate detection , to account for slight differences in noise . AQuA outputs were further analyzed in MATLAB . Event count was quantified using the onset of each event , as detected by AQuA . For ROI analysis , somatic traces were extracted from ROIs hand-drawn using the blow-lasso tool in Fiji . Somatic ROIs were then removed using a mask , and process ROIs were created by applying 10 um2 tiles across the field of view . LFP and EEG recordings were first manually inspected for movement artifacts , which were removed by excluding data exceeding 5 SD from the mean . Similarly , drifting baselines were adjusted by a high-pass filter with a cut-off frequency of 0 . 3 Hz . All electrophysiology acquired on the same day were pooled together and z-scored . A spectrogram was then calculated using a moving window of 10 s , stepping every 5 s . Locomotory data were used to identify each 5 s bin as stationary if no locomotion was detected , or locomotory otherwise . The absolute value of z-scored EMG recordings was used to quantify mean EMG amplitude for each 5 s bin . A bin was identified as NREM sleep if ( 1 ) the slow-wave ratio ( 0 . 5–4 Hz/8–20 Hz ) was > 0 . 5 SD from the mean , ( 2 ) the animal was stationary , and ( 3 ) the EMG was < 5 SD from the mean . Similarly , a bin was identified as REM sleep if it had not been quantified as NREM , and ( 1 ) theta power ( 6–10 Hz ) >0 . 25 SD from the mean , ( 2 ) the animal was stationary , and ( 3 ) the EMG was < 0 . 4 SD from the mean . All remaining times were characterized as wake . Each behavioral period was identified by finding the start and end of consecutive 5 s bins of the same behavioral state . For all analysis , sleep periods < 10–15 s were excluded , with the exception of the sleep/wake transition analysis ( Figure 5 ) , in which sleep periods < 30 s were excluded . Each wake period was further divided into 1 s bins , and characterized as stationary if no movement was detected and locomotory otherwise . Consecutive 1 s bins of locomotion within each wake period were identified as a locomotory wake period and consecutive 1 s bins of no locomotion within each wake period were identified as a stationary wake period . Stationary wake periods < 15 s were excluded . We included a ‘buffer’ in which the first 10 s of a stationary wake period was excluded if that stationary wake period immediately followed a locomotory wake period because Ca2+ bursts during locomotion often persist for ~ 10 s after locomotion ceases . To differentiate slow oscillations and delta waves , we set thresholds for amplitude and peak-to-trough duration ( Kim et al . , 2019; Siclari et al . , 2014; Dang-Vu et al . , 2008; Nir et al . , 2011; Riedner et al . , 2007 ) . First , LFP and EEG was filtered for the slow wave band ( 0 . 1–4 Hz ) using two filters: a high-pass Butterworth filter ( second order , cutoff at 0 . 1 Hz ) and a low-pass Butterworth filter ( fourth order , cut-off at 4 Hz ) . Next , we identified all positive-to-negative zero crossings , preceding peaks , and following troughs that occurred during NREM sleep . Because gamma oscillations are nested in UP states ( Steriade et al . , 1996; Valderrama et al . , 2012; Wolansky et al . , 2006; Mena-Segovia et al . , 2008 ) , we used high-gamma to verify that the identified peaks were DOWN states and the troughs were UP states: LFP and EEG recordings were bandpass filtered for 80–100 Hz , and high-gamma amplitude was quantified during peaks and troughs . If the mean peak high-gamma was greater than the mean trough high-gamma , we inverted the signal and repeated the analysis . Slow oscillations were identified as zero crossings with ( 1 ) preceding peaks >85th percentile , ( 2 ) following troughs <40th percentile , and ( 3 ) peak-to-trough duration of 150–500 ms . Delta waves were identified as zero crossings with ( 1 ) preceding peaks <85th percentile , ( 2 ) following troughs <40th percentile , and ( 3 ) peak-to-trough duration > 100 ms . At the start of each experiment ( ZT 2 ) , mice were weighed and head-fixed on the treadmill . Leads from the amplifier were connected to the LFP , EEG , and EMG electrodes . A 10 min baseline recording was acquired first , prior to any injection . When this baseline recording was completed , CNO or saline ( 0 . 9% ) was administered ( I . P . ) . The imaging/recording began immediately after injection . CNO was diluted in saline from a stock of 60 mM each day and a volume was measured for the desired dose ( 0 . 1–1 . 0 mg/kg ) . An equal volume of saline was injected on control days . The sequence of CNO and saline control days was randomized amongst mice . For acute slice experiments , neonatal mice ( C57Bl/6 , P0–4 ) were anesthetized by crushed ice anesthesia for 3 min and injected with AAV5-GFaABC1D . cyto-GCaMP6f at a rate of 2–3 nL/s . Six injection sites ( 0 . 5 μm apart in a 2 × 3 grid pattern , at 0 . 8 μm and 0 . 15–0 μm below the pial surface ) over assumed V1 were chosen . 30 nL/site ( 360 nL total ) was injected with a microsyringe pump ( UMP-3 , World Precision Instruments ) . Coronal , acute V1 slices ( 400 μm thick ) from P25–P30 mice were cut with a vibratome ( VT 1200 , Leica ) in ice-cold cutting solution ( in mM ) : 27 NaHCO3 , 1 . 5 NaH2PO4 , 222 sucrose , 2 . 6 KCl , 2 MgSO4 , 2 CaCl2 . Slices were incubated in standard continuously aerated ( 95% O2/5% CO2 ) artificial cerebrospinal fluid ( ACSF ) containing ( in mM ) : 123 NaCl , 26 NaHCO3 , 1 NaH2PO4 , 10 dextrose , 3 KCl , 2 CaCl2 , 2 MgSO4 , heated to 37°C and removed from water bath immediately before introducing slices . Slices were held in ACSF at room temperature until imaging . Experiments were performed in continuously aerated , standard ACSF . 2P imaging was carried out as for in vivo imaging described above . Experiments began with a 10 min incubation in 1 μM TTX , followed by a 2 min baseline video to record spontaneous activity . To record responses to the wake cocktail ( 20 μM norepinephrine , 5 μM acetylcholine , 5 μM dopamine , 2 . 5 μM histamine ) , a 5 min video was acquired in which the cocktail was added to the bath at the start . The frame at which the cocktail entered the imaging chamber was recorded for each experiment . A second video was then acquired , repeating bath-application of the wake cocktail . To quantify differences in endogenous astrocyte Ca2+ events across behavioral periods , PCA was used to reduce the 20 AQuA outputs to 5 PCs , accounting for 73% of the variance in the original event features . The 5 PC scores for each astrocyte event in each behavioral state across all mice were used to construct empirical cumulative distribution functions ( CDFs ) for each PC in each behavioral state . The CDFs for stationary wake and sleep were compared for each PC using a two-sample Kolmogorov-Smirnov ( K-S ) test . In Gi-DREADD experiments , separate PCA was performed to reduce the 20 output features , as above , to 5 PCs accounting for 69% of the variance . For each PC , and for each mouse , four empirical CDFs were constructed from the PC scores of recorded events , corresponding to each combination of saline or CNO as intervention , and NREM sleep or wake as behavioral state . Within each behavioral state , the K-S distance was computed between the corresponding saline and CNO distributions , to estimate the effect of CNO administration on the given PC’s score distribution during that behavioral period . To assess differences between these effects in different behavioral states for a given PC , the K-S distances between the distributions in saline and CNO trials across all mice were compared between NREM sleep and wake using a Wilcoxon signed-rank test . To quantify functional connectivity between V1 and FC , we computed spectral coherence between the two signals in saline-administration sessions . Coherence spectra were calculated in a 1 . 5 s window following the onset of astrocyte events using Welch’s averaged periodogram method , utilizing a Hann window and a fast Fourier transform size of 1024 samples , as implemented in SciPy . Event-aligned coherence spectra were compared against coherence spectra aligned to randomly chosen time-points . Random time-points were chosen uniformly between the start of each dataset and 1 . 5 s ( the duration of the coherence analysis window ) from the end , in equal numbers to the original astrocyte events for each dataset . To compare between astrocyte- and random-aligned coherence , each mouse’s median event-aligned and random-aligned coherence spectra were calculated; a paired two-tailed t-test was then performed at each frequency between the event-aligned median values and corresponding random-aligned median values for all mice . To control the familywise error rate across compared frequencies , the Bonferroni correction was applied to the resulting p-values . This analysis was extended to quantify coherence changes at sleep-to-wake transitions . Coherence spectra were computed as above in a 1 . 5 s window following astrocyte events during NREM sleep , and following uniformly randomly selected events that coincided with NREM sleep . These spectra were separated into bins based on proximity of the corresponding event onset ( or random alignment point ) to the next sleep-to-wake transition , with each bin encompassing 2 s relative to the transition; the window most proximal to the transition was truncated to 0 . 5 s to avoid overlap of the 1 . 5 s window for coherence computation with the subsequent wake period . Coherence values within 0–15 Hz were averaged in each time bin; averaged coherence values were then compared between the astrocyte- and random-aligned cases in each bin using an unpaired two-tailed t-test . To control the familywise error rate across compared time bins , the Bonferroni correction was applied to the resulting p-values . After physiology experiments were complete , mice were intracardially perfused with 4% PFA . Brains were collected , immersed in 4% PFA overnight at 4°C and switched to 30% sucrose for 2 days before being frozen on dry ice and stored at −80°C . Brains were sliced coronally ( 40 μm thick ) on a cryostat . Slices were stored in cryoprotectant at −20°C until staining . 17–24 slices/mouse were chosen to span from +2 . 8 to −4 . 24 mm from bregma; each slice was 280 μm from the proximate slice . Slices were washed with PBS , 5 min x 3 , then with 0 . 1% PBS-TX for 30 min . Slices were next washed with 10% NGS ( Invitrogen ) for 1 hr , followed by an overnight incubation of 2% NGS , rat α-mCherry ( 1:1000 , ThermoFischer ) , rabbit α-NeuN ( 1:1000 , EMD Millipore ) , and chicken α-GFP ( 1:3000 , Aves Lab ) in 4°C . Slices were next rinsed with 1x PBS x three before incubating for 2 hr at room temperature with goat α-rat Alexa Fluor 555 ( 1:1000 ) , goat α-rabbit 405 ( 1:1000 ) , and goat α-chicken Alexa Fluor 488 ( 1:1000 ) . Slices were washed again with PBS 3x for 5 min before slide-mounting and coverslipping using Fluoromount . Whole coronal slice images were taken using an AxioImager Z2 upright epiflorescent microscope ( Zeiss ) . 5x images were acquired , and z-stacks were stitched together with Zen Software . Images were segmented using WEKA ( Arganda-Carreras et al . , 2017 ) : A classifier was trained to segment 5x images into three classes: ( 1 ) pixels containing fluorescence , ( 2 ) pixels containing non-fluorescent brain tissue , and ( 3 ) pixels containing background . The classifier was then applied to the full dataset , and images were checked manually for accurate segmentation . Each segmented image was manually divided in Fiji to isolate each hemisphere . Quantification of viral spread was calculated in MATLAB by normalizing the number of fluorescent pixels to the number non-fluorescent pixels within the tissue for each hemisphere . To analyze colocalization of mCherry and NeuN at single-cell resolution , 63x images were taken on a spinning disk confocal ( Zeiss ) . Slides were oil-immersed and two slices/animal ( −3 . 8 and −2 . 3 from bregma ) were imaged . In these slices , eight images were taken at random , spanning the total area in which virus was expressed . Colocalization of mCherry and NeuN was performed using Fiji . All statistical tests used , definition of center and dispersion measurements , and exact n values can be found for each figure in the corresponding figure legend . Additional information regarding statistical tests described in the relevant sections . For all figures , significance levels defined as the following: *: p<0 . 05 , **: p<0 . 005 , ***: p<0 . 0005 . | Sleep has many roles , from strengthening new memories to regulating mood and appetite . While we might instinctively think of sleep as a uniform state of reduced brain activity , the reality is more complex . First , over the course of the night , we cycle between a number of different sleep stages , which reflect different levels of sleep depth . Second , the amount of sleep depth is not necessarily even across the brain but can vary between regions . These sleep stages consist of either rapid eye movement ( REM ) sleep or non-REM ( NREM ) sleep . REM sleep is when most dreaming occurs , whereas NREM sleep is particularly important for learning and memory and can vary in duration and depth . During NREM sleep , large groups of neurons synchronize their firing to create rhythmic waves of activity known as slow waves . The more synchronous the activity , the deeper the sleep . Vaidyanathan et al . now show that brain cells called astrocytes help regulate NREM sleep . Astrocytes are not neurons but belong to a group of specialized cells called glia . They are the largest glia cell type in the brain and display an array of proteins on their surfaces called G-protein-coupled receptors ( GPCRs ) . These enable them to sense sleep-wake signals from other parts of the brain and to generate their own signals . In fact , each astrocyte can communicate with thousands of neurons at once . They are therefore well-poised to coordinate brain activity during NREM sleep . Using innovative tools , Vaidyanathan et al . visualized astrocyte activity in mice as the animals woke up or fell asleep . The results showed that astrocytes change their activity just before each sleep–wake transition . They also revealed that astrocytes control both the depth and duration of NREM sleep via two different types of GPCR signals . Increasing one of these signals ( Gi-GPCR ) made the mice sleep more deeply but did not change sleep duration . Decreasing the other ( Gq-GPCR ) made the mice sleep for longer but did not affect sleep depth . Sleep problems affect many people at some point in their lives , and often co-exist with other conditions such as mental health disorders . Understanding how the brain regulates different features of sleep could help us develop better – and perhaps more specific – treatments for sleep disorders . The current study suggests that manipulating GPCRs on astrocytes might increase sleep depth , for example . But before work to test this idea can begin , we must first determine whether findings from sleeping mice also apply to people . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2021 | Cortical astrocytes independently regulate sleep depth and duration via separate GPCR pathways |
The developmental potential of early embryos is mainly dictated by the quality of the oocyte . Here , we explore the utility of the maternal spindle transfer ( MST ) technique as a reproductive approach to enhance oocyte developmental competence . Our proof-of-concept experiments show that replacement of the entire cytoplasm of oocytes from a sensitive mouse strain overcomes massive embryo developmental arrest characteristic of non-manipulated oocytes . Genetic analysis confirmed minimal carryover of mtDNA following MST . Resulting mice showed low heteroplasmy levels in multiple organs at adult age , normal histology and fertility . Mice were followed for five generations ( F5 ) , revealing that heteroplasmy was reduced in F2 mice and was undetectable in the subsequent generations . This pre-clinical model demonstrates the high efficiency and potential of the MST technique , not only to prevent the transmission of mtDNA mutations , but also as a new potential treatment for patients with certain forms of infertility refractory to current clinical strategies .
Infertility disorders are a growing problem that affects millions of couples worldwide ( WHO , 2017 ) . Although assisted reproductive technologies ( ARTs ) have evolved and can now successfully address many challenging cases ( Huang and Rosenwaks , 2014; Niederberger et al . , 2018 ) , conventional IVF treatment continues to fail a significant percentage of infertile women , with many ultimately ending-up being enrolled in egg donation programs ( Lutjen et al . , 1984; Sauer et al . , 1990; Trounson et al . , 1983 ) . The use of donated oocytes is effective at significantly improving the chances of successful IVF treatment , however , the resultant children are not genetically related to the intended-mothers . Therefore , it is desirable to develop new reproductive strategies that can allow the treatment of these patients with genetically related oocytes . Oocyte quality is defined as the competence of the oocyte to develop into a chromosomally normal blastocyst with potential to sustain a pregnancy up to a healthy live birth . Frequently , poor quality oocytes fail to fertilize or produce embryos that arrest during the first stages of development ( Hardy et al . , 2001; Meskhi and Seif , 2006; Pellicer et al . , 1995 ) either due to nuclear or cytoplasmic defects ( Conti and Franciosi , 2018; Eppig , 1996; Liu and Keefe , 2004 ) . Accumulated evidence suggests that aberrant meiosis or early developmental failure is caused mainly by deficiencies in the oocyte cytoplasmic machinery ( Hoffmann et al . , 2012; Liu et al . , 2003; Liu et al . , 1999; Liu et al . , 2000; Liu and Keefe , 2007; Reader et al . , 2017 ) , which contains a vast diversity of critical components , including organelles , mRNAs , proteins , ribosomes and many other factors ( Bianchi et al . , 2015; Sathananthan , 1997 ) . Mitochondria are the most numerous organelles in the cytoplasm and play an essential role by supplying the ATP needed for the oocyte to support critical events , such as: maturation , spindle formation and segregation of chromosomes and chromatids ( Chappel , 2013; May-Panloup et al . , 2007 ) . Dysfunctions at the mitochondrial level and deficiencies affecting other cytoplasmic factors have been correlated with inadequate oocyte developmental competence ( Eichenlaub-Ritter , 2012; Liu et al . , 2002; Van Blerkom , 2011; Van Blerkom et al . , 1995 ) , particularly in older infertile patients ( Babayev and Seli , 2015; Fragouli et al . , 2015; Igarashi et al . , 2016; Wells , 2017 ) . Techniques like cytoplasmic transfer ( Cohen et al . , 1998; Lanzendorf et al . , 1999 ) or the injection of purified mitochondria ( Fakih MHSM et al . , 2015; Kristensen et al . , 2017 have been proposed as potential methods to restore the viability of compromised oocytes in IVF patients with a history of poor embryo development or repeated implantation failures with conventional treatments . Although live births have been reported following the use of these techniques ( Cohen et al . , 1998; Fakih MHSM et al . , 2015; Huang et al . , 1999; Lanzendorf et al . , 1999 ) their safety and/or benefits to treat infertility has been questioned . Cytoplasmic transfer experiments were abandoned due to concerns that heteroplasmy ( i . e . , the co-existence of two distinct mtDNA genomes ) might have negative clinical consequences ( Darbandi et al . , 2017; Isasi et al . , 2016; Kristensen et al . , 2017 ) . An alternative strategy , which avoided heteroplasmy by utilizing autologous injection of mitochondria from the patient's own germline cells attracted much attention as a possible new treatment to revitalize deficient oocytes ( Johnson et al . , 2004; White et al . , 2012 ) . Multiple studies in animal models showed apparent benefits of the addition of mitochondria to oocytes of compromised quality ( El Shourbagy et al . , 2006; Hua et al . , 2007; Yi et al . , 2007 ) and IVF births were reported after transfer of oogonial precursor cell-derived mitochondria ( Fakih MHSM et al . , 2015 ) . However , the source and quality of the mitochondria used are unclear and a recent randomized clinical study conducted using mitochondria derived from autologous oogonial stem cells failed to demonstrate improvements in embryo developmental or clinical outcomes ( Labarta et al . , 2019 ) . Thus , current data from human clinical research do not support the notion that the addition of further mitochondria derived from the same individual is capable of correcting cytoplasmic deficiencies ( mitochondria or other ) that may be present in poor quality oocytes . Furthermore , the safety of the procedure is yet to be verified . Of note , a recent study suggested that autologous mitochondrial supplementation may induce a phenotypic effect in the heart of resultant mice ( St John et al . , 2019 ) . An approach that may offer greater promise in terms of its capacity to address infertility problems of maternal ( oocyte ) origin is the transfer of the nuclear genome from an affected oocyte or zygote into a new ‘healthy’ cytoplasm . These techniques , known globally as mitochondrial replacement techniques ( MRTs ) were originally proposed to prevent the transmission of inherited mitochondrial diseases ( Craven et al . , 2010; Hyslop et al . , 2016; Paull et al . , 2013; Tachibana et al . , 2009 ) . Indeed , a clinical application of maternal spindle transfer ( MST ) to prevent the transmission of Leigh Syndrome was recently reported , resulting the birth of an unaffected child ( Zhang et al . , 2017 ) . However , the potential of MRTs to overcome infertility remains unclear , as most studies utilizing this approach have not had this as their main focus , instead concentrating on their potential to avoid mitochondrial diseases; examination of nuclear-cytoplasmic interactions in oocytes and zygotes ( Liu and Keefe , 2004; Liu and Keefe , 2007 ) ; the origin of female aneuploidies Palermo et al . , 2002; or the decreased developmental capability of aged oocytes in animal models ( Yamada and Egli , 2017 ) . Here , we explored the feasibility of the MST technique as a reproductive tool to overcome embryo developmental arrest . To test our hypothesis , a detailed series of proof-of-concept experiments were conducted to assess the safety and the efficiency of the technique using mouse models , which , in a clinical context , could represent donors and patients with oocytes of good and poor developmental competence , respectively . Additionally , advanced molecular techniques were used to evaluate in detail the heteroplasmy levels induced by the procedure in early embryonic-stages and in multiple important organs , including some with high metabolic demand , collected from male and female mice generated by MST . The mice were bred and followed up to ascertain their health , fertility and welfare , as well as , to study the fate of the heteroplasmy in the offspring of the MST female progenitors over five generations .
In a first set of experiments we aimed to optimize the MST protocol and to determine whether the manipulation of the spindle-chromosome complex is feasible without impairing the developmental potential of reconstructed oocytes . We performed reciprocal MST among sibling oocytes from the mouse hybrid B6CBAF1 strain ( Figure 1a ) . Enucleation and reconstruction ( karyoplast-cytoplast fusion ) of oocytes were first assessed with freshly collected oocytes . Enucleation was successful in 98 . 9% of oocytes ( n = 790 ) and reconstruction was achieved in 96 . 1% ( n = 321 ) , confirmed using a microscope with polarized light that allows visualization of the birefringence of the spindle microtubules ( Figure 1b–c and Figure 1—figure supplement 1 ) . Next , MST was carried out with both fresh and cryopreserved B6CBAF1 oocytes that were vitrified and warmed using the open Cryotop system ( 97 . 7% survival , n = 600 ) . In this set of experiments , spindles were taken from fresh oocytes and transferred into either fresh ( fresh-sp/fresh-cyt ) or vitrified-warmed cytoplasts ( fresh-sp/vitrified-cyt ) and vice-versa , that is spindles from vitrified oocytes transferred to fresh ( vitrified-sp/fresh-cyt ) or vitrified-warm cytoplasts ( vitrified-sp/vitrified-cyt ) . The resultant oocytes from the different groups were then fixed after reconstruction and processed for evaluation of the spindle apparatus and chromosomes distribution by immunofluorescence microscopy ( Figure 1e–f and Figure 1—figure supplement 2 ) . All oocytes analyzed presented a spindle with a normal barrel shape and with the chromosomes aligned at the MII plate ( fresh-sp/fresh-cyt n = 20 , fresh-sp/vitrified-cyt n = 15 , vitrified-sp/fresh-cyt n = 15 , vitrified-sp/vitrified-cyt n = 16; Figure 1e–f ) , regardless of whether fresh or vitrified gametes were used as spindle or cytoplast donors ( Figure 1—figure supplement 2 ) . These observations indicated that the conditions used to perform the manipulation of the spindle-chromosome complex were neither damaging to its structure nor altering of the distribution of the chromosomes . Furthermore , there was no evidence that the procedure was inducing premature activation of the oocytes . Subsequently , in an independent set of samples , we compared the in vitro development of reciprocal MST experiments using fresh and vitrified B6CBAF1 oocytes , after insemination by ICSI ( Figure 1j and Figure 1—figure supplement 1 ) . High enucleation ( 98 . 7% , n = 399 ) and fusion ( 98 . 2% , n = 394 ) rates were achieved in all MST groups ( see also Table 1 ) and almost all oocytes that were prepared with fresh ( 99% , n = 100 ) or vitrified ( 100% , n = 90 ) spindles , and transferred into fresh cytoplasts , developed to the two-cell stage on the next morning ( Figure 1h–j and Table 1 ) . Interestingly , a significantly lower proportion of inseminated oocytes composed of vitrified spindles transferred into vitrified cytoplasts ( vitrified-sp/vitrified-cyt ) developed to the two-cell stage ( 82 . 4% , n = 85 ) compared with non-manipulated fresh ( 96 . 8% , n = 94 , p=0 . 001 ) or vitrified ( 96 . 7% , n = 90 , p=0 . 001 ) controls . Poorer development was also observed for the fresh-sp/vitrified-cyt group ( 81 . 1% , n = 90 ) ( Figure 1j and Table 1 ) . On the contrary , when spindles from vitrified oocytes were transferred into fresh cytoplasts ( vitrified-sp/fresh-cyt , n = 90 ) , two-cell stage ( 100% ) and blastocyst formation ( 85 . 6% ) rates were high and equivalent to fresh controls ( 96 . 8% and 84 . 1% , respectively ) or to MST oocytes where fresh spindles were transferred into fresh cytoplasts ( fresh-st/fresh-cyt , n = 100 , 99% and 81% , respectively ) ( Figure 1j and Table 1 ) . Additionally , the mean number of total cells ( mean ± SD , n ) in the blastocysts obtained in the fresh-st/fresh-cyt group ( 177 . 8 ± 26 . 7 , n = 81 ) was equivalent to controls ( 192 ± 29 . 5; n = 79 ) . No differences were found either in the number of inner cell mass cells that were positive for the Oct4 pluripotency marker between fresh-st/fresh-cyt and control groups ( 22 . 4 ± 3 . 5; n = 14 versus 25 . 3 ± 5 . 6; n = 10 , see also Figure 1i ) . Taken together , the experiments performed among sibling B6CBAF1 oocytes , showed that MST is technically feasible in the mouse without impacting the in vitro developmental competence of the oocyte . Experiments indicate that vitrification induces changes that make cryopreserved oocytes unsuitable for use as cytoplasts . However , the spindle apparatus does not appear to be damaged during vitrification or MST procedures . When recipient cytoplasts were derived from fresh oocytes , blastocyst development rates were equivalent to those obtained for non-manipulated controls , regardless of whether the spindle originated from a fresh or vitrified oocyte . After careful optimization and validation of the different steps of the MST protocol , the effectiveness of the technique as a strategy to overcome embryo developmental arrest was evaluated . Two different oocyte strains were employed: the hybrid B6CBF1 ( resultant from the cross between C57BL/6JRj females and CBA/Jrj males ) , and the New Zealand Black ( NZB/OlaHsd ) strains . The NZB strain holds two interesting characteristics . Firstly , NZB mice present a poor reproductive performance ( Fernandes et al . , 1973; Hansen CT and Whitney , 1973 ) and , secondly , the genetic background of the NZB strain has diverged genetically from most other mouse laboratory strains , including the hybrid B6CBAF1 strain , accompanied by characteristic differences in mtDNA sequences ( Bielschowsky and Goodall , 1970 ) . These two features are particularly relevant to the experimental design of this study as , in a clinical context , the NZB strain could be considered analogous to a subfertile patient ( especially those with a history of poor in vitro embryo development ) , and the B6CBAF1 strain , a donor of proven fertility . Additionally , single nucleotide polymorphisms in the divergent mtDNA of the NZB strain provides an opportunity to evaluate the carryover of organelles and resultant heteroplasmy induced by MST procedures ( see Materials and methods ) . Experiments were thus carried out between the two mouse strains , so that meiotic spindles were transferred from fresh NZB oocytes into fresh B6CBAF1 cytoplasts and vice-versa ( Figure 2a ) . Once reconstructed , oocytes were inseminated using ICSI in parallel with non-manipulated oocytes from both strains and cultured in vitro until the blastocyst stage ( Figure 2a ) . Enucleation and fusion rates were identical in both MST groups , and no differences were found in terms of survival to ICSI compared to controls ( Figure 2b and Table 2 ) . As expected , NZB control oocytes presented significantly lower fertilization rates than B6CBAF1 control oocytes , measured as two-cell stage development ( Figure 2b and Table 2 ) . Additionally , while blastocyst formation rates were close to 80% in the B6CBAF1 control group ( 77 . 8% , n = 144 ) , most of the injected oocytes from the NZB control group arrested their development before reaching this stage ( 5 . 6% developed into blastocysts , n = 159 , Figure 2b and Table 2 ) . Remarkably , when the meiotic spindles from NZB oocytes were transferred into B6CBAF1 cytoplasts ( NZB-sp/B6-cyt ) , the blastocyst formation rates were 10-fold higher ( 51 . 4% , n = 212 , p<0 . 0001 ) compared to the non-manipulated NZB control ( Figure 2b and Table 2 ) . In the reciprocal MST group , B6CBAF1 spindles transferred into NZB cytoplasts ( B6-sp/NZB-cyt ) , blastocysts were not obtained ( 0% , n = 110 ) , indicating that cytoplasmic factors are likely to be responsible for the lower fertilisation and massive developmental arrest observed at preimplantation stages in the NZB strain ( Figure 2b , c and Table 2 ) . At 96 hr post-insemination , embryos produced in the different experimental groups were vitrified and their competence to develop in vivo determined when synchronized pseudo-pregnant females were available for transfer . A total of 65 MST blastocysts from the NZB-sp/B6-cyt MST group were then warmed ( 100% survival ) and transferred non-surgically into six recipients , which resulted in 14 live pups ( 21 . 5% ) ( Figure 2d , e and Table 3 ) . This birth rate is comparable ( p>0 . 05 ) with results obtained from the B6CBAF1 control group ( 15 live pups ( 25 . 9% ) out of 58 blastocysts transferred into five recipients ) . Consistent with expectations , only six pups developed to term from 44 morulas/blastocysts ( 13 . 6% ) transferred into five recipients from the control NZB group . All living pups were born healthy and respired normally . Caesarean sections at 18 . 5 dpc were performed in two recipients of each group to evaluate the size and weight of the placentas and the corresponding pups , with no significant differences found between groups ( Table 4 ) . These results suggest that MST procedures do not typically induce an overgrowth phenotype of the type described for certain other techniques , such as somatic cell nuclear transfer ( Costa-Borges et al . , 2010 ) . Overall , these experiments confirmed that MST , with cytoplast donation from a distantly related mouse strain , is highly effective at overcoming the in vitro developmental arrest phenotype of NZB mice and that the resultant embryos are competent to develop to term with high efficiency . The extent of mtDNA carryover induced by MST was evaluated in embryos at different preimplantation developmental stages . Spindles from NZB oocytes were transferred into B6CBAF1 cytoplasts and the resultant MST oocytes were fertilized by ICSI and cultured in vitro ( Figure 3a ) . Afterwards , biopsies were performed to remove second polar bodies from embryos at the two-cell stage , single cells ( blastomeres ) from 6 to 8 cell stage embryos , or to excise a cluster of 4–8 trophectoderm cells from blastocysts ( Figure 3—figure supplement 1 ) . The biopsies and their corresponding embryos were then analyzed individually to ascertain whether mtDNA heteroplasmy levels in the biopsied cells are representative of the values found in the complementary embryo ( Figure 3a ) . To determine mtDNA carryover , a high-throughput sequencing protocol was developed based upon quantification of a single nucleotide polymorphism ( SNP ) in mtDNA using Ion PGM sequencer ( ThermoFisher , see Materials and methods for further details ) . The SNP utilized for this purpose is located at position m . 3932 and exists as a guanine ( G ) in the B6CBAF1 strain and an adenine ( A ) in the NZB strain . The presence of different alleles at m . 3932 was confirmed by minisequencing analysis using genomic DNA ( gDNA ) from tail tips of B6CBAF1 and NZB mice ( Figure 3—figure supplement 2 ) . This sequencing protocol was carefully validated . Initially , protocol accuracy and sensitivity was assessed by analyzing different ratios of G and A alleles in artificially constructed samples , composed of gDNA from both mouse strains mixed in different ratios . For the purpose of these experiments , the G base ( derived from B6CBAF1 ) was considered the reference allele and the A base ( from NZB ) the variant allele ( Figure 3—figure supplement 2 and Supplementary file 1 ) . To verify validity of mtDNA carryover assessment by analysis of a single SNP and to ensure reliability of the utilized sequencing platform , four additional SNPs ( B6CBAF1/NZB: m . 2798C/T; m . 2814T/C; m . 3194T/C; m . 3260A/G ) were analyzed on a different sequencer ( Illumina’s MiSeq ) . The presence of different alleles was also confirmed by minisequencing ( see Materials and methods and Supplementary file 2 for further details; and Figure 3—figure supplement 2 ) . Analysis of mtDNA carryover after MST in biopsied cells and the complementary embryos ( Figure 3a ) , revealed that the mean variant ( NZB ) allele frequencies obtained from polar bodies were significantly higher compared to the mean frequencies in the complementary two-cell-stage embryos ( 6 . 2 ± 6 . 2% SD versus 0 . 5 ± 0 . 8% SD; p=0 . 0095 ) ( Figure 3b ) . By contrast , there was no significant difference in mtDNA allele frequencies between biopsied blastomeres and trophectoderm samples when compared to the corresponding embryos ( cleavage-stage: 1 . 3 ± 1 . 0% SD versus 1 . 9 ± 0 . 6% SD , respectively; blastocyst stage: 1 . 7 ± 0 . 9% SD versus 1 . 9 ± 0 . 6% SD , respectively ) . Moreover , the mean heteroplasmy levels were similar between all embryonic samples ( except polar bodies ) ( Figure 3b and Supplementary file 3 ) . These experiments demonstrate that cleavage stage or blastocyst biopsy are preferable over biopsy of second polar bodies as methods for determining the mtDNA carryover levels found in preimplantation embryos . The results also suggest that while some mitochondria remain associated with the meiotic spindle , and are unavoidably transferred to the recipient cytoplast , the vast majority of these organelles do not persist into later developmental stages , with most being expelled into the second polar body at the completion of meiosis II . To ascertain the long-term health status and fertility of the mice generated by MST , follow up studies were then conducted over five generations . Ten mice ( three females and seven males ) generated by MST were selected for mating with wild type ( WT ) mice . At 21 days after birth , the resultant offspring were weaned , and the size and gender ratio of the litters were assessed . All parental MST mice ( F1 ) were fertile and produced a total of 78 pups , with a mean litter size of 7 . 8 ± 1 . 4 pups/animal and no significant deviations in the expected male-female ratio ( 59% and 41% respectively , Supplementary file 4 ) . All pups ( F2 ) were born alive , respired normally and grew to adulthood without manifesting any physiological or behavioral alteration . The fertility of these mice was assessed for a total of 5 generations , by selecting random males and females from litters ( n = 9 in F2 and n = 4 between F3 and F5 ) . Similarly , these mice also displayed normal fertility and produced viable offspring , without alterations in the expected gender ratio ( Supplementary file 4 ) . Gross necropsies of the parents and offspring were performed during the five generations , with no pathological findings observed . In the 239 mice analyzed , all organs showed a normal size , texture and morphological appearance . Additionally , F1 mice generated by MST B6-sp/B6-cyt ( n = 3 ) , MST NZB-sp/B6-cyt ( n = 5 ) and control B6 ( n = 4 ) groups were also processed for histopathological examinations , which were performed in vital organs including heart , kidney , liver and brain , as well as , in tibial and quadriceps skeletal muscle and urinary bladder smooth muscle . Reproductive systems and accessory glands of both males ( testis , epididymis , seminal vesicles , prostate , coagulating glands , ampullary glands and bulbourethral glands ) and females ( ovaries , oviducts , uterine horns ) were also assessed . Except for a pericardium focal inflammation in one animal of the B6 control group , none of the animals showed any lesions or visible abnormalities ( Figure 4—figure supplements 1 and 2 ) . Taken together , these results support the notion that MST can efficiently produce viable and fertile offspring . A source of great concern in MRTs field has been the reversion of mtDNA heteroplasmy observed in embryonic stem cells ( ESCs ) derived from pronuclear transfer or MST generated embryos ( Hyslop et al . , 2016; Kang et al . , 2016; Paull et al . , 2013 ) . To evaluate whether heteroplasmy was transmitted through generations and whether homoplasmy was restored , the ratios of the mtDNA alleles attributable to B6CBAF1 and NZB were assessed through several generations . Multiple organs were assessed , including those with different metabolic demands: brain; heart; liver; kidneys ( Jenuth et al . , 1997; Sharpley et al . , 2012 ) . A total of six mice ( four male and two female ) from F1 were sacrificed at adult age ( 12 weeks old ) . The mean heteroplasmy level in this group of mice was low at 2 . 3 ± 1 . 3% ( mean ± SD , n = 6 ) ranging from mean frequencies of undetectable values to 3 . 5% in individual mice ( Figure 4a , Supplementary file 5 ) . Moreover , heteroplasmy levels were similar among different tissue types from the same mouse ( Figure 4b ) and showed no differences between males and females . Finally , the fate of the heteroplasmy was examined in adult mice derived from the MST female lineage . Four mice ( two males and two females ) were selected at random from each litter , through five generations . Mitochondrial DNA heteroplasmy levels were reduced to 0 . 4 ± 0 . 6% ( mean ± SD , n = 4 ) on average in F2 mice ( Figure 4c and Supplementary file 5 ) and decreased to undetected levels in subsequent generations ( F3 to F5 , Supplementary file 5 ) . These quantifications based on a single SNP in an Ion PGM sequencer were corroborated by using an additional sequencing platform ( Illumina’s MiSeq ) and 5 SNPs , as described above . Artificially constructed samples , composed of gDNA from both mouse strains mixed in different ratios , and gDNA from 5 organs of selected adult mice from F1-3 generations were analyzed ( Figure 3—figure supplement 2 , Figure 4—figure supplement 3 , Supplementary files 2 and 6 ) . These results suggest that low levels of mtDNA heteroplasmy resultant from MST typically result in a homoplasmic state in offspring within a few generations , without reversion ( Supplementary files 5 and 6 ) . However , it is acknowledged that different mtDNA haplogroups or mtDNA genomes affected by specific mutations might have differences in the efficiency with which they replicate , influencing the speed at which homoplasmy is attained as well as the risk if reversion .
MST is a technique that was originally proposed to prevent the transmission of mitochondrial diseases . This proof of concept study provides insights into the feasibility of this technique as a potential new reproductive approach to overcome infertility problems characterized by repeated in vitro embryo development arrest caused by cytoplasmic deficiencies in the oocyte . Herein , it is shown that MST can be carried out with high efficiency in the mouse , with successful enucleation and reconstruction achieved for >95% of oocytes . Furthermore , the data produced indicate that , as long as all the steps of the protocol are well optimized and care is taken to minimize the risk of damage to the oocyte , the procedure does not negatively affect the spindle apparatus or early embryo development . In the event of a future clinical application of MST in humans , it may be difficult to coordinate the retrieval of mature oocytes from patients and donors , due to the inherent variation in ovarian responses to hormonal stimulation . For this reason , the capacity of cryopreserved oocytes to substitute for fresh oocytes , when serving as spindle or cytoplast donors , was evaluated . The results indicated that fresh and vitrified oocytes are equally suitable for use as spindle donors , but superior results are obtained if the recipient cytoplast is fresh . This agrees with a previous report performed in non-human primates that had shown that fresh spindles transplanted into vitrified cytoplasts results in impaired ( 50% ) fertilization after ICSI , while the reciprocal spindle transfer resulted in fertilization ( 88% ) and blastocyst formation ( 68% ) rates similar to fresh controls ( Tachibana et al . , 2009 ) . This also represents an advantage in the clinical setting , where low-responders to ovarian stimulation could vitrify oocytes from repeated oocyte collections , and the accumulated oocytes be used for MST using freshly collected donor cytoplasts . Additionally , MST was conducted between two distantly related mouse strains with the aim of simulating a clinical context , in which donors with oocytes of good reproductive competence provide cytoplasts for patients with a history of poor oocyte fertilization and/or high rates of failed embryo development . The experiments demonstrated how the successful replacement of the entire cytoplasm of compromised oocytes has the potential to overcome the massive embryo development arrest phenotype , which is observed in non-manipulated controls from a sensitive mouse strain ( NZB ) . This strategy resulted in a highly significant ( 10-fold ) increase in blastocyst formation rates , as well as an increased likelihood of embryo development to term , compared to non-manipulated control oocytes . These results highlight the importance of the cytoplasm on the potential of the oocyte to support embryo development in vitro and to lay the foundations for a successful pregnancy . Consistent with this data , Mitsui and colleagues showed that oocyte genomes from mice aged 10–12 months transferred into oocytes of young mice aged 3–5 months , resulted in increased term-development from 6 . 3% for in vivo aged oocytes to 27 . 1% for the reconstructed oocytes ( Mitsui et al . , 2009 ) . Similarly , a recent study demonstrated that in vitro aged oocytes accumulate cytoplasmic deficiencies if they are maintained in culture for an extended period prior to fertilization , and that these deficiencies can be overcome with spindle transfer ( Yamada and Egli , 2017 ) . However , in both cases , studies were performed between oocytes from the same mouse strain and thus the potential of the technique to overcome infertility in a strain with poor fertility competence remained undetermined . It is noteworthy that the current study employed ICSI to inseminate oocytes , rather than conventional IVF , which resembles closely the standard protocol used in humans for oocytes that have been denuded of the surrounding cumulus cells . Levels of mtDNA heteroplasmy caused by carryover of mitochondria in close proximity to spindle during the MST procedure were also evaluated . Clearly , this is an important consideration when utilizing MST technology to avoid transmission of mtDNA mutations responsible for serious inherited disorders , but it is also relevant to other variations in the mtDNA , or other defects affecting the mitochondrial organelle , which may potentially contribute to certain forms of embryonic developmental arrest . As well as assessing the extent of heteroplasmy at different embryonic stages , the levels were also assessed in multiple tissues in adulthood and over several generations . There are conflicting reports regarding the dynamics of mtDNA heteroplasmy during the lifetime of an individual , between organs or even during in vitro culture when ESCs have been derived from MRT embryos with heteroplasmic mtDNA ( Hyslop et al . , 2016; Kang et al . , 2016; Paull et al . , 2013 ) . It is also unclear to what extent divergent mtDNA haplotypes in heteroplasmic organisms might lead to functional incompatibility , either between the two types of mitochondria or between the mitochondrial and nuclear genomes . This study confirms that cells biopsied from MST embryos at the morula or blastocyst stages present minimal levels of heteroplasmy ( <2 . 9% mtDNA from the spindle donor ) and that these biopsy specimens are representative of the remainder of the embryo . On the contrary , the heteroplasmy levels were significantly higher in second polar bodies than in blastomere or trophectoderm biopsies . This agrees with data from Neupane and colleagues , who have shown that , in comparison to second polar bodies , mtDNA heteroplasmy in TE cells is more closely correlated with the levels in the blastocyst as a whole or the corresponding ESCs ( Neupane et al . , 2014 ) . Alternatively , oocytes might be actively removing mitochondria transferred along with the spindle , since they may be disadvantageous as compared to the recipient’s own organelles ( De Fanti et al . , 2017 ) . However , perhaps the most likely explanation is that when the meiotic spindle is transferred , a number of mitochondria accompany it . These mitochondria are likely to remain in the vicinity of the MII spindle and consequently it is inevitable that a disproportionate number of these mitochondria will pass into the second polar body . Regardless of the underlying mechanism , our data suggest that testing of blastomeres or TE biopsies is preferable to second polar body analysis for the quantification of mtDNA heteroplasmy levels . This also has relevance for the preimplantation genetic testing ( PGT , also known as preimplantation genetic diagnosis – PGD ) of mitochondrial disease in the human . The data from our current study revealed that mtDNA heteroplasmy levels were low in all the adult mice produced , regardless of gender , or the type of organ ( range 0–6% ) . Previous studies in monkeys and humans have shown that a minimal number of donor mitochondria are transferred using MRT ( below 1–2% ) ( Craven et al . , 2010; Hyslop et al . , 2016; Paull et al . , 2013; Tachibana et al . , 2009 ) . Nevertheless , since the meiotic spindle in mouse oocytes is much larger than that of the human , and given that multiple mitochondria are found in the vicinity of the spindle , it was expected that MST in mice would lead to higher mtDNA carryover levels . The surprisingly low heteroplasmic levels achieved during this study can likely be attributed to the use of birefringence microscopy during enucleation , which assists in minimizing the carryover of cytoplasm transferred along the meiotic spindle . It has also been suggested that organs with a high-metabolic demand tend to accumulate higher heteroplasmy mtDNA levels ( Jenuth et al . , 1997; Meirelles and Smith , 1997 ) , however , our data do not confirm this observation . This result could be explained by differences in mitochondrial haplotypes from mouse strains used , which could have a differential replication rate . Some studies of heteroplasmic ESC lines derived from embryos carrying mtDNA mutations have shown changes in the levels of normal and mutant mtDNA during prolonged in vitro culture , with reversion back to a situation where mutant mtDNA predominates ( Hyslop et al . , 2016; Kang et al . , 2016; Paull et al . , 2013 ) . This delayed efforts for the direct application of MRT-derived techniques in the clinical setting and raises some concerns for the first baby born using MST ( Zhang et al . , 2017 ) . Nevertheless , the results presented here show a low level of mtDNA carryover in all adult organs analyzed , suggesting that the mechanism seen in ESCs in vitro might not necessarily represent the in vivo process . The results also agree with Sharpley et al , who showed that NZB and 129S6 mtDNA heteroplasmic haplotypes decrease over generations ( Sharpley et al . , 2012 ) . In the current study , heteroplasmy was very low in the F2 progeny and undetected in the offspring of the subsequent generations ( up to F5 ) . The data collected from the analyzed organs suggests that heteroplasmy resultant from MST can be stable within an individual and can lead to an homoplasmic state within a few generations . However , additional work should be done in order to comprehensively assess how mtDNA heteroplasmy segregates in other organs . On the other hand , the MST mice followed over five generations were apparently normal and showed good fertility ( average of 7 . 8 pups per litter ) . This is a notable observation , as based on the literature , NZB/OlaHsd mice are expected to have small litter sizes ( 3 . 8 at weaning ) ( Fernandes et al . , 1973; Hansen CT and Whitney , 1973 ) . Additionally , histological examinations in F1 MST mice did not reveal any lesions in a selection of organs . Whether the MST technique can potentially reveal mitochondrial causes of infertility that are hereditary or aggravated with lifestyle or age is a question that remains to be answered and will require additional studies . In conclusion , this study has demonstrated that MST can overcome a severe developmental arrest phenotype , associated with poor fertility and greatly reduced chances of an individual oocyte producing a pregnancy following in vitro fertilization . The results show that embryos produced using optimized MST techniques can give rise to apparently normal and fertile animals . Levels of heteroplasmy were low in the initial generation and undetectable in subsequent generations , indicating that homoplasmy for the mtDNA of the cytoplast donor is rapidly attained in this model . Given the high proportion of IVF cycles which are unsuccessful due to poor embryo development related to low oocyte quality , we believe that there is a need to further explore the potential of MST as a clinical treatment for infertility . Pre-clinical and clinical trials involving human oocytes , undertaken in a regulated and carefully controlled manner , is desirable , since such a therapy could represent the last chance for infertile patients to have genetically related children .
Animal care and procedures were conducted according to protocols approved by the Ethics Committee on Animal Research ( DAMM-7436 ) of the Parc Cientific of Barcelona ( PCB ) , Spain . Hybrid ( B6/CBA ) and outbred CD1 females of 5–6 weeks of age ( 25–30 g ) , and male mice from the same genetic strains of 8–10 weeks of age ( 25–30 g ) were purchased from Janvier Laboratories ( France ) . New Zealand Black ( NZB/OlaHsd ) mice were purchased from Envigo ( France ) . Upon arrival , all mice were quarantined and acclimated to the PCB Animals´ facility ( PRAL ) for approximately 1 week prior to use . Three to four mice were housed per cage in a room with a 12 hr light/dark cycle with ad libitum access to food and water . For the collection of oocytes , hybrid B6CBAF1 and NZB females were induced to superovulate by intraperitoneal injection of 5 IU of pregnant mare serum gonadotropin ( PMSG ) followed 48 hr later by 5 IU of human chorionic gonadotropin ( hCG ) . Cumulus–oocyte complexes from the both strains were released from the oviducts by 14–15 hr after hCG administration and treated with hyaluronidase ( LifeGlobal ) until cumulus cells dispersed . Once denuded , oocytes with good morphology were washed several times and kept in culture medium ( Global total , LifeGlobal ) under oil ( Lifeguard , LifeGlobal ) at 37 . 3°C , in an atmosphere with 7%CO2% and 7%O2 in air , until use . Sperms were collected from cauda epididymis and then diluted and incubated in medium supplemented with glucose ( Global total for fertilization , LifeGlobal ) at 37 . 3°C , in an atmosphere with 7%CO2% and 7%O2 in air , until use . Oocytes from B6CBAF1 or NZB strains were used as spindle chromosome-complex and cytoplasts donors . Procedures were performed using a piezo-driven ( PiezoXpert , Eppendorf ) micromanipulator . Oocytes first were exposed to small drops of hepes-buffered medium ( Global total w/hepes , LifeGlobal ) containing 5 µg/mL cytochalasin B ( Sigma ) covered with mineral oil for 3–5 min at 37°C . Afterwards , the meiotic spindle was aspirated into an enucleation pipette ( Humagen ) trying to remove the minimum amount of surrounding cytoplasm possible , and enucleation confirmed using a microtubule birefringence system ( PolarAide , Vitrolife ) to visualize the spindle apparatus ( Figure 1—figure supplement 1 ) . If the karyoplast removed contained a larger amount of cytoplasm , the extra cytoplasm was eliminated by pressing the cytoplasm against the zona pellucida . Karyoplasts were inserted below the zona pellucida of another enucleated oocyte ( cytoplast ) and fused using inactivated Sendai virus HVJ-E ( GenomeOne , Cosmo Bio ) . All manipulations were performed on a 37°C heated stage ( Okolab ) of an Olympus IX73 inverted microscope , using Eppendorf micromanipulators . Non-manipulated control oocytes and those generated by MST were inseminated using a modified piezo-actuated ICSI technique , known as the ‘hole removal technique’ . Briefly , this procedure is based on withdrawing the ICSI pipette and applying rapid suction simultaneously just after the sperm head has been injected to seal the oocyte membrane , which increases survival chances . The injected oocytes were then cultured in Global total medium ( LifeGlobal ) under oil at 37 . 3°C , in K-Minc incubators ( Cook Medical ) , in an atmosphere with 7% CO2 and O2 in air . Embryos generated by MST were biopsied at different developmental stages , including: two-cell , morula or blastocyst stage . Regardless of the developmental stage , biopsies were performed in individual 5 µL droplets of Global total w/hepes medium covered with oil using a biopsy pipette with 19 µm of internal diameter ( Eppendorf ) with the assistance of laser shots to open a hole in the zona pellucida or to weaken the trophectoderm cells in the case of the blastocyst biopsy . After biopsy , both the biopsied cells and the complementary embryo were transferred individually to empty PCR tubes and stored at −80°C until processed for mtDNA allele frequencies determination . For analysis of the spindle structure and chromosomes distribution , control and MST oocytes were fixed and extracted for 30 min at 37°C in a microtubule stabilizing buffer ( MTSB-XF ) . A triple-labeling protocol was then used for the detection of microtubules , microfilaments and chromatin by fluorescence microscopy , as described previously ( Messinger and Albertini , 1991 ) . Briefly , fixed oocytes were first incubated in a mixture of mouse monoclonal anti α/β-tubulin antibodies , and then in a mixture of secondary antibody ( chicken anti-mouse IgG ) conjugated to Alexa Fluor 488 and of Alexa Fluor 594 phalloidin . Finally , all oocytes were washed in PBS blocking solution , incubated in Hoechst 33258 , and put on a mounting solution droplet on a glass slide . Blastocysts processed for total cell counts were fixed in 4% PFA and permeabilized in 2 . 5% Triton-X100 for 25 min at room temperature . Afterwards , blastocysts were incubated overnight in blocking solution and then in rabbit monoclonal anti-Oct-4 , washed 3 times in PBS blocking solution for 10 min at 37°C . After , they were incubated in secondary antibody ( goat anti rabbit IgG ) conjugated with Alexa Fluor 594 , washed and incubated in Hoechst ( 10 µg/ml ) for 10 min at room temperature and finally mounting solution droplet on a glass slide . Stained oocytes or blastocysts were examined using an epifluorescence microscope ( Nikon E1000 ) fitted with specific filters for Hoechst , Fluorescein and Texas Red and a 50W mercury lamp . Digital images were acquired with E1000 Nikon software . Oocytes and blastocysts were vitrified following the instructions provided by the manufacturer ( Kitazato BioPharma Japan ) . Briefly , samples were exposed to equilibration solution ( ES ) for 15 min , transferred to VS1 for 30 s and then to VS2 for additional 30 s . Afterwards , they were loaded onto the surface strip of a classic Cryotop ( Kitazato BioPharma Japan ) and directly plunged into liquid N2 . For warming , the Cryotop strip was transferred from the liquid nitrogen into a TS solution for 1 min at 37°C and then gradually moved to dilution solution ( DS ) for 3 min , to washing solution ( WS ) 1 for 5 min and , finally , to WS2 for an additional 1 min . Exposures to DS and WS solutions were performed at room temperature . After warming , samples were extensively washed and kept in culture medium under oil at 37 . 3°C , in an atmosphere with 7% CO2 and O2 in air . Embryo transfers were performed non-surgically using a commercial non-surgical embryo transfer protocol ( NSET , Paratechs ) . Briefly , an NSET device was coupled to a P2 pipette with volume adjusted to 1 . 8 μl . Between 8 and 12 blastocysts were loaded in each device within a culture medium droplet under a stereomicroscope . After loading the blastocysts , the volume in the P2 pipette was re-adjusted to 2 μl to create an air bubble and to avoid the loss of the embryos by capillarity . The recipient female assigned for transfer was then immobilized , and a NSET small speculum was carefully introduced in the vagina . With the animal still immobilized , the NSET device loaded with the embryos was introduced by the speculum through the cervix . When the base of the device got in contact with the speculum , the blastocysts were transferred by pressing the plunger of the pipette . Having the plunger of the pipette still pressed , NSET device was removed and checked under the stereomicroscope to confirm that all embryos had been correctly transferred . Finally , the speculum was removed and the female returned to its corresponding cage . In the majority of transferred females natural delivery was controlled at the day 20 of pregnancy ( P20 ) , while in a few cases , cesarean sections were performed on embryonic day 18 . 5 to collect information on the weight and size of the placentas and pups . Pups ( F1 ) resultant from the embryo transfer procedures were checked for health status and grown up until sexual maturity age was reached . Having reached the adult age , F1 males and females from each experimental group were randomly selected for crossing with wild-type ( WT ) B6CBAF1 mice , so that their health status and fertility competency could be assessed . At day 21 after birth , the offspring of the F1xWT = F2 mice were weaned and the F2 animals were checked and sexed . The same strategy was repeated for a total of 5 generations , by selecting random males and females from litters ( n = 9 in F2 and n = 4 between F3 and F5 ) . For histological evaluation , tissue samples from 4 ICSI-B6 control , 3 B6-sp/B6-cyt MST and 5 NZB-sp/B6-cyt MST mice were collected at 6 weeks of age . Mice were perfused with PBS and 5% formaldehyde solution . Subsequently , tissues were fixed overnight at 4°C in 5% formaldehyde and embedded in paraffin wax , sliced in 4 μm sections and stained with hematoxylin and eosin staining ( H and E ) . The atrium , valves and myocardium of heart , kidney , liver and gall bladder , forebrain , midbrain and hindbrain , tibial and quadriceps muscle , urinary bladder and reproductive organs ( testis , epididymis , accessory glands , ovary and uterus ) were evaluated . Histological analysis was carried out blindly using mouse identification codes for group assignment that were unknown to the evaluator . Analysis of mitochondrial DNA carryover mtDNA carryover in embryo specimens and adult mouse tissues was determined by SNP quantification using a high-throughput sequencing protocol . Prior to sequencing , polymerase chain reaction ( PCR ) was performed to amplify the SNP located at m . 3932 in the mtDNA ( B6CBAF1: A; NZB: G ) . DNA from embryo specimens was obtained by alkaline lysis . After the addition of 0 . 75 μl nuclease-free water , 1 . 25 μl 0 . 1M DL-Dithiothreitol and 0 . 5 μl 1 . 0M Sodium hydroxide solution ( per sample ) , cells were lysed at 65°C for 10 min . Genomic DNA ( gDNA ) from organs ( tail tips , hearts , brains , livers and kidneys ) was extracted using the DNeasy Blood and Tissue Kit from Qiagen . A single PCR mixture consisted of 1 . 5 μl HotMaster Taq DNA Buffer with Magnesium ( 5 Prime ) , 0 . 6 μl of 100 μm primer pool ( 5’-CCATACCCCGAAAACGTTGG-3’ and 5’-GGTTGGTGCTGGATATTGTGA-3’ ) , 0 . 3 μl 10 nM dNTP Mix and 0 . 09 μl HotMaster Taq DNA Polymerase ( 5 Prime ) . The PCR mix was added to lysed embryo specimens along with 7 . 99 μl nuclease-free water and 2 . 5 μl 0 . 4M Tricine ( per sample ) and to 0 . 5 μl of gDNA along with 12 . 49 μl nuclease-free water ( per sample ) . PCRs were performed using the following conditions: 96 . 0°C for one minute; 35 cycles of 94 . 0°C for 15 s , 58°C for 15 s and 65 . 0°C for 45 s; 65 . 0°C for two minutes . Successful amplification was verified by gel electrophoresis . Sequencing libraries were prepared from PCR amplicons using the Ion Plus Fragment Library Kit from ThermoFisher . Libraries were sequenced on the Ion Personal Genome Machine ( PGM; ThermoFisher ) . The Torrent Variant Caller plugin ( ThermoFisher ) was used for SNP allele quantification . In order to increase variant calling accuracy the settings for ‘Somatic’ variants were set to ‘High Stringency’ , to enable low frequency variant detection at a minimal false-positive call rate . The read depth was downsampled to 20 , 000 to increase accuracy of variant calls . A ‘HotSpot Region’ BED file , defining the exact genomic coordinate of the assessed nucleotide , in addition to a ‘Target Region’ BED file , was used . Prior to the analysis of embryo specimens and tissue samples , validation experiments were performed . Minisequencing was used to confirm SNP alleles A and G at position m . 3932 in B6CBAF1 and NZB mouse strains , respectively . PCR amplicons ( 1 μl ) from gDNA ( extracted from tail tips ) of the B6CBAF1 and NZB mouse strains were treated with 0 . 5 μl EXOSAP-it ( Affymetrix ) and incubated at 37°C for 15 min and 80°C for 15 min . PCR amplicons ( 1 . 5 μl ) were combined with 0 . 5 μl water , 2 . 5 μl SNaPshot Multiplex Ready Reaction Mix ( ThermoFisher ) and 0 . 5 μl primer ( 2 μM; 5’-AATAAATCCTATCACCCTT-3’; 5’-ATTGTGAAGTAGATGATGG-3’ ) . Mixtures were incubated using the following conditions: 25 cycles of 96 . 0°C for ten seconds , 50°C for 5 s and 60°C for 30 s . Products were analyzed by capillary electrophoresis on a genetic analyser ( ThermoFisher ) . The resulting data were analyzed using GeneMapper v4 . 0 software ( Applied Biosystems ) . DNA mixing experiments were performed to ensure accuracy and sensitivity of the SNP quantification protocol . Sample mixtures were created by combining gDNA ( extracted from tail tips ) from both mouse strains at different ratios ( B6CBAF1/NZB: 100/0; 98/2; 96/4; 94/6; 92/8; 90/10; 75/25; 50/50; 25/75; 0/100 ) . Samples were sequenced and obtained ratios compared to those expected . To ensure both the validity of assessment of a single SNP for mtDNA carryover analysis and the accuracy of the utilized sequencing platform ( Ion PGM ) ; a second set of experiments was performed , which included analysis of four additional SNPs ( B6CBAF1 > NZB: m . 2798C > T; m . 2814T > C; m . 3194T > C; m . 3260A > G ) utilizing Illumina’s MiSeq System sequencing platform . Minisequencing was used to confirm presence/absence of SNPs in B6CBAF1 and NZB mouse strains . PCR and minisequencing procedures were performed as described above . In brief , PCR was performed to amplify additional SNPs in gDNA from B6CBAF1 and NZB tail tips ( m . 2798 and m . 2814: 5’-AACACTCCTCGTCCCCATTC-3; and 5’-TGGACCAACAATGTTAGGGC-3’; m . 3194 and m . 3260: 5’-GCCGTAGCCCAAACAATTTC-3’ and 5’-GGTCAGGCTGGCAGAAGTAA-3’ ) . Amplicons were subjected to minisequencing with following primers: 5’-TCGTCCCCATTCTAATCGC-3’ and 5’-TGTTAGGAAGGCTAT-3’ ( m . 2798T ) ; 5’-ATAGCCTTCCTAACA-3’ and 5’-AAGATTTTGCGTTCTACTA-3’ ( m . 2814 ) ; 5’-ATGAAGTAACCATAGCTAT-3’ and 5’-ATAGAACTGATAAAAGGAT-3’ ( m . 3194 ) ; 5’-CACTTATTACAACCCAAGA-3’ and 5’-GCAGAAGTAATCATATGTG-3’ ( m . 3260 ) . Sequencing libraries were prepared from PCR amplicons using the TruSeq DNA Nano LT kit from Illumina and sequenced on the MiSeq System . Analysis was performed with Miseq Reporter and Illumina’s Somatic Variant Caller . Again , gDNA mixtures ( B6CBAF1/NZB: 100/0; 99/1; 97/3; 90/10; 75/25; 50/50; 0/100 ) were sequenced and obtained ratios compared to those expected . Furthermore , gDNA samples from organs of three individual mice were sequenced and results compared to those obtained by PGM sequencing . Experiments involving micromanipulation procedures were usually repeated between 6 to 9 times . Results obtained in the different replicates were pooled and analyzed together . Oocytes used for manipulation were always taken randomly from a common pool of oocytes collected from the 4–6 female mice used on each experimental day . In all experiments that involved embryo culture , control groups with non-manipulated oocytes were always processed and cultured in parallel together with the manipulated groups . All statistical analyses were performed using Prism 6 . 0 program ( GraphPad ) . For comparisons of mean cell numbers , placentas and mice weights , mtDNA carryover values in embryo species and adult mouse tissues , a t-test or one-way ANOVA was performed , where the significance was set at p<0 . 05 . For the analysis of oocyte/embryo proportions , chi-square test was performed and a p-value<0 . 05 was considered significant . - Validation of established sequencing protocol for mtDNA carryover analysis measured by Ion PGM sequencer . Allele frequencies at position m . 3932 in homoplasmic samples and artificially constructed heteroplasmic sample mixtures . See also Source data 1 . | Infertility is a growing problem that affects millions of people worldwide . Medical procedures known as in vitro fertilization ( IVF ) help many individuals experiencing infertility to have children . Typically in IVF , a woman’s egg cells are collected , fertilized with sperm from a chosen male and grown for a few days in a laboratory , before returning them to the woman’s body to continue to develop . However , there are some women whose egg cells cannot develop into a healthy baby after they have been fertilized . Many of these patients use egg cells from donors , instead . This greatly improves the chances of the IVF treatment being successful , but the resultant children are not genetically related to the intended mothers . Previous studies suggested that a cell compartment known as the cytoplasm plays a crucial role in allowing fertilized egg cells to develop normally . A new technique known as maternal spindle transfer , often shortened to MST , makes it possible to replace the entire cytoplasm of a compromised egg cell . This is achieved by transplanting the genetic material of the compromised egg cell into a donor egg cell with healthier cytoplasm that has previously had its own genetic material removed . Using this technique , it is possible to generate human egg cells for IVF that have the genetic material from the intended mother without the defects in the cytoplasm that may be responsible for infertility . However , it is not clear whether this approach would be a safe and effective way to treat infertility in humans . Costa-Borges et al . applied MST to infertile female mice and found that the technique could permanently correct deficiencies in the cytoplasms of poor quality egg cells , allowing the mice to give birth to healthy offspring . Further experiments studied the offspring and their descendants over several generations and found that they also had higher quality egg cells and normal levels of fertility . These findings open up the possibility of developing new treatments for infertility caused by problems with egg cells , so experiments involving human egg cells are now being performed to evaluate the safety and effectiveness of the technique . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology",
"cell",
"biology"
] | 2020 | Maternal spindle transfer overcomes embryo developmental arrest caused by ooplasmic defects in mice |
Many biological features are conserved and thus considered to be resistant to evolutionary change . While rapid genetic adaptation following the removal of conserved genes has been observed , we often lack a mechanistic understanding of how adaptation happens . We used the budding yeast , Saccharomyces cerevisiae , to investigate the evolutionary plasticity of chromosome metabolism , a network of evolutionary conserved modules . We experimentally evolved cells constitutively experiencing DNA replication stress caused by the absence of Ctf4 , a protein that coordinates the enzymatic activities at replication forks . Parallel populations adapted to replication stress , over 1000 generations , by acquiring multiple , concerted mutations . These mutations altered conserved features of two chromosome metabolism modules , DNA replication and sister chromatid cohesion , and inactivated a third , the DNA damage checkpoint . The selected mutations define a functionally reproducible evolutionary trajectory . We suggest that the evolutionary plasticity of chromosome metabolism has implications for genome evolution in natural populations and cancer .
The central features of many fundamental biological processes have been conserved since the last common ancestor of all extant organisms . Many of the proteins involved in these processes are essential , and the complex molecular interactions between them have been argued to constrain the evolution of both the processes and the proteins that carry them out ( Hirsh and Fraser , 2001; Jordan et al . , 2002; Wilson et al . , 1977 ) . The strength of these constraints has been questioned by studies that demonstrated that organisms can evolutionary adapt to the removal of important , and sometimes essential cellular genes ( Liu et al . , 2015; Rancati et al . , 2008 ) . Although the mutations that cause some adaptations have been identified , we lack a mechanistic understanding of how they repair the initial defect . Furthermore , in systematic studies , defects in some processes , such as intracellular trafficking , were more easily repaired , by single genetic events , than others , such as ribosome biogenesis , mRNA synthesis and DNA replication ( Liu et al . , 2015; van Leeuwen et al . , 2016 ) . Replication requires multiple enzymes that catalyze individual reactions such as unwinding the double helix , priming replication , and synthesizing new DNA strands ( O'Donnell et al . , 2013 ) . A common feature of replication is the organization of these enzymatic activities in multi-molecular complexes called replisomes , whose function is to coordinate the simultaneous synthesis of DNA from the two anti-parallel template strands ( Yao and O'Donnell , 2016 ) . The temporal and physical interactions amongst the enzymatic machines that performs the different steps of DNA replication are remarkably conserved . Nevertheless , differences in many features of DNA replication have been reported: the number of replisome subunits is higher in eukaryotes than in bacteria , possibly to account for the higher complexity of eukaryotic genomes ( McGeoch and Bell , 2008 ) . Some subunits are only found in some eukaryotic species ( Aves et al . , 2012; Liu et al . , 2009 ) . Notably , there are also biochemical variations in important features , such as the helicase , which encircles the leading strand in eukaryotes and the lagging strand in prokaryotes ( McGeoch and Bell , 2008 ) , or differences in the regulation of DNA replication by the machinery that drives the cell cycle progression ( Cross et al . , 2011; Parker et al . , 2017; Siddiqui et al . , 2013 ) . These differences reveal that although the DNA replication module performs biochemically conserved reactions , its features can change during evolution . This observation poses an apparent paradox: how can such an important process change during evolution without killing cells ? One hypothesis is that the overall organization of DNA replication can change as a consequence of accumulating several mutations , each perturbing a single aspect of replication , in response to a severe initial perturbation . To test this hypothesis , we followed the evolutionary response to a genetic perturbation of DNA replication . Characterizing evolutionary responses to genetic perturbations has informed studies of functional modules ( Filteau et al . , 2015; Harcombe et al . , 2009; Rojas Echenique et al . , 2019 ) , challenged the notion that particular genes are essential ( Liu et al . , 2015; Rancati et al . , 2018 ) , and revealed that initial genotypes can determine evolutionary trajectories ( Lind et al . , 2015; Rojas Echenique et al . , 2019; Szamecz et al . , 2014 ) . We followed the evolutionary response of S . cerevisiae to DNA replication stress , an overall perturbation of DNA replication that interferes with chromosome metabolism , reduces cell viability , and induces genetic instability ( Muñoz and Méndez , 2017; Zeman and Cimprich , 2014 ) . DNA replication stress has been implicated in both cancer progression and aging ( Burhans and Weinberger , 2007; Gaillard et al . , 2015 ) but despite studies investigating the direct effect of replication stress on cell physiology , its evolutionary consequences are unknown . We imposed constitutive replication stress by removing Ctf4 , a component of the replisome and evolved eight populations for 1000 generations . We exploited the ability of experimental evolution to identify , analyze , and compare the mutations that create parallel evolutionary trajectories to increase fitness ( Barrick and Lenski , 2013; Van den Bergh et al . , 2018 ) . We found that populations can recover from the fitness defect induced by DNA replication stress . Genetic analysis revealed that their adaptation is driven by mutations that damage , alter , and improve conserved features of three modules involved in chromosome metabolism: DNA replication , the DNA damage checkpoint , and sister chromatid cohesion . These mutations arise sequentially and collectively allow cells to approach the fitness of their wild-type ancestors within 1000 generations of evolution . The molecular basis of these adaptive strategies and their epistatic interactions produce a mechanistic model of the evolutionary adaptation to replication stress . Our results reveal the short-term evolutionary plasticity of chromosome metabolism . We discuss the consequences of this plasticity for the evolution of species in the wild and cancer progression .
Replication stress refers to the combination of the defects in DNA metabolism and the cellular response to these defects in cells whose replication has been substantially perturbed ( Macheret and Halazonetis , 2015 ) . Problems in replication can arise at the sites of naturally occurring or experimentally induced lesions and can cause genetic instability ( Muñoz and Méndez , 2017 ) . We asked how cells evolve to adapt to constitutive DNA replication stress . Previous work has induced replication stress by using chemical treatments or genetic perturbations affecting factors involved in DNA replication ( Mazouzi et al . , 2016; Tkach et al . , 2012; Zheng et al . , 2016 ) . To avoid evolving resistance to drugs or the reversion of point mutations that induce replication stress , we chose instead to remove CTF4 , a gene encoding an important , but non-essential , component of the DNA replication machinery . Ctf4 is a homo-trimer , that serves as a structural hub within the replisome and coordinates different aspects of DNA replication by binding the replicative helicase , the primase , and other factors recruited to the replication fork ( Figure 1A; Gambus et al . , 2009; Samora et al . , 2016; Simon et al . , 2014; Tanaka et al . , 2009; Yuan et al . , 2019 ) . In the absence of Ctf4 , cells experience several problems in fork progression leading to the accumulation of defects commonly associated with DNA replication stress ( Muñoz and Méndez , 2017 ) , such as single-stranded DNA gaps and altered replication forks ( Abe et al . , 2018; Fumasoni et al . , 2015; Kouprina et al . , 1992 ) . Ctf4 is essential for viability in vertebrates ( Abe et al . , 2018; Yoshizawa-Sugata and Masai , 2009 ) , insects ( Gosnell and Christensen , 2011 ) , and some fungi ( Harris and Hamer , 1995; Williams and McIntosh , 2002 ) but cannot be detected in prokaryotes , where there is a direct physical linkage between the primase ( DnaG ) and the helicase ( DnaB ) ( Lu et al . , 1996 ) . We generated ctf4Δ and wild type ( WT ) ancestor strains by sporulating a heterozygous CTF4/ctf4Δ diploid . As previously reported ( Kouprina et al . , 1992; Miles and Formosa , 1992 ) , ctf4Δ cells display severe growth defects , which we quantified as a fitness decrease of approximately 25% relative to WT ( Figure 1C ) . We then evolved eight parallel populations of each genotype for 1000 generations by serial dilutions in rich media , freezing population samples every 50 generations ( Figure 1B ) . Under this regime , spontaneous mutations that increase cellular fitness and survive genetic drift will be selected and spread asexually within the populations ( Jerison and Desai , 2015; Levy et al . , 2015; Venkataram et al . , 2016 ) . At the end of the experiment , we asked whether cells had recovered from the fitness decrease induced by replication stress by measuring the fitness of the evolved ctf4∆ and WT populations . Expressing the results as a percentage of the fitness of the WT ancestor , the evolved WT populations increased their fitness by an average of 4 . 0 ± 0 . 3% ( Figure 1—figure supplement 1 ) , a level similar to previous experiments ( Buskirk et al . , 2017; Lang et al . , 2013 ) . In contrast , we found that the fitness of the evolved ctf4∆ populations rose by 17 ± 0 . 2% ( Figure 1—figure supplement 1 ) . Clones isolated from these populations showed similar fitness increases ( Figure 1C ) . To understand this evolutionarily rapid adaptation to constitutive replication stress , we whole-genome sequenced all the final evolved populations as well as 32 individual clones ( four from each of the evolved populations ) isolated from the ctf4Δ lineages . During experimental evolution , asexual populations accumulate two types of mutations: adaptive mutations that increase their fitness and neutral or possibly mildly deleterious mutations that hitchhike with the adaptive mutations ( Supplementary file 1 ) . To distinguish between these mutations , we used a combination of statistical and experimental approaches . First , we inferred that mutations in a gene were adaptive if the gene was mutated more frequently than expected by chance across our parallel and independent populations ( Supplementary file 2 ) . Second , we performed bulk segregant analysis on selected evolved clones . This technique takes advantage of sexual reproduction , followed by selection , to separate causal and hitchhiking mutations . In this case , mutations that segregate strongly with the evolved phenotype are assumed to be adaptive ( Figure 1—figure supplement 2 ) . We combined these two lists of mutated genes and looked for enriched gene ontology ( GO ) terms . This analysis revealed an enrichment of genes implicated in several aspect of chromosome metabolism ( Supplementary file 3 ) . Among the genes associated with these terms , many are involved in four functional modules: DNA replication , chromosome segregation ( including genes involved in sister chromatid linkage and spindle function ) , cell cycle checkpoint and chromatin remodeling ( Figure 1—figure supplement 3 ) . The genes in these modules that were mutated in the evolved clones are shown , grouped by function , in Figure 1D . We found several mutations affecting genes involved in cell-cycle checkpoints ( Figure 2B ) . Checkpoints are feedback control mechanisms that induce cell-cycle delays in response to defects that reflect the failure to complete important process and thus guarantee the proper sequence of events required for cell division ( Elledge , 1996; Murray , 1992 ) . Three delays , caused by DNA damage or defects in DNA replication , have been characterized . The first prevent cells from entering S-phase in response to DNA damage occurring in G1 . A second slows progress through S-phase in response to problems encountered during DNA synthesis . The third delays sister chromatid separation ( anaphase ) and the exit from mitosis in response to DNA damage incurred after cells enter S-phase ( Figure 2A; Murray , 1994 ) . The genes listed in Figure 2B are implicated at different levels in either the replication or mitotic delays ( Figure 2—figure supplement 1B; Pardo et al . , 2017 ) . The most frequently mutated gene , RAD9 , encodes an important component of the DNA damage checkpoint , which is required to slow DNA synthesis and delay anaphase in response to DNA lesions ( Weinert and Hartwell , 1988 ) . Four out of the five mutations in RAD9 produced early stop codons , or radical amino acid substitutions in the BRCT domain , which is essential for Rad9’s function ( Figure 2C , Figure 2—figure supplement 1A; Soulier and Lowndes , 1999 ) , arguing that inactivation of Rad9 was repeatedly and independently selected for during evolution . To test this hypothesis , we engineered the most frequently occurring mutation ( 2628 +A , a frameshift mutation leading to a premature stop codon K883* ) into the ancestral ctf4∆ strain ( ctf4∆ anc ) . We suspect that the high frequency of this mutation is due to the presence of a run of 11 As , a sequence that is known to be susceptible to loss or gain of a base during DNA replication . This mutation ( Figure 2C , Figure 2—figure supplement 1A ) produced a fitness increase very similar to the one caused by deleting the entire gene ( Figure 2D ) . We conclude that inactivation of Rad9 is adaptive in the absence of Ctf4 . We asked if the removal of Rad9 eliminated a cell cycle delay caused by the absence of Ctf4 . In the ctf4∆ ancestor , rad9Δ does , indeed , decrease the fraction of cells with a 2C DNA content ( the DNA content in G2 and mitosis ) observed in asynchronously growing ctf4∆ cells ( Tanaka et al . , 2009 ) . This observation suggests that the interval between the end of DNA replication and cell division decreases in ctf4∆ rad9Δ cells . The spindle checkpoint , which blocks anaphase in response to defects in mitotic spindle assembly , can also delay chromosome segregation in cells ( Li and Murray , 1991 ) . But although deleting MAD2 , a key spindle checkpoint component , also decreases the interval between replication and division in ctf4Δ cells ( Hanna et al . , 2001 ) , it reduces rather than increases the fitness of the ctf4∆ ancestor ( Figure 2D ) . These results suggest that ignoring some defects in ctf4∆ cells , such as those that activate the DNA damage checkpoint , improves fitness , whereas ignoring others , such as defects in chromosome alignment on the spindle , reduces fitness . Problems encountered during DNA synthesis also activate the replication checkpoint , which inhibits DNA replication to prevent further lesions ( Zegerman and Diffley , 2009; Zegerman and Diffley , 2010 ) . As many proteins involved in the DNA damage checkpoint are shared with the replication checkpoint ( Figure 2—figure supplement 1B; Pardo et al . , 2017 ) , we followed a single synchronous cell-cycle to ask whether the fitness benefits conferred by RAD9 deletion were due to a faster progression through S-phase or faster progress through mitosis . Loss of Rad9 in ctf4∆ cells did not accelerate S-phase , but it did lead to faster passage through mitosis as revealed by a reduced fraction of 2 C cells ( Figure 2E ) . To separate the role of the replication and DNA damage checkpoints , we genetically manipulated targets of the checkpoints whose phosphorylation delays either anaphase ( Pds1 , Wang et al . , 2001 ) or the completion of replication ( Sld3 and Dbf4 , Zegerman and Diffley , 2010 , Figure 2—figure supplement 1B ) . Fitness measurement in these mutants ( pds1-m9 or the double mutant sld3-A/dbf4-4A ) showed that while decreasing the mitotic delay in ancestral ctf4∆ cells was beneficial , a faster S-phase was highly detrimental ( Figure 2—figure supplement 1C ) . Collectively , these results show that the specific absence of a DNA damage-induced delay of anaphase , rather than generic cell-cycle acceleration , is adaptive in ctf4Δ cells experiencing replication stress . We examined the evolved clones for changes in the copy number across the genome ( DNA copy number variations , CNVs ) . Several clones showed segmental amplifications , defined as an increase in the copy number of a defined chromosomal segment ( Figure 3—figure supplement 1 ) . The most common CNV in evolved ctf4Δ cells ( 17 out of 32 sequenced clones ) was the amplification of a 50–100 kb region of chromosome IV ( chrIV ) . In addition to this segmental amplification , evolved clone EVO2-10 also carried an extra copy of a portion of chromosome V ( chrV , Figure 3A ) . The eight evolved wild type populations had no segmental amplifications , suggesting that changes in copy number were a specific adaptation to constitutive DNA replication stress . Amongst the genes affected by these two CNVs are SCC2 and SCC4 , on the amplified portions of chromosomes IV and V respectively . These two genes encode the two subunits of the cohesin loader complex , which loads cohesin rings on chromosomes to ensure sister chromatid cohesion until anaphase ( Figure 3—figure supplement 2B; Ciosk et al . , 2000; Michaelis et al . , 1997 ) . The amplification of SCC2 and SCC4 , together with the other genes altered by point mutations in our evolved clones ( Figure 3B , Figure 3—figure supplement 2A ) , strongly suggest that the absence of Ctf4 selects for mutations that affect the linkage between sister chromatids . CTF4 was originally identified because mutants in this gene reduced the fidelity of chromosome transmission ( CTF = chromosome transmission fidelity , Spencer et al . , 1990 ) ; later studies showed that this defect was due to premature sister chromatid separation , which resulted in increased chromosome loss at cell division ( Hanna et al . , 2001 ) . We hypothesized that the segmental amplifications of chrIV and chrV were selected to increase the amount of the cohesin loading complex . To test this idea , we reintroduced a second copy of these genes in a ctf4∆ ancestor . As predicted by the more frequent amplification of SCC2 , we found that while an extra copy of SCC4 alone did not significantly affect fitness , an extra copy of SCC2 , or an extra copy of both SCC2 and SCC4 increased fitness by 4–5% ( Figure 3C ) . Consistent with a role of SCC4 amplification only in combination of SCC2 amplification , we found that the segmental amplification of chrV followed that of chrIV in the EVO2 population ( Figure 3—figure supplement 3A–B ) . We examined cells arrested in mitosis to measure the extent of premature sister chromatid separation in the same strains . Adding extra copies of the cohesin loader subunits improved sister chromatid cohesion ( Figure 3D ) and the amplitude of the improvement in sister cohesion for different strains had the same rank order as their increase in fitness ( Figure 3C ) . We conclude that the increased copy number of the cohesin loader subunits is adaptive and alleviates the cohesion defects induced by the lack of Ctf4 . We found mutations in several genes involved in DNA replication ( Figure 4A , Figure 4—figure supplement 1A ) . Among these , we found four independent mutations ( Figure 4B ) that altered three different subunits of the replicative CMG ( Cdc45 , MCM , GINS ) helicase ( Labib and Gambus , 2007; Moyer et al . , 2006 ) . The CMG helicase is bound in vivo by Ctf4 through the GINS subunit Sld5 ( Simon et al . , 2014 ) . This binding allows Ctf4 to coordinate the helicase’s progression with primase , which synthesizes the primers for lagging strand DNA synthesis , and other factors recruited behind the replication fork ( Figure 1A; Samora et al . , 2016; Villa et al . , 2016 ) . A CMG helicase mutation found in one of the evolved clones , sld5-E130K , increased the fitness of the ancestral ctf4∆ strain ( Figure 4C ) . IXR1 , a gene indirectly linked to DNA replication , was mutated in several populations ( Figure 4A ) . IXR1 encodes for a transcription factor that indirectly and positively regulates the concentration of deoxyribonucleotide triphosphates ( dNTPs , Tsaponina et al . , 2011 ) , the precursors for DNA synthesis . The occurrence of multiple nonsense mutations in this gene strongly suggested selection to inactivate Ixr1 ( Figure 4—figure supplement 1B ) . Consistent with this prediction , we found that engineering either a nonsense mutation ( ixr1-Q332* ) or a gene deletion conferred a selective advantage to ctf4∆ ancestor cells ( Figure 4C ) . We asked how mutations in the replicative helicase or inactivation of IXR1 increased the fitness of ctf4∆ cells . One hypothesis is that the absence of Ctf4 reduces the coordination of activities required to replicate DNA and leads to the appearance of large regions of single stranded DNA , which in turn exposes the forks to the risk of nuclease cleavage or collapse . If this were true , slowing the replicative helicase or the synthesis of the leading strand would reduce the amount of single stranded DNA near the replication fork and improve the ability to complete DNA replication before cell division . To test this idea , we used whole genome sequencing at different points during a synchronous cell cycle to compare the dynamics of DNA replication ( Figure 4—figure supplement 2 ) in four strains: WT , the ctf4∆ ancestor , and ctf4∆ strains containing either the sld5-E130K or ixr1Δ mutations . We found that cells lacking Ctf4 experience several defects compared to WT: on average , origins of replication fire later and DNA replication forks proceed more slowly across replicons , often showing fork stalling ( Figure 4D–E , Figure 4—figure supplement 3 ) . As a consequence of these two defects , cells still contain significant regions of unreplicated DNA late in S-phase ( 45 min , Figure 4D–E , Figure 4—figure supplement 3 ) . Both sld5-E130K or ixr1Δ mutations significantly increase the average replication fork velocity primarily by avoiding stalls in DNA replication and thus leading to earlier replication of the regions that replicate late in the ancestral ctf4∆ cells ( Figure 4D–E , Figure 4—figure supplement 3 ) . Altogether , these results show that cells evolved modified DNA replication dynamics to compensate for defects induced by DNA replication stress . Can we explain how the ancestral ctf4∆ strains recovered to within 10% of WT fitness in only 1000 generations ? Although all the mutations that we engineered into ctf4∆ ancestor cells reduce the cost of DNA replication stress , none of them , individually , account for more than a third of the fitness increase observed over the course of the entire evolution experiment ( Figure 1C ) . Sequencing individual evolved clones revealed the presence of mutations in at least two of the three modules whose effects we analyzed in isolation ( Figure 5—figure supplement 1 , Supplementary file 1 ) . We therefore asked if we could recapitulate the fitness of the evolved clones by adding adaptive mutations from multiple different modules to the ctf4∆ ancestor . We obtained all possible combinations of two , three , and four adaptive mutations , in the ctf4∆ ancestor , by sporulating a diploid strain that was heterozygous for all four classes of adaptive mutations: inactivation of the DNA damage checkpoint ( rad9∆ ) , amplification of the cohesin loader ( an extra copy of SCC2 ) , alteration of the replicative helicase ( sld5-E130K ) , and altered regulation of dNTP pools ( ixr1∆ ) . We found that the two mutations that affected DNA replication were negatively epistatic ( Figure 5A ) : in the presence of ctf4∆ , strains that contained both sld5-E130K and ixr1Δ were not significantly more fit than strains that contained only ixr1∆ and the quadruple mutant ( 2X-SCC , rad9Δ , sld5-E130K , ixr1Δ ) was much less fit than the two triple mutants that contained only one of the two mutations that affected DNA replication ( 2X-SCC , rad9Δ , sld5-E130K and 2X-SCC , rad9Δ , ixr1Δ ) . As a result , the two fittest strains carry only three mutations: in both cases , they affected the three modules we previously characterized: sister chromatid linkage and chromosome segregation ( 2X-SCC2 ) , the DNA damage checkpoint ( rad9Δ ) and DNA replication ( sld5-E130K or ixr1Δ ) . These two strains displayed a fitness comparable to the average of the evolved populations ( Figure 1C ) , suggesting that we had recapitulated the major adaptive events in our engineered strains . We asked if the antagonistic interaction between sld5-E130K and ixr1Δ seen in our reconstructed strains had also occurred in our evolution experiment . We focused on an evolved population ( EVO5 ) that carried all the mutations described above and analyzed the allele frequency in the intermediate samples collected across the evolution experiment . By following the frequency of alleles within the population and sequencing individual clones , we found that the mutations in the three modules happened in three consecutive selective waves: first , cells acquired an extra copy of the cohesin loader-encoding gene SCC2 , second , ixr1-Q332* and sld5-E130K appeared , simultaneously , in two different lineages , and finally rad9-N876K appeared independently in the two lineages containing either ixr1-Q332* or sld5-E130K ( Figure 5B , Figure 5—figure supplement 2 ) . After their initial appearance , the two lineages containing ixr1-Q332* or sld5-E130K competed with each other for the remainder of the experiment . In this population , both final lineages accumulated mutations whose interaction was nearly additive or positively epistatic and avoided combinations that show strong negative epistasis ( Figure 5A ) . Thus , although negative epistasis exists , selection finds trajectories that avoid it , as previously observed in a similar experiment perturbing cell polarity ( Laan et al . , 2015 ) .
Cells lacking Ctf4 show an increased frequency of chromosome mis-segregation due to premature sister chromatid separation , but the mechanism underlying this defect is still unclear . Seven of our eight populations amplified SCC2 , which encodes for one of the subunits of the cohesin loader complex . The simplest explanation for this result is that , the absence of Ctf4 restricts the productive loading of cohesin molecules that establish the linkage between sister chromatids . We propose that amplifying the genes for the cohesin loader would increase its expression , increase the productive cohesin loading and improve the linkage between sister chromatids . Improving sister chromatid cohesion allows the evolved cells to segregate their chromosomes more accurately at mitosis , avoiding mitotic delays due to the spindle checkpoint , decreasing cell death and increasing fitness ( Figure 6A ) . Persistent , cohesin-independent linkages between sister chromatids are an alternative source of segregation errors . These links include unreplicated regions of DNA or un-resolved recombination structures ( Ait Saada et al . , 2017; Chan et al . , 2007 ) . If they persist after the removal of cohesin , they become lingering physical links ( anaphase bridges ) between sister chromatids that can lead to chromosome breakage or mis-segregation during anaphase ( Chan et al . , 2009; Gisselsson et al . , 2000 ) . Avoiding these problems requires that replication origins fire efficiently and replication forks move continuously . Our analysis of the dynamics of DNA replication argues that a combination of frequent fork stalling and slower origin firing causes under-replication of certain chromosomal regions in the ancestral ctf4∆ cells . We found that severe fork stalling in ctf4∆ cells frequently occurs near tRNA genes , Long Terminal Repeats ( LTRs ) and transposable elements ( Ty ) ( Supplementary file 4 ) . These chromosomal features were previously found to be associated with replication pausing sites ( Deshpande and Newlon , 1996; Gadaleta and Noguchi , 2017; Zaratiegui et al . , 2011 ) , suggesting that the absence of Ctf4 may exacerbate the problems in replicating these regions . We propose that these defects selected for mutations that stabilize the replication forks , allowing the timely completion of genome replication . We speculate that these mutations have the apparently paradoxical effect of accelerating DNA replication by slowing down the replication forks: mutations like sld5-E130K and ixr1∆ may slow helicase progression , stabilizing the forks by preventing frequent fork stalling or collapse and producing a higher overall fork velocity ( Figure 6B ) . This hypothesis is consistent with two observations: first , although the sld5 mutation is beneficial in ancestor cells , it decreases the fitness of WT cells ( Figure 4—figure supplement 4A ) , a result we would expect from a slower replicative helicase . Second , reduced dNTPs concentrations reduce fork speed by slowing polymerase incorporation rates ( Koren et al . , 2010; Pai and Kearsey , 2017; Poli et al . , 2012 ) and inactivating Ixr1 reduces dNTP concentrations ( Tsaponina et al . , 2011 ) . We tested this prediction by using an experimental system to manipulate dNTP concentrations: decreasing dNTP concentrations increased the fitness of ctf4∆ cells , while inducing higher dNTP production reduced fitness ( Figure 4—figure supplement 4B ) . Our evolved populations also accumulated mutations that inactivated the DNA damage checkpoint ( Figure 2B–D ) . The benefit of these mutations arises from the loss of the DNA damage checkpoint’s ability to delay the start of anaphase ( Figure 2E , Figure 2—figure supplement 1B ) . The absence of Ctf4 induces aberrant DNA structures and ssDNA that induce moderate activation of the checkpoint ( Poli et al . , 2012 ) , which delays the start of anaphase , increasing doubling time and thus decreasing fitness ( Figure 2D–E ) . Inactivating Rad9 eliminates the delay , shortening the time required for mitosis and increasing fitness ( Figure 2E , Figure 2—figure supplement 1C , Figure 6C ) . This solution seems counter-intuitive , as the loss of a safeguard mechanism such as the DNA damage checkpoint should cause genetic instability in cells suffering from replication stress . The resolution of this paradox may lie in the overlapping action of the replication , DNA damage , and spindle checkpoints . We propose that the replication and the spindle checkpoints delay the cell cycle in response to defects that would kill the ancestral ctf4∆ cells , such as excessive replication fork collapses and pairs of sister chromatids attached to the same spindle pole , whereas the damage checkpoint responds to defects , like regions of single-stranded DNA , that can be repaired after cell division . We asked how the mutants we identified and analyzed interacted with each other and whether they could explain the fitness of our evolved populations . Measuring allele frequencies over time and engineering all possible combinations of adaptive mutations allowed us to propose a detailed model for the evolutionary trajectories of our population 5 ( EVO5 ) . Segmental amplifications form at a higher frequency than other types of mutation ( Lynch et al . , 2008; Sharp et al . , 2018; Yona et al . , 2015 ) ; although most are detrimental , the amplification of specific genes can be advantageous and cause rapid adaptation ( Adamo et al . , 2012; Gresham et al . , 2008; Hughes et al . , 2000; Payen et al . , 2014 ) . Thus , the first event in EVO5 is the spread of a segmental amplification of chromosome IV containing SCC2 , which improves fitness by reducing cohesion defects . In this lineage , mutations in the replicative helicase , sld5-E130K , and ixr1-Q332* were then detected almost simultaneously but in different clones . Above , we suggest that both mutations slow replication forks . If there is an optimal fork speed in ctf4∆ cells , the presence of a second mutation of this class might be ineffective or even detrimental if the forks move too slowly , explaining the negative epistasis we observed . Because the ixr1 and sld5 mutations improve DNA replication to a similar extent , the two lineages have comparable fitness , explaining the clonal interference that persists for the rest of the experiment . The last mutation in EVO5 is an identical frameshift mutation in the two lineages that inactivates Rad9 . Interestingly , loss of function mutations in RAD9 , despite the large target size of this gene , only appear relatively late during the experiment ( Figure 5A and Figure 5—figure supplement 2 ) . Furthermore , they happen after other mutations have reduced some of the problems imposed by replication stress . This order suggests that a sustainable fitness advantage of mutations of the DNA damage checkpoint may depend on previous changes in the replication forks stability . Despite being conserved across much of evolution , some of the modules that collectively perform chromosome metabolism and maintain genomes show major important differences between clades , even within the eukaryotic kingdom ( Akiyoshi and Gull , 2014; Gourguechon et al . , 2013; Liu et al . , 2009 ) . For instance , a recent study found species in the yeast genus Hanseniaspora that lack several important genes implicated in cell cycle progress and DNA repair , including checkpoint factors such as RAD9 and MAD2 ( Steenwyk et al . , 2019 ) . Trying to explain these differences is puzzling , especially if ad-hoc selectionist hypotheses are invoked for each different feature . For instance , what could select for a lack of an important safeguard such as the DNA damage checkpoint ? Interestingly , the same lineage of Hanseniaspora also lacks CDC13 ( Steenwyk et al . , 2019 ) , an essential gene in S . cerevisiae , implicated in telomere replication . Studies have shown how the lethality of cdc13Δ mutants , is suppressed by simultaneous mutations in checkpoint factors , including RAD9 ( Ngo and Lydall , 2010 ) . The evolutionary plasticity of chromosome metabolism that we reveal in this work may help to explain differences like these: mutations in ancestral cells , such as the loss of CDC13 , could initiate an evolutionary trajectory that progressively modifies modules that are functionally linked and ultimately leads to increased fitness . But what are the initial perturbations that trigger such changes in fundamental aspects of cell biology ? The ctf4∆ cells that we evolved have a 25% fitness difference relative to their wild type ancestors , meaning that they would rapidly be eliminated from any population of reasonable size . Given the evolutionary rarity of major rearrangements in cell biology we can invoke events that are improbable including passing through very small populations bottlenecks or being attacked by selfish genetic elements whose molecular biology targets an important protein in an essential process . If the processes that were damaged during these events , were part of chromosome metabolism , the consequent evolutionary adaptation could lead to changes in the rates at which the structures of genomes evolve . An increase in these rates , in turn , could potentially accelerating speciation by making it easier for populations to acquire meiotically incompatible chromosome configurations . Remarkably , our experiment recapitulates several phenomena observed during cancer development . Replication stress is thought to be a ubiquitous feature of cancer cells ( Macheret and Halazonetis , 2015 ) with oncogene activation leading to replication stress and genetic instability ( Bartkova et al . , 2006; Di Micco et al . , 2006; Neelsen et al . , 2013 ) . The absence of Ctf4 in our ancestor cells causes several phenotypes observed in oncogene-induced DNA replication stress including late-replicating regions , elevated mutation rates , and chromosome instability ( Fumasoni et al . , 2015; Macheret and Halazonetis , 2015; Muñoz and Méndez , 2017 ) . Furthermore , simply by propagating cells , we generated evolved lines that mimic many features seen in tumors: ( a ) individual final populations contain genetically heterogeneous clones , often with different karyotypes characterized by aneuploidies and chromosomal rearrangements ( Davoli et al . , 2013; Laughney et al . , 2015; Lengauer et al . , 1998 ) , ( b ) evolved lineages display altered DNA replication profiles compared both to WT cells and their mutant ancestors ( Amiel et al . , 1999; Donley and Thayer , 2013 ) , ( c ) several lines have inactivated the DNA damage checkpoint ( Hollstein et al . , 1991; Schultz et al . , 2000 ) , and d ) improved sister chromatid cohesion ( Rhodes et al . , 2011; Sarogni et al . , 2019; Xu et al . , 2011 ) . All these features are adaptive in our populations , suggesting that similar changes in cancer cells may be the result of selection and contribute to the accumulation of other cancer hallmarks during cancer evolution . The similarities between tumorigenesis and our experiment lead us to speculate that a major selective force in the early stages of tumor evolution is the need to counteract the fitness costs of replication stress . Understanding the evolutionary mechanisms and dynamics of the adaptation to replication stress could therefore shed light on the early stage of tumor development . In this work , we identified the main adaptive strategies that cells use to adapt to DNA replication stress induced by the absence of Ctf4 . Our results reveal that defects in one function can be compensated for by two types of mutations: those in the original function and those in functions that are biologically coupled to it . Focusing on less common adaptive strategies , apparently unlinked to chromosome metabolism , could therefore potentially identify novel players that affect genome stability . It would also be interesting to induce DNA replication stress by other means , such as de-regulating replication initiation or by inducing re-replication . Analyzing the response to these challenges will reveal whether the DNA replication module has a common or diverse set of evolutionary strategies to different perturbations . Finally , this approach could be extended to many other types of cellular stress , potentially revealing other molecular adaption aspects that could collectively help understanding cellular evolution .
All strains were derivatives of a modified version ( Rad5+ ) of S . cerevisiae strain W303 ( leu2-3 , 112 trp1-1 can1-100 ura3-1 ade2-1 his3-11 , 15 , RAD5+ ) . Supplementary file 5 lists each strain’s genotype . The ancestors of WT and ctf4Δ strains were obtained by sporulating a CTF4/ctf4Δ heterozygous diploid . This was done to minimize the selection acting on the ancestor strains before the beginning of the experiment . Diploid stains were grown on YPD , transferred to sporulation plates ( sodium acetate 0 . 82% , potassium chloride 0 . 19% , sodium chloride 0 . 12% , magnesium sulfate 0 . 035% ) and incubated for four days at 25°C . Tetrads were re-suspended in water containing zymolyase ( Zymo research , RRID:SCR_008968 , Irvine , CA , US , 0 . 025 u/μl ) , incubated at 37°C for 45 s , and dissected on a YPD plate using a Nikon eclipse E400 microscope equipped with a TDM micro-manipulator . Spores were allowed to grow into visible colonies and genotyped by presence of genetic markers and PCR . Standard rich medium , YPD ( 1% Yeast-Extract , 2% Peptone , 2% D-Glucose ) was used for all experiments except in the experiment in Figure 4—figure supplement 4B where YP + 2% raffinose and YP + 2% raffinose + 2% galactose were also used . Cells were synchronized either in metaphase , for 3 hr in YPD containing nocodazole ( 8 µg/ml , in 1% DMSO ) or in G1 , for 2 hr in YPD , pH 3 . 5 containing α-factor ( 3 µg/ml ) . Synchronization was verified by looking at cell morphology . In the experiment in Figure 2E , cells were then washed twice in YPD containing 50 µg/ml pronase ( Zymo research ) and released in S-phase at 30°C in YPD . α-factor ( 3 µg/ml ) was added again at 30 min to prevent a second cell cycle from occurring . The 16 populations used for the evolution experiment were inoculated in glass tubes containing 10 ml of YPD from eight ctf4Δ colonies ( EVO1-8 ) and 8 WT colonies ( EVO9-16 ) . All the colonies were derived by streaking out MATa ( EVO1-4 and EVO25-28 ) or MATα ( EVO5-9 and EVO29-32 ) ancestors . Glass tubes were placed in roller drums at 30°C and grown for 24 hr . Daily passages were done by diluting 10 μl of the previous culture into 10 ml of fresh YPD ( 1:1000 dilution , allowing for approximately 10 generations/cycle ) . All populations were passaged for a total of 100 cycles ( ≈1000 generations ) . Every five cycles ( ≈50 generations ) 800 μl of each evolving population was mixed with 800 μl of 30% v/v glycerol and stored at −80°C for future analysis ( Figure 1B ) . After 1000 generations four evolved clones were isolated from the each of the eight ctf4Δ evolved populations ( a total of 32 clones ) by streaking cells on a YPD plate . Single colonies were then grown in YPD media and saved in glycerol at −80°C as for the rest of the samples . Genomic DNA library preparation was performed as in Koschwanez et al . ( 2013 ) with an Illumina ( RRID:SCR_010233 , San Diego , CA , US ) Nextera DNA Library Prep Kit . Libraries were then pooled and sequenced either with an Illumina HiSeq 2500 ( 125bp paired end reads ) or an Illumina NovaSeq ( 150 bp paired end reads ) . The SAMtools software package ( RRID:SCR_002105 , samtools . sourceforge . net ) was then used to sort and index the mapped reads into a BAM file . GATK ( RRID:SCR_001876 , www . broadinstitute . org/gatk; McKenna et al . , 2010 ) was used to realign local indels , and VarScan ( RRID:SCR_006849 , varscan . sourceforge . net ) was used to call variants . Mutations were found using a custom pipeline written in Python ( RRID:SCR_008394 , www . python . org ) . The pipeline ( github . com/koschwanez/mutantanalysis ) compares variants between the reference strain , the ancestor strain , and the evolved strains . A variant that occurs between the ancestor and an evolved strain is labeled as a mutation if it either ( 1 ) causes a non-synonymous substitution in a coding sequence or ( 2 ) occurs in a regulatory region , defined as the 500 bp upstream and downstream of the coding sequence ( Supplementary file 1 ) . Three complementary approaches were combined to identify the putative modules and genes targeted by selection . To measure relative fitness , we competed the ancestors and evolved strains against reference strains . Both WT ( Figure 1C , Figure 1—figure supplement 1 , Figure 4—figure supplement 4A , Figure 5—figure supplement 2 ) and ctf4Δ ( Figure 2D , Figure 2—figure supplement 1C , Figure 3C , Figure 4C , Figure 4—figure supplement 4A–B , Figure 5A ) reference strains were used . A pFA6a-prACT1-yCerulean-HphMX4 plasmid was digested with AgeI and integrated at one of the ACT1 loci of the original heterozygous diploid ( CTF4/ctf4Δ ) strain . This allow for the expression of fluorescent protein yCerulean under the strong actin promoter . The heterozygous diploid was then sporulated and dissected to obtain fluorescent WT or ctf4Δ reference haploid strains . For measuring the relative fitness , 10 ml of YPD were inoculated in individual glass tubes with either the frozen reference or test strains . After 24 hr . the strains were mixed in fresh 10 ml YPD tubes at a ratio dependent on the expected fitness of the test strain compared to the reference ( i . e . 1:1 if believed to be nearly equally fit ) and allowed to proliferate at 30°C for 24 hr . 10 μl of samples were taken from this mixed culture ( day 0 ) and the ratio of the two starting strains was immediately measured . Tubes were then cultured following in the same conditions as the evolution experiment by diluting them 1:1000 into fresh medium every 24 hr for 4 days , monitoring the strain ratio at every passage . Strain ratios and number of generations occurred between samples were measured by flow cytometer ( Fortessa , BD Bioscience , RRID:SCR_013311 , Franklin Lakes , NJ , US ) . Ratios r were calculated based on the number of fluorescent and non-fluorescent events detected by the flow cytometer:r= NonFluorescenteventsFluorescentevents Generations between time points g were calculated based on total events measured at time 0 hr . and time 24 hr . :g=log10 ( eventst24/eventst0 ) log102 Linear regression was performed between the ( g , loger ) points relative to every sample . Relative fitness was calculated as the slope of the resulting line . The mean relative fitness s was calculated from measurements obtained from at least three independent biological replicates . Error bars represent standard deviations . The P-values reported in figures are the result of t-tests assuming unequal variances ( Welch’s test ) . Note that the absolute values of relative fitness change depending on the reference strain used: a strain that shows 27% increased fitness when measured against ctf4Δ ( that is 27% less fit then WT ) , does not equate the WT fitness . This is because a 27% increase of 0 . 73 ( ctf4Δ fitness compared to WT ) gives 0 . 93 , hence a 7% fitness defect compared to WT . Cell cycle analysis was conducted as previously described ( Fumasoni et al . , 2015 ) . In brief , 1 × 107 cells were collected from cultures by centrifugation , and resuspended in 70% ethanol for 1 hr . Cells were then washed in 50 mM Tris-HCl ( pH 7 . 5 ) , resuspended in the same buffer containing 0 . 4 μg/ml of RNaseA and incubated at 37°C for at least 2 hr . Cells were collected and further treated overnight at 37°C in 50 mM Tris-HCl ( pH 7 . 5 ) containing proteinase K ( 0 . 4 μg/ml ) . Cells were then centrifuged and washed in 50 mM Tris-HCl ( pH 7 . 5 ) . Samples were then diluted 10–20-fold in 50 mM Tris-HCl ( pH 7 . 8 ) containing 1 mM SYTOX green , and analyzed by flow cytometer ( Fortessa , BD Bioscience ) . The FITC channel was used to quantify the amounts of stained-DNA per cell . 10000 events were acquired for each sample . Cell cycle profiles were analyzed and visualized in FlowJo ( RRID:SCR_008520 , BD Bioscience ) . The percentage of genome replicated at 30 min was calculated based on the cell cycle profile as follow Grep=DNA content mode/2C-1C*100 . The height of the 1C and 2C peaks was obtained as the max cells count reached by the respective peak . For both the percentage of genome replicated at 30 min and the 1C/2C ratio , the mean was calculated from values obtained with three independent biological replicates . Error bars represent standard deviations . The P-values reported are the result of t-tests assuming unequal variances . Whole genome sequencing and read mapping was done as previously described . The read-depths for every unique 100 bp region in the genome were then obtained by using the VarScan copynumber tool . A custom pipeline written in python was used to visualize the genome-wide CNVs . First , the read-depths of individual 100 bp windows were normalized to the genome-wide median read-depth to control for differences in sequencing depths between samples . The coverage of the ancestor strains was then subtracted from the one of the evolved lines to reduce the noise in read depth visualization due to the repeated sequences across the genome . The resulting CNVs were smoothed across five 100 bp windows for a simpler visualization . Final CNVs were then plotted relative to their genomic coordinate at the center of the smoothed window . Since the WT CNVs were subtracted from the evolved CNVs , the y axis refers to the copy number change occurred during evolution ( i . e . +1 means that one an extra copy of a chromosome fragment has been gained ) . The custom pipeline used for the data analysis is available on GitHub: https://github . com/marcofumasoni/Fumasoni_and_Murray_2019 . Logarithmically growing cells were arrested in metaphase as previously described . Samples were then collected and fixed in 4% formaldehyde for 5 min at room temperature . Cells were washed In SK buffer ( 1M sorbitol , 0 . 05 M K2PO4 ) and sonicated for 8 s prior to microscope analysis . Images were acquired with a Nikon eclipse Ti spinning-disk confocal microscope using a 100X oil immersion lens . Fluorescence was visualized with a conventional FITC excitation filter and a long pass emission filter . Images were analyzed using Fiji ( RRID:SCR_002285 , https://fiji . sc/ ) . At least 100 cells were analyzed to calculate the percentage of premature chromatid Separation for each strain . The mean value was calculated from measurements obtained with three independent biological replicates . Error bars represent standard deviations . The P-values reported are the result of t-tests assuming unequal variances . DNA replication profiling was adapted from Müller et al . ( 2014 ) ; Saayman et al . , 2018; Bar-Ziv et al . ( 2016 ) . Genomic DNA and library preparation were performed independently on all the collected samples as previously described . Repeated sequences ( such as telomeres , rDNA and Ty elements ) were excluded from the CNV analysis as non-uniquely mapped reads can alter local read-depth and generate artefacts . A custom python script was used to analyze the CNVs from multiple time points from the same strain to produce DNA replication profiles . Read-depths of individual 100 bp windows were normalized to the genome-wide median read-depth to control for differences in sequencing depths between consecutive samples . To allow for intra-strain comparison , coverage was then scaled according to the sample DNA content measured as the median of the cell-cycle profile obtained by flow cytometry . The resulting coverage was then averaged across multiple 100 bp windows and a polynomial data smoothing filter ( Savitsky-Golay ) was applied to the individual coverage profiles to filter out noise . Replication timing trep is defined as the time at which 50% of the cells in the population replicated a given region of the genome ( Figure 4—figure supplement 2 ) , which is equivalent to an overall relative coverage of 1 . 5x , since 1x corresponds to an unreplicated region and 2x to a fully replicated one . The replication timing trep was calculated by linearly interpolating the two time points with coverage lower and higher than 1 . 5x and using such interpolation to compute the time corresponding to 1 . 5x coverage . Final trep were then plotted relative to their window genomic coordinates . Unreplicated regions at 45 min were calculated as the sum of all regions with trep >45 min . To find DNA replication origins , the trep profiles along the genome were filtered using a Fourier low-pass filter to remove local minima and then used to find local peaks . Only origins giving rise to long replicons were used to measure fork velocity . Fork velocity was calculated by dividing the distance between the origin and the closest termination site by the time required to replicate the region . Duplicate replication profiles were obtained from two experiments performed on biological replicates . Reproducibility was confirmed with qualitatively and quantitatively comparable results across duplicates . The data obtained from the first duplicate are reported . The reliability of the pipeline was assessed by qualitatively and quantitatively comparing our WT results with previously reported measurements ( Müller et al . , 2014; Raghuraman et al . , 2001 ) . The custom pipeline used for the data analysis is available on GitHub: https://github . com/marcofumasoni/Fumasoni_and_Murray_2019 . We first identified chromosomal locations where fork stalling in the ancestral ctf4∆ cells prevented the completion of DNA replication by 45 min ( fork-stall zones ) . The fork position at 45 min was considered the center of the fork-stall zones , while 5 kb upstream and downstream the fork site were included in the analysis to account for features in the proximity of the fork that could have interfered with its progression . We then examined the sequences within these windows to determine whether various chromosomal features were over- or underrepresented . We considered features that previous studies have found to be associated with hotspots for lesions and sources of genetic instability . We first counted how many times a given feature fell in a fork-stall zone . Then we calculated the expected number of features in these zones based on the total number of features in the genome and the percentage of the genome represented by fork-stall zones . We compared these numbers by χ2 analysis and reported the associated p-values ( Supplementary file 4 ) . The number of tRNA genes , transposable elements , LTRs , ARS elements , snRNA and snoRNA genes and centromeres in the genome were determined using YeastMine ( https://yeastmine . yeastgenome . org/ ) . G4 sequences were obtained from Capra et al . ( 2010 ) . Highly- ( top 5% ) and weakly- ( least 5% ) transcribed genes were identified from the data in Nagalakshmi et al . ( 2008 ) . Rrm3 binding sites and regions with high levels of γH2AX were derived from Azvolinsky et al . ( 2006 ) and Szilard et al . , 2010 , respectively . Site of DNA replication termination were derived from valleys in the tRep signal of the wild type strain ( Figure 4—figure supplement 3 , green signal ) . The tandemly repeated sequences , with a minimal repeat tract of twenty-four bases , were obtained from the tandem-repeat-database ( TRDB; https://tandem . bu . edu/cgi-bin/trdb/trdb . exe ) . Allele frequencies within populations were estimated as in Wildenberg and Murray ( 2014 ) . In brief , chromatograms obtained by sanger sequencing were used to estimate the fraction of mutant alleles in a population at different time points during the evolution . The fraction of mutant alleles in the population was assumed to be the height of the mutant allele peak divided by the height of the mutant allele peak plus the ancestor allele peak . The values from two independent sanger sequencing reactions , obtained by primers lying upstream and downstream the mutations , were averaged to obtain the final ratios . Error bar edges represent the ratios obtained by the two independent sequencing reactions . Values below the approximate background level were assumed to be zero , and values above 95% were assumed to be 100% . Droplet digital PCR was used to detect the amplifications of the fragment containing SCC2 at different time points during evolution . Genomic DNA was prepared and diluted accordingly . Bio-Rad ( RRID:SCR_008426 , Hercules , CA , US ) ddPCR supermix for probes ( no dUTP ) was used to prepare probes specific to SCC2 and the centromere of chromosome IV . A Bio-Rad QX200 Droplet Generator was used to generate droplets containing genomic DNA and probes . The droplet PCR was performed in a Bio-Rad thermocycler and analyzed with a Bio-Rad QX200 Droplet Reader . At least 10 , 000 droplets were acquired for each strain . Analysis was performed on biological duplicates with comparable results . Data obtained with the first duplicate are shown . Droplet analysis was performed with QuantaSoft software ( Bio-Rad ) . SCC2/Centromere ratios were then used to quantify SCC2 copy numbers . Error bars represent Poisson 95% confidence intervals . To estimate the percent of cells carrying the SCC2 amplification within a population we assumed that the allele spreading in the population was a duplication of SCC2 ( as indicated by the EVO5 copy number analysis ) . Values above 95% were assumed to be 100% . | All plants , animals and fungi share a common ancestor , and though they have evolved to become very distinct over billions of years , they all share the essential machinery needed for cells to grow and divide . At the heart of this is the complex interaction of proteins involved in DNA replication , the process of duplicating the genetic material every time a cell divides . DNA replication needs to be done with great care , with error rates as small as one mistake in a billion . Otherwise , mutations can accumulate in the genome , causing problems for long-term survival . Despite the overall principles of DNA replication remaining the same , the underlying mechanisms vary across different organisms . Given the precision and complexity of replicating DNA , it was not clear how the process had evolved mechanistic differences . Fumasoni and Murray set out to answer this by forcing a strain of budding yeast to evolve by removing the gene for an important , but not essential , component of DNA replication . The cells were still able to reproduce , but they were hampered by this mutation . Fumasoni and Murray studied the yeast after it had reproduced for a thousand generations , giving it enough time to acquire new mutations that would allow it to compensate for the initial defects . In eight separate samples , the yeast had made many of the same changes in order to overcome the original mutation . These mutations altered conserved features of DNA replication and the segregation of genetic material and inactivated a third feature that would normally protect the cell against the accumulation of damaged DNA . These findings show how reproducible the evolutionary pathways can be in a controlled , laboratory environment and that cells can evolve quickly after conserved processes in the cell are damaged . The behavior of the mutated yeast mimicked that of cancer cells , which are often struggling to adapt to mutations in their replication machinery . Studying the rapid evolution that follows genetic perturbations could help researchers to better deal with challenges in cancer treatment and the development of antibiotic resistance in bacteria , as well as leading to a deeper understanding of both evolution and cell biology . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"evolutionary",
"biology",
"genetics",
"and",
"genomics"
] | 2020 | The evolutionary plasticity of chromosome metabolism allows adaptation to constitutive DNA replication stress |
Intrinsically disordered proteins ( IDPs ) present a functional paradox because they lack stable tertiary structure , but nonetheless play a central role in signaling , utilizing a process known as allostery . Historically , allostery in structured proteins has been interpreted in terms of propagated structural changes that are induced by effector binding . Thus , it is not clear how IDPs , lacking such well-defined structures , can allosterically affect function . Here , we show a mechanism by which an IDP can allosterically control function by simultaneously tuning transcriptional activation and repression , using a novel strategy that relies on the principle of ‘energetic frustration’ . We demonstrate that human glucocorticoid receptor tunes this signaling in vivo by producing translational isoforms differing only in the length of the disordered region , which modulates the degree of frustration . We expect this frustration-based model of allostery will prove to be generally important in explaining signaling in other IDPs .
A cornerstone of biological regulation is the ability of proteins to tune their particular activities in response to the binding of specific ligands at distinct regulatory sites ( Motlagh et al . , 2014 ) . Historically , such tunability has been explained by the concerted ( Monod et al . , 1965 ) or sequential ( Koshland et al . , 1966 ) models of allosteric regulation , which describe the coupling between binding sites in terms of ligand-induced changes in the average structure of the protein . More recent studies reveal that allostery is not restricted to structured proteins . It is widely observed in intrinsically disordered ( ID ) proteins , polypeptides , or regions therein , that lack stable tertiary structure ( Ferreon et al . , 2013; Garcia-Pino et al . , 2010; Lum et al . , 2012; Motlagh et al . , 2014; Sevcsik et al . , 2011 ) . Moreover , ID regions are hyper-abundant in known allosteric proteins such as transcription factors ( Gronemeyer and Bourguet , 2009; Liu et al . , 2006 ) , suggesting that allostery involving ID sequences may represent a major regulatory paradigm . Despite the existing evidence , however , the mechanism by which ID proteins facilitate allostery is not known . Previously , we developed a mathematical model to show how proteins could use intrinsic disorder to facilitate , and even optimize , allosteric control ( Hilser and Thompson , 2007 ) . This model predicts that coupled folding and binding in different ID domains could produce complex coupling mechanisms that result from the simultaneous tuning of both activating and repressing sub-ensembles within the overall conformational ensemble ( Hilser et al . , 2006 , 2012; Motlagh et al . , 2014 ) , a process of ‘energetic frustration’ akin to the well-known physical concept of ‘geometric frustration’ . In condensed matter physics , ‘geometric frustration’ describes a physical system’s inability to simultaneously minimize the competing interaction energies between its components in mean field theory ( Vannimenus and Toulouse , 1977; Villain , 1977 ) . Frustration theory has been invaluable in understanding magnetic and superconducting systems ( Vannimenus and Toulouse , 1977 ) , circuits ( Wang et al . , 2006 ) , protein folding ( Bryngelson and Wolynes , 1987 ) , and even gene networks ( Krishna et al . , 2009 ) . However , whereas numerous biological networks can utilize multiple components ( e . g . repressors and activators ) to control overall activity , it is not known whether a single gene product could encode tunable activity based on an analogous form of frustration , as theory predicts ( Hilser et al . , 2006; Hilser et al . , 2012; Motlagh et al . , 2014 ) . To investigate the relationship between disorder and allostery and to test whether energetic frustration is at the heart of disorder-mediated allostery , we selected the human glucocorticoid receptor ( GR ) as a model system . The GR is a member of the steroid hormone receptor ( SHR ) family of transcription factors and plays key roles in organ development , metabolite homeostasis , and the responses to stress and inflammation ( Griekspoor et al . , 2007 ) . Three major domains segregate the GR’s primary functions ( Hilser and Thompson , 2011 ) . The DNA-binding domain ( DBD ) and ligand-binding domain ( LBD ) are well-structured and are responsible for interacting with DNA ( i . e . GR response element ) and the steroid hormone ( e . g . cortisol ) , respectively ( Hilser and Thompson , 2011 ) . The N-terminal domain ( NTD ) , which consists of the first 420 amino acids , contains the activation function 1 core region ( i . e . AF1c , GR 187–244 ) , which is required for the recruitment of cofactors necessary for transcriptional activation ( Dahlman-Wright et al . , 1994; Ford et al . , 1997 ) and full transcriptional potency . In contrast to the LBD and DBD , the NTD of GR is intrinsically disordered ( Hilser and Thompson , 2011 ) . Importantly , five active isoforms ( Figure 1 , Inset ) among the total of eight translational isoforms of GR ( Figure 1—figure supplement 1 ) , differing only in the lengths of the ID NTDs have been discovered ( Lu and Cidlowski , 2005 ) . These isoforms differ in their relative activities ( Bender et al . , 2013 ) , tissue distributions ( Lu and Cidlowski , 2005; Lu and Cidlowski , 2006 ) , and regulatory specificities ( Bender et al . , 2013; Cao et al . , 2013; Lu and Cidlowski , 2005; Lu and Cidlowski , 2006 ) . Although the effect of binding either different steroid molecules ( Pandit et al . , 2002; Pfaff and Fletterick , 2010 ) or different DNA sequences ( Meijsing et al . , 2009 ) is known to produce a variety of structural changes within their respective binding domains , how such binding events are differentially propagated to the ID NTD of each isoform , and subsequently translated into functional changes , is not known . Fundamentally , it is not clear whether and , if so , how structured domains like the DBD can both receive and transmit allosteric signals to disordered domains like the NTD of GR .
To obtain insight into how allostery tunes GR function , DNA binding and cell-based functional studies were performed on constitutively active ( i . e . steroid-independent ) versions ( Chen et al . , 1997 ) of human GR translational isoforms that lack the C-terminal LBD ( Figure 1 ) . The similarity between the relative transcriptional activities of the different two-domain isoforms studied here and those of the full-length three-domain isoforms studied previously ( Bender et al . , 2013; Lu and Cidlowski , 2005 ) ( Figure 1 and Figure 1—figure supplement 1a and b'' ) , suggests that although the LBD affects the magnitude of activity enhancement ( Godowski et al . , 1987; Hollenberg and Evans , 1988 ) , it does not appear to qualitatively impact the communication between the DBD and the NTD in each isoform . Several features of the activities and DNA-binding properties of the different isoforms are noteworthy . First , the affinities of the isoforms for DNA vary , despite having identical DBDs , indicating that the NTD of each isoform differentially communicates with its respective DBD ( Figure 1 and Figure 1—figure supplement 1c–e ) . Second , the GR C3 isoform ( i . e . 98–525 ) is almost five times more active than the full-length GR A isoform ( i . e . 1–525 ) , indicating that residues 1–97 somehow negatively regulate the activity of the remaining NTD residues , which contain the functionally important AF1c region ( Figure 1 ) . For this reason , we represent the full-length NTD as being composed of two distinct domains , a functional domain ( F-domain ) , and a regulatory domain ( R-domain ) ( Figure 1 , Inset ) , which were experimentally shown to be unfavorably coupled to each other by both osmolyte ( i . e . TMAO ) induced folding and protease sensitivity analyses ( Figure 2a and b ) ( Li et al . , 2012 ) . Conversely , in the same in vitro folding and protease sensitivity experiments the DBD appears to stabilize the folded form of the F-domain ( Figure 2a and b ) . Importantly , the thermodynamic stabilization of the F-domain conferred by the DBD is accompanied by a dramatic increase in activity , as determined from cell-based transcriptional assays that compare the activity of the GR F-DBD construct ( C3 isoform ) with the non-natural chimera that tethers the DBD from the yeast Gal4 transcription factor to the F-domain ( Figure 2c ) . The thermodynamic and activity results suggest that competing factors within GR determine the overall stability ( i . e . the ΔG of folding ) and transcriptional activity of the AF1c region of the F-domain . The coupling between the R-domain and the DBD was further evaluated by use of competitive transfection assays to estimate the DNA-binding affinity of various constructs that connect the DBD and the R-domain of each isoform . In order to only measure the coupling between the R-domain and the DBD , and to avoid the convoluting effects associated with the coupling between the F-domain and the R-domain and DBD , constructs were generated that utilized a series of flexible linkers connecting the R-domain to the DBD ( Figure 3a ) . As Figure 3b reveals , inclusion of the R-domain residues 1–85 , naturally present in GR’s A and B isoforms , significantly increased the DNA-binding affinity of the DBD , while the shorter length R-domains in the C isoforms show no effect . Such increases were not observed for non-natural chimeric constructs that linked the various R-domains to the yeast transcription factor Gal4 DBD ( Kraulis et al . , 1992 ) ( Figure 3b ) . In addition , the fact that different length linkers ( Figure 3c ) give similar results indicates that the linker is , as intended , functionally inert , and serving as a tether that simply connects the R-domain to the DBD . These results support the notion that the R-domain affects DNA binding by the DBD specifically through stabilization of a high-affinity state of the DBD , and is not simply a consequence of a direct interaction between the R-domain ( or the linker ) and the DNA . In other words , our results indicate the R-domain allosterically affects DNA binding of the DBD , serving as a positive intramolecular allosteric effector . The analysis of the thermodynamic couplings in Figures 2 and 3 point to the paradoxical result whereby the binding of DNA to the DBD simultaneously produces two opposing effects on the F-domain . It has long been known that DNA binding to GR DBD stabilizes the DBD ( Lefstin and Yamamoto , 1998 ) . As a consequence of its direct positive coupling to the F-domain ( Figure 2 ) , DBD binding to DNA stabilizes the folded form of the F-domain and therefore promotes activation of transcription ( Figure 4a , counterclockwise green arrow ) . However , as Figure 3 indicates , there is also positive coupling between the DBD and the R-domain . But because of the negative coupling between the R- and F-domains ( Figure 2a and b ) , the same DNA binding ( to the DBD ) that stabilizes/activates the F-domain simultaneously , through stabilization of the R-domain , destabilizes the folded form of the F-domain , promoting repression of transcription ( Figure 4a , red inhibitory semicircle ) , a process that resembles geometric frustration ( Vannimenus and Toulouse , 1977; Villain , 1977 ) . Geometric frustration originates in bi-stable systems wherein competing thermodynamic couplings interact such that no single state acquires significant stability so as to dominate the ensemble probabilities ( Krishna et al . , 2009; Vannimenus and Toulouse , 1977; Villain , 1977 ) . Within the context of the three domains of GR studied here , eight different configurations of coupling energies ( i . e . Figure 4bi–viii ) represent all possible combinations of positive ( + ) and negative ( − ) coupling energies between domains . For each case , a positive interaction energy signifies that a stabilization of one domain would result in a stabilization of the second domain , whereas negative coupling would produce the opposite effect . As Figure 4b reveals , frustration results when the ‘direct’ impact of stabilization of the DBD on the F-domain is opposite in sign to the indirect impact ( i . e . the impact that is mediated through the R-domain ) . Such is the case when one or all three of the inter-domain interactions is/are negative ( Hilser et al . , 2012; Motlagh and Hilser , 2012; Motlagh et al . , 2014 ) . Of the possible configurations that are predicted to produce frustration ( Figure 4b upper ) , GR clearly conforms to case ii , wherein the DBD is positively coupled to the F-domain serving to increase its activity ( Figure 2a–c ) . However , because of the negative coupling between the R-domain and the F-domain ( Figure 2a and b ) , and the positive coupling between the R-domain and the DBD ( Figure 3 ) , the DBD is ultimately also negatively coupled to the F-domain . The net effect of DNA binding on GR transcriptional activity is thus a balance between the strengths of these coupling energies , which could differ among translational isoforms . The results clearly demonstrate that two opposing control mechanisms are at play , and that the classic deterministic models of allostery ( Koshland et al . , 1966; Monod et al . , 1965 ) are insufficient to capture the probabilistic nature of this mechanism . To highlight the analogy with geometric frustration , while simultaneously distinguishing this biological phenomenon from the condensed matter physics model , we term this phenomenon ‘energetic frustration’ . The simplest models for geometric frustration quantify the total energy of a system of spins within a magnet , where the energy of an interacting spin pair of nuclei i and j takes the form Eint = JijSiSj . In this formalism , J is the coupling energy and S is the spin state ( i . e . ‘up +1’ or ‘down −1’ ) . The interaction energy between protein domains i and j can be written in a similar way , Eint = JijS*iS*j , with an interdomain coupling energy and an accounting for the states of the domains . One key difference , with respect to geometric frustration , is that in energetic frustration the sign of Eint is determined solely by the sign of J , which is fixed by the physicochemical nature of the interdomain interaction . Also , the domain state is conditional on whether the protein domain is folded or unfolded , a folded state resulting in a + 1 value for S* and an unfolded state resulting in a value of 0 . Only in the case in which both domains are folded will the coupling energy contribute to the system . To illuminate how the opposing allosteric mechanisms are manifested in our example IDP , the GR , a quantitative characterization of the allosteric coupling was implemented using the previously developed ensemble allosteric model ( EAM ) ( Hilser and Thompson , 2007; Hilser et al . , 2012 ) . An ensemble of states was constructed for each isoform , enumerating all possible combinations of the DBD being in the high-affinity or low-affinity states , and the R and F-domains being in their active ( folded ) or inactive ( unfolded ) states ( see Figure 5a , for the A and C3 isoforms ) . Our choice of model is justified because as shown previously ( Li et al . , 2012 ) the R- and F-domains can fold to globular protein-like structures ( Figure 2—figure supplement 1b ) , consistent with the notion of coupled folding and binding for IDPs , where the folded conformation is the active form ( Dyson and Wright , 2002 ) . In the context of the EAM , the probability of any state is determined by the intrinsic stability of each domain , ∆GR , ∆GF and ∆GD ( the stability of each domain as it would be in isolation ) , as well as the coupling free energies between each domain , ΔgR−F , ΔgR−D and ΔgF−D ( Hilser and Thompson , 2007; Hilser et al . , 2012 ) . For example , in the full length A isoform ( composed of the R-domain , the F-domain , and the DBD ) and the most active C3 isoform ( composed only of the F-domain and the DBD ) , the EAM produces 8 and 4 states in their respective ensembles , representing all combinations ( Figure 5a ) . In the EAM , the experimentally observed transcriptional activity is represented by the summed probabilities of states whose folded F-domain co-occurs with the high-affinity DBD conformation . Similarly , DNA-binding affinity is represented by the summed probabilities of states wherein the DBD is in the high-affinity conformation . Using measurements of transcriptional activities and binding affinities of the five isoforms ( Figure 1 and Figure 1—figure supplement 1a–d ) , measurements of relative binding affinities of RA-linker-DBD and Linker-DBD constructs ( Figure 3b ) , and measurements of conformational stabilities ( Figure 2a and Figure 2—figure supplement 1a ) as constraints , quantitative estimates of the stabilities and coupling energies for each domain were obtained through unbiased comprehensive searches of parameter space ( Figure 5—figure supplement 1a ) . Importantly , the maximum likelihood parameters ( shown in Figure 5—figure supplement 1e ) faithfully reproduce both the relative affinities and transcriptional activities for all five isoforms ( Figure 5b ) . The search results demonstrated clear maximum likelihoods for each thermodynamic parameter ( Figure 5—figure supplement 1e ) , indicative of the qualitative correctness of the activating and repressing scheme outlined in Figure 4a . In particular , in all cases , the signs of the coupling energies between domains are preserved , that is , both ΔgR−D and ΔgF−D are positive while ΔgR−F is negative ( Figure 5—figure supplement 1e ) , demonstrating the robustness and validity of the opposing frustration-based control mechanism underlying allostery in GR . To further test the model , we sought to identify and ablate the repression component of the mechanism and quantitatively evaluated the impact on the natural GR isoforms . To do this , we targeted the interaction between the R-domain of the A isoform ( i . e . RA , residues 1–97 ) and the DBD ( Figure 1 inset ) , by determining the impact of point mutations in the DBD on DNA binding affinity in the presence and absence of the tethered R-domain . For constructs containing only the DBD and the R-domain , DBD mutations could perturb either the stability of the DBD , the coupling between the DBD and the R-domain , or both . Thus , by expressing the mutant forms as linker-DBD and RA-linker-DBD constructs , using an inert linker to substitute for the F-domain ( Figure 6a ) , the impact of the mutations could be clearly discerned . Mutations that affect stability of the DBD are predicted to affect the DNA-binding affinity of both constructs ( Figure 6b , left ) . However , mutations that affect the coupling between the R-domain and the DBD are predicted to impact the DNA-binding affinity of only the RA-linker-DBD construct , leaving the activity of the linker-DBD construct unaffected ( Figure 6b , right ) . Screening of conserved surface residues on the DBD ( Figure 6—figure supplement 1a ) , which did not significantly affect DBD stability or the coupling between the F-domain and DBD ( Figure 6—figure supplement 1b ) , revealed only three positions ( i . e . C431 , V435 , and L436 ) that exhibited the expected signature of DBD residues that mediate coupling to the R-domain ( Figure 6c and Figure 6—figure supplement 1c and d ) . Consistent with these positions exerting their effects through a common mechanism , all three residues mapped to a conserved contiguous surface on the DBD ( Figure 6d ) . To qualitatively and quantitatively test the frustration-based control mechanism , the triple mutation ( C431Y/V435A/L436A ) was introduced into the full-length GR A isoform . Because the mutation should decrease the stabilizing effect of binding DNA on the R-domain , which in turn , should lower the destabilizing effect on the F-domain , the model predicts the counter-intuitive result whereby the activity of the triple mutant should increase , while the DBD-DNA binding affinity should decrease ( Figure 6e Model ) . Importantly , such a prediction is the direct result of the frustration in the system and would represent a compelling argument for the competing energetic couplings shown in Figure 4a . True to the prediction ( Figure 6e Experiment ) , the effect of the triple mutant is a decrease in affinity for DNA while concomitantly increasing the transcriptional activity . This result is particularly important because while the modulatory role of the residues we refer to as the R-domain has been known ( Bender et al . , 2013 ) , previous interpretations attributed the effect to simple steric occlusion ( Bender et al . , 2013 ) . Our results unequivocally demonstrate that the R-domain not only negatively affects the F-domain ( Figure 2a and b ) but also positively affects the wild-type DBD ( Figure 3 ) , increasing the affinity of the DBD for DNA ( Figure 3 ) , and it does so in a manner that is directly related to the stabilities and coupling energies in the system . Furthermore , the facts that the three residues implicated in the coupling were independently identified through mutational analysis ( Figure 6c and Figure 6—figure supplement 1b–d ) , but nonetheless mapped to a conserved contiguous surface on the DBD ( Figure 6d ) , strongly supports a model whereby these residues affect the coupling between the DBD and the R-domain through a common mechanism involving direct interactions between domains . Further supporting this notion , titration of the DBD of the C3 isoform and the C3 triple mutant ( C431Y&V435A&L436A ) with the R-domain ( expressed in trans ) , using the luciferase assay as a reporter , clearly shows a greater concentration dependence of activity for the wild-type C3 isoform over the triple mutant ( Figure 6f ) . This result indicates that the R-domain can exert its stabilization effect on the DBD through mass action , suggestive of a direct interaction involving residues C431 , V435 , and L436 . We note that although the combination of comparatively weak coupling energies ( based on the maximum likelihood parameter estimation for ∆gR-D; Fig . S3e ) and poor solubility of both the DBD and R-domains precluded attempts to structurally characterize the interaction using NMR , such limitations have not adversely affected our efforts to rationally intervene . Indeed , Figure 6 reveals that the coupling between the R-domain and the DBD , identified in isolation ( Figure 3 ) , could not only be leveraged into a comprehensive frustration-based model that quantitatively captures the relative binding and activity of all the GR isoforms ( Figure 5b ) , but could also be rationally manipulated , and the opposing consequences on DNA binding and transcriptional activity predictably altered ( Figure 6e ) . Thus , although physical basis of the couplings between the domains awaits future studies , the fact that GR has evolved the ability to produce isoforms that utilize different degrees of energetic frustration opens entirely new avenues for investigating regulation in IDPs . The regulatory role of the isoform-specific ID R-domain in GR is especially important in light of the observation that the DNA sequences coding for ID regions are enriched in splice sites ( Buljan et al . , 2012 ) , leading to a high degree of variability in the ID regions of the resultant proteins . The studies presented here provide a functional explanation . Alternative splicing , like the alternative translation start sites of GR described here , can produce proteins with different degrees of frustration in their ID regions , and thus differing activities . In addition , isoforms may also have different combinations of post-translational modification sites , which are also enriched in ID segments ( Bah and Forman-Kay , 2016 ) . By combining regulatory elements possessing different stabilities with different numbers and types of modification sites , ID proteins can potentially regulate not only the efficiency of the resultant protein , as shown here for GR , but also how that activity can be tuned by different types of modifications ( Motlagh et al . , 2014 ) . We have shown that GR produces different isoforms , which have different DNA-binding affinities and transcriptional activities that are uncorrelated to each other . Our results show that this uncorrelated behavior is facilitated through ‘energetic frustration’ , wherein opposing energetic couplings compete to modulate the overall response . Recent studies reveal that in addition to being facilitated by structured proteins , allostery can also be mediated by dynamic and even ID proteins ( Freiburger et al . , 2011; Motlagh et al . , 2014; Petit et al . , 2009; Popovych et al . , 2006; Tzeng and Kalodimos , 2009 , 2012 ) . Within these ubiquitous ID systems , significant heterogeneity , both in the apo and ligand-bound states , produces ensembles that cannot be treated using classic deterministic or structure-based allosteric models . Instead , extension of these classic models to account for positive and negative couplings between different regions provides a framework for understanding not only how ID sequences communicate with other structured and ID sequences , but also how such heterogeneity can produce complex regulatory strategies , such as the frustration-based mechanism identified here .
DNA 2 . 0 ( Menlo Park , CA ) synthesized the plasmid used to express the A isoform of human GR in U-2 OS cells . The construct sequence was codon optimized and inserted into the PJ603 mammalian expression vector under CMV promoter control . Plasmids to express isoforms B , C1 , C2 , C3 , D1 , D2 , D3 and DBD were made by inserting the codons for each respective isoform amplified from A isoform vector into the NheI and XhoI sites of the PJ603 vector . The GR F-Gal4 DBD plasmid was also produced by DNA 2 . 0 using the PJ603 plasmid backbone . The RA-linker-DBD ( equivalent to RA-11aa linker-DBD ) /RA-20aa linker-DBD plasmid was made using a PCR that deleted the codons for GR 98–420 from the GR A isoform plasmid , then digesting with BamHI and KpnI , and ligating the sticky ends to an oligo coding for the 11aa linker GTGGSGGSGGS/20 aa linker GTGGSGGSGGSGGSGGSGGS . Plasmids for RAΔ86–97-linker-DBD , RAΔ27–97-linker-DBD , RB -linker-DBD , RC1 -linker -DBD , RC2 -linker and linker-DBD were made by inserting the codons for each construct amplified from RA-11aa linker-DBD plasmid into the NheI and XhoI site of PJ603 vector . For the R domain-nuclear localization sequence-Flag construct , a GeneBlock was synthesized by IDT ( Coralville , IA ) to contain the GR R-region ( 1–97 a . a . ) , a four amino acid GSGS linker , the GR nuclear localization sequence ( 488–505 a . a . ) , a GSGSGS linker , and the FLAG tag ( DYKDDDDK ) . This GeneBlock was restriction digested with NheI and XhoI , then inserted into the pJ603 vector ( DNA2 . 0 ) . The FLAG tag was used for immunostaining to verify nuclear localization ( data not shown ) . All the point mutations on GR constructs were made by site directed mutagenesis ( Hemsley et al . , 1989 ) . To measure transcriptional activity in the dual luciferase reporter assay , two tandem full length GREs ( 5’-aattcAGAACAggaTGTTCTgagatccgtagcAGAACAggaTGTTCTgagatccgtagcg −3’ ) were cloned into the EcoRI and BamHI sites in the promoter region of pGluc-miniTK vector ( NEB , Ipswitch , MA ) , which expresses a secreted Gaussia luciferase ( Tannous et al . , 2005 ) . For the competitive transfection assay , four tandem half-site GREs ( 5’-aattcAGAACAggagagatcgtagc AGAACAggaagatccgtagcAGAACAggagagatccgtagcAGAACAggaagatccgtagcg-3’ ) were cloned into the promoter region of pGluc-miniTK vector . The pCluc-miniTK2 vector ( NEB ) , expressing a secreted Cypridina luciferase ( Nakajima et al . , 2004 ) independent of GR regulation , was utilized as an internal control in the transfection to account for differences in cell density and transfection efficiency in each well . DNA 2 . 0 also synthesized the plasmid for bacterial expression of the two-domain GR construct . Codons for the two-domain construct of A isoform was optimized for bacterial cell expression and inserted into the PJ411 expression vector under T7 promoter control . Plasmids to express isoforms B , C1 , C2 , C3 , D1 , D2 and D3 in E . coli were made by inserting the codons for each respective isoform amplified from A isoform plasmid into the NdeI and XhoI sites of the PJ411 vector . Expression , purification and storage of the two-domain constructs for the eight GR translational isoforms were the same as for the single N-terminal domain construct , as described previously ( Li et al . , 2012 ) , except for the following modifications in the purification steps . The lysis buffer was composed of 100 mM NaH2PO4 , 10 mM Tris , 500 mM NaCl , 20 mM imidazole , pH 8 . 0 . The wash buffer was the lysis buffer containing 60 mM imidazole , and the elution buffer was the lysis buffer containing 200 mM imidazole . Fluorescent Oligos containing half site GRE ( 5’−6-FAM gcgcAGAACAggacgcg-3’ and 5’-cgcgtccTGTTCTgcgc-3’ ) were synthesized by IDT with HPLC grade purification and annealed with each other to get double stranded 6FAM-labeled half site GRE . The binding experiments of GR two domain constructs with the half site GRE were carried out in the following buffer: 10 mM HEPES ( pH7 . 6 ) , 80 mM NaCl , 1 mM EDTA , 5 mM MgCl2 , 1 mM DTT , 200 ug/mL BSA and 5 μM double strand control oligo ( 5’-GCGCCATATGATACGCG-3’ ) . For each data point , 25 nM 6-FAM-labeled half site GRE was incubated with from 0 μM to 10 μM GR two-domain construct at 22°C for 30 min . Fluorescence anisotropy was measured using an Aviv ATF 105 fluorometer equipped with polarizers . A ‘sub micro’ fluorometer cell with 150 μL of solution was allowed to rest at 22°C ( Santa Cells ) for 2 min to allow for temperature stabilization and then excited at 495 nm . Anisotropy at 521 nm was recorded as a function of GR construct concentration and fitted with a single-site-binding model . TMAO-induced protein folding transitions were described previously ( Li et al . , 2012 ) . U-2 OS cells ( American Type Culture Collection , Manassas , VA ) were maintained in modified McCoy's 5a medium ( Corning Cellgro , Tewksbury , MA ) supplemented with 10% fetal bovine serum and 100 U/mL penicillin and 100 µg/mL streptomycin . To transfect U-2 OS cells at about 80–90% confluence , X-tremeGENE HP DNA transfection reagent ( Roche , Indianapolis , IN ) was used at 2 μl per 1 μg DNA according to the manufacturer’s manual . For the transcriptional activity dosage curve , 40 ng of pGluc-miniTK vector with two tandem full length GREs cloned in the promoter region , 40 ng of pCluc-miniTK2 and up to 5 ng ( saturating ) of GR expression vector were co-transfected into U-2 OS cells on 96-well plates . For the competitive transfection assay , 40 ng of pGluc-miniTK vector with four tandem half site GREs cloned in the promoter region , 40 ng of pCluc-miniTK2 , 3 ng of expression vector for C3 isoform , and up to 16 ng of plasmid coding for one of the competitors were co-transfected into U-2 OS cells on 96-well plates . For titration of the C3 isoform wild type and C3 C431Y&V435A&L436A mutant with the R domain-nuclear localization sequence-Flag construct , the method was the same as the competitive transfection assay described above , except up to 12 ng of the R domain-nuclear localization sequence-Flag construct plasmid was used in the titration . After 48 hr , Gaussia Luciferase activity and Cypridina Luciferase activity were measured with the BioLux Gaussia Luciferase Assay Kit ( NEB ) and the BioLux Cypridina Luciferase Assay Kit ( NEB ) , respectively , on a TriStar LB 942 Multidetection Microplate Reader ( Berthold Technologies GmbH & Co . KG , Bad Wildbad , Germany ) , according to the manufacturer’s protocols . In each experiment , the Gaussia luciferase activity ( normalized by the Cypridina luciferase activity ) was measured in triplicate and averaged . The dosage and competitive transfection curves were fitted with dose response curves using Origin . U-2 OS cells were plated on 6-well plates at a density of 5 × 105 cells per well . After 18 hr , 50 ng of GR expression vector and 450 ng of salmon sperm DNA ( Invitrogen , Carlsbad , CA , transfection boosting reagent ) were transfected into each well with X-tremeGENE HP DNA transfection reagent ( Roche ) , following the manufacturer’s protocol . The medium was changed once 24 hr post transfection . After 48 hr , the cells were scraped from each well with PBS , and pelleted by centrifuging at 1500 rpm . For lysis , 50 μL of lysis buffer ( 8 M urea , 20 mM Tris-HCl , 500 mM NaCl , 1 mM Na2EDTA , 1 mM EGTA , 1% Triton , 2 . 5 mM sodium pyrophosphate , 1 mM beta-glycerophosphate , 1 mM Na3VO4 and 1 µg/ml leupeptin , pH 7 . 5 ) was added to each cell pellet . To reduce the viscosity , cells were passed through a 26G 3/8’’ syringe needle 10 times and then the cell lysate was centrifuged at 14000 rpm for 30 min . Supernatant was collected and the total protein concentration was measured by Bradford assay ( Bio-Rad , Hercules , CA ) . In each well of a 4–15% Mini-PROTEAN TGX Precast Gel ( Bio-Rad ) , 5 µg of total protein was loaded , and separated in Tris/Glycine/SDS gel running buffer . Transfer of the protein from the SDS page gel to PVDF film was done in the transfer buffer ( 25 mM Tris-HCl , pH 8 . 3 , 192 mM glycine , 20% methanol ) under 120V for 15 min . After blocking in 5% nonfat milk in PBS with 0 . 1% Tween-20 ( PBST ) for 1 hr , the PVDF film was then incubated at 4˚C overnight with 10000 fold diluted primary antibody for GR ( BD Transduction Laboratories , #611226 , San Jose , CA ) or p150glued ( BD Transduction Laboratories , #610473 ) , which served as loading control . Both antibodies were diluted into the same 5% nonfat milk in PBST . The next morning , after washing with PBST buffer for three times , the PVDF film was incubated in the 20000 fold diluted HRP-linked anti-mouse IgG ( GE healthcare , NA931 , Chicago , IL ) , also in the 5% nonfat milk in PBST . The detection was done with Amersham ECL Prime Western blotting reagent ( GE heathcare , RPN2232 ) and autoradiography film ( Denville Scientific ) . U-2 OS cells were plated on 6-well plates with 15 mm round German coverslips . All the culture and transfection procedures are the same as done for the cells for western blot experiments . After 48 hr , cells were rinsed with PBSM ( PBS with 2 mM MgCl2 ) three times , and fixed with 4% paraformaldehyde in PBSM at room temperature for 10 min . Afterwards , each coverslip was rinsed with PBSM three times again , and quenched with 50 mM NH4Cl in PBSM . Then the slide was placed in PBSTB ( PBS with 0 . 1% Triton X-100 , 1% BSA ) for 30 min at room temperature to permeabilize cells and block nonspecific binding . Thereafter , the slide was incubated for 1 hr at room temperature with primary rabbit antibody for GR ( cell signaling , #3660 ) , which was 5000 fold diluted in PBSTB . Then the slide was washed three times with PBSM and incubated for 30 min at room temperature in the dark with the Alexa Fluor 488 Goat Anti-Rabbit IgG ( Invitrogen , Carlsbad , CA ) , which was 600 fold diluted in PBSTB . Next the slide was incubated for 10 min at room temperature in the dark in PBSM with 0 . 2 μg/mL DAPI ( Invitrogen ) and 5 unit/mL Rhodamine Phalloidin ( Invitrogen ) to stain the nuclei and F-actin respectively . After that , each slide was washed with PBSM twice and mounted onto a microscope slide with Fluoromount ( Sigma , St . Louis , MO ) , and kept in the dark for drying . Images were taken with an inverted light microscope ( Axiovert 200 , JHU Integrated Imaging Center ) . All images were taken the same day using the same gain , exposure times , and filter configurations ( DAPI , FITC , and Texas Red filters ) . The images were analyzed using ImageJ ( Staal et al . , 2004 ) . The Ensemble Allosteric Model ( EAM ) and its usage have been described in detail previously ( Hilser and Thompson , 2007; Hilser et al . , 2012; Motlagh et al . , 2014 ) . | Proteins carry out most of the key tasks inside cells . To perform these roles , proteins must fold up to form complex three-dimensional structures . Researchers used to think that the useful parts of proteins all had set structures . However , we now know that ‘disordered’ proteins with variable structures are common and disordered parts of proteins can have vital roles . In a process called allosteric regulation , regulator molecules can increase or decrease the activity of a protein by binding to it . This binding was thought to work by changing the structure of the protein , but it was not clear how this works in disordered proteins . To investigate , Li et al . studied a disordered protein called glucocorticoid receptor , and found that disordered regions can have opposing effects on other regions of the protein . This creates a ‘tug-of-war’ that Li et al . term “energetic frustration” , whereby the activity of the protein results from the combination of the opposing interactions . Further investigation revealed that the glucorticoid receptor produces different versions of itself that have different degrees of energetic frustration , which alters how effectively the proteins perform their tasks . This means that the protein can regulate its own activity even in the absence of binding to regulator molecules . The concept of energetic frustration could enhance our understanding of the many different proteins that contain disordered regions . Eventually , this knowledge could be used to develop drugs that alter the activity of these proteins and so could form part of treatments for a wide range of conditions including autoimmune diseases ( such as rheumatoid arthritis and lupus ) , cancers , and organ rejection for transplant patients . The results presented by Li et al . suggest where more research is needed to achieve this goal . For example , we need to understand more about the stability of disordered protein regions , and to identify which surfaces of the proteins interact with each other . | [
"Abstract",
"Introduction",
"Results",
"and",
"discussion",
"Materials",
"and",
"methods"
] | [
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2017 | Genetically tunable frustration controls allostery in an intrinsically disordered transcription factor |
Astrocytes have emerged as integral partners with neurons in regulating synapse formation and function , but the mechanisms that mediate these interactions are not well understood . Here , we show that Sonic hedgehog ( Shh ) signaling in mature astrocytes is required for establishing structural organization and remodeling of cortical synapses in a cell type-specific manner . In the postnatal cortex , Shh signaling is active in a subpopulation of mature astrocytes localized primarily in deep cortical layers . Selective disruption of Shh signaling in astrocytes produces a dramatic increase in synapse number specifically on layer V apical dendrites that emerges during adolescence and persists into adulthood . Dynamic turnover of dendritic spines is impaired in mutant mice and is accompanied by an increase in neuronal excitability and a reduction of the glial-specific , inward-rectifying K+ channel Kir4 . 1 . These data identify a critical role for Shh signaling in astrocyte-mediated modulation of neuronal activity required for sculpting synapses .
The organization of synapses into the appropriate number and distribution occurs through a process of robust synapse addition followed by a period of refinement during which excess synapses are eliminated . Failure to establish or maintain appropriate synaptic organization is a hallmark of many neurodevelopmental disorders ( Penzes et al . , 2011 ) . Considerable evidence now shows that , together with neurons , astrocytes are critical regulators of synaptic connectivity and function ( Allen and Eroglu , 2017; Adamsky et al . , 2018; Risher et al . , 2014 ) . Astrocytes interact intimately with synapses to regulate their formation , maturation , and function , and a growing number of astrocyte-secreted proteins that directly mediate synapse formation and elimination have been identified ( Allen et al . , 2012; Kucukdereli et al . , 2011; Christopherson et al . , 2005; Blanco-Suarez et al . , 2018; Chung et al . , 2015 ) . In addition , astrocytes regulate concentrations of K+ and glutamate in the extracellular space , thereby modulating neuronal activity ( Olsen and Sontheimer , 2008 ) . Nevertheless , despite the remarkable progress in our understanding of the essential role for astrocytes in regulating synaptic formation and function , the underlying signaling programs mediating astrocyte-dependent regulation of synapse organization remain poorly understood . The molecular signaling pathway Sonic hedgehog ( Shh ) governs a broad array of neurodevelopmental processes in the vertebrate embryo , including morphogenesis , cell proliferation and specification , and axon pathfinding ( Fuccillo et al . , 2006; Ruiz i Altaba et al . , 2002 ) . However , Shh activity persists in multiple cell populations in the postnatal and adult CNS , including progenitor cells , as well as in differentiated neurons and astrocytes ( Traiffort et al . , 2010; Ahn and Joyner , 2005; Ihrie et al . , 2011; Harwell et al . , 2012; Garcia et al . , 2010 ) , where novel and unexpected roles for Shh activity are emerging ( Garcia et al . , 2018 ) . Following injury , Shh has been shown to mitigate inflammation ( Alvarez et al . , 2011; Allahyari et al . , 2019 ) , and in the cerebellum , Shh derived from Purkinje neurons instructs phenotypic properties of mature Bergmann glia ( Farmer et al . , 2016 ) . In the postnatal cortex , Shh is required for establishing local circuits between two distinct projection neuron populations ( Harwell et al . , 2012 ) . Shh produced by layer V neurons guides the formation of synaptic connections to its layer II/III presynaptic partners , which transduce the Shh signal through non-canonical , Gli-independent mechanisms . We have previously shown that Shh signaling is also active in a discrete subpopulation of cortical astrocytes ( Garcia et al . , 2010 ) , suggesting that Shh signaling mediates both homotypic and heterotypic cellular interactions . Astrocytes engaging in Shh activity are identified by expression of Gli1 , a transcriptional effector of canonical Shh signaling ( Fuccillo et al . , 2006 ) . Whether Shh signaling in cortical astrocytes plays a role in synaptic organization of neurons is not known . In this study , we examined the organization and dynamics of dendritic spines on cortical neurons following selective disruption of Shh signaling in astrocytes . Dendritic spines are the structural hosts of most excitatory synapses and play an important role in the organization and function of neural circuits . We show that deep layer neurons exhibit long-lasting aberrations in the density and turnover of dendritic spines that emerge during postnatal development following selective disruption of Shh signaling in astrocytes . These perturbations in synaptic organization are not observed in upper layer neurons where Gli1 astrocytes are relatively sparse . Chronic in vivo imaging of dendritic spines reveals that mutant mice exhibit lower rates of spine turnover , suggesting impaired structural plasticity . In addition , these mice show a pronounced deficit in expression of the glial-specific inward-rectifying K+ channel , Kir4 . 1 , as well as an increase in excitability of cortical neurons . Taken together , these data demonstrate that astrocytes act as key modulators of neural activity and structure during postnatal development in a Shh-dependent manner and further establish Shh signaling as a fundamental mediator of synaptic connectivity .
In the mature cortex , a subpopulation of astrocytes express the transcription factor Gli1 , indicating active Shh signaling ( Garcia et al . , 2010 ) . Notably , the distribution of Gli1 astrocytes throughout the cortex is non-uniform , showing a laminar-specific pattern ( Figure 1 ) . We analyzed the laminar distribution of Gli1 astrocytes in adult Gli1CreER/+;Ai14 mice , in which tamoxifen administration promotes Cre-mediated recombination of the fluorescent tdTomato reporter protein , permanently marking Gli1-expressing cells . Adult Gli1CreER/+;Ai14 mice received tamoxifen over three days , and their brains were analyzed two – three weeks later . In our previous study , we observed weak expression of βGal reporter protein in Gli1CreER/+;R26lacZ/lacZ mice at time points earlier than 4 weeks after tamoxifen ( Garcia et al . , 2010 ) . In contrast , expression of the tdTomato reporter protein in the Ai14 reporter line shows robust expression as early as two weeks , and a comparable number , distribution and intensity of reporter-positive cells between two and four weeks ( Allahyari et al . , 2019 ) . For these studies , we therefore used a two – three week chase after tamoxifen . The vast majority of marked cells were observed within layers IV and V , with few marked cells observed in layers II/III or VI ( Figure 1 ) . We performed immunostaining with the pan-astrocytic marker S100β and quantified the fraction of marked astrocytes in cortical layers ( Figure 1 ) . This analysis showed that 44% of astrocytes in layer IV express Gli1 ( Figure 1 ) , while 36% of astrocytes in layer V express Gli1 . Interestingly , the distribution of Gli1 astrocytes in layer V was not homogenous , showing an enrichment of marked cells in layer Vb , and a distinctive paucity of marked cells in layer Va ( Figure 1 ) , consistent with the localization of Shh producing neurons in layer Vb ( Harwell et al . , 2012; Garcia et al . , 2010 ) . We analyzed the fraction of Gli1 astrocytes in each sublayer of layer V and found that 75% of marked cells were found in layer Vb whereas only 25% were found in layer Va . Layers II/III and VI showed the lowest fractions of marked astrocytes , with only 11% and 22% of S100β cells expressing tomato , respectively . These data demonstrate that Gli1 astrocytes are preferentially localized in deep cortical layers and suggest their interactions with local neurons may differ from that of Gli1-negative astrocytes in superficial layers . Transduction of canonical Shh signaling begins with binding of Shh to its transmembrane receptor , Patched1 ( Ptch ) , relieving inhibition of a second , obligatory co-receptor Smoothened ( Smo ) . To investigate the role of Shh signaling in astrocyte function , we performed conditional knockout of Smo selectively in astrocytes using Gfap-Cre transgenic mice in which Cre expression is regulated by the full-length mouse Gfap promoter ( Garcia et al . , 2004 ) ( Gfap Smo CKO ) . Because Gfap-mediated Cre recombination targets astrocyte progenitors expressing GFAP which are present at birth , this is an effective tool for targeting the cortical astrocyte population for selective gene deletion . We validated the Cre-mediated recombination pattern in Gfap-Cre mice by crossing them with Ai14 reporter mice ( Gfap-Cre;Ai14 ) , that express the tdTomato reporter protein . Although individual floxed alleles possess distinct recombination efficiencies , owing to specific characteristics of each allele including distance between loxP sites or their accessibility due to chromatin structure ( Vooijs et al . , 2001 ) , analysis of recombination in a reporter line provides a useful approximation of the expression pattern of a given Cre driver . We performed single cell analysis of the identity of recombined cells . The vast majority of tomato-positive cells showed a bushy morphology , typical of protoplasmic astrocytes ( Figure 2—figure supplement 1 ) . Single cell analysis of double staining with S100β showed that nearly 70% were co-labeled , identifying them as astrocytes . We also identified a small fraction ( 6% ) of tomato cells as oligodendrocytes . A small fraction ( 9% ) of tomato cells were co-labeled with the neuronal marker , NeuN , but this represented only 1 . 3% of cortical neurons ( Figure 2—figure supplement 1 ) . Importantly , although a minor fraction of recombined cells was identified as neurons or oligodendrocytes , nearly all cortical astrocytes analyzed expressed the tdTomato reporter protein ( 95%; Figure 2—figure supplement 1 ) , suggesting effective targeting of the cortical astrocyte population using this Gfap-Cre driver . We performed qPCR on whole cortex from Gfap Smo CKO mice and littermate controls and found a 50% reduction in the number of Smo transcripts ( Figure 2—figure supplement 2 ) . As Smo is also expressed in neurons ( Harwell et al . , 2012 ) , the remaining Smo transcripts are likely due to neurons that do not undergo Cre-mediated recombination in these mice . Importantly , Gfap Smo CKO mice show a nearly complete loss of Gli1 activity , with no difference in the number of astrocytes in the mature cortex , demonstrating effective interruption of canonical Shh signaling ( Garcia et al . , 2010 ) . These data demonstrate that this Gfap-Cre driver is an effective tool for selective deletion of Smo in cortical astrocytes , and that canonical , Gli-mediated Shh activity is effectively abolished . During postnatal development , Shh is required for establishing synaptic connectivity between layer V neurons and their presynaptic partners in layer II/III , which is mediated by Gli-independent , non-canonical signaling ( Harwell et al . , 2012 ) . To examine whether canonical , Gli-mediated Shh signaling in astrocytes plays a role in establishing synaptic connectivity of cortical neurons , we evaluated spine density of apical dendrites on layer V neurons in the somatosensory cortex of Gfap Smo CKO mice across postnatal development . Dendritic spines receive the vast majority of excitatory input , thereby serving as a useful readout of synaptic connectivity ( Nimchinsky et al . , 2002 ) . To visualize neurons for reliable identification and tracing , Gfap Smo CKO mice were crossed with Thy1-GFPm transgenic mice , which sparsely express GFP in a subset of layer V cortical neurons ( Feng et al . , 2000 ) . Dendritic spines undergo a period of dynamic reorganization during postnatal development during which there is an initial overproduction of spines over the first 2–3 weeks of postnatal development followed by a period of synaptic pruning that refines the precise connectivity of cortical circuits ( Zuo et al . , 2005a ) . At P14 , spine density was comparable between Gfap Smo CKO mice and wild type ( WT ) littermate controls , suggesting that early synaptogenesis does not require astrocytic Shh signaling ( 0 . 58 ± 0 . 03 spines/µm and 0 . 56 ± 0 . 01 spines/µm , WT [n = 7 cells] and Gfap Smo CKO [n = 6 cells] , respectively , p=0 . 64 , two animals per genotype ) . Between P14 and P21 , spine density remained stable in WT , but showed a dramatic increase at P28 ( Figure 2 ) . This period of synapse addition was followed by a steady reduction in spine density at P42 and into adulthood ( ≥P90 ) , reflecting the developmental elimination of spines . In contrast , Gfap Smo CKO mice showed an accelerated timeline of spine addition in which spine density increased dramatically between P14 and P21 ( Figure 2 ) . Spine density was significantly higher at P21 in Gfap Smo CKO mice compared to controls ( 0 . 53 ± 0 . 05 spines/µm and 0 . 76 ± 0 . 06 spines/µm , WT [n = 9 cells] and Gfap Smo CKO [n = 9 cells] , respectively , p=0 . 008 , three animals per genotype ) . Although there was a modest reduction in spine density as animals matured , spine density remained elevated in adult Gfap Smo CKO mice compared to WT controls ( 0 . 36 ± 0 . 02 spines/µm and 0 . 60 ± 0 . 02 spines/µm in WT [n = 24 cells] and Gfap Smo CKO [n = 23 cells] , respectively , p<0 . 0001 , four animals per genotype; Figure 2 ) . Thus , whereas WT mice experience a 49% reduction in spine density from the peak of spine density at P28 to adulthood , Gfap Smo CKO mice exhibit a 22% reduction from their peak at P21 to adulthood . Interestingly , we find no difference in basal dendrite spine density in Gfap Smo CKO mice ( 0 . 60 ± 0 . 04 spines/um and 0 . 62 ± 0 . 04 spines/µm in WT [n = 10 cells] and Gfap Smo CKO [n = 10 cells] , respectively , p=0 . 71 , three animals per genotype; Figure 2—figure supplement 3 ) . This is in contrast to embryonic deletion of Shh from neurons using the Emx-Cre driver , which produces a reduction in spine density of basal dendrites on layer V neurons ( Harwell et al . , 2012 ) . These data suggest that while the early stages of synapse formation proceed independently of astrocytic Shh signaling , the maturing cortical circuit requires intact Shh activity in astrocytes for the developmental pruning of excess synapses necessary to achieve its mature organization . To confirm that elevated spine density was due specifically to a loss of astrocytic Shh signaling , we interrogated the spine density of pyramidal cells in two different regions where Gli1 astrocytes are relatively sparse ( see Figure 1 ) . Pyramidal neurons in layer II/III of WT mice showed a higher spine density than layer V neurons ( 0 . 79 ± 0 . 06 spines/µm , n = 7 cells , four animals ) consistent with previous studies ( Holtmaat et al . , 2005 ) . However , this was not significantly different from the spine density observed in Gfap Smo CKO mice ( 0 . 76 ± 0 . 06 spines/µm , p=0 . 76 , n = 9 cells , five animals; Figure 2 ) . We also analyzed pyramidal neurons in the hippocampus . Although the dentate gyrus of the hippocampus harbors a population of Gli1-expressing adult neural progenitor cells , mature , differentiated astrocytes expressing Gli1 are relatively sparse ( Ahn and Joyner , 2005; Han et al . , 2008 ) ( Figure 1 ) . The spine density of CA1 pyramidal neurons in adult mice was higher than in cortical neurons , consistent with previous studies ( Attardo et al . , 2015; Perez-Cruz et al . , 2011 ) . However , there was no significant difference in spine density between Gfap Smo CKO mice and WT controls ( 0 . 97 ± 0 . 05 spines/µm and 0 . 90 ± 0 . 05 spines/µm in WT [n = 11 cells] and Gfap Smo CKO [n = 11 cells] , respectively , p=0 . 32 , three animals per genotype; Figure 2 ) . Our analysis of the recombination pattern of the Gfap-Cre driver showed that some cortical neurons undergo recombination ( Figure 2—figure supplement 1 ) . To rule out the possibility that Smo deletion in a small population of neurons is responsible for the elevated spine density of layer V neurons , we deleted Smo in excitatory neurons using the CamKIIα-Cre driver ( CK2 Smo CKO ) , and evaluated spine density of layer V neurons in adult mice . We found no significant difference in spine density between CK2 Smo CKO mice and littermate controls ( 0 . 77 ± 0 . 07 spines/µm and 0 . 59 ± 0 . 06 spines/µm in WT [n = 10 cells] and CK2 Smo CKO [n = 9 cells] , respectively , p=0 . 07 , three animals per genotype; Figure 2 ) . Taken together , these data suggest that Shh activity in astrocytes is necessary for the developmental reorganization of dendritic spines specifically on layer V cortical neurons . Notably , we observed elevated spine density along the apical dendrites of layer V neurons , which traverse layer II/III , where few Gli1 astrocytes are found . This suggests that the modification of synapse organization mediated by Gli1 astrocytes does not occur through direct astrocyte-synapse interactions . Dendritic spines are dynamic structures that undergo rapid formation and elimination during postnatal cortical development . As cortical circuits mature , the fraction of spines undergoing dynamic turnover declines concomitant with an increase in spine stability ( Zuo et al . , 2005a; Holtmaat et al . , 2005 ) . The dynamic turnover of spines has long been considered a structural correlate of synaptic plasticity and can be regulated in an activity-dependent manner ( Lendvai et al . , 2000; Trachtenberg et al . , 2002; Holtmaat and Svoboda , 2009; Zuo et al . , 2005b ) . To evaluate the role of astrocytic Shh signaling in mediating spine dynamics , we performed repeated in vivo imaging of the apical tufts of layer V neurons in the somatosensory cortex through a cranial window ( Holtmaat et al . , 2009 ) . We confirmed the layer V identity of imaged neurons by following individual dendritic segments to their soma and created 3D reconstructions of imaged neurons . Only dendrites with soma in layer V , typically 500 µm – 600 µm from the surface of the brain , were analyzed . In Thy1-GFPm mice , expression of GFP is developmentally regulated such that at P14 , the intensity of GFP expression and the density of fluorescently labeled neurons within the 3 mm cranial window is insufficient for reliable imaging , as has been previously reported ( Holtmaat et al . , 2005 ) . However by P17 , GFP expression was sufficiently dense and bright to enable reliable imaging . We analyzed the fraction of spines undergoing dynamic turnover over 2 days in young mice at P17-P21 and P28-P32 . In WT mice , the turnover ratio was 0 . 14 at P17-P21 ( n = 5 mice ) and 0 . 10 in P28-P32 mice ( n = 3 mice; Figure 3 ) . In Gfap Smo CKO mice however , the turnover ratio was lower at P17-P21 than in WT mice , and remained stable at P28-P32 ( 0 . 11 and 0 . 10 , respectively , n = 4 mice; Figure 3 ) . This suggests that dendritic spines in Gfap Smo CKO mice stabilize earlier than those in WT mice . Consistent with this , we found that the fraction of filopodia was lower in juvenile Gfap Smo CKO mice compared to WT controls ( Figure 3 ) . Filopodia are structural precursors of dendritic spines and exhibit higher rates of dynamic turnover than mature spines ( Ziv and Smith , 1996; Berry and Nedivi , 2017 ) . Accordingly , the density of filopodia declines as the cortex matures ( Zuo et al . , 2005a; Grutzendler et al . , 2002 ) . Indeed , the fraction of all protrusions in juvenile WT mice ( n = 7 mice ) identified as filopodia was 25% , but that fraction declined significantly to 15% in adults ( n = 16 mice , p=0 . 0039; Figure 3 ) . In contrast , we did not observe this age-dependent decline in Gfap Smo CKO mice ( 17% and 12% in juvenile [n = 10 mice] and adult [n = 7 mice] Gfap Smo CKO mice , respectively , p=0 . 398 ) . The filopodial fraction in juvenile Gfap Smo CKO mice was already significantly below WT levels and comparable to adult WT levels , suggesting that nascent spines undergo accelerated stabilization in the absence of astrocyte-mediated Shh signaling . We performed further analysis of spine morphologies and classified spines as mature or intermediate , to determine if there is an increase in mature spines . Protrusions with a mushroom-like morphology were classified as mature , and thin protrusions lacking a discernible spine head , or exhibiting a relatively dim , small head were identified as intermediate spines ( Figure 3—figure supplement 1 ) . In P17-21 mice , WT and CKO mice showed similar proportions of mature spines ( 54% and 55% in WT and CKO , respectively , n = 4 mice per genotype; Figure 3 ) , however Gfap Smo CKO mice showed an apparent increase in protrusions with an intermediate morphology ( 16% and 26% in WT and CKO , respectively; Figure 3 ) , though this was not statistically significant . This suggests that these intermediate protrusions reflect transitional morphologies as spines undergo maturation , and become more stable . Consistent with this , adult Gfap Smo CKO mice trended toward a modest increase in the fraction of mature spines , compared to their WT controls ( 55% and 65% in WT and CKO , respectively , n = 4 mice per genotype; Figure 3—figure supplement 1 ) , whereas the fraction of intermediate spines was comparable at this age ( 27% and 23% in WT and CKO , respectively ) . It is well established that the rate of synaptic turnover declines considerably as animals mature , reflecting an increase in stability of synaptic connections ( Zuo et al . , 2005a; Holtmaat et al . , 2005 ) . We next sought to investigate the long-term stability of individual spines in adult mice by imaging dendrites weekly for up to 6 weeks . We investigated the fraction of spines identified on the first day of imaging that persisted in subsequent imaging sessions using custom written MATLAB code ( Figure 3—figure supplement 2 ) . We calculated the survival fraction by fitting to an exponential decay model . This analysis revealed a larger proportion of long-lived stable spines in Gfap Smo CKO mice compared to WT control ( 71% and 79% in WT [n = 7 mice] and Gfap Smo CKO [n = 5 mice] , respectively , p=0 . 0068 , Extra sum-of-squares F test; Figure 3 ) , suggesting an increase in stability of dendritic spines . Interestingly , among the population of dynamic spines , we found two distinct populations consisting of transient spines , which disappeared and did not reappear for the duration of the study , and recurrent spines , which disappeared and then subsequently reappeared . Our data showed that the shift towards increased spine stability in the Gfap Smo CKO neurons was entirely due to a reduction in the proportion of recurrent spines , with nearly identical fractions of transient spines observed across both genotypes ( Figure 3 ) . These data suggest that structural plasticity is impaired in Gfap Smo CKO mice . This deficit in structural plasticity emerges during the third week of postnatal development and persists into adulthood . Taken together , these data demonstrate that astrocyte-mediated Shh signaling is required for establishing and maintaining the organization of cell type specific cortical synapses . In both the developing and adult CNS , astrocytes directly eliminate synapses in an activity-dependent manner through MEGF10 and MERTK , two phagocytic receptors enriched in astrocytes ( Chung et al . , 2013 ) . To investigate whether these genes are negatively regulated in Gfap Smo CKO mice , we examined their expression in the cortex of adult mice by quantitative PCR ( qPCR ) . Expression of both Mertk and Megf10 was comparable between WT and Gfap Smo CKO mice ( dCq values , Mertk: 5 . 97 ± 0 . 23 and 5 . 87 ± 0 . 06 in WT [n = 3 mice] and Gfap Smo CKO [n = 6 mice] , respectively , p=0 . 54; Megf10: 8 . 02 ± 0 . 38 and 7 . 69 ± 0 . 18 in WT [n = 3 mice] and Gfap Smo CKO [n = 6 mice] , respectively , p=0 . 39; Figure 4—figure supplement 1 ) . These data suggest that direct engulfment of synapses by astrocytes through these pathways is not regulated by Shh signaling and is not responsible for the increased spine density seen in Gfap Smo CKO animals . In the mature cerebellum , Shh signaling between Purkinje neurons and Bergmann glia regulates expression of the glial specific , inward-rectifying K+ channel , Kir4 . 1 ( Farmer et al . , 2016 ) . In the cortex , the distribution of Kir4 . 1 exhibits a laminar pattern similar to that of Gli1 astrocytes , such that expression is enriched in layers IV and Vb , and reduced in layer Va ( Figure 4 ) . High resolution , confocal analysis showed that Kir4 . 1 is localized in the processes of Gli1 astrocytes ( Figure 4 ) . Kir4 . 1 expression is localized to astrocytic endfeet and is found surrounding neuronal somata in the spinal cord and brain ( Kelley et al . , 2018; Cui et al . , 2018; Higashi et al . , 2001 ) . In WT mice , expression of Kir4 . 1 surrounded many NeuN-positive neuronal somata in layer V ( Figure 4 ) . In Gfap Smo CKO mice , however , there was a pronounced reduction in the expression of Kir4 . 1 throughout the cortex , and peri-somal Kir4 . 1 expression was severely diminished ( Figure 4 ) . In order to quantify expression levels of Kir4 . 1 , we measured transcript abundance in the cortex of Gfap Smo CKO mice and WT controls by droplet digital PCR ( ddPCR ) . In Gfap Smo CKO mice , there was a 50% reduction in Kcnj10 transcripts compared to WT controls ( 1 . 99 × 106 ± 1 . 93 x 105 copies/µg and 9 . 92 × 105 ± 8 . 49 x 104 copies/µg in WT [n = 6 mice] and Gfap Smo CKO [n = 4 mice] , respectively , p=0 . 02 ) . There was no difference in Kcnj10 levels in CK2 Smo CKO mice compared to WT ( 1 . 88 × 106 ± 3 . 77 × 105 copies/µg , p=0 . 95 , n = 4 mice; Figure 4 ) , suggesting that downregulation of Kir4 . 1 is due to astrocyte-specific , and not neuronal , disruption of Shh signaling . Conversely , Glast-CreER-mediated deletion of the Shh receptor Ptch1 , a negative regulator of Shh signaling , dramatically upregulated Kir4 . 1 expression in the cortex ( Glast Ptch CKO; Figure 4 ) . Kir4 . 1 expression is associated with glutamate uptake both in vitro and in vivo ( Kucheryavykh et al . , 2007; Djukic et al . , 2007; Tong et al . , 2014 ) . We therefore measured expression levels of the astrocyte-specific glutamate transporters Glast and Glt1 . There was no difference in expression of these genes in Gfap Smo CKO mice compared to WT ( Glt1: 1 . 08 × 107 ± 6 . 6 x 105 copies/µg and 9 . 58 × 106 ± 2 . 1 x 105 copies/µg in WT and Gfap Smo CKO , respectively , p=0 . 15 , n = 3 mice per genotype; Glast: 7 . 4 × 106 ± 4 . 4 x 105 copies/µg and 6 . 7 × 106 ± 6 . 5 x 105 copies/µg in WT and Gfap Smo CKO , respectively , p=0 . 41 , n = 3 mice per genotype; Figure 4—figure supplement 1 ) . It should be noted that Kir4 . 1 was not restricted to Gli1 astrocytes in layers IV and Vb . Nevertheless , single transcript quantification of Kcnj10 transcripts in Gfap Smo CKO show a pronounced reduction , suggesting that Kir4 . 1 expression in cortical astrocytes is regulated , in part , by Shh signaling . To examine whether neurons in Gfap Smo CKO mice exhibit disruptions in physiological activity , we performed whole-cell patch clamp recordings of layer V pyramidal neurons from coronal brain slices at P21 , and examined action potential firing and membrane properties . Using current clamp , we recorded action potential spikes per injected current at multiple step currents from −300 pA to +650 pA . Our results revealed a significant increase in spike numbers with high current injections ( >500 pA ) and total overall spikes in Gfap Smo CKO animals compared to WT control ( 326 ± 54 and 574 ± 71 spikes in WT [n = 11 cells from four animals] and Gfap Smo CKO [n = 7 cells from three animals] , respectively , p=0 . 01; Figure 5 ) . We also measured other membrane properties , including input resistance , resting membrane potential and tau , and found no significant difference between Gfap Smo CKO and WT neurons ( Figure 5—figure supplement 1 ) . However , we observed a significant decrease in action potential threshold ( −30 . 6 ± 2 . 0 and −37 . 9 ± 2 . 0 mV in WT and Gfap Smo CKO , respectively , p=0 . 0441 ) , accompanied by an increase in action potential amplitude ( 59 . 6 ± 2 . 6 and 73 . 3 ± 3 . 0 pA in WT and Gfap Smo CKO , respectively , p=0 . 0083 ) , and a trending decrease in action potential ½ width ( 1 . 0 ± 0 . 92 and 0 . 76 ± 0 . 02 ms in WT and Gfap Smo CKO , respectively , p=0 . 05 ) in Gfap Smo CKO neurons , consistent with an increase in neuron excitability ( Figure 5 ) . Together , these results suggest Gfap Smo CKO neurons exhibit an increase in neuronal excitability consistent with excess extracellular K+ . To examine excitatory synaptic transmission , we recorded spontaneous and miniature excitatory postsynaptic currents ( sEPSCs and mEPSCs ) in layer V pyramidal neurons . We found an increase in both the frequency ( 0 . 74 ± 0 . 08 and 1 . 18 ± 0 . 19 Hz in WT [n = 16 cells] and Gfap Smo CKO [n = 16 cells] , respectively , p=0 . 042 ) and amplitude ( 8 . 54 ± 0 . 38 and 11 . 33 ± 0 . 91 pA in WT and Gfap Smo CKO , respectively , p=0 . 008 ) of sEPSCs in slices from Gfap Smo CKO mice compared to WT control slices ( four animals per genotype; Figure 5 ) . Furthermore , we observed an increase in the frequency ( 0 . 40 ± 0 . 05 and 1 . 10 ± 0 . 23 Hz in WT and Gfap Smo CKO , respectively , p=0 . 003 ) and amplitude ( 7 . 54 ± 0 . 57 and 10 . 34 ± 0 . 74 pA in WT and Gfap Smo CKO , respectively , p=0 . 005 ) in mEPSCs in the presence of tetrodotoxin ( TTX ) , indicating an increase in postsynaptic response of neurons in Gfap Smo CKO mice ( n = 14 and 11 cells from WT and Gfap Smo CKO , respectively , four animals per genotype; Figure 5 ) . These data suggest that loss of Shh activity in astrocytes produces disturbances in both pre and postsynaptic activity . Moreover , these data suggest that astrocytes modulate both neuronal excitability and excitatory synaptic transmission in a Shh-dependent manner . We previously demonstrated that cortical astrocytes in Gfap Smo CKO mice upregulate GFAP expression and exhibit cellular hypertrophy , two classic hallmarks of reactive astrogliosis ( Garcia et al . , 2010 ) . Astrocytes exhibiting these features were broadly distributed across cortical layers ( Figure 6 ) , in contrast to the distribution of Gli1 astrocytes which are found predominantly in layers IV and V . In addition to upregulation of GFAP , astrocytes in Gfap Smo CKO mice show dramatic changes in morphological structure . Sholl analysis revealed no difference in the number of primary branches in Gfap Smo CKO mice compared to WT controls ( 7 . 2 and 7 . 5 branches , WT [n = 9 cells] and Gfap Smo CKO [n = 9 cells] , respectively , three animals per genotype ) . However , astrocytes from Gfap Smo CKO animals did show an increase in the number of higher order branches , the length of the longest tree ( 136 . 6 ± 17 and 212 . 1 ± 28 µm for WT and Gfap Smo CKO , respectively , p=0 . 035 ) , and the total length of all processes ( 397 ± 66 and 674 ± 91 µm for WT and Gfap Smo CKO , respectively , p=0 . 025 ) compared to WT controls ( Figure 6 ) . Consistent with this , there was a significant increase in the number of branches intersecting concentric shells at various distances from the soma ( Figure 6 ) . Notably , in CK2 Smo CKO mice , GFAP staining in the cortex was indistinguishable from WT controls ( Figure 6—figure supplement 1 ) , suggesting that astrocytes exhibit dramatic changes in morphological structure following astrocytic , but not neuronal , disruption of Shh signaling . Such changes in morphology are consistent with mild reactive gliosis , a complex cellular response of astrocytes to disturbances in physiological homeostasis ( Sofroniew , 2015 ) . Notably , the distribution of reactive astrocytes in Gfap Smo CKO mice is broader than that defined by Gli1 expression , arguing against the idea that Shh signaling regulates the intrinsic state of astrocytes . Rather , reactive gliosis may instead reflect a cellular response to the mild , but persistent , aberrations in neuronal activity observed in Gfap Smo CKO mice .
The dynamic processes underlying organization of synapses during postnatal development are essential for establishing functional neural circuits . In this study , we demonstrate that astrocytes modulate refinement and reorganization of synaptic connectivity in the postnatal cortex in a Shh-dependent manner . Our data show that Gli1 astrocytes are enriched in deep cortical layers with a relative paucity in upper cortical layers . Selective disruption of Shh signaling in astrocytes leads to an overabundance of spines on the apical dendrites of layer V , but not layer II/III , cortical neurons . The overabundance of spines emerges during postnatal development , continues into adulthood and is accompanied by a reduction in dynamic turnover . Gfap Smo CKO mice exhibit a pronounced reduction in Kir4 . 1 as well as an increase in neuronal excitability . Finally , we show that cortical astrocytes exhibit dramatic changes in morphology and GFAP expression , a phenotype consistent with mild reactive gliosis . Our data identify an essential role for astrocyte modulation of neuronal activity during postnatal development , facilitating activity-dependent reorganization of synaptic connectivity . To accomplish selective disruption of Shh signaling in astrocytes , we used a Cre driver in which expression is regulated by the full-length mouse Gfap gene ( Garcia et al . , 2004 ) . Single cell analysis in Gfap-Cre;Ai14 mice showed that nearly all cortical astrocytes undergo Cre-mediated recombination , while the fraction of recombined cortical neurons is less than 2% . Importantly , we observed recombined neurons predominantly in superficial layers , consistent with early Cre activity during late embryogenesis when layer II/III neurons are being generated . Notably , despite a small population of Smo null neurons in layer II/III , changes in spine density are not observed in these cells . In addition , a more complete deletion of Smo in excitatory neurons using the CamKIIαCre driver fails to alter spine density in layer V neurons , arguing against non-specific effects arising from the small population of recombined neurons in Gfap Smo CKO mice . Importantly , Gli1 expression is nearly completely lost in Gfap Smo CKO mice , with no loss in the number of total astrocytes ( Garcia et al . , 2010 ) , demonstrating effective disruption of Shh activity in these mice . Because a small number of recombined oligodendrocytes were observed in GfapCre;Ai14 mice , the possibility that these cells may contribute to some of the observed phenotypes cannot be ruled out . However Gli1 expression is restricted predominantly to astrocytes in the adult forebrain ( Garcia et al . , 2010 ) . Thus , any contribution from other cell types must be mediated by non-canonical , Gli-independent mechanisms . In the postnatal cortex , Shh signaling from layer V neurons mediates synaptic connectivity with their layer II/III presynaptic partners ( Harwell et al . , 2012 ) . Here , we show that , in addition to neuronal Shh signaling , Shh activity in astrocytes is required for the establishment and maintenance of cortical circuits . This is supported by two key observations in our study . First , we find that in Gfap Smo CKO mice , apical dendrites of cortical neurons exhibit an increase in spine density that emerges during postnatal development and persists into adulthood . Importantly , genetic deletion of Smo in mature neurons instead of astrocytes failed to show any differences in spine density , pointing to a specific role for Shh signaling in astrocyte regulation of synaptic connectivity . Interestingly , deletion of Shh ligand using an EmxCre driver produces a reduction in spine density of basal dendrites ( Harwell et al . , 2012 ) , whereas we found no difference in basal dendrite spine density , suggesting that Shh acts in distinct , cell type-specific ways to regulate synaptic connectivity of local cortical circuits . Indeed , whereas astrocytes express Gli1 , neurons do not ( Garcia et al . , 2010 ) , indicating differential transduction of Shh through canonical and non-canonical , Gli-independent pathways , respectively , in different cell types . Second , our data show fewer filopodia in juvenile Gfap Smo CKO mice compared to WT controls along with a concomitant increase in spine stability . In the adult , spines continue to show deficits in dynamic turnover , demonstrating long-term impairments in structural plasticity . This suggests that Shh signaling is required not only for identifying appropriate synaptic partners during development , but also plays an important role in mediating synaptic plasticity . In the postnatal and adult brain , we and others identified neurons as the source of Shh ( Harwell et al . , 2012; Garcia et al . , 2010; Farmer et al . , 2016; Álvarez-Buylla and Ihrie , 2014 ) . However non-neuronal sources , including astrocytes have been identified in the injured brain ( Amankulor et al . , 2009; Sirko et al . , 2013 ) . Beyond the CNS , epithelial cells have been reported as a source of Shh in the developing dentate gyrus ( Choe et al . , 2015 ) . It will be interesting to examine whether other non-neuronal sources are available to trigger astrocyte transduction of Shh . We previously demonstrated that Gli1 expression is restricted to a subpopulation of cortical astrocytes ( Garcia et al . , 2010 ) . Here , we extend these findings and demonstrate that the distribution of Gli1 astrocytes in the cortex is not uniform across cortical layers , but rather , occurs in a laminar-specific fashion . Cortical layers have been historically defined by neuronal populations and their specific functional properties and connectivity . However , emerging evidence suggests that astrocytes also exhibit cortical lamination patterns based on distinct gene expression profiles ( Bayraktar , 2018; Miller et al . , 2019 ) . Our data show that Gli1 astrocytes are enriched in layers IV and V with a relatively sparse distribution in layer II/III . Interestingly , a recent study reported a population of astrocytes with a similar distribution that show enrichment of the Shh pathway ( Miller et al . , 2019 ) . These cells were also shown to regulate spine density of Layer V cortical neurons . It will be interesting to examine whether these cells correspond to Gli1 astrocytes reported in this study . Whereas astrocytes transducing Shh are enriched in deep cortical layers , the BMP antagonist , chordin-like 1 ( Chrdl1 ) is preferentially enriched in upper layer cortical astrocytes ( Blanco-Suarez et al . , 2018 ) , suggesting that Gli1 and Chrdl1 may identify two distinct , but regionally complementary , astrocyte populations . Alternatively , Shh signaling may actively repress Chrdl1 expression in deep layer astrocytes . Notably , our observation that Gli1 astrocytes only comprise a fraction of astrocytes in layers IV and V suggests that , in addition to the laminar-specific distribution of these cells , there may be additional heterogeneity of astrocytes even within a given cortical layer . Whether and how the laminar distribution of astrocytes with distinct gene expression profiles confers functional specialization is not well understood and requires further study . Interestingly , we observed that the disturbances in synaptic organization of cortical neurons in Gfap Smo CKO mice occurs selectively in apical dendrites of layer V , but not layer II/III , neurons . Since apical dendrites of layer V neurons traverse through layer II/III , this suggests that Shh-dependent regulation of synaptic refinement is not mediated by direct interaction between Gli1 astrocytes and synapses . One possibility is that Gli1 astrocytes interact with neuronal soma in layer V and effectively modulate their excitability . Our observation that Kir4 . 1 expression in cortical astrocytes is regulated by Shh signaling suggests that Gfap Smo CKO mice may experience deficits in local buffering of extracellular K+ leading to perturbations in neuronal activity . Indeed , a recent study demonstrated that Kir4 . 1 surrounds neuronal soma in the lateral habenula and regulates their firing properties ( Cui et al . , 2018 ) . Astrocytes play essential roles in buffering extracellular K+ through Kir4 . 1 , and several studies demonstrate that loss of function of Kir4 . 1 impairs K+ uptake and increases neuronal excitability ( Djukic et al . , 2007; Tong et al . , 2014; Sibille et al . , 2014 ) . Alternatively , increased neuronal excitability may be due to disruptions in glutamate clearance . Although we did not observe changes in expression of either Glt1 or Glast , deficits in Shh-dependent trafficking , localization or transporter function would lead to impairments in astrocyte-mediated glutamate clearance . Further work is needed to definitively identify the precise mechanism underlying the abnormal firing properties of neurons in Gfap Smo CKO mice . The establishment of appropriate synapse number and connectivity are tightly regulated processes that are profoundly shaped by experience or activity ( Zuo et al . , 2005b; Yang et al . , 2009 ) . Our observation that Gfap Smo CKO mice show increases in spontaneous neuronal activity and excitability , accompanied by a pronounced increase in spine density suggests that astrocytes contribute to activity-dependent processes that shape the organization of developing neural circuits in a Shh-dependent manner . The deficits in synaptic organization observed in Gfap Smo CKO mice persist for the life of the animal . These deficits are not limited to synapse number but also include lower rates of spine turnover , suggesting long-lasting deficits in structural plasticity . Together with the observation that Shh guides local synaptic connectivity of cortical neurons ( Harwell et al . , 2012 ) , these studies reveal that Shh signaling exerts considerable influence on the establishment and maintenance of cortical circuits .
All experiments were approved by Drexel University’s Institutional Animal Care and Use Committee and were conducted according to approved protocols . We used the following strains of transgenic mice on the C57BL/6 background: Gli1CreER/+ ( Ahn and Joyner , 2005 ) , Ai14 ( Madisen et al . , 2010 ) , Smofl/fl ( Long et al . , 2001 ) , Gfap-Cre ( Garcia et al . , 2004 ) , CaMKIIa-Cre ( Tsien et al . , 1996 ) and Thy1-GFPm ( Feng et al . , 2000 ) . Mice of either sex were included in these studies . Tamoxifen was administered as previously described ( Garcia et al . , 2010; Allahyari et al . , 2019 ) . Briefly , tamoxifen ( Sigma , T5648 ) was dissolved in corn oil to a final concentration of 20 mg/ml . Adult Gli1CreER/+;Ai14 mice received 250 mg/kg tamoxifen by oral gavage for three consecutive days , and tissues were analyzed two – three weeks later . Mice were deeply anesthetized by intraperitoneal injection of ketamine/xylazine/acepromazine . Animals were then transcardially injected with 100 units heparin and perfused with phosphate-buffered saline ( PBS ) followed by 4% paraformaldehyde ( PFA , Sigma-Aldrich , St . Louis , MO ) . Brains were fixed overnight at 4°C and then transferred to 30% sucrose for at least 48 hr or until sectioned . Brains were cryosectioned ( Leica CM3050S , Wetzlar , Germany ) and collected in 40 µm sections . Sections were stored in 0 . 1M Tris-buffered saline ( TBS ) with 0 . 05% sodium azide at 4°C . Immunohistochemistry was performed using the following primary antibodies: rabbit anti-GFP ( AB3080; Millipore , Burlington , MA ) , rabbit anti-Kir4 . 1 ( AB5818; Millipore ) , rabbit anti-GFAP ( Z033429-2; Agilent DAKO , Santa Clara , CA ) , mouse anti-NeuN ( MAB377; Millipore ) , rabbit anti-RFP ( PM005; MBL International , Woburn , MA ) , rat anti-Ctip2 ( ab18465; Abcam , Cambridge , UK ) , sheep anti-CAII ( AHP206; Bio-Rad ) , rabbit anti-S100β ( Z0311 , Agilent ) , and rabbit anti-Olig2 ( AB9610; Millipore ) . Sections were washed three times in 0 . 1M TBS at room temperature ( 10 min/wash ) . Sections were then blocked in 10% Normal Serum with 0 . 5% Triton-X ( Sigma-Aldrich ) for one hour at room temperature and incubated with primary antibody in 0 . 5% Triton-X at 4°C overnight . For fluorescent immunohistochemistry , sections were rinsed the following day in 0 . 1M TBS and incubated in Alexafluor-conjugated secondary antibodies with 10% Normal Serum and 0 . 1M TBS and incubated at room temperature for two hours . The sections were then rinsed in 0 . 1 TBS and incubated in DAPI ( 1:50 , 000; Life Technologies , Carlsbad , CA ) for 15 min . Sections were rinsed again in 0 . 1M TBS and mounted onto microscope slides ( Fisherbrand , Waltham , MA ) and coverslipped using ProLong Gold Antifade Mountant ( Invitrogen , Carlsbad , CA ) and Fisherfinest Premium Cover Glass . For brightfield immunohistochemistry , sections were rinsed as previously described but were incubated with biotinylated secondary antibodies against the primary antibody species ( Vector Laboratories , Burlingame , CA ) for one hour . Sections were then placed in Avitin-Biotin Complex solution ( Vector ) for one hour and visualized using 3 , 3’-diaminobenzidine ( DAB Peroxidase Substrate Kit , Vector ) . Sections were then mounted onto slides and coverslipped with DPX Mountant ( Fisher ) . Analysis of spine density was performed in a blinded study design . WT controls were derived from Cre-negative littermates . Neurons were traced using Neurolucida ( MicroBrightField Biosciences , Williston , VT ) and individual spines were marked . Apical dendrites of layer V neurons were counted from ~100 µm below the primary bifurcation through the apical tuft . Apical dendrites of layer II/III cells were analyzed just below the primary bifurcation through the apical tuft . Basal dendrites were analyzed beginning ~50 µm from the soma through the end of the processes . CA1 hippocampal neurons were analyzed from just below the primary bifurcation through the apical tuft . All analysis was performed on an upright Zeiss microscope using a 63X oil objective . Cranial window surgeries were adapted from a published protocol ( Holtmaat et al . , 2009 ) . Immediately prior to each surgery , mice received a subcutaneous injection of carprofen ( Rimadyl , 5 mg/kg ) , and a follow-up injection was given 24 hr after surgery . Mice were deeply anesthetized with isoflurane ( 5% ) in an induction chamber , and then transferred to a stereotaxic apparatus where they inhaled 1 . 5% isoflurane continuously for the duration of the surgery . The scalp was then removed and a metal bar ( ~1 cm long ) with threadings for screws was affixed to the skull , using superglue and dental acrylic to form a head post that sealed off the skin while leaving the right parietal bone exposed . After allowing 24 hr for the glue and acrylic to set , the mouse was returned to the stereotaxic apparatus and head-fixed using the headpost . A circular craniotomy ~3 mm in diameter and centered 2 . 5 mm lateral and posterior to bregma was performed . The exposed dura was treated with saline-soaked gel foam until any minor bleeding ceased . Finally , a 3 mm glass coverslip was gently pressed onto the dura and sealed in place using superglue and dental acrylic . Adult mice were allowed at least 1 week for recovery . To mitigate the tendency of rapid bone growth to destabilize the windows at younger ages , juvenile mice were allowed 24 hr to recover . A commercial two-photon microscope ( Bruker , Billerica , MA ) with a tunable Ti:Sapphire laser ( Coherent , Santa Clara , CA ) was used for all experiments . The laser was set to 920 nm and power was tuned on the sample as necessary to obtain consistently high-quality images without damaging the tissue . Mice were head-fixed and anesthetized with continuous isoflurane ( 1–1 . 5% ) during imaging . Under a 20X water immersion objective , GFP-expressing neurons were identified visually and then imaged down to the cell body to determine layer . From the apical dendrite tuft , individual dendrite segments that projected primarily along the xy-plane were selected for analysis , and z-stacks ( 0 . 5 µm step size ) were collected . To track individual dendrites , repeated imaging was generally performed on the following schedule ( in days , relative to first imaging session ) : 0 , 1 , 2 , 7 , 14 , 21 , 28 , 42 . Some data from studies using other schedules were also included . For juvenile mice , this schedule was: 0 , 1 , 2 . Every effort was made to use Cre-negative , littermate controls for these experiments . However , it was not always possible to match littermate controls in the final data sets presented here , either due to the absence of the GFP allele in the Cre-negative controls , or due to the loss of head caps securing the cranial windows , over time . However , it should be noted that all data presented here represent Cre-negative controls from the appropriate strain , if not necessarily from the same litter . To analyze spine turnover , we performed side-by-side manual comparison of dendritic protrusions using ImageJ . Z-stacks from two time points were analyzed , and individual protrusions were identified as present or absent in each stack . For each mouse , a minimum of 150 protrusions were analyzed , although the average number of protrusions analyzed per mouse was 250–300 . These comparisons were conducted by trained observers blinded to mouse genotype . Long , thin protrusions lacking a bulbous head were identified as filopodia , and any other prominent dendritic protrusion that extended >0 . 4 µm from the shaft was counted as a spine . The filopodial fraction was calculated as the proportion of all analyzed features present at the first imaging timepoint that were identified as filopodia . The position and dynamic status ( stable/eliminated/formed ) of each spine was recorded using ImageJ’s CellCounter plugin . Each neuron’s turnover ratio was calculated as:TO= ( Nelim+Nform ) / ( Nelim+Nform+2Nstable ) For longitudinal analysis of spine lifetime , the individual features identified in our manual ImageJ analysis were then imported into custom written MATLAB ( Mathworks , Natick , MA ) code and manually tracked across all imaged time-points using custom-written software . This code facilitated the efficient manual tracking of each feature by iteratively estimating the z-plane , within each stack , in which that feature would be expected to appear , assuming a linear relationship . Using a graphical user interface , the user was presented with a series of frames centered on that plane and allowed to make a determination of the feature’s presence or absence simply by left-clicking on the feature , or in the case of absence right-clicking on the location it otherwise would have been , using the best frame presented ( Figure 3—figure supplement 2 ) . Due to the iterative nature of this procedure , each dataset was run through at least twice to ensure that not only were the determinations of presence/absence accurate , but that the marked locations of every feature were optimal . Spines that were observed consistently across all time points were classified as stable . Spines that disappeared at some time point and were observed again in a subsequent time point were classified as recurrent . Spines that disappeared and never reappeared were classified as transient . To compare long-term dynamics between WT and Gfap Smo CKO mice , we examined the survival curves , S ( t ) , for each mouse , where:S ( t ) =#ofspinesconsistentlypresenttotimepoint/#ofspinesinitiallyobserved Survival curves for each mouse were pooled by genotype and fit to a single-phase exponential decay model:S ( t ) =Sp+Si∗exp ( −t/τ ) where Sp and Si represent the relative fractions of permanent and impermanent spines respectively and τ is the characteristic lifespan of impermanent spines . Fitting and statistical comparison of survival data was performed with PRISM 6 ( GraphPad , San Diego , CA ) software . To analyze morphology , we examined data from the first imaged day , and classified individual protrusions as filopodia , intermediate , or mushroom spines . For each protrusion , a mean projection image was generated , averaging the frame on which the protrusion was manually marked in the original analysis and the two neighboring frames ( 1 above and one below , effectively averaging over 1 . 5 um of depth in the z axis ) . Each image was cropped , centering around the protrusion's XY coordinates and a marker was drawn on the image to indicate which protrusion was under consideration . These images were then presented to the scorer sequentially for scoring . Protrusions exhibiting a bright , bulbous head were scored as mushroom . Thin protrusions exhibiting a relatively dim , small head , or no head , were scored as intermediate . Dim , elongated protrusions that lacked any head were scored as filopodia . Mice were deeply anesthetized by intraperitoneal injection of ketamine/xylazine/acepromezine before rapid decapitation . Cortices were dissected into TRIzol reagent ( Thermo-Fisher , Waltham , MA ) and RNA was extracted according to the standard protocol . RNA was then purified with an RNeasy Micro Kit ( Qiagen , Hilden , Germany ) and subsequently reverse transcribed to cDNA using the High-Capacity cDNA Reverse Transcription Kit ( Thermo-Fisher ) . Droplet digital PCR ( ddPCR ) was performed with the QX200 Droplet Digital PCR System using Evagreen Supermix ( Bio-Rad , Hercules , CA ) . Alternatively , for Megf10 and Mertk , qPCR was performed with the CFX96 Touch Real-Time PCR Detection System ( Bio-Rad ) using PowerUp SYBR Green Master Mix ( Thermo-Fisher ) . PrimePCR ddPCR Expression EvaGreen Assay ( Bio-Rad ) was used for Gapdh primers; all other primers were designed using NCBI Primer-BLAST as follows: Kcnj10 F , GCTGCCCCGCGATTTATCAG; Kcnj10 R , AGCGACCGACGTCATCTTGG; Glast F , TCCTCTACTTCCTGGTAACCC; Glast R , TCCACACCATTGTTCTCTTCC; Glt1 F , CATCAACAGAGGGTGCCAAC; Glt1 R , CACACTGCTCCCAGGATGAC; Megf10 F , CTCACTGCTCTGTCACTGGGTG; Megf10 R , GGTAGCTGATTCTGTGCCGTGT; Mertk F , AAACTGCATGTTGCGGGATGAC; Mertk R , TCCCACATGGTCACGCCAAA . ddPCR absolute quantification of targets was calculated using Quantasoft Version 1 . 7 software ( Bio-Rad ) . Samples were run in triplicate , and Gapdh was quantified for each sample as a loading control with no difference detected between sample groups . Mice aged P21 were anesthetized with intraperitoneal injection of euthasol ( 0 . 2 ml/kg ) and brains were rapidly removed . 300 µm coronal slices were cut using a vibratome tissue slicer ( Leica ) and transferred to a holding chamber , submerged in oxygenated artificial cerebrospinal fluid ( ACSF , in mM: 124 NaCl , 2 . 5 KCl , 1 . 25 NaH2PO4 , 2 CaCl2 , 1 MgSO4 , 26 NaHCO3 , and 10 dextrose , pH 7 . 4 ) at 36°C for 1 hr and then maintained at room temperature . Individual slices containing the barrel cortex were placed into a recording chamber immersed in oxygenated ACSF mounted on an Olympus upright microscope ( BX51 ) . Neurons were visualized with infrared differential interference video microscopy . Whole-cell current clamp was used to record action potentials from layer V pyramidal cells using patch electrodes with an open tip resistance of 7–10 MΩ . Patch electrodes were filled with potassium gluconate internal solution ( in mM ) : 120 potassium gluconate , 20 KCl , 4 ATP-Na , 0 . 3 Na2GTP , 5 Na-phosphocreatine , 0 . 1 EGTA , 10 HEPES , pH 7 . 3 , 305 mOsmol/l ) and action potential responses were measured in response to various step currents from −300 pA to +650 pA with 50 pA increments . Action potentials were described as spike numbers per depolarized current injections . The resting membrane potential , input resistance , action potential ( AP ) threshold , AP half-width , and peak AP amplitude were also measured . Whole-cell voltage-clamp recordings were obtained from layer V pyramidal cells using patch electrodes with Cs+-containing solution ( in mM ) : 120 Cs-gluconate , five lidocaine , 6 CaCl2 , 1 Na2ATP , 0 . 3 Na2GTP , and 10 HEPES ( pH 7 . 3 adjusted by CsOH ) . To record spontaneous excitatory postsynaptic currents ( sEPSCs ) , cells were held at a membrane potential of −70 mV in the presence of GABAA receptor antagonist picrotoxin ( PTX; 100 µM , Sigma-Aldrich ) and recorded for 5 min . Miniature EPSCs ( mEPSCs ) were then recorded for an additional 5 min in the presence of both picrotoxin and tetrodotoxin ( TTX , 0 . 5 μM , Hello Bio , Princeton , NJ ) . All recordings were conducted with Axon MultiClamp 700B amplifier ( Molecular Devices , San Jose , CA ) . The sEPSCs and mEPSCs frequency and amplitudes were measured by averaging five sweeps from the onset of recording with Clampfit 9 . 2 software ( Molecular Devices ) . The s/mEPSCs were detected and characterized using a sample template for the 5 min data period . Based upon the selected sample template , the frequency ( number of event detections ) and amplitude of events were measured using a threshold set in Clampfit . Statistical analyses used for various datasets are indicated both in the figure legends and in the text . Prism six software ( GraphPad ) was used for all analyses and to generate graphs . | A central system of neurons in the spinal cord and brain coordinate most of our body’s actions , ranging from regulating our heart rate to controlling our movement and thoughts . As the brain develops , neurons form specialized contacts with one another known as synapses . If the number of synapses is not properly regulated this can disrupt communication between the neurons , leading to diseases like schizophrenia and autism . As the brain develops , it first forms an excess of synapses and later eliminates unnecessary or weak connections . Various factors , such gene expression or a neuron’s level of activity , regulate this turnover process . However , neurons cannot do this alone , and rely on other types of cells to help regulate their behavior . In the central nervous system , for example , a cell called an astrocyte is known to support the formation and activity of synapses . Now , Hill and Blaeser et al . show that astrocytes also exert influence over synaptic turnover during development , leading to long lasting changes in the number of synapses . Hill , Blaeser et al . revealed that disrupting activity of the signaling pathway known as Sonic hedgehog , or Shh for short , in the astrocytes of mice led to disordered synaptic connections . Notably , neurons produce Shh , suggesting that neurons use this signaling pathway to communicate to specific astrocyte partners . Further experiments showed that reducing astrocyte’s ability to respond to Shh impaired synaptic turnover as the brain developed , leading to an overabundance of synapses . Importantly , these effects were only found to influence neuron populations associated with astrocytes that actively use Shh signaling . This suggests that distinct populations of neurons and astrocytes interact in specialized ways to build the connections within the nervous system . To address how astrocytes use Shh signaling to regulate synaptic turnover , Hill , Blaeser et al . examined gene expression changes in astrocytes that lack Shh signaling . Astrocytes with a reduced capacity to respond to Shh were found to have lower levels of a protein responsible for transporting potassium ions into and out of the cell . This impairs astrocyte’s ability to regulate neuronal activity , which may lead to a failure in eliminating unnecessary synapses . Understanding how synapses are controlled and organized by astrocytes could help identify new ways to treat diseases of the developing nervous system . However , further studies would be needed to improve our understanding of how this process works . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2019 | Sonic hedgehog signaling in astrocytes mediates cell type-specific synaptic organization |
Most kinesins transport cargoes bound to their C-termini and use N-terminal motor domains to move along microtubules . We report here a novel function for KIF1C: it transports Rab6A-vesicles and can influence Golgi complex organization . These activities correlate with KIF1C's capacity to bind the Golgi protein Rab6A directly , both via its motor domain and C-terminus . Rab6A binding to the motor domain inhibits microtubule interaction in vitro and in cells , decreasing the amount of motile KIF1C . KIF1C depletion slows protein delivery to the cell surface , interferes with vesicle motility , and triggers Golgi fragmentation . KIF1C can protect Golgi membranes from fragmentation in cells lacking an intact microtubule network . Rescue of fragmentation requires sequences that enable KIF1C to bind Rab6A at both ends , but not KIF1C motor function . Rab6A binding to KIF1C's motor domain represents an entirely new mode of regulation for a kinesin motor , and likely has important consequences for KIF1C's cellular functions .
Kinesin superfamily proteins ( KIFs ) are microtubule-based motors that are responsible for the motility of membrane-bound compartments and transport vesicles ( Hirokawa et al . , 2009b; Verhey and Hammond , 2009 ) . Of fundamental interest is how these motor proteins link to specific membrane cargoes and how they are regulated . Rab GTPases represent a family of more than 60 human proteins that mark distinct membrane-bound compartments and function in transport vesicle formation , motility , docking , and fusion ( Stenmark , 2009; Hutagalung and Novick , 2011 ) . Rabs help connect motors to their cargoes , usually via an intermediary linking protein . For example , the Rab27 Slac2 effectors recruit myosin Va ( reviewed by Fukuda , 2013 ) , Rab3 effector , DENN/MADD links KIF1β and KIF1A to Rab3 on synaptic vesicles ( Niwa et al . , 2008 ) and both Rab6 and Rab7 interact with cytoplasmic dynein via the dynactin complex ( Short et al . , 2002 ) , bicaudal-D ( Matanis et al . , 2002 ) , or RILP ( Jordens et al . , 2001 ) proteins . KIF5B also links to Rab6-containing membranes via the Rab6 effector , bicaudal-D2 ( Grigoriev et al . , 2007 ) . Rab6 binds to myosin II ( Miserey-Lenkei et al . , 2010 ) and Rab5 GTPase participates indirectly in the recruitment of the plus-end directed kinesin , KIF16B to early endosomes ( Nielsen et al . , 1999; Hoepfner et al . , 2005 ) . KIF1C is a member of the Kinesin-3 family that includes the Unc104/KIF1A motor that transports synaptic vesicles to growth cones ( Hirokawa et al . , 2009b ) . KIF1C has been reported to be a Golgi-localized , tyrosine phosphorylated protein that interacts with the protein tyrosine phosphatase PTPD1 ( Dorner et al . , 1998 ) and bicaudal-D-related protein 1 ( BICDR-1 ) ( Schlager et al . , 2010 ) . Phosphorylation of a carboxy-terminal serine allows binding to 14-3-3 proteins ( Dorner et al . , 1999 ) . KIF1C was initially reported to participate in the transport of proteins from the Golgi to the endoplasmic reticulum ( ER; Dorner et al . , 1998 ) , but subsequent gene disruption in mice yielded animals with no apparent abnormalities , and fibroblasts from these mice showed normal Golgi to ER transport ( Nakajima et al . , 2002 ) . More recent studies have shown that KIF1C acts to regulate podosome dynamics in macrophages ( Kopp et al . , 2006; Efimova et al . , 2014; Bhuwania et al . , 2014 ) and is also important in vesicle transport in neurons ( Schlager et al . , 2010 ) , MHC class II antigen presentation in myeloid cells ( del Rio et al . , 2012 ) , and α5β1-integrin transport ( Theisen et al . , 2012 ) . Consistent with these findings , KIF1C was identified in a genome-wide screen for proteins important for VSV-G transport to the cell surface; its depletion also led to fragmentation of the Golgi ribbon as monitored by GM130 localization ( Simpson et al . , 2012 ) . We show here that Rab6A regulates the function of KIF1C by direct interaction with both KIF1C's C-terminal cargo binding domain and , more surprisingly , with its N-terminal motor domain . Rab6A binding to the motor domain blocks KIF1C interaction with microtubules and inhibits KIF1C motility . We confirm that depletion of KIF1C leads to fragmentation of the Golgi and show that while both N- and C-terminal Rab6A binding sites are required for KIF1C rescue , KIF1C motor activity is not: neither ATP nor microtubule binding is required . Finally , Rab6A-decorated vesicles are less confined and less directed in cells depleted of KIF1C , consistent with its role as a Rab-regulated motor that aids in intra- and post-Golgi vesicle transport . These data reveal a novel form of motor regulation with unexpected consequences for motor function in Golgi organization .
KIF1C is comprised of an N-terminal motor domain that is highly homologous to KIF1A and KIF1B , followed by several coiled coil stretches that are interrupted by a Forkhead homology domain ( FHA; Figure 1A ) . KIF1C was identified as a protein tyrosine phosphatase D1 ( PTPD1 ) binding partner whose binding domain is located between the third and fourth coiled coil ( Dorner et al . , 1998 ) . The KIF1C C-terminal domain binds both 14-3-3 proteins ( Dorner et al . , 1999 ) and the Rab6A effector , bicaudal-D-related protein 1 ( Schlager et al . , 2010 ) . 10 . 7554/eLife . 06029 . 003Figure 1 . KIF1C binds Rab6A GTPase . ( A ) KIF1C schematic showing the N-terminal motor ( red ) , predicted coiled-coil ( cc , gray ) , Forkhead homology ( FHA , blue ) , protein tyrosine phosphatase D1 binding ( PTPD1 , orange ) , and C-terminal Rab binding ( CBD , yellow ) domains ( AA 1060-1103 ) . ( B ) KIF1C co-immunoprecipitation by Rab6A . HEK293 cells were transfected with GFP-Rab6A , myc-KIF1C , or both . Lysates were immunoprecipitated with llama GFP-binding protein and immunoblotted with anti-myc antibody . Left panel , total soluble cell lysates , 5%; right panel , bound fraction , 33% . ( C ) Domain specificity of Rab6A binding . His-Rab6A-35S-GTPγS ( black bars ) or His-Rab1A-35S-GTPγS ( gray bars ) ( 1 μM ) binding to GST-KIF1C constructs ( 15 μM ) pulled-down with glutathione Sepharose and quantified by liquid scintillation counting ( error bars = SD ) . ( D ) Nucleotide specificity of Rab6A binding . Binding of His-Rab6A or His-Rab9A ( 500 nM ) preloaded with 35S-GTPγS or 3H-GDP to GST or GST-KIF1C CBD ( 15 μM ) was assayed as in C ( error bars = SD ) . ( E ) His-Rab6A Q72L binds to GST-KIF1C CBD in a concentration-dependent manner . GTPγS-loaded His-Rab6A Q72L ( 0 . 58 μM ) binding to GST-KIF1C CBD ( immobilized on glutathione Sepharose ) as determined by quantitative fluorescent antibody immunoblot , presented as a fraction of maximal binding ( 2 . 2% of total ) . Data were fit using GraphPad Prism software . DOI: http://dx . doi . org/10 . 7554/eLife . 06029 . 003 We identified KIF1C in a two hybrid screen for Rab GTPase binding partners ( Reddy et al . , 2006 ) . More detailed studies showed that transiently expressed , full-length myc-KIF1C and GFP-Rab6A could be co-immunoprecipitated ( Figure 1B ) . However , this interaction may have been indirect as BICDR-1 is both an effector of Rab6A and a KIF1C binding partner ( Schlager et al . , 2010 ) . Direct and specific binding between Rab6A and KIF1C was tested using purified GST-constructs . As shown in Figure 1C , KIF1C's C-terminal 40 amino acids , termed the C-terminal binding domain ( CBD; Figure 1A ) , bound to purified His-Rab6A protein ( Figure 1C–E ) . Binding to His-Rab6A by N-terminally truncated KIF1C was observed only with constructs containing the CBD ( Figure 1C ) . Importantly , binding of the CBD required that His-Rab6A be in its active , GTP-bound state , as would be expected for a Rab effector ( Figure 1D ) ; binding was also Rab6A-specific as neither His-Rab1A ( Figure 1C ) nor His-Rab9A ( Figure 1D ) bound this domain strongly . His-Rab6A binding to the CBD was saturable and concentration dependent with a KD of 0 . 9 µM , consistent with the affinity of most Rab protein:effector interactions ( Burguete et al . , 2008 ) . To verify whether the CBD was the only site of Rab6A interaction , we tested Rab6A binding to KIF1C lacking the CBD ( Figure 2A , KIF1C ΔCBD ) , synthesized by in vitro transcription and translation . ( Significant amounts of active full-length protein could not be obtained upon expression in bacteria , consistent with work from Hirokawa on KIF1A motor protein [Nitta et al . , 2004] . ) Surprisingly , deletion of the CBD in the context of full-length KIF1C did not abolish binding to GST-Rab6A ( Figure 2B ) . Similarly , KIF1C constructs containing the first 500 , 400 ( data not shown ) , or 350 amino acid residues comprising only the motor domain ( KIF1C-350 ) bound specifically to GST-Rab6A ( Figure 2C , left ) . In contrast , a KIF1C construct lacking both the motor domain and the CBD ( Δmotor ΔCBD ) failed to bind GST-Rab6A ( Figure 2C right ) . KIF1C motor domain binding to Rab6A was specific as no significant binding was seen with either GST-Rab9A or GST-Rab5A proteins ( Figure 2D ) . In concentration-dependent binding analyses , Rab6A was capable of binding up to 25% of in vitro translated KIF1C motor domain , suggesting only a quarter of the molecules synthesized by in vitro translation were active . These experiments yielded an apparent KD = 0 . 23 µM ( Figure 2E ) , strong for a Rab:effector interaction . These data show that Rab6A binds specifically to KIF1C at both its N- and C-termini . 10 . 7554/eLife . 06029 . 004Figure 2 . Rab6A binds KIF1C at two locations . ( A ) KIF1C schematic showing CBD truncation ( ΔCBD , AA 1-1060 ) , N-terminal motor domain ( 1–350 ) , and a construct lacking both motor domain and CBD ( ΔmotorΔCBD , AA 450-1060 ) . ( B ) Binding of in vitro translated myc-KIF1CΔCBD to GTPγS-loaded GST-Rab6A Q72L ( 5 μM ) and pulled down using glutathione Sepharose . Left , input ( 1% ) compared to bound fraction ( 50% ) , right . ( C ) Binding of in vitro translated myc-KIF1C constructs to GTPγS-loaded GST-Rab6A Q72L ( 0 . 2 μM ) . Input ( 7% ) , left , compared to bound ( 50% ) , right . ( D ) Rab specificity of motor domain binding . GTPγS-preloaded GST-Rab6A Q72L , Rab9A , and Rab5A Q79L ( 5 μM ) incubated with in vitro translated myc-KIF1C motor domain . Input , 1% , on left compared to bound fraction ( 48% ) on right . ( E ) myc-KIF1C-350 binds to GST-Rab6A Q72L in a concentration-dependent manner . In vitro translated myc-KIF1C motor domain binding to GTPγS-loaded GST-Rab6A , presented as a fraction of maximal binding detected ( 25% of total ) . Data were fit using KaleidaGraph software . In vitro translation optimally yields a reaction concentration of 11 . 25 nM product . DOI: http://dx . doi . org/10 . 7554/eLife . 06029 . 004 Because of the novelty of a Rab GTPase–motor domain interaction , it was essential to determine if Rab6A binding to the motor domain is direct . For this purpose , we took advantage of the strategy employed by Hirokawa and coworkers in their studies of KIF1A to create a stable KIF1C motor domain ( Nitta et al . , 2004 ) . Specifically , we expressed in bacteria a construct comprised of KIF1C residues 1–349 followed by 6 residues ( 329–334 ) of KIF5C heavy chain ( plus 6 His residues; ‘KIF1C-355-His’ ) . This KIF1C motor domain was purified by Ni-NTA-chromatography followed by ion exchange chromatography on Q Sepharose FF ( Figure 3A ) . This method yielded a single , ∼40 kD polypeptide upon Coomassie-stained SDS-PAGE . 10 . 7554/eLife . 06029 . 005Figure 3 . Rab6A binds directly to the KIF1C motor domain . ( A ) Coomassie-stained SDS-PAGE of bacterially expressed , KIF1C-355-His motor domain purification using Ni-NTA followed by Q Sepharose FF . Lane 1 , flowthrough; lane 2 , wash; lane 3–5 , NaCl gradient elution . Protein from lanes 3 and 4 were used in subsequent experiments . ( B ) KIF1C-355-His binds to GST-Rab6A Q72L in a concentration-dependent manner . Purified KIF1C-355-His motor domain ( Figure 3A , lane 4; 78 . 8 nM ) binding to GTPγS-loaded GST-Rab6A Q72L ( immobilized on glutathione Sepharose ) as determined by quantitative fluorescent antibody immunoblot , presented as a fraction of maximal binding detected ( 61% of total ) . Data were fit using GraphPad Prism . ( C ) Immunoblot determination of binding of purified KIF1C-355-His ( 0 . 56 µM ) to GTPγS-loaded GST-tagged QL mutant Rabs ( 0 . 2 μM ) after collection on glutathione Sepharose . Input , 20% of sample; bound fraction , 37 . 3% of sample . Bottom panels show Ponceau S-staining to detect glutathione resin-bound and eluted Rabs . ( D ) Binding specificity of GST-Rab6A mutants T27N , GDP-preferring , and Q72L , GTP-hydrolysis deficient ( 0 . 25 μM ) to purified KIF1C-355-His ( 45 nM ) quantified as in B , as a percentage of maximal binding ( 69 . 1% of total ) . ( E ) Nucleotide binding specificity of GST-Rab6A Q72L ( 2 . 5 μM ) to purified KIF1C-355-His ( 160 nM ) in the presence of AMP-PNP or ADP as quantified by fluorescent antibody immunoblot as a percentage of maximal binding ( 37% of total ) . Error bars represent SD . Mobility of marker proteins is shown in KD . DOI: http://dx . doi . org/10 . 7554/eLife . 06029 . 005 Binding of this recombinant KIF1C-355-His motor domain to Rab GTPases was tested using GST-tagged Rab proteins bound to glutathione resin . Quantitative analysis of Rab6A binding to the E . coli-produced motor domain showed that binding was of high affinity ( Figure 3B ) , essentially identical with that measured for KIF1C motor domain produced by in vitro translation ( Figure 2 ) . As shown in Figure 3C , recombinant KIF1C motor domain bound directly to GST-Rab6A but not to GST-Rab5A . Moreover , binding was reduced with Rab6A protein lacking its C-terminal hypervariable domain . We have shown previously that hypervariable domains are important for the binding of many Rab effectors ( Aivazian et al . , 2006; Burguete et al . , 2008 ) . GST-Rab6A exchanged bound GTP less readily than His-Rab6A , making it difficult to compare His-tagged motor binding to Rab6A in GDP vs GTP without losing overall Rab6A activity . Instead , we compared motor domain binding for the active , Rab6A hydrolysis-deficient mutant ( GST-Rab6A Q72L ) vs the inactive , GDP-preferring form ( GST-Rab6A T27N ) and found strong preference for the active Rab6A mutant ( Figure 3D ) . We also tested Rab discrimination of the motor's nucleotide state: GST-Rab6A showed strong preference for the KIF1C motor domain with AMP-PNP bound , compared with ADP ( Figure 3E ) . Thus , Rab6A binds with preference to KIF1C's strong microtubule-binding state . Crystal structures of the KIF1A motor domain with either AMP-PNP or ADP bound indicate a change in the projection angle of loop 10 ( Kikkawa and Hirokawa , 2006 ) , which as described below , may explain the strong preference of Rab6A for AMP-PNP bound-KIF1C . These data demonstrate that active Rab6A can bind the KIF1C motor domain strongly and directly , without adaptor proteins , with preference for KIF1C in its strong microtubule-binding state . KIF1A and KIF1C motor domains are 81% identical ( Figure 4—figure supplement 1B ) . When the KIF1C sequence is superimposed onto the structure of KIF1A bound to tubulin ( PDB , 2HXH , Figure 4—figure supplement 1A , Kikkawa and Hirokawa , 2006 ) , the sequence differences localize primarily to loops 2 , 3 , 6 , and 10 , which are positioned away from the microtubule-binding interface ( Figure 4—figure supplement 1A ) . To facilitate identification of KIF1C sequences needed for Rab6A binding , we tested whether the KIF1A motor domain binds to Rab6A . Figure 4A , C show that Rab6A bound to the KIF1A motor domain much more weakly ( if at all ) , compared with KIF1C . We thus created motor domain chimeras containing N-terminal portions of the KIF1A motor domain fused to the C terminus of the KIF1C motor domain ( KIF1A/1C ) as well as the reverse chimera , KIF1C/1A , to narrow down the Rab6A interaction site ( Figure 4B , Figure 4—figure supplement 1B ) . The C-terminal portion of KIF1C restored binding to Rab6A in the KIF1A/1C chimera , however , the KIF1C/1A chimera showed diminished binding ( Figure 4A ) . 10 . 7554/eLife . 06029 . 006Figure 4 . KIF1C loops 6 and 10 are necessary for Rab6A binding . ( A ) Binding of in vitro translated myc-KIF1A ( 1–361 ) , myc-KIF1A/1C , myc-KIF1C/1A , and myc-KIF1C-350 motor domain chimera constructs to GTPγS-loaded GST-Rab6A Q72L ( 0 . 2 μM ) presented as a percentage of maximal KIF1C bound ( 24 . 7% of input ) . ( B ) Crystal structure of KIF1A showing the regions swapped between KIF1A and KIF1C . The C-terminal portion is colored red . ( C ) Binding of in vitro translated myc-KIF1A ( 1–361 ) , myc-KIF1C-350 Loop 2/3 swap , myc-KIF1C-350 Loop 6/10 swap , myc-KIF1C-350 , and myc-KIF1C-350 K103A constructs to GTPγS-loaded GST-Rab6A Q72L ( 0 . 2 μM ) presented as a percent of myc-KIF1C-350 bound ( 35 . 3% of input ) . ( D ) Predicted crystal structure of KIF1C with Loop 2/3 ( at left ) , 6/10 ( at right ) labeled in red . The KIF1C sequence was overlaid onto the KIF1A crystal structure ( PDB 2ZFI ) using PHYRE2 . ( E ) Binding of endogenous Rab6A to transiently expressed CFP-KIF1C-350 , CFP-KIF1C-350 Loop 6/10 swap , and GFP immunoprecipitated with llama GFP-binding protein , as determined by anti-Rab6A immunoblot . Total endogenous Rab6A ( 2% ) at left compared to bound ( 33% ) at right . Below , bound CFP-KIF1C and GFP as measured by Ponceau S staining . ( F ) Molecular docking of Rab6A onto the predicted structure of KIF1C . The predicted structure ( blue ) was docked to the crystal structure of Rab6A ( PDB 2Y8E , gray ) using ClusPro2 . The model for the largest cluster containing 144 members is shown . The switch regions of Rab6 are labeled in yellow and loop 6 and 10 of KIF1C are labeled in purple . Error bars represent SD . DOI: http://dx . doi . org/10 . 7554/eLife . 06029 . 00610 . 7554/eLife . 06029 . 007Figure 4—figure supplement 1 . The motor domains of KIF1A and KIF1C share 81% identity . ( A ) Structure of KIF1A motor domain bound to tubulin dimer ( PDB 2HXH ) . Loops 2 , 3 , 6 , and 10 are highlighted . ( B ) Alignment of human KIF1C ( top ) and KIF1A ( bottom ) with boxes around regions corresponding to the loop mutants generated and indicated in A . The indicated break demarcates the region swapped between KIF1A and KIF1C to generate chimeric proteins . DOI: http://dx . doi . org/10 . 7554/eLife . 06029 . 00710 . 7554/eLife . 06029 . 008Figure 4—figure supplement 2 . Mutations in the KIF1C K-loop do not affect KIF1C loop 6/10 mutant motor domain microtubule localization . HeLa cells transfected with the indicated CFP-KIF1C 350 construct were precipitated in MeOH and imaged by immunofluorescence microscopy . Two cells each are shown . The NRSK mutant changes 291KKRK to 291NRSK . The GTKT mutant changes 288MQSKKRKSD to 288GTKT , which mimics the absence of the K-loop and makes the mutant more similar to conventional kinesin , KIF5C ( Okada and Hirokawa , 2000 ) . Tubulin staining is on the left . Scale bar is 20 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 06029 . 008 We next generated KIF1C chimeras containing just the divergent loop sequences derived from KIF1A . Consistent with the Rab6A binding capacity of the KIF1C/1A chimera , conversion of KIF1C's loops 6 and 10 into KIF1A sequences yielded a ‘loop mutant’ protein that showed significantly reduced Rab6A binding ( Figure 4C , D ) . In contrast , conversion of loops 2 and 3 , in the N-terminal portion , was of lower consequence ( Figure 4C , D ) . Mutation of either loop 6 or 10 alone did not completely abolish binding , suggesting that both regions form contacts with Rab6A ( data not shown ) . These regions are spatially adjacent in the KIF1A crystal structure and are likely also similarly situated in KIF1C ( Figure 4D ) . Importantly , the KIF1C motor domain loop 6/10 mutant that did not bind Rab6A retained its ability to bind microtubules ( Figure 4—figure supplement 2 ) , indicative of proper protein folding . Furthermore , microtubule localization was not just from alternate kinesin 3 family ‘K-loop’ binding , as KIF1C loop 6/10 mutant 350 with further mutations in the K-loop ( NRSK or GTKT ) were still able to localize to microtubules ( Figure 4—figure supplement 2 ) . These in vitro binding results were confirmed by expression of KIF1C wild-type or loop mutant motor domain proteins in HEK293 cells . The wild-type motor domain was able to co-immunoprecipitate endogenous Rab6A protein , whereas similar expression of the KIF1C loop 6/10 motor domain mutant did not ( Figure 4E ) . These data show that Rab6A requires loops 6 and 10 to interact with the KIF1C motor domain . Structure prediction of the KIF1C motor domain was accomplished using PHYRE2 by taking advantage of the known KIF1A crystal structure ( PDB 2ZFI , 1 . 55 Å resolution , PHYRE2 confidence = 100% ) ( Figure 5D; Nitta et al . , 2008; Kelley and Sternberg , 2009 ) . It is noteworthy that sequences surrounding and contributing to loop 10 are predicted to be more structured in the KIF1C model ( Figure 4D ) than in the KIF1A structure ( Figure 4B ) . 10 . 7554/eLife . 06029 . 010Figure 5 . Rab6A inhibits KIF1C microtubule co-sedimentation . ( A ) Binding of full-length KIF1C to microtubules in the presence of Rab6A . In vitro synthesized 35S-myc-KIF1C was desalted , incubated with GTPγS-preloaded His-tagged Rabs ( 4 . 2 μM ) , and then with 0 . 8 µg/µl Paclitaxel stabilized pre-polymerized microtubules in 2 . 5 mM ADP and 0 . 5 mM GTPγS . Reactions were centrifuged through a 10% sucrose cushion and pellets were analyzed by scintillation counting . The fraction of full-length KIF1C cosedimenting with microtubules in the presence of the indicated Rabs is shown ( error bars = SE [n ≥ 2] ) . ( B ) Rab6 affects KIF1C motor domain microtubule co-sedimentation in a concentration-dependent manner . KIF1C-355-His ( 160 nM ) was incubated with increasing concentrations of His-Rab6A Q72L , and then with 2 . 1 µM Paclitaxel stabilized pre-polymerized microtubules in 2 . 6 mM ADP and 0 . 35 mM GTPγS . Reactions were centrifuged through a 35% sucrose cushion . Pellets were analyzed by fluorescent antibody immunoblot . ( C ) Rab6A affects the strong microtubule binding state of KIF1C . Purified KIF1C-355-His ( 80 nM ) was incubated with His-Rab6A Q72L ( 4 . 86 µM ) or BSA ( 7 . 6 µM ) , and then with increasing concentrations of microtubules in 2 . 6 mM AMP-PNP and 0 . 35 mM GTPγS . Samples were processed and analyzed as in B . ( D ) Rab6A affects the weak microtubule binding state of KIF1C . Purified KIF1C-355-His ( 160 nM ) was incubated with His-Rab6A Q72L , His-Rab33 , or BSA ( 3 . 42 µM ) , and then with increasing concentrations of microtubules in 2 . 6 mM ADP and 0 . 35 mM GTPγS . Data were fit using GraphPad Prism software . The fraction of motor sedimented is normalized to the amount of microtubules pelleted , determined by Coomassie blue staining . DOI: http://dx . doi . org/10 . 7554/eLife . 06029 . 01010 . 7554/eLife . 06029 . 011Figure 5—figure supplement 1 . Rab6A inhibits KIF1C binding and motility . KIF1C ( 9 nM ) , labeled with rabbit anti-KIF1C and Dylight 649 anti-rabbit Fab , was incubated with Rhodamine labelled-microtubules ( 7 . 5 µg/ml ) adhered to coverslips with anti-β tubulin ± Rab6A ( 15 µM ) and imaged by TIRF microscopy ( on average 14 microtubules/75 s/region , 15 regions per condition quantified ) . ( A ) Images from a single frame ± Rab6A . Top panels show microtubule-localized KIF1C molecules ± His-Rab6A; bottom shows KIF1C ( green ) on microtubules ( red ) . Scale , 10 μm . ( B ) Rab6A decreases the number of motors on microtubules normalized by microtubule length ( p < 0 . 0001 ) . ( C ) Rab6A decreases the percentage of moving motors ( p < 0 . 0001 , bars = SE ) . Experiments were done in series so that incubation and imaging times were controlled . Rab6A did not affect KIF1C processivity or speed . DOI: http://dx . doi . org/10 . 7554/eLife . 06029 . 011 In silico molecular docking was performed using ClusPro 2 . 0 to dock the predicted structure onto Rab6A ( PDB 2Y8E ) ( Kozakov et al . , 2010; Walden et al . , 2011 ) . Consistent with our binding studies , the top clustered model showed interaction interfaces between Rab6A switch I and II regions and loops 6 and 10 of KIF1C ( Figure 4F ) . Rab6A was also docked onto KIF1A and KIF1C loop 6/10 mutant mapped onto KIF1A . Notably , while the docking confirmation of wild-type KIF1C and Rab6A involving loops 6 and 10 was highly enriched ( 125% increase over the next conformation ) , the most populated confirmations between Rab6A and KIF1A or KIF1C loop 6/10 mutant was only enriched 2% and 12% , respectively , over their next conformations . While conformation cluster sizes are not definitive , they lend credence to the binding between KIF1C and Rab6A , as well as the importance of loops 6 and 10 . The most obvious explanation for Rab6A binding to the KIF1C motor domain would be to influence KIF1C's microtubule binding and/or motility properties . To test if Rab6A interferes with KIF1C motor domain binding to microtubules , in vitro transcribed and translated , radiolabelled , full-length KIF1C was tested for microtubule binding in the presence and absence of purified Rab6A using a microtubule co-sedimentation assay . As a close family member to KIF1A , full-length KIF1C is likely auto-inhibited and is predicted to interact less stably with microtubules ( Hammond et al . , 2009 ) . Nevertheless , in the presence of His-Rab6A ( but not His-Rab33B ) , the amount of full-length KIF1C bound to microtubules decreased by more than twofold ( Figure 5A ) . Especially dramatic was the ability of His-Rab6A to inhibit binding of the purified KIF1C-355-His motor domain to microtubules: KIF1C motor domain binding to microtubules could be abolished in the presence of increasing concentrations of active His-Rab6A ( Figure 5B ) . Furthermore , His-Rab6A influenced both the AMP-PNP and ADP bound states of KIF1C-355-His; the presence of His-Rab6A increased the KD of KIF1C to microtubules by more than 10 fold for both AMP-PNP and ADP states ( Figure 5C , D ) . The microtubule affinities observed in the absence of Rab6A ( 0 . 46 µM and 2 . 1 µM in AMP-PNP and ADP , respectively ) are also very similar to those reported for KIF1A protein ( Soppina and Verhey , 2014 ) . Importantly , the specificity of this effect was confirmed by the finding that His-Rab33 did not influence the KIF1C-microtubule interaction significantly ( Figure 5D ) . These data demonstrate a new mode of regulation: Rab6A can regulate the binding of KIF1C to microtubules by direct interaction with the KIF1C motor domain . Because Rab6A binds to loop regions of the KIF1C motor domain that face away from the microtubule binding face , inhibition of microtubule binding is likely to be due to a conformational change in KIF1C , rather than direct steric hindrance . Indeed , superposition of the predicted KIF1C structure over the predicted structure of a KIF1C-Rab6A complex indicated changes in regions important for microtubule binding ( not shown; site one of Kikkawa and Hirokawa , 2006 ) . To better understand the molecular mechanism by which Rab6A may regulate KIF1C , we used TIRF microscopy to analyze Rab6A's influence on the attachment of purified full-length KIF1C to microtubules . Again , as KIF1A is auto-inhibited , we expected that the majority of KIF1C would not be bound to microtubules and/or would not be fully processive ( Hammond et al . , 2009 ) . KIF1C molecules were monitored using fluorescently labeled antibody ( Video 1 ) . Example frames from a video of KIF1C motility are shown in the absence ( control ) or presence of Rab6A protein ( Figure 5—figure supplement 1A ) . The identified motors ( green ) that overlaid microtubules ( red ) are highlighted using automated segmentation ( top row ) . 10 . 7554/eLife . 06029 . 012Video 1 . Time-lapse imaging of KIF1C ( green ) motility on microtubules ( blue ) in vitro in the absence ( right ) or presence ( left ) of Rab6 . More KIF1C motors can be seen under control conditions . More motors are moving . The time lapse covers 300 frames ( 12 . 3 s ) and is sped up slightly ( video is 10 s ) . Scale bar , 5 μm . ( QuickTime; 11 . 6 MB ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06029 . 012 Immediately apparent was the decrease in microtubule-associated KIF1C molecules in reactions containing Rab6A . Automated segmentation ( Spot Detector , Olivo-Marin , 2002 ) and tracking ( u-track , Jaqaman et al . , 2008 ) were used to quantify motility to support this conclusion . In a representative experiment , over 14 , 000 motors were quantified for each condition . Consistent with observations obtained from microtubule co-sedimentation , the presence of Rab6A resulted in a 30% decrease in the mean number of motors detected along microtubules: from 0 . 321 to 0 . 225 motors/µm ( p < 0 . 0001 ) ( Figure 5—figure supplement 1B ) . The percentage of moving motors also decreased 60% in the presence of Rab6A ( from 2 . 13% to 0 . 86%; p < 0 . 0001; Figure 5—figure supplement 1C ) . Analysis of the population of moving motors yielded no significant differences in the speed or processivity of those motors in the absence of presence of Rab6A ( distribution of motors overlapped for both metrics , data not shown ) and likely Rab6A only affects microtubule attachment of KIF1C and no other characteristics . In summary , the presence of Rab6A led to both a reduction in the number of motors present on microtubules and a reduction in the percentage of motors that were moving . To determine if the effects of Rab6A on KIF1C could also be detected in cells , we examined the localization of KIF1C constructs in HeLa cells using methanol fixation , which favors cytoskeletal structure ( Figure 6A ) . While both full-length KIF1C and KIF1C-400 were localized on microtubules primarily at the cell periphery , KIF1C-350 localized along microtubules throughout the cell ( Figure 6A ) . Unlike the KIF1C-350 construct , the KIF1C-400 construct contains the neck linker region that in KIF1A is important for regulating microtubule association and likely dimerization ( Hammond et al . , 2009; Hirokawa et al . , 2009a ) . Thus , the localization observed is consistent with earlier findings . As KIF1C-350 displayed an easily monitored distribution and was localized to regions capable of maximal contact with Rab6A , we used this construct to examine the consequences of concurrent Rab6A over-expression . KIF1C-350 wild-type or loop mutant proteins were expressed in Vero cells , with or without Rab6A . As shown in Figure 6B , both the KIF1C wild-type and loop mutant constructs localized to microtubules in the absence of Rab6A ( top two rows ) . However , upon Rab6A co-expression , the overall distribution of wild-type KIF1C-350 over microtubules was lost and the staining was also less continuous ( row 3 , Figure 6B , C ) . Moreover , Rab6A only influenced the localization of the wild-type KIF1C and not the loop mutant ( row 4 , Figure 6B , C ) . These results were quantified using CellProfiler ( Carpenter et al . , 2006 ) by determining the Pearson's correlation coefficient as a measure of the colocalization between KIF1C and microtubules ( Figure 6C , top ) . Co-expression of Rab6A significantly reduced the correlation of wild-type KIF1C-350 and microtubules by 33% ( p < 0 . 001 ) but did not significantly affect the correlation between loop mutant KIF1C and microtubules . Importantly , the overall expression of KIF1C in co-transfected cells did not change in response to Rab6A , as the total intensity of KIF1C , quantified in paraformaldehyde fixed cells to capture the total KIF1C pool , was similar under all transfection conditions ( Figure 6C , bottom ) . These experiments show that Rab6A influences microtubule association of the KIF1C motor domain in cells . 10 . 7554/eLife . 06029 . 009Figure 6 . Rab6A inhibits KIF1C microtubule colocalization in cells . ( A ) Localization of KIF1C in MeOH-fixed HeLa cells transiently transfected with CFP-KIF1C-350 , CFP-KIF1C-400 , or full-length CFP-KIF1C . Scale bar , 10 µm . ( B ) Localization of CFP-KIF1C-350 or CFP-KIF1C-350 Loop 6/10 mutant ± Rab6A in MeOH-fixed Vero cells . Three examples are shown for each condition . Scale bar , 10 µm . Image levels were adjusted identically . ( C ) Mean KIF1C-microtubule co-localization measured by Pearson's correlation over KIF1C segmented objects from cells such as those shown in B ( error bars = SE , >80 cells/condition ) . CFP-KIF1C-350 + Rab6A was statistically different from the other populations ( p < 0 . 001 ) . Below , average total fluorescence intensity of CFP-KIF1C and mCherry-Rab6A from similarly treated cells fixed in paraformaldehyde and quantified ( error bars = SE , >100 cells/condition ) . Differences in total KIF intensity were not statistically significant between any of the populations ( p > 0 . 05 ) . Total Rab6A intensity was not statistically significant between the non-treated populations or between the treated populations ( p > 0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06029 . 009 Rab6A is needed for overall Golgi structure and for transport of proteins to and from the trans-Golgi network ( Storrie et al . , 2012 ) ; KIF1C seems to perform similar roles ( Simpson et al . , 2012 ) . We tested specifically whether KIF1C participates in protein transport from the Golgi to the cell surface using KIF1C siRNA which led to a greater than 90% depletion 72 hr post-transfection ( Figure 7—figure supplement 1A ) . Previous work showed that KIF5B participates in this process and suggested that an additional kinesin contributes as well ( Grigoriev et al . , 2007 ) . The plasma membrane delivery of Vesicular stomatitis virus ( VSV ) G glycoprotein in HeLa cells was monitored using a cell surface antibody-binding assay . Cells expressing YFP-VSV-G ts045 protein were held at 39°C to accumulate this temperature-sensitive protein in the ER , post-synthesis . Cells were then incubated at 32°C to release the block and permit cell surface delivery . VSV-G protein began to appear at the cell surface within 35 min after release of the block in control cells ( Figure 7—figure supplement 1B ) . In cells depleted of KIF1C , however , cell surface delivery was significantly slowed ( Figure 7—figure supplement 1B ) and resulted in a 4 . 5-fold decrease in the overall delivery rate . Thus , KIF1C participates in cell surface delivery of proteins from the Golgi complex , consistent with work from other labs . As KIF1C participates in VSV-G transport , we explored the role of KIF1C in Golgi-derived vesicle transport using time-lapse epi-fluorescence microscopy of live cells expressing GFP-Rab6A , in the presence or absence of KIF1C ( Video 2 ) . GFP-Rab6A-labeled vesicles were segmented and tracked using u-track 2 . 1 . 0 ( Jaqaman et al . , 2008 ) . Motion analysis was then used to categorize vesicles as being either confined ( undergoing only localized random motion ) or linearly moving . 10 . 7554/eLife . 06029 . 013Video 2 . Live cell imaging of pEGFP-Rab6 transfected Vero cells after KIF1C- ( left ) or control- ( right ) siRNA treatment . The large bright mass near the center is the Golgi and vesicles are seen as smaller punctate structures . The time lapse covers 71 frames ( 142 s ) and is sped up 14 fold . Scale bar , 10 μm . ( QuickTime; 5 . 7 MB ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06029 . 013 While the absolute fraction of vesicles characterized as being ‘confined’ did not change in the presence or absence of KIF1C siRNA ( not shown ) , confined vesicles in KIF1C-depleted cells occupied , on average , a twofold larger area when compared with the same population in control-depleted cells ( Figure 7A , C ) . KIF1C depletion also affected Rab6A vesicles that were moving linearly: life histories of linearly moving vesicles showed that KIF1C depletion yielded more backwards movement events ( Figure 7B , E ) although no statistically significant differences in vesicle stalling were detected ( 0 . 084 vs 0 . 10 pauses per second , ± KIF1C siRNA , respectively ) . Similar findings were reported by Grigoriev et al . ( 2007 ) for neuropeptide Y vesicle motility in Rab6A-depleted cells . 10 . 7554/eLife . 06029 . 014Figure 7 . Loss of KIF1C affects Rab6A vesicle motility . Vero cells transfected with control- ( 30 cells ) or KIF1C- ( 29 cells ) siRNA followed by pEGFP-Rab6A ( ∼300 vesicles imaged per cell ) . U-track 2 . 1 . 0 was used to segment , track , and characterize vesicles . Motion analysis was used to segment tracks into confined ( >750 per condition ) and linear populations ( >70 per condition ) . ( A ) Traces of confined tracks in representative KIF1C- and control-siRNA cells . One frame from pEGFP-Rab6A videos ( cells outlined in white ) was overlaid with the identified confined tracks ( blue ) . The confinement area for the random motion of each track is outlined ( white circles ) . Scale , 10 µm . ( B ) Life history plots of linearly moving tracks from cells transfected with the indicated siRNAs . Five tracks were taken randomly for each population . ( C ) Quantitation of the confinement area as shown in A , error bars = SE ( p < 0 . 05 ) . ( D ) Quantitation of the speed of linear tracks as shown in B . The speed of each track was computed as the mean of the track's displacement divided by time between frames; error bars = SE ( p < 0 . 0001 ) . ( E ) Mean number of direction changes per second in linear tracks . Principle component analysis was used to find the main components of each track . Direction changes were tabulated each time a track shifted from positive to negative along the main component excluding periods of pausing ( speeds less than 0 . 1 µ/s ) , error bars = SE ( p < 0 . 0001 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06029 . 01410 . 7554/eLife . 06029 . 015Figure 7—figure supplement 1 . Depletion of KIF1C impairs Golgi-to-cell surface transport . ( A ) Immunoblot from siRNA-treated HeLa cells . ( B ) Surface VSV-G protein normalized to total VSV-G of the sample in KIF1C- ( gray square ) or control- ( black circle ) siRNA-treated HeLa cells , n = 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 06029 . 015 While the total distance traversed did not change significantly upon KIF1C depletion ( data not shown ) , the mean speed of Rab6A vesicles increased by 31% ( Figure 7D ) . However , this speed increase was not correlated with productive movement , as the number of directional changes along the principle axis of motion increased 74% upon KIF1C depletion ( Figure 7E ) . Speed increases , loss of confinement , and loss of directionality upon KIF1C depletion indicate that the role of KIF1C is most evident when Rab6A vesicles are moving least—KIF1C promotes vesicle directionality . When vesicles are confined , KIF1C helps to maintain that confinement , and when vesicles are stalled , KIF1C prevents backwards motion . An increase in vesicle directionality may be due to precluding the vesicle from binding a faster , less processive motor , or by influencing the tension on the vesicle , thereby changing the activity of other motors on the vesicle , and may come at the expense of overall vesicle speed . It was not possible to achieve plasmid rescue of KIF1C levels after siRNA depletion as KIF1C over-expression and artificially induced recruitment also influenced Rab6A vesicle motility ( data not shown , Schlager et al . , 2014 ) . Taken together , these results show that KIF1C is involved in both holding Rab6A vesicles within confined areas and also in maintaining the directionality of linearly moving Rab6A vesicles . Depletion of KIF1C was previously reported to cause Golgi fragmentation in HeLa cells ( Simpson et al . , 2012 ) . We reasoned that loss of Golgi ribbon structure may be related to KIF1C's ability to bind Rabs at both ends , which could in theory , be used to bind and link two adjacent membrane compartments . We therefore tested whether KIF1C's role in Golgi morphology maintenance correlated with its ability to bind Rab6A within the motor domain . HeLa cells were transfected with either a control or KIF1C-targeting siRNA for 72 hr prior to fixation . After fixation , Golgi complexes were visualized using anti-p115 antibody . As expected , cells depleted of KIF1C displayed fragmented Golgi complexes that very likely represent peri-nuclear mini-stacks ( Figure 8A , Figure 8—figure supplement 1 ) . Upon rescue by siRNA-resistant , wild-type KIF1C plasmid transfection ( 24 hr prior to fixation ) , Golgi morphology returned to normal . In contrast , cells rescued with the full-length KIF1C motor domain loop mutant showed Golgi fragmentation similar to KIF1C depletion alone ( Figure 8A ) . We also tested a construct , KIF1C K103A , that should not bind ATP ( Dorner et al . , 1998; Li et al . , 1998 ) and unlike wild-type KIF1C , does not localize to the cell periphery ( Figure 8—figure supplement 2 ) or to microtubules ( Figure 4—figure supplement 2 ) but is still capable of binding to Rab6A ( Figure 4C ) . Remarkably , KIF1C K103A rescued Golgi ribbon morphology in a manner similar to wild-type KIF1C rescue ( Figure 8 ) . Over-expression of these rescue constructs in cells treated with control siRNA did not significantly alter Golgi morphology ( data not shown ) . 10 . 7554/eLife . 06029 . 016Figure 8 . KIF1C's N-terminal Rab6A binding site is required for the maintenance of Golgi morphology . ( A ) Golgi morphology ( p115 ) of HeLa cells transfected with control siRNA or KIF1C siRNA and the indicated rescue plasmid . Scale bar , 10 µm . ( B ) Golgi ribbon rescue , defined as the mean fraction of Golgi staining present as large objects ( >4 . 11 µm² ) , normalized by KIF1C intensity , quantified from cells such as those shown in A ( bars = SE , >90 cells/condition ) . KIF1C wild-type rescue cells were statistically different from KIF1C depleted and loop mutant rescue cells but not those rescued with KIF1C K103A ( p < 0 . 001 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06029 . 01610 . 7554/eLife . 06029 . 017Figure 8—figure supplement 1 . siRNA rescue shows that KIF1C's N-terminal Rab6A binding site is required for the maintenance of Golgi morphology . Golgi morphology ( p115 ) of HeLa cells transfected with control siRNA or KIF1C siRNA and the indicated rescue plasmid . Scale bar , 10 µm . One cell of each type is also shown in Figure 8 . DOI: http://dx . doi . org/10 . 7554/eLife . 06029 . 01710 . 7554/eLife . 06029 . 018Figure 8—figure supplement 2 . Full-length KIF1C mutant protein localization . Vero cells transfected with the indicated CFP-tagged full-length KIF1C construct were MeOH fixed and imaged by fluorescence microscopy . Scale bar is 20 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 06029 . 018 The effects of KIF1C depletion and subsequent rescue were quantified using CellProfiler ( Carpenter et al . , 2006 ) and custom Matlab algorithms ( https://github . com/lee-ohlson-pfeffer/kif_golgi_fragmentation ) . To quantify the degree of Golgi fragmentation , we measured the percent of each cell's Golgi found in large objects ( >4 . 11 µm² ) ; a high value would be seen for control cells . As expected , cells treated with KIF1C siRNA alone or rescued with the KIF1C 6/10 loop mutant showed a decrease in large Golgi objects . In contrast , the KIF1C wild-type and K103A constructs efficiently rescued Golgi morphology upon KIF1C depletion ( Figure 8 ) despite a predominantly cytosolic localization for the K103A construct ( Figure 4—figure supplement 2 , Figure 8—figure supplement 2 ) . Importantly , the motor domain loop mutation did not alter KIF1C localization , as both wild type and loop-mutant proteins localized largely at the cell periphery , indicating that the loop-mutant was well folded and able to migrate in an anterograde direction along microtubules ( Figure 8—figure supplement 2 ) . Over-expression of both wild-type KIF1C and KIF1C K103A increased the percent of large Golgi objects compared with control siRNA , which suggests that additional KIF1C may increase Golgi compactness . These results show that the sequences comprising the Rab6A binding site within the KIF1C motor domain contribute , in some way , to the maintenance of normal Golgi morphology and that KIF1C influences Golgi morphology , independent of its function as a motor . An alternative means to assay Golgi ribbon stability is to test whether a protein influences the organization of Golgi mini-stacks in nocodazole-treated cells . Upon nocodazole addition , microtubules begin to depolymerize and the Golgi is detected as characteristic mini-stacks that distribute throughout the cytoplasm ( Thyberg and Moskalewski , 1985 ) . As shown in Figure 9 ( and Figure 9—figure supplement 1 ) , Golgi ribbons in cells over-expressing wild-type KIF1C were somewhat protected from nocodazole-induced fragmentation . In contrast , expression of the KIF1C loop 6/10 mutant that binds microtubules ( Figure 4—figure supplement 2 ) but shows diminished Rab6A binding ( Figure 4C ) did not protect the Golgi from nocodazole-triggered ribbon breakdown . Similarly , KIF1C lacking the C-terminal Rab binding domain also failed to protect the Golgi from nocodazole-induced fragmentation . In addition , the KIF1C K103A protein that can bind Rab6A at both ends but not microtubules , was able to protect the Golgi ribbon from disassembly ( Figure 9A and 9B ) . KIF1C E170A , mutated in a conserved kinesin microtubule binding site ( Woehlke et al . , 1997; Grant et al . , 2011 ) , was also able to protect the Golgi ribbon . This mutant showed diminished localization at the cell periphery compared with wild-type KIF1C ( Figure 8—figure supplement 2 ) . 10 . 7554/eLife . 06029 . 019Figure 9 . KIF1C over-expression stabilizes the Golgi in nocodazole treated cells . ( A ) Golgi morphology of HeLa cells incubated with or without nocodazole in the presence of the indicated over-expressed KIF1C construct . Scale bar , 10 µm . ( B ) Golgi structure maintenance , defined as the mean fraction of Golgi staining present in large objects ( >2 . 74 µm² ) , normalized by KIF1C intensity , in nocodazole-treated cells ( bars = SE , >85 cells/condition ) . KIF1C wild type over-expressing cells were statistically different from KIF1C loop mutant and ΔCBD cells but not KIF1C K103A expressing cells ( p < 0 . 01 ) . ( C ) Perinuclear KIF1C forms contacts with the Golgi . CFP-KIF1C-transfected cells in the presence ( left column ) or absence ( right columns ) of co-transfected mCherry-Rab6A were incubated with nocodazole for the indicated times and treated with liquid nitrogen prior to fixation to release cytosolic proteins . KIF1C localization ( wild type or K103A as indicated , green ) was monitored in relation to the trans Golgi marker , GCC185 or to the co-expressed mCherry-Rab6A ( red ) . Magnified areas are indicated with a box and shown in the middle two rows . Scale bars , 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 06029 . 01910 . 7554/eLife . 06029 . 020Figure 9—figure supplement 1 . KIF1C over-expression stabilizes the Golgi in nocodazole-treated cells . Golgi morphology of HeLa cells incubated with or without nocodazole in the presence of the indicated over-expressed KIF1C construct . Scale bar , 10 µm . One cell of each type is also shown in Figure 9 . DOI: http://dx . doi . org/10 . 7554/eLife . 06029 . 020 To quantify these observations , we measured the percent of each Golgi found in large objects ( >2 . 74 µm² ) . These objects were smaller than those scored in siRNA rescue because the over-expression of wild-type KIF1C was not able to fully rescue nocodazole-induced fragmentation , despite a readily apparent rescue phenotype . Nocodazole treatment of control cells , as well as cells expressing KIF1C loop 6/10 or ΔCBD mutants , resulted in a decrease in the mean percentage of large Golgi objects . In contrast , wild-type KIF1C , K103A , or E170A mutant proteins protected the Golgi from fragmentation , even after nocodazole-triggered , microtubule depolymerization , as seen by the increase in percentage of Golgi detected in large objects . These data demonstrate that KIF1C has the capacity to affect Golgi morphology , independent of its motor stepping and microtubule binding activities . Influencing Golgi morphology via sequences capable of Rab6 binding suggested that a small pool of KIF1C might form contacts with the Golgi . As noted earlier , KIF1C was first identified as a Golgi-localized motor protein ( Dorner et al . , 1998 ) but in addition to its abundant cytosolic pool , it has since been found at the cell periphery at podosome ends ( Bhuwania et al . , 2014; Efimova et al . , 2014; Kopp et al . , 2006; Figure 6A and Figure 8—figure supplement 2 ) and peri-centrosomally ( Schlager et al . , 2010; Theisen et al . , 2012; Figure 8—figure supplement 2 ) . Organelle association of predominantly cytosolic proteins can be difficult to detect . This challenge can be ameliorated by permeabilizing cells under conditions in which the cytosol is released , prior to fixation . For this purpose , we immersed coverslips in liquid nitrogen , which breaks plasma membranes and can reveal smaller pools of membrane-associated , cytosolic proteins ( Seaman , 2004 ) . Upon expression of CFP-KIF1C in HeLa cells in the absence or presence of limited mCherry-Rab6A co-expression , deconvolution light microscopy in conjunction with liquid nitrogen permeabilization revealed that wild-type KIF1C , and more clearly , the non-motile KIF1C K103A mutant , form at least some contacts with the Golgi , monitored with antibodies to the trans Golgi GCC185 protein ( left column ) or exogenously expressed mCherry-Rab6A ( right columns , Figure 9C ) . In the absence of nocodazole , residual KIF1C was distributed throughout the cell , with a clear perinuclear pool forming contacts with the Golgi , best detected at high magnification ( middle rows ) . Golgi contacts were also detected at intermediate times of nocodazole treatment ( 45 min ) . KIF1C K103A also showed Golgi contacts ( Figure 9C , bottom two rows ) . These experiments demonstrate that a small pool of KIF1C is appropriately situated to contribute to the process of stabilizing Golgi structure . Whether the mechanism of stabilization is direct or indirect , KIF1C is clearly important for normal Golgi structure maintenance .
The classical model for motor–cargo interaction consists of N-terminal domains interacting with the cytoskeleton and C-terminal domains interacting with cargo . A number of motors link to membranes via Rab GTPases , but few bind Rabs directly and none , to our knowledge , bind Rabs via their motor domain . We have shown here that Rab6A can bind directly to the KIF1C C-terminus ( KD = 0 . 9 µM ) , consistent with the role of Rab GTPases as cargo adaptors . Surprisingly , we have also discovered a new mode of motor protein regulation: direct binding of a normally membrane-associated Rab GTPase to the KIF1C motor domain ( KD = 0 . 2 µM ) . Our results show that Rab6A binds more tightly to the motor domain than to the cargo binding domain in vitro . However , in cells , the situation may be more complex as other proteins and/or KIF1C post-translational modifications may contribute to Rab6A's domain binding preferences . Rab GTPase binding to the motor domain impairs KIF1C's ability to bind to microtubules , both in equilibrium binding assays and in cells . Indeed , Rab6A led to a greater than 10 fold decrease in the affinity of the motor for microtubules in both AMP-PNP and ADP states . Thus , KIF1C may have a choice: the motor domain can bind either microtubules or Rab6A on membranes with very similar affinities ( KD 0 . 46 µM vs ∼ 0 . 2 µM , respectively ) . This feature , combined with KIF1C's ability to interact with Rab GTPases via its C-terminus ( KD = 0 . 9 µM ) , may enable KIF1C to transition from a transport vesicle motor to a protein capable of binding a Rab GTPase at both ends . Rab6A may influence control on KIF1C-driven vesicle traffic in novel ways . If KIF1C is an auto-inhibited motor like conventional KIF5C ( Dietrich et al . , 2008 ) and KIF1A ( Hammond et al . , 2009 ) , direct binding of Rab6A to the C-terminus may couple cargo binding with relief of motor protein auto-inhibition . Such a mechanism ensures that motor function is coupled to successful cargo engagement . Once on vesicles , additional Rab6A binding to the motor domain binding may regulate the load that other motors on the vesicle experience , by blocking KIF1C engagement with microtubules . This regulation may influence the speed , run length , and directionality of Rab6A vesicles , and may also play a role in KIF1C's confinement of non-moving vesicles . In addition to the direct interaction between Rab6A and KIF1C described here , additional Rab6A effectors have been reported to interact at least indirectly with KIF1C . These include BICDR-1 , which interacts in lysates and by yeast two-hybrid at the C-terminus of KIF1C ( Schlager et al . , 2010 ) , and myosin IIA , which in lysates , interacts in the region before the C-terminus ( residues 713–811; Kopp et al . , 2006 ) . Such interactions between Rab6 , its effectors , and the C-terminus of KIF1C highlight the importance of the linkage between KIF1C and Rab6 , all of which may strengthen the overall interaction of this motor with its cargo and provide an additional layer to regulate Rab6–KIF1C interactions . During neuronal development , BCDR-1 decreases significantly ( Schlager et al . , 2010 ) . The ability of KIF1C to interact with Rab6A GTPase directly ( as reported here ) or via Rab binding proteins such as BCDR-1 suggests that KIF1C may have both constitutive and development-specific functions . It would be difficult for a single Rab6A molecule to bind to both the KIF1C motor domain and to KIF1C's C-terminus , as both binding sites involve interaction with the Rab6A ‘switch domains’ that report the identity of the bound guanine nucleotide on the Rab protein surface . Multiple binding sites offer the possibility for Rab6A concentration-dependent effects on KIF1C , which will be of interest to explore in future experiments . In principle , KIF1C should be able to sense the concentration of Rab6A on transport vesicles , on the donor compartments from which they derive , or on the acceptor membrane once at their destination and choose between engaging microtubules or additional Rabs . Other adapters , such as BICDR-1 , may also play a role in influencing how concentration affects each Rab6A binding site . We have shown that sequences comprising Rab6A binding sites at both ends of KIF1C enable KIF1C to contribute to Golgi morphology maintenance , in a manner that is separable from its function as a motor . This explains the Golgi fragmentation phenotype seen here and previously upon KIF1C depletion ( Simpson et al . , 2012 ) . Fibroblasts from mice lacking KIF1C did not appear to show a morphological Golgi defect ( Nakajima et al . , 2002 ) , but such cells likely compensate by expressing other proteins at higher levels and this subtle phenotype may have been missed . Myosin V is a well-characterized example of a motor protein that serves to tether bound cargo near the terminal actin web ( Wu et al . , 2000 ) . In membrane traffic , tethers hold transport vesicles near their targets to permit engagement of SNARE proteins that mediate vesicle fusion . As KIF1C appears to be able to bind to Rabs directly at both its N- and C-termini , KIF1C might be employed similarly to help hold vesicles near the Golgi during the transition between microtubule based movement and docking at a membrane . The capacity of KIF1C to bind to Rabs at both ends also suggests a possible mechanism for KIF1C's role in maintaining Golgi morphology . Additional work will be needed to confirm these models for KIF1C function . While motor function was not necessary for KIF1C's role in contributing to Golgi structure maintenance , cellular depletion of KIF1C had several additional , interesting consequences . KIF1C was needed both to confine relatively stationary vesicles and to sustain the momentum of moving vesicles . Rab6A vesicles lacking KIF1C move faster from frame to frame; they also change directions more often than vesicles in control cells . These results suggest that KIF1C has at least two functions: ( 1 ) to hold vesicles; ( 2 ) to increase transport directionality , which surely requires motor activity . The most visible role for KIF1C is to function as a motor , as seen by the motor's localization primarily at the cell periphery . Rab6A regulation of KIF1C could allow KIF1C to release from microtubules when bound to Rab6A at both ends . On vesicles , this might help to avoid vesicle traffic jams by allowing for backwards motion or adjust vesicle speed by allowing other motors to dominate vesicle motion . By aiding in directional motion and decreasing speed , it is also possible that KIF1C helps vesicles recognize their targets , as slower moving vesicles would have more time for the protein:protein interactions that drive vesicle docking and SNARE-mediated fusion . Finally , it is important to remember that multiple motor proteins work in concert to transport vesicles . Rab6A participates in the motility of post-Golgi , exocytic transport vesicles ( Grigoriev et al . , 2007 ) in cooperation with the kinesin motor , KIF5B , and with cytoplasmic dynein via its effector , bicaudal D1 and D2 proteins ( Matanis et al . , 2002 ) . Depletion of KIF5B did not fully block the motility of Rab6A vesicles in previous work ( Grigoriev et al . , 2007 ) , thus , KIF1C likely functions in conjunction with these motors to drive Rab6A vesicle motility . The novel regulation by Rab6A of the KIF1C motor domain may allow Rab6A to shift the dynamics between these motors . The conventional KIF5B kinesin might play the primary role in moving Rab6A vesicles , while the interplay between KIF1C and dynein/dynactin might contribute to processivity . In summary , this is the first example to our knowledge of a Rab GTPase ( or any non-cytoskeletal protein ) binding to a motor domain in a manner that has the potential to influence whether that protein chooses to bind to a membrane or a microtubule track . KIF1C may attach to membranes via Rab6A binding at the C-terminus . Once bound to membranes , the motor domain can ( 1 ) bind a microtubule for conventional vesicle motility , or ( 2 ) bind to another Rab6A molecule on the vesicle to allow other motors to drive vesicle transport , or ( 3 ) bind to adjacent Rab6A-containing membranes like the Golgi complex where it can hypothetically , be held . Competition for Rab6A on the vesicle will be high as the absolute number of Rab GTPases on an individual transport vesicle is likely to be small ( ∼10 per vesicle; Takamori et al . , 2006 ) and many effectors , including two domains of KIF1C , will be vying to bind to Rabs . Moreover , as Rab6A is abundant on Golgi mini-stacks , it is in theory , possible that KIF1C could also serve to hold these mini-stacks together to maintain normal Golgi ribbon structure . This KIF1C-driven mechanism could explain how depletion of KIF1C leads to increased Golgi fragmentation that can be rescued by full-length KIF1C independent of motor activity but dependent on Rab6A binding domains . Finally , because the motor domain can potentially choose between Rab6A or microtubules , KIF1C could play a role during transport vesicle formation: when the tip of a newly forming transport vesicle projects away from a Rab6A-enriched membrane surface , the motor could instead latch onto microtubule filaments . KIF1C could then contribute to Rab6A vesicle motility en route to the cell surface . Future experiments will clarify how the Rab GTPase-assisted gymnastics accomplished by KIF1C contribute to membrane traffic within and beyond the Golgi complex .
HeLa and Vero cells were cultured in Dulbecco's modified Eagle's medium and transfected with siRNA and plasmids as described ( Aivazian et al . , 2006 ) . Rabbit anti-KIF1C antibody and goat anti-rabbit-horseradish peroxidase were from Cytoskeleton , Inc . ( Denver , CO ) and Bio-Rad ( Hercules , CA ) , respectively . HEK293 cells were transfected with KIF1C and Rab6A constructs 48 hr prior to cell lysis in 50 mM Hepes , pH 7 . 4 , 150 mM NaCl , 1% CHAPS , and protease inhibitors ( cOmplete , EDTA free , Roche , Indianapolis , IN ) . Clarified cell lysate was incubated with llama anti-GFP binding protein ( Rothbauer et al . , 2008 ) conjugated to NHS-activated Sepharose 4 Fast Flow ( GE Healthcare Biosciences , Pittsburgh , PA ) . After washing , the bound fraction was eluted in sample buffer and analyzed by immunoblot with mouse anti-Myc ( 9E10 ) or rabbit anti-Rab6A antibody ( Santa Cruz Biotechnology , Dallas , TX ) . Purification of Rab GTPases was described ( Aivazian et al . , 2006 ) . GST-tagged KIF1C constructs were expressed in bacteria and homogenized in 20 mM Tris , 400 mM NaCl , 1 mM DTT , pH 7 . 4 . Clarified homogenates were bound to glutathione-Sepharose and eluted in 20 mM glutathione . Purified Rabs ( 10 μM ) were preloaded with 35S-GTPγS , 3H-GDP , or cold nucleotide ( for immunoblot analysis ) in 50 mM HEPES pH 7 . 4 , 150 mM KCl , 10 mM EDTA , 0 . 1% BSA , 1 mM DTT , and 100 μM nucleotide for 10 min at 37°C; 20 mM MgCl2 was then added . Glutathione–Sepharose beads were incubated with 1 μM Rab and 15 μM GST or GST-KIF1C for 1 hr at room temperature in 20 mM HEPES pH 7 . 4 , 150 mM KCl , 4 mM MgCl2 , 0 . 1% BSA , and 0 . 2% Triton X-100 ( Buffer A ) . Beads were washed 3× and scintillation counting ( LS 6500 , Beckman Coulter , Inc . , Indianapolis , IN ) determined Rab bound . Alternatively , beads were eluted with 20 mM glutathione , pH 7 . 5 and analyzed by fluorescent immnuoblot using His-tag Rabbit antibody ( Cell Signaling , Danvers , MA ) and Alexa Flour 647 Goat anti-Rabbit antibody ( Life Technologies , Grand Island , NY ) . Fluorescence signal was captured using a Typhoon imager ( GE Healthcare Biosciences ) and analyzed by ImageJ ( Schneider et al . , 2012 ) . Myc-KIF1C constructs synthesized using TNT Quick Coupled Transcription/Translation System ( Promega , Madison , WI ) , optimally 11 . 25 nM , were incubated with GTPγS- or GDP-loaded , GST-Rabs ( 0 . 2–5 μM ) in Buffer A plus 1 mM DTT , and 100 μM GTPγS or GDP for 1 hr , 25°C , added to glutathione-Sepharose at 25°C , 1 hr and washed 3× ( Buffer A +1 mM DTT , 400 mM NaCl ) before elution with 20 mM glutathione , pH 7 . 5 . Eluates were immunoblotted with mouse anti-myc antibody ( 9E10 ) . KIF1C motor domain ( residues 1–349 ) was fused to murine KIF5C residues 329–334 followed by 6×His tag ( Nitta et al . , 2004 ) and purified based on Romberg et al . ( 1998 ) . After expression in Rosetta 2 cells using 0 . 5 mM isopropyl B-p-thiogalactopyranside for 16 hr at 16°C , cells were suspended in Buffer B ( 50 mM NaPO4 , pH 7 . 4 , 15 mM imidazole , 250 mM NaCl , 1 mM MgCl2 , 25 µM ATP , protease inhibitors ) and disrupted by Emulsiflex C-5 ( Avestin , Ottawa , ON ) . Clarified lysate was incubated with Ni-NTA ( Qiagen , Santa Clarita , CA ) for 1 . 5 hr at 4°C and after washing with Buffer B was eluted with Buffer B + 200 mM imidazole . The eluate was diluted fivefold with Buffer C ( 30 mM Hepes , pH 7 . 4 , 1 mM MgCl2 , 1 mM EGTA , 25 µM ATP ) and applied to HiTrap Q FF ( GE Healthcare ) , washed with Buffer C + 100 mM NaCl , before being eluted by gradient to 500 mM NaCl in Buffer C . Binding of the purified components was assayed in the same manner as in vitro translation-synthesized constructs . Full-length myc-KIF1C synthesized in vitro with 35S-methionine ( EasyTag , Perkin Elmer , San Jose , CA ) was desalted and incubated with GTPγS-preloaded Rabs ( 4 . 2 μM ) for 1 hr in Buffer A ( −BSA ) , 1 mM DTT , 2 . 5 mM ADP , 0 . 5 mM GTPγS . Microtubules ( 0 . 8 mg/ml ) , polymerized in 80 mM PIPES , 1 mM MgCl2 , 1 mM EGTA , pH 6 . 8 ( BRB80 ) , 1 mM DTT , 1 mM GTP , 10% DMSO , spun through a 40% glycerol cushion , and resuspended in BRB80 , 1 mM DTT , 0 . 2 mM Paclitaxel ( Cytoskeleton , Inc . ) , were incubated with the KIF1C-Rab complexes for 1 hr before being spun through a 10% sucrose , 20 µM Paclitaxel , 1 mM DTT , at 65K for 5 min ( Optima TLX , Beckman Coulter , Inc . , Indianapolis , IN ) . Scintillation counting and SDS-PAGE and radiography using a Typhoon imager ( GE Healthcare Biosciences ) revealed the amount of 35S-labeled-KIF1C in fractions . Purified His-tagged motor domain was incubated with Rabs in BRB80 , 1 mM DTT , 0 . 1 mg/ml BSA and indicated nucleotides for 30 min at room temperature with agitation . Microtubules were added for another 30 min . Reactions were centrifuged through a 35% sucrose cushion at 65K for 20 min ( Beckman Coulter , Inc . ) . Pellets were visualized by SDS-PAGE and immunofluorescence using a Typhoon imager ( GE Healthcare Biosciences ) and analyzed by ImageJ ( Schneider et al . , 2012 ) . CFP-KIF1C constructs were transfected into Vero or HeLa cells in the absence of presence of mCherry-Rab6A 24 hr before fixation . To measure total cell fluorescence intensity , cells were fixed with paraformaldehyde ( 3 . 7% , RT , 15 min ) and permeabilized with 0 . 1% Triton X-100 . To observe motor-microtubule localization , cells were permeabilized with MeOH ( 100% , −20°C , 4 min ) . To observe motor-Golgi localization , cells were treated with liquid nitrogen to permit release of cytosolic proteins , fixed with paraformaldehyde ( 3 . 7% , RT , 15 min ) and permeabilized with 0 . 1% Triton X-100 ( Seaman , 2004 ) . Cells were stained with chicken anti-GFP ( Abcam , Cambridge , MA ) , rabbit anit-Rab6A ( GeneTex , Irvine , CA ) , mouse anti-GCC185 ( produced for us by Cocalico Biologicals , Reamstown , PA ) , and mouse anti-tubulin ( Sigma–Aldrich , St . Louis , MO ) antibodies and visualized using an Olympus IX70 microscope with a 60× 1 . 4 NA Plan Apochromat oil immersion lens ( Olympus , Center Valley , PA ) and a charge-coupled device camera ( CoolSNAP HQ , Photometrics , Tucson , AZ ) . Maximum intensity projections were generated using softWoRx 4 . 1 . 0 software ( Applied Precision , Issaquah , WA ) . ImageJ ( Schneider et al . , 2012 ) was used to measure the total fluorescence intensity of traced cells . CellProfiler ( Carpenter et al . , 2006 ) was used to segment MeOH-treated cells and measure the Pearson's correlation coefficient between tubulin and KIF1C intensity over KIF1C-segmented objects . Two-sample t-test was used to determine significance . Full-length His-KIF1C was purified as described ( Hirokawa and Noda , 2001 ) . Coverslips coated with anti-β tubulin antibody and blocked with 0 . 75% Pluronic F-127 ( Sigma–Aldrich ) were incubated with 7 . 5 μg/ml rhodamine-labeled pre-formed microtubules + 15 μM Paclitaxel and blocked by casein . GTPγS-loaded Rab6A ( 15 μM ) or exchange buffer was mixed with 10 nM KIF1C plus rabbit anti-KIF1C antibody ( Cytoskeleton , Inc . ) and Dylight 649 goat anti-rabbit Fab fragment ( Jackson ImmunoResearch , West Grove , PA ) . Samples were visualized using a Nikon Ti-E inverted microscope with a 100× 1 . 49 NA APO-TIRF lens and an EMCCD ( Andor iXON+; Andor technology , South Windsor , CT ) . Motors were segmented using Spot Detector ( Olivo-Marin , 2002 ) and tracks were found using u-track ( 110523 ) ( Jaqaman et al . , 2008 ) . The Wilcoxon rank sum test was used to determine statistical significance . Samples were visualized in series such that proteins in each condition were on the coverslip for similar amounts of time . Moving motors were defined as those moving >0 . 12 µm/s for a distance of >0 . 2 µm that were tracked for >0 . 9 s . In these experiments , a large pool of motors was immobile on microtubules . However , the median life span of this pool was 0 . 7 s , suggesting that most motors do not accumulate on microtubules but rather bind and release . Importantly , both immobile and motile pools of KIF1C were susceptible to Rab6A action . A small pool of motors ( 6% ) was long-lived ( ± Rab6A ) ; this pool did not increase over the time of these experiments ( ∼15 min ) . Auto-inhibition likely explains why only 2% of microtubule-bound motors showed motility in these experiments . Vero cells were transfected with siRNA ( 72 hr total ) followed by pEGFP-Rab6A ( for the final 36 hr ) . Coverslips were transferred to a 37°C heated stage in Leibovitz's L-15 medium ( Invitrogen , Grand Island , NY ) and filmed using a Nikon Eclipse 80i microscope using a 100× numerical aperture 1 . 40 plan apochromat oil immersion objective and an EMCCD ( Andor technology ) . Vesicles in cropped stacks were segmented and tracked using u-track 2 . 1 . 0 ( Jaqaman et al . , 2008 ) . Track parameters were not normally distributed and the Wilcoxon rank sum test was used to determine statistical significance . HeLa cells were transfected with either a KIF1C ( CCUCAUGGAC UGUGGAAAUUU ) or a non-targeting siRNA ( GUUCAAUAGGCUUACUAAUUU ) ( Thermo Scientific , Lafayette , CO ) for 20 hr followed by pVSV-G-YFP ( ts045 ) for 2 hr at 37°C . VSV-G was accumulated in the ER ( 39°C , 16 hr ) , incubated at 32°C , then washed with ice-cold PBS , blocked with 0 . 2% BSA ( 30 min ) and incubated with mouse anti-VSV-G ( 8G5F11 ) ( 1 hr , 4°C ) . Cells were scraped into RIPA buffer with protease inhibitors ( Roche ) ; the post-nuclear supernatant was immunoblotted with goat anti-mouse horseradish peroxidase ( BioRad , Hercules , CA ) ; bound anti-VSV-G antibody was normalized to total VSV-G . HeLa cells were transfected with siRNA ( 72 hr total ) followed by CFP-tagged rescue constructs 24 hr before fixation or transfected with CFP-tagged rescue constructs ( 24 hr ) and then incubated with nocodazole ( 5 µg/ml , 1 hr , 37°C ) . Rescue constructs were made insensitive to siRNA by replacing seven nucleotides while retaining coding identity ( GACCTCATGGACTGTGGAAAT to GATTTAATGGATTGCGGTAAC ) . To observe Golgi morphology , cells were fixed with paraformaldehyde ( 3 . 7% , RT , 15 min ) and permeabilized with 0 . 1% Triton X-100 . Cells were stained with chicken anti-GFP ( Abcam ) and mouse anti-p115 ( mouse ascites ) antibodies and visualized using an Olympus IX70 microscope with a 40× 1 . 35 NA Apochromat oil immersion lens ( Olympus ) and a charge-coupled device camera ( CoolSNAP HQ ) . Maximum intensity projections were generated using softWoRx 4 . 1 . 0 software ( Applied Precision ) . Cells and Golgi were segmented using CellProfiler ( Carpenter et al . , 2006 ) . Golgi complex morphology was scored for each cell as the percent of p115-positive structures defined as ‘large’ ( 4 . 11 µm2 for siRNA and 2 . 74 µm2 for nocodazole-treated cells ) and normalized by the mean KIF1C intensity using Matlab ( Mathworks , Natick , MA; https://github . com/lee-ohlson-pfeffer/kif_golgi_fragmentation ) . The two-sample t-test was used to determine statistical significance . | Within our cells there are many compartments that play important roles . Small bubble-like packages called vesicles carry proteins and other molecules between these compartments . These vesicles can be driven around cells by a family of motor proteins called kinesins , which move along a network of filaments called microtubules . Kinesin proteins have two sections known as the N-terminus and the C-terminus . In most cases , the N-terminus contains the motor that binds to and walks along microtubules , while the C-terminus binds to vesicles or other cell compartments . Attached to the compartments are members of another family of proteins called the Rab GTPases . These proteins help the kinesins bind to a compartment , but it was not clear if , or how , these proteins control the activity of the kinesins . Here , Lee et al . studied a kinesin called KIF1C . The experiments show that this kinesin can move vesicles that contain a Rab-GTPase called Rab6A along microtubules . Unexpectedly , Rab6A controls the activity of KIF1C by directly interacting with the motor as well as the C-terminus . Loss of the kinesin from the cell slows down the delivery of cargo carried in vesicles to the surface of the cell . The experiments also show that KIF1C is involved in organizing another compartment within cells called the Golgi . This role relies on Rab6A binding to both the N-terminus and C-terminus of the kinesin , but does not require the kinesin to act as a motor . Lee et al . 's findings reveal a new way in which the activity of kinesins can be controlled . Future challenges will be to find out if other kinesins are also controlled in this way and discover when and where the Rab GTPases bind motor domains in cells . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"cell",
"biology"
] | 2015 | The Rab6-regulated KIF1C kinesin motor domain contributes to Golgi organization |
Most DNA in the genomes of higher organisms does not code for proteins . RNA Polymerase II ( Pol II ) transcribes non-coding DNA into long non-coding RNAs ( lncRNAs ) , but biological roles of lncRNA are unclear . We find that mutations in the yeast lncRNA CUT60 result in poor growth . Defective termination of CUT60 transcription causes read-through transcription across the ATP16 gene promoter . Read-through transcription localizes chromatin signatures associated with Pol II elongation to the ATP16 promoter . The act of Pol II elongation across this promoter represses functional ATP16 expression by a Transcriptional Interference ( TI ) mechanism . Atp16p function in the mitochondrial ATP-synthase complex promotes mitochondrial DNA stability . ATP16 repression by TI through inefficient termination of CUT60 therefore triggers mitochondrial genome loss . Our results expand the functional and mechanistic implications of non-coding DNA in eukaryotes by highlighting termination of nuclear lncRNA transcription as mechanism to stabilize an organellar genome .
RNA Polymerase II ( Pol II ) transcribes protein-coding DNA sequences into mRNA . Pol II also transcribes many non-coding DNA ( ncDNA ) sequences into long non-coding RNA ( lncRNA ) molecules ( Ponting et al . , 2009 ) . ncDNA sequences vastly outnumber protein coding DNA sequences in most eukaryotic genomes . Genome-wide Pol II activity at ncDNA is referred to as ‘pervasive transcription’ of eukaryotic genomes ( Jensen et al . , 2013 ) . The purpose of pervasive transcription is currently unclear . The ‘transcriptional noise’ hypothesis postulates that widespread Pol II transcription of ncDNA is a by-product of promiscuous Pol II activity and may often not be functionally relevant ( Struhl , 2007 ) . Indeed , the expression levels of lncRNA are generally lower than those of genes ( Schlackow et al . , 2017 ) . Reasons for low expression levels of lncRNA include a combination of tissue-specific expression and targeted lncRNA degradation ( Deveson et al . , 2017; Jensen et al . , 2013 ) . Measurements of nascent Pol II transcription reveal that lncRNA transcription can be strong , for example divergent lncRNA transcription from gene promoters ( Churchman and Weissman , 2011 ) . The strength of divergent lncRNA transcription is actively regulated at the level of initiation through chromatin-based mechanisms ( Marquardt et al . , 2014 ) . Profiling RNA expression in RNA decay pathway mutants supports the view that nascent lncRNA transcription is often stronger than expected based on steady-state RNA measurements in wild type ( Jensen et al . , 2013 ) . The regulation of lncRNA levels through targeted RNA decay and at the level of initiation argue that lncRNA levels in cells are carefully balanced . However , some intensely characterized human lncRNA such as H19 , Xist , NORAD and Malat1 represent relatively stable lncRNA molecules ( Schlackow et al . , 2017 ) . As a class , lncRNA molecules are poorly conserved on sequence level , which challenges the idea of an evolutionarily conserved function of lncRNA ( Graur et al . , 2015 ) . However , additional considerations such as secondary RNA structure , splicing or genomic location may be important factors when considering lncRNA conservation ( Seemann et al . , 2017; Ulitsky , 2016 ) . As some lncRNA molecules can serve biological roles , the distinction between functional lncRNA transcription and transcriptional noise requires experimental testing . Progress on this question is aided by a community consensus on the experiments needed to assign functions to ncDNA ( Bassett et al . , 2014; Goff and Rinn , 2015 ) . In budding yeast , profiling transcripts in RNA decay pathway mutants reveals many classes of ‘cryptic’ lncRNA molecules that accumulate specifically in these mutants . Cryptic Unstable Transcripts ( CUTs ) accumulate in mutants of the nuclear exosome 3’-to-5’ RNA decay pathway ( Davis and Ares , 2006; Wyers et al . , 2005; Xu et al . , 2009 ) . Mutations in the 5’-to-3’ cytoplasmic decapping-based RNA degradation pathway define Xrn1-sensitive Unstable Transcripts ( XUTs ) ( van Dijk et al . , 2011 ) . Stable Uncharacterized Transcripts ( SUTs ) were initially defined based on their environment-specific expression pattern ( Xu et al . , 2009 ) . SUTs are targeted for degradation by both the nuclear exosome and the cytoplasmic Xrn1 pathway ( Marquardt et al . , 2011 ) . Even though lncRNAs classified into the same subgroup based on RNA decay pathway sensitivity reveal some common characteristics , it is currently unclear to what extent this classification indicates similar functions . Processes linked to the act of Pol II transcription play an important role in targeting RNA decay pathways . Different stages in the ‘transcription cycle’ model of Pol II transcription are defined by the localization of molecular hallmarks that instruct stage-specific Pol II activities ( Buratowski , 2009 ) . The essential budding yeast Nrd1-Nab3-Sen1 ( NNS ) pathway mediates termination of Pol II transcription of short non-coding transcripts including CUTs and SUTs ( Porrua and Libri , 2015; Schulz et al . , 2013 ) . A combination of three molecular hallmarks associated with early stages of Pol II transcription recruit NNS to nascent lncRNA transcripts: Histone 3 lysine 4 trimethylation ( H3K4me3 ) , RNA consensus motifs in nascent lncRNA , and phosphorylation of the serine five residue in the Pol II largest subunit C-terminal repeat domain ( Pol II CTD ) ( Carroll et al . , 2007; Terzi et al . , 2011; Vasiljeva et al . , 2008 ) . Interestingly , phosphorylation of the Pol II CTD at tyrosine 1 has the opposite effect and decreases NNS recruitment ( Yurko et al . , 2017 ) . NNS-mediated transcriptional termination of CUTs is directly linked to the recruitment of the nuclear exosome RNA degradation machinery ( Vasiljeva and Buratowski , 2006 ) . While NNS components are specific to budding yeast , equivalent pathways ensure that early transcription termination of lncRNA is linked to co-transcriptional degradation by the nuclear exosome in mammals ( Ntini et al . , 2013; Ogami et al . , 2017; Preker et al . , 2008 ) . lncRNA abundance in cells is therefore determined by molecular checkpoints operating co-transcriptionally that are tightly linked to the termination of Pol II transcription . The termination of Pol II transcription for polyadenylated RNAs is currently best described by the ‘Torpedo Model’ ( Kim et al . , 2004; Richard and Manley , 2009; West et al . , 2004 ) . This mechanism relies on the cleavage of the nascent transcript . The continuing polymerase , which is transcribing beyond the mature transcript end site , is chased by 5’-to-3’ exonucleases that dislodge Pol II from the DNA template . The system relies on the identification of termination specifying RNA sequences by polyadenylation/termination factors , transcription stage specification , and reducing the velocity of the polymerase after transcript cleavage . Therefore , gene expression relies on efficient termination of Pol II transcription , for example through tight linkage of termination and mRNA polyadenylation ( Proudfoot , 2016 ) . Failures in the Pol II termination process can impact on genomic stability ( Aguilera and García-Muse , 2012 ) . Inefficient termination can blur the boundaries of neighboring transcription units with negative effects on gene expression ( Ard et al . , 2017 ) . Examples of this phenomenon have been initially described through studies of two α-globin gene copies oriented in tandem and termed ‘Transcriptional Interference’ ( TI ) ( Proudfoot , 1986 ) . TI of gene promoters by read-through transcription of upstream protein coding genes has since been detected in additional organisms ( Greger et al . , 2000; Hedtke and Grimm , 2009 ) . TI repression of gene promoters can also be triggered by overlapping lncRNA transcription ( Ard et al . , 2014; Kim et al . , 2016; Mellor et al . , 2016; Touat-Todeschini et al . , 2017 ) , or by termination failures of transcripts embedded within lncRNA such as pri-miRNA ( Dhir et al . , 2015 ) . Overall , these studies support an important role of Pol II termination to maintain genomic integrity and gene expression . Upstream lncRNA transcription across the downstream gene promoter represents a mechanism to adjust gene expression to the environmental conditions in budding yeast ( Martens et al . , 2004 ) . Even though Pol II is actively transcribing across the promoter of the downstream SER3 gene required for serine biosynthesis , no initiation from the SER3 promoter occurs through co-transcriptional changes in the chromatin structure associated with Pol II elongation ( Hainer et al . , 2011; Martens et al . , 2004 ) . Similarly , gene expression can be attenuated by depositing methylation marks over promoters by cryptic transcription ( Kim et al . , 2012; Pinskaya et al . , 2009 ) . Regulation of gene expression by TI is intriguing , because regulation by it could reconcile equivalent functions of lncRNA that are expressed at low levels and are poorly conserved at sequence level . Here , we find that transcriptional termination of the lncRNA CUT60 in budding yeast insulates the downstream gene from interfering transcription resulting from promoter bi-directionality . Efficient termination of CUT60 promotes expression of the downstream ATP16 by preventing TI . Reduced ATP16 expression causes poor growth and triggers the loss of mitochondrial DNA ( mtDNA ) . Our findings highlight the efficiency of nuclear lncRNA termination as a mechanism to promote functional gene expression and mtDNA stability .
The lncRNA CUT60 is divergently transcribed from the promoter of the mediator complex subunit MED2 ( Poss et al . , 2013 ) . In addition , CUT60 is located upstream in tandem of the ATP16 gene encoding the δ-subunit of the mitochondrial ATP-synthase that couples proton translocation to ATP synthesis ( Duvezin-Caubet et al . , 2003; Giraud and Velours , 1994 ) ( Figure 1A ) . Native Elongating Transcript sequencing ( NET-seq ) data ( Marquardt et al . , 2014 ) indicates higher nascent Pol II transcription of the two flanking genes than of CUT60 ( Figure 1—figure supplement 1A ) . To test if this low level of CUT60 transcription may be functionally significant , we replaced CUT60 with the URA3 coding sequence ( cut60Δ::URA3 ) ( Figure 1—figure supplement 1B–C ) . We detect growth on selective media arguing for functional transcription ( Figure 1—figure supplement 1D ) . We note that multiple independently generated isolates of cut60Δ::URA3 mutants had a strongly reduced growth rate compared to wild type ( Figure 1B , Figure 1—figure supplement 1E ) . Common growth defects in multiple isolates argue against a second-site mutation . Reduced growth may be attributed to reduced mitochondria function . To test mitochondria function , we assayed yeast growth on non-fermentable carbon sources such as ethanol and glycerol . Strikingly , cut60Δ::URA3 fails to grow on non-fermentable carbon sources , indicating mitochondrial defects ( Figure 1C , Figure 1—figure supplement 1F ) . To test lesions in the mitochondrial genome , we performed test crosses using cut60Δ::URA3 to wild type and the mip1Δ mutant that has lost the mitochondrial genome ( i . e . genotype ρ0 ) through DNA replication defects ( Foury , 1989 ) . As mitochondria are inherited through the cytoplasm , the resulting diploids can grow on non-fermentable carbon sources if one of the parents contains functional mitochondria ( Merz and Westermann , 2009 ) . We find that diploids resulting from crosses of cut60Δ::URA3 with mip1Δ fail to grow on non-fermentable carbon sources , while they show growth on the fermentable carbon source glucose ( Figure 1D ) . These data strongly support that cut60Δ::URA3 mutants have mitochondrial defects . To test if the cut60Δ::URA3 mutation may result in a reduction ( ρ- ) or loss ( ρ0 ) of mitochondrial DNA , we measured mitochondrial DNA using quantitative PCR with probes specific to the mitochondrial genome ( Figure 1—figure supplement 1G ) . We find that the mitochondrial DNA level in cut60Δ::URA3 and mipΔ1 controls are reduced compared to wild type yeast . We detect no qPCR signal corresponding to mtDNA , suggesting a ρ0 phenotype in cut60Δ::URA3 . If cut60Δ::URA3 mutants were p0 , we would expect that expression of transcripts encoded on the mitochondrial genome cannot be detected . To test this hypothesis , we assayed expression of the highly abundant COX3 transcript encoded by the mitochondrial genome ( Figure 1E ) . While we can detect strong COX3 expression in wild type , we fail to detect any COX3 expression in the p0 control mutant mip1Δ and cut60Δ::URA3 . Remarkably , our collective results suggest the loss of mtDNA when the CUT60 lncRNA sequence of the nuclear genome is disrupted . To understand how CUT60 stabilizes the mitochondrial genome , we hypothesized that CUT60 may affect the expression of genes involved in mtDNA maintenance . Mitochondrial genome loss can be triggered in cells lacking Atp16p ( Duvezin-Caubet et al . , 2006 ) . ATP-synthase proton channel formation requires multiple genes located on the mitochondrial genome; therefore the loss of mtDNA represents a strategy to prevent uncoupled proton translocation from ATP synthesis . We hypothesized that the observed mitochondrial genome loss could be triggered if mutating CUT60 affected expression of neighboring ATP16 ( Duvezin-Caubet et al . , 2006 ) . To test cis-acting functions of CUT60 we analyzed expression of the flanking genes . Expression of the downstream ATP16 is strongly reduced in cut60Δ::URA3 mutants , while we detect no expression changes of the MED2 gene ( Figure 2A , Figure 2—figure supplement 1A ) . To enhance detection of low-level or unstable transcripts , we included the 5’-to-3’ RNA decay pathway mutant xrn1Δ , and the 3’-to-5’ nuclear exosome RNA decay pathway mutant rrp6Δ in our analysis ( Houseley and Tollervey , 2009 ) . We noticed a different size profile of ATP16 RNA in cut60Δ::URA3 mutant strains with RNA degradation defects through rrp6Δ or xrn1Δ ( Figure 2A ) . While we detect no ATP16 mRNA expression of the expected size in the mutants , we observe an extended ATP16 transcript in rrp6Δ , and at higher levels in xrn1Δ . The size of the extended ATP16 transcript is approximately ( 1 . 9 kb ) , which would be in agreement with a hypothesized URA3-ATP16 fusion transcript . To test if the extended ATP16 transcript represents a 5’-extension , we used northern blotting probes against URA3 and the ATP16 promoter region downstream of the annotated CUT60 3’-end ( Figure 2—figure supplement 1B ) . Both probes detect a transcript coinciding in size with the extended ATP16 transcript . This data suggest that the extended ATP16 transcript detected in the cut60Δ::URA3 mutant represents 5’-extension that originates from the CUT60 promoter and fails to efficiently terminate , resulting in read-through transcription into ATP16 . A similar repression of ATP16 is observed in mutants replacing CUT60 with SUT129 ( cut60Δ::SUT129 ) , indicating that this effect is not dependent on the URA3 sequence ( Figure 2A ) . CUT60 can be expressed from the SUT129 regulatory region ( sut129Δ::CUT60 ) with CUT-specific expression characteristics ( Marquardt et al . , 2014 ) . However , expressing CUT60 in trans from the SUT129 regulatory regions ( sut129Δ::CUT60 ) does not restore ATP16 expression ( Figure 2A ) . These data indicate a cis-acting function of the CUT60 locus to promote ATP16 expression , either as DNA element or non-coding transcript . ATP16 expression could rely on DNA sequence elements located in CUT60 ( ‘DNA element model’ ) , or from CUT60 lncRNA function by a cis-acting mechanism ( ‘cis-acting lncRNA model’ ) . To distinguish these two models , we cloned four versions of ATP16 expression constructs with variable 5’ extensions in yeast expression plasmids and transformed them into cut60Δ::URA3 ( Figure 2B ) . We measured ATP16 expression derived from plasmids by northern blotting and compared expression to wild type containing the empty vector control ( Figure 2C ) . We quantified ATP16 expression relative to the SCR1 loading control to test if the constructs restore ATP16 expression to wild-type level ( Figure 2D ) . Our results indicate that all constructs restore ATP16 expression to the level detected in wild type . A comparison of ATP16 expression between construct V1 ( lacking CUT60 ) and construct V2 ( including CUT60 ) most directly tests the effect of potential DNA cis-elements located within the CUT60 sequence . Our statistical tests show no significantly different ATP16 expression between these constructs . These data strongly argue against DNA sequence elements providing promoter function located in CUT60 as explanation for reduced ATP16 expression and growth in cells lacking CUT60 sequences . We note that in the only construct lacking CUT60 sequences a second , longer transcript accumulates ( see discussion ) . Even though ATP16 expression can be restored using constructs V1-4 , a growth defect compared to wild type remains . Interestingly , growth in strains carrying the constructs is slightly improved compared to cut60Δ::URA3 mutants ( Figure 2—figure supplement 1C ) . The growth defects are consistent with mitochondrial genome loss: once mtDNA is lost , reconstitution of ATP16 expression is insufficient to restore wild type growth . In conclusion , our findings suggest a cis-acting function for the CUT60 lncRNA to promote functional ATP16 expression , which is necessary to maintain the mitochondrial genome . lncRNA species are commonly defined based on expression profiling in characteristic mutant backgrounds , but it remains unclear if these lncRNA definitions carry functional implications . The cis-acting function of CUT60 , in combination with strengths of our experimental system , allows us to address this question . To test if the function of CUT60 can be replaced by other lncRNA in the same genomic location as CUT60 we generated replacement mutants by site-specific homologous recombination and assayed ATP16 expression . To test if a lncRNA of the SUT class can restore ATP16 expression , we replaced CUT60 with SUT129 ( cut60Δ::SUT129 ) ( Xu et al . , 2009 ) . We removed endogenous SUT129 ( sut129Δ::URA3 ) to avoid potentially confounding effects in the same background ( Figure 3A ) . In cut60Δ::SUT129 , an extended ATP16 transcript with reduced expression is detected ( Figure 3A ) . We conclude that SUT129 cannot substitute CUT60 function . Transcriptional termination of SUTs is often inefficient , resulting in 3’-extended SUTs ( eSUTs ) ( Marquardt et al . , 2011 ) . This suggests that a SUT lncRNA might promote ATP16 expression if it was terminated like a CUT . To test this hypothesis , we generated CUT60-SUT129 hybrids by fusing these sequences in the transcript middles ( Figure 3—figure supplement 1A–B ) . Expressing the 5’-CUT60-3’-SUT129 hybrid instead of CUT60 ( cut60Δ::5’C-3’S ) resulted in low ATP16 expression as part of an extended transcript as in the cut60Δ::URA3 mutant ( Figure 3A ) . We note the size of ATP16-containing fusion transcript in cut60Δ::5’C-3’S is slightly reduced compared with cut60Δ::SUT129 , consistent with read-through transcription of SUT129 3’-end into ATP16 . Strikingly , ATP16 expression is restored when the 5’-SUT129-3’-CUT60 hybrid replaces CUT60 ( Figure 3A , Figure 3—figure supplement 1C ) . Moreover , a short transcript with SUT-specific accumulation profile in RNA decay mutants can be detected with a probe against the SUT129 5’-region ( Figure 3—figure supplement 1D ) . The detection of a SUT-like transcript resulting from the 5’-SUT129-3’-CUT60 hybrid maps efficient transcriptional termination to the 3’-half of CUT60 . These data suggest that efficient transcriptional termination of CUT60 promotes ATP16 expression . The essential Nrd1-Nab3-Sen1 ( NNS ) Pol II transcription termination pathway terminates CUT transcription ( Mischo and Proudfoot , 2013 ) . A role for NNS-mediated transcriptional termination of CUT60 is supported by hypomorphic nrd1 mutations that reveal 3’-extended CUT60 RNA species ( Marquardt et al . , 2011 ) . Nascent transcription data ( Pol II PAR-CLIP ) following conditional Nrd1 depletion support a direct role of Nrd1 in CUT60 termination as depletion results in more Pol II-associated RNA downstream of CUT60 ( Schaughency et al . , 2014 ) ( Figure 3—figure supplement 2A ) . To test if transcription termination defects in CUT60 resulting from mutations in trans-acting factors share phenotypes observed in cut60Δ mutations , we assayed growth of two hypomorphic nrd1 mutant alleles in different genetic backgrounds , nrd1-1 and nrd1Δ151–214 ( Steinmetz and Brow , 1998; Vasiljeva et al . , 2008 ) ( Figure 3B , Figure 3—figure supplement 1E ) . Strikingly , nrd1-1 , nrd1Δ151–214 and the mip1Δ ρ0 control fail to grow on non-fermentable carbon sources , while we can recapitulate the reduced growth phenotype of both mutant alleles on fermentable carbon sources . These data show that nrd1-1 and nrd1Δ151–214 mimic growth defects of cut60 mutations . The shared growth defects suggest that ATP16 expression may be reduced when Nrd1 function and thus NNS termination is impaired . We assayed ATP16 expression by northern blotting to test the effects of nrd1-1 and nrd1Δ151–214 ( Figure 3C ) . Both nrd1 mutant alleles reduce ATP16 expression by about 40% compared to their respective isogenic controls ( Figure 3—figure supplement 1F ) . While this expression is higher compared to cut60Δ mutations ( Figure 2 ) , we note that the hypomorphic nrd1 mutant alleles provide residual Nrd1 termination function . Moreover , it has been estimated that reducing ATP16 by around 50% is sufficient to trigger mtDNA loss ( Duvezin-Caubet et al . , 2003 ) . In conclusion , these data suggest that CUT60 termination by the NNS pathway promotes ATP16 expression and mtDNA maintenance . NNS-mediated termination is tightly connected to the classification of budding yeast CUTs as NNS-termination is linked to the characteristic RNA degradation through the nuclear exosome pathway ( Arigo et al . , 2006; Fox and Mosley , 2016; Porrua and Libri , 2015; Vasiljeva and Buratowski , 2006 ) . To test if other CUTs can functionally substitute for CUT60 , we screened CUTs based on their genomic location , and selected CUTs located upstream in tandem of a protein coding gene along with distance filtering ( >0 . 1 kb;<1 . 5 kb ) ( Xu et al . , 2009 ) . Furthermore , we excluded CUTs with transcripts annotated on either strand between the CUT and the downstream gene , resulting in a list of 68 CUTs ( Figure 3—source data 1 ) . A combination of increased Pol II PAR-CLIP signal downstream of the CUTs following Nrd1 depletion ( Schaughency et al . , 2014 ) and low divergent non-coding transcription of the downstream gene in NET-seq data ( Marquardt et al . , 2014 ) identified CUT95 , CUT277 , CUT48 and CUT217 that shared many characteristics with CUT60 ( Figure 3—figure supplement 2 ) . The genomic CUT60 sequences were seamlessly replaced with the selected CUTs ( Figure 3—figure supplement 1G ) . In addition , we replaced CUT60 with CUT#78 resulting from synthetic selection for strong NNS termination ( Porrua et al . , 2012 ) and CUT170 that lacks a downstream gene within our limits ( Figure 3—figure supplement 2 ) . ATP16 expression increased compared to cut60Δ::URA3 controls in the cut60Δ::CUT217 , cut60Δ::CUT170 and cut60Δ::#78 even though not fully to wild type level ( Figure 3D ) . Our selection for similarity to CUT60 identified CUT217 that results in relatively high ATP16 expression in cut60Δ::CUT217 . However , the cut60Δ::CUT95 , cut60Δ::CUT277 and cut60Δ::CUT48 replacements result in no detectable ATP16 expression increase ( Figure 3D ) . To test if these results could be explained by differences in termination efficiency between the inserted CUTs , we quantified known binding sites for the Nab3 and Nrd1 RNA binding protein components of NNS ( Carroll et al . , 2004; Jamonnak et al . , 2011; Porrua et al . , 2012; van Nues et al . , 2017; Wlotzka et al . , 2011 ) ( Figure 3—figure supplement 3A–B ) . The results suggest that CUTs increasing ATP16 expression are enriched for NNS-targeting sites , even though the high number of sites for CUT277 illustrates that this analysis is not fully predictive . We analyzed Nrd1 and Nab3 PAR-CLIP data to examine NNS-targeting to the tested CUTs in vivo ( Figure 3—figure supplement 3C–D ) ( Schulz et al . , 2013 ) . We find high Nrd1 and Nab3 PAR-CLIP signal at CUT60 consistent with efficient NNS termination and sensitivity to reduced NNS function . Interestingly , CUT277 shows relatively high NNS occupancy by PAR-CLIP yet does not terminate efficiently enough to promote ATP16 expression in the cut60Δ::CUT277 replacement . To rule out epigenetic effects resulting from homologous recombination we reintroduced CUT60 following cut60Δ::URA3 replacement , essentially recreating a wild type yeast strain that has undergone the procedure of site-specific recombination ( cut60Δ::CUT60 ) ( Figure 3—figure supplement 1G ) . Reintroducing CUT60 restores ATP16 expression , arguing against epigenetic effects preventing ATP16 expression in CUT replacement strains ( Figure 3E ) . CUT60 function can be partially provided by other lncRNA defined as CUTs , however ATP16 expression is highest when CUT60 sequences trigger termination . We consider a combination of strong CUT60 termination efficiency and locus-specific effects relaying upstream termination to the activation of ATP16 expression as the most parsimonious explanation for these results . All in all , our site-specific CUT replacement data support that NNS termination upstream of ATP16 promotes functional ATP16 expression . To directly test if transcription termination upstream of ATP16 is required for functional ATP16 expression , we replaced CUT60 with URA3 including three sequences mediating efficient termination of Pol II transcription downstream of protein coding genesFigure 3—figure supplement 1E; Figure 3—figure supplement 1F; Figure 3—figure supplement 1G . We compared ATP16 expression in wild type , cut60Δ::URA3 , cut60Δ::URA3+ , cut60Δ::URA3++ and cut60Δ::URA3-trp1-terminator ( Figure 3F ) . Strikingly , ATP16 expression increases with all three strategies to enhance termination of URA3 transcription . These results strongly argue for a model in which the process of transcriptional termination promotes ATP16 expression . Our data are consistent with a cis-acting model for CUT60 function that emphasizes the role of efficient transcriptional termination . To understand how CUT60 promotes ATP16 expression , we tested if chromatin signatures at the ATP16 promoter are affected by read-through transcription . We performed Chromatin Immuno Precipitation ( ChIP ) coupled to quantitative PCR ( qPCR ) using indicated primer pairs and strains ( Figure 4A ) . Read-through transcription can repress downstream gene expression by Transcriptional Interference ( TI ) ( Proudfoot , 1986 ) . If ATP16 was repressed by TI , we expect chromatin signatures associated with Pol II elongation at the ATP16 promoter in cut60Δ::URA3 mutants compared to cut60Δ::URA3++ mutants that terminate transcription or wild type controls . Promoter regions are characterized by a nucleosome depleted region ( NDR ) where nucleosome occupancy can be elevated by TI ( Hainer et al . , 2011 ) . To test if the ATP16 promoter NDR is affected by upstream termination , we performed ChIP experiments with an antibody against histone 3 ( H3 ) and a tri-methylated version of H3 lysine 36 ( H3K36me3 ) ( Figure 4—figure supplement 1A–C ) . H3K36me3 marks Pol II elongation zones in gene bodies and performs an important function to repress cryptic initiation from within transcription units ( Carrozza et al . , 2005; Keogh et al . , 2005; Venkatesh et al . , 2012 ) . The levels of H3K36me3/H3 are unchanged between wild type , cut60Δ::URA3++ and cut60Δ::URA3 in the gene-body region . However , H3K36me3/H3 levels are significantly increased in cut60Δ::URA3 compared to controls in the ATP16 promoter region ( Figure 4A ) . Our ChIP analyses indicate that ATP16 repression is associated with elevated chromatin signatures of Pol II elongation at the ATP16 promoter . These data are consistent with regulation by TI if termination of upstream transcription is inefficient . Previous characterizations of TI suggest that this mechanism relies on efficient Pol II elongation and associated chromatin signatures at promoter regions ( Ard and Allshire , 2016; Hainer et al . , 2011 ) . If ATP16 was indeed sensitive to repression by TI , we would expect increased expression from the ATP16 promoter when the efficiency of Pol II elongation is reduced . As the loss of mtDNA precludes growth assays to monitor functional ATP16 expression we designed synthetic circuits to quantitatively assay expression from the ATP16 promoter using fluorescent proteins ( Figure 4B ) . We replaced the MED2 coding sequence with mCherry and the ATP16 coding sequence with YFP by cloning . We used the wild type DNA region containing CUT60 ( circuit I ) where we expect efficient transcriptional termination upstream of the ATP16 promoter . Consistently , we detected high levels of mCherry as well as YFP fluorescence ( Figure 4C ) . We generated a second circuit based on cut60Δ::SUT129 ( circuit II ) where we expect low termination efficiency upstream of the ATP16 promoter ( Figure 3A ) . YFP expression from the ATP16 promoter in circuit II is significantly reduced compared to circuit I , while mCherry expression appears largely unaffected ( Figure 4C ) . RNA expression analysis in cut60Δ::SUT129 represents a population average measurement . The reduced ATP16 expression could be explained by a gradual reduction in a homogenous population , or a higher proportion of cells lacking ATP16 expression in a heterogeneous population . YFP fluorescence is normally distributed across thousands of individual cells , supporting the hypothesis that ATP16 expression is gradually reduced in a homogenous population ( Figure 4—figure supplement 1D ) . All in all , the repression of the ATP16 promoter by read-through transcription is supported by reduced YFP fluorescence in circuit II . To test the role of candidate factors facilitating Pol II elongation in TI-repression of the ATP16 promoter we combined elongation factor mutants with our synthetic circuits . We expect mutations in elongation factors required for TI to elevate YFP expression in circuit II . Since we observed elevated H3K36me3/H3 levels at the ATP16 promoter , we tested the involvement of Set2p mediating H3K36me3 in budding yeast ( Carrozza et al . , 2005; Keogh et al . , 2005 ) . While mCherry expression is unchanged in set2Δ mutants , YFP expression increases ( Figure 4D ) . This effect of set2Δ is not observed in circuit I ( Figure 4—figure supplement 1E ) . Gene repression by TI relies on the activity of chromatin remodeling by the Spt6p and Spt16p histone chaperones ( Hainer et al . , 2011; Kaplan et al . , 2003 ) . Hence , we assayed fluorescence resulting from the synthetic circuits in spt16-197 and spt6-1004 mutants ( Kaplan et al . , 2003; Prelich and Winston , 1993 ) . Both mutants increase YFP expression specifically in circuit II , consistent with TI suppression ( Figure 4E ) . While the changes in fluorescence in set2Δ are specific to YFP fluorescence , spt6-1004 and spt16-197 reduce mCherry fluorescence in circuit I and circuit II ( Figure 4—figure supplement 1F ) . These data show that Pol II elongation factors previously associated with gene repression by TI are required for full repression of the ATP16 promoter region when the termination of upstream lncRNA transcription is inefficient . These results support the notion that efficient CUT60 termination prevents ATP16 repression by a TI mechanism . In conclusion , our data suggest that efficient termination of upstream lncRNA transcription ensures ATP16 promoter activity . CUT60 termination promotes functional ATP16 expression , the maintenance of mitochondrial DNA and yeast growth . All in all , the dependence of mitochondrial DNA stability on CUT60 termination provides a compelling example for the biological significance of pervasive non-coding transcription in genomes .
While pervasive transcription of Pol II in eukaryotic genomes results in the production of many lncRNA , their functional significance often remains elusive . Our analysis of the lncRNA CUT60 reveals how efficient transcriptional termination fulfills an important biological role . Inefficient transcriptional termination of CUT60 causes read-through transcription across the promoter of the downstream ATP16 gene , which represses ATP16 transcription ( Figure 5 ) . CUT60 read-through transcription represses functional ATP16 expression by a Transcriptional Interference mechanism . ATP16 repression results in a striking biological defect: mitochondrial genome loss ( ρ0 ) . The function of Atp16p as δ-subunit of the mitochondrial ATP-synthase can reconcile the strong phenotypic response resulting from inefficient CUT60 termination . Interestingly , in atp16 mutant backgrounds the loss of the mitochondrial genome represents a survival strategy ( Duvezin-Caubet et al . , 2006 ) . At first glance the positive selection for the loss of the mitochondrial genome in atp16 mutants seems counter-intuitive . However , acquiring ρ0 efficiently blocks proton leakage , as the two subunits encoding the core of the ATP-synthase proton channel ( ATP6 and ATP9 ) are located on the mitochondrial genome . Hence , loss of the mitochondrial genome disrupts proton channel formation to prevent uncoupling of the ATP synthase in the absence of a functional δ-subunit ( Duvezin-Caubet et al . , 2006 ) . Future research will be required to identify if modulating CUT60 termination efficiency is used as mechanism for regulated mtDNA loss . Mutant mtDNA can out-compete wild type mtDNA copies through increased replication efficiency ( Contamine and Picard , 2000 ) . It is tempting to speculate that the loss of mtDNA triggered by transiently reduced CUT60 termination efficiency could be a temporary state to detoxify cells from deleterious mtDNA . Healthy mtDNA variants could be subsequently acquired through mating and would be maintained as long as CUT60 transcription is terminated efficiently . In human , mutant mtDNA has emerged as common cause for metabolic disease and mechanisms to acquire healthy mtDNA would offer a promising avenue for biotechnology ( DiMauro et al . , 2013 ) . All in all , our findings provide a compelling example for a biologically significant lncRNA transcription termination event . Promoting the stability of organellar DNA adds a new function to the growing list of ‘junk DNA’-derived lncRNA with functional roles in cells . Divergent transcription of lncRNA from eukaryotic gene promoters represents an important source of lncRNA ( Sigova et al . , 2013 ) . Divergent lncRNA transcription is repressed by lncRNA degradation ( Almada et al . , 2013; Ntini et al . , 2013; Xu et al . , 2009 ) , and by a repressive chromatin architecture ( Jin et al . , 2017; Marquardt et al . , 2014; Rege et al . , 2015; Tan-Wong et al . , 2012 ) . Transient inhibition of these pathways and a reduction of termination efficiency could amplify read-through transcription of divergent lncRNA ( Mellor et al . , 2016 ) . Inefficient transcriptional termination and environmentally regulated expression define the species of budding yeast SUTs ( Marquardt et al . , 2011; Xu et al . , 2009 ) . These characteristics poise SUTs for gene regulation by read-through transcription , for example at IME1 and GAL80 ( van Werven et al . , 2012; Xu et al . , 2011 ) . The TI mechanism described in our study is reminiscent of the regulation of FLO11 expression and yeast flocculation by a divergently transcribed SUT lncRNA ( Bumgarner et al . , 2009 ) . While regulation of FLO11 expression is regulated at the level of individual cells resulting in heterogeneous expression states ( Bumgarner et al . , 2012 ) , our analysis of circuit II suggests that ATP16 expression is reduced in a homogenous population . Replacement of the efficiently terminated CUT60 with inefficiently terminated SUT129 supports a role for SUTs as source for read-through regulation . The 5’-SUT129-3’CUT60 hybrid experiments in particular suggest that the underlying molecular cause may be differences in termination efficiencies of CUTs and SUTs ( Figure 3 ) . While the SUT nomenclature is specific to budding yeast , it should be noted that many human lncRNA share characteristic cell-type-specific and environmentally regulated expression ( Deveson et al . , 2017 ) and cis-acting functions are increasingly appreciated ( Engreitz et al . , 2016 ) . Divergent lncRNA transcription of human promoter upstream transcripts ( PROMPTs ) may overlap with downstream genes reminiscent of the MED2-CUT60-ATP16 circuitry ( Chen et al . , 2016 ) . Efficient transcriptional termination of a divergent lncRNA has therefore functional implications beyond the specific example described here . The sensitivity to RNA degradation pathways has been used as a defining criterion to distinguish lncRNA species in budding yeast ( van Dijk et al . , 2011; Xu et al . , 2009 ) . It remains unclear how frequently lncRNA classification based on RNA decay pathways is indicative of equivalent lncRNA function . Our study addresses this question by testing experimentally which lncRNAs of the CUT species can functionally substitute for CUT60 ( Figure 3 ) . To enable these experiments we systematically identified CUTs with similar genomic location to examine if they can replace CUT60 function ( Figure 3—figure supplement 2 ) . Indeed , ATP16 expression increases in a subset of CUTs , but none of the CUTs we tested fully restores ATP16 expression . Our results argue that there is some yet limited functional overlap between lncRNA classified as members of the same species . Importantly , our analysis of complementation plasmids shows that sequences downstream of CUT60 are sufficient to fully restore ATP16 expression ( Figure 2 ) . DNA sequence elements in CUT60 that could hypothetically provide promoter function can therefore not explain reduced ATP16 expression in the replacement mutants . We detect strong binding of Nrd1 and Nab3 to CUT60 by PAR-CLIP , suggesting CUT60 represents a strong transcriptional terminator for the NNS pathway ( Figure 3—figure supplement 3 ) . Consistently , we find that hypomorphic mutations in Nrd1 result in reduced ATP16 expression , CUT60 read-through transcription and common growth phenotypes with CUT60 termination defective strains ( Figures 1–3 ) . While CUT60 scores highly in accessible metrics to predict NNS termination these differences cannot fully account for the termination efficiency and resulting ATP16 expression , particularly in comparison to CUT277 ( Figure 3—figure supplements 2 , 3 ) . We map sequences mediating strong termination of Pol II transcription to the 3’-half of CUT60 where a cluster of NNS-targeting cis-elements is located ( Figure 3—figure supplement 3 ) . Moreover , we can partially restore ATP16 expression by the inclusion of three sequences to trigger transcriptional termination by NNS-independent pathways ( Figure 3F ) . The most parsimonious interpretation of our experimental results is that an architecture linking efficient transcriptional termination of CUT60 with the initiation of ATP16 expression is under strong locus-specific selection , presumably to maintain the mitochondrial genome . Even though most promoters in eukaryotic genomes initiate transcription bi-directionally ( Preker et al . , 2008; Seila et al . , 2008 ) , the ATP16 promoter shows little evidence of divergent lncRNA transcription ( Figure 1—figure supplement 1A , Figure 3—figure supplement 2A ) . Low levels of divergent lncRNA transcription may offer a partial explanation for the need of CUT60 as strong ‘insulator’ for the ATP16 promoter . Divergent lncRNA transcription could conceivably provide an opposing wave of Pol II transcription that may trigger termination of incoming read-through transcripts by polymerase collisions , as proposed for the yeast BAT2 promoter ( Hobson et al . , 2012; Mellor et al . , 2016; Nguyen et al . , 2014 ) . The extended ATP16 transcript we detect in the background of construct V1 lacking CUT60 may be a manifestation of this phenomenon ( Figure 1D ) . The extended transcript likely results from upstream transcription initiation in the plasmid , as it is not observed in constructs containing CUT60 sequences that would lead to termination . However , our site-specific swaps of CUTs do not support that lncRNAs upstream of unidirectional gene promoters are generally characterized by particular efficient NNS termination ( Figure 3 ) . The ATP16 promoter lacks binding sites for the Reb1p and Rap1p transcription factors that prevent read-through transcription by a road-blocking mechanism , which may also contribute to the sensitivity for read-through repression ( Candelli et al . , 2018 ) . It has previously been established that promoter repression by TI can be partially overcome by increasing the levels of promoter-specific transcription factors to trigger transcriptional initiation ( Greger et al . , 2000 ) . A locus-specific balance of transcriptional read-through across promoters and their strength to mediate transcriptional initiation likely determines the susceptibility of promoters for repression by TI . Low primary sequence conservation combined with low steady-state abundance of lncRNA challenges functions of pervasive transcription through the resulting lncRNA molecules . Our experimental characterizations of CUT60 reveal a cis-acting lncRNA function through the mechanism of NNS-termination to promote ATP16 expression . We propose that ATP16 promoter function is inhibited through a TI mechanism by upstream transcription extending across this region when CUT60 termination is inefficient ( Figure 5 ) . A TI mechanism is consistent with an increase in the Pol II elongation-specific H3K36me3 mark at the ATP16 promoter region that is usually detected in the middle of transcription units . In addition , the Set2p histone-methyl transferase mediating H3K36me3 is genetically required for efficient repression of the ATP16 promoter ( Figure 4 ) . Set2p is required for TI by sufficiently long upstream interfering transcripts ( Kim et al . , 2016 ) . The histone chaperones Spt6p and Spt16p are required for TI without an apparent size limit of the interfering transcript ( Ard and Allshire , 2016; Hainer et al . , 2011; Winston et al . , 1984 ) . We find that both histone chaperones are also required to enforce TI repression of the ATP16 promoter . While gene regulation by TI may be most intuitive in dense genomes , this mechanism has been first described for two tandem copies of the human α-globin genes ( Proudfoot , 1986 ) . Human lncRNA sequences are rich in cis-elements triggering polyadenylation and could help to protect from TI akin to CUT60 ( Almada et al . , 2013; Ntini et al . , 2013 ) . Read-through transcription in human is associated with cellular responses to viral infection ( Rutkowski et al . , 2015 ) and osmotic stress ( Vilborg et al . , 2015 ) . Future research elucidating the molecular mechanism of gene repression through the act of transcription will be required to detect and predict where this powerful mechanism operates in genomes . While previous studies largely focused on lncRNA transcription as a cause for TI ( Ard et al . , 2017 ) , our results suggest that lncRNA transcription may also offer protection . In summary , our findings expand the functional and mechanistic implications of ‘junk DNA’ transcription in eukaryotes .
Yeast strains used in this study are listed in Supplementary file 1 . Strains were constructed using standard procedure . To introduce cut60∆::URA3 mutation primers SB2803 and SB2804 were used with SMC88 as template DNA . Primers SB2803 and SB2804 amplify URA3 coding region and contain overhangs that are homologous to the region upstream and downstream of the annotated CUT60 sequence ( Xu et al . , 2009 ) . These homologous overhangs integrate PCR fragments into the desired locus as described in Figure 3—figure supplement 1G . Transformants were selected using SC-URA plates . To generate cut60∆::URA3#1 , cut60∆::URA3#2 and cut60∆::URA3#3 primers MLO515 and MLO516 were used on genomic DNA using SMY81 as template . The PCR fragments containing the cut60∆::URA3 replacement and homologous overhangs to the region upstream and downstream of CUT60 were transformed into SMY2132 and selected for growth on SC-URA plates . To generate cut60∆::CUT60#1 and cut60∆::CUT60#1 primers MLO488 and MLO489 were used SMY2132 genomic DNA template . The generated PCR fragments containing the endogenous CUT60 sequence and homologous overhangs to the region upstream and downstream of CUT60 and were transformed into SMY81 and selected using 5-FOA plates . cut60∆::SUT129 mutants were generated starting from cut60∆::URA3 mutant strains . URA3 was replaced with a PCR fragment from genomic DNA using primers SB2875 and SB2876 . These primers contain homologous overhangs to the region upstream and downstream of CUT60 . Transformants were selected using 5-FOA plates . To generate cut60∆::5’C-3’S mutants , we generated PCR fragments using primers SB2823 and SB2899 on genomic CUT60 DNA as template DNA , and primers SB2898 and SB2876 on genomic SUT129 on genomic DNA as template . These fragments were fused by PCR using primers SB2823 and SB2876 , and these fusion fragments were then transformed into cut60∆::URA3 strains . Transformants were selected using 5-FOA plates . To generate cut60∆::5’S-3’C mutants , we generated PCR fragments using primers SB2900 and SB2902 on genomic CUT60 template DNA and primers SB2875 and SB2901 on genomic SUT129 template DNA . These fragments were fused by PCR using primers SB2875 and SB2902 , and the resulting fusion fragment was transformed into cut60∆::URA3 strains . Transformants were selected using 5-FOA plates . To introduce rrp6Δ::KanMX mutations , we generated PCR fragments for transformation using primers SB1055 and SB1056 on rrp6Δ::KanMX template DNA . Transformants were selected on G418 plates . To introduce xrn1Δ::KanMX mutations , we generated PCR fragments for transformation using primers SB2877 and SB2878 on xrn1Δ::KanMX template DNA . Transformants were selected on G418 plates . To generate strains where CUT60 was replaced with different lncRNA sequences , we transformed cut60∆::URA3 strains with PCR fragments that contains the lncRNA sequence of interest and homologous overhangs to the region upstream and downstream of CUT60 as in Figure 3—figure supplement 1G . Primers MLO1319 and MLO1320 were used for cut60Δ::CUT170 , primers MLO496 and MLO497 were used for cut60Δ::CUT95 , primers MLO498 and MLO499 were used for cut60Δ::CUT277 , MLO1376 and MLO1377 were used for cut60Δ::CUT48 , MLO1370 and MLO1371 were used for cut60Δ::CUT217 , MLO1328 and MLO1329 were used for cut60Δ::CUT#78 . Transformants were selected using 5-FOA plates . cut60Δ::URA3 ++ contains the coding sequence of URA3 and the 300 bp downstream of the URA3 coding sequence on pRS316 as terminator . Primers MLO933 and MLO935 are used on pRS316 template DNA to generate a PCR fragment containing the URA3 coding sequence including terminator and homologous overhangs to the sequence upstream and downstream of CUT60 . Transformants were selected using SC-URA plates . cut60Δ::URA3+ contains the coding sequence of URA3 and the 80 bp downstream of the endogenous coding sequence as terminator . Primers MLO933 and MLO934 are used on pRS316 template DNA to generate a PCR fragment containing the URA3 coding sequence including the 80 bp terminator with homologous overhangs sequences upstream and downstream of CUT60 . Transformants were selected using SC-URA plates . ut60Δ::URA3 +trp1 terminator contain the coding sequence of URA3 and the 80 bp downstream of the coding sequence of the endogenous TRP1 gene as terminator . Primers MLO933 and MLO1368 are used on pRS316 template DNA to generate a PCR fragment containing the URA3 coding sequence with homologous overhangs to the sequence upstream of CUT60 and the TRP1 terminator 5’ end . MLO1335 and MLO1334 are used on SMC147 template to generate a PCR fragment containing the 80 bp TRP1 terminator that has homologous overhangs to the 3’ sequence of URA3 and the sequence downstream of CUT60 . Transformants were selected using SC-URA plates . Standard yeast media were used , 100 µg/ml ampicillin was added to liquid media to avoid bacterial contaminations . YPD liquid media and plates containing 300 μg/ml G418 Sulfate ( Geniticin , American Bioanalytical ) were used to select for KanMX or/and 100 μg/ml ClonNat ( Jena Bioscience , Jena , Germay ) to select for NatMX . SC media lacking leucine was used for growth of yeasts containing the complementation plasmids ( V0-V4 ) . SC media containing uracil was used for selection for growth in presence of 1 mg/ml 5-Fluoro Orotic Acid ( 5-FOA ) , to select against cells with functional URA3 . SC-Glycerol and SC-Ethanol plates are used to test for mitochondrial function . Single colonies of yeast strains were incubated in 5 ml of liquid medium ( YPD or SC-Leu medium ) overnight in a shaker at 150 RPM at 30°C . The titer was determined , using spectrophotometer . Cultures were diluted and harvested at a titer of 2 × 107 cells/ml . Per transformation reaction , 1/10 of the harvested pellet was used . Transformation mix ( 240 µl PEG ( 50% w/v ) , 36 µl 1M LiAc , 10 µl single stranded salmon sperm carrier DNA ( 10 mg/ml ) , 74 µl PCR product or 5–10 µl of plasmid , 34 µl H2O minus the volume of PCR product or plasmid was added to the cell pellets and mixed by pipetting . Cells were then heat shocked at 42°C for 30 min . For selection using URA3 or LEU markers cells were plated on SC-URA or SC-LEU plates directly . Selection for G418 , NAT or 5-FOA resistance was performed following growth on YPD plates for 1 day and then replica-plated to an appropriate selection plate . Transformants were mostly visible after 3 days . SMC340 , SMC342 , SMC361 and SMC362 were constructed using one step isothermal assembly of four overlapping dsDNA fragments as described ( Gibson , 2011 ) . Fragments that were inserted into HindIII digested SMC339 backbones , were generated by PCR reaction using primers MLO431 , MLO433 , MLO486 , MLO487 as forward primers and MLO434 as reverse primer . Vectors were sequenced with primers MLO434 , MLO486 , MLO487 , MLO489 and MLO467 . SMC425 and SMC426 were constructed using one step isothermal assembly . The fragments that were inserted into AscI digested SMC50 backbones were generated by PCR reactions . The mCherry-NatMX fragment was made using primers MLO517 and MLO777; YFP fragment was made using primers MLO517 and MLO778 both with SMC50 as template . The central fragment containing the bidirectional promoter of MED2 and either the CUT60 sequence ( for circuit I ) or the SUT129 sequence ( for circuit II ) and were amplified using primers MLO776 and MLO779 , using BY4741 as template for the CUT60 version and SMY131 ( cut60Δ::SUT129 ) as template for the SUT129 version . The three fragments were fused using PCR with splicing overlapping ends , using each other and more of the flanking primer MLO517 to make one continuous fragment , which was identified on an agarose gel . Plasmids were genotyped by PCR and sequenced . Overnight cultures were grown at 30°C in media . The yeast pellets were washed with water , and diluted to OD600 = 0 . 1 . An equal starting amount of cells were spotted in three-fold serial dilutions and growth assayed at 30°C for 2–3 days . Single colonies of yeast strains were incubated in 5 ml of liquid medium overnight in a shaker at 150 RPM at 30°C . Subsequently 25 ml of liquid medium were inoculated to an OD600 = 0 . 1 and grown at 30°C to OD600 = 0 . 4–0 . 8 . RNA was isolated using the MasterPureTM Yeast RNA Purification Kit ( Epicentre , Wisconsin , USA ) . Final RNA concentration was determined using a Nanodrop spectrophotometer . Ten micrograms of total RNA was separated by electrophoresis on 1 . 5% agarose-formaldehyde-MOPS gels . The RNA was transferred to a nylon transfer membrane via capillary blotting in 10 X SSC buffer for 16 hr . RNA was cross-linked to the nylon membrane by UV irradiation . The membrane was incubated in a rotating oven at 65°C in 15 ml hybridization buffer ( 0 . 5 g BSA , 15 ml phosphate buffer , 35 ml 10% SDS , 100 µl 0 . 5 M EDTA pH 8 . 0; 100 ml phosphate buffer consisting of 68 . 4 ml 1M Na2HPO468 . 4 ml and 31 . 6 ml 1 M Na2H2PO4*H2O ) 60 min prior to addition of radioactive probe . Single-stranded probes were generated by incorporation of radioactive dTTP into DNA using a unidirectional thermocycling reaction . Twenty microliters of reaction mix were assembled containing five units Taq DNA polymerase , 200 μM each of dCTP , dGTP and dATP , 5 ng of DNA template consisting of a 100–300 bp long PCR product of the probed region , 0 . 4 μM primer oligonucleotide ( antisense to the RNA to be detected ) and 4 μl of α−32P-dTTP added . The tube was placed into a PCR machine and subjected to 35 cycles of the following series: denaturation at 94°C for 30 s , annealing at an experimentally determined temperature for 20 s , and extension for 45 s at 72°C . Upon completion , 20 μl of water was added to the reaction and unincorporated nucleotides were removed using a Spin-50 gel-filtration column ( BiomaxInc . , Planegg , Germany ) according to manufacturers’ instructions . The purified probe was denatured by incubation at 95°C for 5 min and chilled on ice . The probe was added to the membrane in 15 ml hybridization solution and incubated overnight at 65°C in a rotating oven . The membranes were washed with low stringency wash buffer ( 0 . 1 x SSC , 0 . 1% SDS ) and when necessary to reduce background , high stringency wash buffer ( 2 x SSC , 0 . 1% SDS ) at 65°C . A Typhoon scanner and ImageQuant software ( GE Healthcare Chicago , Illinois ) was used for analysis and quantification . ATP16 probe template was amplified using primers MLO466 and MLO467 . COX3 probe template was amplified using primers MLO843 and MLO844 . URA3 probe template was amplified using primers SB1777 and SB1778 . MED2 probe template was amplified using primers SB3080 and SB3081 . ATP16-promoter probe template was amplified using primers SB2451 and SB2452 . SUT129 probe template was amplified using primers SB2453 and SB2454 . The blots were stripped using 0 . 1% SDS and re-probed with SCR1 loading controls . SCR1 probes generated using MLO689 and MLO690 . Single colonies of yeast strains were incubated in 25 ml of YPD overnight in a shaker at 150 RPM at 30°C . 250 ml YDP cultures were inoculated with overnight culture to OD600 = 0 . 1 and grown until OD600 = 0 . 6 . Cells were fixed with 1% Formaldehyde for 15 min at room temperature . 37 . 5 ml of 3M Glycine was added and incubated for 5 min . Fixed cells were centrifuged at 3 . 500 RPM for 2 min , washed twice with ice-cold PBS ( 137 mM NaCl , 2 . 7 mM KCl , 10 mM Na2HPO4 , 1 . 8 mM KH2PO4 , pH 7 . 4 ) . Cell pellets were flash frozen in liquid nitrogen and stored at −80°C . The pellet was resuspended in 1 ml ice cold FA lysis buffer +Proteinase inhibitor ( Roche , Basel , Switzerland ) /0 . 5% SDS ( 2X FA buffer: 100 mM Hepes-KOH pH7 . 5 , 300 mM NaCl , 2% Triton-X-100 , 0 . 2% Na-Deoxycholate , 2 mM EDTA ) and lysed by bead beating . Crude whole cell extract was collected by puncturing small holes in the tube using flame-headed needle and centrifugation into a new microfuge tube at 1 . 000 RPM for 1 min . Crude whole cell extract was briefly vortexed and then sonicated at 5°C using Q700 sonicator ( Qsonica , Connecticut ) on Amplitude 100 for a total of 20 min ( 30 s ON/OFF cycles ) . Insoluble material was removed by centrifugation at 13 . 500 RPM 2 × 10 min . 25 µL samples of whole cell extracts were collected as total input controls ( ‘Input’ ) and frozen at −20°C . 250 µL of soluble lysates were pre-cleared with Protein A beads ( Gen Script , New Jersey ) for 1 hr at 4°C and then incubated with appropriate antibody and beads overnight at 4°C . 2 μL of H3 antibody ( ab1791; Abcam , Cambridge , UK ) , and 2 μL of H3K36me3 antibody ( ab9050; Abcam ) were used for IPs . IPs were washed in four steps , each wash step was performed for 4 min at 4°C . First wash step was done in 1x FA buffer/0 . 1% SDS +275 mM NaCl , 2nd wash step was done in 1x FA buffer/0 . 1% SDS +500 mM NaCl , 3rd wash step was done in 10 mM TRIS-HCL ( pH8 ) +0 . 25 M LiCl , 1 mM EDTA , 0 , 5% Na-Deoxycholate and 4th wash was done in TE buffer . Beads following IP and 25 μL of Input samples were incubated with 100 μL of 1% Chelex100 Resin ( Bio-Rad ) in dH2O , boiled to remove DNA-protein crosslinks for 15 min , and then treated with proteinase K ( 10 mg/mL ) for 30 min at 55°C . Samples were boiled for an additional 10 min to denature proteinase K . Samples were centrifuged at 10 . 000 RPM for 1 min . 60 μL of supernatant was carefully pipetted into new microfuge tubes . Quantitative analysis was performed by qPCR on diluted samples: Input DNA samples were diluted 1/100 in dH2O while IP DNA samples were diluted 1/50 . ChIP enrichments per antibody was calculated using the following formula: 2^ ( ( CT Value of input ) - ( CT Value of IP ) ) * ( Dilution factor input/Dilution factor IP ) . This gives the %IP per antibody , these values were normalized to the %IP calculated for the household gene ( ACT1 ) , to give the %IP relative to actin . Triplicate biological samples were each analysed using three technical repeats . For normalization to the ACT1 household gene using the primers MLO1247 and MLO1250 were used . qPCR was performed using 5 µl of GoTaq qPCR Master Mix ( Promega , Wisconsin ) , 1 µl primers ( 20 µM ) , 3 µl H2O and 1 µl of DNA in C1000 Touch Thermal Cycler ( BioRad , California ) . For each sample , at least three biological repeats and three technical replicates were used . The amplification efficiency of each amplicon used was determined , and only amplicons with efficiencies between 97 and 110% were retained . A BD Fortessa ( BD biosciences , New Jersey ) flow cytometer with a high-throughput resolution sampler was used to quantitate YFP and mCherry fluorescence . YFP was excited at 450 nm and fluorescence collected through a 535/45 band pass and 525 LP emission filter . mCherry was excited at 600 nm and assayed with a 632/22 band pass filter . Between 30000 and 50000 events were sampled for each well . Flow cytometry data were exported from the acquisition program ( FACSDiva , Beckton Dickinson , San Jose , CA , USA ) in the FCS3 . 0 format . R studio was used to process the data , importing using a custom modified version of ‘flowCore’ package from Bioconductor . org . To compare cells with the same average sizes and filter out cellular debris and aggregates , the median side scatter ( SSC ) was used . Data within ±25% of the median SSC value were used for all the comparisons . The first third of events and the last ninth of data in time were removed to minimize errors due to unstable sample flow through the flow cytometer . Any well that had fewer than 500 counted cells was excluded from the analysis . BY4741 without a fluorescent reporter ( FPR ) served as background control for experiments in Figure 4D and E . SMY921 without FPR served as background control for experiments in Figure 4F . After background subtraction , the data were represented normalized to co-assayed isogenic wild-type FPR control strains . Values for circuit I were normalized to wild type FPR control strains containing circuit I , values for circuit II were normalized to wild type FPR control strains containing circuit II . Statistical tests were performed on means generated from at least three biological replicates . Repeats per individual experiment shown in legend ( n = number of biological replicates ) . Results are expressed as the mean ±standard error of the mean ( s . e . m . ) between experimental groups . Significance is assessed using a two-tailed unpaired Student’s t-tests . The *p<0 . 05 , **p<0 . 01 , and ***p<0 . 001 levels were considered significant . Genomic coordinates of CUTs ( n = 925 ) and SUTs ( n = 847 ) were obtained from ( Xu et al . , 2009 ) . The coordinates were converted from SacCer2 to SacCer3 using kentUtils liftOver tool and the chain file sacCer2ToSacCer3 . over . chain . gz . Coordinates of SGD genes in SacCer3 ( n = 6692 ) were downloaded from the UCSC Table Browser . To obtain CUTs for Figure 3—source data 1 , CUTs overlapping other transcription units are removed . For the remaining CUTs ( n = 238 ) , the nearest downstream gene on the same strand was found , and the distance to this gene was calculated . The final 68 CUT-gene pairs were selected based on the following criteria: ( i ) the distance between the 3’ end of the CUT and the TSS of the downstream gene ( gap interval ) is between 100 bp and 1500 bp; ( ii ) there are no transcription units overlapping with the gap interval on either strand . This pipeline was implemented in the 01-Candidate_CUTs . R script . The directionality of the downstream genes was obtained using yeast NET-Seq data from ( Marquardt et al . , 2014 ) ( accession GSE55982 ) . The raw reads were extracted from SRA archives using SRA Toolkit v2 . 6 . 3 , quality and adapter trimming was performed using Trim Galore v0 . 4 . 3 and alignment to the yeast genome was done using Bowtie2 ( --very-sensitive-local mode ) . The SAM files were filtered from unmapped reads and reads with low mapping quality ( MAPQ <= 10 ) and then converted to sorted BAM files using Samtools v1 . 3 . 1 . BAM files were filtered from reads aligning to unwanted sequences such as rRNA , tRNA , snRNA and snoRNA genes and then converted into stranded Bedgraph files using Bedtools v2 . 25 . 0 . Only 5' positions of reads were used to calculate the coverage . Bedgraph files were then normalized to 1 million tags . For more details , see the 02-Reanalysis_of_NET-Seq_raw_data . sh script . At the next step , these Bedgraph files were used to compute the NET-Seq coverage around TSS of genes found downstream of our candidate CUTs ( n = 68 ) . The signal at the first 100 bp of downstream gene on the coding strand gave transcription intensity into the coding direction and the signal at the last 100 bp of the gap interval on the opposite strand gave transcription intensity into the divergent direction . To avoid possible division by zero , a pseudocount of 0 . 25 was added to the latter values . This pipeline is detailed in the 03-Directionality_of_downstream_genes . R script . Pol II PAR-CLIP data from Schaughency et al . , 2014 were used to assess Nrd1-dependent transcription termination of each CUT . The level of nascent transcription over the CUTs and their gap intervals were assessed using the original Wig files provided by authors ( accession GSE56435 ) . The read-through index is defined as the ratio of Pol II PAR-CLIP signal at the gap interval divided by Pol II PAR-CLIP signal at their CUT in the Nrd1-AA sample compared the control sample . The Rbp2 strain treated with rapamycin was used as the control . For details see the 04-Nrd1-dependent_termination_of_CUTs . R pipeline . The Pol II PAR-CLIP data were also used to visualize the genome-wide changes in the nascent transcription profile which occur due to the depletion of Nrd1 in Figure 3—figure supplement 2 . We subtracted the rapamycin-treated negative control track from the Nrd1-AA track at all genomic positions . This was done by the 07-Subtract_PolII_PAR-CLIP_tracks . R script . The resultant Bedgraph files contain both positive ( black ) and negative ( gray ) values . Nrd1 PAR-CLIP data from ( Schulz et al . , 2013 ) were used to assess the Nrd1 and Nab3 occupancy at each CUT . This study provides both Nrd1 PAR-CLIP and 4tU-Seq data for the same samples ( accession E-MTAB-1766 ) . The raw NGS reads were aligned to the yeast genome using STAR v2 . 5 . 2 in the local mode . The Nrd1 PAR-CLIP BAM files were converted to the mpileup format using samtools . The cross-linking positions ( defined as T-to-C and A-to-G conversion sites on the forward and reverse strands , respectively ) were detected with single nucleotide resolution using VarScan v . 2 . 4 . 3 . The resultant VCF files were converted to Bedgraph files and normalized to 1 million tags . This pipeline is described in the 05-Reanalysis_of_4sU-Seq_and_Nrd1_PAR-CLIP_raw_data . sh . The PAR-CLIP signal was also normalized by the transcription level of the underlying locus . To this end , the 4sU-Seq coverage representing nascent transcription was smoothed using the sliding window approach . A pseudocount of 0 . 1 was added to genomic positions with zero 4sU-Seq coverage , then mean values were taken over 51 bp windows centered around each genomic position . The Nrd1 PAR-CLIP signal was divided by the smoothed 4tU-Seq signal , giving the normalized Nrd1 occupancy track . The 4sU-normalized Nrd1 PAR-CLIP Bedgraph files were used to quantify Nrd1 occupancy over each CUT . These calculations were implemented in the 06-Quantification_of_Nrd1_occupancy_in_CUTs . R script . All custom scripts and pipelines for data processing were deposited at ( Ivanov , 2018 ) . A copy is archived at https://github . com/Maxim-Ivanov/du_Mee_et_al_2018_eLife . | Genes are made up of DNA and contain the information to make proteins , which carry out a variety of roles in the cell and the body . First , the information found on DNA needs to be transcribed into RNA molecules , which then act as a template to build the actual proteins . However , the vast majority of DNA does not encode proteins . Nevertheless , these non-coding regions of DNA ( often given the popular but misleading name ‘junk-DNA’ ) are still transcribed into non-coding RNA . The purpose of this type of RNA is largely unclear , although some are known to activate certain genes . The transcription of non-coding RNA is also sensitive to environmental changes , suggesting it may play other important roles in the cell . Not all DNA – including non-coding DNA– is copied into RNA in one go . Usually , every DNA sequence is transcribed separately as one unit . These units have clearly marked start and end points . If these marker points are overridden the transcription process can overlap onto the next sequence . Thus , in the case of coding DNA , proteins may not form properly . However , until now it was unclear if missed marker points in non-coding RNA may also have consequences . To investigate this further , du Mee et al . mutated several non-coding parts of the DNA in yeast . The experiments showed that a non-coding RNA sequence called CUT60 , appeared to be important to help yeast cells grow . When CUT60 was modified so that it lacked the end marker , its RNA transcript fused with the neighbouring gene called ATP16 . As a result , the protein of the ATP16 gene could no longer be produced properly . Normally , ATP16 plays important roles in a cell structure called the mitochondrion , also known as the energy powerhouse of the cell . The mitochondrion has its own DNA , and without CUT60 and ATP16 , the yeast cells lost their mitochondrial DNA and could not grow as quickly . This shows that non-coding DNA sequences can have a purpose and can affect other parts of the cell . Moreover , start and end markers of transcription are also important in non-coding DNA sequences . The same mechanism could be at play in other genes or even other organisms . As well as revealing a new role for non-coding DNA , the findings could also help to develop a new method to cleanse yeast cells of disease-causing mutations in their mitochondrial DNA . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression",
"genetics",
"and",
"genomics"
] | 2018 | Efficient termination of nuclear lncRNA transcription promotes mitochondrial genome maintenance |
Within the confines of tissues , cancer cells can use blebs to migrate . Eps8 is an actin bundling and capping protein whose capping activity is inhibited by Erk , a key MAP kinase that is activated by oncogenic signaling . We tested the hypothesis that Eps8 acts as an Erk effector to modulate actin cortex mechanics and thereby mediate bleb-based migration of cancer cells . Cells confined in a non-adhesive environment migrate in the direction of a very large ‘leader bleb . ’ Eps8 bundling activity promotes cortex tension and intracellular pressure to drive leader bleb formation . Eps8 capping and bundling activities act antagonistically to organize actin within leader blebs , and Erk mediates this effect . An Erk biosensor reveals concentrated kinase activity within leader blebs . Bleb contents are trapped by the narrow neck that separates the leader bleb from the cell body . Thus , Erk activity promotes actin bundling by Eps8 to enhance cortex tension and drive the bleb-based migration of cancer cells under non-adhesive confinement .
Cell migration mediates critical physiological processes including development and the immune response , and is de-regulated during cancer metastasis . Cells move by orchestrating their extracellular interactions with intracellular cytoskeleton-dependent changes in shape and force generation; protruding forward , grasping sites in the environment , and pulling themselves along ( Lauffenburger and Horwitz , 1996; Papusheva and Heisenberg , 2010 ) . Although the pulling forces driving cell movement are well known to be mediated by myosin-based contraction ( Vicente-Manzanares et al . , 2009 ) , recently it has become clear that migrating cells can utilize multiple mechanisms to drive protrusion and interaction with their environment ( Lammermann and Sixt , 2009 ) . Protrusion of the cell's plasma membrane boundary can be driven by either actin polymerization or by pressure-driven membrane blebbing ( Charras and Paluch , 2008 ) , while grasping the environment can be mediated by specific adhesion receptors that bind to the extracellular matrix ( ECM ) or other cells , or by non-specific friction ( Lammermann and Sixt , 2009; Bergert et al . , 2015 ) . These multiple modes of migration are adopted by different cell types in different contexts . For example immune cells utilize adhesion-independent migration during tissue surveillance ( Lammermann et al . , 2008 ) , while endothelial cells utilize polymerization-driven protrusion during angiogenesis ( Lamalice et al . , 2007 ) . Recent evidence indicates that physical confinement in a non-adhesive environment may drive a change of migration modes . In particular , contractility and extracellular pressure can drive a switch from polymerization/adhesion-based to bleb/friction-based motilities known as the mesenchymal-to-amoeboid transition ( MAT ) ( Bergert et al . , 2015; Liu et al . , 2015; Ruprecht et al . , 2015 ) . Cancer cells are known to be highly contractile , making them prone to blebbing ( Bergert et al . , 2012 ) , while tumors are known to be sites of high turgor pressure ( Jain , 1987 ) , both properties being conducive to MAT . Indeed , intravital imaging revealed that melanoma and breast cancer cells migrate by blebbing in live mice ( Tozluoglu et al . , 2013 ) . Thus , metastatic cancer cells are hypothesized to be susceptible to MAT in vivo , with the plasticity of their migration modes and lack of specificity in adhesion contributing to their high invasivity and difficulty targeting ( Lammermann and Sixt , 2009 ) . Polymerization-driven mesenchymal migration and bleb-based amoeboid migration are mechanistically distinct . In mesenchymal migration , actin polymerization initiated by localized activation of a filament-nucleating factor drives the formation of actin networks or bundles that mediate lamellipodial or filopodial membrane protrusion ( Skau and Waterman , 2015 ) . In contrast , in blebbing cells , membrane protrusion is mediated by myosin II contractility-induced hydrostatic pressure that drives a bubbling-out of the plasma membrane at a site of local weakness in the cortical actin cytoskeleton ( Charras et al . , 2005; Charras and Paluch , 2008 ) . While the molecular mechanisms mediating polymerization-based protrusion and contractility-induced pressure are reasonably well understood , the molecules responsible for local changes in the actin cortex that allow bleb formation are not known . It is hypothesized that blebs could form either by local down-regulation of membrane-cytoskeleton linkers such as those of the ezrin/radixin/moesin family ( Estecha et al . , 2009; Lorentzen et al . , 2011 ) , or by local weakening of sites in the cortical actin network itself ( Charras et al . , 2006 ) . Indeed , experiments using actin depolymerizing drugs have shown cortex integrity to be an important factor in regulating blebbing ( Charras et al . , 2006 ) . Local inhomogeneity in cortical integrity could be mediated by regulation of actin filament number or organization . In turn , filament number could be controlled by modulating the activity of filament nucleators , cappers or depolymerizers , while actin organization could be regulated by filament crosslinkers , bundlers or motor proteins . However , the proteins critical to regulating actin cortical organization and mechanics during blebbing are not known . In this study , we focused on identifying the mechanism for regulating the actin cortex of cancer cells during MAT and bleb-based migration . Many highly invasive cancers are known to be caused by mutations that activate the oncogenic EGF/Ras/Raf/MEK/Erk pathway ( Downward , 2003; Roberts and Der , 2007 ) . It is well-known that Erk-mediated activation of myosin II promotes cell contractility ( Klemke et al . , 1997 ) . Accordingly , we concentrated on actin regulatory proteins downstream of this pathway that may synergize with contractility to affect blebbing . Epidermal growth factor receptor pathway substrate 8 ( Eps8 ) is an actin bundling and capping protein ( Disanza et al . , 2004 , 2006; Hertzog et al . , 2010 ) that plays a critical role in development of the nervous , auditory and reproductive systems ( Lie et al . , 2009; Menna et al . , 2009 , 2013; Manor et al . , 2011 ) and its upregulation in cancers correlates with invasivity and poor prognosis ( Griffith et al . , 2006; Wang et al . , 2009; Kang et al . , 2012 ) . Eps8 was originally identified as an actin binding protein downstream of the EGF receptor and to regulate actin through a complex with SOS1 and Abi1 ( Fazioli et al . , 1993; Scita et al . , 1999 ) . Additionally , the actin filament capping activity of Eps8 is inhibited by Erk through phosphorylation ( Menna et al . , 2009 ) . Thus , the dual actin regulatory functions and targeting by Erk make Eps8 a good candidate for regulating the cortex downstream of oncogenic signaling . By combining imaging of fluorescently tagged proteins and atomic force microscopy ( AFM ) with targeted mutations in Eps8 , the work described here tests the hypothesis that Eps8 acts as a key effector of Erk to modulate actin cortex mechanics and thereby mediate bleb-based migration of cancer cells . We find that Eps8 bundling activity promotes cortex tension and pressure to drive MAT in non-adherent , confined cells , which migrate with a characteristic ‘leader-bleb’ morphology . Within leader blebs , Eps8 capping and bundling activities act antagonistically , and phosphorylation by Erk mediates this effect . Using a FRET biosensor for Erk , we document a massive concentration of kinase activity within leader blebs and find that leader bleb contents are trapped by the bleb neck that separates the bleb from the cell body . Our results identify a mechanism by which Eps8 may promote the transition to rapid , unregulated migration of cancer cells in confinement that may be critical to their highly invasive behavior in vivo .
To determine the role of Eps8 in cancer cell blebbing and migration , we utilized human A375 melanoma cells which carry a mutation in B-Raf ( V600E ) that activates the Raf/MEK/Erk pathway ( Davies et al . , 2002 ) . We imaged Eps8 and the actin cytoskeleton by spinning disk confocal microscopy in cells that were fixed and stained with fluorescent phalloidin and either immunolabeled with antibodies to Eps8 or expressing Emerald-tagged mouse Eps8 ( Emerald-mEps8 ) . When cells were plated on fibronectin-coated coverslips to promote adhesion and spreading , actin formed a dense meshwork in the lamellipodia near the cell edge , and circumferential arcs and stress fibers in the lamella and cell body ( Figure 1A ) . Both endogenous and Emerald-mEps8 localized primarily to lamellipodia and arcs , but were absent from stress fibers ( Figure 1A , Figure 1—figure supplement 1A ) . To determine Eps8 distribution in a non-adhesive environment , A375 cells co-expressing Emerald-mEps8 and FusionRed-tagged F-tractin ( an actin filament binding peptide , [Schell et al . , 2001] ) were plated on uncoated glass . Here , cells were rounded , and confocal sections midway through the cell Z-axis revealed a dense band of cortical actin at the periphery of the cell body . From this cortical band extended many blebs that possessed thin , continuous rims of cortical actin at their membranes ( Figure 1D ) . Eps8 was absent from the dense cortical band of the cell body , but localized with actin in punctae along the peripheral rim of actin in blebs ( Figure 1B , D ) . Co-expression of FusionRed-F-tractin and EGFP-tagged myosin II regulatory light chain ( GFP-MII-RLC , a marker of myosin II isoforms ) in non-adherent cells showed that myosin II was concentrated on the dense cortical band and was at very low levels or absent from the bleb periphery ( Figure 1F ) . Thus , Eps8 shows differential localization to actin structures in adherent spread and non-adherent blebbing melanoma cells . 10 . 7554/eLife . 08314 . 003Figure 1 . Eps8 is recruited early to bleb membranes and forms a gradient across the length of a ‘leader bleb . ’ ( A ) Confocal image of the ventral Z-plane of human A375 melanoma cells expressing Emerald-tagged mouse Eps8 ( green ) plated on fibronectin-coated glass and stained with phalloidin ( red ) . ( B–G ) Confocal images through the central Z-plane of A375 cells plated on uncoated glass . ( B ) Cell expressing Emerald-mEps8 . ( C ) Time-lapse series of Emerald-mEps8 ( green ) and rhodamine-dextran ( red ) used as a negative stain in the culture media to detect the position of the cell boundary . Times in ( C , E , G ) indicate seconds after the formation of a new bleb . ( D ) Cell expressing Emerald-tagged F-tractin to label actin filaments . ( E ) Time-lapse series of Emerald-F-tractin and rhodamine-dextran . ( F ) Cell expressing EGFP-tagged myosin II regulatory light chain ( EGFP-RLC ) . ( G ) Time-lapse series of EGFP-RLC and rhodamine-dextran . ( H ) Quantification of the average time of appearance of EGFP or Emerald-tagged cortical proteins ( EGFP-Ezrin , Emerald-mEps8 , Emerald-F-tractin , and EGFP-RLC ) relative to the time of maximal membrane protrusion determined from time-lapse series similar to those shown in ( C , E , G ) . ( I–K–N ) Confocal images through the ventral Z-plane of A375 cells plated on uncoated glass and confined under an agar slab . ( I ) Cell co-expressing Emerald-mEps8 ( green ) and FusionRed-F-tractin ( red ) . Boxed area is shown zoomed in ( K ) . ( J ) Average ratio of fluorescence in the leader bleb to that in the cell body for Emerald-mEps8 and FusionRed-F-tractin . ( K ) ( Left ) Example of 5 regions of interest ( ROIs ) , each 20% of the length of the leader bleb , used for ( Right ) regional analysis of the average fluorescence ( normalized to maximum ) of Emerald-mEps8 ( green ) and FusionRed-F-tractin ( red ) along leader blebs . ( L ) ( Left ) Image and ( right ) regional analysis of FusionRed-RLC ( red ) and Emerald-mEps8 ( green , image not shown ) fluorescence in leader blebs . ( M ) ( Top ) Image showing the position ( dotted line ) along which kymographs ( bottom ) of Emerald-mEps8 and FusionRed-F-tractin were made from time-lapse videos of their dynamics in leader blebs . Scale bar: 2 min ( N ) Color encoded time-overlay of images of Emerald-F-tractin in a migrating cell . Error is SEM , *p ≤ 0 . 05 , **p ≤ 0 . 01 , ***p ≤ 0 . 001 , ****p ≤ 0 . 0001 , NS: p > 0 . 05 . See also Videos 1–3 . DOI: http://dx . doi . org/10 . 7554/eLife . 08314 . 00310 . 7554/eLife . 08314 . 004Figure 1—figure supplement 1 . Eps8 and ezrin localization . ( A ) Confocal image of the ventral Z-plane of human melanoma A375 cells spread on fibronectin-coated glass and immunostained for endogenous Eps8 ( green ) and phalloidin to stain actin ( red ) . ( B ) ( Left ) Central Z-plane confocal image of EGFP tagged ezrin in cells plated on uncoated glass . ( Right ) time-lapse series of EGFP-ezrin ( green ) and rhodamine-dextran ( red ) used as a negative stain in the culture media to detect the position of the cell membrane . Time ( sec ) relative to the formation of a bleb . ( C , D ) Cells confined between uncoated glass and an agar slab , boxed areas shown zoomed to the right . ( C ) Confocal image of the central Z-plane of Emerald-mEps8 ( green ) co-expressed with FusionRed-F-tractin ( red ) . ( D ) Central Z-plane of Emerald-mEps8 ( green ) and FusionRed-myosin II regulatory light chain ( RLC , red ) in cells confined between uncoated glass and an agar slab . DOI: http://dx . doi . org/10 . 7554/eLife . 08314 . 00410 . 7554/eLife . 08314 . 005Figure 1—figure supplement 2 . Detailed view of Eps8 and actin localization . ( A ) Confocal image of the ventral Z-plane of an A375 cell confined between uncoated glass and an agar slab that was expressing Emerald-mEps8 ( green , upper left ) and FusionRed-F-tractin to stain actin ( red , lower left ) . Higher magnification views of color-encoded overlay shown at ( center ) and ( right ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08314 . 005 We next sought to determine the dynamics of Eps8 in blebbing cells . Previous studies have shown that shortly after bleb protrusion , the membrane-cytoskeleton linker protein ezrin is recruited to the bleb membrane , followed by the assembly of actin , then myosin II , which induces bleb retraction ( Charras et al . , 2006 ) . To determine the timing of Eps8 arrival at the bleb membrane relative to these proteins , we subjected A375 cells expressing EGFP tagged mEps8 , ezrin , F-tractin , or MII-RLC to time-lapse confocal imaging at 5 s intervals . To mark the position of the cell membrane in negative-image , cells were mounted in media containing red fluorescent dextran ( Figure 1C , E , G , Figure 1—figure supplement 1B ) . Analysis of time-lapse image series showed that similar to previous studies , ezrin appeared rapidly on the newly protruded bleb membrane at ∼5 s after bleb formation ( Figure 1H , Figure 1—figure supplement 1B ) . Recruitment of Eps8 and actin to the membrane occurred on a similar timescale as ezrin ( Figure 1C , E , H ) . In contrast , MII-RLC appeared ∼30 s after protrusion , and coincided with the onset of bleb retraction ( Figure 1G , H ) . These results indicate that in non-adherent cells , Eps8 recruitment to bleb membranes occurs concurrently with actin assembly . To induce migration of non-adherent melanoma cells , we confined A375 cells expressing Emerald-mEps8 and FusionRed-F-tractin between an agarose pad and uncoated glass ( Bergert et al . , 2012 ) . Strikingly , cells in these conditions had reduced blebbing on the cell body , but formed a single very large sausage-shaped bleb that generally excluded the nucleus , and which was defined by a thin neck at the junction between the large bleb and the spherical cell body ( Figure 1I and Video 1 ) . This morphology was very similar to the ‘A2’ or ‘stable bleb’ phenotypes recently described for non-adherent cells migrating under confinement ( Bergert et al . , 2015; Liu et al . , 2015; Ruprecht et al . , 2015 ) . Long-term imaging showed that ∼40% of non-adherent , confined cells exhibited apparently rapid migration in the direction of the very large bleb , which we will thus call a ‘leader bleb’ ( Video 2 ) . Indeed , A375 cells adhered to fibronectin-coated coverslips ( either 5 or 50 µg/ml ) did not migrate , while leader bleb-based migration of confined non-adherent cells was very fast at 1 . 15 ± 1 . 02 µm/min ( mean ± SE , Figure 1N ) . 10 . 7554/eLife . 08314 . 006Video 1 . Human melanoma A375 cells form a single prominent bleb when confined under an agar pad . Comparison of central Z-plane confocal time-lapse videos of Emerald-mEps8 dynamics in A375 cells plated on human plasma fibronectin coated glass ( left ) , uncoated glass ( middle ) and confined between uncoated glass and an agar pad ( right ) . Scale bar: 5 µm , elapsed time in seconds shown . DOI: http://dx . doi . org/10 . 7554/eLife . 08314 . 00610 . 7554/eLife . 08314 . 007Video 2 . Human melanoma A375 cells migrate in the direction of a ‘leader bleb . ’Central Z-plane confocal time-lapse video of Emerald-F-tractin showing actin dynamics during ‘leader bleb’ based migration of an A375 cell confined between uncoated glass and an agar pad . Scale bar: 5 µm , elapsed time in minutes shown . DOI: http://dx . doi . org/10 . 7554/eLife . 08314 . 007 Confocal imaging revealed that both Eps8 and F-actin were concentrated within the leader bleb relative to the cell body ( Figure 1J , Figure 1—figure supplement 1C , Figure 1—figure supplement 2 ) . To analyze the spatial distribution of actin and Eps8 within the leader bleb , we determined their average intensities within five regions of interest ( ROIs ) , each representing 20% of the length of the leader bleb from neck ( region 1 ) to distal tip ( region 5 ) ( Figure 1K ) . This showed that actin was distributed as circumferential bundles around the short axis of most of the length of the leader bleb , but was reduced at the bleb tip ( Figure 1I , J , Figure 1—figure supplement 1C , Figure 1—figure supplement 2 ) . Eps8 localized in a punctate manner along actin bundles , forming a gradient , with the highest concentration near the neck connecting the bleb to the cell body and the lowest level at the distal tip of the leader bleb ( Figure 1K , Figure 1—figure supplement 1D , Figure 1—figure supplement 2 ) . Time-lapse imaging and kymograph analyses showed that Eps8 and F-actin underwent coordinated retrograde movement from the distal leader bleb tip towards the bleb neck ( Figure 1M and Video 3 ) . Similar imaging of EGFP-MII-RLC showed that myosin II was distributed around the cortical band of the cell body , highly concentrated in the narrow neck at the bleb neck where the bleb connected to the cell body , and nearly absent from the distal half of the leader bleb , similar to previous reports ( Figure 1L ) ( Bergert et al . , 2015; Liu et al . , 2015; Ruprecht et al . , 2015 ) . Together , these results show that in non-adherent cells , Eps8 localizes rapidly to bleb membranes as actin assembles . When non-adherent cells are confined , actin and Eps8 concentrate in leader blebs where they exhibit a directional assembly gradient and rearward flow towards the contractile bleb neck , and this cortical flow is coordinated with rapid cell movement directed by the leader bleb . 10 . 7554/eLife . 08314 . 008Video 3 . Eps8 and actin flow towards the bleb neck in leader blebs . Ventral Z-plane confocal time-lapse video of Emerald-mEps8 ( green ) and FusionRed-F-tractin ( red ) dynamics showing their coordinated flow towards the bleb neck in an A375 cell confined between uncoated glass and an agar pad . Scale bar: 5 µm , elapsed time in seconds shown . DOI: http://dx . doi . org/10 . 7554/eLife . 08314 . 008 We next sought to determine the requirement for Eps8 in the promotion of leader bleb-based migration and organization of the cortical cytoskeleton in confined , non-adherent cells . To test this , we used a small interfering RNA ( siRNA ) targeted to human Eps8 that resulted in 76 ± 2 . 1% ( mean ± SE ) depletion of the protein in A375 cells after 24 hr ( Eps8-KD , Figure 2A ) . Expression of FusionRed-F-tractin in non-adherent Eps8-KD cells confined under an agar pad followed by confocal imaging revealed that Eps8 depletion generally inhibited the formation of large leader blebs ( Figure 2B ) . Instead , confined non-adherent cells resembled unconfined non-adherent cells , with a distribution of various-sized protruding and retracting blebs around their perimeter ( Figure 2B ) , and the fraction of cells that underwent leader bleb-based migration was reduced by nearly half ( Figure 2D ) . To quantify the effects of Eps8-KD on bleb size , we used a simple definition of a leader bleb as the single largest bleb made by the cell , expressed as a percent of cell body area . This showed that Eps8-KD decreased leader bleb area by nearly half compared to non-targeting siRNA ( Figure 2C and Supplementary file 1A ) . Expression of Emerald-mEps8 fully rescued leader bleb size and migration in Eps8-KD cells ( Figure 2C–D , Figure 2—figure supplement 1C , Supplementary file 1A ) . Thus , Eps8 promotes formation of a large leader bleb to drive migration of confined , non-adherent A375 cells . 10 . 7554/eLife . 08314 . 009Figure 2 . Eps8 promotes leader bleb-based migration by maintaining actin bundles towards the distal bleb tip . ( A ) ( Left ) Representative blot and ( right ) quantitation of Western blot analyses of Eps8 and Erk in lysates of A375 ( gray ) , U20S ( blue ) and A549 ( green ) cells that were treated with non-targeting siRNAs ( non-targeting ) or siRNAs targeting human Eps8 to deplete Eps8 ( hEps8 siRNA ) . ( B–H ) Images and analyses of cells plated on glass and confined under an agar slab . ( B ) Color encoded time-overlay of confocal images through the central Z-plane of an A375 cell depleted of Eps8 and expressing soluble EGFP . ( C , G ) Tukey box plots showing ( C ) quantification of leader bleb area expressed as a % of cell body area for cells treated with non-targeting or Eps8 siRNAs , with or without the additional expression of mouse Eps8 ( mEps8 WT ) . ‘+’ and line denote the mean and median , respectively . ( D ) Quantitation of the percent of cells that migrate from time-lapse phase contrast videos , treatments as in ( C ) . ( E , F ) Confocal images through the ventral Z-plane of an A375 cell depleted of Eps8 and expressing Emerald-F-tractin ( E ) and FusionRed-myosin II regulatory light chain ( ( F ) , RLC ) , boxed area in E shown at right , overlay of ( E ) and ( F ) shown at left of ( F ) . ( G ) Analysis of cortical actin density in A375 cells ( see Materials and methods ) in the cell body from images of phalloidin , treatments as in ( C ) , normalized to the mean value of non-targeting control . ( H ) Regional analysis of the average fluorescence intensity ( H , normalized to maximum ) and bundle anisotropy ( H′ ) of FusionRed-F-tractin along leader blebs in A375 cells treated with either non-targeting or human Eps8 siRNAs . Each point represents the average value in a region of interest ( ROI ) that is 20% of the length of the leader bleb . ( I ) Western blot analyses of Eps8 , myosin II regulatory light chain ( MLC ) , and myosin II regulatory light chain phosphorylated on serine 19 ( pMLC ( S19 ) ) in lysates of A375 cells that were treated with non-targeting siRNAs ( non-targeting ) or siRNAs targeting human Eps8 to deplete Eps8 ( hEps8 siRNA ) . Error in ( A , H ) is SEM , *p ≤ 0 . 05 , **p ≤ 0 . 01 , ***p ≤ 0 . 001 , ****p ≤ 0 . 0001 , NS: p > 0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 08314 . 00910 . 7554/eLife . 08314 . 010Figure 2—figure supplement 1 . Myosin II localization is unperturbed in human melanoma A375 cells depleted of and rescued with Eps8 . ( A–C ) Confocal images of A375 cells confined between uncoated glass and an agar slab , magnification is the same in all panels . ( Left ) Central and ( right ) ventral Z-plane confocal images of FusionRed-myosin II regulatory light chain ( RLC ( red ) ) in cells treated with ( A ) non-targeting siRNA or ( B , C ) siRNA directed towards human Eps8 ( hEps8 siRNA ) with ( C ) or without ( A , B ) the additional expression of Emerald-tagged mouse Eps8 ( mEps8 ) and FusionRed-RLC ( red ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08314 . 01010 . 7554/eLife . 08314 . 011Figure 2—figure supplement 2 . Eps8 is required for leader bleb formation in A549 and U2OS cells . ( A ) Phase contrast image of an A549 cells ( Left ) or a U20S cells ( right ) plated on uncoated glass coverslip . ( B ) Color encoded time-overlay of confocal images through the central Z-plane of an A549 cell ( left ) or a U20S cell ( right ) depleted of Eps8 and expressing soluble EGFP and confined between uncoated glass and an agar slab . DOI: http://dx . doi . org/10 . 7554/eLife . 08314 . 011 We then sought to determine whether the requirement for Eps8 in leader bleb-based migration was a more general property of cancer cells or specific to A375 cells . We tested this in U2OS human osteosarcoma that are deleted for the cell cycle regulatory gene CDKN2A ( Catalogue of somatic mutations in cancer ) , and human lung cancer A549 cells that carry the K-Ras G12S oncogenic mutation ( Catalogue of somatic mutations in cancer ) . When plated on uncoated glass , both U20S and A549 cells rounded up and exhibited blebbing around their peripheries ( Figure 2—figure supplement 2 ) . When transfected with EGFP as a soluble marker and confined between uncoated glass and agar and subjected to time-lapse confocal microscopy , 36% of U20S cells and 46% of A549 cells took on a leader bleb morphology ( Figure 2—figure supplement 2 ) and underwent rapid migration ( Figure 2D ) . To test the requirement for Eps8 in this transition , we used siRNA to reduce Eps8 level by 53% and 57% in U20S and A549 cells , respectively ( Figure 2A ) . Analysis of confocal videos showed that Eps8 depletion generally inhibited the formation of large leader blebs ( Figure 2—figure supplement 2 ) , and quantitation confirmed that the fraction of both cell types that underwent leader bleb-based migration was reduced by more than half , and leader bleb area was reduced by 34% and 43% in U20S and A549 cells , respectively ( Figure 2C , D ) . Expression of Emerald-mEps8 rescued leader bleb size and migration in U20S and A549 cells that had been treated with Eps8 siRNA ( Figure 2C–D , Figure 2—figure supplement 2 , Supplementary file 1A ) . Thus , Eps8 is required for leader bleb formation to drive migration of confined , non-adherent cells in several cancer cell types , independent of the defect driving transformation . To analyze the effects of Eps8-KD on cytoskeletal organization in A375 cells ( Figure 2E–H ) , we utilized metrics that quantified both the density and bundle organization of the cytoskeleton in confined , non-adherent cells . For filament density , we measured average F-tractin intensity in both the dense cortical actin band of the cell body and in the leader bleb ( Figure 2G , H ) . To quantify bundle organization , we determined the local alignment ( ‘anisoptropy’ ) of F-tractin bundles within the leader bleb via the ImageJ plugin , FibrilTool , which uses the concept of nematic tensor to provide a local quantitative description of the anisotropy of fiber arrays and their average orientation in cells ( Boudaoud et al . , 2014 ) ( Figure 2H ) . To quantify the spatial variation in these parameters , we determined their values in five ROIs along the leader bleb length ( Figure 2H ) . This showed that compared to control , Eps8-KD induced a slight but significant reduction in the density of the cortical actin band in the cell body ( Figure 2E , G ) , but had no significant effect on the density or distribution of actin within leader blebs , where actin exhibited a gradient with the highest concentration at the neck ( Figure 2E , H ) . Regional analysis of F-tractin anisotropy showed that in control cells , actin bundle alignment also formed a gradient in the leader bleb , but in the opposite direction as the density gradient , such that bundles consistently were most highly aligned towards the distal regions ( Figure 2H″ ) , and decreased towards the neck . Knockdown of Eps8 did not destroy the anisotropy gradient , but made actin bundle organization highly variable along the largest bleb ( Figure 2H″ ) . Examination of the localization of EGFP-MII-RLC in Eps8-KD cells showed that similar to control , myosin II concentrated in the dense cortical band of actin around the cell body and at the neck of the largest bleb ( Figure 2F , Figure 2—figure supplement 1B ) . In agreement , western blotting for myosin II regulatory light chain phosphorylated on serine 19 ( pS19 MLC ) as a marker of activated myosin II showed no significant difference in myosin II activity between control and Eps8-KD ( Figure 2I ) . Together , these results show that in confined , non-adherent cells , Eps8 promotes enlargement of leader blebs independent of myosin II activity , where it acts to promote actin bundling towards the distal bleb tip , and is required for leader bleb-based migration . Since leader bleb formation is mediated by intracellular pressure induced by contractility in the cytoskeleton ( Bergert et al . , 2015; Liu et al . , 2015; Ruprecht et al . , 2015 ) , we next sought to determine the role of Eps8 in regulation of cortical cytoskeleton mechanical properties . We utilized an atomic force microscope ( AFM ) -based assay in which rounded , non-adherent cells are subjected to a minimal deformation with a tipless cantilever ( Figure 3A , B ) ( Fischer-Friedrich et al . , 2014; Ramanathan et al . , 2015 ) , and cortex tension and intracellular pressure can be extracted from the resulting force–displacement curves using measurements of the cell radius and actin cortex thickness ( Clark et al . , 2013 ) and a theory derived from a simple force balance of the applied cantilever normal force with the force due to intracellular pressure and the force from cortex tension ( see ‘Materials and methods’ ) . We first validated the approach using cells treated with the myosin II ATPase inhibitor blebbistatin . This showed that inhibition of myosin II did not significantly affect cell radius or cortical actin thickness , but cortex tension and intracellular pressure were both significantly reduced ( Figure 3C–D , Figure 3—figure supplement 1 , Supplementary file 1 ) , consistent with previous reports ( Tinevez et al . , 2009 ) . Similar measurements showed that compared to non-targeting control , Eps8-KD resulted in significant decreases in cortex tension and intracellular pressure , although these effects were not as strong as those induced by blebbistatin ( Figure 3C–D , Figure 3—figure supplement 1 , Supplementary file 1 ) . The effects of Eps8-KD on mechanical properties could be rescued by re-expression of Emerald-mEps8 ( Figure 3C–D , Figure 3—figure supplement 1 , Supplementary file 1 ) . Thus , Eps8 promotes cortex tension and increases intracellular pressure . Together with our above results , this suggests that Eps8 regulates cortical cytoskeletal organization to enhance cortical tension and increase intracellular pressure to mediate leader bleb-based migration . 10 . 7554/eLife . 08314 . 012Figure 3 . Cortex tension and intracellular pressure are maintained by Eps8 . ( A ) Schematic representation of the Atomic Force Microscope ( AFM ) based assay for determining cortex tension and intracellular pressure in A375 cells plated on uncoated glass . ‘kc’ cantilever spring constant , ‘d’ cantilever deflection , ‘z’ piezo Z displacement ( B ) ( Left ) confocal image of the central Z plane or ( right ) x-y projection of a 3D reconstruction of a Z-stack of a cell expressing EGFP as a volume marker . ( C , D ) Tukey box plots of cortex tension ( C ) and intracellular pressure ( D ) determined from AFM analysis . ‘+’ and line denote the mean and median , respectively . Cells were treated with 50 µM blebbistatin , non-targeting siRNAs , siRNAs targeting human Eps8 ( hEps8 ) with or without the additional expression of Emerald-tagged mouse Eps8 ( mEps8 ) . *p ≤ 0 . 05 , **p ≤ 0 . 01 , ***p ≤ 0 . 001 , ****p ≤ 0 . 0001 , NS: p > 0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 08314 . 01210 . 7554/eLife . 08314 . 013Figure 3—figure supplement 1 . Actin cortex thickness and cell radii are not significantly affected by depletion of and rescue with wild type Eps8 or its mutants . ( A ) Example central Z-plane confocal images of a bleb-free region of an A375 cell cortex plated on uncoated glass and stained with Alexa Fluor 568-conjugated Wheat Germ Agglutinin ( WGA , green ) to mark the position of the cell membrane and Alexa Fluor 647-conjugated phalloidin ( Phall , red ) to mark the position of the actin cortex for making thickness measurements . ( B ) Example line-scans of the cell membrane ( WGA , green ) and actin cortex ( phalloidin , red ) from ( A , white bar ) used for determining actin cortex thickness . ( C ) Average actin cortex thickness determined from confocal images of A375 cells plated on uncoated glass and stained as in ( A ) . ( D ) Cell radii measurements from phase contrast images of live A375 cells acquired during atomic force microscopy analyses . Error is SD . NS: p > 0 . 05 . Conditions are as follows: 50 µM blebbistatin , non-targeting siRNAs , siRNAs targeting human Eps8 ( hEps8 ) with or without the additional over-expression ( OE ) of Emerald-tagged mouse Eps8 ( mEps8 ) variants ( ‘mEps8Δcap , ’ ‘mEps8Δbund , ’ ‘mEps8-SATA , ’ Emerald-tagged versions of the respective constructs ( see Figure 4A ) or 10 µM U0126 . DOI: http://dx . doi . org/10 . 7554/eLife . 08314 . 013 We next sought to test whether the actin bundling activity of Eps8 is required for leader bleb formation and migration , cytoskeletal organization , and cellular mechanical properties . To accomplish this , we made alanine substitutions within the C-terminus of Emerald-tagged mouse Eps8 ( L757A/K759A , referred to as Emerald-mEps8Δbund , Figure 4A ) which have been previously shown to specifically block Eps8 bundling activity ( Hertzog et al . , 2010 ) . We co-expressed this together with FusionRed-F-tractin or FusionRed-MII-RLC in either wild type or Eps8-KD cells in a non-adherent , confined environment . This showed that Emerald-mEps8Δbund co-localized with actin at the thin cortical rim on bleb membranes , similar to wild type Eps8 ( Figure 4B , F , Figure 4—figure supplement 1 ) . However , in confined Eps8-KD cells , expression of Emerald-mEps8Δbund failed to rescue the defect in leader bleb size induced by loss of Eps8 , and these cells remained round and exhibited small blebs around their periphery with no dominant leader bleb ( Figure 4B–C , Figure 4—figure supplement 1 , Supplementary file 1A and Video 4 ) . Accordingly , Eps8-KD cells expressing Emerald-mEps8Δbund exhibited a more than 50% decrease in the fraction of cells migrating under confinement compared to Eps8-KD cells reconstituted with Emerald-mEps8 ( Figure 4E ) . Even stronger effects on leader bleb size were observed in wild-type A375 cells over-expressing Emerald-mEps8Δbund ( Figure 4D and Supplementary file 1B ) , indicating it acts as a dominant negative , likely by dimerizing with endogenous Eps8 ( Kishan et al . , 1997 ) . We thus used a dominant negative approach where appropriate to negate the possibility of off-target effects induced by siRNAs . Because cells carrying defects in Eps8 bundling activity lacked a large bleb , we were unable to determine the effects of this mutation on actin density or anisotropy in the leader bleb . However , analysis of actin density in the cortical band of the cell body showed that over-expression of Emerald-mEps8Δbund had no effect compared to over-expression of Emerald-mEps8 alone , and caused no detectable changes in the organization of MII-RLC ( Figure 4F , Figure 4—figure supplement 1 ) . AFM analysis showed that while over-expression of Emerald-mEps8 had no effect on cortical tension and intracellular pressure ( Figure 4H–I and Supplementary file 1 ) , over-expression of Emerald-mEps8Δbund significantly reduced cortex tension and intracellular pressure compared to untreated controls ( Figure 4H–I and Supplementary file 1 ) . These results show that the actin bundling activity of Eps8 facilitates the formation of large leader blebs by promoting cortical tension and intracellular pressure . 10 . 7554/eLife . 08314 . 014Figure 4 . Actin bundling by Eps8 promotes cortex tension and intracellular pressure to drive leader bleb formation . ( A ) Schematic representation of wild-type ( WT ) mouse Eps8 ( mEps8 ) mutant constructs . Top: wild-type Eps8 . ‘*’ indicates one part of the split PH domain , ‘dSH2’ indicates degenerate SH2 , ‘capping’ indicates domain required for actin capping and which is subject to negative regulation by the Raf/MEK/Erk pathway , ‘bundling’ indicates the domain required for actin bundling . ‘Δbund’ indicates the bundling defective double point mutant L757A/K759A ( red bars ) , ‘Δcap’ indicates the capping defective double point mutant V689D/L693D ( red bars ) , ‘SATA’ indicates double alanine point mutation of Erk phosphorylation sites S624A/T628A ( red bars ) . ( B–I ) Images and analyses of A375 cells plated on glass and confined under an agar slab . ( C , D , G–I ) Tukey box plots in which ‘+’ and line denote the mean and median , respectively . ( C , D ) Quantification of leader bleb area expressed as a % of cell body area for ( C ) cells treated with non-targeting or Eps8 siRNAs , with or without the additional expression of Emerald-mEps8 , Emerald-mEps8Δbund or EGFP human α-actinin or ( D ) cells over-expressing ( OE ) Emerald-mEps8 , Emerald-mEps8Δbund or EGFP human α-actinin . Data for mEps8 are re-displayed from Figure 2 for comparison . ( E ) Quantitation of the percent of cells that migrate from time-lapse phase contrast videos , treatments as in ( C ) . ( F ) Confocal images through the ventral Z-plane of a cell depleted of Eps8 and co-expressing Emerald-mEps8Δbund and either FusionRed-F-tractin ( top ) or FusionRed-myosin II regulatory light chain ( bottom , RLC ) . ( G ) Analysis of cortical actin density ( see Materials and methods ) in the cell body from images of phalloidin , treatments as in ( D ) , normalized to the mean value of over-expression of soluble EGFP ( EGFP alone ) . ( H , I ) Cortex tension ( H ) and intracellular pressure ( H ) determined from AFM analysis of cells under the conditions described in ( D ) . *p ≤ 0 . 05 , **p ≤ 0 . 01 , ***p ≤ 0 . 001 , ****p ≤ 0 . 0001 , NS: p > 0 . 05 . See also Video 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 08314 . 01410 . 7554/eLife . 08314 . 015Figure 4—figure supplement 1 . Eps8 bundling activity is not required for myosin II localization to the cortex of A375 cells . ( A ) Confocal images of A375 cells confined between uncoated glass and an agar slab . ( A ) Central and ( A′ ) ventral Z-plane confocal images of FusionRed-myosin II regulatory light chain ( RLC , red ) in cells treated with siRNA directed towards human Eps8 ( hEps8 siRNA ) and additionally expressing Emerald-tagged mouse Eps8 bearing double point mutations L757A/K759A that block its actin bundling activity ( green , ‘mEps8 Δbund’ ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08314 . 01510 . 7554/eLife . 08314 . 016Figure 4—figure supplement 2 . Ectopically expressed α-actinin localizes to the leader bleb cortex . ( A ) Confocal images of A375 cells treated confined between uncoated glass and an agar slab that were treated with siRNA directed towards human Eps8 ( hEps8 siRNA ) and additionally expressing FusionRed-F-tractin ( red ) to mark actin filaments and GFP-tagged human α−actinin ( green ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08314 . 01610 . 7554/eLife . 08314 . 017Video 4 . Formation of large leader blebs requires actin bundling by Eps8 . Central Z-plane confocal time-lapse video showing Emerald-Eps8 dbund ( L757A/K759A mutations in mouse Eps8 that block its bundling activity ) dynamics and small blebs in an A375 cell that has been depleted of Eps8 by siRNA and confined between uncoated glass and an agar pad . Scale bar: 5 µm , elapsed time in seconds shown . DOI: http://dx . doi . org/10 . 7554/eLife . 08314 . 017 To further test the notion that F-actin bundling is critical to leader bleb formation and cell mechanical properties , we sought to determine if expression of a different actin bundling protein affected leader blebs or could rescue the defect in leader bleb size induced by loss of Eps8 . We over-expressed the actin bundling protein α-actinin ( Podlubnaya et al . , 1975 ) tagged with EGFP together with FusionRed-F-tractin in either WT or Eps8-KD cells under non-adherent confinement . Confocal imaging of EGFP-α-actinin in either WT or Eps8-KD cells showed a remarkably similar localization as Emerald-mEps8 ( Figure 1I , Figure 1—figure supplements 1 , 2 ) , with α-actinin concentrated in leader blebs where it was localized in a gradient along circumferential actin bundles , although the labelling of the bundles was more continuous and less punctate than that of Emerald-mEps8 ( Figure 4—figure supplement 2 ) . Quantitative analysis showed that in Eps8-KD cells , over-expression of α-actinin was sufficient to partially rescue their defect in leader bleb size , but not to the same extent as expression of Emerald-mEps8Δbund ( Figure 4C–D and Supplementary file 1 ) . In addition , α-actinin over-expression in Eps8-KD cells partially rescued the effects of loss of Eps8 in cell migration ( Figure 4E ) . AFM analysis showed that over-expression of α-actinin increased cortical tension and intracellular pressure by more than 50% compared to either untreated cells or cells over-expressing Emerald-mEps8 ( Figure 4H–I and Supplementary file 1 ) . These results demonstrate the critical role of actin bundling in promoting cortical tension and generating intracellular pressure to mediate the formation of leader blebs , however they show that other functions or regulation of Eps8 cannot be compensated by α-actinin . We next sought to determine whether the actin filament capping activity of Eps8 is required for leader bleb formation , cytoskeletal organization , and cellular mechanical properties . We made aspartic acid substitutions within the C-terminus of Emerald-tagged mouse Eps8 ( V689D/L693D , referred to as Emerald-mEps8Δcap ) which have been previously shown to specifically block Eps8 capping activity ( Figure 4A ) ( Hertzog et al . , 2010 ) and regulate blebbing during cell division ( Werner et al . , 2013 ) . Strikingly , when confined under agar , Eps8-KD cells expressing Emerald-mEps8Δcap formed significantly larger leader blebs and the proportion of cells migrating nearly doubled compared to either untransfected controls or to Eps8-KD cells expressing Emerald-mEps8 ( Figure 5A , B , D , Supplementary file 1 and Video 5 ) . Over-expression of Emerald-mEps8Δcap similarly increased leader bleb area compared to over-expression of Emerald-mEps8 ( Figure 5C and Supplementary file 1B ) . Confocal imaging of confined Eps8-KD cells and analysis of re-expressed protein distribution together with FusionRed-F-tractin or -MII-RLC showed that unlike the gradient of Emerald-mEps8 with the highest level at the leader bleb base , Emerald-mEps8Δcap was equally distributed along the length of the leader bleb or slightly concentrated at the tip ( Figure 5E , G ) . Analysis of actomyosin organization showed that compared to expression of Emerald-mEps8 , expression of Emerald-mEps8Δcap in Eps8-KD cells decreased the level of actin in the dense cortical band of the cell body , although the density gradient of actin along the leader bleb and the distribution of MII-RLC was unchanged ( Figure 5E , F , G′ and Figure 5—figure supplement 1 ) . In contrast , the actin bundle anisotropy gradient in the leader bleb was highly enhanced , with Eps8-KD cells expressing Emerald-mEps8Δcap exhibiting nearly fivefold higher actin bundle anisotropy near the tips of their leader blebs compared to Eps8-KD cells expressing Emerald-mEps8 ( Figure 5G″ ) . AFM analysis of cortex mechanics showed that compared to over-expression of Emerald-mEps8 , over-expression of Emerald-mEps8Δcap significantly increased cortex tension and intracellular pressure ( Figure 5H–I and Supplementary file 1 ) . Together , these results show that the capping activity of Eps8 decreases actin density and mechanical properties of the cortex in the cell body , but acts to antagonize actin bundle formation in the distal region of leader blebs , and together this limits leader bleb size . This further suggests that the capping activity of Eps8 may be regionally regulated to maintain a gradient of actin bundle organization in the leader bleb . 10 . 7554/eLife . 08314 . 018Figure 5 . Eps8 actin capping activity limits leader bleb size by decreasing actin density and mechanical properties in the cell body cortex , and antagonizing actin bundling at the leader bleb tip . ( A–I ) Images and analyses of A375 cells plated on glass and confined under an agar slab . ( A ) Color encoded time-overlay of confocal images through the central Z-plane of a cell depleted of Eps8 by siRNAs targeting human Eps8 ( hEps8 siRNA ) and expressing Emerald-mEps8Δcap ( see Figure 4A ) . ( B , C , F , H , I ) Tukey box plots in which ‘+’ and line denote the mean and median , respectively . ( B , C ) Quantification of leader bleb area expressed as a % of cell body area for ( B ) cells treated with hEps8 siRNAs and additionally expressing wild type ( WT ) Emerald-mEps8 or Emerald-mEps8Δcap or ( C ) cells over-expressing ( OE ) Emerald-mEps8-WT or Emerald-mEps8Δcap . Data for mEps8 WT are re-displayed from Figure 2 for comparison . ( D ) Quantitation of the percent of cells that migrate from time-lapse phase contrast videos , treatments as in ( B ) . ( E ) Confocal images through the ventral Z-plane of cells depleted of Eps8 and co-expressing Emerald-mEps8Δcap and either FusionRed-F-tractin ( top ) or FusionRed-myosin II regulatory light chain ( bottom , RLC ) . ( F ) Analysis of cortical actin density ( see Materials and methods ) in the cell body from images of phalloidin , treatments as in ( C ) , normalized to the mean value of over-expression of soluble EGFP ( EGFP alone ) . ( G ) Regional analysis of the average fluorescence intensity ( G , G′ , normalized to maximum ) Emerald-mEps8 or Emerald-mEps8Δcap ( G ) or FusionRed-F-tractin ( G′ ) and bundle anisotropy ( G″ ) of FusionRed-F-tractin along leader blebs in cells treated with hEps8 siRNA . Each point represents the average value in a region of interest ( ROI ) that is 20% of the length of the leader bleb . ( H , I ) Cortex tension ( H ) and intracellular pressure ( H ) determined from AFM analyses of cells under the conditions described in ( D ) . *p ≤ 0 . 05 , **p ≤ 0 . 01 , ***p ≤ 0 . 001 , ****p ≤ 0 . 0001 , NS: p > 0 . 05 . See also Video 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 08314 . 01810 . 7554/eLife . 08314 . 019Figure 5—figure supplement 1 . Eps8 capping activity is not required for myosin II localization to the cortex or leader bleb in A375 cells . ( A ) Confocal images of A375 cells confined between uncoated glass and an agar slab . ( A ) central and ( A′ ) ventral Z-plane confocal images of FusionRed-myosin II regulatory light chain ( RLC , red ) in cells treated with siRNA directed towards human Eps8 ( hEps8 siRNA ) and additionally expressing Emerald-tagged mouse Eps8 bearing double point mutations V689D/L693D that inactivate its actin capping activity ( green , ‘mEps8 Δcap’ ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08314 . 01910 . 7554/eLife . 08314 . 020Video 5 . Actin capping by Eps8 limits leader bleb size . Ventral Z-plane confocal time-lapse video showing Emerald-Eps8 dcap ( V689D/L693D mutations in mouse Eps8 that block its actin capping activity ) dynamics and large leader bleb formation in an A375 cell that has been depleted of Eps8 by siRNA and confined between uncoated glass and an agar pad . Scale bar: 5 µm , elapsed time in minutes shown . DOI: http://dx . doi . org/10 . 7554/eLife . 08314 . 020 Erk-dependent phosphorylation of Eps8 on S624 and T628 has been shown to inhibit actin capping by Eps8 without altering its filament binding activity ( Menna et al . , 2009 ) , and our above results suggest that capping may be regionally regulated in melanoma cells migrating under non-adhesive confinement . Because A375 cells are known to have hyperactivation of the Erk pathway , we sought to test whether Erk activity or phospho-regulation on S624 and T628 of Eps8 regulates leader bleb formation , actin organization , and cortex mechanics . We first exploited the highly specific inhibitor U0126 to block Erk activity ( Figure 6A ) . We validated that U0126 inhibited Eps8 phosphorylation in A375 cells by expressing either Emerald-mEps8 or EGFP as a control in cells treated with or without 10 μM U0126 for 90 min , performing immunoprecipitation of the expressed proteins with anti-GFP antibodies from cell lysates , followed by western blot analysis with antibodies specific to Erk , activated Erk phosphorylated on T202/Y204 , or phospho-serine and phospho-threonine ( pS/T ) . This analysis showed , as expected , that U0126 inhibited Erk activation independent of the expression of Eps8 constructs . Similarly , U0126 strongly reduced the pS/T level in immunoprecipitated Emerald-mEps8 compared to untreated cells . This indicates that U0126 blocks Erk-mediated Eps8 phosphorylation . 10 . 7554/eLife . 08314 . 021Figure 6 . Inhibition of Erk activity perturbs leader bleb-based migration . ( A ) Top two panels: Western blot analysis of anti-GFP immunoprecipitates from lysates of A375 cells that were expressing EGFP or Emerald-mEps8 that were or were not treated with 10 µM U0126 to inhibit Erk . Blot was probed with antibodies specific to phospho-serine and phospho-threonine ( pS/T , upper panel ) or GFP ( upper middle panel ) . Bottom two panels: Western blot analysis of Erk and Erk phosphorylated on T202/Y204 ( pErk ) in lysates of untreated A375 cells that were expressing EGFP or Emerald-mEps8 that were or were not treated with 10 µM U0126 . ( B ) Immuno-localization of endogenous Eps8 ( green ) and phalloidin staining of actin ( red ) in untreated and U0126 ( 10 µM ) -treated A375 cells that were non-specifically adhered to poly-L-Lysine coated glass . Arrows: Eps8 localizing to the tips of filopodia . ( C–H ) Images and analysis of A375 cells plated on glass and confined under agarose . ( C ) Time-lapse confocal image series of a cell that was co-expressing FusionRed-F-tractin ( top ) and EGFP tagged myosin II regulatory light chain ( RLC , bottom ) . Time indicates minutes relative to perfusion with 10 µM U0126 . ( D , F , G , H ) Tukey box plots in which ‘+’ and line denote the mean and median , respectively , treatments as in ( A ) . ( D ) Quantification of leader bleb area expressed as a % of cell body area . ( E ) Quantitation of the percent of cells that migrate from time-lapse phase contrast videos . ( F ) Analysis of cortical actin density ( see Materials and methods ) in the cell body from images of phalloidin , normalized to the mean value of untreated cells . ( G , H ) Cortex tension ( G ) and intracellular pressure ( H ) determined from AFM analyses of cells . *p ≤ 0 . 05 , **p ≤ 0 . 01 , ***p ≤ 0 . 001 , ****p ≤ 0 . 0001 , NS: p > 0 . 05 . See also Video 6 . DOI: http://dx . doi . org/10 . 7554/eLife . 08314 . 02110 . 7554/eLife . 08314 . 022Figure 6—figure supplement 1 . Erk inhibition causes A375 cell flattening . ( A ) X-Y projections of 3D reconstructions of Z-stacks of paraformaldehyde fixed A375 cells plated on poly-L-Lysine coated glass to mediate non-specific adhesion to the coverslip during immunostaining for endogenous Eps8 ( green ) and stained with phalloidin to visualize actin ( red ) . ( Top ) An untreated cell non-specifically adhered poly-L-Lysine coated glass . ( Bottom ) A cell treated for 90 min with 10 µM U0126 to inhibit Erk . Z-stacks consisted of 30 confocal slices separated by 0 . 5 microns . Reconstructions were rendered using ‘Volume Viewer’ in Fiji ( http://fiji . sc/Fiji ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08314 . 022 We then examined the role of Erk activity in cytoskeletal organization and Eps8 localization . We localized endogenous Eps8 and actin in cells plated on poly-L-Lysine to mediate non-specific adhesion during immunostaining . This showed that U0126 treatment caused cells to flatten out at their base where they attached to the coverslip and to form actin bundles in the cell center and lamellipodia and filopodia with Eps8 on their tips , although they still possessed small blebs containing Eps8 on their dorsal surface ( Figure 6B and Figure 6—figure supplement 1 ) . When cells co-expressing FusionRed-F-tractin and EGFP-MII-RLC were confined under non-adherent conditions and perfused with U0126 , this remarkably induced rapid retraction of leader blebs and formation of actin bundles in the center of the cell body that lacked myosin II ( Figure 6C and Video 6 ) . Quantification showed that compared to untreated control , U0126 significantly reduced leader bleb area and density of the cortical actin band in the cell body , and completely blocked leader-bleb based migration ( Figure 6D , E , F ) . AFM analysis of cortical mechanics showed that U0126 reduced both cortical tension and intracellular pressure compared to control ( Figure 6G–H and Supplementary file 1 ) . These results illustrate two important points . First , they show that Erk activity mediates spatial regulation of the actin cytoskeleton , such that it promotes actin density in the cortex and inhibits actin bundling in the cell center . Second , Erk activity is required for maintaining cortical tension and intracellular pressure to drive leader bleb formation for adhesion-independent migration under confinement . 10 . 7554/eLife . 08314 . 023Video 6 . MEK/Erk activity regulates the organization of the actin cortex and is required to form large leader blebs . Ventral Z-plane confocal time-lapse video showing actin ( FusionRed-F-tractin , red ) and myosin II ( EGFP-myosin II regulatory light chain , RLC , green ) dynamics and leader bleb retraction in an A375 cell confined between uncoated glass and an agar pad . When the word ‘U0126’ appears , 10 µM U0126 was perfused into the media to inhibit MEK/Erk . Scale bar: 5 µm , elapsed time in minutes shown . DOI: http://dx . doi . org/10 . 7554/eLife . 08314 . 023 We next sought to test specifically if phospho-regulation on the Erk sites ( S624 and T628 ) of Eps8 were critical to leader bleb formation and cytoskeletal regulation . We generated a non-phosphorylatable mutant of Emerald-mEps8 ( Figure 4A , S624A/T628A , referred to as Emerald-mEps8-SATA ) that has been shown to have constitutive capping activity in vitro and inhibit filopodia formation in neurons ( Menna et al . , 2009 ) . We expressed Emerald-mEps8 or Emerald-mEps8-SATA in either control or Eps8-KD cells and subjected them to confinement under non-adhesive conditions . This showed that either reconstitution of Eps8-KD or over-expression with Emerald-mEps8-SATA blocked leader bleb formation , significantly reducing leader bleb size by ∼50% and strongly inhibiting cell migration compared to expression of Emerald-mEps8 in either wild type or Eps8-KD cells ( Figure 7A–D , Supplementary file 1 and Video 7 ) . Thus , S624 and T628 in Eps8 are required for leader bleb formation and migration under non-adhesive confinement . This further suggests that down-regulation of capping activity by phosphorylation at these sites promotes leader blebs , in agreement with our finding that the capping-deficient Eps8Δcap enhances leader blebs . 10 . 7554/eLife . 08314 . 024Figure 7 . MEK/Erk-mediated phosphorylation of S624 and T628 coordinates Eps8 capping and bundling activities to mediate leader bleb-based migration . ( A–J ) Images and analyses of A375 cells plated on glass and confined under an agar slab . ( A ) Color encoded time-overlay of confocal images through the central Z-plane of a cell depleted of Eps8 by siRNAs targeting human Eps8 ( hEps8 siRNA ) and expressing Emerald-mEps8 bearing double alanine point mutation of Erk phosphorylation sites S624A/T628A ( Emerald-mEps8-SATA ) . ( B , C , F , I , J ) Tukey box plots in which ‘+’ and line denote the mean and median , respectively . ( B–C ) Quantification of leader bleb area expressed as a % of cell body area for ( B ) cells treated with hEps8 siRNAs and additionally expressing wild type ( WT ) Emerald-mEps8 or Emerald-mEps8-SATA or ( C ) cells over-expressing ( OE ) Emerald-mEps8-WT or Emerald-mEps8-SATA . ( D ) Quantitation of the percent of cells that migrate from time-lapse phase contrast videos , treatments as in ( B ) . ( E ) Confocal images through the central Z-plane of cells depleted of Eps8 and co-expressing Emerald-mEps8-SATA and either FusionRed-F-tractin ( top ) or FusionRed-myosin II regulatory light chain ( bottom , RLC ) . ( F ) Analysis of cortical actin density ( see Materials and methods ) in the cell body from images of phalloidin , treatments as in ( C ) , normalized to the mean value from over-expression of Emerald-mEps8 WT . ( G ) Confocal images through the central Z-plane of cells treated with 10 µM U0126 and over-expressing either Emerald-mEps8 WT or Emerald-mEps8Δbund ( see Figure 4A ) and actin stained with fluorescent phalloidin . ( H ) Quantitation of the percent of cells containing central actin bundles from confocal images of fluorescent phalloidin-stained cells . ‘F-tractin’ indicates over-expression of FusionRed-F-tractin , ‘UO’ indicates treatment with 10 µM U0126 , ‘mEps8 , ’ ‘mEps8Δcap , ’ ‘mEps8Δbund , ’ ‘mEps8-SATA , ’ indicate over-expression of EGFP-tagged versions of the respective constructs ( see Figure 4A ) . ( I , J ) Cortex tension ( I ) and intracellular pressure ( J ) determined from AFM analyses of cells under the conditions described in ( C ) . *p ≤ 0 . 05 , **p ≤ 0 . 01 , ***p ≤ 0 . 001 , ****p ≤ 0 . 0001 , NS: p > 0 . 05 . See also Video 7 . DOI: http://dx . doi . org/10 . 7554/eLife . 08314 . 02410 . 7554/eLife . 08314 . 025Figure 7—figure supplement 1 . Non-phosphorylatable Eps8 co-localizes with F-actin and not intermediate filaments . ( A ) Paraformaldehyde fixed A375 cell plated on uncoated glass over-expressing Emerald-tagged mouse Eps8 with the Erk phosphorylation sites mutated ( S624A/T628A , Eps8-SATA , green ) and stained with phalloidin to visualize actin ( red ) . ( B ) Live A375 cell confined between uncoated glass and an agar pad and expressing Emerald-mEps8-SATA ( green ) and FusionRed tagged human vimentin ( FusionRed-vimentin ) to mark intermediate filaments ( red ) . ( A , B ) Confocal images of the central ( left ) or ventral ( right ) Z-planes . Boxed area shown zoomed at bottom . DOI: http://dx . doi . org/10 . 7554/eLife . 08314 . 02510 . 7554/eLife . 08314 . 026Figure 7—figure supplement 2 . Inhibition of Eps8 capping activity is not sufficient to maintain leader blebs in the presence of Erk inhibitor . ( A ) Time-lapse confocal image series of an A375 cells plated on glass and confined under agarose cell that was treated with siRNA directed towards human Eps8 ( hEps8 siRNA ) and additionally expressing Emerald-tagged mouse Eps8 bearing double point mutations V689D/L693D that inactivate its actin capping activity ( green , ‘mEps8 Δcap’ ) . Time indicates minutes relative to perfusion with 10 µM U0126 to inhibit Erk . DOI: http://dx . doi . org/10 . 7554/eLife . 08314 . 02610 . 7554/eLife . 08314 . 027Video 7 . Erk phosphorylation of Eps8 is necessary for large leader bleb formation . Ventral Z-plane confocal time-lapse video showing Emerald-Eps8 SATA ( S624A/T628A mutations in mouse Eps8 that block its ability to be phosphorylated by Erk ) dynamics and small blebs in an A375 cell that has been depleted of Eps8 by siRNA and confined between uncoated glass and an agar pad . Scale bar: 5 µm , elapsed time in seconds shown . DOI: http://dx . doi . org/10 . 7554/eLife . 08314 . 027 We then examined the effects of the Eps8 phospho-mutant on cytoskeletal organization . Interestingly , expression of Emerald-mEps8-SATA together with FusionRed-F-tractin showed that the Eps8 phospho-mutant was depleted from the cortex and instead formed thick bundle structures in the cell center ( Figure 7E ) , similar to the actin bundles seen in cells treated with U0126 ( Figure 6C ) , but the bundles were not labeled with F-tractin ( Figure 7E ) . However , fixation and staining with antibodies to vimentin or with fluorescent phalloidin showed that the Emerald-mEps8-SATA cables did not co-localize with intermediate filaments , but were dense with actin , suggesting that F-tractin and Eps8 may compete for the same binding site in actin bundles ( Figure 7—figure supplement 1A , B ) . Analysis of fluorescent phalloidin intensity in the cortical band of the cell body showed that Emerald-mEps8-SATA did not alter actin density compared to Emerald-mEps8 ( Figure 7F ) . In addition , co-expression of Emerald-mEps8-SATA together with FusionRed MII-RLC in Eps8-KD cells showed that myosin II continued to localize to the peripheral cortex , but the central Eps8/actin bundles lacked myosin II ( Figure 7E ) . Quantification of the percentage of cells with Eps8/actin bundles showed that expression of F-tractin did not induce bundle formation , while over-expression of Emerald-mEps8 or Emerald-mEps8-SATA induced bundles in ∼30% or ∼70% of cells , respectively ( Figure 7H ) . This indicates that Emerald-mEps8-SATA promotes the formation of excessive non-contractile actin bundles in the cell center . Together with our previous results , this suggests that the S624 and T628 Erk phosphorylation sites in Eps8 , are required for suppressing central actin bundles , while other targets of MEK/Erk regulate cortical actin density . We then sought to determine how the S624A/T628A mutations in Eps8 mediated their effects on the cytoskeleton . We first tested whether this was due to a lack of Erk-mediated regulation on these sites . Treatment of non-adherent cells expressing either Emerald-mEps8 or Emerald-mEps8-SATA with 10 μM U0126 for 90 min followed by fixation and phalloidin staining showed that Erk inhibition induced Eps8/actin cables to the same extent , independent of the S624A/T628A mutations ( Figure 7G , H ) . Because phosphorylation of S624 and T628 are known to inhibit the actin capping activity of Eps8 , and yet non-phosphorylatable mutation of these sites ( which would be expected to have constitutive capping ) produced an ectopic actin cable effect , we wondered if the formation of Eps8/actin cables induced by Erk inhibition required either the capping or bundling activities of Eps8 . Over-expression of either Emerald-mEps8Δcap or Emerald-mEps8Δbund in either the presence or absence of U0126 did not induce Eps8/actin cables , indicating that both capping and bundling activities are required for cable formation induced by Erk inhibition ( Figure 7G , H ) . Furthermore , over-expression of Emerald-mEps8Δcap in cells treated with U0126 and confined under agar could not rescue the loss of leader blebs induced by perfusion of Erk inhibitor ( Figure 7—figure supplement 2 ) , indicating that Erk promotes leader blebs by other mechanisms in addition to down-regulation of Eps8 capping activity . Finally , we found that over-expression of an S624E/T628E Eps8 mutant ( Menna et al . , 2009 ) could not recapitulate the effects of Emerald-mEps8Δcap in cells , suggesting that the E substitutions do not act as phospho-mimics to inhibit Eps8 capping activity in cells ( not shown ) . However , together our results show that MEK/Erk-mediated phosphorylation of Eps8 on S624 and T628 inhibits the bundling- and capping-dependent formation of central actin cables by Eps8 . This also suggests that in addition to directly regulating actin capping ( Menna et al . , 2009 ) , these sites may affect the bundling activity of Eps8 in cells by an indirect mechanism . We then tested the role of S624 and T628 of Eps8 and their regulation by Erk in cortex mechanics . AFM analysis showed that compared to untreated or cells expressing Emerald-mEps8 , Emerald-mEps8-SATA expression reduced cortex tension and intracellular pressure to a level similar to treatment with U0126 ( Figures 7I , J , 6G , H and Supplementary file 1 ) . This was surprising , considering that this mutant had no effect on cortical actin density . However , it is possible that the cortical mechanics could be altered by the redistribution of mutant Eps8 from the cortex to central actin bundles . Together , these results suggest that Erk-mediated phosphorylation of S624 and T628 coordinates the local co-regulation of Eps8 capping and bundling activities to control cortical tension and intracellular pressure to mediate leader bleb formation and migration of confined cells . Our demonstrations that Erk activity and regional regulation of Eps8 activity are essential to leader bleb formation suggests that Erk activity itself may be regionally regulated in migrating melanoma cells . Because immunofluorescense ( of phospho-Eps8 or active Erk ) is not possible in cells confined under agar , we turned to a Fluorescence Resonance Energy Transfer ( FRET ) -based biosensor of active Erk called ‘Erk Kinase Activity Reporter containing EV linker’ ( EKAREV ) ( Harvey et al . , 2008; Komatsu et al . , 2011 ) . In short , a CFP-tagged phospho-peptide-binding domain is connected by a FRET-optimized ‘EV linker’ to a YFP-tagged Erk substrate peptide , and FRET is obtained when the substrate is phosphorylated by Erk . We first used the reporter to localize Erk activity in cells adhered to fibronectin-coated glass . Confocal imaging showed that EKAREV ( CFP channel ) was soluble , excluded from membranous organelles , and diffusely localized throughout the cell , while YFP/CFP ratio imaging revealed a low FRET signal indicating low Erk activity throughout the cell ( Figure 8A ) . In contrast , in non-adherent , blebbing cells , although the EKAREV reporter was evenly distributed , the ratio image showed heightened levels of FRET indicating higher Erk activity specifically in blebs ( Figure 8B , C ) . Importantly , treatment with U0126 reduced the EKAREV FRET ratio to minimal levels ( Figure 8—figure supplement 1 ) . Remarkably , time-lapse ratio imaging at 5 s intervals showed that a flash of high FRET appeared in the bleb periphery just after protrusion , and was maintained until bleb retraction ( Figure 8C ) . Thus , EKAREV reveals spatially and temporally localized Erk activity in blebs of non-adherent cells . 10 . 7554/eLife . 08314 . 028Figure 8 . Erk activity is concentrated in a gradient across leader blebs by a diffusion barrier at the bleb neck . ( A–E , G ) A375 cells expressing the EKAREV biosensor in which a CFP-tagged phospho-peptide-binding domain is connected by the EV linker to a YFP-tagged Erk substrate peptide and FRET is obtained when the substrate is phosphorylated by Erk . ( A ) ( Left ) Confocal image at the ventral Z-plane of the distribution of EKAREV ( CFP channel ) in a cell plated on fibronectin-coated glass . ( Right ) Pseudocolored ratio image of EKAREV FRET ( YFP over CFP emission ) . ( B ) ( Left ) Confocal image at the central Z-plane of the distribution of EKAREV ( CFP channel ) in a cell plated on glass . ( Right ) Pseudocolored ratio image of EKAREV FRET . ( C ) ( Top ) Time-lapse confocal image series at the central Z-plane of the distribution of EKAREV ( CFP channel ) in a cell plated on glass . ( Bottom ) Pseudocolored ratio image of EKAREV FRET . ( D–H ) Images and analysis of A375 cells cell plated on glass and overlaid with an agar slab . ( D ) ( Left ) Confocal image at the central Z-plane of the distribution of EKAREV ( CFP channel ) . ( Center ) Pseudocolored ratio image of EKAREV FRET . ( Right ) Zoom of the boxed area . ( E ) Regional analysis of the average fluorescence intensity of EKAREV ( CFP channel ) or the EKAREV FRET ( YFP over CFP emission ) ( normalized to maximum ) along leader blebs . Each point represents the average value in a region of interest ( ROI ) that is 20% of the length of the leader bleb . ( F ) Confocal image at the central Z-plane of the distribution of EGFP tagged Erk , MEK or B-Raf ( V600E ) in cells that are confined under agarose . ( G ) Quantification of the average ratio of signal in the leader bleb to that in the cell body for ( D , F ) . ‘Emission ratio’ indicates EKAREV FRET signal ( YFP over CFP emission ) , ‘EKAREV’ indicates EKAREV CFP channel , ‘EGFP alone’ indicates soluble EGFP . ( H ) A375 cell expressing soluble mEos2 . Time-lapse confocal image series at the central Z-plane . Box indicates the region near the neck of the leader bleb to which a pulse of 405 nm light was applied to locally photo-convert mEos2 from green to red fluorescence . Pseudocolor indicates the magnitude of red fluorescence from mEos2 after photo-conversion . ( I ) Speculative model for Eps8 function during leader bleb-based migration . *p ≤ 0 . 05 , **p ≤ 0 . 01 , ***p ≤ 0 . 001 , ****p ≤ 0 . 0001 , NS: p > 0 . 05 . See also Videos 8 , 9 . DOI: http://dx . doi . org/10 . 7554/eLife . 08314 . 02810 . 7554/eLife . 08314 . 029Figure 8—figure supplement 1 . The Erk inhibitor U0126 reduces EKAREV FRET and A375 cell blebbing . ( A ) Central Z-plane confocal image series of a live A375 cell plated on uncoated glass expressing the EKAREV biosensor treated with 10 µM U0126 . Pseudocolor reflects the magnitude of YFP/CFP ratio FRET signal induced by Erk-mediated phosphorylation of the biosensor ( red = high , blue = low ) . Time ( minutes ) relative to perfusion of U0126 . DOI: http://dx . doi . org/10 . 7554/eLife . 08314 . 029 We next sought to localize Erk kinase activity in non-adherent cells confined under agarose . Strikingly , confocal YFP/CFP ratio imaging revealed a strong concentration of high FRET signal indicating high Erk activity within leader blebs compared to the cell body ( Figure 8D and Video 8 ) . Quantification of the magnitude of FRET signal in the leader bleb relative to the cell body showed >250% enrichment of Erk activity in the leader bleb ( Figure 8G ) . Furthermore , regional analysis of the magnitude of FRET signal distribution within leader blebs showed a shallow gradient , with Erk activity highest at the distal tip ( Figure 8E ) . To determine if this effect could be caused by an enrichment of signaling proteins within leader blebs , we localized either soluble EGFP , or EGFP tagged versions of Erk , MEK or activated B-Raf ( V600E ) in cells confined under agarose ( Figure 8F ) . Quantification showed that EGFP alone as well as EKAREV were enriched by ∼10–12% in the leader bleb relative to the cell body ( Figure 8G ) . Although Erk , MEK and B-Raf ( V600E ) were all enriched within leader blebs by 15–20% relative to the cell body ( Figure 8F , G ) , this was not significantly higher than the enrichment of EGFP alone . Therefore , slight enrichment of soluble proteins within leader blebs may be a general phenomenon , likely because membranous organelles that exclude cytoplasmic markers are largely excluded from the leader bleb and maintained in the cell body . Together , our results show that in confined non-adherent cells , Erk activity is highly concentrated in leader blebs where it forms a gradient with peak activity at the bleb tip . 10 . 7554/eLife . 08314 . 030Video 8 . The EKAREV biosensor reveals concentrated Erk kinase activity within large leader blebs . Comparison of central Z-plane confocal time-lapse videos of Erk kinase activity in A375 cells plated on human plasma fibronectin coated glass ( left ) , uncoated glass ( middle ) and confined between uncoated glass and an agar pad ( right ) . Cells were expressing the EKAREV Erk activity biosensor . Pseudocolor reflects the magnitude of YFP/CFP ratio FRET signal induced by Erk-mediated phosphorylation of the biosensor ( red = high , blue = low ) . Scale bar: 5 µm , elapsed time in seconds shown . DOI: http://dx . doi . org/10 . 7554/eLife . 08314 . 030 Our observation that membranous organelles were excluded and soluble proteins were slightly concentrated in leader blebs led us to hypothesize that leader blebs trap their contents with a diffusion barrier formed by the constriction between the bleb and cell body . To test this , we utilized soluble monomeric Eos which converts from green to red fluorescence upon exposure to UV light . Using a localized pulse of 405 nm light , we photo-converted mEos in a rectangular region within the leader bleb a short distance away from the neck and then imaged the redistribution of red fluorescence by diffusion within the bleb and cell body . This showed that the majority of the red fluorescence was retained within and diffused throughout the leader bleb before equilibrating with the cell body ( Figure 8H and Video 9 ) . These results indicate that the bleb neck slows diffusion of soluble components between the leader bleb and cell body . Together , these results show that Erk activity concentrates in and forms a gradient across leader blebs , and may be trapped there by a diffusion barrier at the bleb neck . 10 . 7554/eLife . 08314 . 031Video 9 . The bleb neck restricts the diffusion of photo-converted mEos2 between large leader blebs and the cell body . Central Z-plane confocal time-lapse video of freely diffusing mEos2 before and after activation with a 0 . 5 s pulse of 405 nm laser light within a defined region ( box ) of a leader bleb showing slow diffusion between a large leader bleb and the cell body . Pseudocolor reflects the magnitude of red fluorescence from mEos2 after photo-conversion ( red = high , blue = low ) . Scale bar: 5 µm , elapsed time in seconds shown . DOI: http://dx . doi . org/10 . 7554/eLife . 08314 . 031
Our study shows for the first time a critical role for Eps8 and its regulation of the actin cytoskeleton in promoting cell cortex tension and intracellular pressure to induce the rapid , adhesion-independent migration of confined cancer cells . We demonstrate that , in contrast to cells adhered to ECM where Eps8 associates with filopodia and lamellipodia , in non-adherent melanoma cells , Eps8 localizes to bleb membranes as actin assembles . A novel , bleb-based mode of migration was recently described by several labs that occurs when non-adherent , highly contractile cells are squeezed in 3D confinement . The pressure induced by such an environment induces cells to undergo a MAT and adopt a characteristic polarized morphology where rapid cell motility is driven by cortical cytoskeletal flow that generates extracellular friction along a large sausage-shaped ‘leader bleb’ that drags along the cell body ( Bergert et al . , 2015; Liu et al . , 2015; Ruprecht et al . , 2015 ) . We show here that in cancer cells , the actin capping and bundling scaffold protein Eps8 is required to promote this morphological transition under confinement by enhancing cortical tension and promoting increased intracellular pressure . Our manipulations of Eps8 level and activities that defined its requirement for leader bleb formation and migration did not perturb the organization or activity of myosin II . This , together with previous work ( Bergert et al . , 2015; Liu et al . , 2015; Ruprecht et al . , 2015 ) , indicates that both contractility and proper organization of the cytoskeleton are critical to generating the cortical mechanical properties conducive to leader bleb-based migration under non-adhesive confinement . Our experimental results suggest a speculative model for how Eps8 mediates leader bleb-based migration ( Figure 8I ) . We find that the bundling activity of Eps8 is required for promoting the mesenchymal-to-leader bleb transition in confined , non-adherent melanoma cells . This identifies a critical role for actin bundles in promoting the cortical inhomogeneity and cellular pressure that have been postulated to mediate symmetry breaking to drive this drastic polarized cell shape change ( Ruprecht et al . , 2015 ) . Once the morphology transition occurs , we find that Eps8 and actin both form a gradient within the leader bleb , decreasing with distance from the neck , but absent from the very tip . Our demonstration the Eps8 localizes to bleb membranes on a similar timescale as ezrin suggests that their interaction ( Zwaenepoel et al . , 2012 ) may be critical to Eps8 cortical recruitment . Analysis of actin bundle anisotropy shows that actin bundling forms an opposing gradient , with the most organized actin bundles towards the bleb tip . Actin bundles wrapping around the short axis of the bleb are likely responsible for maintaining the sausage/allantoid shape . Our demonstration that inhibition of Eps8 capping activity further enhances actin bundle organization towards the leading bleb tip suggests that Eps8's bundling and capping activities act antagonistically , such that when capping is down-regulated , Eps8 may be a more efficient bundler . Thus , local regulation of Eps8 capping activity could promote polarization of actin organization with opposing gradients in actin density and bundling along the leader bleb . However , by manipulating Erk activity or the Erk phosphorylation sites in Eps8 , we demonstrate that Erk-mediated phosphorylation of S624 and T628 not only down-regulates the capping activity of Eps8 as shown previously ( Menna et al . , 2009 ) , but may in fact coordinate the regulation of Eps8 capping and bundling activities . Our use of a FRET biosensor surprisingly shows that Erk activity is massively concentrated in leader blebs , although the mechanism for this concentration is not clear . However , the high activity is likely trapped there by a diffusion barrier at the bleb neck , and within the leader bleb forms a shallow gradient with peak activity at the bleb tip . This suggests that Erk-mediated down-regulation of Eps8 capping activity at the leader bleb tip could drive the formation of opposing gradients in actin bundling and density along the bleb length . This gradient in actin organization , in turn , may be required for spatial organization of cortical tension , which we predict would mirror the gradient in actin bundling , to maintain the allantoid shape and drive cortical flow in the leader bleb . The increased cortical tension at the bleb tip would be opposed at the bleb neck by myosin II contractility , and the depletion of myosin II and actin from the very tip of the leader bleb would provide a site where high intracellular pressure could promote local protrusion of the leading membrane to mediate the leader-bleb-based migration of confined cells . Thus , our results identify a mechanism by which spatial regulation of Eps8 actin regulatory activities by Erk may promote the rapid , unregulated migration of melanoma cells that may be critical to their highly invasive behavior in vivo . The confinement of cells simulated here by an under agarose assay is thought to occur in animal cells during intraepithelial migration , within the perivascular space and between muscle fibers ( Charras and Sahai , 2014 ) . Additionally , leader bleb-based migration may be important to migration between tightly packed cells found in solid tumors . Our observation that inhibition of Erk activity is capable of blocking formation of large leader blebs is consistent with the notion that effectors of the pathway , such as Eps8 , are important to the migration of confined cancer cells . The Ras/Raf/MEK/Erk pathway is one of the most frequently upregulated pathways in cancer . In melanoma , this pathway is particularly active because of a commonly found activating mutation in B-Raf V600E ( Davies et al . , 2002 ) that bypasses Ras and the negative feedback that normally restrains MEK/Erk activity ( Logue and Morrison , 2012 ) . This feature predisposes melanoma cells to blebbing by supporting a Raf/MEK/Erk/MLCK/RLC/myosin contractility cascade . Drugs that target B-Raf V600E ( e . g . , vemurafenib ) have been shown to have remarkable effects in the short-term . Compensatory upregulation of the pathway by cancer cells frequently limits their effectiveness ( Logue and Morrison , 2012 ) . However , our work shows that other mechanisms for activating Erk , including via oncogenic Ras mutations in lung cancer cells or via serum factors in osteosarcoma cells also lead to Eps8-dependent bleb-based migration . The work described here elucidates the importance of a specific Erk effector , Eps8 , in the migration of confined melanoma cells . Therefore , targeting specific effectors as opposed to canonical signaling enzymes may have therapeutic value . Effectors that impact cell architecture through the regulation of the cytoskeleton could be particularly attractive targets especially for prevention of metastasis .
A375 , A549 and U2OS cells were all obtained from American Type Culture Collection ( ATCC , Manassas , VA ) and all were maintained for up to 15 passages in DMEM supplemented with 10% FBS , GlutaMAX ( Life Technologies , Carlsbad , CA ) , antibiotic-antimycotic ( Life Technologies ) and 20 mM Hepes pH 7 . 4 . Lipofectamine 2000 ( Life Technologies ) and RNAiMAX ( Life Technologies ) were used to transfect plasmids and small interfering RNAs , respectively . Cells were plated on 6-well glass bottom plates ( In Vitro Scientific , Mountain View , CA ) either directly , or after coating with either 5 or 50 µg/ml human plasma fibronectin ( Millipore , Billerica , MA ) or poly-L-Lysine ( Millipore ) , as noted . Agarose slabs for cell confinement ( Bergert et al . , 2012 ) were made by adding 750 mg of ultrapure agarose ( Life Technologies ) to 50 ml of 20 mM Hepes ( pH 7 . 4 ) , microwaving briefly , and pouring 4 ml into each well of a 6-well glass bottom plate ( In Vitro Scientific ) . After gelation , a hole was punched in the agarose using a 5 ml plastic test tube . Prior to confining cells , 3 ml of media was pipetted into each well and equilibrated with the agarose overnight . Before use , media was thoroughly vacuumed off and 200 microliters of media containing cells was added to the empty hole punch . To get cells under the agarose , a 1 ml pipette tip was placed into the hole punch containing media and cells and slid just under the agarose to gently lift a portion of the agarose , sucking the cells underneath , and the pad was gently set down . The remaining media and cells were then thoroughly vacuumed out of the hole punch . To prevent drying , the plate was sealed using parafilm . Prior to imaging , the plate was brought up to temperature for 1 hr in an incubator . To inhibit the Erk pathway , a working concentration of 10 µM U0126 ( Cell Signaling Technology , Beverly , MA ) was prepared by diluting a 1000× stock solution in DMSO into complete media and dissolved using a vortex mixer for 30 s before adding to cells . Phospho-Erk ( T202/Y204 ) ( #4370 ) and Erk ( #9102 ) antibodies purchased from Cell Signaling Technology were used to confirm inhibition of MEK by Western blotting of whole-cell lysates . Cells were treated with U0126 for 30 min for AFM assays and 90 min for light microscopy assays . Cells under agarose were treated by applying media containing U0126 directly on top of the agar pad . Blebbistatin ( Sigma Aldrich , St . Louis , MO ) was used at 50 µM and applied directly to the cells for 5 min before AFM analyses . EGFP-tagged mouse Eps8 ( GFP-mEps8 ) and Emerald-mEps8-SATA ( S624A/T628A ) were the kind gift of Giorgio Scita ( University of Milan ) . Bundling ( L757A/K759A , Δbund ) and capping ( V689D/L693D , Δcap ) defective versions of Emerald-mEps8 were made using Quick Change II XL ( Agilent Technologies , Santa Clara , CA ) and the following primers: The human-specific Eps8 siRNA ( #s4770 ) used during this study was from Life Technologies . Cells were incubated with siRNA for 24 hr prior to performing experiments . Knockdown was confirmed by Western blotting of whole-cell lysates for the presence of Eps8 . Whole-cell lysates were prepared by scraping cells into ice cold RIPA buffer ( 50 mM Hepes pH 7 . 4 , 150 mM NaCl , 5 mM EDTA , 0 . 1% SDS , 0 . 5% deoxycholate and 1% Triton X-100 ) containing protease and phosphatase inhibitors ( Roche , Switzerland ) . Before loading onto 4–12% NuPAGE Bis-Tris gradient gels ( Life Technologies ) , lysates were cleared by centrifugation . Following SDS-PAGE , proteins in gels were transferred to nitrocellulose using an iBlot ( Life Technologies ) . Before blocking , proteins were fixed to the nitrocellulose by air drying the membrane overnight at room temperature . Blocking of membranes was then done in blocking buffer ( Hepes buffered saline containing 0 . 1% Triton X-100 , 1% BSA , 1% fish gelatin and 5 mM EDTA ) . Antibodies against Eps8 ( BD Biosciences , Franklin Lakes , NJ; #610143 ) , phospho-Erk ( Cell Signaling Technology #4370 ) , Erk ( Cell Signaling Technology #9102 ) , pMLC ( Rockland , Limerick , PA; #600-401-416 ) and MLC ( Rockland #600-401-938 ) were used at a 1:1000 dilution and incubated overnight at 4C in blocking buffer . IRDye 680RD and 800CW secondary antibodies ( LI-COR Biosciences , Lincoln , NE ) were then used at 1:5000 in blocking buffer for 2 hr at room temperature after extensive washing in Hepes Buffered Saline ( HBS ) containing 0 . 1% Triton X-100 . Bands were then resolved on an Odyssey scanner ( LI-COR Biosciences ) . 12 hr after transfection of plasmids encoding EGFP or Emerald-tagged mouse Eps8 , A375 whole-cell lysates were prepared by scrapeing into ice cold RIPA buffer ( 50 mM Hepes pH 7 . 4 , 150 mM NaCl , 5 mM EDTA , 0 . 1% SDS , 0 . 5% deoxycholate and 1% Triton X-100 ) containing protease and phosphatase inhibitors ( Roche ) . Before immunoprecipitation , lysates were cleared by centrifugation . EGFP and Emerald tagged proteins were immunoprecipitated using a mouse GFP antibody ( Roche #11814460001 ) and Protein-G magnetic beads ( Life Technologies ) for 2 hr at 4C . Immunoprecipitated proteins were washed 4 times with ice cold RIPA buffer before running on 4–12% NuPAGE Bis-Tris gradient gels ( Life Technologies ) . Following SDS-PAGE , proteins in gels were electrotransferred to nitrocellulose using an iBlot ( Life Technologies ) . For U0126 treatments , A375 cells were treated with 10 µM U0126 for 90 min and inhibition of Erk activity was confirmed by Western blotting for phospho-Erk and Erk as described under ‘Western blotting . ’ Proteins with phosphorylated serine or threonine were detected using a rabbit anti-pS/T antibody ( Abcam , United Kingdom; #ab17464 ) used at 1:500 . Immunoprecipitated EGFP and Emerald tagged proteins was confirmed using a rabbit GFP antibody ( Life Technologies #A6455 ) used at 1:1000 . Cells were cultured , stained and imaged in 6-well glass bottom dishes ( In Vitro Scientific ) . Where noted , glass was first coated with 5 or 50 µg/ml human plasma fibronectin ( Millipore ) or poly-L-Lysine ( Millipore ) for 1 hr at 37C . Samples were fixed using 4% paraformaldehyde ( Electron Microscopy Sciences , Hatfield , PA ) in HBS for 20 min at room temperature . Permeabilization/blocking were performed using blocking buffer for an hour . Eps8 antibody ( BD Biosciences #610143 ) was used at a 1:250 dilution , fluorescently conjugated phalloidin ( Life Technologies ) at 1:200 , and fluorescently conjugated wheat germ agglutinin ( Life Technologies ) at 1:1000 . Each was incubated in blocking buffer overnight at room temperature . Samples were gently washed several times in HBS containing 0 . 1% Triton X-100 . An anti-mouse Alexa Fluor 488 secondary antibody ( Life Technologies ) was used to detect Eps8 . Imaging was performed in HBS . Immunofluorescence and time-lapse live-cell fluorescence microscopy was performed using the imaging system that is described in ( Shin et al . , 2010 ) . Briefly , this consisted of an automated Eclipse Ti microscope equipped with the Perfect Focus System ( Nikon , Japan ) , a servomotor driven X-Y stage with a piezo top plate ( Applied Scientific Instruments , Eugene , OR ) and a CSU-X1-A3 spinning disk confocal scan head ( Yokogawa , Japan ) . Illumination for confocal imaging was provided by solid state lasers ( 40 mW 442 nm for CFP; 100 mW 488 nm for EGFP; 100 mW 523 nm for YFP and 500 mW 561 nm for FusionRed ) that were directed to the microscope by a custom-designed optical fiber-coupled laser combiner ( Spectral Applied Research , Canada; Shin et al . , 2010 ) . For phase contrast imaging , illumination was provided by a quartz-halogen lamp using a 546 nm bandpass filter . Images were acquired using either a CoolSNAP HQ2 or MYO cooled CCD camera ( Photometrics , Tucson , AZ ) using either a 100× or 60× ( 1 . 4 NA , Plan Apo PH ) oil immersion objective lens and 0 . 9 NA condensor . Illumination , image acquisition , and microscope functions were controlled by Metamorph software ( Molecular Devices , Sunnyvale , CA ) . For time-lapse FRET imaging of the EKAREV biosensor , a confocal image through the central plane of the cell was first acquired using 442 nm excitation and the YFP emission filter ( FRET image ) , followed by an image acquired using 442 nm excitation and the CFP emission filter ( CFP image ) . For photo-activation of mEos2 , a Nikon A1R laser-scanning confocal microscope equipped with dual resonant scanners was used with a 60× ( 1 . 4 NA Plan Fluor ) oil immersion lens . A 0 . 5 s pulse of the 405 nm laser at 50% power within a defined region was used to photo-convert mEos2 . Images were then acquired using the 488 and 561 nm lasers in the red and green channels at 130 ms intervals to image the redistribution of fluorescence after photo-conversion . For all experiments , a stage-top incubator ( Tokai-Hit , Japan ) was used to maintain samples at 37C . Cells expressing EGFP or Emerald-tagged protein were plated on uncoated 6-well glass bottom plates ( In Vitro Scientific ) in media containing 0 . 1 mg/ml rhodamine-labeled dextran ( MW = 70 kDa , Sigma Aldrich ) . Time of arrival along the bleb perimeter was determined to be when EGFP or Emerald fluorescence was above background . Maximal protrusion of the bleb membrane as judged by displacement of rhodamine-dextran ( negative stain ) was set as time 0 . For determination of leader bleb area , confocal images of EGFP or Emerald through the central plane of confined cells were acquired for 4 hr at 5 min intervals . Leader bleb and cell body area were measured using images of soluble EGFP for non-targeting and hEps8 siRNA treated cells and Emerald tagged to Eps8 in rescue and over-expression experiments by outlining cells in each frame using Fiji ( http://fiji . sc/Fiji ) . The percent of migratory cells was determined from time-lapse phase-contrast images acquired for 4 hr at 5 min intervals . The number of moving vs stationary cells was counted during the course of the video . Round cells on uncoated 6-well glass bottom plates were prepared as described in ‘immunofluorescence . ’ To measure actin cortex density , cells were stained with Alexa Fluor 568 conjugated phalloidin ( Life Technologies ) . To measure cortex thickness , we used the methods of ( Clark et al . , 2013 ) . Briefly , cells were double-labeled with Alexa Fluor 568-conjugated wheat germ agglutinin ( Life Technologies ) and Alexa Fluor 647-conjugated phalloidin . Spinning disk confocal images through the central Z-plane of the cell were acquired using a 100× ( 1 . 4 NA ) objective lens . For cortex density , a 5 pixel wide line was drawn along a region of the cortex that was free of blebs and the mean fluorescence intensity was measured using Fiji ( http://fiji . sc/Fiji ) . Additionally , background fluorescence was measured by selecting a region inside the cell . Actin cortex density was then calculated as the mean fluorescence intensity at the cortex minus background fluorescence . For determining cortex thickness , images were analyzed by performing an intensity line-scan perpendicular to and across the cell edge on a dual color image , and the distance between the peaks of the phalloidin and wheat germ agglutinin fluorescence intensity was recorded and input into the equation reported by Clark et al . ( 2013 ) to calculate the cortex thickness . For regional analyses of protein distribution and anisotropy , five ROIs each representing 20% of the length of the leader bleb were drawn using Fiji ( http://fiji . sc/Fiji ) . Using FibrilTool ( Boudaoud et al . , 2014 ) , we measured the anisotropy of F-tractin fluorescence signal within the same regions used for determining protein distribution . FusionRed-F-tractin images having similar contrast , as determined by SD/mean , were used for anisotropy analyses . To generate emission ratio images , we wrote a Fiji ( http://fiji . sc/Fiji ) macro based on previously described work ( Pertz et al . , 2006 ) . Briefly , CFP and FRET images were first background-subtracted and corrected for bleaching by exponential fitting . CFP images were then thresholded to generate a binary mask . After multiplication by this mask , the FRET image was divided by the CFP image to yield an emission ratio reflecting Erk kinase activity throughout the cell . Ratio images were then pseudo colored using the ‘16 color’ LUT . Similarly , images of diffusing mEos2 were pseudo colored using the ‘16 color’ LUT . Force spectroscopy on non-adherent cells was performed using a Bioscope II AFM system ( Veeco , Plainview , NY ) mounted on an automated inverted epi-fluorescence microscope ( Nikon Eclipse TE2000 ) controlled by Metamorph software ( Molecular Devices ) . Illumination was provided by a mercury arc lamp and wavelengths selected by a Sedat filterset ( Semrock , Rochester , NY ) . The hybrid microscope instrument was placed on an acoustic isolation table ( Kinetic Systems , Boston , MA ) . A heating stage ( Veeco ) was used to maintain cells at 37C . Cells were located by EGFP fluorescence while cell radii were measured using bright-field images taken with a 40× ( 0 . 6 NA Plan Fluor ) objective lens and a QuantEM EMCCD camera ( Photometrics ) . After location of cells , a soft and tipless rectangular silicon nitride cantilever ( length: 350 ± 5 µm , width: 32 . 5 ± 3 µm , thickness: 1 ± 0 . 5 µm; MikroMasch , Bulgaria ) was brought to close proximity . The cantilever stiffness was first calibrated by performing a force curve on the stiffer glass-bottom dish to estimate the photodetector deflection sensitivity , and then by using the thermal noise fluctuation method ( Hutter and Bechhoefer , 1993 ) . The estimated spring constants for cantilevers used in force curves were 0 . 08–0 . 11 N/m . After calibration , the AFM cantilever was moved on top of a cell and lowered to gently deform it . Approximately ten successive force curves were performed in the same location on each cell per condition using 4 µm ramps with up to ∼1 nN applied force at 0 . 5 Hz . All AFM force spectroscopy measurements were analyzed to extract the cell mechanical properties using MATLAB ( Mathworks , Natick , MA ) . Before importing the force curves into MATLAB for analysis , each individual acquired curve was preconditioned by offsetting the y-axis to 0 and reformatted to a text file format using the NanoScope Analysis software ( Bruker , Billerica , MA ) . We discarded noisy force curves and/or curves that presented jumps possibly due to cantilever slippage or very weakly adhered and moving rounded cells . For initial contact estimation , user-dependent determination was employed , selecting the location when the force curves increased substantially from zero . This method does not require prior knowledge or assumption about the material and geometrical properties of the cell . For fitting , Z distances between 0–400 nm were relatively consistent in yielding good fits ( R2 > 0 . 9 ) . Curves with poor fits R2 < 0 . 9 were discarded from the analysis . A simple theory was derived for the quantitative determination of mechanical properties of non-adherent melanoma cells confined between two flat surfaces . This will be described in detail elsewhere . Briefly , when a spherical cell is rapidly ( 0 . 5 Hz ) but gently pressed down upon by the soft tipless cantilever , the cell is assumed to deform to an ellipsoid shape . The surface tension can be estimated by force balance in the vertical Z-direction . This force balance enables the relationship of the external applied cantilever force to the internal hydrostatic pressure and cortical tension . The Laplace's law describes the proportional relationship between tension and pressure . The derived expressions for the cortical tension and intracellular pressure of a non-adherent cell are: ( S1 ) T=kcπ ( 1Zd−1 ) , ( S2 ) P=2TR , where T is the cortical tension , P is the intracellular pressure , kc is the calibrated effective cantilever spring constant , Z is the Z-piezo extension distance , d is the cantilever deflection and R is the sample radius . Statistical significance between means was determined using a two-tailed Student's t-test in GraphPad Prism ( La Jolla , CA ) . All differences were considered significant if p ≤ 0 . 05 . | Cells within an animal have to be able to move both during development and later stages of life . For example , white blood cells have to move around the body and into tissues to fight off infections . Normally , cell movement is heavily controlled and will only happen when it is necessary to keep an animal healthy . However , cancer cells can bypass these controls and ‘metastasize’ , or spread to new sites in the body . Cells can move in several different ways: on the one hand , cells can use ‘mesenchymal’ movement , in which the skeleton-like scaffolding of molecules within a cell rearranges to push the cell forward . On the other hand , cells can employ ‘amoeboid’ movement , in which they squeeze their way forward by building up pressure in the cell . Although these different types of movement are only used by some healthy cells and not others , cancer cells can switch between the two . How they do this is still unclear , but now Logue et al . have studied this question using several microscopy techniques . Logue et al . watched skin cancer ( or melanoma ) cells migrating between a glass plate and a slab of agar , which mimics the confined spaces that cancer cells have to move through within the body . The images showed that the cancer cells formed so-called ‘leader blebs’ , finger-like projections that put cells on the right track . The experiments revealed that a protein called Eps8 was responsible for making the skin cancer cells move in this amoeboid fashion . The ‘blebbing’ caused by Eps8 is turned on by another protein called Erk that is often overactive in melanoma cells . Furthermore , Erk can accumulate near and within the cell blebs and this leads to the increased movement of the skin cancer cells . Studying cell movement in melanoma is particularly important because it is the metastatic tumors that kill patients . Therefore , increasing our basic understanding of how cells migrate could eventually lead to better treatment options that stop cancer cells from spreading . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology"
] | 2015 | Erk regulation of actin capping and bundling by Eps8 promotes cortex tension and leader bleb-based migration |
Cystic kidney diseases ( CKDs ) affect millions of people worldwide . The defining pathological features are fluid-filled cysts developing from nephric tubules due to defective flow sensing , cell proliferation and differentiation . The underlying molecular mechanisms , however , remain poorly understood , and the derived excretory systems of established invertebrate models ( Caenorhabditis elegans and Drosophila melanogaster ) are unsuitable to model CKDs . Systematic structure/function comparisons revealed that the combination of ultrafiltration and flow-associated filtrate modification that is central to CKD etiology is remarkably conserved between the planarian excretory system and the vertebrate nephron . Consistently , both RNA-mediated genetic interference ( RNAi ) of planarian orthologues of human CKD genes and inhibition of tubule flow led to tubular cystogenesis that share many features with vertebrate CKDs , suggesting deep mechanistic conservation . Our results demonstrate a common evolutionary origin of animal excretory systems and establish planarians as a novel and experimentally accessible invertebrate model for the study of human kidney pathologies .
The vertebrate kidney plays a pivotal role in the maintenance of organismal homeostasis in the face of changing external and internal conditions . Its myriad individual functions , including the removal of metabolic waste products , regulation of ion concentrations and acid/base balance , are all tied to two basic physiological processes: ( 1 ) the pressure-driven ultra-filtration of blood plasma across the glomerulus , whereby molecular sieves prevent the passage of large macromolecules ( e . g . , plasma proteins ) ; and ( 2 ) the subsequent modification of the resulting filtrate during its passage through the epithelial nephron tube ( Ruppert and Smith , 1988; Ruppert , 1994 ) . The parallel operation of many millions of glomerulus/nephron units allows formidable filtration rates , amounting to 170 liters of primary filtrate/day in a healthy human adult . In line with the pivotal homeostatic roles of the kidney , kidney diseases pose a serious health problem . The most common human kidney disorders are cystic kidney diseases ( CKDs ) , affecting nearly 12 million people worldwide ( Priolo and Henske , 2013 ) . CKDs encompass a wide range of hereditary , developmental , and acquired conditions ( Bisceglia et al . , 2006 ) , all of which share the pathological hallmark of fluid-filled cysts developing in the kidney . This has led to the suggestion that the molecular mechanisms causing cyst formation are similar , or at least , share a common pathway ( Watnick and Germino , 2003 ) . The molecular cloning of multiple CKD mutations and the realization that the affected genes all function at the primary cilia , basal bodies or centrosomes , has given rise to the ciliary hypothesis as a unifying disease mechanism of CKDs ( Yoder et al . , 2002; Mollet et al . , 2005; Fliegauf et al . , 2006 ) . Accordingly , the primary cilia of tubule cells are thought to act as flow sensors , eliciting intracellular calcium fluxes through stretch sensitive polycystin channels in response to flow-driven bending ( Praetorius and Spring , 2001 , 2003; Nauli et al . , 2003; Praetorius et al . , 2004 ) . These signals are thought to constitutively dampen cell proliferation , such that loss of filtrate flow or interruptions in the signal transduction process precipitate chronic overproliferation and consequently cyst formation ( Deane and Ricardo , 2012 ) . However , major mechanistic aspects of the ciliary hypothesis remain poorly understood , including the integration of the calcium signal with downstream transcriptional regulation of cell behavior ( Wilson and Goilav , 2007; Uhlenhaut and Treier , 2008; Deane and Ricardo , 2012; Kotsis et al . , 2013 ) , the extent by which cyst development can be understood as chronic activity of endogenous repair mechanisms ( Deane and Ricardo , 2012 ) , and the identity and origins of the ectopically overproliferating cells ( Murer et al . , 2002; Weimbs , 2007; Lodi et al . , 2012 ) . Further , these questions present an investigative challenge , given the poor experimental accessibility of the mammalian kidney as an internal and essential organ . The Xenopus pronephros and zebrafish pro- and mesonephric kidneys , therefore , are increasingly being explored as model systems for human kidney disease ( Drummond , 2005; Ebarasi et al . , 2011 ) . Compounding this problem is the fact that it has not been possible to bring the full power of invertebrate models in solving fundamental cell biological processes to the analysis of human kidney disease ( Igarashi , 2005; Dow and Romero , 2010 ) . Both Caenorhabditis elegans and Drosophila melanogaster have highly derived excretory organs in which ultrafiltration is either entirely lacking ( C . elegans; [Buechner , 2002] ) or uncoupled from reabsorption/secretion ( D . melanogaster; [Dow and Romero , 2010] ) . Furthermore , the excretory cells of both organisms are lacking cilia as a further requirement for modeling CKDs . However , C . elegans or Drosophila are but two of the myriad invertebrate species and multiple studies have documented the existence of more complex excretory systems outside the Ecdysozoa ( Ruppert and Smith , 1988 ) . One such example is the excretory system of planarian flatworms . We and others have previously reported on intriguing similarities between planarian protonephridia and the vertebrate nephron ( Rink et al . , 2011; Scimone et al . , 2011 ) . Here , we carried out a systematic structure function comparison to systematically gauge the potential of planarian protonephridia as a model system for human kidney diseases . Our results demonstrate the structural coupling of cilia-driven ultrafiltration and filtrate modification in planarian protonephridia , as well as extensive topological homology of solute carrier expression domains with the vertebrate nephron . These structure/function homologies extend to common pathologies , including shared requirements of nephrin in the maintenance of the ultrafiltration barrier , and of nephrocystins in preventing the development of tubular cysts . Our results therefore establish planarian protonephridia as a novel and viable invertebrate model for studying human kidney development and diseases .
The planarian excretory system consists of branched epithelial tubules ( protonephridia ) distributed throughout the entire body plan ( Figure 1A ) ( Rink et al . , 2011 ) . The barrel-shaped flame cells capping the proximal tubule ( PT ) ends have been proposed to act as unicellular ultrafiltration devices solely on the basis of morphological evidence ( Figure 1B ) ( Wilhelmi , 1906; Wilson and Webster , 1974 ) . To functionally test this premise , we adapted an assay previously used to demonstrate the ultrafiltration capacity of Drosophila nephrocytes ( Weavers et al . , 2009; Zhuang et al . , 2009 ) . We co-injected two inert and differentially labeled tracer molecules of different sizes into the anterior planarian mesenchyme ( 10 kDa and 500 kDa molecular weight dextrans ) . Already at 2-hr post injection , we found robust tracer accumulation in protonephridia throughout the body , confirming their active role in extracellular fluid processing . Interestingly , only the small molecular weight tracer produced intense and continuous protonephridial labeling , whereas the large dextran displayed weak and patchy labeling ( Figure 1C , D ) . Because the two tracer molecules in the injection mix carried equal numbers of fluorophores , the preferential accumulation of the small over the large dextran particles indicates molecular size filtration upon entry into the protonephridial system . We conclude from these experiments that planarian protonephridia , like vertebrate nephrons , combine ultrafiltration with filtrate modification in the same structure . 10 . 7554/eLife . 07405 . 003Figure 1 . Protonephridia are ultrafiltration devices in planarians . ( A ) Whole-mount acetylated tubulin ( AcTub ) staining . Scale bars: 500 μm . Inset shows depth-coded projection of AcTub staining . Superficial structures are in blue and deeper structures are in red . Scale bars: 50 μm . ( B ) Cross-section through a flame cell . Inset shows a high magnification of filtration diaphragm . Scale bar: 1 μm . ( C , D ) Ultrafiltration assay assessing ultrafiltration capacity in the planarian protonephridia . ( C ) Fluorescent overlay showing dextran uptake in the animals that co-injected with 10 kDa and 500 kDa fluorescently labeled dextran . Inset showing a high magnification of tubule structure labeled by dextran . Scale bar: 100 μm . ( D ) Quantification of small and large dextran uptake . DOI: http://dx . doi . org/10 . 7554/eLife . 07405 . 003 We next sought to investigate the filtrate modification capacities of the planarian protonephridial system . In the vertebrate nephron , the expression of a large number and diverse types of solute carrier ( slc ) transporters recovers essential molecules from the primary filtrate or secretes waste products into the tubule lumen ( Landowski , 2008; Raciti et al . , 2008 ) . The known substrate specificity of slc families together with their restricted expression in specific nephron segments establishes a structure/function topology of filtrate modification processes along the nephron . Towards the dual goal of identifying and mapping solute modification processes in planarian protonephridia , we set out to identify , clone and expression-map all solute carriers in the planarian genome . A systematic sequence homology search of the planarian Schmidtea mediterranea genome identified 318 slc genes . Reciprocal BLAST analysis and sequence alignments revealed that S . mediterranea slcs represent 43 slc families ( Figure 2—figure supplement 1 , Figure 2—figure supplement 2 , Figure 2—figure supplement 3 , Figure 2—figure supplement 4 , Figure 2—figure supplement 5 , Figure 2—figure supplement 6 , Figure 2—figure supplement 7 , Supplementary file 1 ) . Expression patterns of all slc genes were analyzed by in situ hybridization in intact asexual planarians . We obtained expression patterns for 287 genes in various tissues ( Figure 2—figure supplement 8 , Figure 2—figure supplement 9 , Figure 2—figure supplement 10 , Figure 2—figure supplement 11 , Figure 2—figure supplement 12 , Figure 2—figure supplement 13 ) , with 49 of these displaying putative protonephridial expressions . The expression of such a large fraction of slc genes in protonephridial tubules already indicated a rich potential for solute modifications . In order to establish a comprehensive structure-function map of protonephridia , we next mapped the expression domain of each protonephridial slc relative to two previously characterized markers ( Figure 2A , top; Supplementary file 2 ) : ( 1 ) acetylated tubulin ( AcTub ) antibody staining , which marks flame cells and the adjoining PT segment; and ( 2 ) Smed-CAVII-1 , which is expressed in the adjacent distal tubule ( DT ) segment ( Rink et al . , 2011 ) . Markers for the domain distal to CAVII-1 expression were not available at the beginning of this study . Fluorescent in situ hybridization ( FISH ) mapping of putative protonephridial slc genes against the two markers and general tubule anatomy ( e . g . , branched vs coiled PT segments ) revealed a significantly greater complexity of protonephridial cell types than previously appreciated ( Figure 2A; Figure 2—figure supplement 14 , Figure 2—figure supplement 15 , Figure 2—figure supplement 16 , Figure 2—figure supplement 17 , Figure 2—figure supplement 18 , Figure 2—figure supplement 19 , Figure 2—figure supplement 20 ) . slc expression domains define at least three sub-domains within the PT ( PT1 , PT2 , and PT3; Figure 2A–D ) and the non-overlapping expression of representative slc genes in 3-color FISH experiments demonstrates the significance of the inferred PT subdivisions ( Figure 2—figure supplement 14 , Figure 2—figure supplement 15 , Figure 2—figure supplement 16 , Figure 2—figure supplement 17 ) . Similarly , we found that slc expression domains divide the DT into 2 sub-domains ( DT1 and DT2; Figure 2A , D–F; Figure 2—figure supplement 18F ) . Interestingly , the slc12a-4 expression domain extended beyond CAVII-1 expression , where it was co-expressed with a further cohort of 14 slc genes , including Smed-slc24a-3 ( Figure 2A , G , Figure 2—figure supplement 19 , [Scimone et al . , 2011] ) . Together , these 14 slc genes therefore define the so far unknown continuation of protonephridia beyond the CAVII-1 expression domain , which for reasons detailed below we refer to as the ‘Collecting Duct’ ( CD ) . Interestingly , CD marker expressing segments were exclusively located close to the dorsal body surface , supporting early reports suggesting that the protonephridial terminus was located in the dorsal epithelium ( Wilhelmi , 1906 ) . Consistently , sagittal sections revealed occasional CD segments crossing the basement membrane and appearing to terminate in the single-layered outer epithelium ( e . g . , Smed-slc12a-1 , Figure 2H ) . To confirm this finding , we performed electron microscopy ( EM ) on serial thin sections and succeeded in visualizing multiple examples of ducts connecting into the dorsal epithelium and opening directly to the exterior ( Figure 2I , Video 1 ) . The presence of mitochondria and numerous small vesicles is ultrastructural characteristics of this region , similar to that of type B intercalated cells in the vertebrate CD . 10 . 7554/eLife . 07405 . 004Figure 2 . Unexpected complexity of protonephridial tubules is revealed by systematic gene expression mapping of slc genes along the protonephridial tubules . ( A ) Cartoon shows previous segmentation model of the protonephridial tubule and expression map of slc genes along the protonephridial tubule . ( B–G ) Representative images show expression domains of selected slc genes in ( B ) PT1 , PT2 and PT3 , ( C ) PT2 and PT3 , ( D ) PT3 , ( E ) DT1 , DT2 and CD , ( F ) DT2 and CD and ( G ) CD . Fluorescent overlay of the indicated gene ( red ) with PT marker ( AcTub ) and distal tubule ( DT ) marker ( CAVII-1 ) . A color-coded scheme of the protonephridial tubule at the end of each panel represents the expression domain of the indicated gene . Images are maximum projections of confocal Z-sections . Scale bars: 50 μm . ( H ) Longitudinal-section through a worm shows a dorsal-bias expression of slc12a-1 . Fluorescent overlay of slc12a-1 with DT marker ( CAVII-1 ) and streptavidin ( which labels the basement membrane of several planarian epithelial structures , including the outer epithelium ) . Inset shows a magnification of CD , visualized by slc12a-1 , crossing the basement membrane of the dorsal epithelia . Yellow arrowhead , exterior opening of the CD . Scale bars: 200 μm . ( I ) TEM image shows CD connected to the dorsal epithelia . Inset shows a magnification of CD connected to the dorsal epithelia . e , epithelia; bm , basement membrane; m , mesenchyme; sj , septate junction; l , lumen; ms , muscle . Scale bars: 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 07405 . 00410 . 7554/eLife . 07405 . 005Figure 2—figure supplement 1 . Solute carrier gene families in the planarian Schmidtea mediterranea . DOI: http://dx . doi . org/10 . 7554/eLife . 07405 . 00510 . 7554/eLife . 07405 . 006Figure 2—figure supplement 2 . Schematic representation of phylogenetic clusters of γ- ( A ) , δ- ( B ) groups of slcs and the Tim barrel- ( C ) , IT- ( D ) , Drug/Metabolite ( E ) transporter clans of slcs . Planarian homologs are colored in red . DOI: http://dx . doi . org/10 . 7554/eLife . 07405 . 00610 . 7554/eLife . 07405 . 007Figure 2—figure supplement 3 . Schematic representation of phylogenetic clusters of α-groups of slcs . Planarian homologs are colored in red . DOI: http://dx . doi . org/10 . 7554/eLife . 07405 . 00710 . 7554/eLife . 07405 . 008Figure 2—figure supplement 4 . Schematic representation of phylogenetic clusters of β-groups of slcs . Planarian homologs are colored in red . DOI: http://dx . doi . org/10 . 7554/eLife . 07405 . 00810 . 7554/eLife . 07405 . 009Figure 2—figure supplement 5 . Schematic representation of phylogenetic clusters of Smed-slc1a ( A ) , Smed-slc5a ( B ) , Smed-slc22a ( C ) , Smed-slc6a ( D ) . Planarian homologs are colored in red . DOI: http://dx . doi . org/10 . 7554/eLife . 07405 . 00910 . 7554/eLife . 07405 . 010Figure 2—figure supplement 6 . Schematic representation of phylogenetic clusters of Smed-slc4a ( A ) , Smed-slc7a ( B ) , Smed-slc12 ( C ) , Smed-slc15 ( D ) , Smed-slc20 ( E ) , Smed-slc23 ( F ) , Smed-slc26 ( G ) . Planarian homologs are colored in red . DOI: http://dx . doi . org/10 . 7554/eLife . 07405 . 01010 . 7554/eLife . 07405 . 011Figure 2—figure supplement 7 . Schematic representation of phylogenetic clusters of Smed-slc28 ( A ) , Smed-slc30 ( B ) , and Smed-slc42 ( C ) . Planarian homologs are colored in red . DOI: http://dx . doi . org/10 . 7554/eLife . 07405 . 01110 . 7554/eLife . 07405 . 012Figure 2—figure supplement 8 . Expression patterns of slc genes that belong to solute carrier families 1–6 in an asexual strain of the planarian S . mediterranea . Whole-mount expression patterns of slc genes by in situ hybridization ( NBT/BCIP development ) . Scale bars: 500 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 07405 . 01210 . 7554/eLife . 07405 . 013Figure 2—figure supplement 9 . Expression patterns of slc genes that belong to solute carrier families 7–15 in an asexual strain of the planarian S . mediterranea . Whole-mount expression patterns of slc genes by in situ hybridization ( NBT/BCIP development ) . Scale bars: 500 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 07405 . 01310 . 7554/eLife . 07405 . 014Figure 2—figure supplement 10 . Expression patterns of slc genes that belong to solute carrier families 16–23 in an asexual strain of the planarian S . mediterranea . Whole-mount expression patterns of slc genes by in situ hybridization ( NBT/BCIP development ) . Scale bars: 500 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 07405 . 01410 . 7554/eLife . 07405 . 015Figure 2—figure supplement 11 . Expression patterns of slc genes that belong to solute carrier families 24–29 in an asexual strain of the planarian S . mediterranea . Whole-mount expression patterns of slc genes by in situ hybridization ( NBT/BCIP development ) . Scale bars: 500 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 07405 . 01510 . 7554/eLife . 07405 . 016Figure 2—figure supplement 12 . Expression patterns of slc genes that belong to solute carrier families 30–38 in an asexual strain of the planarian S . mediterranea . Whole-mount expression patterns of slc genes by in situ hybridization ( NBT/BCIP development ) . Scale bars: 500 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 07405 . 01610 . 7554/eLife . 07405 . 017Figure 2—figure supplement 13 . Expression patterns of slc genes that belong to solute carrier families 40–47 in an asexual strain of the planarian S . mediterranea . Whole-mount expression patterns of slc genes by in situ hybridization ( NBT/BCIP development ) . Scale bars: 500 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 07405 . 01710 . 7554/eLife . 07405 . 018Figure 2—figure supplement 14 . Expression of slc genes in the PT . Fluorescent overlay of indicated gene ( in red ) ( A-F ) with PT2 and PT3 marker ( slc6a-13 ) , DT marker ( CAVII-1 ) and AcTub staining . Images are maximum projections of confocal Z-sections . Scale bars: 50 μm . A color-coded scheme of protonephridial tubule at the end of each panel represents the expression domain of the indicated gene . DOI: http://dx . doi . org/10 . 7554/eLife . 07405 . 01810 . 7554/eLife . 07405 . 019Figure 2—figure supplement 15 . Expression of slc genes in the PT1 segment of the PT . ( A ) Fluorescent overlay of indicated gene ( in red ) with PT1 and PT2 marker ( CUBN1 ) , PT2 and PT3 marker ( slc6a-13 ) and AcTub staining . ( B–D ) Fluorescent overlay of indicated gene ( in red ) with PT2 and PT3 marker ( slc6a-13 ) , DT marker ( CAVII-1 ) and AcTub staining . Images are maximum projections of confocal Z-sections . Scale bars: 50 μm . A color-coded scheme of protonephridial tubule at the end of each panel represents the expression domain of the indicated gene . DOI: http://dx . doi . org/10 . 7554/eLife . 07405 . 01910 . 7554/eLife . 07405 . 020Figure 2—figure supplement 16 . Expression of slc genes in the PT2 and PT3 segments of the PT . Fluorescent overlay of indicated gene ( in red ) with PT2 and PT3 marker ( slc6a-13 ) , DT marker ( CAVII-1 ) and AcTub staining . Images are maximum projections of confocal Z-sections . Scale bars: 50 μm . A color-coded scheme of protonephridial tubule at the end of each panel represents the expression domain of the indicated gene . DOI: http://dx . doi . org/10 . 7554/eLife . 07405 . 02010 . 7554/eLife . 07405 . 021Figure 2—figure supplement 17 . Expression of slc genes in PT3 segment of the PT . Fluorescent overlay of indicated gene ( in red ) ( A-C ) with PT2 and PT3 marker ( slc6a-13 ) , DT marker ( CAVII-1 ) and AcTub staining . Images are maximum projections of confocal Z-sections . Scale bars: 50 μm . A color-coded scheme of protonephridial tubule at the end of each panel represents the expression domain of the indicated gene . DOI: http://dx . doi . org/10 . 7554/eLife . 07405 . 02110 . 7554/eLife . 07405 . 022Figure 2—figure supplement 18 . Expression of slc genes in the DT . Fluorescent overlay of indicated gene ( in red ) ( A-G ) with PT marker ( slc6a-13 or CUBN1 ) , DT marker ( CAVII-1 ) or CD marker ( slc12a-1 or slc24a-9 ) together with AcTub staining . Images are maximum projections of confocal Z-sections . Scale bars: 50 μm . A color-coded scheme of protonephridial tubule at the end of each panel represents the expression domain of the indicated gene . DOI: http://dx . doi . org/10 . 7554/eLife . 07405 . 02210 . 7554/eLife . 07405 . 023Figure 2—figure supplement 19 . Expression of slc genes in the collecting duct . Fluorescent overlay of indicated gene ( in red ) ( A-P ) with PT2 and PT3 marker ( slc6a-13 ) , DT marker ( CAVII-1 ) or CD marker ( slc24a-9 ) together with AcTub staining . Images are maximum projections of confocal Z-sections . Scale bars: 50 μm . A color-coded scheme of protonephridial tubule at the end of each panel represents the expression domain of the indicated gene . DOI: http://dx . doi . org/10 . 7554/eLife . 07405 . 02310 . 7554/eLife . 07405 . 024Figure 2—figure supplement 20 . Expression of slc genes weakly expressed in both proximal and DTs . Fluorescent overlay of indicated gene ( in red ) ( A-B ) with PT2 and PT3 marker ( slc6a-13 ) , DT marker ( CAVII-1 ) and AcTub staining . Images are maximum projections of confocal Z-sections . Scale bars: 50 μm . A color-coded scheme of protonephridial tubule at the end of each panel represents the expression domain of the indicated gene . DOI: http://dx . doi . org/10 . 7554/eLife . 07405 . 02410 . 7554/eLife . 07405 . 025Video 1 . Protonephridial collecting duct opens to the dorsal epithelia . Serial TEM images showing the protonephridial collecting duct connected to the dorsal epithelia . DOI: http://dx . doi . org/10 . 7554/eLife . 07405 . 025 For the first time , our results trace the complete course of protonephridial tubules from the ultrafiltrating flame cells as proximal entry point to their terminus in the dorsal epithelium . Further , our systematic mapping of expression domains of slc genes defined 6 molecularly distinct segments along the proximal-distal axis of protonephridia . We next took advantage of our expression data and the known transport activities of slc families in vertebrates to infer possible functional specializations of these 6 protonephridial segments . Clustering a subset of slc genes with known substrate specificity by substrate class and site of expression revealed a striking segregation of similar transport activities into similar regions of the protonephridial tubule , indicating the functional specialization of different segments ( Figure 3A , top ) . Because this subset of slc genes was intentionally chosen due to its known representation for transport activities of specific segments of the nephron ( Raciti et al . , 2008 ) , this map afforded a basis for direct structure/function comparisons with the vertebrate nephron . Constructing a similar map of slc expression in the rodent metanephros based on published data ( Figure 3A , bottom; Supplementary file 3 ) revealed striking parallel: Not only is the sequence of slc family expression very similar along the filtrate flow axis , but almost all nephron segments have clearly identifiable homologous segments in protonephridia . In vertebrates , the PT is responsible for the reabsorption of more than 70% of filtered solutes from the primary urine , including inorganic/organic ions and vital nutrients ( glucose , amino acids , and vitamins ) . The homologous slc expression of planarian PT1-3 and the preferential labeling of PT1-2 by injected dextran ( Figure 3B ) provide strong evidence that the proximal protonephridial segments are likewise primarily responsible for the recovery of filtered substances . The DT plays an important role in acid-base homeostasis by reabsorbing bicarbonates and secreting protons into the urine ( Carraro-Lacroix and Malnic , 2010 ) . The corresponding expression of bicarbonate ( e . g . , Smed-slc4a-6 , Figure 2—figure supplement 18A ) or proton transporters ( e . g . , Na+/H+ exchanger Smed-slc9a-3 , Figure 2E ) in DT1 and DT2 suggests a similar function of these protonephridial segments . Consistently , the RNAi knockdown of slc4a-6 caused a measurable acidification of the intercellular milieu ( Figure 3C ) , supporting a functionally conserved role of DT1-2 in planarian pH homeostasis . Moreover , the vertebrate CD comprises distinct cortical and medullary segments and mediates the bulk of water recovery/urine concentration ( Nielsen et al . , 2002 ) . The shared expression of the bicarbonate transporter Smed-slc4a-7 and the ammonia transporter Smed-slc42a-2 in the terminal segment ( Figure 3A ) supports a basal homology between the CD and the corresponding protonephridial segment , which is why we have chosen to adopt the vertebrate nomenclature . However , the large number of additional slc genes expressed in the protonephridial CD ( Figure 2A ) and the lack of aquaporin expression ( not shown ) suggest divergent functions . 10 . 7554/eLife . 07405 . 026Figure 3 . Extensive structural and functional homology between planarian protonephridia and vertebrate nephrons . ( A ) Tables summarize expression domains of selected slc genes in planarian protonephridia and rodent metanephros . Cartoons showing segmental organization of planarian protonephridia and rodent metanephros are on the left . Gray color in the tables indicates expression domain of slc in planarian protonephridia and rodent metanephros . Planarian slc sequence nomenclature ( e . g . , slc1a-3 ) doesn't reflect direct orthology to the mammalian counterparts . Abbreviations for segments of protonephridia are as follows: PT1 , PT2 , and PT3 , segments of the proximal tubule ( PT ) ; DT1 and DT2 , segments of the DT; CD , the collecting duct . Abbreviations for segments of the metanephros are as follows: S1 , S2 , and S3 , segments of the PT; DTL , descending thin limb; ATL , ascending thin limb; TAL , thick ascending limb; DCT , distal convoluted tubules; CNT , connecting tubule; CD , collecting duct . ( B ) Fluorescent overlay of reabsorbed dextran with PT marker ( AcTub ) . ( C ) pHi reporter assay using SNARF-5F-AM in Control ( RNAi ) and slc4a-6 ( RNAi ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07405 . 026 The only nephron segment for which our analysis did not identify a protonephridial homologue was the intermediate tubule ( IT ) . In terrestrial vertebrates , IT and CD have tightly linked functions in water conservation , whereby urea secretion by the IT establishes high extracellular solute concentrations that aid in water reabsorption from the CD ( Pannabecker , 2012 ) . As freshwater animals , planarian protonephridia have to clear , rather than conserve water , providing a compelling rationale for why specifically IT and CD are divergent . Such functional diversity of IT/CD segments is also observed in the pronephric kidneys of freshwater vertebrates , such as zebrafish ( Wingert and Davidson , 2008 ) . Altogether , our analyses reveal a striking structural and functional homology between the vertebrate nephron and the planarian protonephridium . We next asked whether the homologies between nephrons and protonephridia extend to common pathologies . The striking structural similarities between the ultrafiltration sites in the two systems , podocyte foot processes ( Pavenstadt et al . , 2003 ) and flame cell filtration barriers ( Figure 1B ) could reflect a requirement for common components . In humans , mutations in the large IgG-repeat transmembrane proteins NPHS1 and NEPH1 cause slit diaphragm loss by fusion of neighboring foot processes into a continuous cytoplasmic sheet ( foot process effacement ) , resulting in proteinuria and edema ( Kestila et al . , 1998; Donoviel et al . , 2001 ) . Systematic sequence homology searches of the S . mediterranea genome identified 7 NPHS1 homologs and 3 NEPH homologs ( Figure 4—figure supplement 1 ) . Interestingly , Smed-NPHS1-6 and Smed-NEPH-3 were expressed in flame cells ( Figure 4A , B ) and RNAi of both genes produced strong bloating and partial clearing of body pigmentation ( Figure 3C ) . Both phenotypes have previously been identified as characteristic hallmarks of tissue edema ( Rink et al . , 2011 ) , thus providing a strong indication that the genes are required for the function of the planarian excretory system . 10 . 7554/eLife . 07405 . 027Figure 4 . Vertebrate slit-diaphragm components are expressed in planarian flame cells and are required for the maintenance of their filtration diaphragm . ( A ) Whole-mount expression patterns of indicated marker genes by in situ hybridization ( NBT/BCIP development ) . Scale bars: 500 μm . ( B ) Fluorescent overlay of indicated gene ( red ) with flame cell marker EGFR-5 and AcTub staining . Images are maximum projections of confocal Z-sections . Scale bars: 50 μm . ( C ) Live images show edema in intact NPHS1-6 ( RNAi ) and NEPH-3 ( RNAi ) animals . Scale bars: 500 μm . ( D ) TEM images show cross-section through a flame cell in intact Control ( RNAi ) , NPHS1-6 ( RNAi ) and NEPH-3 ( RNAi ) animals . Inset shows a high magnification of the filtration diaphragm . Scale bar: 1 μm . ( E , F ) Ultrafiltration assay assesses the ultrafiltration capacity in NPHS1-6 ( RNAi ) animals . ( E ) Representative images show dextran uptake in the animals that co-injected with 10 kDa and 500 kDa fluorescently labeled dextran . Scale bar: 50 μm . ( F ) Quantification of small and large dextran uptake . DOI: http://dx . doi . org/10 . 7554/eLife . 07405 . 02710 . 7554/eLife . 07405 . 028Figure 4—figure supplement 1 . Slit-diaphragm components in the planarian S . mediterranea . ( A ) Cartoon showing the glomerular filtration barrier . Top: A schematic view of the podocyte . The podocyte wraps around the capillary wall on the outer surface of the glomerular basement membrane with its extended inter-digitating foot processes . Podocyte foot processes are then bridged by a slit diaphragm . Middle: A close-up view of the glomerular filtration barrier consisting of three components: porous endothelium , glomerular basement membrane , and podocyte foot processes with the interposed slit diaphragm . The endothelial pores are not bridged by a diaphragm . Bottom: Schematic drawing of the molecular equipment of slit diaphragm . NPHS1 undergoes homophilic interaction on neighboring podocyte foot processes . The intercellular junction also contains the adhesion molecule NEPH-1 . ( B ) Homology analysis of the planarian homologs of NPHS1 and NEPH . Domains predicted by SMART for planarian and human proteins . Best reciprocal BLAST hits in human , C . elegans , and fly refseq protein database . ( C ) Whole-mount expression patterns of NPHS1 and ( D ) NEPH by in situ hybridization . Scale bars: 500 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 07405 . 02810 . 7554/eLife . 07405 . 029Figure 4—figure supplement 2 . NPHS1-6 is not required for flame cell viability during normal homeostasis , as well as regeneration . ( A , B ) Fluorescent overlay of flame cell markers ( CXCRL and EGFR-5 ) with AcTub staining in intact ( A ) and regenerating ( B ) Control ( RNAi ) and NPHS1-6 ( RNAi ) . Images are maximum projections of confocal Z-sections . Scale bars: 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 07405 . 02910 . 7554/eLife . 07405 . 030Figure 4—figure supplement 3 . NPHS1-6 is required for de novo formation of filtration diaphragm during regeneneration . TEM images showing cross-section through a flame cell in regenerating Control ( RNAi ) and NPHS1-6 ( RNAi ) animals . Inset shows high magnification of filtration diaphragm . Scale bar: 1 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 07405 . 030 Because flame cell numbers appeared normal in both intact and regenerating animals ( Figure 4—figure supplement 2 ) , we examined the ultrastructure of the filtration diaphragm in NPHS1-6 and NEPH-3 ( RNAi ) planarians . Wild type flame cells display slit-shaped 35–40 nm wide fenestrae that form between 90–150 nm wide foot processes ( Figures 1B , 4D ) . Under RNAi knockdown of either NPHS1-6 or NEPH-3 , the filtration diaphragm was completely absent and the foot processes underwent apparent effacement in both intact ( Figure 4D , Videos 2 , 3 ) and regenerating animals ( Figure 4—figure supplement 3 ) . The dextran injection assay confirmed the loss of ultrafiltration capability in NPHS1-6 ( RNAi ) planarians , which displayed equal uptake of both small and large molecular tracer in the PT ( Figure 4E , F ) . Why would the fusion of foot processes into a continuous sheet result in loss of filtration size selectively , rather than a general block of filtration ? In human nephrotic syndrome patients , the loss of ultrafiltration capability in thought to occur as a consequence of podocyte detachment or apoptosis and subsequent filtrate leakage ( Tojo and Kinugasa , 2012 ) . However , many nephrotic syndrome patients present ultrafiltration deficiencies without podocyte detachment or loss ( Furness et al . , 1999; Lahdenkari et al . , 2004 ) . Consistently , in NPHS1-6- and NEPH-3 ( RNAi ) planarians , we could not observe ultrastructural evidence of flame cell detachment at the time that loss of ultrafiltration was observed ( not shown ) . Evidence from a rat model of nephrotic syndrome indicates that the upregulation of transcytotic transport processes across the effaced podocyte envelope could maintain a basal level of non-size-selective fluid flow ( Tojo et al . , 2008; Kinugasa et al . , 2011; Tojo and Kinugasa , 2012 ) , yet the exact mechanisms remain to be determined in both humans and planarians . Regardless , the striking parallels between the NPHS1-6- and NEPH-3 ( RNAi ) phenotypes in planarians and human nephrotic syndrome demonstrate that the functional homology between planarian flame cells and vertebrate podocytes extends to common pathologies . 10 . 7554/eLife . 07405 . 031Video 2 . Flame cell morphology in Control ( RNAi ) animal . Serial TEM images showing flame cell in Control ( RNAi ) animal . DOI: http://dx . doi . org/10 . 7554/eLife . 07405 . 03110 . 7554/eLife . 07405 . 032Video 3 . Flame cell morphology in NPHS1-6 ( RNAi ) animal . Serial TEM images showing flame cell in NPHS1-6 ( RNAi ) animal . DOI: http://dx . doi . org/10 . 7554/eLife . 07405 . 032 Encouraged by these results , we expanded our search for conserved pathologies to the protonephridial tubules . The most common class of human inherited disorders affecting the nephron are the CKDs . We assembled a small library of putative planarian orthologues of human CKD genes ( Supplementary file 4 ) . This list included the nephrocystins , causative genes of nephronophthisis ( NPHP ) , one of the most frequent genetic causes of chronic renal failure in children and young adults ( Hildebrandt and Otto , 2000; Salomon et al . , 2009 ) . The S . mediterranea genome harbors homologs to human NPHP1-9 , except for NPHP2 and NPHP3 ( Figure 5—figure supplement 1 ) . RNAi-screening of the library revealed strong edema formation in Smed-NPHP5 , Smed-NPHP6 , and Smed-NPHP8 knockdown animals ( Figure 4A ) , suggesting a protonephridial function for these genes . Consistently , we detected severe structural alterations of protonephridial tubules in NPHP ( RNAi ) animals , particularly of the proximal segment . RNAi animals presented striking clump-like accumulations of PT marker expressing cells ( Figure 5A , B , Videos 4 , 5 ) instead of the fine terminal ramifications of PTs in controls . High-resolution imaging confirmed the presence of abnormally high numbers of densely packed PT cells ( Figure 5—figure supplement 2 ) . The protonephridial lumen was severely disorganized within such aggregates ( Figure 5C ) . Instead of strong and continuous luminal labeling throughout the coiled PT segments of controls , labeling was weak and fragmented . The weak single-line labeling outside of aggregates ( Figure 5C ) and the much weaker cilia staining ( AcTub ) in NPHP ( RNAi ) animals ( Figure 5A ) suggested general lumen defects . EM images revealed frequent basal body mislocalizations to non-luminal membrane domains and cell intrusions into the lumen , which both indicate a loss of normal tubular cell polarity ( Figure 5—figure supplement 3 ) . Altogether , the accumulation of morphologically abnormal tubule cells and concomitant loss of luminal connectivity present striking morphological parallels to the NPHP loss-of-function phenotype in humans , suggesting that planarian protonephridia can develop cyst-like structures . 10 . 7554/eLife . 07405 . 033Figure 5 . Down-regulation of nephrocystin members leads to the formation of cyst-like structure in protonephridia . ( A ) Protonephridial defects in NPHP5 ( RNAi ) , NPHP6 ( RNAi ) and NPHP8 ( RNAi ) animals . Top panel: live images show edema in intact RNAi animals . Scale bars: 500 μm; middle panel: monochrome showing AcTub staining; bottom panel: fluorescent overlay of AcTub staining with PT2 and PT3 marker ( slc6a-13 ) and DT marker ( slc6a-12 ) . Scale bars: 50 μm . ( B ) 3D rendering images showing normal tubule and cystic-like tubule in Control ( RNAi ) and NPHP8 ( RNAi ) animals , respectively . 3D rendering was performed in IMARIS . Scale bars: 50 μm . ( C ) Fragmented lumen in enlarged protonephridial tubule . Fluorescent overlay of PT2 and PT3 marker slc6a-13 and lumen marker ( a customized rabbit antiserum recognized unknown epitope ) in intact Control ( RNAi ) and NPHP8 ( RNAi ) animals . Scale bars: 50 μm . Images in ( A ) and ( C ) are maximum projections of confocal Z-sections . DOI: http://dx . doi . org/10 . 7554/eLife . 07405 . 03310 . 7554/eLife . 07405 . 034Figure 5—figure supplement 1 . Nephrocystins in the planarian S . mediterranea . ( A ) Homology analysis of planarian nephrocystins . Domains predicted by SMART for planarian and human proteins . Best reciprocal BLAST hits in human , C . elegans , and fly refseq protein database . ( B ) Whole-mount expression patterns of genes encoding nephrocystins by in situ hybridization . Scale bars: 500 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 07405 . 03410 . 7554/eLife . 07405 . 035Figure 5—figure supplement 2 . Abnormal tubular enlargement in NPHP8 ( RNAi ) animals . ( A ) Fluorescent overlay of lumen marker with PT2 and PT3 marker slc6a-13 and nuclei ( DAPI ) in Control ( RNAi ) and NPHP8 ( RNAi ) animals . Scale bars: 25 μm . ( A ) Fluorescent overlay of PT marker ( slc6a-13 ) and ( B ) DT marker ( CAVII-1 ) in intact Control ( RNAi ) and NPHP8 ( RNAi ) animals . Images are maximum projections of confocal Z-sections . Scale bars: 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 07405 . 03510 . 7554/eLife . 07405 . 036Figure 5—figure supplement 3 . Ultrastructure of the PT in NPHP ( RNAi ) animals . TEM images showing cross-section through a tubule of protonephridia in indicated RNAi animals . Inset in red box shows abnormally localized basal body in indicated RNAi animals . Inset in green box shows ultrastructure of cilia . c , cilia; n , nucleus; bb , basal body; sj , septate junction; l , lumen . DOI: http://dx . doi . org/10 . 7554/eLife . 07405 . 03610 . 7554/eLife . 07405 . 037Video 4 . 3D rendering of normal protonephridial tubule in Control ( RNAi ) animal . DOI: http://dx . doi . org/10 . 7554/eLife . 07405 . 03710 . 7554/eLife . 07405 . 038Video 5 . 3D rendering of cystic-like protonephridial tubule in NPHP8 ( RNAi ) animal . DOI: http://dx . doi . org/10 . 7554/eLife . 07405 . 038 Sustained cell proliferation in the renal tubules is a hallmark of cystic kidneys in humans and the severity of the phenotype correlates with the ectopic proliferation level ( Wilson and Goilav , 2007 ) . We used BrdU pulse labeling to determine whether cell proliferation was involved in the formation of the observed cyst-like structures ( Figure 6A ) . In controls , we found occasional cells double positive for BrdU and the protonephridial progenitor marker Smed-POU2/3 ( Scimone et al . , 2011 ) in the vicinity of tubules ( Figure 6A ) , consistent with the emerging view that all planarian cell types derive from the proliferation of specific progenitor classes within the neoblast population ( Cowles et al . , 2013; Adler et al . , 2014; Scimone et al . , 2014; van Wolfswinkel et al . , 2014 ) . In NPHP8 ( RNAi ) animals , the number of BrdU/POU2/3 double-positive cells in the vicinity of cell accumulations was notably increased ( Figure 6A ) , and in situ analyses further confirmed the progressive accumulation of protonephridial progenitors ( Figure 6—figure supplement 1A–D ) . To probe the magnitude of the overproliferation effect , we carried out whole-mount staining with the G2/M-phase marker phospho-Histone H3 ( H3P ) and found a global increase in cell proliferation in NPHP ( RNAi ) animals ( Figure 6B ) . To ask whether these effects were specific to protonephridial progenitors or globally affected all progenitor classes , we quantified the relative fraction of proliferation in protonephridial- ( POU2/3+/smedwi-1+/H3P+ ) , neuronal- ( pax6A+/smedwi-1+/H3P+ ) ( Wenemoser et al . , 2012; Scimone et al . , 2014 ) and intestinal ( HNF4+/smedwi-1+/H3P+ ) ( Wagner et al . , 2011; Scimone et al . , 2014 ) progenitor classes ( Figure 6—figure supplement 1C , D , Figure 6—figure supplement 2 ) . Whereas the fraction of proliferating protonephridial progenitors was increased in both NPHP5 ( RNAi ) and NPHP8 ( RNAi ) animals , we found no change in the fraction of proliferating neuronal progenitors and even a slight decrease in intestinal progenitor proliferation ( Figure 6C–E ) . The observation that all cases of ectopic BrdU-incorporation in the normally division-devoid area anterior to the photoreceptors were limited to POU2/3+ protonephridial progenitors ( Figure 6—figure supplement 1E ) further supports the protonephridial specificity of the overproliferation response . Altogether , these results demonstrate that loss of function of planarian NPHP genes selectively increased the proliferation of protonephridial progenitors . 10 . 7554/eLife . 07405 . 039Figure 6 . Cystogenesis in planarian protonephridia results from direct proliferation of protonephridia progenitors and requires the presence of stem cells . ( A ) BrdU pulse-chase experiment shows the presence of diving protonephridial progenitors in the proximity of protonephridial tubule in Control ( RNAi ) and NPHP8 ( RNAi ) animals . Yellow arrowhead indicates POU2/3+/BrdU+ cell . Scale bars: 25 μm . ( B ) Increased global proliferation in NPHP5 ( RNAi ) and NPHP8 ( RNAi ) animals is displayed by immunostaining of mitotic marker phospho-Histone H3 ( H3P ) . Scale bars: 500 μm . * p < 0 . 05; ** p < 0 . 01; *** p < 0 . 001 vs control . The time points in the bar graph indicate the number of day after the last dsRNA introduction . ( C–E ) Quantification of ( C ) dividing protonephridial progenitors ( POU2/3+/H3P+ ) , ( D ) diving neuronal progenitors ( pax6A+/H3P+ ) and ( E ) diving gut progenitors ( HNF4+/H3P+ ) among diving cells ( H3P+ ) in indicated RNAi animals at 18 day after last RNAi introduction . * p < 0 . 05 vs control . ( F–J ) Effect of proliferation and the requirement of neoblasts on cyst formation in planarian protonephridia . ( F ) Schematics demonstrates experimental strategy for panel H–J . 7 day post RNAi feeding animals were either fed with liver to induce cell proliferation or subjected to sublethal or lethal doses of irradiation to reduce or eliminate neoblasts . Scoring live phenotype as well as measuring the average size of each protonephridial unit was used to evaluate the severity of cystic phenotype . Temporal succession of indicated phenotypes ( left ) and quantification of average area of each slc6a13+/CAVII-1+ tubule ( right ) in Control ( RNAi ) and NPHP8 ( RNAi ) animals under ( G ) basal condition ( only RNAi feeding ) , ( H ) basal condition plus extra feeding with liver , ( I ) basal condition plus sub-lethal irradiation to reduce the number of neoblasts , and ( J ) basal condition plus lethal irradiation to completely eliminate neoblasts . The time points in the bar graphs ( G–J ) indicate the number of the day after the first dsRNA introduction . DOI: http://dx . doi . org/10 . 7554/eLife . 07405 . 03910 . 7554/eLife . 07405 . 040Figure 6—figure supplement 1 . Increase of protonephridial progenitors during cystogenesis in planarian protonephridia . ( A ) Left: Whole-mount expression patterns of POU2/3 by in situ hybridization . Scale bars: 500 μm . Right: Fluorescent overlay of POU2/3 with DT marker ( CAVII-1 ) and AcTub . Scale bar: 50 μm . ( B ) Left: Whole-mount expression patterns of six1/2 by in situ hybridization . Scale bars: 500 μm . Right: Fluorescent overlay of six1/2 with DT marker ( CAVII-1 ) and AcTub . Scale bar: 50 μm . ( C , D ) Magnified view showing the region surrounding photoreceptor . Fluorescent overlay of POU2/3 and six1/2 with pan stem cell marker Smedwi-1 and mitotic marker H3P . Scale bar: 50 μm . ( E ) Increase of S-phase protonephridial progenitors during cystogenesis in planarian protonephridia . Intact Control ( RNAi ) and NPHP8 ( RNAi ) animals were pulsed with BrdU ( 1 hr ) , followed by 2 hr-chase . Fluorescent overlay of POU2/3 with BrdU showing the abnormal increase of POU2/3+/BrdU+ in the head region anterior to the photoreceptors . Images are maximum projections of confocal Z-sections . Scale bar: 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 07405 . 04010 . 7554/eLife . 07405 . 041Figure 6—figure supplement 2 . Gut and brain progenitors in NPHP8 ( RNAi ) animals . ( A , B ) Left panel: whole-mount expression patterns of pax6A ( A ) and HNF4 ( B ) by in situ hybridization . Scale bars: 500 μm; Right panel: fluorescent overlay of ( A ) pax6A and ( B ) HNF4 with pan stem cell marker ( Smedwi-1 ) and mitotic marker ( H3P ) . Scale bar: 50 μm . ( C ) Magnified view showing the head region . Fluorescent overlay of pax6A and HNF4 with pan stem cell marker Smedwi-1 and mitotic marker H3P . Scale bar: 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 07405 . 04110 . 7554/eLife . 07405 . 042Figure 6—figure supplement 3 . The severity of cystic phenotype in protonephridia depends on proliferation rate and requires the presence of stem cells . ( A ) Quantification of mitoses in Control ( RNAi ) and NPHP8 ( RNAi ) animals . Experimental paradigm is described in Figure 5F . The time point in the bar graph indicates the number of the day after the first dsRNA introduction . ( B ) The severity of cystic phenotype in protonephridia depends on the rate of proliferation and requires the presence of stem cells . Fluorescent overlay of PT marker ( slc6a-13 ) with DT marker ( CAVII-1 ) . Images are maximum projections of confocal Z-sections . Scale bar: 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 07405 . 042 To test whether the level of proliferation determined the severity of the phenotype as observed in humans , we made use of the facile manipulation of global cell proliferation levels in the planarian system ( Figure 6F , Figure 6—figure supplement 3A ) . Lethally or sub-lethally irradiated animals were used to examine the effects of abolished or reduced proliferation , respectively ( Wagner et al . , 2012 ) , while animals on an increased feeding regiment provided an opportunity to examine the effects of above-baseline proliferation ( Kang and Sanchez Alvarado , 2009 ) . We found that edema development in NPHP8 ( RNAi ) animals was faster and more severe under the increased proliferation condition , yet significantly diminished or even abolished under reduced or no proliferation , respectively ( Figure 6G–J , left ) . The quantification of the confocally projected area of protonephridial marker expression domains ( slc6a-13 and CAVII-1 ) as a direct cell accumulation metric ( Figure 6G–J , right; Figure 6—figure supplement 3B ) also showed equal dependency on proliferation rates , thus indicating that the development of planarian NPHP phenotypes is tightly associated with cell proliferation . In light of the striking morphological and ontological parallels between protonephridial and human NPHP loss-of-function phenotypes , we now refer to the observed structural alterations in planarian protonephridia as cysts . Cilia as flow sensors play a critical role in the ontogeny of human CKDs ( Hildebrandt and Otto , 2005; Hildebrandt and Zhou , 2007; Kotsis et al . , 2013 ) . NPHP ( RNAi ) planarians display severe defects in cilia-driven gliding motility ( Figure 7A , B ) , prompting us to investigate a possible involvement of cilia in the ontogeny of planarian protonephridial defects . Direct visualization of axonemes in NPHP ( RNAi ) animals indeed confirmed structural cilia defects , which appeared shorter ( NPHP5 ( RNAi ) ) or much reduced in density ( NPHP6/8 ( RNAi ) ) ( Figure 7C ) . EM images revealed abnormal localization of centrioles as well as axonemal abnormalities in ciliated cells under NPHP5/6/8 ( RNAi ) ( Figure 5—figure supplement 3 ) . Together with the broad resemblance between NPHP5/6/8 expression patterns and typical cilia genes ( Rink et al . , 2009; Glazer et al . , 2010 ) ( Figure 5—figure supplement 1 ) , these data conclusively demonstrate that knockdown of planarian NPHP-genes causes not only protonephridial cyst formation , but also structural defects in cilia . 10 . 7554/eLife . 07405 . 043Figure 7 . Cystic phenotype in protonephridia is cilia-and fluid flow-dependent . ( A ) Series of live images show gliding mobility in Control ( RNAi ) and NPHP8 ( RNAi ) animals . Yellow dot line provides a spatial reference to illustrate progress of animal . Scale bar: 1 mm . ( B ) Quantification of translocation speed in indicated RNAi animals . Error bar , SD; *** p < 0 . 001 vs control . ( C ) Fluorescent overlay of ventral cilia ( AcTub ) with nucleus marker ( DAPI ) in indicated RNAi animals . Scale bar: 10 μm . ( D–G ) Live images show bloating phenotype in IFT88 ( RNAi ) , DNAHβ-1 ( RNAi ) , and LRRC50 ( RNAi ) animals . Scale bar: 500 μm . ( D′–G′ ) Fluorescent overlay of ventral cilia ( AcTub ) with nucleus marker ( DAPI ) in IFT88 ( RNAi ) , DNAHβ-1 ( RNAi ) , and LRRC50 ( RNAi ) animals . Scale bar: 10 μm . ( D′′–G′′ ) 3D rendering showing fluorescent overlay of AcTub staining with PT2 and PT3 marker ( slc6a-13 ) and DT marker ( CAVII-1 ) in Control ( RNAi ) , IFT88 ( RNAi ) , DNAHβ-1 ( RNAi ) , and LRRC50 ( RNAi ) animals . Scale bar: 50 μm . ( D′′′–G′′′ ) Magnified view shows fluorescent overlay of POU2/3 with pan stem cell marker ( smedwi-1 ) in the region surrounding photoreceptor . White arrowhead shows POU2/3+/smedwi-1+ cell . Scale bar: 25 μm . ( H–I ) Abnormal cilia beating in DNAHβ-1 ( RNAi ) , and LRRC50 ( RNAi ) animals . ( H ) Left panel: live images show cilia beating along the lateral body edge of the planarian head region; Right panel: vector map generated by Spatiotemporal image correlation spectroscopy ( STICS ) analysis shows velocity magnitude and beating pattern of cilia . The brightness of the vector represents the velocity magnitude of the cilia: brighter vector , stronger ciliary beating or vice versa . ( I ) Quantification of ciliary velocity magnitude in indicated RNAi animals . * p < 0 . 05 vs control . ( J ) Cartoon represents working model of cyst formation in the planarian protonephridia . In normal tubule , protonephridial tubular cell turnover is maintained by integration of protonephridial progenitors , originated from the neoblasts , into the tubule . During this process , cilia-driven fluid flow is required for the maintenance of tubular geometry . Obstruction of fluid flow by disrupting cilia function leads to protonephridial cystogenesis that characterized by abnormal proliferation of protonephridial progenitors , tubular enlargement and disorganization . DOI: http://dx . doi . org/10 . 7554/eLife . 07405 . 04310 . 7554/eLife . 07405 . 044Figure 7—figure supplement 1 . Primary Ciliary Dyskinesia ( PCD ) genes in the planarian S . mediterranea . ( A ) Quantification of translocation speed in indicated RNAi animals . Error bar , SD; *** p < 0 . 001 vs control . ( B ) Schematic drawing shows the structure of 9 + 2 motile cilia in planarians . Right panel: schematic representation of the expanded view of the ODA depicts several light , intermediate , and heavy chains . The planarian homologs of human PCD genes with the ODA defects indicated in this study are labeled in red ( DNAHβ-1 and LRRC50 ) . ( B′ ) TEM image shows cross-section through a cilium of the protonephridial tubule . IDA , inner dynein arm; ODA , outer dynein arm; DHC , dynein heavy chain; LC , dynein light chain; IC , dynein intermediate chain; DC , docking complex . ( C ) Whole-mount expression patterns of DNAHβ-1 and LRRC50 by in situ hybridization . Scale bars: 500 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 07405 . 04410 . 7554/eLife . 07405 . 045Figure 7—figure supplement 2 . Cystogenesis in planarian protonephridia under IFT88 ( RNAi ) is correlated with increased proliferation . ( A ) Quantification of average area of each slc6a13+ tubule ( right ) in Control ( RNAi ) , IFT88 ( RNAi ) and NPHP8 ( RNAi ) animals to evaluate the severity of protonephridial cystogenesis . ( B ) Quantification of mitoses in Control ( RNAi ) , IFT88 ( RNAi ) and NPHP8 ( RNAi ) animals . NPHP8 ( RNAi ) animals served as a positive control in this set of experiment . The time points in the bar graph indicate the number of day after the last dsRNA introduction . ( C ) Quantification of average area of each slc6a13+ tubule ( right ) in Control ( RNAi ) , IFT88 ( RNAi ) and LRRC50 ( RNAi ) animals to evaluate the severity of protonephridial cystogenesis . ( D ) Quantification of mitoses in Control ( RNAi ) , IFT88 ( RNAi ) and LRRC50 ( RNAi ) animals . * p < 0 . 05; ** p < 0 . 01; *** p < 0 . 001 vs control . DOI: http://dx . doi . org/10 . 7554/eLife . 07405 . 045 Therefore , we decided to systematically test possible mechanistic roles of cilia in planarian cyst ontogeny . If cilia were generally required for maintaining the structure/function of protonephridia , then all disruptions of cilia structure should cause cystic phenotypes . We therefore knocked down Smed-IFT88 , a component of the intraflagellar transport machinery . As previously shown ( Rink et al . , 2009 ) , IFT88 ( RNAi ) animals lost their cilia-dependent gliding ability ( Figure 7—figure supplement 1A ) , developed massive tissue edema and had severely shortened cilia ( Figure 7E , E′ ) . Interestingly , IFT88 ( RNAi ) animals also developed cystic protonephridia ( Figure 7E′′ ) and cystogenesis in IFT88 ( RNAi ) animals was also associated with increased proliferation ( Figure 7—figure supplement 2 ) and the abnormal accumulation of protonephridial progenitors ( Figure 7E′′′ ) . These results therefore demonstrate that disruption of cilia is sufficient for cyst development in planarians . In contrast to adult mammalian kidneys that contain only immotile sensory cilia ( Webber and Lee , 1975 ) , the excretory systems of planarians and many other lower vertebrates possess motile cilia ( Figure 7—figure supplement 1B; [McKanna , 1968a , 1968b; Ishii , 1980a , 1980b; Lacy et al . , 1989; Kramer-Zucker et al . , 2005; Rink , 2013] ) that collectively drive fluid flow into and through the tubules ( White , 1929; Pontin , 1964; Warner , 1969; Kramer-Zucker et al . , 2005 ) . Hence cilia-dependent cyst formation might reflect either a requirement of cilia as flow generators and/or as flow sensors . We first sought to disrupt ciliary beating without gross changes in cilia length or structure . We therefore targeted two planarian homologues of Primary Ciliary Dyskinesia ( PCD ) disease genes , a rare ciliopathy causing general cilia immobility in humans ( Badano et al . , 2006 ) ( Figure 7—figure supplement 1B , C ) . Disrupting the function of Smed-DNAHβ-1 and Smed-LRRC50 by RNAi led to abnormal gliding ability ( Figure 7—figure supplement 1A ) due to loss of ciliary beating ( Figure 7H–I; Videos 6–8 ) , while cilia length and structure appeared unaffected ( Figure 7F′–G′ ) . Interestingly , DNAHβ-1- and LRRC50 ( RNAi ) animals also developed edema and formed protonephridial cysts ( Figure 7F , G , F′′ , G′′ , F′′′ , G′′′ ) . These results indicate that reduced ciliary beating rate without change in cilia structure is sufficient to cause the cystic phenotype in planarian protonephridia . Together , these results suggest that cilia-driven fluid flow is crucial to orchestrate tubular cell homeostasis in planarian protonephridia . 10 . 7554/eLife . 07405 . 046Video 6 . Cilia beating in Control ( RNAi ) animal . Serial images show beating of the cilia along the lateral body edge of the planarian head region in Control ( RNAi ) animal . DOI: http://dx . doi . org/10 . 7554/eLife . 07405 . 04610 . 7554/eLife . 07405 . 047Video 7 . Cilia beating in DNAHβ-1 ( RNAi ) animal . Serial images show beating of the cilia along the lateral body edge of the planarian head region in DNAHβ-1 ( RNAi ) animal . DOI: http://dx . doi . org/10 . 7554/eLife . 07405 . 04710 . 7554/eLife . 07405 . 048Video 8 . Cilia beating in LRRC50 ( RNAi ) animal . Serial images show beating of the cilia along the lateral body edge of the planarian head region in LRRC50 ( RNAi ) animal . DOI: http://dx . doi . org/10 . 7554/eLife . 07405 . 048
An important consequence of these findings is that planarian protonephridia provide an invertebrate model for human kidney diseases . Both human nephrons and planarian protonephridia rely on pressure-driven ultrafiltration with subsequent filtrate modification during its passage through the nephridial tubules . This common design principle is significant for two reasons: first , many human kidney pathologies involve filtrate flow ( see below ) ; and second , the excretory systems of existing invertebrate model systems do not recapitulate this critical aspect . In fruitflies and other insects , ultrafiltration and filtrate modifications are carried out in separate cell types and organs ( ultrafiltration: nephrocytes; filtrate modification: Malpighian tubules ) ( Dow and Romero , 2010 ) . C . elegans has a highly derived excretory system consisting of just a single cell ( Buechner , 2002 ) . Both are therefore unsuitable for modeling flow-related kidney diseases , including polycystic kidney disease ( PKD ) . PKD is one of the most common , life-threatening genetic diseases that affects millions worldwide ( Wilson , 2004a , 2004b , 2011 ) . The pathology of PKD is characterized by the massive overproliferation of kidney cells , coupled with aberrant differentiation and the formation of fluid-filled cysts ( Wilson , 2004a , 2004b , 2011 ) . In contrast to C . elegans and D . melanogaster , our results demonstrate that ultrafiltration and filtrate flow are as important in maintaining excretory organ homeostasis in both planarians and humans . Such parallels between human nephrons and planarian protonephridia may seem surprising at first , given that the two excretory systems are phenotypically very different . In vertebrates , individual nephron units combine within the kidney as a single organ , whereas in planarians , protonephridial units are pervasively present throughout the body ( Rink et al . , 2011 ) . Further , in vertebrates , it is blood pressure and , ultimately , the heart that drives primary filtrate across the glomerular basement membrane and into the nephron . Planarians lack a circulatory system and therefore cannot solely rely on pressure-driven flow generation . The dense and vigorously beating cilia bundles of flame cells , and the multiciliated lining of the PT suggest instead that motile cilia are responsible for driving protonephridial ultrafiltration and flow . In fact , previous studies of protonephridia in other invertebrates strongly support this conclusion , including the direct observation of fluid flow cessation upon pharmacological inhibition of ciliary beating in the rotifer Asplanchna ( Pontin , 1964; Warner , 1969 ) . The nephrons of the adult human kidney also bear cilia . However , in contrast to the multiciliated and flow generating cilia of the planarian PT , the cells of the adult mammalian nephron bear a single , immotile primary cilium protruding into the lumen ( Webber and Lee , 1975 ) . Despite these structural and functional differences , we find that cilia are similarly crucial for the maintenance of organ homeostasis in both systems . In humans , loss of function of many genes involved in cilia biogenesis/function ( Eley et al . , 2005; Hildebrandt and Otto , 2005; Winyard and Jenkins , 2011; Kotsis et al . , 2013 ) leads to overproliferation and cyst formation , including for example the NPHP proteins , that anchor the basal body to the cell membrane and are required for primary cilia formation ( Betleja and Cole , 2010; Craige et al . , 2010; Omran , 2010; Williams et al . , 2011 ) . The importance of cilia is thought to be due to their function as flow sensors ( Winyard and Jenkins , 2011; Kotsis et al . , 2013 ) . Flow-induced bending is hypothesized to cause calcium influx via stretch sensitive polycystin channels , which subsequently inhibits cell proliferation via still unknown mechanisms ( Praetorius and Spring , 2001 , 2003; Nauli et al . , 2003; Praetorius et al . , 2004 ) . Unexpectedly , in spite of the very different structure and function of planarian protonephridial cilia , we found that they are similarly critical in maintaining form and function of the organ system . First , knockdown of the planarian homologues of NPHP proteins lead to overproliferation of protonephridial progenitors . Second , the ablation of cilia by IFT88 ( RNAi ) or the inhibition of ciliary beating by interfering with axonemal dyneins leads to protonephridial progenitor overproliferation . Moreover , in both cases , accumulating protonephridial progenitors form dense aggregates with disorganized lumens that quickly compromise the osmoregulatory functions of the organs , leading to edema formation . Hence the disruption of planarian cilia leads to phenotypically very similar alterations as PKD in vertebrates . We therefore conclude that cilia-mediated flow sensing constitutes an ancient mechanism for maintaining excretory organ homeostasis . Planarians have lost central components of centrosome duplication , which should therefore preclude their ability to form primary cilia ( Azimzadeh et al . , 2012 ) . Since planarians only appear to harbor motile cilia , the corollary of this argument would be that the cilia of the protonephridial tubule have a dual role as flow generators and as flow sensors . Our data cannot conclusively address this point , because the currently available experimental paradigms do not allow the requisite uncoupling between cilia and flow . NPHP proteins are generally thought to only be essential for immotile sensory cilia ( Eley et al . , 2005; Badano et al . , 2006; Ferkol and Leigh , 2012 ) based on their involvement in anchoring basal body to the cell membrane ( Betleja and Cole , 2010; Craige et al . , 2010; Omran , 2010; Williams et al . , 2011 ) . Loss of NPHP ‘sensory’ machinery in the multiciliated tubule cells leads to cyst formation in planarians , indicating the putative flow-sensing role of protonephridial cilia . However , bronchiectasis , a classic phenotypic feature of PCD , has been found in many cases of NPHP ( Bagga et al . , 1989; Bergmann et al . , 2008; Moalem et al . , 2013; Wolf , 2015 ) as well as autosomal dominant PKDs ( Driscoll et al . , 2008; Moua et al . , 2014 ) , indicating impaired motor cilia function . Conversely , phenotypic features of defective sensory cilia , including retinitis pigmentosa and cystic kidneys , have been reported in PCD patients ( Kartagener and Horlacher , 1935; Saeki et al . , 1984; Moore et al . , 2006 ) . These observations suggest an under-appreciated overlap between NPHP gene functions in motile cilia with that of PCD genes in immotile cilia , which makes it difficult to segregate flow generating and flow sensing roles of protonephridial cilia . However , recent work in vertebrate systems indicates that the transduction of sensory cues from the environment is a universal characteristic of all cilia , including motile cilia ( Teilmann and Christensen , 2005; Shah et al . , 2009 ) , suggesting the likely existence of flow sensing role of motile cilia in planarian protonephridia ( Schwartz et al . , 1997; Teilmann and Christensen , 2005; Shah et al . , 2009; Quarmby and Leroux , 2010; Takeda and Narita , 2012 ) . In humans , polycystin-1 and polycystin-2 are key components of cilia-mediated flow sensing ( Praetorius and Spring , 2001 , 2003; Nauli et al . , 2003; Praetorius et al . , 2004 ) . As stretch-activated calcium channels , these proteins are thought to transduce the mechanical bending of the cilium into a chemical signal . Planarians have homologues of both polycystins and their knockdown leads to defects in ciliated dorsal mechanosensory neurons , but interestingly , not in protonephridia ( Vu HTK and Rink JC , unpublished ) . Hence flow sensing in planarian protonephridia likely utilizes different mechanisms . The future elucidation of the underlying mechanisms may uncover general mechanistic links between flow sensing and flow generating in motile cilia . Given the extensive evolutionary homology between excretory systems , how can the exclusive reliance on motile cilia driven flow generation in invertebrates be reconciled with their absence in adult vertebrate kidneys ? The loss of motile cilia in vertebrate kidneys has been observed in coincidence with the increase in blood pressure in birds and mammals ( Marshall , 1934 ) . This raises the possibility that the evolution of an extensively developed circulatory system in mammals rendered the flow-generating role of motile cilia redundant , and subsequently led to their loss from the adult kidney . The evidence for this hypothesis can be found in the embryology of the vertebrate kidney , which , as Ernst Haeckel argued , may indeed hint at phylogeny ( Haeckel , 1866 ) . Many vertebrates display multiciliated flow generation during early developmental stages ( White , 1929; Lacy et al . , 1989; Kramer-Zucker et al . , 2005 ) . For instance , the pronephric tubules of zebrafish contain multiciliated epithelial cells , and interference with cilia obstructs flow ( Kramer-Zucker et al . , 2005 ) . In humans , motile cilia may even play a role at early kidney development , since multiciliated tubular cells have been observed at fetal stages ( Zimmermann , 1971; Katz and Morgan , 1984 ) . Furthermore , a number of human pathologies such as crescentic and membranoproliferative glomerulonephritis ( Katz and Morgan , 1984; Hassan and Subramanyan , 1995 ) , as well as focal segmental glomerulosclerosis ( Katz and Morgan , 1984 ) are associated with the appearance of multiciliated tubular cells . This suggests that the gene regulatory programs for the development of multiciliated cells remain intact and accessible even in vertebrates . Altogether , these observations suggest that cilia-driven pressure generation may be the ancestral state for excretory systems , and that the immotile primary cilia of the adult vertebrate kidney represents an evolutionary vestige of the cilia-driven flow filtration observed in more basal excretory organs . Even though the exact flow sensing mechanisms remain to be identified , our data show that in planarians , these signals also affect cell proliferation . We observed a global increase of mitoses due to the specific overproliferation of protonephridial progenitors whenever cilia or fluid flow were affected ( Figure 6B–E ) . The aberrant proliferation and differentiation of these cells ultimately lead to dense cell accumulations with discontinuous lumens . In NPHP knockdown planarians , cysts primarily arise in the PT at early stage , but over time , progressively affect the DT as well ( Figure 6—figure supplement 2B ) . The severity of the cystic phenotype intimately depends on global proliferation rates ( Figure 6F–J ) , which , interestingly , is also the case in human CKDs ( Wilson and Goilav , 2007 ) . However , not all aspects of the planarian NPHP ( RNAi ) phenotypes phenocopy human NPHP . In human NPHP , cysts arise at the corticomedullary junction of the kidneys , meaning that they mostly develop from the distal convoluted and collecting tubules ( Hildebrandt and Zhou , 2007 ) . This discrepancy could be due to the differential distribution of cilia , which in planarians are only present in the PT ( Rink et al . , 2011 ) . Accordingly , PT cysts could be the direct consequence of cilia dysfunction , while structural alterations of the DT could be a secondary consequence of overproliferation of protonephridial progenitors . Conversely , cilia are found on the apical surface of most epithelial cells lining the human nephrons with the exception of the intercalated cells interspersed along the CD ( Webber and Lee , 1975 ) . Nevertheless , why cysts predominantly affect the distal convoluted and collecting tubules only , but not other tubule parts in human NPHP is still unknown . Overall , the interplay between cilia and cell proliferation in protonephridial cyst formation is remarkably similar between humans and planarians . Therefore , in planarian protonephridia , cilia appear to generate a non-cell autonomous signal capable of regulating the proliferation of protonephridial progenitors and orchestrating their integration into the protonephridial tubules ( Figure 7J ) . The search for the flow-regulated progenitor populations in the adult vertebrate kidneys is currently an active area of research ( Elger et al . , 2003; Diep et al . , 2011; McCampbell and Wingert , 2012; Rinkevich et al . , 2014 ) . The mechanisms constituting the non-cell autonomous signal between tubule cells and progenitor cells is a second missing element in our current understanding of human CKDs and the homeostatic role of flow sensing in general . Given the striking evolutionary conservation of flow sensing and flow-dependent progenitor proliferation between planarians and vertebrates , it seems likely that also this elusive signal is conserved . In this regard , the high speed and low cost of high-throughput RNAi screening in planarians therefore provides a novel experimental paradigm for gene discovery and mechanistic studies of human kidney diseases .
The CIW4 clonal line of S . mediterranea was maintained as described ( Cebria and Newmark , 2005 ) . 1 week starved animals were used for all experiments . For irradiation experiments , animals were exposed to 1250 or 6000 rads on a GammaCell 40 Exactor irradiator . Human , mouse , Xenopus and zebrafish protein sequences were used to find planarian homologs from S . mediterranea genome database via TBLASTN . Planarian homologs were then used for reciprocal BLAST against the human refseq to verify the homology . All genes were cloned from an 8 day regeneration time course cDNA library prepared as described previously ( Gurley et al . , 2008 ) . Primers used for cloning are described in Supplementary files 1 , 4 . The complete set of protein sequences were retrieved for human , mouse , and fly from Ensembl ( release 76 ) ( Flicek et al . , 2014 ) . The mosquito protein sequences were retrieved from Ensembl metazoa ( release 23 ) . Only the proteins corresponding to the longest isoform of each gene were considered for the analysis . The PFAM protein domains ( PfamA-27 . 0 ) ( Finn et al . , 2014 ) were predicted for all those proteins from human , mouse , fly and mosquito and the planarian homologs of solute carriers using the InterProScan ( version 5 . 4–47 . 0 ) tool ( Jones et al . , 2014 ) . The solute carrier proteins were classified into their corresponding solute carrier family or clan groups based on the presence of the corresponding PFAM protein domain as described in the literature ( He et al . , 2009; Hoglund et al . , 2011 ) . The predicted domain regions were extracted from those proteins and multiple sequence alignment was then performed for those extracted regions using clustalw2 ( version 2 . 1 , with default parameters ) ( Larkin et al . , 2007 ) . Using the sequence alignment , the bootstrapped neighbor joining trees ( positions with gaps removed and corrected for multiple substitution ) were constructed using clustalw2 ( version 2 . 1 ) ( Larkin et al . , 2007 ) . Colorimetric and FISHs were performed as previously described ( Pearson et al . , 2009; King and Newmark , 2013 ) . Following fluorescent or NBT/BCIP development , animals were incubated with anti-acetylated-Tubulin antibody ( 1:1000 , Cell Signaling , Danvers , MA ) , anti-H3P ( 1:1000 , Millipore ) , or a rabbit antiserum recognized unknown epitope to visualize the lumen of PT ( 1:500 ) . Primary antibodies were detected with either Alexa-conjugated anti-rabbit antibodies ( 1:1000; Abcam ) or HRP-conjugated anti-rabbit antibodies ( 1:1000; Jackson ImmunoResearch ) . NBT/BCIP developed whole-mount in situ specimens were mounted in mounting media containing 75% glycerol and 2 M urea . Fluorescent whole-mount in situ specimens were mounted in modified ScaleA2 containing 20% glycerol , 2 . 5% DABCO and 4 M urea ( Hama et al . , 2011 ) . For cryosectioning , fluorescently stained whole-mounted animals were fixed overnight in 4% paraformaldehyde ( in PBS ) at 4°C , washed three times in PBS , equilibrated in 30% sucrose , frozen in OCT , and cryosectioned ( 10–20 μm ) . A Leica M205 Stereo Microscope was used for documenting live images , videos , and NBT/BCIP developed whole-mount in situ specimens . Zeiss LSM-510 VIS or LSM-700 Upright confocal microscopes were used to capture fluorescent whole-mount in situ specimens and image projections . To quantify the average size of each protonephridial unit and mitotic activity , individual worm was imaged and tiled on a Perkin Elmer Ultraview spinning disk microscope . Stitching and mitotic activity quantification was performed in FiJi using standard plugins ( Schindelin et al . , 2012 ) . Worm area , protonephridial size and number were measured/counted using a custom signal to noise thresholding and seeded region grow plugins . Batching was performed using macros . Movement speed quantification was performed on video sequences ( acquired at 17 . 5 Hz ) using a custom thresholding plugin and Mtrack2 ( Klopfenstein and Vale , 2004 ) . For each tracked object , the initial position was subtracted from the final to determine an average translocation velocity . Average velocities were computed by weighting track averages by the length of the track . Plugins and macros are available at https://github . com/jouyun . BrdU was administered by soaking animals in 15 mg/ml BrdU and 3% DMSO ( diluted in 0 . 1× Montjuic salts ) for 1 hr as previously described ( Cowles et al . , 2012 ) and chasing for specified time . Animals were fixed and processed as in situ hybridization protocol except they were bleach in 6% H2O2 in PBSTx ( 0 . 5% Triton ) for 3–4 hr under direct light . After in situ development , specimens were treated with 2 N HCl for 45 min at room temperature , and washed four times with PBSTx ( 0 . 3% Triton ) for 1 hr . BrdU was detected using rat anti-BrdU antibody ( 1:1000; Abcam , Cat . No . ab6326 ) . Primary antibody was detected with HRP-conjugated anti-rat antibody ( 1:1000; Jackson ImmunoResearch ) . To assay ultrafiltration capacity of planarian protonephridia , 10 kDa tetramethylrhodamine-dextran ( Molecular Probes , D-1817 ) and 500 kDa fluorescein-dextran ( Molecular Probes , D-7136 ) at the concentration of 1 mg/ml were co-injected into the mesenchyme of the animals . After 2 hr , the animals were rinsed with an excess of 1× Montjuic salts , fixed in cold 4% paraformaldehyde ( in 1× Montjuic salts ) , mounted in modified ScaleA2 and photographed using a Zeiss LSM-510 VIS confocal microscope . Dextran uptake was quantified by measuring the average fluorescence intensity per unit area using a standard signal to noise thresholding in Fiji ( Schindelin et al . , 2012 ) . For immunostaining , after fixation , the samples were rinsed 3–4 times with PBSTx ( 0 . 3% Triton ) , incubated in blocking solution containing 5% horse serum in PBSTx ( 0 . 5% Triton ) for 2 hr at room temperature , and then in anti-acetylated-Tubulin antibody ( 1:1000 , Cell Signaling ) . Primary antibody was detected using Alexa-conjugated anti-rabbit antibodies ( 1:1000; Abcam ) . Intracellular pH was measured using ratiometric pH dye SNARF-5F-AM ( Molecular Probes , Cat . No . S-23923 ) at 5 μM ( in DMSO with 20% wt/vol Pluronic F-127 ) as previously described ( Beane et al . , 2011 , 2013 ) . Animals were soaked for 1 hr , rinsed three times with an excess of 1× Montjuic salts , immobilized on the glass bottom dish using the microfluidic device and imaged at both 640 nm ( pH sensitive ) and 580 nm ( pH insensitive ) wavelengths using a LSM-700 Falcon confocal microscope . The ratio of 580/640 ( used for controlling uneven dye uptake ) was shown . To visualize cilia beating along the lateral body edge of the planarian head region , live worms are immobilized on the glass bottom dish using a microfluidic device and imaged on a Zeiss Axiovert 200 microscope under DIC optics using 63× objective . Series of images were captured at 250 frames per second with pixel number of 800 × 800 ( exposure time is 3 . 97 ms ) using an ORCA-Flash4 . 0 V2 C11440-22CU camera from Hamamatsu . Spatiotemporal image correlation spectroscopy was used to determine the speed of the cilia for each animal . In each time-lapse , 100 consecutive frames were manually selected in which the animal was stationary so that no image registration was required . A region of interest was manually drawn around the cilia in each time-lapse . The area outside this region was uniformly filled with the average intensity inside the region . Spatiotemporal correlation was then carried out in 32 × 32 pixel regions with a 16 pixel overlap between the regions to allow for highly localized motions to be accurately represented using the fast Fourier transform method . The average cilia displacement within the correlation image is represented by the maximum of the spatial cross-correlation between two images separated in time . The time correlation shift was a single frame , and all velocities were converted to micrometers per minute . This method was adapted from a previous paper ( Yi et al . , 2011 ) , where it was implemented with custom plugins written in Java for ImageJ , available for download at ( http://research . stowers . org/imagejplugins ) . Statistical analysis of the data was carried out in Excel . p values were determined using Student's t-test . Specimens were prepared as following at 4°C on orbital rotator: ( 1 ) fix in cold 2 . 5% glutaraldehyde in 0 . 05 M or 0 . 1 M sodium cacodylate ( contained 1 mM CaCl2 ) for overnight; ( 2 ) wash in wash buffer ( 0 . 1 M sodium cacodylate buffer; 1 mM CaCl2; and 1% sucrose ) for 1 hr ( 3–4 exchanges ) ; ( 3 ) fix in 1% Osmium tetroxide in 0 . 1 M sodium cacodylate buffer ( +1 mM CaCl2 ) for 2 hr; ( 4 ) wash in wash buffer for 1 hr ( 3–4 exchanges ) and in distill water for 30 min ( 3–4 exchanges ) ; ( 5 ) fix in 0 . 5% aqueous Uranyl Acetate ( in dark ) overnight; ( 6 ) wash in distill water for 30 min ( 3–4 exchanges ) ; ( 7 ) and dehydrate in acetone 30% ( 20 min ) , 50% ( 20 min ) , 70% ( overnight ) , 90% ( 20 min , 2 times ) , and 100% ( 20 min , 3 times ) . Specimens were then embedded in epon-araldite or Spurr's resin as follows: 25% resin/acetone for 3 hr; 50% resin/acetone for 2 . 5 hr; 75% resin/acetone overnight; 100% resin without accelerator with microwave at 350 W for 3 min on/3 min off/3 min on for 1 day ( 2 exchanges ) ; 100% resin with accelerator with microwave at 350 W for 3 min on/3 min off/3 min on for 1 day ( 2 exchanges ) and placed in 60°C oven for polymerization for 2 days . Ultra-thin 50–100 nm sections were collected using a Leica UC6 Ultramicrotome . TEM specimens were stained with Sato's lead ( 3 min ) /4% Uranyl Acetate in 70% methanol ( 4 min ) /Sato's lead ( 6 min ) prior to imaging on a FEI Technai BioTwin at 80 kV equipped with a Gatan UltraScan 1000 digital camera . RNAi feedings were performed as described previously ( Gurley et al . , 2008; Rink et al . , 2009 ) . Feeding and amputation schedules were tailored for each experiment and described in detail as following: Figure 3C: 5 dsRNA feedings ( 3 days in between ) . Figure 4C , D , Figure 4—figure supplement 2A: 8 dsRNA feedings ( 3 days in between ) . Figure 4E , F: 9 dsRNA feedings ( 3 days in between ) . Figure 4—figure supplement 2B , Figure 4—figure supplement 3: 6 dsRNA feedings ( 3 days in between ) prior to amputation . Figure 5A–C: 3 dsRNA feedings ( 3 days in between ) . Figure 6 , Figure 6—figure supplement 1C–E , Figure 6—figure supplement 2C , Figure 6—figure supplement 3: 2 dsRNA feedings ( 3 days in between ) . Figure 7A–C: 3 dsRNA feedings ( 2 days in between ) . Figure 7D–I , Figure 7—figure supplement 1A: IFT88- and LRRC50 ( RNAi ) : 3 dsRNA feedings ( 2 days in between ) ; DNAHβ-1 ( RNAi ) : 8 dsRNA feedings ( 2 days in between ) . | Millions of people around the world are affected by cystic kidney diseases , which are amongst the most common inherited genetic disorders . Throughout their life , people with these diseases develop fluid-filled cysts in their kidneys , which stop these organs from working properly and can eventually lead to organ failure . Healthy kidneys perform many vital roles in the body , including removing waste products and keeping the concentration of salts in the blood in balance . These activities depend on the kidneys filtering the blood , and then reabsorbing useful chemicals from the filtered fluid as it passes through small tubes called tubules . Cysts disrupt both of these processes . Mutations in many different genes can cause cystic kidney diseases . Many of these genes encode proteins that are involved in the formation of cilia: hair-like structures that project from some cell membranes . Cilia on the cells that line tubules are thought to bend in response to the flow of fluid and then generate signals that dampen cell proliferation . This would explain how the loss of cilia could cause too many cells to develop , which would lead to the formation of cysts . But many of the molecular details are missing from this explanation . Previous studies in mammals and simple model organisms ( such as fruit flies and roundworms ) have sought to fill in the gaps , but each model has its own limitations . Now , Thi-Kim Vu et al . propose that a flat worm called a planarian could provide a new and experimentally accessible animal model to study cystic kidney diseases . These flat worms get rid of their waste products via an excretory system that consists of branched tubules that spread throughout the body . Thi-Kim Vu et al . found that , like the tubules in the kidney , these tubules filter and then reabsorb chemicals from body fluids . Moreover , these processes are performed in different parts of the tubules , exactly as they are in the tubules in kidneys . The genome of a flat worm called Schmidtea mediterranea contains many genes that cause cysts to form when their equivalents are mutated in humans . Reducing the expression of these genes ( and others that are involved in cilia formation ) also caused cysts to form in the flat worms . These findings indicate that it is likely that the excretory systems of different animals have a shared evolutionary history . If so , the findings support the idea that cilia in kidney tubules send signals in response to fluid flow that affect kidney-specific stem cells . They also suggest that problems with these signals could be at the core of some human cystic kidney diseases . One of the next challenges will be to identify these cilia-associated signals . Finally , given that studies involving thousands of flat worms can be carried out with minimal cost , the ultimate goal is to develop flat worms into a new model to discover and investigate genes linked to human kidney diseases . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"stem",
"cells",
"and",
"regenerative",
"medicine"
] | 2015 | Stem cells and fluid flow drive cyst formation in an invertebrate excretory organ |
Clathrin-mediated endocytosis is an essential process that forms vesicles from the plasma membrane . Although most of the protein components of the endocytic protein machinery have been thoroughly characterized , their organization at the endocytic site is poorly understood . We developed a fluorescence microscopy method to track the average positions of yeast endocytic proteins in relation to each other with a time precision below 1 s and with a spatial precision of ∼10 nm . With these data , integrated with shapes of endocytic membrane intermediates and with superresolution imaging , we could visualize the dynamic architecture of the endocytic machinery . We showed how different coat proteins are distributed within the coat structure and how the assembly dynamics of N-BAR proteins relate to membrane shape changes . Moreover , we found that the region of actin polymerization is located at the base of the endocytic invagination , with the growing ends of filaments pointing toward the plasma membrane .
Clathrin-mediated endocytosis is a key membrane trafficking process for the uptake of cargo molecules from the cell surface . It is involved in numerous different biological contexts from the generation of cellular polarization , to virus uptake and the regulation of neural signaling . The formation of clathrin-coated endocytic vesicles is driven by a complex molecular machinery composed of more than 50 different proteins ( Doherty and McMahon , 2009; Boettner et al . , 2012; Weinberg and Drubin , 2012 ) . These proteins recruit cargo molecules and reshape the plasma membrane to generate the endocytic vesicle . The assembly dynamics of the endocytic protein machinery has been described in detail by live-cell imaging ( Gaidarov et al . , 1999; Merrifield et al . , 2002; Kaksonen et al . , 2003 , 2005; Newpher et al . , 2005; Sirotkin et al . , 2010; Loerke et al . , 2011; Taylor et al . , 2011; Cocucci et al . , 2012; Berro and Pollard , 2014 ) . However , the spatial organization of the endocytic proteins during vesicle budding remains poorly understood . The complexity and the size of the endocytic machinery , below the resolution limit of light microscopy , have made it challenging to reveal how the endocytic proteins are organized during vesicle budding . Budding yeast Saccharomyces cerevisiae has been used extensively as a model organism to study endocytosis . The protein machinery of clathrin-mediated endocytosis is largely conserved between mammals and yeast ( Conibear , 2010 ) . In yeast , endocytosis starts with the recruitment of clathrin and several clathrin adaptors and accessory proteins to the plasma membrane to initiate the assembly of the vesicle coat ( Carroll et al . , 2012; Godlee and Kaksonen , 2013 ) . During this initial early phase , which can last between 40 and 90 s , the coat assembles on a flat plasma membrane and cargo molecules are recruited to the endocytic site . Later , additional coat proteins and actin interactors are recruited . The actin interactors include Las17 ( homolog of mammalian N-WASP ) and type I myosins Myo3 and Myo5 , which both activate the actin filament nucleating Arp2/3 complex ( Winter et al . , 1999; Evangelista et al . , 2000; Lechler et al . , 2000 ) . In yeast , the invagination of the plasma membrane , and the consequent inward movement of the coat , is concomitant with the appearance of actin at the endocytic site ( Kukulski et al . , 2012 ) . A functional dependence on actin polymerization is possibly due to high turgor pressure ( Aghamohammadzadeh and Ayscough , 2009 ) . In mammals , actin is similarly critical under conditions of high membrane tension ( Boulant et al . , 2011 ) . Two of the coat-associated proteins , Sla2 ( homolog of mammalian Hip1R ) and the epsin Ent1 , interact directly with lipids , clathrin and actin , and are essential for mediating the forces from the actin cytoskeleton to deform the membrane ( Baggett et al . , 2003; Sun et al . , 2005; Skruzny et al . , 2012 ) . The shape of the endocytic invagination is regulated by proteins containing BAR domains , such as the Rvs161/167 heterodimer , which are localized at the neck region of the membrane invagination ( Idrissi et al . , 2008; Youn et al . , 2010; Kishimoto et al . , 2011 ) . After vesicle scission the endocytic machinery quickly disassembles and the free vesicle is trafficked further into the cell . The exact molecular mechanisms of endocytosis remain unknown , largely due to lack of knowledge about the functional organization of the protein components of the endocytic machinery . Here , we developed a novel imaging approach to characterize the spatial and temporal relationship between the different protein components of the endocytic machinery during vesicle budding . Furthermore , we integrated this data with time-resolved membrane shapes that were obtained previously by correlated light and electron microscopy ( Kukulski et al . , 2012 ) . With this approach , we determined several key features of the dynamic architecture of the endocytic machinery during vesicle budding ( Figure 1A ) . 10 . 7554/eLife . 04535 . 003Figure 1 . Tracking procedure . ( A ) The rational behind our approach . The centroid positions of endocytic proteins were correlated with the plasma membrane intermediates derived from CLEM ( Kukulski et al . , 2012 ) . We could thus position the protein complexes along the plasma membrane invagination to reconstruct the molecular architecture of the endocytic machinery . ( B ) A yeast cell expressing the coat protein Sla2 tagged N-terminally ( GFP-Sla2 ) and a collection of trajectories of GFP-Sla2 centroid in different endocytic events . The trajectories are oriented so that the plasma membrane lies horizontally at the bottom and the inward movement axis is vertical . The triangle and the square mark the start and the end of each trajectory , respectively . ( C ) The information content of the GFP-Sla2 average trajectory: The average trajectory movement on the focal plane . The trajectory is aligned so that the Y-axis represents the inward movement along the invagination and the X-axis represents the movement along the plasma membrane ( left panel ) . The inward movement of the trajectories over time ( right panel ) . The number of molecules recruited at the endocytic site over time ( bottom panel ) . The 53 individual trajectories that were used to generate this average are plotted ( gray ) together with the average ( blue ) . The contour lines highlight point densities . ( D ) A yeast cell expressing GFP-Sla2 and the reference protein Abp1-mCherry and a collection of trajectories derived from the simultaneous acquisition of the target and reference proteins . ( E ) A diagram summarizing the steps that led to the spatial and temporal alignment of the average trajectories . Abp1 is used as the spatial and temporal reference . The average trajectory of the target protein ( GFP-Sla2 in this example ) is aligned to the average trajectory of Abp1-GFP ( the reference protein ) by aligning each of them to the respective trajectories acquired simultaneously in cells expressing GFP-Sla2 and Abp1-mCherry . More than 50 endocytic events are used to derive the average spatial and temporal transformation that aligns the average trajectories together . The shading represents the confidence interval ( See ‘Materials and methods’ ) . Scale bars in images are 1 µm long . See also Figure 1—figure supplements 1 , 2 , 3 , 4 , 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 04535 . 00310 . 7554/eLife . 04535 . 004Figure 1—figure supplement 1 . Average trajectories of endocytic events . ( A–G ) The average trajectories and the individual trajectories that were used to generate the average trajectories are shown together . The contour lines highlight point densities . The distribution of the individual trajectories is unimodal confirming that the dynamics of the proteins investigated in this study do not show substantial heterogeneity during the process of endocytosis . The shading represents the confidence interval . DOI: http://dx . doi . org/10 . 7554/eLife . 04535 . 00410 . 7554/eLife . 04535 . 005Figure 1—figure supplement 2 . Experimental controls for the alignment procedure . ( A ) Lifetimes of Abp1 patches in strains expressing Abp1-mCherry and a target protein labeled with GFP were used to control the functionality of the tagged proteins . Only Las17-GFP strain showed a significant difference in Abp1 patch lifetime ( pvalue <0 . 05 ) . Error bars represent the SE . ( B ) The time required by Abp1-GFP to reach its peak in fluorescence intensity . This gives an estimate of the time needed to invaginate the plasma membrane ( Figure 2A–C ) . The time distribution is unimodal and normal ( Shapiro–Wilk test of normality , p value = 0 . 07 , H0: the distribution is normal . ) ( C ) The movement of Abp1 centroid was used as a reference for the alignment of the trajectories of different endocytic proteins . Its movement is on average directed perpendicularly to the plasma membrane . The histogram shows the distribution of angles between the vector representing the direction of movement of Abp1 trajectories ( blue arrow ) and the vector tangent to the plasma membrane ( red arrow ) . Mean angle is given as mean ± SEM . ( D ) Yeast cells were imaged at the equatorial plane therefore the invagination movement is projected on the focal plane . An endocytic event imaged at the edge of a depth of field of 500 nm ( dashed lines , right plot ) would induce a maximal underestimation of 0 . 5% of the invagination movement in a yeast cell of 2 . 5 μm in radius . ( E ) Repeated alignments of the same average trajectories ( Abp1-GFP , Rvs167-GFP and GFP-Sla2 ) using five independent data sets of simultaneous two color acquisitions ( Abp1-mCherry-GFP , Rvs167-GFP with Abp1-mCherry , and GFP-Sla2 with Abp1-mCherry ) . σx and σyare the standard deviations of the five alignments in time and space respectively . The shading represents the confidence interval . DOI: http://dx . doi . org/10 . 7554/eLife . 04535 . 00510 . 7554/eLife . 04535 . 006Figure 1—figure supplement 3 . Simulation of the accuracy of the two color alignment procedure . ( A–D ) Lag τ , rotation Tθ , and translations , Tx and Ty , that align a virtual trajectory to its virtual reference . The trajectory and its reference were generated already aligned and the expected values are 0 for each component of the transformation . The transformation was computed 30 times , each time using 100 trajectory pairs that were generated randomly from the virtual trajectory and the virtual reference with equal incremental values of σp and σr . σp and σr indicate respectively the standard deviation used to generate the trajectories for the target and for the reference protein in each trajectory pair . ( E–H ) The average values for τ , Tθ , Tx and Ty , computed as in ( A–D ) with incremental values of σp . σr is kept 19 nm , which is the noise we encountered experimentally for the reference trajectory in the trajectory pairs . ( I–L ) The average values for τ , Tθ , Tx and Ty , computed as in ( E–H ) but the position of the trajectory of the target protein is 30 nm away from the reference . The error bars represent the standard deviation of the distributions . DOI: http://dx . doi . org/10 . 7554/eLife . 04535 . 00610 . 7554/eLife . 04535 . 007Figure 1—figure supplement 4 . Simulation of the robustness to systematic shifts between the two channels during two color acquisition . ( A ) The shift from its correct position of a trajectory aligned to its reference in the presence of systematic color aberration . Each point shows the mean and standard deviation of 30 repeats of the alignment . Each alignment was computed using 100 pairs of virtual trajectories each of which was generated with σp = 16 nm , for the trajectory of the target protein , and σr = 19 nm , for the trajectory of the reference protein . The reference trajectories were systematically shifted along one direction with incremental color shifts , which are reported along the X-axis . ( B ) As in ( A ) but with the real trajectory pairs that we acquired for Abp1-mCherry and Sla2-GFP . Those trajectory pairs were used to align Sla2-GFP to Abp1-GFP . The trajectories of Abp1-mCherry are shifted along one direction in all the trajectory pairs used to compute the alignment of Sla2-GFP . The error bars represent the confidence interval for the position of Sla2-GFP . ( C–D ) Examples of Sla2-GFP trajectories aligned using trajectory pairs affected by different color shifts . The shading represents the confidence interval . DOI: http://dx . doi . org/10 . 7554/eLife . 04535 . 00710 . 7554/eLife . 04535 . 008Figure 1—figure supplement 5 . Simulation of the accuracy of the trajectory averaging . ( A ) The average trajectory computed from 65 virtual trajectories that were generated from a trajectory template ( ground truth trajectory , shown in red ) adding noise normally distributed around the points of the ground truth trajectory with σ = 10 nm . The trajectories were aligned in space and time to compute the averaged together . The average trajectory was then aligned to its reference using 100 trajectory pairs that were generated from the ground truth trajectory , with noise σp = 10 nm and from the reference trajectory , with noise σr = 19 nm . The average trajectory is compared with the ground truth trajectory shown in red , after the alignment . ( B–D ) Same as ( A ) but with σ = 13 nm , σ = 17 nm and σ = 20 nm respectively . The shading represents the confidence interval . DOI: http://dx . doi . org/10 . 7554/eLife . 04535 . 008
To understand how the endocytic machinery is organized during vesicle budding , we developed an imaging approach to track the movements of different fluorescently labeled endocytic proteins in relation to each other in living cells . We used a two-step procedure: in the first step , we analyzed different endocytic proteins individually to measure their average dynamic behavior and abundance . In the second step , we aligned the resulting data from these different proteins relative to each other in space and time using simultaneous two color imaging . This two-step procedure allowed us to separately optimize the acquisition rate and the alignment precision for single and double channel imaging respectively . We used wide-field epifluorescence microscopy to image GFP-tagged endocytic proteins expressed from their endogenous genomic loci in haploid cells of yeast S cerevisiae . We acquired videos at the equatorial plane of the yeast cells , where the average direction of endocytic vesicle budding is planar with the focal plane ( Kaksonen et al . , 2003; Kukulski et al . , 2012 ) . We thus directly visualized the movements of the endocytic proteins along the axis of membrane invagination ( Kaksonen et al . , 2003; Galletta et al . , 2008 ) . For each GFP-tagged target protein we tracked the centroid position and measured the fluorescence intensity during 50–80 individual endocytic events ( Figure 1B , Table 1 ) . In yeast , the dynamic behavior of the endocytic proteins during vesicle budding is highly stereotypical ( Kaksonen et al . , 2003; Mooren et al . , 2012 ) . We could therefore align the individual centroid trajectories in time and space to calculate average trajectories ( Figure 1C , Figure 1—figure supplement 1 ) . For this purpose we developed an algorithm that finds the translation , rotation and temporal shift minimizing the weighted mean square displacement between the time points of all pairs of trajectories , where the weights are the fluorescence intensities at each time point ( see ‘Materials and methods’ ) . The centroid gives an estimate of the position of the center of mass of the protein distribution; therefore the average trajectory describes the stereotypic dynamic behavior of the center of mass of the target protein molecules during endocytosis . 10 . 7554/eLife . 04535 . 009Table 1 . Number of trajectories used in this studyDOI: http://dx . doi . org/10 . 7554/eLife . 04535 . 009Number of single color trajectories to generate the average trajectoryNumber of trajectory pairs used for spatial and temporal alignmentGFP-Sla253271Sla2-GFP55158Sla1-GFP6658End3-GFP7557Abp1-GFP65NAGFP-Act18392Arc18-GFP6990Rvs167-GFP58340Las17-GFPNA49Myo5-GFPNA50TOTAL5241165The number of events that were tracked and used to compute the average trajectories and to align those average trajectories together . We estimated the average number of molecules present at the endocytic site by calibrating the fluorescence signal using a kinetochore protein of known abundance ( Joglekar et al . , 2006; Lawrimore et al . , 2011 ) . For each protein , we used the average number of molecules to calibrate its fluorescence intensity profile with the absolute numbers of molecules ( Figure 1C; see ‘Materials and methods’ ) . To control that the GFP-tagging did not impair the functionality of the tagged protein we monitored the lifetime of Abp1 patches , which is very sensitive to defects in the endocytic function ( Kaksonen et al . , 2005 ) . We analyzed the lifetime of Abp1-mCherry patches , when Abp1-mCherry was coexpressed with each of the GFP-tagged proteins ( see ‘Materials and methods’ ) . The Abp1-mCherry lifetimes were unaffected by all the GFP-tagged proteins except Las17-GFP , for which Abp1-mCherry lifetime was slightly longer ( Figure 1—figure supplement 2A ) . Las17-GFP , however , must be mostly functional because it does not lead to the strong block of endocytosis observed with LAS17 deletion ( Sun et al . , 2006 ) . In the second step , we aligned the average trajectories of all proteins in space and time to a common reference . We chose Abp1 as a reference protein because of its abundance at the endocytic site , which leads to a strong fluorescence signal , and its highly regular dynamic behavior , which is directed perpendicular to the cell surface thereby defining the direction of the invagination movement ( Figure 1—figure supplement 2B , C ) . We imaged strains coexpressing each GFP-tagged protein of interest together with Abp1-mCherry . The two colors were imaged simultaneously . Chromatic aberration was measured using multicolored fluorescent microbeads and corrected for . For each protein pair we recorded 50–350 endocytic events from which we extracted the centroid positions and fluorescence intensities of the two proteins as paired trajectories ( Figure 1D , Table 1 ) . From these paired trajectories we estimated the rotation , translation and time shift that optimally align the average trajectory of the protein of interest to the average trajectory of Abp1 ( Figure 1E; see ‘Materials and methods’ ) . We thus produced a dataset in which all average trajectories are aligned to Abp1 . The axis of the invagination might not be perfectly aligned with the focal plane and would therefore be imaged as a projection , which leads to an underestimate of the true centroid movement along the invagination axis . Considering the distribution of the angles of Abp1 trajectories ( Figure 1—figure supplement 2C ) , the spherical geometry of the cells and the depth of field ( Figure 1—figure supplement 2D ) we estimated that these projection effects would lead maximally to about 10% underestimation of the true movement along the invagination axis . To measure the reproducibility of the alignment procedure we repeated multiple times the acquisition of paired trajectories for GFP-Sla2 and Rvs167-GFP expressed together with Abp1-mCherry . We also used paired trajectories of Abp1 derived from a strain expressing Abp1 tagged C-terminally with both mCherry and GFP ( see ‘Materials and methods’ ) . The standard deviations in space and time of the repeated alignments were 2 . 8 nm and 0 . 1 s for Abp1 , 5 . 5 nm and 0 . 3 s for Rvs167 and 12 . 9 nm and 0 . 6 s for Sla2 ( Figure 1—figure supplement 2E ) . The accuracies of the alignments correlate with the signal intensities of these different proteins . To further test the alignment procedure we created virtual ‘ground truth trajectories’ from which we generated sets of virtual trajectories by adding different levels of random noise . We first tested the accuracy of the two color alignment procedure using sets of virtual paired trajectories . With experimentally relevant noise levels the trajectories aligned within 3 nm from the ground truth trajectory ( Figure 1—figure supplement 3 ) . To test the effect of misalignment between the color channels we took paired trajectories and shifted them artificially in relation to each other to different extents . Due to random orientations of the trajectories the color shifts average out and the increase in the shift is only manifested as increased uncertainty of track positions . With realistic color shifts ( up to 50 nm ) the average trajectories are only about 3 nm from the ground truth trajectory ( Figure 1—figure supplement 4 ) . Finally , we tested the full two-step alignment procedure using virtual trajectories . This test demonstrated that the alignment procedure robustly generates representative average trajectories from noisy data ( Figure 1—figure supplement 5 ) . In summary , with the average trajectories we can resolve the position and movement of different endocytic components in relation to each other along the direction of invagination with a temporal precision below 1 s and with a spatial precision of ∼10 nm . In addition , we could complement the tracking data with time-resolved estimates of the numbers of molecules present at the endocytic site . The endocytic protein machinery progressively changes the shape of the plasma membrane , ultimately leading to vesicle budding . To provide a framework for interpreting the tracking data , we aligned the trajectories with time-resolved average membrane shapes obtained previously by CLEM ( Kukulski et al . , 2012 ) , starting from a flat membrane and ending with a vesicle ( Figure 2 ) . 10 . 7554/eLife . 04535 . 010Figure 2 . Alignment of trajectories and interpretation . ( A ) Abp1-GFP , GFP-Sla2 and Rvs167-GFP average trajectories aligned with each other . ( B ) The average number of molecules recruited at the endocytic locus varies from protein to protein , ranging from ∼40 molecules at the peak of GFP-Sla2 to ∼410 molecules at the peak of Abp1-GFP . ( C ) The lifetime of Abp1-GFP and Rvs167-GFP , as well as the centroid position of GFP-Sla2 are used to align in space and time the average plasma membrane profiles obtained by correlative light and electron microscopy ( Kukulski et al . , 2012 ) . The centers of mass of the Sla2 model are marked by an ‘X’ . The grey vertical bar represents the estimated time window during which scission happens ( Kukulski et al . , 2012 ) . The shading represents the confidence interval . The trajectories plotted are listed in Supplementary file 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 04535 . 010 We used Sla2 , Rvs167 and Abp1 proteins as spatial and temporal landmarks to align the membrane shapes from the CLEM study to our tracking data . These three proteins represent different endocytic substructures: the coat ( Sla2 ) , the invagination neck ( Rvs167 ) and the actin cytoskeleton ( Abp1 ) ( Kaksonen et al . , 2005; Idrissi et al . , 2008 ) . We generated aligned average trajectories and measured the average number of molecules for the proteins tagged N-terminally ( GFP-Sla2 ) or C-terminally ( Rvs167-GFP and Abp1-GFP; Figure 2A , B ) . We first defined a common frame by setting the initial position of the Sla2 centroid to coincide with the position of the flat plasma membrane . In GFP-Sla2 the GFP molecule is adjacent to Sla2's lipid-binding ANTH domain . The binding of Sla2's ANTH domain to PIP2 at the plasma membrane is essential for productive endocytosis ( Sun et al . , 2005 ) . We could therefore assume that before membrane bending starts ( first time point in Figure 2A , C ) the centroid of GFP-Sla2 is likely to be within few nm from the plasma membrane surface , whose position we took as the origin of the position axis . We then calculated the positions of the center of mass of Sla2 molecules on the shapes of the different membrane intermediates taking into account that Sla2 molecules cover a ∼30–40 nm long region at the tip of the invagination ( Idrissi et al . , 2012 ) . We used these positions to shift the membrane intermediates in time so that they coincided with the centroids in the average trajectory of GFP-Sla2 ( Figure 2A , C ) . Our alignment agreed well with the time resolved data from CLEM study: the start of Abp1 assembly coincides with initial membrane bending , and the assembly of Rvs167 starts when the invaginations are ∼50 nm long ( Kukulski et al . , 2012 ) ( Figure 2C ) . The CLEM data also indicated that on average the vesicle scission takes place when ∼59% of the Rvs167-GFP patch lifetime has passed ( Kukulski et al . , 2012 ) . This time point coincides with the peak number of Rvs167-GFP molecules ( Figure 2B ) . We therefore used the time at which the number of Rvs167-GFP molecules peaks as an estimate of the scission time and we defined it as the origin of the time axis ( Figure 2A , B ) . To gain insights into the organization of the endocytic coat we focused on three coat-associated proteins: Sla2 , Sla1 and End3 . Sla2 is critical for membrane-actin coupling during vesicle budding ( Sun et al . , 2005; Boettner et al . , 2011; Skruzny et al . , 2012 ) , whereas Sla1 and End3 are involved in the regulation of the initiation of actin polymerization ( Tang et al . , 2000; Rodal et al . , 2003; Kaksonen et al . , 2005 ) . Sla2 is composed of the N-terminal lipid-binding ANTH domain and a C-terminal actin-binding domain , which are separated by a long coiled-coil region . In Hip1R , the mammalian homolog of Sla2 , the coiled-coil region has been measured to be ∼40 nm long ( Engqvist-Goldstein et al . , 2001 ) . We imaged cells expressing C-terminally GFP-tagged Sla2 ( Sla2-GFP ) . Both GFP-Sla2 and Sla2-GFP average trajectories exhibited an initial motionless phase followed by inward movement . However , the Sla2-GFP trajectory was 29 ± 5 nm ( mean ± SE ) distant from the GFP-Sla2 trajectory during the initial non-motile phase ( Figure 3A ) . 10 . 7554/eLife . 04535 . 011Figure 3 . Coat dynamics and organization . ( A ) Left panel: the inward movement of Sla2 coat protein , tagged at its N- or C-terminus ( GFP-Sla2 and Sla2-GFP respectively ) . Right panel: our model of Sla2 organization at the tip of the plasma membrane invagination . The centers of mass of the Sla2 model are marked by an ‘X’ ( N-terminus ) and an open circle ( C-terminus ) . ( B ) Left panel: the inward movement of Sla1-GFP and End3-GFP . GFP-Sla2 ( dashed line ) is plotted for comparison . Right panel: our model of Sla1 and End3 organization at the outer rim of the coat . An open circle and an open diamond mark the centers of mass of the Sla1 and End3 models respectively . The center of mass of Sla2 N-terminus is marked by an ‘X’ and it is plotted for comparison . ( C ) The average number of molecules for GFP-Sla2 , Sla2-GFP , Sla1-GFP and End3-GFP . ( D ) Sla1 ring structures imaged at endocytic sites , using superresolution microscopy . ( E ) Sla1 ring structures imaged at endocytic sites , using superresolution microscopy , in yeast cells treated with LatA . The shading of the trajectories represents the confidence interval . The plotted trajectories are listed in Supplementary file 1 . Scale bar is 100 nm long . See also Figure 3—figure supplement 1 and Figure 3—figure supplement 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 04535 . 01110 . 7554/eLife . 04535 . 012Figure 3—figure supplement 1 . Organization of the coat and experimental control for the two color alignment . ( A ) Sla2 N- and C-termini were tagged simultaneously with green ( GFP ) and red ( RFP ) fluorescent proteins respectively ( GFP-Sla2-RFP ) . The estimated distance d between the GFP and RFP centroids gives an estimate of Sla2 length . ( B ) The orientation of GFP-Sla2-RFP molecules at endocytic sites in LatA treated cells . The histogram shows the angles between the vectors determined by the GFP and mCherry centroid pairs and their closest tangent to the plasma membrane . Mean angle is given as mean ± SEM . ( C ) The distribution of the distances s between the centroids of the N- and C-terminal tags of GFP-Sla2-RFP , measured from cells treated with LatA . The red curve marks the non-gaussian distribution that the distances follow and which is used to determine the distance d and σ ( Stirling Churchman et al . , 2006 ) . ( D ) The distribution of the distances between centroids of TetraSpecks emitting on both GFP and mCherry channels . The red curve marks the non-gaussian distribution that the distances follow and which is used to determine d and σ ( Stirling Churchman et al . , 2006 ) . ( E ) As an additional control the C-terminus of Sla2 was tagged with both red and green fluorescent protein in tandem ( Sla2-RFP-GFP ) . ( F ) The distribution of the distances between Sla2-RFP-GFP centroids . The red curve marks the non-gaussian distribution that the distances follow and which is used to determine d and σ ( Stirling Churchman et al . , 2006 ) . ( G ) The distance between Sla2-GFP and GFP-Sla2 average trajectories and the distance between End3-GFP and Sla1-GFP average trajectories over time . The error bars represent the confidence interval . DOI: http://dx . doi . org/10 . 7554/eLife . 04535 . 01210 . 7554/eLife . 04535 . 013Figure 3—figure supplement 2 . Imaging of Sla1 assemblies by localization microscopy . ( A ) Overview of a yeast cell expressing Sla1-SNAP imaged using localization microscopy ( left ) and conventional , diffraction-limited wide-field microscopy ( right ) . The dashed lines represent the cell outline . ( B–D ) We observed structural heterogeneity among the Sla1 assemblies . In addition to clear ring-shaped structures ( B ) , a subset of Sla1 sites was found in more diverse and irregular shapes that we classified as possible rings ( C ) and not rings ( D ) ( See Materials and methods ) . ( E–G ) Yeast cells expressing Sla1-SNAP were treated with LatA and imaged using localization microscopy . Similar distribution of clear ring-shapes ( E ) , possible rings ( F ) and not rings ( G ) was observed . ( H ) Yeast cells expressing Sla1-SNAP and treated with LatA were imaged on the equatorial plane of cells to obtain a side view of Sla1 structures . Scale bars are 1 µm ( A ) and 100 nm ( B–H ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04535 . 013 To confirm the spatial separation between GFP-Sla2 and Sla2-GFP we used a different imaging strategy based on double labeling: we tagged the Sla2 molecule at its N-terminus with GFP and at its C-terminus with mCherry ( GFP-Sla2-mCherry; Figure 3—figure supplement 1A ) . We immobilized the Sla2 patches by treating the cells with latrunculin A , which inhibits actin polymerization and prevents membrane invagination at endocytic sites ( Kukulski et al . , 2012 ) , and acquired still images of both GFP and mCherry signals at equatorial planes of the cells . The GFP-Sla2-mCherry molecules were oriented perpendicularly to the plasma membrane ( Figure 3—figure supplement 1B ) and the distance between the centroids of the GFP and mCherry tags was 33 ± 3 . 4 nm ( mean ± SE ) ( Figure 3—figure supplement 1C–F , ‘Materials and methods’ ) . This compares well with the average distance between GFP-Sla2 and Sla2-GFP average trajectories during the initial motionless phase . These results indicate that Sla2 molecules are oriented within the coat so that the PIP2 binding N-terminal domain is at the membrane and the actin-binding C-terminal domain projects into the cytoplasm . When the membrane invagination starts , the GFP-Sla2 and Sla2-GFP centroids start moving inward and the distance between them becomes shorter ( Figure 3A , Figure 3—figure supplement 1G ) . A simple explanation for this convergence is that the Sla2 molecules maintain a constant orientation with respect to the membrane . Therefore , during invagination the Sla2 molecules would reorient to accommodate membrane bending and the center of mass positions of their N and C-termini would move closer to each other ( Figure 3A , right panel , time points −7 , −6 and −4 s ) . Alternatively , a conformational change of Sla2 could also contribute to the convergence of the trajectories by bringing the N- and C-termini closer to each other . We studied Sla1 and End3 proteins with C-terminal GFP-tags . Sla1 and End3 are multi-domain proteins but the locations of their N- and C-termini in the tertiary structure are less well characterized than in Sla2 . The average trajectories of Sla1-GFP and End3-GFP had very similar shapes , but were separated by about ∼10 nm or more ( Figure 3B and Figure 3—figure supplement 1G ) . The number of Sla2 molecules remained relatively constant at ∼40 molecules during vesicle budding and started to drop ∼2–3 s before scission ( Figure 3C ) . Sla1 and End3 molecules peaked at ∼90 and ∼60 molecules respectively and started disassembling already during invagination ( Figure 3C ) . The overall dynamic behaviors of the Sla1 and End3 centroids appeared similar to that of Sla2 . However , the Sla2 centroid started moving ∼2 s before the Sla1 and End3 centroids and moved past the Sla1 centroid during invagination ( Figure 3B , left panel ) . This suggests that Sla2 , Sla1 and End3 are not distributed similarly within the coat . The observed trajectories could result from Sla2 being located on the invagination tip and Sla1 and End3 being located at the rim of the coat ( Figure 3B , right panel ) . With such distributions , when the membrane starts bending , Sla2 molecules would start moving first , followed slightly later by Sla1 and End3 molecules . The average trajectories describe the center of mass position of the tagged protein molecules , but the individual proteins could be distributed in various ways around the center of mass . We used superresolution microscopy ( Betzig et al . , 2006; Hess et al . , 2006; Rust et al . , 2006 ) to resolve the distribution of individual Sla1 proteins at the endocytic site: We fixed cells expressing Sla1 fused to a C-terminal SNAP tag and imaged structures on the bottom membrane of the cell close to the coverslip surface . With an average localization precision of approximately 10 nm , we could directly observe that Sla1 was localized in ring shapes ( Figure 3D and Figure 3—figure supplement 2A , B , E ) . We also observed some less defined shapes , which likely correspond to states either during coat assembly or disassembly ( Figure 3—figure supplement 2C , D , F , G ) . To exclude the possibility that a ring was observed because we were imaging a curved membrane , we treated Sla1-SNAP expressing cells with Latrunculin A , which inhibits actin polymerization and prevents membrane bending ( Kukulski et al . , 2012 ) . Again , we observed ring-shaped Sla1 structures ( Figure 3E ) even though the plasma membrane remained flat ( Figure 3—figure supplement 2H ) . These results show that Sla1 is organized in a circular pattern at the endocytic site prior to membrane invagination . The BAR domain proteins Rvs161 and Rvs167 form stable heterodimers , which localize transiently to the neck region of the endocytic invagination where they are thought to regulate vesicle scission ( Kaksonen et al . , 2005; Ren et al . , 2006; Idrissi et al . , 2008; Youn et al . , 2010; Kishimoto et al . , 2011; Kukulski et al . , 2012 ) . We thus aimed to better characterize the dynamics of the Rvs proteins during vesicle budding . The average trajectory of Rvs167-GFP showed three different phases ( Figure 4A ) . During the first phase the centroid moved inward linearly less than 15 nm while the Rvs167-GFP molecules assembled at a constant rate of ∼40 molecules/s ( Figure 4A , B ) . In the next phase the centroid moved rapidly for ∼100 nm ( Figure 4A ) in less than a second . This rapid centroid movement coincided with a transition to disassembly and with the scission of the vesicle ( Figure 4B ) . In the last phase the centroid continues to move inward , but with a reduced rate ( Figure 4A ) . The fast centroid movement of the Rvs proteins has been described before , but its mechanism remained unknown ( Kaksonen et al . , 2005 ) . 10 . 7554/eLife . 04535 . 014Figure 4 . BAR protein dynamics . ( A ) The inward movement of Rvs167-GFP . ( B ) The average number of molecules of Rvs167-GFP . ( C ) Our model of Rvs coverage of the plasma membrane invagination during the invagination growth . When scission happens , Rvs molecules are released rapidly and remain in proximity of the vesicle only . The centers of mass of the Rvs model are marked by an ‘X’ . The grey vertical bar represents the estimated time window during which scission happens ( Kukulski et al . , 2012 ) . The shading represents the confidence interval . The plotted trajectories are listed in Supplementary file 1 . See also Figure 4—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 04535 . 01410 . 7554/eLife . 04535 . 015Figure 4—figure supplement 1 . The variability in the BAR protein coverage of the plasma membrane . The points show the upper and lower bound of the Rvs coverage shown in Figure 4C . The error bars show the variability in the calculation of the extent of Rvs coverage of the plasma membrane as a function of the uncertainty in the number of molecules ( Figure 4B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04535 . 015 To understand better how the Rvs molecules are assembled on the membrane we supplemented our tracking data with data from structural studies of BAR domain proteins homologous to Rvs ( Peter et al . , 2004; Mizuno et al . , 2010; Mim et al . , 2012 ) . These studies have shown that BAR domains can form regular oligomeric structures on membrane tubules of varying diameters , giving us an estimate of the density of dimers on the membrane . With our estimate of the numbers of Rvs167 molecules ( Figure 4B ) , such density yields the possible membrane area covered by the Rvs dimers , assuming that all dimers are membrane bound . We then used the Rvs167 average trajectory to locate the area of the membrane profiles covered by Rvs . These simple assumptions allowed us to generate a dynamic model of the membrane area covered by the Rvs molecules ( Figure 4C; see ‘Materials and methods’; see also Arasada and Pollard , 2011 for a similar approach ) . To obtain an estimate of Rvs coverage immediately prior to scission we extrapolated the coordinates of the invagination profile at time 0 from the available membrane profiles . This reconstruction of Rvs localization revealed the dynamics of Rvs assembly and disassembly in relation to membrane invagination and scission ( Figure 4C ) . Rvs proteins start assembling at the midpoint of the endocytic invagination and the Rvs covered membrane area increases linearly and reaches its maximum at about the time of scission . Immediately prior to scission there are enough Rvs molecules to cover the whole tubular part of the invagination from the base of the invagination to the coated tip ( Figure 4C; time point 0 ) . After reaching their peak number , the Rvs molecules disassemble rapidly ( Figure 4B ) and their center of mass moves quickly inward toward the region where the newly formed vesicle is located . Our reconstruction suggests that the rapid movement of the centroid is the result of a rapid disassembly of the Rvs molecules at the neck of the invagination , where scission occurs ( Kukulski et al . , 2012 ) , while some of the Rvs molecules are kept at the vesicle surface where they stay ∼2 s longer . Polymerization of actin filaments at the endocytic site is critical for membrane invagination and vesicle budding ( Merrifield et al . , 2002; Kaksonen et al . , 2003; Boulant et al . , 2011; Idrissi et al . , 2012; Kukulski et al . , 2012 ) . The endocytic actin filaments are nucleated by the Arp2/3 complex , which binds to existing filaments and nucleates new filaments in a branched configuration . Thus , actin filaments are organized as a dense , branched network , which surrounds the endocytic site and the newly formed vesicle ( Mulholland et al . , 1994; Idrissi et al . , 2012; Kukulski et al . , 2012 ) . To gain further insight into the assembly of the endocytic actin patches we tracked Act1 , the yeast actin protein , and Arc18 , a component of the Arp2/3 complex . Arc18 was GFP-tagged C-terminally at its genomic locus . As the GFP fusion of actin is not fully functional we expressed GFP-Act1 in addition to the endogenously expressed untagged Act1 ( Doyle and Botstein , 1996 ) . In actin patches the GFP-Act1 localization was comparable to general actin staining ( Kaksonen et al . , 2003 ) and the lifetime of Abp1-mCherry patches was unaffected ( Figure 1—figure supplement 2A ) . GFP-Act1 is thus likely to be a good marker for actin localization and assembly at the endocytic site . We also tracked Las17 , the yeast homolog of mammalian N-WASP , and Myo5 , a type I myosin , which were both C-terminally GFP-tagged at their genomic loci . Las17 and Myo5 can activate the Arp2/3 complex and are critical for the initiation of actin polymerization and for membrane invagination ( Sun et al . , 2006 ) . The average trajectories of both Act1 and Arc18 followed closely the trajectory of Abp1 ( Figure 5A ) . All three average trajectories started at 30–40 nm above the level of the plasma membrane when Sla2 movement began . Their centroids moved for ∼100 nm until the scission time ( Figure 5A ) . The tight colocalization of the centroids of Act1 , Arc18 and Abp1 suggests that they are homogeneously distributed within the actin network . After scission the centroid trajectories became noisier ( Figure 5A ) . The increased noise probably reflects the free diffusion of the actin-covered vesicle in the cytoplasm ( Berro and Pollard , 2014 ) . 10 . 7554/eLife . 04535 . 016Figure 5 . Actin cytoskeleton dynamics . ( A ) The inward movement of the actin cytoskeleton components GFP-Act1 , Arc18-GFP and Abp1-GFP , together with the nucleation factors Las17-GFP and Myo5-GFP . GFP-Sla2 ( dashed line ) is plotted for comparison . ( B–C ) The average number of molecules of Las17-GFP , GFP-Act1 , Myo5-GFP , Abp1-GFP and Arc18-GFP . The grey vertical bar represents the estimated time window during which scission happens ( Kukulski et al . , 2012 ) . The shading represents the confidence interval . The plotted trajectories are listed in Supplementary file 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 04535 . 016 The trajectories of individual Las17-GFP and Myo5-GFP patches did not exhibit any clear inward movement . To align these trajectories with respect to the invagination axis we thus used two-color imaging of Las17-GFP or Myo5-GFP with Abp1-mCherry , and we aligned the Las17-GFP and Myo5-GFP trajectories based on the alignment of the corresponding Abp1-mCherry trajectories to Abp1-GFP average trajectory ( see ‘Materials and methods’ ) . At the beginning of the membrane invagination Las17 and Myo5 average trajectories are close to Act1 , Arc18 and Abp1 ( Figure 5A ) . The Myo5 trajectory remained almost stationary during the invagination of the plasma membrane , whereas the Las17 trajectory moved inward , but much less than Abp1 , Arc18 and Act1 trajectories . The amount of Las17 molecules reached a maximum of ∼45 molecules . At ∼2 s before scission the amount of Las17 molecules started decreasing while the number of Act1 molecules reached a plateau which was maintained until the scission event ( Figure 5B ) . Also the amount of Arc18 and Myo5 molecules peaked at ∼2 s before scission . However , the amount of Myo5 molecules started decreasing before scission , while Arc18 , similarly to Act1 , maintained a plateau level until scission . The amount of Abp1 molecules peaked at the scission time and then rapidly declined ( Figure 5C ) . We measured by quantitative Western blotting that GFP-Act1 represented 9 ± 1% of the total cellular actin ( See ‘Materials and methods’ ) . If we assume that GFP-Act1 is recruited to endocytic sites with the same efficiency as the untagged actin , we can estimate that the peak number of actin molecules at the endocytic site is ∼3000 . These results document the coupling between the assembly dynamics of the actin cytoskeleton and the changes in membrane shape . However , the homogeneous distribution of the actin cytoskeleton components across the actin network did not allow us to resolve its organization . To understand how actin polymerization is distributed over the network , we next combined centroid tracking with local photobleaching . The mechanism by which the polymerizing actin filaments contribute to membrane invagination is poorly understood . Different models have been suggested , but there is no conclusive data in favor of any of them ( Kaksonen et al . , 2003; Merrifield , 2004; Takenawa and Suetsugu , 2007; Suetsugu , 2009; Collins et al . , 2011; Idrissi et al . , 2012; Mooren et al . , 2012 ) . Figure 6A summarizes three possible models for the organization of actin filaments at the endocytic site . In the first model the polymerizing ends of actin filaments are growing against the coat , thereby pushing it inward . In the second model filaments are growing against the plasma membrane at the base of the invagination , and the older filaments are connected to the coat to mediate the pushing force . In the third model actin polymerization is not preferentially oriented . In all models , the centroid of labeled actin would move similarly inward during endocytosis and thus could not be used to discriminate between the models . However , the models can be distinguished if the centroid is tracked after local photobleaching , because the location of the centroid is then determined by the fluorescence of the newly polymerized filaments . Model 1 predicts that after photobleaching the centroid would jump inward , away from the cell surface . Model 2 predicts a jump in the opposite direction , toward the cell surface , while model 3 predicts no jump . The centroid positions at later time points should be parallel to the reference trajectory , and be shifted by the distance of the post-bleach jump ( Figure 6A ) . 10 . 7554/eLife . 04535 . 017Figure 6 . Region of assembly of actin filaments . ( A ) A schematic cartoon to show how local photobleaching would affect the centroid position of a fluorescent actin patch , given three possible scenarios for the nucleation of new actin filaments ( in red ) at the endocytic locus: ( 1 ) New actin filaments are nucleated in the proximity of the coat; photobleaching would shift the centroid away from the plasma membrane . ( 2 ) New filaments are nucleated at the base of the plasma membrane invagination; photobleaching would shift the centroid toward the plasma membrane . ( 3 ) There is no preferential direction for actin nucleation; photobleaching would not affect the centroid position . ( B ) Local photobleaching of an endocytic event in a cell expressing GFP-Act1 . Scale bar: 1 μm . See also Video 1 . ( C–E ) Centroid positions of endocytic patches in yeast cells expressing GFP-Act1 ( C ) , Arc18-GFP ( D ) or Abp1-GFP ( E ) before and after photobleaching . Error bars represent the Standard Deviation . Plots show the respective average trajectories for comparison . ( F ) Left panel: The locations of the first post-bleach centroids of Act1-GFP ( red triangles ) and Arc18-GFP ( green circles ) aligned relative to the unbleached average trajectories . Nucleation promoting factors , Myo5-GFP and Las17-GFP , localize to the same region where actin monomers are added . Average trajectories of GFP-Act1 and Arc18-GFP are plotted for comparison . Right panel: Our model for the region of actin polymerization . The dashed boxes highlight the region where actin and Arp2/3 complexes are recruited . The centers and the heights of the boxes correspond respectively to the average and to the standard deviation of the centroid positions of GFP-Act1 and Arc18-GFP after photobleaching . The cartoon highlights the regions where the new Arp2/3 complexes and actin molecules are recruited . The shading of the trajectories represents the confidence interval . The plotted trajectories are listed in Supplementary file 1 . See also Figure 6—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 04535 . 01710 . 7554/eLife . 04535 . 018Figure 6—figure supplement 1 . The position of Abp1 fluorescence recovery after photobleaching . ( A ) The quantification of the jump in GFP-Act1 patches photobleached 3–4 s after the appearance of the patch ( data showed in Figure 6C ) , compared with the jump in GFP-Act1 patches that where next to the region photobleached and thus were only partially affected by the photobleaching . Error bars represent the SEM . ( B ) The centroid positions of endocytic fluorescent patches recovering after photobleaching ( FRAP ) , performed at different time points during the plasma membrane invagination process , in yeast cells expressing Abp1-GFP . The shading represents the confidence interval . GFP-Act1 FRAP are the data showed in Figure 6F and serve as a comparison . DOI: http://dx . doi . org/10 . 7554/eLife . 04535 . 01810 . 7554/eLife . 04535 . 019Video 1 . Photobleaching experiment . Local photobleaching of an endocytic event in a cell expressing GFP-Act1 . The video plays in real time . Scale bar is 2 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 04535 . 019 We used a laser beam focused on a ∼0 . 5 µm diameter spot to photobleach an individual fluorescence patch at the cell equator ( Figure 6B ) and we tracked its centroid position before and after photobleaching . Multiple trajectories were averaged and plotted with the average trajectory of the unbleached patch , as a comparison . We first photobleached GFP-Act1 patches ∼3–4 s after their appearance . After photobleaching the reappearing centroids were located closer to the plasma membrane than the centroids at the corresponding time point in the unbleached average trajectory ( Figure 6C ) . Neighbor trajectories to the photobleaching site , which were only partially affected by photobleaching , did not show any jump ( Figure 6—figure supplement 1A ) . We next photobleached Arc18-GFP and observed a similar , although slightly smaller , shift of the centroid position toward the plasma membrane ( Figure 6D ) . These results suggest that actin filament nucleation and polymerization take place at the base of the invagination , in agreement with model 2 . In addition , we photobleached Abp1-GFP , a proposed inhibitor of actin filament nucleation ( D'Agostino and Goode , 2005 ) . However , the trajectories of the Abp1-GFP centroid with or without photobleaching did not differ significantly ( Figure 6E ) suggesting that Abp1 molecules are recruited throughout the actin network unlike actin and the Arp2/3 complex . We next performed similar photobleaching experiments at different time points during endocytosis , and recorded the centroid position right after local photobleaching ( Figure 6F ) . At all time , GFP-Act1 and Arc18-GFP were added at the base of the invagination , but the exact localizations of the two proteins differed: Act1 was added on average at 27 ± 18 nm ( mean ± SD , n = 19 ) and Arc18 at 41 ± 15 nm ( mean ± SD , n = 17 ) above the plasma membrane surface ( p value = 0 . 02 ) . We note that this is also where the Arp2/3 activators Las17 and Myo5 are located , according to the average centroid trajectories ( Figure 6F ) . The post bleach centroid position of Abp1 followed closely the Abp1 reference trajectory ( Figure 6—figure supplement 1B ) . These results suggest that both actin filament nucleation and polymerization take place at the base of the endocytic invagination . The Arp2/3 complex is assembled slightly higher above the membrane plane compared to actin , which is consistent with the dendritic nucleation model ( Pollard and Borisy , 2003 ) where the actin plus ends are oriented toward the plasma membrane ( Figure 6F ) . The putative inhibitor of actin nucleation , Abp1 , would be distributed uniformly within the network .
The organization of clathrin molecules in the endocytic coat is extensively documented ( Brodsky et al . , 2001; Fotin et al . , 2004; Kirchhausen , 2009 ) , but the organization of the numerous coat-associated adaptor and scaffold proteins is less well understood . However , several studies indicate that many of these proteins have distinct distributions within the coat ( Tebar et al . , 1996; Saffarian and Kirchhausen , 2008; Boettner et al . , 2011 ) . We showed that three coat-associated proteins exhibit specific orientation and distribution within the endocytic coat ( Figure 3 ) . Sla2 molecules are oriented so that the N-terminal membrane binding-domain is at the plasma membrane , while the C-terminal actin-binding domain is ∼30 nm away from the plasma membrane . The coiled-coil region separating the terminal domains contains a clathrin-binding motif ( Engqvist-Goldstein et al . , 2001; Boettner et al . , 2011 ) . Therefore , Sla2 molecules could span the clathrin lattice , which is about 22 nm from the membrane surface ( Vigers et al . , 1986 ) , and interact simultaneously with the membrane , with clathrin , and with actin filaments on the cytoplasmic side of the clathrin lattice . Sla1 and End3 , together with a third protein Pan1 , form a complex , which is implicated in the regulation of actin assembly ( Holtzman et al . , 1993; Bénédetti et al . , 1994; Tang et al . , 2000 ) . Sla1 is localized in a ring of ∼50 nm in diameter , which can fit well around the tip of the endocytic membrane invagination of ∼30 nm in diameter ( Idrissi et al . , 2008; Kukulski et al . , 2012 ) . As Sla1 and End3 are late-assembling coat proteins ( Kaksonen et al . , 2005; Newpher et al . , 2005 ) , they could assemble at the rim of a pre-existing early coat . The disassembly of Sla1 and End3 begins after membrane invagination has started , but several seconds before scission , suggesting that these proteins may be important during early stages of vesicle budding . Interestingly , Sla2 and Pan1 , have been suggested to capture short cytosolic actin filaments that could serve as ‘mother filaments’ to recruit the Arp2/3 complex and to initiate the assembly of the actin network ( Chen and Pollard , 2013 ) . Taken together , our data demonstrate that the endocytic coat exhibits a highly organized and dynamic molecular architecture during vesicle budding . BAR domain proteins can sense membrane curvature and they can actively induce membrane curvature at high concentration ( Peter et al . , 2004; Mizuno et al . , 2010; Mim et al . , 2012 ) . In yeast , the heterodimeric BAR domain proteins Rvs161/167 assemble transiently at the endocytic site during the formation of the membrane invagination ( Kaksonen et al . , 2005; Idrissi et al . , 2008; Kukulski et al . , 2012 ) to regulate the invagination or scission steps ( Kaksonen et al . , 2005; Youn et al . , 2010; Kishimoto et al . , 2011; Boucrot et al . , 2012; Kukulski et al . , 2012 ) . However , the exact mechanism of action of Rvs is unknown . Our results suggest that there are enough Rvs molecules to form a dense BAR domain lattice covering the whole tubular part of the invagination below the coated tip ( Figure 4 ) . At this density the Rvs proteins could exert significant bending forces on the membrane ( Sorre et al . , 2012 ) . This Rvs lattice could therefore support the membrane tubule and prevent premature fission ( Kukulski et al . , 2012 ) . In addition , an Rvs lattice could act as a lipid diffusion barrier that could form a lipid phase boundary to promote vesicle scission ( Liu et al . , 2006; Zhao et al . , 2013 ) . Furthermore , our model explains the rapid movement of the Rvs167-GFP centroid upon scission: due to their affinity for curved membranes , the Rvs molecules are likely to first disassemble at the tubular part of the invagination that collapses after scission , while some of the Rvs molecules may remain bound to the membrane that forms the vesicle . Therefore , the center of mass of Rvs molecules would rapidly shift due to localized disassembly without actual movement of the Rvs lattice . The exact timing of the Rvs disassembly , in respect to the scission event , remains an interesting open question . Currently , our data does not have the time resolution to discern whether the initiation of Rvs disassembly precedes scission or follows it . If Rvs disassembly follows scission it could be a passive response to the collapse of the membrane tubule . Alternatively , if Rvs disassembly precedes scission it might destabilize the membrane tubule and thereby regulate the timing of scission . In either case , the extensive Rvs lattice could support the membrane tubule and thereby promote its extension until the scission . Actin nucleation and polymerization are precisely coordinated with membrane shape changes . Our previous CLEM results suggested that the initiation of membrane bending coincides with actin polymerization ( Kukulski et al . , 2012 ) . Another EM study , however , suggested that the membrane bending begins already ∼15 s prior to actin polymerization ( Idrissi et al . , 2012 ) . The average trajectory of the coat proteins Sla2 shows directed inward movement only after actin polymerization has started ( Figure 1 ) supporting the idea that the main membrane bending phase starts only after actin polymerization . However , a very shallow initial membrane curvature might be unresolvable by the tracking approach . The data about numbers of molecules allow us to estimate some key parameters of the endocytic actin network . If each Arp2/3 complex nucleates an actin filament , new filaments are generated at a rate of ∼30 filaments per second , reaching a maximum number of ∼200 filaments . With the estimated maximum of ∼3000 actin molecules the average filament length would be ∼40 nm . The actin polymer , the Arp2/3 complex , and the Arp2/3 activators , Myo5 and Las17 , reach the maximum level of molecules about 2 s before vesicle scission . Actin and the Arp2/3 complex maintain their levels of molecules until scission , whereas Las17 and Myo5 start to disassemble right after reaching their peak levels . These data suggest that the rate of nucleation of new actin filaments decreases before vesicle scission occurs . The initiation of actin filament nucleation has been studied extensively ( Mooren et al . , 2012 ) , but the mechanisms that terminate nucleation are not well understood . The termination of nucleation could involve feedback from membrane shape or tension . Different models have been proposed for the organization of actin filaments at the endocytic site ( Takenawa and Suetsugu , 2007; Mooren et al . , 2012 ) . However , conclusive data about the location of nucleation and polymerization during endocytosis has not been available . We showed here that actin filaments are nucleated and polymerize at the base of the invagination , where the Arp2/3 activators Las17 and Myo5 are also localized ( Figure 6 ) . Capping proteins , which bind to the growing ends of actin filaments and stop polymerization ( Mooren et al . , 2012 ) , are likely important for restricting actin polymerization to the base of the invagination . By combining all our data we built a comprehensive model that reveals the dynamic interplay of the functional modules of the endocytic machinery during the final ∼10 s of endocytosis ( Figure 7 ) . During this time window the flat plasma membrane is reshaped into an endocytic vesicle . The coat , which is initially assembled on flat membrane , starts internalizing and reorganizing when actin polymerization begins . Actin polymerization is initiated close to the plasma membrane where Las17 is located and the region of polymerization remains at the base of the invagination throughout vesicle budding . When the invagination has formed , actin is likely polymerizing in a circular region surrounding the invagination . Initially , actin polymerization might also be triggered in a circular region at the rim of the coat , where Sla1 , a regulator of actin nucleation , is also located . Continued filament nucleation and polymerization at the plasma membrane would push the older filaments into the cell generating a flow of actin filaments . The organization of the endocytic actin network bears striking similarity to the organization of actin at the leading edge of migrating animal cells where actin filaments nucleated by the Arp2/3 complex polymerize and push against the plasma membrane ( Pollard and Borisy , 2003 ) . This polymerization pushes the leading edge of the cell forward , but also results in a retrograde flow of actin filaments into the cell . The actin binding domains of Sla2 are ideally positioned at the surface of the endocytic coat to capture the actin filaments that are moving away from the plasma membrane . We suggest that the actin scaffold is rigid enough to transmit the forces produced by the polymerization at the plasma membrane so that Sla2 can , together with Ent1 ( Skruzny et al . , 2012 ) , transmit the force from the actin polymerization to the membrane . Detailed quantitative analyses of assembly and disassembly of the endocytic machinery , especially the actin cytoskeletal proteins , have been previously performed on fission yeast Schizosaccharomyces pombe ( Wu and Pollard , 2005; Berro et al . , 2010; Sirotkin et al . , 2010; Arasada and Pollard , 2011; Chen and Pollard , 2013 ) . Comparing the S . pombe and S . cerevisiae data shows that the overall assembly dynamics and the stoichiometries of the endocytic proteins are highly similar in these very distantly related fungal species . For example , the numbers of the S . pombe homologs of Las17 and Myo3/5 also peak briefly before the peak of actin molecules ( Berro et al . , 2010; Sirotkin et al . , 2010 ) , suggesting that the assembly dynamics of the actin cytoskeleton are functionally optimized . However , there are also some clear differences: in S . pombe the Las17 homolog Wsp1 is recruited together with myosin just prior to actin polymerization ( Berro et al . , 2010; Sirotkin et al . , 2010; Arasada and Pollard , 2011 ) whereas in S . cerevisiae Las17 localizes >10 s before myosins are recruited and actin polymerization starts ( Kaksonen et al . , 2003; Figure 5 ) . Analyzing these evolutionary differences in detail could reveal mechanistic insights into the regulation of the endocytic assembly . Rvs assembly and disassembly are tightly coupled to the membrane shape changes ( Figure 4 ) . Rvs molecules may get recruited by a specific membrane curvature at the invagination , but protein–protein interactions may also have a role in the recruitment ( Ren et al . , 2006 ) . Just before scission , the membrane invagination is densely covered by Rvs molecules . Scission occurs within the Rvs covered region at ∼1/3 of the invagination length ( Kukulski et al . , 2012 ) . However , the exact molecular mechanism of vesicle scission is still not understood . Neither Rvs , nor the yeast homolog of dynamin Vps1 , are essential for scission , but they may regulate its timing or location ( Kishimoto et al . , 2011; Kukulski et al . , 2012; Smaczynska-de Rooij et al . , 2012 ) . Interestingly , the location of actin polymerization and Myo5 corresponds closely to the site of scission , therefore polymerization and motor activities could have a role in scission ( Jonsdottir and Li , 2004 ) . After scission the Rvs molecules that remain on the newly formed vesicle might have additional roles , for example , in uncoating as shown for the mammalian homolog endophilin ( Milosevic et al . , 2011 ) .
Yeast strains ( Table 2 ) were generated by homologous recombination of the target genes with PCR cassettes . C-terminal tagging was performed using plasmids pFA6a-EGFP-His3MX6 , pFA6a-mCherry-KanMX4 and pYM12-PKS134 for monomeric GFP ( myEGFP ) . pMaM17 ( Khmelinskii et al . , 2012 ) was used for the tandem tag of mCherry and sfGFP . pMK03-SNAP-His3MX6 was generated by replacing EGFP with SNAPf ( Sun et al . , 2011 ) in pFA6a-EGFP-His3MX6 . N-terminal tagging was performed using plasmid pMaM173 which was transformed into yeast strains with a Gal L - ISCE1 integration used to loop out the URA marker and the TEF promoter ( Khmelinskii et al . , 2011 ) . pMaM175 was used to tag Nuf2 C-terminally with sfGFP ( Khmelinskii et al . , 2011 ) . The strains used for protein abundance measurements were confirmed by sequencing of the integrated tags . 10 . 7554/eLife . 04535 . 022Table 2 . Yeast strainsDOI: http://dx . doi . org/10 . 7554/eLife . 04535 . 022Strain #GenotypeMKY0216MATa , his3-∆200 , leu2-3 , 112 , ura3-52 , lys2-801 , NUF2-EGFP::HIS3MX6MKY0217MATα , his3-∆200 , leu2-3 , 112 , ura3-52 , lys2-801 , NUF2-EGFP::HIS3MX6MKY0711MATa , his3-∆200 , leu2-3 , 112 , ura3-52 , lys2-801 , MYO5-EGFP::HIS3MX6 , ABP1-mCherry::kanMX4MKY0822MATa , his3-∆200 , leu2-3 , 112 , ura3-52 , lys2-801 , SLA1-EGFP::HIS3MX6 , ABP1-mCherry::kanMXMKY1304MATa , his3-∆200 , leu2-3 , 112 , ura3-52 , lys2-801 , END3-EGFP::HIS3MX6 , ABP1-mCherry::kanMX4MKY1318MATa , his3-∆200 , leu2-3 , 112 , ura3-52 , lys2-801 , RVS167-EGFP::HIS3MX6 , ABP1-mCherry::kanMX4MKY1368MATα , his3-∆200 , leu2-3 , 112 , ura3-52 , lys2-801 , LAS17-EGFP::HIS3MX6 , ABP1-mCherry::kanMX4MKY2119MATα , his3-∆200 , leu2-3 , 112∆::GalL-ISce1-natNT2 , ura3-52 , lys2-801 , sfGFP-SLA2-mCherry::hphNT1MKY2653MATa , his3-∆200 , leu2-3 , 112 , ura3-52 , lys2-801 with pMK0100[CEN , URA3 GFP-ACT1]MKY2655MATa , his3-∆200 , leu2-3 , 112 , ura3-52 , lys2-801 , ABP1-mCherry::kanMX4 with pMK0100[CEN , URA3 GFP-ACT1]MKY2689MATα , his3-∆200 , leu2-3 , 112∆::GalL-ISce1-natNT2 , ura3-52 , lys2-801 , sfGFP-SLA2MKY2720MATa , his3-∆200 , leu2-3 , 112 , ura3-52 , lys2-801 , ARC18-myEGFP::natNT2 , ABP1-mCherry::kanMXMKY2747MATa , his3-∆200 , leu2-3 , 112 , ura3-52 , lys2-801 , ARC18-myEGFP::natNT2MKY2832MATa , his3-∆200 , leu2-3 , 112 , ura3-52 , lys2-801 , RVS167-EGFP::HIS3MX6MKY2833MATa , his3-∆200 , leu2-3 , 112 , ura3-52 , lys2-801 , SLA1-EGFP::HIS3MX6MKY2834MATα , his3-∆200 , leu2-3 , 112 , ura3-52 , lys2-801 , ABP1-EGFP::HIS3MX6MKY2836MATα , his3-∆200 , leu2-3 , 112∆::GalL-ISce1-natNT2 , ura3-52 , lys2-801 , sfGFP-SLA2 , ABP1-mCherry::kanMXMKY3135MATα , his3-∆200 , leu2-3 , 112∆::GalL-ISce1-natNT2 , ura3-52 , lys2-801 , sfGFP-SLA2 , ABP1-mCherry::kanMX , Sla1-SNAP::HIS3MX6MKY2859MATa , his3-∆200 , leu2-3 , 112 , ura3-52 , lys2-801 , SLA2-EGFP::HIS3MX6MKY2863MATa , his3-∆200 , leu2-3 , 112 , ura3-52 , lys2-801 , CSE4-EGFP::HIS3MX6MKY2864MATa , his3-∆200 , leu2-3 , 112 , ura3-52 , lys2-801 , LAS17-EGFP::HIS3MX6MKY2876MATa , his3-∆200 , leu2-3 , 112 , ura3-52 , lys2-801 , MYO5-EGFP::HIS3MX6MKY2880MATa , his3-∆200 , leu2-3 , 112 , ura3-52 , lys2-801 , ABP1-mCherry-sfGFP::kanMX4MKY2893MATa , his3-∆200 , leu2-3 , 112 , ura3-52 , lys2-801 , END3-EGFP::HIS3MX6MKY2918MATα , his3-∆200 , leu2-3 , 112 , ura3-52 , lys2-801 , SLA2-EGFP::HIS3MX6 , ABP1-mCherry::kanMX4MKY2919MATa , his3-∆200 , leu2-3 , 112∆::GalL-ISce1-natNT2 , ura3-52 , lys2-801 , NUF2-sfGFP::KIURA3MKY2920MATα , his3-∆200 , leu2-3 , 112∆::GalL-ISce1-natNT2 , ura3-52 , lys2-801 , NUF2-sfGFP::KIURA3MKY3136MATa , his3-∆1 , leu2-∆0 , ura3-∆0 , met15-∆0 , LAS17-myEGFP::natNT2MKY3137MATa , his3-∆1 , leu2-∆0 , ura3-∆0 , met15-∆0 , RVS167-myEGFP::natNT2The yeast strains used in this study . Yeast cells were grown to logarithmic phase on SC-Trp medium at 25°C . They were adhered to ConA coated coverslips . Cells were incubated for 10 min at room temperature on the ConA coated coverslip and then washed with SC-Trp medium . Cells were imaged on the coverslip in 40 µl of SC-Trp medium . All samples were imaged at room temperature using an Olympus IX81 wide-field epifluorescence microscope equipped with a 100×/1 . 45 objective . For single channel live cell imaging 488 nm laser light and images were acquired with 80–100 ms exposure time . Emission light was filtered using the GFP-3035C-OMF single-band filter set ( Semrock , Rochester , NY ) . Fluorescence was detected using the Hamamatsu ImagEM EMCCD camera . For two color live cell imaging the samples were excited simultaneously with 488 nm and 561 nm laser light for 250 ms exposure . Excitation light was reflected with a OBS-U-M2TIR 488/561 ( Semrock , Rochester , NY ) dichroic mirror . Emission light was split and filtered with the DUAL-view ( Optical Insights , LLC , Tucson , AZ ) beam splitter . The beam splitter created two separated images , one for each channel , on the Hamamatsu ImagEM EMCCD camera sensor . Photobleaching was performed using a custom built setup with a 488 nm laser , focused on a ∼0 . 5 µm spot . The wide-field epifluorescence microscope setup was controlled by Metamorph 7 . 5 ( Molecular Devices , Sunnyvale , CA ) . Before tracking the endocytic patches and the photobleaching experiments , the extracellular background was subtracted from the images using the Background Subtraction function in ImageJ ( with rolling ball radius equal to 90 pixels , corresponding to 9 µm ) . To correct for photobleaching , each image was then scaled such that the average fluorescence intensity within the cell remains constant in successive frames . To correct for the uneven cytoplasmic background signal at the edge of the cell , we estimated the cytoplasmic contribution by further processing the images with a median filter ( of kernel 6 pixels , corresponding to 0 . 6 µm ) . We then subtracted the cytoplasmic background and tracked endocytic patches using the Particle Tracker plugin in ( Sbalzarini and Koumoutsakos , 2005 ) . Each protein trajectory p is a list of points , pi={pix , piy , pif} , defined from pix and piy , the centroid 2D coordinates in the focal plane , and pif , the corresponding fluorescence intensity , where the index i denotes time . Trajectories were aligned in space and time by a custom made software written in R ( www . CRAN . org ) . An isometric transformation of space T={Tx , Ty , Tθ} is defined asT:{x , y}→{cos ( Tθ ) x−sin ( Tθ ) y+Tx , sin ( Tθ ) x+cos ( Tθ ) y+Ty} . The best alignment between two trajectories , p and q , is computed by minimizing the sum of the squared difference between all overlapping points:{Tbest , τbest}=arg minT , τ ( ∑iwi ( ( qi+τx− ( Tpi ) x ) 2+ ( qi+τy− ( Tpi ) y ) 2 ) ∑iwi ) , with wi=qi+τfpif . In practice , we computed for each possible time shift τ , the best spatial transformation ( Horn , 1987 ) , and selected the best overall result . The weights wi are the product of the fluorescence intensities of the spot pairs at each time point . Being proportional to the cross-correlation in time of the fluorescence intensities , they helped to refine the temporal alignment of the trajectories . To not bias the alignment by the choice of a reference , each trajectory was separately used as a reference to which all the remaining trajectories were aligned . Given n trajectories we thus computed n ( n−1 ) alignments . Each alignment gave us an estimate of the transformation that aligns a trajectory to its reference . We thus estimated the average transformation that aligns all the trajectories together as seen from a common reference point in the field of view . Once aligned , all trajectories were averaged to obtain the average trajectory . The average trajectory of each protein was derived from 50 to 80 individual trajectories ( Figure 1—figure supplement 1 , Table 1 ) . See Source code 1 . The alignment of the trajectories of Abp1 , Arc18 and Act1 was computed using only the trajectory data associated with the invagination dynamics , up to the trajectory peak in fluorescence intensity . The number of molecules over time was obtained after calibrating the fluorescence intensity curve mean integral with the average number of molecules estimated at the endocytic spot ( see ‘Materials and Methods’: calibration of the fluorescence intensity curve of the trajectories with the number of molecules ) . All data presented in this work are listed in the Supplementary file 1 . All the curves in Figure 7 , except Las17 and Myo5 , were smoothened using a Savitzky-Golay filter over 11 time points . Las17 and Myo5 were smoothened using a moving average filter of length 5 . The alignment is performed using pairs of trajectories acquired simultaneously with two labeled proteins: Abp1-mCherry serving as a reference , and a protein of interest tagged with GFP . Simultaneous acquisition of the two colors is made with the DUAL-view beam splitter ( Optical Insights , LLC , Tucson , AZ ) . Image un-splitting and correction of chromatic aberration are done following the DUAL-view recommended procedures . A sample with TetraSpeck microsphere ( 0 . 1 µm , Invitrogen , Carlsbad , CA ) was imaged in both channels and the centroids were identified with Particle Tracker ( Sbalzarini and Koumoutsakos , 2005 ) in ImageJ and then processed in MATLAB ( TheMathworks , Natick , MA ) to generate a spatial warping transformation that was applied on the raw coordinates . The non-linear transformation was generated by the local weighted mean ( lwm ) method of cp2tform ( See Matlab help and references thereafter ) . The average trajectory of a protein , P , is represented by a temporal collection of 6-dimensional vectors , Pi={Pix , Piy , Pif , δix , δiy , δif} with i={1 , … , N} . P contains Px and Py , the average position of the centroid of the fluorescence patch , and the corresponding average fluorescence Pf , the standard error of the means δx and δy , associated with the position coordinates , and δf , the standard error of the mean associated with the fluorescence intensity . The average trajectory of the reference protein , that we name R , follows the same notation . The trajectory pairs are not necessarily defined over the same time interval as the average trajectories P and R . They were thus smoothed with a moving average , to reduce noise , and interpolated with a cubic spline to estimate values at the missing time points . The resulting trajectory , p is a temporal collection of 3-dimensional vectorspi={pix , piy , pif} with i={1 , … , N} , built from the positions , px and py , and the fluorescence intensities pf of each endocytic patch . The notation is similar for the reference trajectory r . r and p are already aligned together because they were acquired simultaneously . First , individual trajectories are aligned to their corresponding average trajectory in time . The resulting lag τp , that aligns in time p and P , is the one that maximizes the cross-correlation of the fluorescence intensities:τp=arg maxτ ( ∑ipi+τfPif ) . The same procedure is used to compute the lag τr between r and R . An isomeric transformation of space T={Tx , Ty , Tθ} is the combination of a translation and a rotationT:{x , y}→{cos ( Tθ ) x−sin ( Tθ ) y+Tx , sin ( Tθ ) x+cos ( Tθ ) y+Ty} . The optimal transformation Tp that aligns p to P is calculated asTp=arg minT ( ∑iwi ( ( Pix− ( Tpi+τp ) x ) 2+ ( Piy− ( Tpi+τp ) y ) 2 ) ∑iwi ) , with wi=Pifpi+τpfδixδiy . The same procedure is used to compute Tr that aligns r to R . The spatial alignment of the trajectories of Abp1 , Arc18 and Act1 was computed using only the trajectory data associated with the invagination dynamics , up to the trajectory peak in fluorescence intensity . Between 50 and 350 trajectory pairs were used to compute the average alignment ( Table 1 ) . For each trajectory pair , we got an estimate of the transformations that align r to R and p to P . An estimate of the transformation that aligns P to R is thus the combination of the inverse transformation that aligns p to P and of the transformation that aligns r to R:T=Tr ( Tp ) −1 . The average transformation is then the average of the estimates of the individual transformations that align P to R , computed from each of the trajectory pairs . The average time alignment τ is computed as:τ=median ( τr−τp ) and δτ is the estimate of the standard error for the median computed as:δτ=1 . 4826×MADτMMADτ is the median absolute deviation for τ and M is the number of trajectory paires used to compute the average transformation . Tx and Ty are computed as:Tx=median ( Trx−cos ( Trθ−Tpθ ) Tpx+sin ( Trθ−Tpθ ) Tpy ) , Ty=median ( Try−sin ( Trθ−Tpθ ) Tpx−cos ( Trθ−Tpθ ) Tpy ) . Tθ is computed as:Tθ=median ( Trθ−Tpθ ) . δTx , δTy and δTθ are computed as δτ using their respective MAD . Note that the approximation used to estimate Tθ is only valid if Trθ≈Tpθ . To ensure that this is the case , R and P are first aligned to their axis of symmetry , which is the direction of the invagination that we defined as the X-axis . As r and p are already aligned along the axis of invagination , because they were acquired simultaneously , the final rotation that align P to R is close to the identity ( Tθ≈0 ) . Centering P to its center of mass further minimize any error that might be induced by the above approximation . The trajectory of the protein of interest aligned to the reference trajectory is thus:Pi′={ ( TPi ) x , ( TPi ) y , Pif , ζix , ζiy , ζif} , where ζix and ζiy are the standard errors computed as:ζix= ( δix ) 2+ ( PiyδTθ ) 2+ ( δTx ) 2 , ζiy= ( δiy ) 2+ ( PixδTθ ) 2+ ( δTy ) 2 . All plots describing the trajectories inward movement along the invagination direction report the 95% confidence intervals , computed as 1 . 96×ζix and 1 . 96 × δτ . See Source code 2 . Trajectory pairs of Myo5-GFP or Las17-GFP and Abp1-mCherry were used to generate Myo5 and Las17 average trajectories . The rotations and translations that aligned the Abp1-mCherry trajectories and Abp1-GFP average trajectory together were used to align the corresponding Myo5-GFP or Las17-GFP trajectories . Once aligned , the coordinates of the GFP trajectories were averaged together at each time point . To control the accuracy of our alignment procedure we generated virtual trajectory pairs using two trajectories that we generated as a template ( ground truth trajectories ) : one ground truth trajectory represents the reference protein while the other represents the target protein . The points of the trajectory pairs were randomly generated and were normally distributed around the ground truth trajectories with sigma σp for the trajectories of the target protein and σr for the trajectories of the reference protein . All trajectories were generated with the same sampling rate as the real trajectories . The virtual trajectory pairs were then processed with the same pipeline we used for the real data to align the ground truth trajectories together . As the ground truth trajectories were already aligned all transformations are expected to be 0 ( Figure 1—figure supplement 4A–D ) . When we arbitrarily increased the noise used to generate both the reference and the target proteins we observed that the alignment procedure remains faithful up to ∼25 nm of noise . For higher values a small systematic shift occurs ( Figure 1—figure supplement 4C ) . We then tested the robustness of the alignment procedure keeping σr fixed at 19 nm while we varied σp . A σr fixed mimics the noise in the real trajectory pairs , as the reference protein was always imaged under the same conditions . The alignment showed negligible errors ( Figure 1—figure supplement 3E–L ) . Note that in the real trajectory pairs σp span a range from 10 nm to 24 nm , however the algorithm was stable also for higher errors . We observed a systematic shift , of maximally ∼3 nm , that occurs when the trajectories are getting far from the reference ( Figure 1—figure supplement 3K ) . This is expected as any small error in the alignment of the trajectories leads to an underestimate of the separation between the reference and the target protein . This underestimate grows larger the bigger the separation is , resulting in the trajectory of the target protein being shifted little closer to its reference than it should be . A shift of 3 nm is reached when the trajectories are 30 nm away far from the reference protein , which is the maximal distance of a trajectory from its reference in our model . To test for the robustness against systematic shifts between the two channels , we shifted the reference trajectories in the virtual trajectory pairs that we generated on the computer ( see ‘Materials and Methods’: Simulation of the accuracy of the two color alignment procedure; σr = 16 nm and σr = 19 nm ) , by shifts up to 150 nm along one direction . These shifts simulate different amounts of aberration between the two channels . On average , the alignment is not affected ( Figure 1—figure supplement 4A ) . In fact , the trajectory pairs , as well as the endocytic events that they mimic , are oriented in all possible directions and once their average relative position is calculated , the chromatic aberration contribution is averaged out . However , the incertitude in the average position increases , as expected ( Figure 1—figure supplement 4A ) . It is important to note that our pixel size correspond to 100 nm , therefore any shift larger than 50–100 nm would have been easily noticeable by eye . To assess how the alignment of the real data would be affected by a systematic color shift we also shifted the real trajectory pairs we used to align Sla2-GFP tp Abp1-GFP . Again , we induced shifts up to 150 nm to the trajectories of Abp1-mCherry to simulate different amounts of aberration between the two channels . The average position of Sla2-GFP does not significantly change but the incertitude in its average position increases ( Figure 1—figure supplement 4B–D ) . To test the accuracy of the trajectory averaging we generated 65 virtual trajectories starting from a trajectory template ( ground truth trajectory ) . The trajectories were generated adding noise that was normally distributed around the points of the ground truth trajectory with a standard deviation σ . We sampled values of σ covering the range of noise encountered experimentally ( between 10 nm and 20 nm ) . The average trajectories where then aligned in time and in space to a reference trajectory that was generated together with the ground truth trajectory . To compute the alignment we used virtual trajectory pairs generated from the ground truth trajectory and the reference trajectory as described in ‘Materials and Methods: simulation of the accuracy of the two color alignment procedure’ , with noises σp = 10 nm and σr = 19 nm respectively . The averaging procedure and the complete alignment procedure are very robust: after the alignment the average trajectories were reproducing very closely the ground truth trajectory ( Figure 1—figure supplement 5 ) . As expected , the average of very noisy trajectories slightly underestimates the full length of the ground truth trajectory ( Figure 1—figure supplement 5D ) . To quantify the protein amounts we imaged a sample containing cells from both a yeast strain expressing a fluorescently tagged protein of interest and a yeast strain expressing fluorescently tagged Nuf2 , which was used as a reference to calibrate the fluorescence intensity ( Joglekar et al . , 2006 ) . Both strains expressed the same fluorescent tag and were imaged together . For the quantification of the Nuf2 fluorescence intensity we used cells in anaphase-telophase only . Excitation light and emission light were directed through the U-MGFPHQ ( Olympus , Japan ) . Samples were imaged as a z-stack of 21 frames , 200 nm spaced , using the Hamamatsu Orca-ER CCD camera . Each frame was excited with X-Cite 120Q lamp for 400 ms . Frames were not processed for background subtraction and the spots were quantified by quantifying the fluorescence intensity of the patches , in the frame of the z-stack in which they were brighter , and subtracting their local background , as described in Joglekar et al . ( 2006 ) . The intensities of the fluorescent patches of the target endocytic proteins where in general dimmer than Nuf2 , and their distribution was in general not symmetric . The intensities measured from the patches of the target proteins were thus processed after a logarithmic transformation . The average number of molecules of the target protein np was derived as:np=fgnr . nr is the known number of molecules of the reference , f is the median of the fluorescence intensity of the patches of the target protein and g is the median fluorescence intensity of the patches of the reference protein . The uncertainty in the number of molecules was derived as: ( 1 ) δnp= ( nrfgδl ) 2+ ( nrfg2δg ) 2+ ( fgδnr ) 2 . δl is the estimate of the standard error for the median of the fluorescence intensity of the target protein after logarithmic transformation , l = log ( f ) , and is computed as:δl=1 . 4826×MADlN . MADl is the median absolute deviation for l and N is the number of observations . The distribution of the intensities of the target endocytic proteins where not symmetric , they were thus processed after a logarithmic transformation of their intensities . δg is computed as δl , using MADg , the median absolute deviation of the fluorescence intensity of the reference protein . δnr is the uncertainty in Nuf2 number of molecules . The average number of molecules np was then used to rescale fluoresce intensity curves of the average trajectories ( see ‘Materials and Methods’: calibration of the fluorescence intensity curve of the trajectories with the number of molecules ) . Nuf2 number of molecules was quantified using Cse4 as a reference . The number of Cse4 molecules used for the calibration was 5 molecules/kinetochore ( Lawrimore et al . , 2011 ) . The measured average number of Nuf2 molecules , per fluorescent spot , was 280 . 6 ± 16 . 1 molecules . Its error was quantified as in Eq . ( 1 ) with δnr = 0 . The quantification of Nuf2 molecules with Cse4 served as a control for our procedure as there is no difference in the ratio between our number of Nuf2 molecules and the number of Cse4 molecules , which is 3 . 5 ± 0 . 2 Nuf2 molecules each Cse4 , and Joglekar's ratio , which is 3 . 5 . To quantify the protein abundance of Arc18 , which was tagged with myEGFP , we compared the fluorescence intensity of myEGFP and EGFP tags and we measured the myEGFP tags to be 68% ± 14% of the fluorescence intensity of an EGFP tag . The uncertainty in Arc18 number of molecules was then computed as:δnp= ( cfgnrδl ) 2+ ( cfg2δg ) 2+ ( cfgδnr ) 2+ ( fgnrδc ) 2 . c is the estimate of the correction for the fluorescence intensity and δc is its uncertainty . We calibrated the fluorescence intensity curve Pf for each average trajectory , Pi={Pix , Piy , Pif , δix , δiy , δif} with i={1 , … , N} , to estimate of the number of molecules Pn over time . Pn is computed by rescaling the fluorescence intensity curve of the protein of interest with the average number of molecules np at the endocytic site ( see ‘Materials and methods’: quantification of the number of molecules ) :Pin=nPPif−Pminf1N∑j=1N ( Pjf−Pminf ) , with Pminf=mink ( Pkf ) The error in the estimate of the number of molecules is:δin= ( Pif−PminfF¯δnP ) 2+ ( npNNF¯−Pif+PminfF¯2δif ) 2+ ( npPif−Pminf−F¯F¯2δm ) 2 , δm=δlf with l=arg minj ( Pjf ) , F¯=1N∑i=1N ( Pif−Pminf ) . where δnp is the standard error of the average number of molecules measured in the endocytic fluorescent patches ( see ‘Materials and Methods: quantification of the number of molecules ) . All plots describing the number of molecules report the 95% confidence interval , computed as 1 . 96 × δin . To quantify the ratio between GFP-Act1 and the total actin we run western blots with cell extract of yeast cells expressing GFP-Act1 ( MKY2653 ) ( Wu and Pollard , 2005 ) . MKY2653 cells where grown overnight in SC-URA media . As primary antibody against actin we used Sigma A2066 . The secondary antibody was a AP-1000 Alkaline Phosphatase anti-rabbit igG ( H + L ) from Vector Laboratories . The quantification was repeated 8 times . The ratio r˜ between GFP-Act1 and the total actin was computed as:r˜=rr+1where r is the average ratio between the GFP-Act1 and the endogenous actin that we measured from the 8 quantifications . The error was computed as:σr˜=σr ( r+1 ) 2where σr is the standard error of the mean of r . To determine the angle between Abp1 trajectories and the yeast cell surface , we determined the closest membrane tangent for each Abp1 track , using a binary mask of the yeast cell in which the trajectory was acquired . We then measured the angle between the vector tangent to the membrane and the vector whose direction was determined by the interpolation of Abp1 trajectory points on the focal plane ( Figure 1—figure supplement 2B ) . Individual photobleaching experiments were tracked with the Particle Tracker ( Sbalzarini and Koumoutsakos , 2005 ) plugin in ImageJ after background subtraction , normalization and cytoplasmatic background subtraction of the images ( see ‘Materials and methods’: Image analysis ) . The tangent to the plasma membrane was determined in ImageJ using a binary mask of the cell . A custom written software in R was then used to extrapolate the trajectory along the direction orthogonal to the plasma membrane , which represents the inward movement of the photobleached endocytic spot . The resulting trajectories were aligned in time and in space to the average trajectory of the corresponding protein using the fluorescence intensity curve and the inward movement of the average trajectory and of the photobleached trajectory before the photobleaching . After alignment , the photobleached trajectories were thus aligned in space and time with all the average trajectories and with the plasma membrane profiles ( Figure 6C–F ) . To check whether Sla2 was oriented on average perpendicularly to the plasma membrane , we tagged simultaneously the N-terminus with sfGFP and the C-terminus with mCherry ( GFP-Sla2-RFP , Figure 3—figure supplement 1A ) . In cells treated with 2 µM LatA , we recorded the closest membrane tangent to each GFP spot . We then measured the angle between the vector tangent to the membrane and the vector whose direction was determined by the centroids of the corresponding GFP and mCherry spots , for each GFP and mCherry pairs ( Figure 3—figure supplement 1B ) . In order to determine whether the displacement between the N- and C-terminal trajectories of Sla2 is a measure for Sla2 length we measured the distance between the N- and C-terminus of Sla2 , by tagging simultaneously the N- terminus with GFP and the C-terminus with mCherry ( Figure 3—figure supplement 1A ) . We arrested the membrane invagination by treating cells with 2 µM of LatrunculinA ( LatA ) ( Kukulski et al . , 2012 ) and we imaged cells on the GFP and mCherry channels . Images were corrected for chromatic aberration as for the two color trajectories ( see ‘Materials and methods’: Two color alignment procedure ) . The separations between the centroids of the GFP and mCherry pairs follow a non-gaussian distribution ( Stirling Churchman et al . , 2006 ) , which was used to compute the distance between the fluorophores and thus estimate the displacement of the N- and C-terminal tags of Sla2 ( Figure 3—figure supplement 1C ) . The distance is reported in the text together with the estimate of the standard error of the mean obtained from the observed Fisher information matrix . As a control , we performed the same analysis on a TetraSpeck sample ( Figure 3—figure supplement 1D ) and on Sla2 tandemly tagged at its C-terminus with both mCherry and sfGFP ( Figure 3—figure supplement 1E , F ) . To estimate the membrane area covered by the Rvs161/167 proteins , we assumed that all protein molecules are membrane bound and homogeneously distributed on the surface of the endocytic invagination . Their distribution was computed considering that BAR proteins form a dimer that is ∼13 nm long ( Peter et al . , 2004 ) and tubulate in vitro forming spirals spaced by 50 Å ( Mim et al . , 2012 ) . We used those data , together with our estimate of the number of Rvs molecules and with the average membrane shapes , to model all possible coverages of the invagination along the invagination length . We then chose the coverage whose center of mass matched the position of the Rvs167-GFP average trajectory . We extrapolated the plasma membrane profiles to estimate the plasma membrane shape at time 0 in order to determine the Rvs coverage just prior to scission . Abp1 lifetimes were estimated from the lifetime of the trajectories ( i . e . the time difference between the first and the last time point in the trajectory ) . We considered only the trajectories that were complete and , to minimize as much as possible the effect of photobleaching , we considered only the trajectories that were recorded in a number of frames , at the beginning of the video , covering ∼3 times the known lifetime of the protein . Note that to achieve robust automatic patch detection the thresholding of the fluorescent patches was stringent , resulting in Abp1 patch lifetimes that are an underestimate of the true patch lifetime . ConA crosslinked glass coverslips were prepared as described previously ( Mund et al . , 2014 ) . 24 mm coverslips were cleaned for at least 12 hr in 1:1 methanol/hydrochloric acid , and washed in ddH2O until the pH of the solution remained neutral . 20 µl of Bioconext ( UCT , Bristol , PA ) was spread out over the coverslip and incubated for 30 min , followed by washing twice with ethanol , twice with H2O , drying at 65°C for 30 min and incubation for 60 min with 20 µl of 2% ConA . Coverslips were then rinsed three times with H2O , air dried and stored until usage . Sample preparation was performed as described previously ( Mund et al . , 2014 ) . In summary , yeast cells were grown in SC-Trp medium until reaching log phase , pelleted by centrifugation , resuspended in a small volume of ddH2O and pipetted on the ConA crosslinked glass coverslips . After settling for 15 min , the supernatant was removed and the coverslips were submerged in a fixative containing 4% formaldehyde , 2% sucrose in PBS for 15 min . Cells were then incubated two times for 15 min in PBS containing 50 mM NH4Cl to stop the fixation . Coverslips were subsequently put face down on a 100 μl drop of blocking solution consisting of 50% ImageIT FX ( Invitrogen , Carlsbad , CA ) to block background due to unspecific binding and 0 . 25% Triton-100 in PBS for 60 . Coverslips were briefly washed three times with PBS and put face down on a 100 µl drop of SNAP labeling solution containing 1 µM SNAP-Surface Alexa Fluor 647 , 1% BSA , 0 . 25% Triton X-100 , 0 . 004% NaN3 in PBS and incubated for 120 min . Finally , coverslips were washed three times by gentle shaking in PBS for at least 5 min . Localization microscopy was performed on a custom-built microscope . Single-mode output from an iChrome MLE-L laser box equipped with 405 nm , 488 nm , 561 nm and 640 nm laser lines ( Toptica Photonics , Germany ) was focused onto the back–focal plane of a 60× NA 1 . 49 TIRF objective ( Nikon , Japan ) and adjusted for epi illumination . Emission light was filtered using an ET 700/100 bandpass filter ( Chroma , Bellows Falls , VT ) and a FF01-446/523/600/677 multi bandpass filter ( Semrock , Rochester , NY ) , and focused by a 400 mm tube lens onto the chip of an EMCCD camera ( Ixon Ultra , Andor , United Kingdom ) that was air-cooled to −75°C . Images were acquired using MicroManager ( Edelstein et al . , 2010 ) . A piezo objective positioner ( Physikinstrumente , Karlsruhe , Germany ) was used to move the z-focus . The focus was stabilized by an electronic feedback loop based on an infrared laser that was totally internally reflected at the coverslip and detected by a quadrant photodiode . The z stability was better than ±10 nm over several hours . Lateral drift , typically smaller than 50 nm/hr , was corrected for in the analysis software . Image acquisition was performed as described ( Mund et al . , 2014 ) . In brief , the samples were mounted in a custom-made holder and covered with at least 200 µl of imaging buffer consisting of 150 mM Tris–HCl pH7 . 5 , 2% glucose , 60 mM cysteamine , 40 µg/ml catalase and 0 . 5 mg/ml glucose oxidase . We used an exposure time of 25 ms and an EM gain of 100 . Imaging laser intensity at 640 nm was 2 . 5 kW/cm2 , the activation laser intensity was automatically adjusted to ensure a constant number of localizations per frame . Typically 30 , 000–50 , 000 frames were recorded . Localization analysis was performed as previously described ( Ries et al . , 2012 ) . In summary , photon counts were obtained by subtraction of the constant offset from the pixel count and multiplication with the inverse gain . Initially , approximate positions of bright spots were determined by smoothing , nonmaximum suppression and thresholding . Selected peaks were fitted by a pixelized Gaussian function and a homogenous photonic background with a maximum likelihood estimator for Poisson distributed data using a freely available , GPU based fitting routine ( Smith et al . , 2010 ) on a Geforce GTX670 ( Nvidia , Santa Clara , CA ) . Lateral drift correction was performed using image correlation as previously described . Localizations with an uncertainty of >15 nm were discarded . The images in Figure 3 and Figure 3—figure supplement 2 were rendered using a Gaussian with a width according to the respective localization precision . All analysis software was written in Matlab ( TheMathworks , Natick , MA ) . The observed structures in Figure 3—figure supplement 2 were visually classified . ‘Clear ring structures’ show both localizations in circular arrangement and a pore zone in the middle with virtually no localizations , while for ‘Possible ring structures’ one of the two criteria was less striking . ‘No ring structures’ exhibit other geometries . | Cells take up proteins and other useful material ( called cargo ) from their external environment through a process known as endocytosis . To start with , the cargo accumulates in a patch on the surface of the cell . On the inner side of the cell's membrane , a protein called clathrin gathers around the patch of cargo . Clathrin molecules and many other proteins bind together to make a lattice-like coat that causes the membrane to curve inwards and form a pocket that contains the cargo . This continues until the cargo is completely surrounded by membrane and eventually forms a bubble-like structure , or ‘vesicle’ , that moves into the cell . More than 50 other proteins are involved in the endocytosis . These proteins arrive at the site of endocytosis in a particular order , complete their tasks and then move away to be used in further rounds of endocytosis . It is not clear how these proteins are organized to complete these steps because it is technically difficult to track the movements of many proteins at the same time . Here , Picco et al . developed a new fluorescence microscopy method that enabled them to track the positions of many of the proteins involved in endocytosis in yeast cells in real time . The experiments revealed when the proteins arrived at the site of endocytosis and how they assembled in relation to the membrane . For example , a group of proteins called N-BAR proteins formed an extended lattice covering the sides of the pocket that forms as the membrane curves inwards . To transform the flat membrane into a vesicle , a network of filaments made of a protein called actin needs to form at the site of endocytosis . The new method shows that the actin filaments grow in a small region at the base of the developing vesicle . By combining different types of microscopy data , Picco et al . were able to build a comprehensive model describing when the proteins involved in endocytosis move and assemble . The next challenge will be to understand the physics behind the molecular machine composed of these many proteins and the cell membrane . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology",
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2015 | Visualizing the functional architecture of the endocytic machinery |
Breathing is a vital rhythmic behavior generated by hindbrain neuronal circuitry , including the preBötzinger complex network ( preBötC ) that controls inspiration . The emergence of preBötC network activity during prenatal development has been described , but little is known regarding inspiratory neurons expressing pacemaker properties at embryonic stages . Here , we combined calcium imaging and electrophysiological recordings in mouse embryo brainstem slices together with computational modeling to reveal the existence of heterogeneous pacemaker oscillatory properties relying on distinct combinations of burst-generating INaP and ICAN conductances . The respective proportion of the different inspiratory pacemaker subtypes changes during prenatal development . Concomitantly , network rhythmogenesis switches from a purely INaP/ICAN-dependent mechanism at E16 . 5 to a combined pacemaker/network-driven process at E18 . 5 . Our results provide the first description of pacemaker bursting properties in embryonic preBötC neurons and indicate that network rhythmogenesis undergoes important changes during prenatal development through alterations in both circuit properties and the biophysical characteristics of pacemaker neurons .
Rhythmic motor activities are generated and controlled by neuronal networks organized as central pattern generators ( CPG ) ( Marder and Bucher , 2001; Harris-Warrick , 2010 ) . Considerable data accumulated over the last decades from both invertebrate and vertebrate models have established the general mechanistic principle that rhythmogenesis relies on an interplay between intrinsic neuronal membrane properties and intercellular synaptic connectivity . Two main processes that may operate in varying combinations underlie motor rhythm generation: ( i ) the CPG network in question contains endogenously oscillatory neurons , so-called pacemakers , which drive the wider circuit cell population , and/or ( ii ) , the rhythm emerges from the pattern of synaptic connections within the network . In addition to these intrinsic rhythmogenic mechanisms , the dynamics of network function can be conferred by extrinsic neuromodulatory actions . By acting on the membrane properties of constitutive neurons or their synaptic interconnections , modulators can ensure the operational plasticity that enables network motor output to remain adapted to organismal needs and changing environmental conditions ( for reviews , see Harris-Warrick , 2011; Marder , 2011; Marder et al . , 2014 ) . A major physiological function of such a CPG is breathing . Respiratory movements are driven by rhythmic motor activity generated by neuronal circuits located in the brainstem . The respiratory rhythm generator is composed of two interacting CPG circuits distributed bilaterally in the ventral part of the medulla . The first is the parafacial respiratory group ( RTN/pFRG; Onimaru and Homma , 2003 ) that appears to generate preinspiratory activity in neonates in vitro , active expiration in adults and plays a prominent role in central chemosensitivity ( Guyenet and Bayliss , 2015 ) . The second network is the preBötzinger complex ( preBötC; Smith et al . , 1991 ) which has now been established to be both sufficient and necessary for generating the inspiratory phase of respiration ( Smith et al . , 1991; Gray et al . , 2001; McKay et al . , 2005; Tan et al . , 2008; Bouvier et al . , 2010 ) . The excitatory glutamatergic preBötC network contains ~800 neurons , some of which ( ≤15% ) in neonatal mouse exhibit intrinsic pacemaker properties ( Koshiya and Smith , 1999; Thoby-Brisson and Ramirez , 2001; Pena et al . , 2004 ) . To date , a leading hypothesis , inscribed in the 'group pacemaker hypothesis' , proposes that the rodent postnatal respiratory rhythm derives from an interaction between membrane properties ( including pacemaker cellular properties ) and synaptic coupling ( Rekling and Feldman , 1998; Feldman and Del Negro , 2006; Feldman et al . , 2013 ) . It has been shown in rodents that the preBötzinger complex becomes functional during the last third of gestation ( Pagliardini et al . , 2003; Thoby-Brisson et al . , 2005 ) . Already at early embryonic stages , glutamatergic synaptic signaling is required for preBötC network output ( Thoby-Brisson et al . , 2005; Wallen-Mackenzie et al . , 2006 ) , although the presence of embryonic inspiratory neurons endowed with intrinsic bursting properties has only been inferred ( Thoby-Brisson et al . , 2005; Bouvier et al . , 2008 ) . Therefore , the aim of this study was to establish the presence and biophysical characteristics of pacemaker neurons in mouse embryonic preBötC circuitry in order to understand their development and contribution to respiratory network activity in the critical period immediately prior to birth .
To identify pacemaker neurons in preBötC respiratory circuitry of mouse embryos between E16 . 5 and E18 . 5 , we combined electrophysiological recordings of population rhythmic activity on one side with individual cell calcium imaging on the contralateral side of brainstem slice preparations ( Figure 1A ) . For this , slices were previously incubated en bloc with the Calcium Green 1-AM indicator , allowing fluorescence fluctuations due to somatic Ca2+ fluxes resulting from spontaneous impulse burst generation to be monitored ( see Materials and methods ) . Initially , rhythmic fluorescent changes in cells occurring in phase with the population electrical activity allowed the localization of inspiratory neuron somata ( Figure 1A , right and Figure 1B ) . We identified an endogenous pacemaker neuron by its ability to produce spontaneous membrane potential oscillation and rhythmic action potential burst discharge even in synaptic isolation from its network partners ( Koshiya and Smith , 1999 ) . Accordingly , neurons expressing fluorescence fluctuations in time with fictive inspiration in control conditions were classified as pacemakers if they remained rhythmically active during subsequent exposure to a cocktail of agents known to block chemical synaptic transmission in the preBötC network ( see Material and methods ) . Under such conditions of synaptic blockade , network electrical activity ceased as did rhythmic fluorescent changes in most of the previously identified inspiratory neurons ( Figure 1C , black traces ) . However , a small proportion of monitored cells continued to express spontaneous fluorescence fluctuations at unrelated frequencies ( Figure 1C , red traces ) . These specific neurons were therefore considered to be pacemaker cells and were targeted for patch-clamp recording . 10 . 7554/eLife . 16125 . 003Figure 1 . Functional isolation of embryonic inspiratory pacemaker neurons . ( A ) left , Schematic representation of an in vitro embryonic slice preparation used for making electrophysiological macroelectrode recording of preBötC network activity on one side simultaneously with calcium imaging of the contralateral inspiratory network ( red square ) . Right , Image of fluorescence expression by a Ca2+-dye loaded slice obtained with a 40X objective . Numbered yellow circles indicate rhythmically active neurons . ( B ) Simultaneous recordings of inspiratory population electrical activity ( integrated trace in violet , Int preBötC ) and calcium transients ( ΔF/F , black and red traces ) in 16 individual neurons ( numbered 1 to 16 ) . All the cells displayed fluorescent changes synchronized to rhythmic electrical activity recorded on the contralateral side in control conditions . ( C ) After blockade of chemical synapses by a bath- applied cocktail of CNQX ( 20 µM ) , AP5 ( 10 µM ) , strychnine ( 1 µM ) and bicuculline ( 10 µM ) , neurons that remained rhythmically active ( red traces ) were considered to be pacemakers while those falling silent were considered as non-pacemaker cells ( black traces ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16125 . 003 Of the 84 pacemaker neurons identified in 69 embryo slice preparations , three distinct types of discharge pattern were observed that differed in the characteristics of the spontaneous depolarizing waveforms - or drive potentials ( DPs ) – that underlie their intrinsic bursting activities . In a first group ( n = 23 ) , the cells expressed long-lasting plateau-like DPs with action potentials occurring at the beginning of the plateau followed by a depolarization block during which the neuron remained at a depolarized membrane potential without further spike generation prior to a spontaneous return to resting potential ( Figure 2A ) . The mean amplitude of such square-wave DPs was 30 . 5 ± 6 . 6 mV ( 343 burst cycles measured from 12 neurons; Figure 2D1 , left ) and their mean duration was 2 . 9 ± 0 . 1s ( Figure 2D1 , right ) . The mean membrane potential of these neurons measured between their DPs was −52 . 7 ± 0 . 7 mV . In a second group ( n = 45 ) , the pacemaker neurons generated short-lasting oscillatory DPs and associated bursts with depolarizing amplitude and duration means of 12 . 8 ± 3 . 8 mV and 0 . 79 ± 0 . 01 s , respectively ( 1296 bursts measured from 23 neurons; Figure 2B , D2 ) . Their mean membrane potential between bursts was −49 . 2 ± 0 . 5 mV . In the third group ( n = 16 ) , cells expressed a mixture of long- and short-lasting DPs , which overall had amplitude and duration means of 14 . 5 ± 7 . 5 mV and 1 . 74 ± 0 . 08 s , respectively ( 739 bursts measured from 15 neurons; Figure 2C , D3 ) . The mean inter-burst membrane potential of the mixed phenotype was −49 . 1 ± 0 . 7 mV . The durations of the drive potentials were statistically different between the three groups ( Mann-Whitney test , p<0 . 001; Figure 2D4 , right ) , while the mean DP amplitude of plateauing pacemakers was significantly larger compared to the amplitudes of two other types ( p< 0 . 001 ) , which themselves were not significantly different ( Figure 2D4 , left ) . However , for the mixed pacemaker phenotype , when we discriminated between the two types of bursting ( light green bars in Figure 2D4 ) , the mean values for both DP amplitude and duration were comparable to those of the separate plateauing and oscillatory bursters . Moreover , as evident in the DP amplitude vs duration relationship of Figure 2E ( which note was plotted exclusively from measurements of long-term single cell recordings; see Material and methods ) , the values for the mixed phenotype were bimodally distributed , with its shorter bursts lying in the range for oscillatory only pacemaker neurons and its longer bursts overlapping values for plateauing only cells . Finally , the mean resting membrane potential values were statistically different between both plateauing and oscillatory neurons ( t-test , p<0 . 001 ) and between plateauing and mixed pacemaker cells ( t-test , p<0 . 002 ) . Thus , together these data show that the respiratory network at embryonic stages between E16 . 5 and E18 . 5 already contains a subpopulation of pacemaker neurons and that these cells are endowed with heterogeneous burst-generating intrinsic properties . 10 . 7554/eLife . 16125 . 004Figure 2 . Different discharge patterns expressed by embryonic inspiratory pacemaker neurons . Patch-clamp recordings of individual pacemaker neurons from three different preparations expressing spontaneous plateau ( A ) , oscillatory ( B ) or mixed plateau/oscillatory ( C ) burst firing patterns . ( D ) Histograms showing frequency distributions of the burst-generating dirve potential ( DP ) amplitude ( D1–3 , left column ) and duration ( D1–3 , right column ) for the three types of pacemaker neurons . D4: Histograms representing mean DP amplitude ± SEM ( left ) and mean DP duration ± SEM ( right ) for pacemaker neurons expressing plateau-like ( blue bars ) , burst-like ( red bars ) or mixed ( green bars; light green bars correspond to short-lasting and long-lasting bursts when grouped separately ) activity . Asterisks indicate significant differences ( Mann-Whitney test; p<0 . 001 ) , while numbers of neurons analyzed for each pacemaker phenotype are indicated in the corresponding bar . ( E ) Distribution plot for DP duration vs DP amplitude measured for neurons recorded for 8 to 10 min ( blue , plateau bursting pacemakers ( n = 8 ) ; red , oscillatory bursting pacemakers ( n = 5 ) ; green , mixed pacemakers ( n = 5 ) . Note that the latter expressed two types of burst duration/amplitude relationship that overlapped with either the purely plateau or oscillatory bursters . DOI: http://dx . doi . org/10 . 7554/eLife . 16125 . 004 We next investigated the membrane properties involved in the pacemaker activities of these embryonic inspiratory neurons . I-V curves obtained from isolated plateauing ( n = 6 ) and oscillatory ( n = 14 ) pacemaker neurons expressed a non-linear deviation at hyperpolarized membrane potentials , indicating the activation of voltage-dependent membrane conductances ( Figure 3A ) . Consistent also with the voltage-dependence of an endogenous pacemaker mechanism , membrane potential depolarization with current injection caused a cycle frequency increase in both types of bursting neuron , with the rates of plateauing and oscillatory bursting increasing as a function of injected current intensity ( Figure 3B , C ) . It is also noteworthy , however , that the I-V relationships of the two pacemaker phenotypes were statistically different ( t-test; p<0 . 001 ) in the range of more depolarized membrane potential levels between −20 mV and 0 mV , thus indicating differences in properties that govern their membrane excitability . No difference ( t-test , p= 0 . 7 ) was found in the membrane input resistance measured around resting potential for plateau-like ( 1756 ± 667 MΩ ) and burst-like pacemaker neurons ( 1574 ± 973 MΩ ) . 10 . 7554/eLife . 16125 . 005Figure 3 . Voltage-dependence of pacemaker neuron properties . ( A ) Current-voltage ( I-V ) relationships for plateau ( black circles ) and oscillatory bursting ( open circles ) pacemaker neurons . Individual circles indicate mean ( ± SEM ) current response to an injected step-change voltage pulse; n = number of neurons analyzed for each pacemaker phenotype . Asterisks indicate statistically different values ( t-test; p<0 . 001 ) . ( B , C ) Patch-clamp recordings of plateau ( B ) and oscillatory bursting ( C ) pacemaker neurons displaying cycle frequencies that varied with membrane voltage when held at different indicated membrane potential levels with continuous depolarizing current injection . Right inset: A single burst ( indicated by shading ) in each pacemaker neuron subtype displayed at the same time scale for burst duration comparison . DOI: http://dx . doi . org/10 . 7554/eLife . 16125 . 005 In order to identify the major membrane conductances contributing to the intrinsic burst activity of these embryonic pacemakers , we blocked the persistent sodium current ( INaP ) and the calcium-activated non-specific cationic conductance ( ICAN ) that are known to be widely implicated in the rhythmogenic mechanisms of mammalian motor circuits ( van Drongelen et al . , 2006; Zhong et al . , 2007; Tazerart et al . , 2008; Tsuruyama et al . , 2013 ) , including the preBötC respiratory network of the postnatal mouse ( Thoby-Brisson and Ramirez , 2001; Pena et al . , 2004; Del Negro et al . , 2005; Paton et al . , 2006; Pace et al . , 2007b , 2007a ) . We therefore bath-applied Riluzole ( Ril; 10 µM ) a blocker of INaP ( Urbani and Belluzzi , 2000 ) , and Flufenamic Acid ( FFA; 50 µM ) a blocker of ICAN ( Guinamard et al . , 2004 ) onto synaptically-isolated , patch-clamp recorded neurons . Note that since to our knowledge none of these two drugs are fully washable , we applied FFA and Ril alone , or in combination , but never sequentially on a given slice . Out of the 10 embryonic preBötC neurons identified as plateauing pacemakers , the bursting activity of 8 of these was completely blocked in the presence of 50 µM FFA applied either alone ( n = 4; Figure 4A , right panel ) or in co-application with 10 µM Ril , which by itself was ineffective in blocking bursting activity ( n = 4 , Figure 4A , left panel ) . For the remaining two plateauing cells , bursting activity was blocked in the presence of Ril alone ( n = 2 ) . These findings therefore suggest that pacemaker activity of the plateauing inspiratory neurons relies on a combination of both INaP and ICAN , with a predominant role played by the latter . 10 . 7554/eLife . 16125 . 006Figure 4 . Membrane conductances underlying the different discharge patterns . Patch-clamp recordings of plateau ( A ) , oscillatory ( B ) and mixed ( C ) bursting pacemaker neurons in control conditions ( top traces ) , in the presence of either 10 µM Ril or 50 µM FFA ( middle traces ) or in the presence of 10 µM Ril + 50 µM FFA ( bottom traces ) . The left and right panels in A illustrate two different plateau bursters under different indicated pharmacological treatments . Rhythmic burst activity of the plateau pacemaker was mostly affected ( i . e . , reduced ) by blockade of ICAN channels with FFA ( A ) , whereas the oscillatory pacemaker activity was more sensitive to persistent sodium channel ( INaP ) blockade with Ril ( B ) . Correspondingly , longer duration plateau driven discharge in the mixed pacemaker phenotype was blocked with FFA alone , whereas remaining short-duration oscillatory bursts were blocked by the addition of Ril ( C ) . Shaded activity at right of each trace: recording excerpts on expanded time-scale to facilitate comparison . DOI: http://dx . doi . org/10 . 7554/eLife . 16125 . 006 For the vast majority ( 13/17 ) of the oscillatory bursting neurons tested , pacemaker activity was blocked in the presence of 10 µM Ril alone ( n = 8; data not shown ) or when co-applied with FFA ( n = 5; Figure 4B ) . In the remaining four neurons , bursting was blocked by 50 µM FFA alone . For pacemaker neurons exhibiting a mixed DP pattern ( n = 10 ) , bursting activity was differentially sensitive to FFA and Ril , with plateau-like bursts being suppressed by FFA while short-lasting oscillatory bursts were blocked by Ril ( Figure 4C ) . Thus , the heterogeneous patterns of discharge in pacemaker neurons of the embryonic preBötC network within the period from E16 . 5 to E18 . 5 appear to be associated with different combinations of membrane conductances contributing to underlying drive potential activity: ICAN plays a major role for plateau pacemakers , INaP predominates in oscillatory pacemakers , whereas the relative conductance contribution in the mixed pacemaker phenotype is intermediate between these two . We next determined whether the distinct types of pacemakers ( plateauing , oscillatory; INaP-dependent , ICAN-dependent ) occur similarly throughout the prenatal period tested . Initially , pacemaker neurons at E16 . 5 ( n = 27 ) and E18 . 5 ( n = 35 ) were classified according to their discharge patterns obtained with single cell patch-clamp recordings at the two embryonic stages . Plateau and oscillatory pacemakers were found to be almost equally present at E16 . 5 , whereas oscillatory bursters were predominant ( 65 . 7% , n = 23/35 ) at E18 . 5 ( Figure 5A ) . At this later stage furthermore , the activity of only 8% ( n = 3/35 ) of neurons was exclusively plateau-like , with the remaining 25 . 5% ( n = 9/35 ) of cells expressing a mixed oscillatory/plateauing phenotype . These proportions of the different pacemaker cell types were significantly different between the two embryonic stages ( chi-square test , p<0 . 001 ) . 10 . 7554/eLife . 16125 . 007Figure 5 . Developmental changes in inspiratory network pacemaker composition . ( A ) Bar histograms indicating the mean proportions ( as% ) of individual pacemaker neuron subtypes recorded with patch-clamp at E16 . 5 ( n = 27 ) and E18 . 5 ( n = 35 ) . Plateau bursting neurons , black shading; mixed plateau/oscillatory bursters , dark gray; oscillatory bursting neurons , light gray . ( B ) Calcium imaging protocol ( see also Figure 1 ) to test the sensitivity of the preBötC pacemaker neuron subpopulation to INaP blockade at E16 . 5 and E18 . 5 . Calcium transients in 12 monitored inspiratory neurons at E16 . 5 in control conditions ( left column ) , after synapse blockade ( middle column ) , then under additional exposure to 10 µM riluzole ( right column ) . Black traces correspond to non-pacemaker neurons ( i . e . , inactive under cocktail ) ; the blue trace denotes a pacemaker neuron whose spontaneous activity was subsequently blocked under riluzole ( i . e . , an INaP-dependent pacemaker ) ; the red trace corresponds to a pacemaker neuron that remained rhythmically activity under both cocktail and riluzole ( i . e . , an ICAN-dependent pacemaker ) . ( C ) Bar histograms representing the proportions of riluzole-sensitive ( blue bar ) vs riluzole-resistant ( red bars ) neurons detected at E16 . 5 and E18 . 5 . Note the higher proportion of riluzole-sensitive ( INaP-dependent ) pacemaker neurons at the later embryonic stage . DOI: http://dx . doi . org/10 . 7554/eLife . 16125 . 007 To discriminate between INaP-dependent and ICAN-dependent pacemaker neuron activities in the two age groups , we compared sensitivity to Ril exposure in multiple-cell calcium imaging recordings at E16 . 5 and E18 . 5 ( Figure 5B ) . It should be noted here that we were unable to use complementary FFA application in these imaging experiments because the ICAN blocker’s effect on overall cell calcium signaling leads to a reduction of any initially detectable fluorescent changes , thereby preventing unequivocal interpretation . Of 32 pacemaker neurons identified at E16 . 5 , 19 were sensitive to Ril exposure ( Figure 5C , left histogram ) indicating that the preBötC pacemaker population contains significant proportions ( ratio 60:40 ) of both INaP and presumed ICAN at this earlier embryonic age . In contrast , at E18 . 5 , pacemaker neurons were found to be predominantly sensitive to Ril ( 22/27 neurons; Figure 5C , right ) , indicating that INaP is now the effective rhythmogenic mechanism in the majority of the pacemaker population at this older stage . The proportions of INaP- and ICAN-dependent pacemaker types found at the two developmental stages examined , despite not being statistically different in our imaging data ( chi-square , p= 0 . 1 ) , nevertheless , expressed a tendency that corresponded to our findings from patch-clamp recordings ( see Figure 5A ) . Therefore , altogether these data support the conclusion that at E16 . 5 both types of pacemakers ( riluzole sensitive INaP-dependent and riluzole-insensitive ICAN-dependent ) are present in comparable proportions , while 2 days later in prenatal development , the same preBötC cell subset is mainly comprised of INaP-dependent pacemaker neurons . In principle , two developmental processes ( or their combination ) could underlie the different distributions of inspiratory pacemaker subtypes at different embryonic ages . One possibility is that ICAN-dependent pacemakers progressively switch to an INaP-dependence as their membrane properties mature , thus involving a modification in the balance between the two conductances at the single cell level . Another alternative possibility is that ICAN-dependent pacemaker neurons disappear progressively with embryo maturation and are replaced by a distinct INaP-dependent pacemaker population that emerges over the same period . To directly test the former possibility ( also see below ) , a computational approach was used to assess the outcome of changing the proportion of INaP and ICAN distribution on the discharge of a previously reported model inspiratory neuron ( Table 1 , Figure 6A; Toporikova et al . , 2015 ) . In a first trial , the total persistent sodium conductance ( gNaP ) of the artificial cell's membrane was held constant at 2 . 5 nS while the total conductance for ICAN channels ( gCAN ) was made variable . With gCAN at 2 . 5 nS , the model neuron produced repetitive , long-duration oscillations and burst discharge events ( Figure 6B ) that strongly resembled the plateau-like rhythmic activity recorded from biological pacemaker neurons ( c . f . , Figure 2A ) . In contrast , when gCAN was reduced to 0 nS , higher frequency short-lasting bursts were now produced ( Figure 6D ) in a manner that was strikingly similar to the oscillatory pacemaker phenotype observed in vitro ( c . f . , Figure 2B ) . On the other hand , a hybrid model activity pattern ( Figure 6C ) that closely resembled the mixed biological pacemaker phenotype ( c . f . , Figure 2C ) was observed when gCAN was set at an intermediate conductance value of 1 nS . 10 . 7554/eLife . 16125 . 008Figure 6 . The balance between simulated ICAN and INaP in silico can determine pacemaker discharge phenotype . ( A ) Computational model of an inspiratory pacemaker neuron with the main burst-generating currents ICAN and INaP highlighted in yellow . For other abbreviations , see Material and methods and ( Toporikova et al . , 2015 ) . ( B–D ) Model neuron voltage traces obtained with INaP conductance ( gNaP ) held constant at 2 . 5 nS and gCAN set to different steady state values ( indicated in orange ) . Depending on gCAN magnitude , the model neuron generated plateau bursting discharge ( B , gCAN = 2 . 5 nS ) , a mixed firing pattern ( C , gCAN = 1 nS ) or a rhythmic oscillatory burst pattern ( D , gCAN = 0 nS ) . ( E ) Graph showing the type of model neuron discharge as a function of the proportions of gNaP and gCAN magnitudes . The gray dashed lines correspond to the values of constant gNaP and gCAN used , respectively , in the variable gCAN simulations illustrated in ( B-D ) and in complementary simulations in which INaP was varied with gCAN held constant ( data not shown ) . Depending on the balance between gNaP and gCAN amplitudes , the model neuron could exhibit the three different patterns observed with patch-clamp recordings in vitro . DOI: http://dx . doi . org/10 . 7554/eLife . 16125 . 008 Comparable results were also found when the opposite simulation paradigm was applied , whereby gCAN was held constant ( at 1 . 5 nS ) while gNaP was now varied between 0 and 5 nS ( data not shown ) . Here again , as the proportion of artificial INaP conductance relative to gCAN was increased , the model pacemaker neuron switched from inactive to plateauing states , then transcended a mixed plateau/oscillatory state until eventually the transition to a regularized oscillatory bursting condition occurred ( Figure 6E ) . These in silico findings thus support the possibility that differences in proportion of the two conductances INaP and ICAN could underlie the different discharge patterns expressed by actual preBötC pacemaker neurons . Additionally , they suggests that the developmental process responsible for the difference in functional composition of the inspiratory pacemaker population at E16 . 5 and E18 . 5 could involve a switch in this proportion , and thereby resultant rhythmic burst patterning , within individual neurons . The change in pacemaker neuron conductance proportions during late embryonic development led us to also ask whether this transition is associated with age-dependent changes in the overall mechanism by which preBötC circuitry generates rhythmic output . To assess this possibility , we examined how overall network rhythm generation is affected by blockade of either INaP or ICAN , or both , at the two studied embryonic ages . Using transverse brainstem slice preparations , we recorded network electrical activity in control conditions and in the presence of Ril or/and FFA . In these experiments , where entire network activity was monitored , we used riluzole at 20 µM to ensure blockade of INaP throughout the network . It should be noted , however , that very similar results were obtained with Ril applied at 10 µM . At E16 . 5 , application of 50 µM FFA ( Figure 7A , left panel ) or 20 µM Ril ( Figure 7A , right panel ) significantly decreased the frequency of the ongoing spontaneous inspiratory rhythm by 46 ± 9% ( t-test , p= 0 . 006 ) and 89 . 7 ± 5% ( Mann-Whitney test , p<0 . 001 ) , respectively ( Figure 7B , left histogram bars ) . Unexpectedly , however , the same drug treatments performed at E18 . 5 had no significant effect ( t-test , p= 0 . 1 and 0 . 4 for FFA and Ril treatments , respectively ) , with the frequencies of the ongoing preBötC rhythms in each case remaining unchanged under either FFA ( Figure 7C , left middle trace ) or Ril ( Figure 7C , right middle trace ) application ( Figure 7D , left bars ) . Furthermore , when the drugs were applied concomitantly , rhythmic preBötC activity was fully blocked at E16 . 5 ( Figure 7A , lower traces; 7B , right histogram bars; Mann-Whitney test , p<0 . 001 ) , but rhythmic activity persisted at E18 . 5 ( Figure 7C , lower traces ) albeit at a significantly reduced cycle frequency ( Figure 7D , right bars; Mann-Whitney test , p<0 . 01 ) . Finally , to confirm that the effects involving FFA were not due to target actions other than on ICAN ( Guinamard et al . , 2013 ) , we performed a series of equivalent experiments using 10 µM 9-phenanthrol instead of 50 µM FFA: comparable results at E16 . 5 ( i . e . , a frequency decrease ( by 39 ± 15%; t-test , p= 0 . 05 ) when applied alone and full blockade of rhythmic activity when co-applied with Riluzole; n = 3 ) and E18 . 5 ( no significant effect on frequency when applied alone ( t-test , p= 0 . 1 ) and persistence of rhythmic activity but at a lower frequency ( 63 ± 2%; t-test , p<0 . 001 ) when co-applied with Riluzole; n = 6 ) were obtained ( data not shown ) . To ensure that the apparent reduced sensitivity of the network rhythm at E18 . 5 to blockers was not due to decrease in their diffusion into the older tissue , we repeated these experiments using 300 µm instead of 450 µm slices . However , the finding that the expression of rhythmic activity persisted in such thinner preparations in the presence of FFA + Ril , albeit at a lower frequency compared to control conditions ( 51 . 2 ± 15%; n = 6; Mann-Whitney test , p = 0 . 02 ) argues against this possibility . Altogether , these data therefore show that at E16 . 5 , inspiratory network rhythmogenesis requires co-activation of INaP and ICAN , whereas later at E18 . 5 , blockade of either conductance fails to fully prevent the generation of rhythmic activity . This in turn indicates that a developmental change has indeed occurred in the mechanistic basis for respiratory network rhythm generation , which evidently is linked to a less predominant role played by endogenous pacemaker neurons at the older embryonic stage . 10 . 7554/eLife . 16125 . 009Figure 7 . Mechanisms for inspiratory network rhythmogenesis change during embryonic development . ( A ) Schematic representation of an in vitro slice preparation used for electrophysiological recording of spontaneous preBötC network activity ( integrated traces , Int preBötC ) in control conditions ( top traces ) or in the presence of either FFA ( 50 µM , middle trace , left ) or Ril ( 20 µM , middle trace , right ) alone or in combination ( bottom traces ) at E16 . 5 . ( B ) Bar histograms representing the percentages of cycle frequency change ( mean ± SEM ) compared to control under the different pharmacological conditions illustrated in A . Asterisks indicate significant differences ( t-test , p<0 . 05 ) ; n = number of experiments in each case . ( C , D ) same layout as ( A , B ) for experiments performed at E18 . 5 . Note that unlike at E16 . 5 where both FFA and Ril significantly reduced rhythmogenesis , blockade of either ICAN or INaP alone did not significantly perturb rhythm generation at E18 . 5 . Also , blockade of both conductances at E16 . 5 was sufficient to completely suppress rhythm generation , but not at E18 . 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 16125 . 009 To further explore this hypothesis , we sought evidence for a possible developmental increase in the role played by synaptic transmission in network rhythmogenesis by measuring the amplitude of the glutamatergic synaptic drive for inspiratory burst production in individual neurons recorded from slices at E16 . 5 and E18 . 5 . Inspiratory cells were first identified in current clamp conditions when their spontaneous burst discharges occurred in phase with overall preBötC network activity . Cells were then held at −50 mV in voltage clamp and the synaptic currents generated during up to 15 consecutive inspiratory bursts were monitored and measured ( Figure 8A , B , upper traces ) . The synaptic current magnitudes obtained at E16 . 5 and E18 . 5 ( red traces in lower Figure 8A and B , respectively ) were found to be significantly different ( Mann-Whitney test , p<0 . 001 ) , with the mean amplitude value obtained for 17 neurons at E18 . 5 ( 55 ± 0 . 9 nA ) being >3 fold greater than that measured in 14 cells at E16 . 5 ( 15 ± 0 . 9 nA ) , ( Figure 8C ) . The finding that excitatory post-synaptic currents do indeed appear to increase developmentally supports the hypothesis that the embryonic inspiratory network operates through different age-dependent strategies: one based principally on a subset of neuronal pacemakers and another in which overall circuit connectivity also plays a critical role . 10 . 7554/eLife . 16125 . 010Figure 8 . Developmental increase in burst-generating synaptic excitation amongst inspiratory network neurons . Integrated preBötC network activity ( top traces , Int preBötC ) with simultaneous intracellular recordings of membrane current fluctuations ( lower traces ) in rhythmically active preBötC neurons under voltage clamp ( holding potential set at −50 mV ) obtained at E16 . 5 ( A ) and E18 . 5 ( B ) . Expanded bottom traces in A and B show the excitatory synaptic currents ( left side in each case , single burst cycle; right side in red , average of 15 cycles ) recorded from individual inspiratory neurons during rhythmic network activity . Dashed lines indicate baseline and maximum amplitude levels of the synaptic currents . Note their larger amplitude at E18 . 5 for both single cycle and averaged events . ( C ) Bar histograms representing the mean maximal amplitude ( ± SEM ) of excitatory synaptic current underlying burst discharge measured in 14 neurons at E16 . 5 ( unshaded bar ) and 17 neurons at E18 . 5 ( gray bar ) . The asterisk indicates significant difference ( p< 0 . 001 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16125 . 010
Here , we describe for the first time distinct pacemaker properties expressed by a subpopulation of inspiratory neurons in the embryonic preBötC network . As also found in early postnatal rodents ( Thoby-Brisson and Ramirez , 2001; Del Negro et al . , 2002; Pena et al . , 2004; Del Negro et al . , 2005 ) , two major types of prenatal preBötC pacemaker neurons are present: one whose bursting activity depends mainly on the activation of a Ril-sensitive , voltage-dependent persistent sodium current ( INaP ) and another that is dominated by a FFA-sensitive , Ca2+-activated nonspecific cationic current ( ICAN ) . Furthermore , a proportion of pacemaker neurons express bursting activity that arises from both mechanisms , indicative of an intermediary pacemaker phenotype . We also found that in the embryonic network , the two main pacemaker mechanisms produce a depolarizing potential drive with distinctly different durations and cycle frequencies - ICAN is prevalent in generating long and slowly repeating plateau-like activity , while INaP tends to underlie shorter and more rapid burst discharge - thus indicating that the pacemaker subtypes are distinguishable according to both the conductance mechanisms responsible for the initiation of their bursts as well as for burst termination . Burst terminating mechanisms have not yet been fully identified in neonatal inspiratory network neurons , but it has been suggested that Na- and ATP-dependent outward currents play a key role ( Krey et al . , 2010 ) together with INaP and ICAN inactivation/deactivation processes ( Del Negro et al . , 2002; Del Negro et al . , 2005 ) and the activation of voltage- and calcium-dependent K+ conductances ( Butera et al . , 1999 ) . Additionally , we observed that plateau-like inspiratory pacemaker bursting is considerably more prevalent at earlier embryonic ages ( representing ~50% of the total pacemaker population ) but is then largely superseded by faster , short-duration oscillatory bursting as development progresses . Perhaps unsurprisingly , the shape , duration and regularity of inspiratory pacemaker bursts occurring at E18 . 5 , and in contrast to the considerably slower , yet robust , plateauing activity expressed at E16 . 5 , more closely resemble the discharge patterns expressed by preBötC pacemakers in the neonatal animal ( cf . , Pena et al . , 2004; Del Negro et al . , 2005 ) . It is likely , therefore , that together with developmental changes in the balance of membrane conductances engaged in initiating burst-driving potentials ( see below ) , the conductance mechanisms required for their appropriate cycle-by-cycle termination in the postnatal respiratory network are also being established and refined at late embryonic stages . Our cell patch clamp and imaging data associated with differential pharmacological manipulation revealed that both INaP- and ICAN-dependent pacemaker neuron subtypes are equally present at E16 . 5 . However , the proportion of INaP-dependent ( Ril-sensitive ) pacemakers increases significantly by E18 . 5 , with FFA-sensitive or Ril-resistant ( therefore presumed ICAN-dependent ) cells being rarely discernible at this later stage . Significantly , previous studies have reported that cadmium-sensitive ( therefore presumed ICAN-dependent ) pacemaker neurons are more readily detected in the preBötC network of juvenile mice ( P8 - P10 ) than at neonatal ( P0 – P5 ) stages ( Pena et al . , 2004 ) where they were found to represent < 1% of recorded inspiratory cells compared to 7 . 5% in the older animals ( Del Negro et al . , 2005 ) . These observations are therefore consistent with our own finding that in a developmental period around the time of birth embryonic ICAN-dependent inspiratory pacemakers are much less prevalent than INaP-dependent ones . However , the functional significance of the re-emergence of the ICAN-dependent pacemaker subtype during the second post-natal week ( Pena et al . , 2004; Del Negro et al . , 2005 ) remains unknown . The possibility that the developmental changes in the discharge patterns of individual inspiratory pacemaker neurons could be a direct consequence of changes in the balance between INaP and ICAN expression was supported by our simulation data . Increasing the strength of membrane gNaP relative to gCAN ( or vice versa ) within a single model pacemaker neuron could reversibly switch its spontaneous rhythmogenic activity from slow , plateau-like drive potentials as found at E16 . 5 to the accelerated , short-duration oscillatory bursting observed at E18 . 5 . Whether an equivalent developmental alteration in channel balance from an ICAN to INaP predominance occurs at the single cell level in real embryonic pacemaker neurons , or that ICAN-dependent pacemakers within inspiratory network are replaced by a de novo population of INaP-dependent pacemakers , remains unknown . Although such processes are not mutually exclusive , the finding that a small group of our in vitro recorded pacemaker cells could express a mixture of both plateau- and oscillatory-based discharge patterns , as similarly produced by our model neuron with intermediate gNaP and gCAN proportions ( see also Dunmyre et al . , 2011 ) , was indicative of a mechanistic switch that was transitioning within individual neurons . What role does the embryonic pacemaker neuron subpopulation play in respiratory network rhythmogenesis ? Two observations in the present study are relevant to this question . First , pharmacological attenuation of ICAN and INaP at E16 . 5 completely blocked rhythmic preBötC network activity , whereas the same treatment of slices at E18 . 5 reduced , but failed to completely suppress ongoing network activity . Notwithstanding the fact that the vast majority of inspiratory network neurons , including pacemakers and non-pacemakers alike , possess both NaP and CAN conductances ( Del Negro et al . , 2011 ) and are therefore affected by our drug treatments , these findings indicate that a fundamental developmental change had occurred whereby network rhythmogenesis has shifted from an obligatory pacemaker-driven mechanism to one in which pacemaker neurons were no longer uniquely essential . Second , as shown by in vitro experimental data , the strength of excitatory glutamatergic synaptic transmission is significantly stronger throughout the preBötC circuitry of E18 . 5 compared to E16 . 5 slices . On this basis , therefore , we propose that at earlier embryonic stages , pacemaker neurons play an essential role in network rhythm generation but that later in the prenatal period , the contribution of glutamatergic synaptic interactions becomes increasingly important so that nearer the time of birth , both cellular pacemaker properties and synaptic excitation are critical to network rhythmogenesis . Interestingly , such a developmental plasticity occurring in the late embryo embodies essential features of both the 'pacemaker neuron' ( Pena et al . , 2004; Ramirez et al . , 2011 ) and 'group-pacemaker' hypotheses ( Rekling and Feldman , 1998; Pena et al . , 2004; Feldman and Del Negro , 2006 ) that have been proposed to account for respiratory rhythm generation in the early postnatal rodent ( for review , see Feldman et al . , 2013 ) . In conclusion , the present study provides the first description of a heterogeneous population of neurons bestowed with pacemaker properties in the embryonic respiratory network . The composition of this population , both in terms of pacemaker subtypes and their implication in overall network rhythmogenesis undergo significant maturational changes during the prenatal period . As yet , we have not been able to establish a direct link between these two transitional processes , nor has our data allowed us to determine the specific contribution that each pacemaker subtype makes to inspiratory network output as a whole . However , our in vitro results from differential pharmacology lead to the conclusion that the pacemaker cell subpopulation is exclusively responsible for network rhythmogenesis at earlier embryonic stages , but then assumes a cooperative role with recurrent synaptic excitation throughout the wider network immediately prior to birth . A better understanding of such prenatal developmental processes could in turn provide relevant insights into clinical aspects of early post-natal respiratory disorders .
Brainstem transverse slice preparations that isolated the preBötC network were obtained from mouse embryos using the following procedures: pregnant mice were killed by cervical dislocation on embryonic day ( E ) 16 . 5 or E18 . 5 , the day of the plug being considered as E0 . 5 . Embryos were excised from the uterine horns and their individual uterine bags , and prior to experimental use , were either placed in artificial cerebrospinal fluid ( aCSF , see below ) that was continuously supplemented with oxygen at a temperature not exceeding 24°C for E16 . 5 embryos or were manually stimulated to activate spontaneous breathing behavior and then kept under a warm heating light for E18 . 5 embryos . Slice preparations were dissected in cold oxygenated aCSF composed of ( in mM ) : 120 NaCl , 8 KCl , 1 . 26 CaCl2 , 1 . 5 MgCl2 , 21 NaHCO3 , 0 . 58 NaH2PO4 , 30 glucose , pH 7 . 4 . In a first step , the hindbrain was isolated from the embryo's body by a rostral section made at the level of the rhombencephalon and a caudal section made at the level of the first cervical roots . The isolated hindbrain was then placed in a low melting point agar block and carefully oriented to enable serial transverse sectioning in a rostral-to-caudal direction using a vibratome ( Leica , Germany ) . A 450-µm-thick slice , with its anterior limit set 250–300 µm caudal to the more caudal extension of the facial nucleus , was isolated . Other anatomical landmarks , such as the wide opening of the fourth ventricule , the presence of the inferior olive , the nucleus ambiguus and the hypoglossal nucleus were also used to determine the appropriate sectioning axis ( as also referred to in newborn mice by Ruangkittisakul et al . , 2011 ) . Such slice preparations encompass the region containing a significant portion of the preBötC network capable of spontaneously generating fictive rhythmic inspiratory activity ( Thoby-Brisson et al . , 2005; Toporikova et al . , 2015 ) are devoid of the more rostral Bötzinger complex and contain only the most rostral region of the more caudal rVRG . On this basis , therefore , all neuronal populations examined were considered to be integral components of the preBötzinger network . Slices were then transferred to a recording chamber , rostral surface upwards , and continuously superfused with oxygenated aCSF at a temperature of 30°C . Preparations were allowed to recover from the slicing procedure and subsequent calcium indicator loading ( see below ) for 20–30 min before any recording sessions commenced . Global preBötzinger network activity in slice preparations was recorded using glass micropipettes ( tip diameter 80–100 µm ) positioned on the upper surface of the slice in a region ventral to the nucleus ambiguus where respiratory circuitry is located . The micropipettes were fabricated from aCSF-filled borosilicate glass tubes ( Harvard Apparatus , Germany ) broken at the tip and used as suction electrodes connected to a high-gain amplifier ( AM Systems , USA ) . The collected signals were filtered ( bandwidth 3 Hz - 3 kHz ) , integrated ( time constant 100 ms; Neurolog , Digitimer , England ) and stored on a computer via a Digidata 1440 interface and PClamp10 software ( Molecular Devices , USA ) . The stored files were analyzed off-line . Whole cell patch-clamp recordings of pacemaker neurons were performed under visual control using differential interference contrast . Patch pipettes were fabricated with borosilicate glass tubes using a puller ( Sutter Instrument , USA ) and filled with a solution composed of ( in mM ) : 140 K-gluconate acid , 1 CaCl2 . 6H2O , 10 EGTA , 2 MgCl2 , 4 Na2ATP , 10 HEPES , pH 7 . 2 and had tip resistance of 5–7 MΩ when filled with this solution . Electrophysiological signals were recorded using an Axoclamp 2A amplifier ( Molecular Devices ) and the same digitizing interface and software as stipulated above . Neurons were selected for patch-clamp recording on the basis of their rhythmic impulse discharge that had to be in phase with the extracellularly-recorded population activity both in control conditions and under conditions where network chemical synapses were subsequently blocked ( by a cocktail of pharmacological agents , see below ) . Recorded neurons were located close to the extracellular macroelectrode or in the contralateral preBötC network that had been initially identified in control conditions as exhibiting rhythmically organized fluorescent changes in phase with the electrophysiological recording of its contralateral network partner . Only recordings lasting longer than 8 min were considered for the burst parameter analysis presented in Figure 2 . I-V curves were plotted by measuring the amplitude of the membrane current evoked by 0 . 5 s duration voltage steps from −100 to +30 mV . Current amplitude was measured at steady state , 400 ms after the onset of a given voltage step . The magnitude of the synaptic drive underlying individual inspiratory bursts that results from the summation of glutamatergic excitatory post-synaptic currents was measured as the maximal amplitude of the envelop ( i . e . , the difference between the horizontal dashed line pairs in Figure 8 ) and averaged for 15 consecutive bursts per neurons in slices at E16 . 5 and at E18 . 5 . The method used for calcium imaging that allows monitoring the activity of multiple preBötC neurons simultaneously has been fully described elsewhere ( Thoby-Brisson et al . , 2005; Toporikova et al . , 2015 ) . Briefly , slice preparations were first incubated in the dark for 45 min at room temperature in a solution of oxygenated aCSF containing the cell-permeable calcium indicator dye Calcium Green-1 AM ( 10 µM; Life Technologies , France ) . After dye loading , preparations were positioned rostral side up in the recording chamber . Before subsequent image acquisition , a 30 min delay was observed to wash out excess dye and enable the preparation to stabilize in oxygenated aCSF at 30°C . Fluorescent signals were captured through a FN1 upright microscope ( Nikon , Japan ) equipped with an epifluorescent illumination system and a fluorescein filter coupled to an Exiblue camera ( QImaging , Surrey , Canada ) . Images ( 100 ms exposure time ) were acquired over periods lasting 120 s and analyzed using customized software provided by Dr N . Mellen ( Mellen and Tuong , 2009 ) . Pharmacological agents were obtained from Sigma or Tocris ( France ) and dissolved in oxygenated aCSF for bath-application for 15 – 30 min at their final concentration of 10–20 µM Riluzole ( Ril ) to block persistent sodium currents ( INaP ) , and 50 µM Flufenamic acid ( FFA ) to block calcium-dependent non-specific cationic currents ( ICAN ) . Although such a relatively low FFA concentration was used in order to limit potential non-specific effects ( Guinamard et al . , 2013 ) , complementary experiments were also performed in which CAN conductances were blocked by 10 µM 9-phenanthrol in place of FFA . The effects of all drugs on rhythm generation were assessed at the end of each exposure period . To synaptically isolate respiratory network neurons , we applied a cocktail of synaptic blockers containing 20 µM 6-cyano-7-nitroquinoxaline-2 , 3-dione ( CNQX ) , 10 µM DL-2-amino-5-phosphonovaleric acid ( AP5 ) , 1 µM strychnine and 10 µM bicuculline . Burst frequency and the amplitude and duration of the depolarizing drive potentials underlying bursts are given as mean ± SEM . Statistical significance was assessed by Student’s t-test , or Mann-Whitney . Mean values were considered as statistically different when p<0 . 05 . Differences in pacemaker type proportions were assessed using the Chi-square test and considered significant for p<0 . 05 . | Babies need to start breathing immediately when they are born . Researchers have detected rhythmic movements in the fetus that are related to breathing , which supports the idea that the nervous system circuits needed for breathing are still being established and refined shortly before birth . Neurons in a part of the brain called the brainstem control the muscles that generate the movements required for breathing . These neurons are organized into groups , with each group forming an independent network that can carry information in the form of electrical signals . However , it is not clear how these networks form and operate before birth . Chevalier et al . tracked electrical activity in slices of brainstems from mouse embryos . The experiments show that these embryos already have ‘pacemaker’ neurons that can drive rhythmic activity in the networks of neurons related to breathing . There are several types of pacemaker neurons that produce different patterns of electrical firing . The amount of each type of pacemaker in the brainstem changes in the later stages of the pregnancy . The experiments also show that the way in which pacemaker neurons control the networks of breathing-related neurons changes as the embryo develops . Early on in development , pacemaker neurons play an essential role in generating rhythms in the other neurons . However , in older embryos , the connections between each neuron in the breathing network become more important . Further work is now needed to map out the exact sequence of events in embryos that allow mice to breathe as soon as they are born . This could help us to develop therapies for human babies that are born with breathing difficulties . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2016 | Development of pacemaker properties and rhythmogenic mechanisms in the mouse embryonic respiratory network |
Organismal phenotypes frequently involve multiple organ systems . Histology is a powerful way to detect cellular and tissue phenotypes , but is largely descriptive and subjective . To determine how synchrotron-based X-ray micro-tomography ( micro-CT ) can yield 3-dimensional whole-organism images suitable for quantitative histological phenotyping , we scanned whole zebrafish , a small vertebrate model with diverse tissues , at ~1 micron voxel resolutions . Micro-CT optimized for cellular characterization ( histotomography ) allows brain nuclei to be computationally segmented and assigned to brain regions , and cell shapes and volumes to be computed for motor neurons and red blood cells . Striking individual phenotypic variation was apparent from color maps of computed densities of brain nuclei . Unlike histology , the histotomography also allows the study of 3-dimensional structures of millimeter scale that cross multiple tissue planes . We expect the computational and visual insights into 3D cell and tissue architecture provided by histotomography to be useful for reference atlases , hypothesis generation , comprehensive organismal screens , and diagnostics .
Histology has been used for over a century to visualize cellular composition and tissue architecture in millimeter- to centimeter-scale tissues from diverse multicellular organisms ( Virchow , 1860 ) . Its diagnostic power is dependent on the detection and description of changes in cell and tissue architecture . Normal and abnormal cytological features indicative of physical , inflammatory , and neoplastic causes of disease are readily distinguished using established staining procedures ( Kumar et al . , 2015 ) . Despite its power , histology has practical limitations in throughput and quantitative phenotyping of cell and tissue volume and shape . Physical sectioning introduces tissue loss and distortions that compromise complete reconstruction of tissues from serial histology sections ( Arganda-Carreras et al . , 2010 ) . The technical demands and physical properties of paraffin blocks limit slice thickness to ~5 µm , leading to incomplete sampling and imperfect visualization of elongated structures such as vessels and large cells that extend beyond the section thickness . As a result , only a small fraction of any given tissue sample is studied in histology . Moreover , it is intractable to generate complete sets of sections for large numbers of whole organisms ( as needed for a genetic or chemical screen ) and impractical to align them for volumetric analysis . A routine method for histological phenotyping that avoids these problems and enables comprehensive three-dimensional ( 3D ) analysis of whole organisms would transform large-scale studies . X-ray micro-tomography ( micro-CT ) is a potential means of achieving complete , 3D histological phenotyping for large numbers of specimens . Micro-CT is commonly used to study hard , mineralized tissues like bones and fossils ( Donoghue et al . , 2006; Ketcham and Carlson , 2001 ) . Soft-tissue imaging typically requires contrast-enhancing , heavy-metal stains such as osmium tetroxide , iodine , phosphotungstic acid ( PTA ) , or gallocyanin-chromalum ( Betz et al . , 2007; Metscher , 2009a; Metscher , 2009b ) . Phase-based synchrotron imaging of unstained samples allows the study of development in live whole Xenopus embryos over time ( Moosmann et al . , 2013 ) and of dense human cerebellar tissue ( Töpperwien et al . , 2018 ) . Commercial micro-CT devices have been used to interrogate soft-tissue structure in diverse stained specimens ( Badea et al . , 2008; Batiste et al . , 2004; Cheng et al . , 2016a; Metscher , 2009a; Staedler et al . , 2013 ) , but these devices utilize polychromatic , low-flux X-ray tube sources that limit resolution and throughput . Synchrotron X-ray sources have monochromatic , high-flux beams ( Winick , 1994 ) that allow rapid imaging of mm-scale samples such as insects ( Betz et al . , 2007; Mizutani et al . , 2013; van de Kamp et al . , 2018 ) , vertebrate embryos ( Khonsari et al . , 2014; Raj et al . , 2014 ) , zebrafish ( Seo et al . , 2015 ) , and mouse somatosensory cortex ( Dyer et al . , 2017 ) . Micro-CT of larger samples such as whole mice or whole mouse organs have lacked adequate tissue contrast and/or resolution ( ~10 µm3 ) for histological phenotyping ( Busse et al . , 2018; Hsu et al . , 2016 ) . Conversely , nano-CT enables imaging at higher resolution ( ~100 nm3 ) but does not provide sufficient field-of-view for mm-scale samples . To our knowledge , no existing method has the combination of throughput , resolution , field-of-view , and soft-tissue contrast necessary for whole-organism 3D phenotypic screens that are inclusive of histopathological evaluation . Here , we present a 3D quantitative histological analysis of features of whole fixed and stained zebrafish , at sub-micron resolution , using synchrotron micro-CT optimized for soft tissue differentiation ( histotomography ) . We chose zebrafish , the largest established vertebrate model to fit in our field-of-view ( juveniles are <3 mm in diameter ) , to determine conditions for attaining a degree of tissue contrast required for histopathological analysis across a range of tissues . X-ray histotomography as developed here is pan-cellular , like histology , allowing for the creation of histology-like virtual sections , but unlike histology , can be used to study millimeter-scale phenotypes involving convoluted and branched structures such as gills , gut , vessels , and nerve tracts . Histotomographic images provide a mechanism for quantitative histopathological analysis that facilitates the objective and reproducible study of volume , shape , and texture of cells across organisms and tissue types . Histotomography’s potential throughput and analytical power may be applied to computational histological phenotyping of whole organisms to probe the diversity of organismal tissue architecture and relationships between genotype , environment , and microanatomy ( Austin et al . , 2004; Cheng et al . , 2016b; Varshney et al . , 2013 ) .
Synchrotron based micro-CT was chosen based on two demands of phenome projects: the potential to achieve resolutions and contrast required for histological evaluations , and high-throughput potential for phenotyping whole organisms . Fine-tuning of X-ray energies and bandwidth , and adjustment of sample-to-scintillator distance , were used to optimize volumetric reconstructions of optically-opaque zebrafish at isotropic resolutions ( equal in all three dimensions ) for histopathological interpretations . Synchrotron X-ray flux is orders of magnitude more brilliant than commercial tube sources , allowing imaging times short enough for high sample-number imaging projects ( Winick , 1994 ) . Micro-CT studies were performed on the sector two bending magnet , hutch B ( 2-BM-B ) of the Advanced Photon Source at Argonne National Laboratory for both single-energy and multi-energy acquisitions ( Figure 1 ) . Beam energies for monochromatic imaging were optimized for contrast-to-noise ratio across a range of sample diameters and concentrations of tungsten ( Figure 1—figure supplement 1 ) . A beam energy of 13 . 8 keV was used for the smaller , larval samples and 16 . 2 keV for juveniles due to their greater thickness ( see Materials and methods ) . Sample-to-scintillator distance was adjusted to optimize the magnitude of phase contrast-based edge enhancement ( Figure 1—figure supplement 2 ) . A 30 mm sample-to-scintillator distance provided sufficient edge enhancement to achieve histology-like contrast . For whole-organism imaging , the zebrafish were vertically translated to capture segments over the full length of the sample . Each zebrafish segment was reconstructed and the multiple segments stitched into a single 3D volume ( Video 1 ) . Large-scale phenotyping studies , such as those focusing on series of mutants or chemical exposures , require short imaging times to facilitate throughput . Monochromatic acquisitions took ~20 min , ~36 fold faster than acquisition by commercial tabletop sources ( e . g . , ~720 min for Xradia 500 series machines ) ( Metscher , 2009b ) . The use of polychromatic ‘pink-beam’ increases X-ray flux , which greatly reduces exposure times . Each pink-beam acquisition took ~20 s , ~60 fold faster than monochromatic acquisition and ~2 , 000 fold faster than commercial sources . Monochromatic acquisitions have better image quality than pink-beam , as defined by signal-to-noise ratio and pixel intensity profile ( Figure 1—figure supplement 3 ) . Notably , no image degradation over years of repetitive synchrotron imaging was detected ( Lin et al . , 2018 ) . This high degree of sample stability across multiple scans suggests the potential use of pink-beam pre-screening followed by higher resolution monochromatic reacquisition . Digital zebrafish were reconstructed from projections after staining all tissues with PTA . The similarity of digital slices to conventional histological outputs is demonstrated by representative transverse ( axial ) , sagittal , and coronal cross-sections for larval ( 4 days post-fertilization , dpf ) and juvenile ( 33 dpf ) zebrafish imaged at 0 . 743 and 1 . 43 µm3 isotropic voxel resolution , respectively ( Figure 2 ) . Volume renderings illustrate the orientation of individual planes of section in 3D . Full sets of cross-sections of the zebrafish in the sagittal orientation illustrate one way to use micro-CT data to phenotype full tissue volumes ( Videos 2–3 ) . Notably , the z-axis resolution of histotomography , presented here at 0 . 74 or 1 . 43 microns , is superior to the ~5 micron slice thickness of histology’s conventional paraffin sections . Increased z-axis resolution decreases the likelihood of two separate nuclei being scored as one when they are crowded , as is the case in the retinal and brain neurons of larval zebrafish . The problem of overlapping nuclei in the z-axis makes it difficult to accurately count nuclei from traditional histological images . Our x-y histotomographic resolution is comparable with that of optical microscopy using a 10X objective lens ( Figure 3 ) . When direct comparisons between histotomography and histology are desired , maximum intensity projections ( MIPs ) totaling ~5 microns in thickness can be created from image stacks . Just as in hematoxylin and eosin ( H and E ) stained sections from ~3 . 5 mm-long larval ( five dpf ) ( Figure 3A ) and ~1 cm-long juvenile ( 33 dpf ) zebrafish ( Figure 3C ) , cell and tissue types can be recognized histologically from five micron MIPS of age-matched fish ( Figure 3B and D ) . As an internal measure of resolution , we calculated the spacing of striations in skeletal muscle fibers encircling the larval swim bladder ( Figure 3—figure supplement 1 ) . The average of 293 measures = 2 . 16 μm ( SD = 0 . 55 μm ) is comparable with published values ( Burghardt et al . , 2016; Dou et al . , 2008 ) . While both traditional histology and histotomography have the resolution needed to distinguish cellular features in 2D slices , only the latter is able to reveal elongated , complex 3D tissue structures such as vessels , nerve tracts , and bones . These advantages are particularly well-illustrated by the study of gill architecture ( Figure 3C and D , insets ) . Histotomography makes it possible to interrogate primary and secondary lamellae from multiple angles without spatial distortion , revealing the delicate leaf-like structure of gill filaments on any pharyngeal arch and even bulges in epithelial cells caused by their nuclei ( Video 4 ) . Whole animal histotomographic reconstructions thus provide a level of organismal and anatomical context that is not practical using histology alone . The isotropic nature of histotomography enables reslicing in any plane of section without distortion . Dynamic reslicing ( cutting the same volume digitally in multiple planes ) allows the study of either longitudinal , perpendicular , or other planar virtual sections of the gut at cellular resolution ( Figure 4 ) . As far as we are aware , the ability to generate full sets of cross-sections of convoluted structures ( such as the gut of juvenile zebrafish ) for histological phenotyping within a single modality is unprecedented . This approach can be applied to any convoluted structure requiring thorough evaluation . Virtual sectioning and 3D rendering can be used to study any tissue in the fish , as illustrated for the neuronal cells in the eye , cartilaginous elements of the notochord , the squamous patch of the dorsum of the pharynx , nucleated red blood cells in the heart , and the pneumatic duct and goblet cells of the intestine of whole juvenile ( 33 dpf ) and larval ( five dpf ) specimens in Figure 5 and Figure 6 , respectively . In sum , our digital zebrafish allow tissues and cell nuclei to be visualized across organ systems , including the integumentary , hematopoietic , respiratory , genitourinary , musculoskeletal , gastrointestinal , cardiovascular , nervous and sensory systems ( Figure 5—figure supplement 1 , Video 5 ) . Successful histopathological analysis requires the detection of subtle differences in the characteristics of specific micron-scale structures . To probe the potential ability of X-ray histotomography to detect subtle histopathological change in a whole organism , we used a mutant , huli hutu ( hht ) , that is known to show a range of subtle to obvious histological changes across all cell types and tissues ( Mohideen et al . , 2003 ) ( Figure 6 , Video 6 ) . Reconstructions of larval hht at 0 . 743 µm3 voxel resolution allowed us to detect all of hht’s known histological changes , including nuclear fragmentation ( karyorrhexis ) that is associated with cell death , nuclear atypia ( increased size , deviation from typical ovoid shape , and irregular nuclear membrane contour ) in the gut , and tissue degeneration . Nuclear fragments are commonly two microns or less in diameter . We were able to reliably establish the absence of an easily missed , micron-scale structure , the pneumatic duct , since every virtual slice of entire fish is available at sub-micron resolution . This structure cannot be reliably assessed by histology on account of its small size and tortuous shape . Age-matched wild-type and hht fish ( 2 , 3 , 4 and 5 dpf ) can be examined , using our web-based data sharing interface , to visually track progression of the histopathological changes characteristic of hht ( Table 1 ) . Additionally , computational detection of nuclei allows global assessment of the density of nuclei across the entire animal ( second half of Video 6 ) . Due to the combination of histotomography’s field-of-view and resolution , we are able to compute features of individual cells and pattern of cells throughout the whole organism . Anatomic pathologists use cytological characteristics in qualitative tissue assessments ( Al-Abbadi , 2011; Cheng and Bostwick , 2002 ) . Quantitative assessment of these morphological attributes is best-accomplished in 3D , bringing added precision in distinguishing disease states from the range of normal tissue and cellular architecture . To evaluate the power of computational phenotyping enabled by histotomography , we combined regional segmentation with automated cell detection for characterization of the zebrafish brain ( Figure 7 ) . For example , brain cell nuclei can be distinguished from other stained objects based on shape ( elongation ) and volume . Elongation is measured as the ratio of the major axis over the minor axis of a 3D object . As anticipated , computational measurements of shape and volume of manually segmented red blood cells ( RBCs ) and motor neurons ( n = 20 each ) were different , in agreement with morphological differences that are readily apparent in histology ( Figure 8A ) . In addition , RBCs and motor neurons are more elongated than typical brain nuclei , which are more spherical . Specifically , RBCs are typically flat and oval , while motor neurons are more ‘tear drop’ shaped . Other cell differences include the volume , with motor neurons being larger than typical brain nuclei . The PTA staining patterns of the cytoplasm of RBCs and motor neurons are different in appearance from those of other cells . From the measurements shown , it is apparent that our current resolution is sufficient to readily distinguish cell types . We have assigned brain nuclei to major brain regions including the olfactory epithelia , telencephalon , diencephalon , hypothalamus , mesencephalon , metencephalon , myelencephalon , white matter , and spinal cord by cross-atlas registration ( Raj et al . , 2018; Ronneberger et al . , 2012; Wullimann and Mueller , 2005 ) . In addition , cell nuclei were automatically detected from head scans of intact fish using a supervised learning approach with a random forest classifier . Three 75 µm3 regions of neural cell nuclei were manually annotated and compared against classifier results for location and counts . The F1 score , optimized for probability threshold to balance precision and recall , showed ~90% correspondence between manual and automated detections ( Figure 7—figure supplement 1 ) . The distribution of cell nuclei in our samples corresponded well with those seen in 54 nm thick transmission electron microscopy sections ( Hildebrand et al . , 2017 ) ( Figure 7—figure supplement 2 ) . Whole-animal histotomography enables quantitative analyses of both individual cells and groups of cells . The average number of nuclei in the brain region out of 5 larvae studied was 75 , 413 ( SD = 8 , 547 ) ( Table 2 ) . This number corresponds well to the ~80 , 000 reported brain cell ROIs in similarly-aged zebrafish imaged with light sheet microscopy ( Chen et al . , 2018 ) . In contrast to the general agreement of size and elongation distributions of brain cell nuclei between individual fish ( Figure 8 ) , nuclear density varied between brain regions , as reflected by cell density distributions and 3D patterning ( Figure 9 ) . Notably , heat-maps of cell densities revealed obvious differences between five wild-type siblings . Comparatively , fish 1 and 2 exhibit a shift towards lower cell density as compared with fish 3 , 4 , and 5 , which correlates with the larger brain volumes of fish 1 and 2 ( Tables 2–3 ) . These differences in cell density may be explained by differences in exact developmental stage .
Large-scale studies of the effects of genes and environment on phenotype ( phenome projects ) are ideally comprehensive and quantitative in nature , covering all cell and tissue types across length scales . We pursued synchrotron micro-CT to enable the vision of computationally phenotyping small organisms in 3D , at a throughput and resolution that is compatible with phenome projects ( Cheng et al . , 2011 ) . Other groups have utilized X-ray micro-CT for quantitative morphometric analyses of juvenile and adult teleost fish ( Babaei et al . , 2016; Seo et al . , 2015; Weinhardt et al . , 2018 ) . Histological analysis across cell types would add significant powerful to phenotypic screens , but would require higher resolution and contrast-to-noise ratios than available at the time . This need motivated us to systematically optimize details of sample preparation , sample mounting , imaging acquisition settings , X-ray optics , image processing , and visualization parameters to make highly informative histological signatures apparent . Histotomography offers resolving power over more than four orders of magnitude , providing both anatomical and cellular detail from single images that encompass , for zebrafish , the whole organism . High-throughput adaptations would be necessary for whole-organism chemical and genetic phenome projects . Genetic phenomics may involve , for example , the study of mutants for each zebrafish protein-encoding gene ( >26 , 000 ) and non-coding functional element . By our estimates , screening 10 , 000 zebrafish mutants in replicate would require ~20 years with a monochromatic synchrotron source and , if fully optimized , <1 year using pink-beam X-rays . High-throughput primary screens by pink-beam could be followed by higher-contrast monochromatic imaging of samples that show evidence of phenotypic change . The fact that some 70% of human protein-coding genes have at least one zebrafish orthologue makes this throughput potential relevant to disease modeling and the probing of human gene function ( Howe et al . , 2013 ) . The unprecedented depth of zebrafish tissue phenotyping enabled by histotomography has the potential to enrich conclusions from mouse phenome projects ( Dickinson et al . , 2016; Hsu et al . , 2016 ) . Plastic embedding of specimens , while not essential for synchrotron micro-CT , adds sample stability ( Lin et al . , 2018 ) f image re-acquisition with improved micro-CT implementations . Optical imaging modalities such as light sheet fluorescence microscopy are ideal for in vivo studies ( McDole et al . , 2018 ) , and resolve 3D sub-micron features . Fluorescence-based imaging , however , depends on sample transparency , and for optically more opaque samples , dissection and/or physical slicing to maintain resolution without diffraction or loss of signal ( Chung et al . , 2013; Hama et al . , 2011; Susaki et al . , 2014; Watson et al . , 2017 ) . Image quality is compromised in the presence of significant melanin pigmentation , even in transparent samples with diameters of more than about a millimeter . The molecular specificity of fluorescence-based phenotyping is poorly suited for large-scale screens that require phenotyping across all cell types . Three-dimensional tissue imaging methods based on ultrathin serial tissue sections or block-face imaging have ultrastructural cellular resolution , but becomes intractable for large scale studies of specimens larger than ~1 cubic millimeter in minimum dimension . Serial block-face scanning electron microscopy has been used to image and reconstruct a mouse neocortex ( Kasthuri et al . , 2015 ) and the larval zebrafish brain at nanometer-scale ( 56 . 4 × 56 . 4 × 60 nm3 resolution from 16 , 000 sections ) ( Hildebrand et al . , 2017 ) . These studies made it possible to define precise neurological and circulatory relationships across the entire brain . However , 3D reconstructions based on serial sectioning are time-consuming enough to be infeasible for interrogating all tissues in the numbers of specimens required for toxicology or genetic screens . For example , a multi-beam scanning electron microscope , imaging a single cubic mm of tissue at 20 nm3 resolution can require ~3 months of continuous imaging ( Dyer et al . , 2017 ) . Ideal comprehensive anatomical phenotyping would include quantitative characterization of tissue architecture and cellular features in addition to measures of larger features such as organ size , vascular structure , nerve structure , and body shape . Striking individual phenotypic variation was revealed here by the computational measurement of cell density across larval zebrafish brains . This finding illustrates how histotomography’s unique combination of field-of-view and resolution can be useful for distinguishing individual from experimentally-determined phenotypic variation . Imaging and analysis of multiple specimens at multiple ages is necessary to establish a knowledge of normal statistical variation in gross and microscopic anatomy across organ systems that is needed for automated detection of ‘abnormal’ phenotypes . Comprehensive computational phenotyping will thus require the development of detailed 3D atlases . Whole specimen histotomography datasets are well-suited for automated organ and cellular level detection that can be used as a scaffold for morphological lifespan atlases , capturing the extent of normal phenotypic variation at cellular , tissue , and organismal scales . The pan-cellular nature of histotomography makes it ideal for cross-referencing with images from focused ( e . g . fluorescence-based ) modalities involving transgenic or whole-mount stained organisms; this would involve fixation , staining and histotomographic imaging of the fluorescently studied specimens . Conversely , the addition of tissue- , cell- , and/or protein-specific micro-CT stains would enable targeted analyses and cross-atlas comparisons of explicit constituents within the full context of a scanned sample or organism ( Metscher and Müller , 2011 ) . Further optimization will be needed to resolve a key characteristic of proliferating cells: mitotic chromosomes . Condensed chromosomes are about one micron in diameter , requiring true 0 . 5 micron pixel/voxel resolution for detection . Increasing field-of-view beyond ~2 . 7 mm ( used for histotomography of juveniles at 1 . 4 micron isotropic voxel resolution ) will be needed to image of the full width of mature zebrafish , and will be facilitated by larger imaging chip arrays . The work shown here approaches the ideal of comprehensive and quantitative phenotyping for tissue samples and model organisms that lie in the mm-to-cm length scale , such as the zebrafish . Indeed , X-ray histotomography allows for full volume imaging of replicate samples at cellular resolution , distinguishing neighboring cells and resolving nuclear morphologies in optically opaque larval and juvenile zebrafish specimens . Extending this work to Drosophila , C . elegans , and Arabidopsis , and to tissue samples from larger organisms such as mouse and human would enable cross-species phenotypic analyses , leading to a more universal understanding of tissue architecture and morphological phenotype . To make our data accessible to users with standard computational resources , we have developed an open-access , web-based data sharing platform , ViewTool , that allows 2D and 3D visualizations of full sample volumes . Histotomographic data are also ideal for real-time visualization using virtual reality technologies such as syGlass ( Pidhorskyi et al . , 2018 ) . Beyond data exploration , the rate of development of methods and tools for computational analysis can be enhanced by long-term accessibility of histotomographic data through an internationally available portal analogous to those used for DNA or transcriptome-based resources . Repositories of high-resolution images from a full complement of human and other organismal tissues would facilitate cross-correlations between model system and human phenotypes ( Regev et al . , 2017; Rozenblatt-Rosen et al . , 2017 ) . Such a resource would have the potential to increase analytical precision , sensitivity , reproducibility , and data sharing as we address important questions across basic and clinical sciences .
Wild-type zebrafish ( Ekkwill strain ) and the huli hutu mutant ( Mohideen et al . , 2003 ) were reared at an average temperature of 28°C in a recirculating system with a 14:10 hr light:dark cycle ( Copper et al . , 2018 ) . Fish were fed three times a day a diet consisting of brine shrimp and flake food . All fish were staged according to the zebrafish developmental staging series of Kimmel et al . ( 1995 ) . After staging , larval ( 2 , 3 , 4 , and 5 dpf ) and juvenile ( 33 dpf ) zebrafish specimens were euthanized in pre-chilled 2x Finquel ( MS-222 or tricaine-S , 400 mg/L ) solution ( Argent Chemical Laboratories , Redmond , WA ) buffered in 1% phosphate-buffered saline ( PBS ) , and fixed in chilled 10% neutral buffered formalin ( NBF ) ( Fisher Scientific , Allentown , PA ) overnight in flat-bottom containers at room temperature . To improve fixation and reduce the volume of gut contents , we starved juvenile fish for at least 24 hr for a 10 mm specimen . Fixed zebrafish specimens were rinsed 3 times in 1X PBS for 10 min followed by being submerged in 35% ethyl alcohol ( EtOH ) for 20 min at room temperature with gentle agitation . The samples were then submerged in 50% EtOH for 20 min at room temperature with gentle agitation . Specimens were stained with 0 . 3% phosphotungstic acid ( PTA ) ( diluted from a 1% w/v stock solution at a ratio of 3:7 in 100% EtOH ) for 24 hr at room temperature with gentle agitation . This metallic stain is used widely in histology and electron microscopy for staining collagen and other connective tissue ( Bloom and Aghajanian , 1968; Bulmer , 1962 ) . The complete infiltration and embedding protocol is described elsewhere ( Lin et al . , 2018 ) . Briefly , after staining , each specimen was infiltrated by sequential submerging in increasing concentrations of EtOH ( 70% , 90% , 95% , and 100% ) for 30 min intervals at room temperature with gentle agitation . The samples were then embedded in glycol methacrylate ( Polysciences , Inc , Warrington , PA ) or EMBed-812 ( Electron Microscopy Sciences , Hatfield , PA ) in Kapton tubing ( Small Parts , Inc , Logansport , IN ) . The tubing provides structural support with high thermal stability . Total sample preparation time takes 5 days . All procedures on live animals were approved by the Institutional Animal Care and Use Committee ( IACUC ) at the Pennsylvania State University . Generation of the ENU-mutagenized mutant hht was previously described ( Mohideen et al . , 2003 ) . The hht line was maintained as heterozygotes due to the recessive larval-lethal nature of the mutation . Mating was carried out by placing male and female heterozygotes in Aquatic Habitat breeding tanks with dividers the afternoon prior to egg collection . Collected eggs were disinfected in 10% Ovadine ( Syndel ) for 1 min at room temperature followed by three washes with charcoal-filtered water . Larvae were incubated at 28 . 5°C to maintain consistent speed of development . Homozygous mutant larvae were identified by a combination of gross phenotypes , including small eyes , small head , dorsally curved body , and enlarged yolk , that are easily detected under a low-power stereomicroscope at ≥3 dpf . Gross mutant phenotype at two dpf is limited to small eyes and requires screening at 40X . Synchrotron micro-CT studies were performed on the beamline 2-BM-B Advanced Photon Source at Argonne National Laboratory . The beamline’s quasi-parallel X-rays were used to acquire projection images of larval and juvenile zebrafish in sets of 2048 digital slices . After passing through the object , the X-rays impinge on a thin scintillator , which converts X-rays to optical photons that are magnified by a microscope objective lens onto a cooled CCD . The volumetric field of view was 1 . 5 mm3 for 0 . 743 µm3 voxels using a 10X objective lens , and 3 mm3 with 1 . 43 µm3 voxels using a 5X objective . A vertical stage was used to translate the sample between multiple acquisitions . For monochromatic imaging , a multilayer monochromator with bandwidth ΔE/E ~ 1 . 5% , or 200–250 eV , was used to obtain discrete X-ray energies . 1501 projections were obtained over 180 degrees ( 1 projection every 0 . 12 degrees ) with a 2048-by-2048 pixel CoolSnap HQ CCD camera ( Photometrics , AZ , USA ) . Larval and juvenile zebrafish scans were obtained at 13 . 8 keV and 16 . 2 keV , respectively . Additionally , two flat-field ( gain ) images ( one at the beginning and one at the end of the acquisition ) and one dark-field image are also acquired . Flat-field and dark-field corrections , ring artifact reduction , and image reconstruction were done using the open source TomoPy toolkit . Reconstructing the projection data generates a 3D data set comprised of a 2048-by-2048-by-2048 voxel cube . A voxel in the PTA-stained larval zebrafish has nominal 0 . 743 µm x 0 . 743 µm x 0 . 743 µm resolution , while a voxel in the juvenile zebrafish has a nominal 1 . 43 µm x 1 . 43 µm x 1 . 43 µm resolution , corresponding with larval and juvenile zebrafish fields-of-view of 1 . 5 mm x 1 . 5 mm x 1 . 5 mm and 3 mm x 3 mm x 3 mm , respectively . Whole-organism zebrafish reconstructions were created by combining series of segmental reconstructions at each vertical position along the full length of each specimen . Exposures of 400 ms per projection yielded acquisitions of ~20 min . Whole zebrafish scans take 3–5 acquisitions , depending on asample size . A polychromatic ‘pink-beam’ covering 10–30 keV , was used to obtain 1600 projections over 192 degrees ( 1 per 0 . 12 degrees ) with a 2048-by-2048 pixel CoolSnap HQ CCD camera . Whole-organism imaging polychromatic reconstructions were generated from segmental reconstructions over its full length , taking ~20 s per segment . Stripes present in the raw pink-beam projection data were removed using a Fourier-Wavelet based method with a Haar filter and sigma of 3 pixels ( Münch et al . , 2009 ) . Rings and bands were corrected in the polar transform domain of the reconstructed images . Free parameters in this method were optimized by using minimum entropy approaches . The acquisition times using monochromatic or polychromatic ‘pink-beam’ sources are about 20 min or 20 s per scanned segment , respectively . Assuming: ( i ) a 5 min set up time per specimen , ( ii ) 10 replicates per condition , ( iii ) three scans per fish , ( iv ) a synchrotron availability of 100 scanning days per year and 10 hr of scanning time per scanning day , and ( v ) ability to multiplex five larval specimens at a time without increasing acquisition time , screening 20 , 000 larval mutants would require ~40 years with monochromatic source or ~1 year using pink-beam . There is potential for higher throughput via robotic sample swapping for monochromatic and polychromatic sources and/or increased multiplexing with pink-beam . Estimates of throughput would also be affected by the proportion of specimens that are juvenile or adult , since those are likely to be imaged individually , and require either multiple segmental scans per fish or spiral CT as used for humans . In X-ray based imaging modalities , including micro-CT , the choice of imaging energy determines both the contrast between different tissue types and the noise level in the image . The contrast is based on the difference in energy-dependent absorption coefficient between the materials of interest , and the noise is determined by the transmitted flux of X-rays . In medical imaging , the contrast is often maximized at lower energies , for which differential absorption is maximized , while relative noise is reduced at higher energies , for which transmitted flux is maximized . A contrast-to-noise ratio ( CNR ) is generally maximized at some intermediate energy that depends on the concentration and amounts of various tissues present . Metal-stained tissue tomography is associated with the potential addition of X-ray absorption edges introducing additional structure into the energy-dependence of attenuation coefficients . To investigate these tradeoffs , we made use of a simple but powerful model from Spanne ( 1989 ) that assumes the presence of a small circular contrasting detail at the center of a circular absorbing background . The background linear attenuation coefficient for energy E , μ1E=μbgE . The attenuation coefficient for the contrasting material μ2E=μbgE+μcE . Note that each attenuation coefficient is given by the product of the material density in g/cm3 and of a tabulated mass-attenuation coefficient in cm2/g , that is μcE=ρcμρc ( E ) . The contrast detail is assumed to be large enough that blurring during reconstruction does not bias the reconstructed attenuation coefficient from its true value . The reconstructed noise variances at the center of the image in the presence and absence of the contrast detail are denoted varμ1E and varμ2E , respectively . Because of circular symmetry , the variance at the center of a reconstructed image can be calculated in closed form ( Kak and Slaney , 1988 ) and shown to be inversely proportional to the transmitted intensity . var{μi ( E ) }∝1I¯i ( E ) , where I-1E=I0Eexp-μbgEtbg and I-2E=I0Eexp-μbgEdbg+μcEdc are the transmitted intensities in the absence and presence of the contrast detail , respectively . Here I0E is the incident intensity at energy E and dc and dbg are the diameters of the contrast detail and background circle , respectively . With this model , we can then readily calculate a CNR defined asCNRE=μ1E-μ2 ( E ) varμ1E+varμ2E . We used this model to calculate CNR ratios for the PTA stain as a function of background material thickness and tungsten contrast detail concentration . These are shown in Figure 1—figure supplement 1 and both sets of plots show a clear optimal energy range just above the tungsten L1 edge at 12 keV with strong fall off below the L3 edge at 10 . 2 and again above about 16 keV . We thus chose to acquire all PTA stained larval data at the pre-calibrated 13 . 8 keV monochromator setting . For larger juvenile fish of diameter >3 mm , the CNR peak is broader , justifying the use of higher , more penetrating X-ray energies . Therefore , we chose the 16 . 2 keV monochromator setting for juvenile specimens . The coherence of synchrotron radiation allows for imaging that is sensitive to phase shifts in the incident X-ray wave front caused by variations in the real part of the complex index of refraction . A variety of approaches have been explored for performing quantitative phase-contrast tomography involving the use of grating analyzers or multiple measurements with different sample-scintillator distances ( Paganin , 2006 ) . The latter techniques rely on the fact that interference fringes develop in the transmitted intensity pattern as it propagates beyond the sample . Here , we were seeking a degree of edge perception resembling that seen in histology rather than quantitative phase-contrast . Juvenile ( 33 dpf ) reconstructions at sample-to-scintillator distances ( SSDs ) of 10 , 40 , and 80 mm are reported in La Rivière et al . ( 2010 ) . Herein , we generated larval ( five dpf ) reconstructions at SSDs of 20 , 30 , 40 and 50 mm . We focused on the zebrafish eye to take advantage of the periodicity of retinal photoreceptors ( Figure 1—figure supplement 2 ) . The SSD of 20 mm caused edges of nuclei to appear blurry . SSDs of 30 or 40 mm yielded a degree of edge perception resembling that achieved in glass slide histology . Subjectively SSDs of 50 mm and larger ( data not shown ) caused edge effects to begin to look ‘artificial’ compared with traditional histology and transmission electron microscopy . Exaggerated phase effects can diminish perceived resolution . Notably , an ‘ideal’ SSD depends upon the specific structures being defined . Line profiles through the periodic retinal cells show that higher SSDs give rise to line profiles with a higher modulation depth , with the greatest benefit achieved around 30 and 40 mm SSD and strong evidence of overshooting and bias above 50 mm SSD . Based on these considerations , we used an SDD of 30 mm for all subsequent acquisitions . CT reconstruction was done using TomoPy , an open-source package from Argonne National Labs ( http://tomopy . readthedocs . io ) . Some of the image processing was conducted in Fiji/ImageJ2 ( https://fiji . sc/ ) . Software used for volumetric visualizations in this manuscript include Avizo version 9 . 4 ( Thermo Fisher Scientific , Waltham , MA ) and VGStudio Max 2 . 1 ( Volume Graphics , Heidelberg , Germany ) . Videos 1 and 4 were generated in Avizo . Videos 2 , 3 , 5 and 6 were generated in VGStudio Max 2 . 1 using an intensity histogram adjusted to better discern otherwise faint or overlapping structures . The file sizes associated with synchrotron micro-CT are on the order of ~100 GB per larval or juvenile zebrafish , making scans difficult to view for people with standard computational resources . To allow users to inspect the data without having to download the full resolution volumes , we developed ViewTool , an open-access and web-based multiplanar viewer ( http://3D . fish ) ( Figure 5—figure supplement 1 ) . ViewTool is partly based on the open-access project OpenSeaDragon ( https://openseadragon . github . io/ ) , and combines radiology and digital pathology workflows into a seamless experience to provide user-friendly access to our 3D data . Orthogonal image z-stacks are downsampled using JPG compression before being served to end users through Amazon Web Services’ Simple Storage Service ( AWS 3 ) . For bandwidth considerations , only every fourth slice is currently shown in the viewer by default . These images are presented in three windows in a user’s browser that are spatially linked to the user’s mouse or multi-touch interface . The two planes orthogonal to the interrogated plane sync to the corresponding x , y mouse location . ViewTool’s code was written in client-side JavaScript , HTML and CSS , requires no download , and has no server requirement to run the basic implementation . Semi-automated segmentation was performed using Elastix version 4 . 8 , an open-source registration software ( Klein 2010 , http://elastix . isi . uu . nl/ ) . All registrations were performed in two parts: an affine registration for optimizing initialization positions of both images followed by a landmark based thin-plate spline registration . Landmarks were manually marked on corresponding anatomical regions . The detected brain regions were derived from segmented volumes of the larval ( 4 dpf ) reference fish from the ViBE-Z zebrafish brain atlas ( Ronneberger et al . , 2012 ) , which was used for the primary registration onto one of our 5 dpf samples . Validation of region detection was performed by cross-referencing to a histology-based developmental brain atlas ( Wullimann and Mueller , 2005 ) and manual segmentation was used to correct areas of inaccurate registration . This process generated a foundation for the detection of brain regions in our remaining larval ( 5 dpf ) specimens ( n = 5 total samples ) . All other registrations used this segmented 5 dpf sample as the moving image for better detection accuracy with less extensive manual segmentation . Manual segmentations were performed using the open-source software , ITK-SNAP version 3 . 4 ( Yushkevich et al . , 2006 ) ( http://www . itksnap . org ) . The ViBE-Z database and atlas are publicly available ( http://vibez . informatik . uni-freiburg . de/ ) ( Ronneberger et al . , 2012 ) . Using a supervised learning approach , we distinguished brain cell nuclei from background by training examples using Ilastik version 1 . 3 ( Sommer et al . , 2011 ) , a simple , user-friendly , open-source tool for interactive image classification , segmentation and analysis ( http://ilastik . org/ ) . Specifically , three 75 µm3 regions were selected for training across the larval ( 5 dpf ) zebrafish brain and notochord regions . Nuclei were manually segmented in 2D in each of three orthogonal views ( sagittal , coronal , and axial ) . Features that were used in the classification including intensity ( pixel value with various smoothing ) , edges ( gradient , Laplacian of Gaussian , and difference of Gaussians ) and texture ( structured tensor of eigenvalues , Hessian of Gaussian eigenvalues ) . The random forest classifier assigned a probability of being a nucleus to any given pixel . We combined probabilities assigned from each orthogonal plane to obtain the probabilities in a given volume , Ptotal ( x , y , z ) =Pcoronal ( x , y , z ) ×Psaggitalx , y , z×Ptransverse ( x , y , z ) . Original data were thresholded to show approximate locations of nuclei and their probabilities . The results of nuclear labeling based on the nuclear training set demonstrate that , as expected , brain cell nuclei have high probabilities , while blood vessels and background have low probabilities . From these measures , a probability threshold and size filter ( 80% or higher probability containing at least 8 voxels , ~3 . 5 µm3 ) were used to detect individual cell nuclei . Cell nuclei in the same training regions were manually segmented in order to assess the accuracy of the automated segmentation results and to justify the probability threshold and size filter settings . Automatically detected cell nuclei were checked against those manually detected and categorized as true-positive ( manual and automatic detected ) , false negative ( manually detected but automatically undetected ) , or false positive ( manually undetected but automatically detected ) . F1 score was optimized over different settings to balance precision and recall between manual and automated segmentation . ViewTool is publically available ( http://3D . fish ) . Digital histology is publicly available from our Zebrafish Lifespan Atlas ( http://bio-atlas . psu . edu ) ( Cheng , 2004 ) . Registered and unregistered 8-bit reconstructions of the heads of five zebrafish larvae involved in analysis are available on the Dryad Digital Repository ( https://doi . org/10 . 5061/dryad . 4nb12g2 ) , along with scripts written for cell nuclei detection , analysis , and sample registration . Other code used for analysis can be also be downloaded from the Dryad repository . Full resolution scans , including raw projection data , are available from researchers upon request as a download or by transfer to physical media . | Diagnosing diseases , such as cancer , requires scientists and doctors to understand how cells respond to different medical conditions . A common way of studying these microscopic cell changes is by an approach called histology: thin slices of centimeter-sized samples of tissues are taken from patients , stained to distinguish cellular components , and examined for abnormal features . This powerful technique has revolutionized biology and medicine . But despite its frequent use , histology comes with limitations . To allow individual cells to be distinguished , tissues are cut into slices less than 1/20th of a millimeter thick . Histology’s dependence upon such thin slices makes it impossible to see the entirety of cells and structures that are thicker than the slice , or to accurately measure three-dimensional features such as shape or volume . Larger internal structures within the human body are routinely visualized using a technique known as computerized tomography , CT for short – whereby dozens of x-ray images are compiled together to generate a three-dimensional image . This technique has also been applied to image smaller structures . However , the resolution ( the ability to distinguish between objects ) and tissue contrast of these images has been insufficient for histology-based diagnosis across all cell types . Now , Ding et al . have developed a new method , by optimizing multiple components of CT scanning , that begins to provide the higher resolution and contrast needed to make diagnoses that require histological detail . To test their modified CT system , Ding et al . created three-dimensional images of whole zebrafish , measuring three millimeters to about a centimeter in length . Adjusting imaging parameters and views of these images made it possible to study features of larger-scale structures , such as the gills and the gut , that are normally inaccessible to histology . As a result of this unprecedented combination of high resolution and scale , computer analysis of these images allowed Ding et al . to measure cellular features such as size and shape , and to determine which cells belong to different brain regions , all from single reconstructions . Surprisingly , visualization of how tightly the brain cells are packed revealed striking differences between the brains of sibling zebrafish that were born the same day . This new method could be used to study changes across hundreds of cell types in any millimeter to centimetre-sized organism or tissue sample . In the future , the accurate measurements of microscopic features made possible by this new tool may help us to make drugs safer , improve tissue diagnostics , and care for our environment . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology",
"tools",
"and",
"resources"
] | 2019 | Computational 3D histological phenotyping of whole zebrafish by X-ray histotomography |
Non-centrosomal microtubule organizing centers ( MTOCs ) are important for microtubule organization in many cell types . In fission yeast Schizosaccharomyces pombe , the protein Mto1 , together with partner protein Mto2 ( Mto1/2 complex ) , recruits the γ-tubulin complex to multiple non-centrosomal MTOCs , including the nuclear envelope ( NE ) . Here , we develop a comparative-interactome mass spectrometry approach to determine how Mto1 localizes to the NE . Surprisingly , we find that Mto1 , a constitutively cytoplasmic protein , docks at nuclear pore complexes ( NPCs ) , via interaction with exportin Crm1 and cytoplasmic FG-nucleoporin Nup146 . Although Mto1 is not a nuclear export cargo , it binds Crm1 via a nuclear export signal-like sequence , and docking requires both Ran in the GTP-bound state and Nup146 FG repeats . In addition to determining the mechanism of MTOC formation at the NE , our results reveal a novel role for Crm1 and the nuclear export machinery in the stable docking of a cytoplasmic protein complex at NPCs .
Non-centrosomal microtubule organizing centers ( MTOCs ) are critical to the morphology and function of many types of cells ( Petry and Vale , 2015; Sanchez and Feldman , 2017; Wu and Akhmanova , 2017 ) , especially cells in which interphase microtubules ( MTs ) are arranged in linear rather than radial arrays ( Bartolini and Gundersen , 2006 ) . Examples include differentiated animal cells such as neurons ( Kapitein and Hoogenraad , 2015 ) , muscle ( Mogessie et al . , 2015; Tassin et al . , 1985 ) , and epithelial cells ( Wu and Akhmanova , 2017 ) , and many higher plant cells ( Masoud et al . , 2013; Oda , 2015 ) , as well as some single-celled eukaryotes , such as fission yeast Schizosaccharomyces pombe ( Chang and Martin , 2009; Sawin and Tran , 2006 ) . The mechanisms underlying non-centrosomal MTOC formation are just beginning to be understood . Some non-centrosomal MTs are thought to be generated by nucleation-and-release from the centrosome , followed by minus-end stabilization and anchoring elsewhere in the cell ( Bartolini and Gundersen , 2006; Sanchez and Feldman , 2017; Wu and Akhmanova , 2017 ) . However , in many cases , MTs are nucleated directly from non-centrosomal sites by the γ-tubulin complex , the primary microtubule-nucleation complex in eukaryotic cells ( Kollman et al . , 2011; Petry and Vale , 2015 ) . Understanding how the γ-tubulin complex is recruited to these sites is thus key to deciphering the fundamental mechanisms of non-centrosomal MT organization ( Lin et al . , 2015 ) . Sites of non-centrosomal γ-tubulin complex recruitment include pre-existing microtubules themselves , as well as membrane-bound compartments such as the Golgi apparatus and the nuclear envelope ( NE ) . Recruitment of the γ-tubulin complex to pre-existing microtubules depends on the multi-subunit augmin complex , in both animals and plants ( Goshima et al . , 2008; Liu et al . , 2014; Sánchez-Huertas et al . , 2016 ) . Microtubule nucleation and organization by the Golgi apparatus is orchestrated largely by AKAP450 , which recruits not only the γ-tubulin complex but also its activators , as well as MT minus-end stabilizers ( Rivero et al . , 2009; Wu et al . , 2016 ) . Combined recruitment of γ-tubulin complex and MT minus-end stabilizers/anchoring proteins is also important for MTOC organization at the cell cortex in diverse types of epithelial cells ( summarized in [Sanchez and Feldman , 2017; Wu and Akhmanova , 2017] ) . MTOC formation at the NE remains poorly understood . The NE is an important MT nucleation site both in muscle cells ( Tassin et al . , 1985 ) and in higher plants ( Ambrose and Wasteneys , 2014; Masoud et al . , 2013; Stoppin et al . , 1994 ) , as well as in fission yeast ( Lynch et al . , 2014; Sawin and Tran , 2006 ) . In muscle , γ-tubulin complex components and associated proteins are redistributed from the centrosome to the NE during development/differentiation , coincident with a decrease in centrosomal MT nucleation and large-scale changes in intracellular MT organization ( Bugnard et al . , 2005; Fant et al . , 2009; Srsen et al . , 2009; Zebrowski et al . , 2015 ) . In plant cells , which lack centrosomes altogether , many of the same proteins are similarly observed on the NE , especially before and/or after cell division ( Erhardt et al . , 2002; Janski et al . , 2012; Nakamura et al . , 2012; Seltzer et al . , 2007 ) . However , the mechanisms that regulate their recruitment are largely a mystery . Fission yeast nucleate MTs from multiple non-centrosomal sites through the cell cycle and thus provide an excellent system to study non-centrosomal MTOCs , including those on the NE ( Sawin and Tran , 2006 ) . During interphase , linear arrays of MTs are nucleated from the spindle pole body ( SPB; the yeast centrosome equivalent ) , from MTOCs on the NE and on pre-existing microtubules , and from ‘free’ MTOCs in the cytoplasm . As cells enter mitosis , non-centrosomal MT nucleation is switched off ( Borek et al . , 2015 ) and the duplicated SPBs become the only active MTOCs , nucleating both intranuclear spindle MTs and cytoplasmic astral MTs . Toward the end of cell division , microtubules are nucleated from the cytokinetic actomyosin ring ( CAR ) . By contrast , in budding yeast Saccharomyces cerevisiae , the SPBs are the only MTOCs throughout the cell cycle . In fission yeast , all types of MT nucleation in the cytoplasm ( i . e . both centrosomal and non-centrosomal nucleation ) depend on the Mto1/2 complex ( Janson et al . , 2005; Samejima et al . , 2005; Sawin et al . , 2004; Venkatram et al . , 2005; Venkatram et al . , 2004 ) . Mto1/2 contains multiple copies of the proteins Mto1 and Mto2 and directly recruits the γ-tubulin complex to prospective MTOC sites . Mto1/2 interacts with the γ-tubulin complex via Mto1’s Centrosomin Motif 1 ( CM1 ) domain , which is conserved in higher eukaryotic MTOC regulators such as Drosophila centrosomin , and human CDK5RAP2 and myomegalin ( Samejima et al . , 2008; Sawin et al . , 2004; Zhang and Megraw , 2007 ) . Interaction of CM1-domain proteins with the γ-tubulin complex can also serve to activate the γ-tubulin complex ( Choi et al . , 2010; Lynch et al . , 2014 ) , although the detailed mechanisms remain unclear . Because Mto1/2 localizes to prospective MTOC sites independently of interacting with the γ-tubulin complex ( Samejima et al . , 2008 ) , Mto1/2 localization effectively determines where and when all cytoplasmic MTOCs are generated , and thus understanding Mto1/2 localization is critical to understanding MTOC formation more broadly . Mto1/2 localization is mediated primarily by domains within Mto1 ( Figure 1A; [Samejima et al . , 2010] ) , although Mto2 contributes indirectly by helping to multimerize the Mto1/2 complex ( Lynch et al . , 2014; Samejima et al . , 2005 ) . Mto1/2 association with pre-existing MTs depends on a broadly defined region near the Mto1 C-terminus , while localization to the CAR and the SPB is mediated by overlapping modular sequences within the conserved MASC domain at the Mto1 C-terminus ( Samejima et al . , 2010 ) . Localization to the CAR involves interaction of MASC with the unconventional myosin Myp2 , while localization to the SPB involves the Septation Initiation Network protein Cdc11 ( Samejima et al . , 2010 ) . Here , we determine the mechanism of Mto1/2 localization to the NE . Using a comparative-interactome mass spectrometry approach , we find that NE localization depends on the Mto1 N-terminus interacting with exportin Crm1 , a nuclear transport receptor , and nucleoporin Nup146 , a component of the nuclear pore complex ( NPC ) . We further find that although Mto1 is an exclusively cytoplasmic protein , it becomes stably docked at the NPC by mimicking a nuclear export cargo . In addition to revealing the mechanism of MTOC formation at the fission yeast NE , our work demonstrates a completely novel role for the nuclear export machinery , in which the exportin is repurposed to create NPC-docking sites for cytoplasmic proteins with functions unrelated to nuclear export .
Mto1 localization to the NE is enhanced in the C-terminal truncation mutant Mto1[NE] , which lacks MASC and MT-localization domains ( [Lynch et al . , 2014]; Figure 1A ) . Previously , we deleted amino acids 1–130 from Mto1[NE] and from full-length Mto1 to make Mto1[bonsai] and Mto1[∆130] , respectively ( Figure 1A ) , and we showed that these deletions lead to loss of Mto1/2 complex from the NE , accompanied by loss of MT nucleation from the NE ( Lynch et al . , 2014 ) . However , in that work the consequences of this altered MT nucleation were not investigated . In fission yeast , MT-dependent pushing forces are thought to center the interphase nucleus precisely in the cell middle ( Tran et al . , 2001 ) . Because nuclear position during early mitosis determines the future cell division plane , this ensures equal size of daughter cells after cell division ( Daga and Chang , 2005 ) . To investigate whether MT nucleation from the NE contributes to nuclear positioning , we measured interphase nuclear position in mto1-GFP , mto1[NE]-GFP , mto1[∆130-GFP] and mto1[bonsai]-GFP cells ( Figure 1A; Figure 1—figure supplement 1 ) . ( In these and all subsequent experiments , mto1 mutants replace endogenous wild-type mto1+ at the mto1 locus , and in this particular experiment , all versions of mto1 were GFP-tagged to equalize protein expression levels [Lynch et al . , 2014] ) . Interestingly , nuclear positioning was less accurate in mto1[bonsai]-GFP and mto1[∆130]-GFP cells compared to mto1[NE]-GFP and mto1-GFP cells , indicating that MT nucleation from the NE contributes to nuclear positioning . By contrast , there was no difference in nuclear positioning between wild-type and mto1[NE] cells , or between mto1[131–1115] and mto1[bonsai] cells , indicating that MT nucleation from the SPB is not particularly important for nuclear positioning . To identify proteins involved in recruiting Mto1 to the NE , we wanted to compare interactomes of Mto1[NE] vs . Mto1[bonsai] . Initially , we attempted to use SILAC mass spectrometry ( MS ) ( Bicho et al . , 2010; Ong et al . , 2002 ) to compare anti-GFP immunoprecipitates of Mto1[9A1-NE]-GFP and Mto1[9A1-bonsai]-GFP , which are otherwise identical to Mto1[NE]-GFP and Mto1[bonsai]-GFP except for the additional mutation of nine consecutive amino acids in the CM1 domain to alanine ( Samejima et al . , 2008 ) , Figure 1A ) ; the 9A1 mutation disrupts interaction with the γ-tubulin complex and thereby enhances localization of Mto1[NE] to the NE ( [Lynch et al . , 2014]; Figure 1—figure supplement 2 ) . In preliminary experiments , however , we found that the immunoprecipitation approach yielded low peptide counts for many Mto1-interactors of potential interest ( Supplementary file 2 ) . We therefore decided to develop a more robust method to capture interactors even when they may be low-abundance and/or low-affinity interactors . We tagged Mto1[9A1-NE] and Mto1[9A1-bonsai] at their N-termini with GFP and at their C-termini with an HTB ( His-TEV-biotin ) tag , which allows for two-step purification of a tagged protein under fully denaturing conditions after cross-linking to interactors ( Tagwerker et al . , 2006 ) ( Figure 1B ) . As expected , GFP-Mto1[9A1-NE]-HTB localized to the NE in vivo , while GFP-Mto1[9A1-bonsai]-HTB was present only in the cytoplasm ( Figure 1C ) . Disuccinimidyl suberate ( DSS ) cross-linking of cell cryogrindates shifted a significant proportion of HTB-tagged Mto1 into higher molecular-weight species ( Figure 1D ) . After DSS cross-linking and denaturing purification ( Figure 1E; see Materials and methods ) , we analyzed cross-linked adducts of GFP-Mto1[9A1-NE]-HTB and GFP-Mto1[9A1-bonsai]-HTB by label-free quantification ( LFQ ) MS ( [Cox and Mann , 2008; Tyanova et al . , 2016]; Figure 1F; Table 1; Supplementary file 3 ) . Among the proteins significantly enriched in the Mto1[9A1-NE] interactome vs . the Mto1[9A1-bonsai] interactome , we identified nucleoporin Nup146 ( Asakawa et al . , 2014; Chen et al . , 2004 ) , exportin Crm1 ( Adachi and Yanagida , 1989; Fung and Chook , 2014; Hutten and Kehlenbach , 2007; Stade et al . , 1997 ) , the fission yeast TACC homolog , Alp7 ( Sato et al . , 2004 ) , and , to a lesser extent , polo kinase Plo1 ( Ohkura et al . , 1995 ) . Neither Alp7 nor Plo1 is known to localize to the NE , and Plo1 was not investigated further . The interaction of Mto1[NE] with Alp7 was of potential interest because of the role of Alp7 in microtubule organization ( Ling et al . , 2009; Sato et al . , 2009; Zheng et al . , 2006 ) , and an interaction between Mto1 and Alp7 has been confirmed independently ( M . Sato , Waseda University , personal communication , July 2017 ) . However , we found that in alp7∆ deletion mutants , Mto1[9A1-NE]-GFP was present on the NE just as in wild-type ( alp7+ ) cells ( Figure 1—figure supplement 3 ) . This indicates that Alp7 is not required for Mto1 localization to the NE . The interaction of Mto1[9A1-NE] with Nup146 suggested that Mto1 may localize to nuclear pore complexes ( NPCs ) on the NE . We therefore imaged Mto1[9A1-NE]-GFP together with Nup146-3mCherry in a nup132∆ background , in which NPCs can become clustered on the NE ( BaiBaï et al . , 2004 ) . We observed extensive colocalization of Mto1[9A1-NE]-GFP with Nup146-3mCherry clusters ( Figure 2A ) , indicating specific association with NPCs . We also examined Mto1[9A1-NE]-GFP localization by immunoelectron microscopy . Close homologs of Nup146 in budding yeast ( Nup159; referred to here as Sc Nup159 ) and humans ( Nup214; referred to as Hs Nup214 ) are both located exclusively at the cytoplasmic face of NPCs ( Gorsch et al . , 1995; Kraemer et al . , 1994; Kraemer et al . , 1995 ) , and indirect evidence suggests that this is also the case for Nup146 ( Lo Presti et al . , 2012 ) . Consistent with this , we observed Mto1[9A1-NE]-GFP specifically at the cytoplasmic face of NPCs ( Figure 2B ) . The interaction of Mto1[NE] with Crm1 was both surprising and puzzling . As the major transport receptor for nuclear export of proteins ( as well as some RNAs ) , Crm1 normally forms a trimeric complex with export cargo and RanGTP within the nucleus , which facilitates transit of cargo through the permeability barrier of the NPC and into the cytoplasm ( Dong et al . , 2009; Fung and Chook , 2014; Hutten and Kehlenbach , 2007 ) . However , to date , there is no evidence that Mto1 is a nuclear export cargo or indeed is ever present in the nucleus . Because deletion of crm1+ is lethal ( Adachi and Yanagida , 1989 ) , we investigated the significance of the Mto1-Crm1 interaction by asking whether inhibition of Crm1 cargo-binding activity affects Mto1 localization to NPCs . Nuclear export cargos typically bind to Crm1 via hydrophobic nuclear export signals ( NESs ) ( Dong et al . , 2009; Fung and Chook , 2014; Fung et al . , 2015; Güttler et al . , 2010; Hutten and Kehlenbach , 2007; Kutay and Güttinger , 2005 ) . This can be inhibited by the drug leptomycin B ( LMB ) , which binds within the hydrophobic NES-binding cleft of Crm1 ( Dong et al . , 2009; Fornerod et al . , 1997a; Fukuda et al . , 1997; Fung and Chook , 2014; Ossareh-Nazari et al . , 1997 ) . As a result , when cells are treated with LMB , nuclear export cargos accumulate within the nucleus . Interestingly , after LMB treatment , we found that Mto1[9A1-NE]-GFP was lost from NPCs ( Figure 3A ) . Strikingly , however , rather than accumulating in the nucleus , Mto1[9A1-NE]-GFP became dispersed in the cytoplasm . Given the unusual behavior of Mto1[9A1-NE]-GFP after LMB treatment , we confirmed that LMB was inhibiting nuclear export . We assayed localization of Alp7 , which shuttles continuously in and out of the nucleus during interphase , in complex with its partner protein Alp14 ( ch-TOG homolog ) ( Okada and Sato , 2015; Okada et al . , 2014 ) ( Figure 3—figure supplement 1 ) . In the absence of LMB , Alp7-3GFP was present in the cytoplasm , primarily as puncta on cytoplasmic MTs . As expected , after LMB treatment , Alp7-3GFP accumulated in the nucleoplasm and on an intranuclear MT bundle that has been described to form upon LMB treatment of fission yeast ( Matsuyama et al . , 2006 ) ( Figure 3—figure supplement 1 ) . To rule out the possibility that loss of Mto1[9A1-NE]-GFP from NPCs was due to an off-target effect of LMB ( i . e . unrelated to Crm1 inhibition ) , we generated an LMB-resistant crm1 mutant . LMB is a particularly potent inhibitor of Crm1 because it reacts covalently with cysteine 529 ( C529 ) in Crm1’s NES-binding cleft ( Kudo et al . , 1999 ) . We mutated C529 in the endogenous crm1 coding sequence to alanine ( crm1-C529A ) , as well as to serine ( crm1-C529S ) , threonine ( crm1-C529T ) and valine ( crm1-C529V ) ( Figure 3B , Figure 3—figure supplement 2 ) . All four mutants were viable , indicating that they preserve essential functions of crm1 for nuclear export , and three out of the four were resistant to high concentrations of LMB ( Figure 3—figure supplement 2A ) . Interestingly , we found that in crm1-C529A cells , Mto1[9A1-NE]-GFP localized to NPCs both in the absence and in the presence of LMB ( Figure 3B ) . This demonstrates that loss of Mto1 from NPCs after LMB treatment can be specifically attributed to inhibition at the Crm1 cargo-binding cleft . In addition to these experiments , we used immunofluorescence to compare MT regrowth after cold-induced MT depolymerization in control vs . LMB-treated wild-type cells ( i . e . cells expressing full-length , untagged Mto1 ) . Previous work showed that cold-induced MT depolymerization causes the pool of Mto1 normally associated with cytoplasmic MTs to redistribute to the NE , and as a result , when cells are rewarmed , nearly all MT regrowth initiates from the NE ( Sawin et al . , 2004 ) . As expected , we found that in control cells , MT regrowth occurred from the NE . However , in LMB-treated cells , MT regrowth occurred randomly in the cytoplasm , consistent with a failure of Mto1 to localize to the NE after LMB treatment ( Figure 3—figure supplement 3 ) . How might Crm1 cargo-binding activity be involved in Mto1 localization to the NPC ? We hypothesized that Mto1 itself might bind to Crm1 as an unconventional ‘cargo’ and somehow exploit this interaction to localize to the cytoplasmic face of the NPC . To test this , we used LFQ MS to compare GFP-Mto1[9A-NE]-HTB interactomes prepared from untreated vs . LMB-treated cells ( Figure 3C; Table 2; Supplementary file 4 ) . Interestingly , only 3–4 out of nearly 500 quantified proteins were significantly enriched in the GFP-Mto1[9A1-NE]-HTB interactome from untreated cells compared to LMB-treated cells . Among these , Crm1 showed the greatest enrichment ( ~20X ) . Nup146 also showed enrichment , but to a lesser extent ( ~2 . 8X ) , which may indicate that Mto1 can bind weakly to Nup146 independently of Crm1 ( see Discussion ) . These results demonstrate that , like Mto1 localization to NPCs , Mto1 interaction with Crm1 requires Crm1 cargo-binding activity . Based on these findings , we next used the LocNES algorithm ( Xu et al . , 2015 ) to search for NES-like sequences within the N-terminal 130 amino acids of Mto1 , which are present in Mto1[NE] but absent from Mto1[bonsai] . The sequence spanning Mto1 amino acids 9–25 contained two closely overlapping candidate NESs ( Figure 4A ) . Interestingly , the spacing of hydrophobic amino acids within this NES-like sequence is similar to that of several non-natural high-affinity NESs ( Figure 4B; [Engelsma et al . , 2004; Güttler et al . , 2010] ) . To investigate the role of the Mto1 NES-like sequence , we deleted the first 25 amino acids of Mto1 from GFP-Mto1[9A1-NE]-HTB . The truncated protein , termed GFP-Mto1[∆NES-9A1-NE]-HTB , failed to localize to NPCs and instead was present in the cytoplasm ( Figure 4C ) . In parallel , we used LFQ MS to determine how the ∆NES truncation affected the GFP-Mto1[9A1-NE]-HTB interactome . As with LMB treatment , very few proteins were enriched in the GFP-Mto1[9A1-NE]-HTB interactome compared to GFP-Mto1[∆NES-9A1-NE]-HTB interactome ( Figure 4D; Table 3; Supplementary file 5 ) . However , we observed strong enrichment of both Crm1 ( ~85X ) and Nup146 ( ~20X ) . The importance of the Mto1 NES-like sequence both for localization to NPCs and for interaction with Crm1 strongly suggests that Mto1 is a direct but unconventional cargo for Crm1 . Because of the unusual role of the Mto1 NES-like sequence , we will refer to it as a ‘NES-mimic’ ( NES-M ) . We next asked whether the Mto1 NES-M is sufficient to localize a reporter protein to the NPC . We replaced endogenous Mto1 with GFP-Mto1[1-29]-GST , which contains only the first 29 amino acids of Mto1 . Strikingly , GFP-Mto1[1-29]-GST localized to puncta on the NE , which we interpret to be NPCs ( Figure 4E ) . By contrast , GFP-Mto1[1-12]-GST , which lacks the NES-M , did not show any specific localization . We further found that after LMB treatment , GFP-Mto1[1-29]-GST was lost from NPCs ( Figure 4F ) ; moreover , like Mto1[9A1-NE]-GFP , GFP-Mto1[1-29]-GST was present exclusively in the cytoplasm after LMB treatment . Compared to GFP fusions with Mto1[NE] , GFP-Mto1[1-29]-GST had a weaker punctate localization at NPCs . We hypothesized that this may be due an avidity effect , because Mto1[NE] can form higher order multimers , via its coiled-coil region and via interaction with Mto2 ( Lynch et al . , 2014 ) , whereas GFP-Mto1[1-29]-GST would be expected to form only dimers , via the GST domain . To investigate whether dimerization may contribute to NPC localization , we analyzed localization of a GFP-Mto1[1-29]−13Myc fusion protein , which should be monomeric . GFP-Mto1[1-29]−13Myc did not localize to NPCs ( Figure 4E ) , suggesting that dimerization/multimerization may be an important factor for Mto1 NPC localization . Collectively , these results indicate the Mto1 NES-M is both necessary and sufficient for localization to NPCs , without ever being present in the nucleus . To further investigate similarities between the mechanism of Mto1 localization to NPCs and nuclear export , we tested whether Mto1 localization depends on the nucleotide state of Ran . Net directional transport of conventional cargos through the NPC depends on a RanGTP gradient across the NE , generated by Ran GTPase activating protein ( RanGAP ) in the cytoplasm and Ran guanine-nucleotide exchange factor ( RanGEF ) in the nucleus ( Aitchison and Rout , 2012; Görlich and Kutay , 1999; Wente and Rout , 2010 ) . Importins bind import cargos in the cytoplasm , where RanGTP concentration is low , and release them in the nucleus , where RanGTP concentration is high . By contrast , exportins bind cooperatively to export cargos and RanGTP within the nucleus to form trimeric export complexes , which then dissociate after export , accompanied by RanGTP hydrolysis aided by RanGAP ( Fornerod et al . , 1997a; Fung and Chook , 2014; Güttler and Görlich , 2011; Koyama and Matsuura , 2012; Monecke et al . , 2014 ) . The role of Ran can be addressed by expressing mutant versions of Ran ( encoded by the spi1+ gene in fission yeast; [Matsumoto and Beach , 1991] ) that mimic either GTP or GDP states ( Bischoff et al . , 1994; Klebe et al . , 1995 ) . Constitutively active Ran ( RanQ69L in humans ) is defective in GTP hydrolysis and thus ‘locked’ in the RanGTP state , while inactive/dominant-negative Ran ( RanT24N in humans ) has low affinity for nucleotide and competes with endogenous RanGDP for binding to RanGEF . We expressed wild-type spi1+ , spi1[Q68L] ( equivalent to human RanQ69L ) , and spi1[T23N] ( equivalent to human RanT24N ) as integrated transgenes from a thiamine-repressible promoter . All cells were viable under repressing conditions , but growth was impaired by expression of spi1[Q68L] or spi1[T23N] ( Figure 5—figure supplement 1 ) , consistent with phenotypes of the equivalent mutants in vertebrate cells ( Clarke et al . , 1995; Dasso et al . , 1994; Kornbluth et al . , 1994; Ren et al . , 1994 ) . To avoid any indirect effects on Mto1[9A1-NE]-GFP localization as a result of growth impairment , we assayed localization as early as possible during expression ( Figure 5 , Figure 5—figure supplement 2 ) . Expression of spi1+ had no effect on Mto1[9A1-NE]-GFP localization . Expression of spi1[Q68L] impaired import of a nuclear localization signal ( NLS ) reporter protein , as expected ( Figure 5—figure supplement 2 ) , but did not alter Mto1[9A1-NE]-GFP localization to NPCs . Interestingly , expression of spi1[T23N] , which had only minor effects on NLS reporter localization , led to strong loss of Mto1[9A1-NE]-GFP from NPCs ( Figure 5 , Figure 5—figure supplement 2 ) . These results indicate that , like nuclear export , Mto1 localization to NPCs requires RanGTP . Moreover , at least in the short-term , neither RanGDP nor Ran nucleotide cycling is required for Mto1 NPC localization . We next asked whether Nup146 contributes to Mto1 NPC localization . Like approximately one-third of all nucleoporins , Nup146 and its homologs Sc Nup159 and Hs Nup214 contain multiple phenylalanine-glycine ( FG ) repeats , which bind directly to importins and/or exportins ( Figure 6—figure supplement 1A; [Aitchison and Rout , 2012; Wente and Rout , 2010] ) . Because of their location on the cytoplasmic face of the NPC , these nucleoporins are classified as ‘cytoplasmic FG-Nups’ , distinguishing them from the ‘symmetric FG-Nups’ present within the central permeability barrier of the NPC . While FG repeats of symmetric FG-Nups directly facilitate cargo transport through the NPC , FG repeats of cytoplasmic FG-Nups are thought not to be important for transport per se ( Adams et al . , 2014; Strawn et al . , 2004; Zeitler and Weis , 2004 ) , although their other ( non-FG ) regions recruit proteins for processes linked to transport ( e . g . mRNP processing after export ( Napetschnig et al . , 2007; Schmitt et al . , 1999; Weirich et al . , 2004 ) ; Figure 6—figure supplement 1A ) . Nevertheless , the FG repeats of Sc Nup159 and Hs Nup214 have been shown to bind to Crm1 with high specificity relative to other importins/exportins . ( Allen et al . , 2002; Fornerod et al . , 1997b; Hutten and Kehlenbach , 2006; Port et al . , 2015; Roloff et al . , 2013; Zeitler and Weis , 2004 ) . We therefore focused attention on the Nup146 FG repeats . We deleted the 50-amino-acid region comprising FG repeats 5–12 ( out of a total of 16 FG repeats ) from the endogenous nup146 coding sequence ( Figure 6A ) . Although complete deletion of nup146 is lethal ( Chen et al . , 2004 ) , the nup146[∆FG5-12] strain was viable , and Nup146[∆FG5-12]−3mCherry was localized to NPCs . ( Figure 6—figure supplement 1B ) . Strikingly , in nup146[∆FG5-12] cells , Mto1[9A1-NE]-GFP no longer localized to NPCs and instead was present only in the cytoplasm ( Figure 6B; Figure 6—figure supplement 1B ) . We also analyzed MTOC activity at the NE in wild-type ( nup146+ ) cells vs . nup146 [∆FG5-12] cells . First , we assayed MT regrowth after cold-induced depolymerization , in cells expressing full-length , untagged Mto1 ( Figure 6C ) . In wild-type cells , MT regrowth occurred from the NE , while in nup146[∆FG5-12] cells MT regrowth occurred randomly in the cytoplasm ( Figure 6C ) , similar to LMB-treated wild-type cells ( Figure 3—figure supplement 3 ) . Second , we used live-cell imaging of GFP-tubulin to assay steady-state MT nucleation , in cells expressing Mto1[NE]-GFP ( Figure 6D , E; in these cells , Mto1[NE]-GFP is too faint to be seen relative to GFP-tubulin ) . In nup146[∆FG 5–12] cells , MT nucleation frequency in the vicinity of the NE was decreased by ~90% relative to wild-type ( nup146+ ) cells , while nucleation frequency away from the NE was unchanged . Collectively , these results indicate that Nup146 FG repeats 5–12 are essential for Mto1 docking at the NPC and , consequently , for MTOC nucleation from the NE . To our knowledge , this is the first biological function that can be uniquely attributed to the FG repeats of the Nup146/Sc Nup159/Hs Nup214 class of cytoplasmic FG-Nups , in any organism ( see Discussion ) . Our results thus far indicate that a RanGTP-dependent Mto1-Crm1 ‘cargo-like’ complex docks at the cytoplasmic face of the NPC via a mechanism involving Nup146 FG repeats ( see Figure 7 ) . Interestingly , a subset of FG repeats in Hs Nup214 have been shown to bind to Crm1 in a manner that stabilizes the Crm1-RanGTP-cargo interaction in vitro ( Askjaer et al . , 1999; Fornerod et al . , 1997b; Hutten and Kehlenbach , 2006; Kehlenbach et al . , 1999; Port et al . , 2015; Roloff et al . , 2013 ) . We therefore asked whether Nup146 FG repeats 5–12 are important for Mto1 interaction with Nup146 , and whether these repeats contribute to Crm1 association with Mto1 in vivo . We used LFQ MS to compare GFP-Mto1[9A1-NE]-HTB interactomes from wild-type ( nup146+ ) vs . nup146[∆FG5-12] cells . Among more than 500 quantified proteins , only five to six proteins were significantly enriched in the GFP-Mto1[9A1-NE]-HTB interactome from nup146+ cells compared to nup146[∆FG5-12] cells . Nup146 itself showed the greatest enrichment ( ~11X ) , while Crm1 was also enriched , although to a lesser extent ( ~3X ) ( Figure 6F; Table 4; Supplementary file 6 ) . This suggests that Nup146 FG repeats are essential for interaction of Mto1 with Nup146 . In addition , while Nup146 FG repeats may not be absolutely essential for formation of an Mto1-Crm1 complex , they may help to stabilize it .
How does Mto1 , a nuclear export cargo ‘mimic’ , become docked at the NPC , while conventional export cargos are released into the cytoplasm ? Ultimately , a detailed understanding of this issue will require in vitro biochemistry with purified proteins . However , based on previous work in mammalian cells ( Engelsma et al . , 2004; Port et al . , 2015 ) , we speculate that docking may depend on: 1 ) the Mto1 NES-M acting as a high-affinity NES; and 2 ) the stability of interaction between Mto1-Crm1 and Nup146 FG repeats . The Mto1 NES-M is necessary and sufficient for docking at the NPC ( Figure 4 ) . Interestingly , in human cells , cargo containing a non-natural , high-affinity NES ( a ‘supraphysiological NES’ ) was shown to accumulate at the cytoplasmic face of the NPC and also to enhance Crm1 accumulation at the same site ( Engelsma et al . , 2004; Engelsma et al . , 2008 ) . We hypothesize that the Mto1 NES-M may be a natural high-affinity NES . In recent years , the NES ‘consensus’ has evolved in concert with new experimental findings ( Dong et al . , 2009; Fung et al . , 2015; Fung et al . , 2017; Güttler et al . , 2010; Monecke et al . , 2009 ) . In particular , relative to an original consensus involving four spaced hydrophobic residues ( Kutay and Güttinger , 2005 ) , several high-affinity NESs depend on a fifth hydrophobic residue , which may also be present in the Mto1 NES-M ( Figure 4B ) . In this context , it is interesting that we found that Mto1[9A1-NE]-GFP localizes to NPCs in crm1-C529A mutants ( Figure 3B ) but not in crm1-C529S , crm1-C529T , and crm1-C529V mutants , even though these mutants are viable and thus competent for nuclear export ( Figure 3—figure supplement 2 ) . This may indicate that , relative to conventional NESs , the binding of the Mto1 NES-M to Crm1 involves recognition of additional and/or distinct features within the Crm1 NES-binding cleft . Assuming that the Mto1 NES-M interacts with Crm1 as a high-affinity NES , clues as to how this could lead to accumulation at the NPC can be found in structural studies of Crm1 alone and Crm1 in complex with RanGTP , cargo , and an FG-repeat fragment of Hs Nup214 ( Figure 7—figure supplement 1A; [Dong et al . , 2009; Güttler et al . , 2010; Monecke et al . , 2009; Monecke et al . , 2013; Port et al . , 2015; Saito and Matsuura , 2013] ) . Crm1 can exist in two conformations: an unliganded extended , superhelical conformation , which is inhibitory to cargo and RanGTP binding , and a compact , ring-like conformation , which is stabilized by cooperative binding to cargo and RanGTP . Importantly , the FG-repeat fragment of Hs Nup214 , which binds Crm1 cooperatively with RanGTP and cargo , interacts with the compact conformation of Crm1 at multiple sites , spanning the junction between the Crm1 N- and C-termini in a manner similar to an adhesive bandage ( Port et al . , 2015 ) ( Figure 7—figure supplement 1A ) . The Hs Nup214 FG repeats have therefore been described as a ‘molecular clamp’ that can stabilize Crm1-RanGTP-cargo complex in the compact conformation ( Port et al . , 2015 ) . However , from an energetic perspective , cooperative binding also implies that anything that stabilizes the Crm1 compact conformation ( including a high-affinity NES ) will correspondingly reinforce association of Crm1 with Hs Nup214 FG repeats . As a result , a sufficiently high-affinity NES cargo would be expected to stabilize interaction of Crm1 with Nup146 , leading to docking of Crm1 ( and the NES cargo itself ) at the cytoplasmic face of the NPC ( Figure 7—figure supplement 1A ) . In addition to a ‘high-affinity NES’ mechanism , other factors may also contribute to docking of the Mto1/2 complex at the NPC . For example , if Mto1 ( or its partner Mto2 ) were to bind Nup146 independently of binding to Crm1 , such multivalent binding would decrease the off-rate from the NPC; currently our MS data cannot distinguish between direct and indirect Mto1 interactors . Interactions between Mto1/2 and the NPC could also be stabilized by avidity effects ( Figure 4E ) . Mto1/2 is multimeric in vivo , containing multiple ( >10 ) copies of both Mto1 and Mto2 ( Lynch et al . , 2014 ) , while nucleoporins are also present in multiple copies within the NPC , because of its eight-fold symmetry ( Aitchison and Rout , 2012; Görlich and Kutay , 1999; Wente and Rout , 2010 ) . As a result , multiple Mto1 molecules in a single Mto1/2 complex could bind to multiple nucleoporins ( and/or Crm1 ) in a single NPC . Interestingly , localization of Mto1/2 to the SPB and the CAR also depends on avidity effects ( Samejima et al . , 2010 ) . Given that conventional Crm1-dependent export complexes form in the nucleus , where RanGTP concentration is high ( Aitchison and Rout , 2012; Görlich and Kutay , 1999; Wente and Rout , 2010 ) , how might an Mto1/2 docking complex form in the cytoplasm , where RanGTP concentration is low ? We speculate that if the Mto1 NES-M acts as a high-affinity NES , it may be possible for Mto1/2 to replace a conventional nuclear export cargo at the final stages of export , via a ‘cargo-handover’ mechanism ( Figure 7—figure supplement 1B ) . Alternatively , a docking complex involving Mto1/2 , Crm1 , Nup146 and RanGTP could in principle form de novo at the cytoplasmic face of the NPC . While the low concentration of RanGTP in the cytoplasm makes this unlikely , it is formally possible that in the immediate vicinity of the NPC , the local concentration of RanGTP is higher than in the cytoplasm in general , because in yeast , RanGAP is freely soluble in the cytoplasm rather than associated with the NPC ( Aitchison and Rout , 2012; Hopper et al . , 1990 ) . Accordingly , immediately after RanGTP dissociates from export complexes ( but prior to GTP hydrolysis ) , it might be available to cytoplasmic Mto1/2 . Our proposition that an Mto1/2 docking complex includes RanGTP ( see Figure 7; Figure 7—figure supplement 1 ) is based on the observed requirement for RanGTP for docking in vivo and on analogy to the known mechanisms underlying stable NES-dependent binding of conventional nuclear export cargo to Crm1 ( Dong et al . , 2009; Güttler et al . , 2010; Monecke et al . , 2009; Monecke et al . , 2013; Port et al . , 2015; Saito and Matsuura , 2013 ) . We note that in cross-linking MS experiments , we did not observe increased association of Ran with Mto1 that is localized to the NE compared to Mto1 that is not localized to the NE ( Supplementary file 3–6 . One possible reason for this is that Ran may not be readily cross-linked to Mto1 or to Mto1’s immediate interactors ( by analogy to conventional cargo export , Ran would not be expected to bind directly to Mto1 ) . Alternatively , it is possible that the Mto1/2 docking complex does not contain Ran and that the requirement for Ran-GTP for docking is only indirect . This might be the case if the main role of Ran-GTP in Mto1/2 docking is to generate sufficiently high levels of conventional export complexes at the cytoplasmic face NPC such that components of these complexes ( e . g . Crm1 ) can subsequently be used for Mto1/2 docking by an unconventional , Ran-independent mechanism . Further work with purified proteins may help to address this issue . In this work , we identified a very specific phenotype associated with deletion of Nup146 FG repeats 5–12: Mto1 is lost from NPCs , with a concomitant loss in MT nucleation from the NE . Moreover , this is correlated with a strong decrease in interaction of Mto1 with Nup146 and , to a lesser extent , with Crm1 , consistent with our model of a cargo-like complex of Mto1/2 , Crm1 and RanGTP docking at the cytoplasmic face of the NPC . In this context , it is interesting that extensive analysis in budding yeast has shown that the FG regions of cytoplasmic FG-Nups ( as well as nucleoplasmic FG-Nups ) can be deleted without almost any discernible effects on nuclear transport ( Adams et al . , 2014; Strawn et al . , 2004; Zeitler and Weis , 2004 ) . In human cells , the role of Hs Nup214 in protein export appears to be somewhat controversial ( Bernad et al . , 2006; Hutten and Kehlenbach , 2006 ) ; however , similar to budding yeast , in at least one instance where Hs Nup214 was found to be important for export—namely , export of the 60S pre-ribosome— the FG repeats of Hs Nup214 were found not to be required [Bernad et al . , 2006] ) . Based on these results , and on the conservation of FG repeats in Nup146 , Sc Nup159 and Hs Nup214 , we propose that an important but previously unrecognized role for cytoplasmic FG-Nups may be to dock cytoplasmic proteins at the NPC for non-export-related functions , as described here for generation of non-centrosomal MTOCs by the Mto1/2 complex . It will be interesting to see how widespread this type of repurposing of the nuclear export machinery is in eukaryotic cells more generally .
Fission yeast methods and growth media were as described ( Forsburg and Rhind , 2006; Petersen and Russell , 2016 ) . Strains were normally grown in YE5S-rich medium or PMG minimal medium ( like EMM2 , but using 5 g/L sodium glutamate acid instead of ammonium chloride as nitrogen source ) . For preliminary experiments using SILAC mass spectrometry , cells were grown in low-nitrogen EMM2 medium ( ‘LowN’; using 0 . 3 g/L ammonium chloride as nitrogen source; [Bicho et al . , 2010] ) . For purification of HTB-tagged Mto1 variants for LFQ MS , Mto1 variants were expressed from the nmt81 promoter , and cells were grown in PMG medium , except for experiments involving leptomycin B ( Figure 3 ) , in which case Mto1 variants were expressed from the repressed nmt1 promoter , and cells were grown in 4xYE5S medium . For electron microscopy , cells were grown in EMM2 minimal medium . Nutritional supplements were normally used at 175 mg/L , except for arginine and lysine in SILAC experiments , in which unlabeled arginine or L-13C6-arginine ( Sigma Isotec , Gillingham , UK ) was used at 80 mg/L , and unlabeled lysine or L-13C615N2-lysine ( Sigma Isotec ) was used at 60 mg/L ( Bicho et al . , 2010 ) . Solid media contained 2% Bacto agar ( Becton Dickinson , Wokingham , UK ) . For mating , SPA plates containing 45 mg/L each of adenine , leucine , uracil , histidine and lysine were used . For repression of thiamine-regulated promoters , sterile-filtered thiamine was added to media at a final concentration of 15 µM . Strains used in this study are listed in Supplementary file 1 . For experiments purifying HTB-tagged Mto1 for LFQ MS , strains contained the mto2[17A] allele; this allele contains 17 phosphorylation sites in Mto2 mutated to alanine , which helps to stabilize the Mto1/2 complex ( Borek et al . , 2015 ) . The mto2[17A] allele was also present in strains imaged in Figures 1C and 4C ( see Supplementary file 1 ) . Genetic crosses used either tetrad dissection or random spore analysis ( Ekwall and Thon , 2017 ) . Except for the cases described below , genome manipulations such as gene tagging , truncation and/or deletion were made by homologous recombination of PCR products ( Bähler et al . , 1998; Hentges et al . , 2005; Van Driessche et al . , 2005 ) . PCR was performed with either Phusion High-Fidelity polymerase or Q5 High-Fidelity polymerase ( NEB , Hitchin , UK ) . Desired strains were confirmed by yeast colony PCR , western blot and/or fluorescence microscopy as appropriate . For all cloning experiments , E . coli strain DH5alpha was used . To generate crm1-C529A/S/T/V mutants , a one-step approach was used , in which mto1[9A1-NE]-GFP nup146-3mCherry cells were transformed with mutated crm1 DNA fragments and selected directly for leptomycin ( LMB ) resistance . Mutant crm1 fragments were designed with the mutation site in the center , ~650 base pairs of crm1 sequence on either side of the mutation site , and BstXI sites at each end of fragment . Plasmids containing the mutant fragments were synthesized by GeneArt ( plasmids pKS1735 , pKS1734 , pKS1737 , and pKS1738 , for crm1-C529A , crm1-C529S , crm1-C529T , and crm1-C529V mutants , respectively ) . The crm1 fragments were released from plasmids by BstXI digestion , purified and transformed into strain KS7255 . Cells from the transformation were plated onto YE5S plates containing 300 nM LMB ( LC Laboratories , Woburn , MA ) , and LMB-resistant colonies were easily identified . A negative-control transformation conducted in parallel did not yield any LMB-resistant colonies . Stable LMB-resistant colonies from each transformation were then used for sequencing to confirm specific mutations in crm1 genomic DNA . The mutant strains were named KS9340 ( crm1-C529A ) , KS9221 ( crm1-C529S ) KS9338 ( crm1-C529T ) , and KS9336 ( crm1-C529V ) . Strains overexpressing spi1+ , spi1[Q68L] and spi1[T23N] from the nmt41 promoter were generated by targeted integration of transgenes at the hph . 171k locus ( Fennessy et al . , 2014 ) . First , pJET-spi1+/[Q68L]/[T23N] plasmids were constructed . To construct pJET-spi1+ , spi1+ genomic DNA was amplified from fission yeast genomic DNA using primer pair OKS3290/OKS3291 , and the PCR product was ligated into vector pJET1 . 2 ( Thermo Fisher Scientific , Paisley , UK ) . The resulting pJET-spi1+ plasmid was confirmed by sequencing and named pKS1603 . To construct pJET-spi1[Q68L] , the Q68L mutation was introduced into pKS1603 by PCR , using primer pair OKS3139/OKS3140 . The PCR product was recircularized using T4 polynucleotide kinase and T4 DNA ligase . The resulting plasmid was confirmed by sequencing and named pKS1596 . To construct pJET-spi1[T23N] , pKS1603 was used as template to introduce the T23N mutation into the spi1 sequence , using QuikChange II Site-Directed Mutagenesis kit ( Agilent , Stockport , UK ) and primer pair OKS 3336/OKS3337 . After DpnI treatment and transformation , the resulting plasmid was confirmed by sequencing and named pKS1595 . Next , the spi1 inserts from pKS1603 , pKS1596 , and pKS1595 were each subcloned into the fission yeast integration vector pINTH41 ( Fennessy et al . , 2014 ) after restriction digest with BamHI and NdeI . The resulting pINTH41-spi1+/[Q68L]/[T23N] plasmids were confirmed by restriction digest and named pKS1597 , pKS1599 , and pKS1598 , respectively . For transformation into fission yeast , pKS1597 , pKS1599 , and pKS1598 were digested with NotI , and the relevant fragments were purified and used to transform strain KS 7742 . Stable nourseothricin-resistant , hygromycin-sensitive integrants were identified , indicating replacement of the hygromycin-resistance marker at the hph . 171k locus by the transgene . Colonies were then tested on PMG plates ( also containing adenine and uracil ) with or without 15 µM thiamine . After two days of growth at 32°C , nmt41:spi1+ colonies were similar with and without thiamine , while nmt41:spi1[Q68L] and nmt41:spi1[T23N] colonies appeared normal on plates with thiamine but formed only very tiny colonies on plates without thiamine . nmt41: spi1+/[Q68L]/[T23N] overexpression strains were named KS8578 , KS8581 and KS8580 , respectively . Strains with internal deletions of nup146 FG repeats were constructed by a two-step approach ( Fennessy et al . , 2014 ) . For the first step , an rpl42:natMX6 cassette was amplified by PCR using primer pair OKS2460/OKS2461 and the PCR product was used to transform cycloheximide-resistant rpl42 . sP56Q strain KS8072 . The amplified cassette was at the end of the nup146 coding sequence to generate a nourseothricin-resistant , cycloheximide-sensitive nup146:rpl42:natMX6 rpl42 . sP56Q strain , which was named KS8254 . For the second step , a 5 . 1 kb wild-type nup146 genomic DNA fragment ( containing 5’ and 3’ untranslated regions as well as coding sequence ) was amplified by PCR using primer pair OKS3063/OKS3067 . The PCR product was ligated into pJET1 . 2 vector , and the resulting pJET-nup146 plasmid was sequenced and named pKS1511 . Internal deletions of FG repeats were made within pKS1511 by PCR , using primer pair OKS3093/OKS3094 to make nup146 [∆FG5-12∆] . The PCR product was recircularized using T4 polynucleotide kinase and T4 DNA ligase , and after transformation , the resulting plasmid was confirmed by sequencing . The pJET-nup146[∆FG5-12] genomic DNA plasmid was named pKS1514 . DNA sequence of nup146 [∆FG5-12] was amplified from pKS1514 by PCR using primer pair OKS3098/OKS3099 . The resulting PCR product was transformed into strain KS8254 . Nourseothricin-sensitive , cycloheximide-resistant colonies were selected , and colony PCR using primer pair OKS3154/OKS3155 was used to identify the desired strains . The correct nup146[∆FG5-12] rpl42 . sP56Q strains was named KS8305 . Immunoelectron microscopy was carried out as described previously ( Tange et al . , 2016 ) , with some modifications . Briefly , strain KS5750 ( mto1[9A1-NE]-GFP ) was cultured in EMM2 medium with supplements . After washing with 0 . 1 M phosphate buffer ( PB , pH7 . 4 ) , cells were fixed for 20 min at room temperature with 4% formaldehyde and 0 . 01% glutaraldehyde dissolved in PB , and washed with PB three times for 5 min each . Cells were then treated with 0 . 5 mg/mL Zymolyase 100T ( Seikagaku Co . , Tokyo , Japan ) in PB for 30 min . Because the cell walls were not removed well , the cells were further treated with 1 mg/mL Zymolyase 100T in PB for 30 min at 30°C , with 0 . 2 mg/mL Lysing Enzyme for 30 min , and washed with PB three times . After treatment with 100 mM lysine HCl in PB twice for 10 min and subsequent washing with PB , cells were permeabilized for 15 min with PB containing 0 . 2% saponin and 1% bovine serum albumin ( BSA ) , and incubated at 4°C overnight with primary antibody ( rabbit polyclonal anti-GFP antibody; Rockland , Limerick , PA ) diluted 1:400 in PB containing 1% BSA and 0 . 01% saponin . After washing with PB containing 0 . 005% saponin three times for 10 min each , cells were incubated for 2 hr at room temperature with secondary antibody ( goat anti-rabbit Alexa 594 FluoroNanogold Fab’ fragment , Nanoprobes , Yaphank , NY ) diluted 1:400 in PB containing 1% BSA and 0 . 01% saponin , washed with PB containing 0 . 005% saponin three times for 10 min each , and with PB once . Then , the cells were fixed again with 1% glutaraldehyde in PB for 1 hr , washed with PB once and treated with 100 mM lysine HCl in PB twice for 10 min each . The cells were stored at 4°C until further use . Before use , the cells were incubated with 50 mM HEPES ( pH5 . 8 ) three times for 3 min each , washed with distilled water ( dH2O ) once , and then incubated at 25°C for 3 min with the Silver enhancement reagent ( an equal-volume mixture of the following solutions A , B and C: A . 0 . 2% silver acetate solution . B . 2 . 8% trisodium citrate-2H2O , 3% citric acid-H2O , and 0 . 5% hydroquinone . C . 300 mM HEPES , pH 8 . 2 ) . Cells were then washed with dH2O three times . Cells were embedded in 2% low melting agarose dissolved in dH2O . Then , cells were post-fixed with 2% OsO4 in dH2O for 15 min at room temperature , washed with dH2O three times , stained with 1% uranyl acetate in dH2O for 1 hr , and washed with dH2O three times . Cells were dehydrated by sequential incubation in 50% and 100% ethanol for 10 min each , and with acetone for 10 min . For embedding in epoxy resin , cells were incubated sequentially with mixtures of acetone: Epon812 ( 1:1 ) for 1 hr , acetone:Epon812 ( 1:2 ) for 1 hr , and Epon812 overnight , and then Epon812 again for another 3 hr , and left to stand until solidified . The block containing cells was sectioned with a microtome ( Leica Microsystems , Tokyo , Japan ) , and the ultra-thin sections were doubly stained with 4% uranyl acetate for 20 min and lead citrate ( Sigma , Tokyo , Japan ) for 1 min as usually treated in EM methods . Images were obtained using a JEM1400 transmission electron microscope ( JEOL , Tokyo , Japan ) at 120kV . Images shown are taken from one of three independent biological replicate experiments . | Nearly all cells contain networks of filaments called microtubules that play many different roles . They provide internal structure; they serve as ‘tracks’ for transporting materials from one region of the cell to another; and they help to separate chromosomes during cell division . To understand how microtubules work , it is important to know how they are organized and distributed in cells . In several types of cells , including muscle cells in humans , and most plant cells , microtubules form on the surface of the cell nucleus – the membrane-bound compartment that stores genetic information . A key protein involved in microtubule formation , called Mto1 , is present on the surface of the nucleus , but it was not clear how Mto1 localizes there . Using fission yeast cells , Bao et al . have devised a new method to identify the proteins that recruit Mto1 to the surface of the nucleus . This revealed that Mto1 is recruited to the nuclear pores – large channels on the surface of the nucleus through which proteins can be transported . Unexpectedly , the proteins that recruit Mto1 to the nuclear pores do so using a mechanism that is also used to transport proteins out of the nucleus . Thus , in addition to determining how microtubules are organized at the cell nucleus , Bao et al . have identified a previously unknown role for the ‘nuclear transport machinery’ . Currently , drugs that inhibit the nuclear transport machinery form potential treatments for some cancers and viral infections . A better understanding of the multiple roles performed by the nuclear transport machinery may help researchers to design more effective inhibitor drugs . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology"
] | 2018 | Exportin Crm1 is repurposed as a docking protein to generate microtubule organizing centers at the nuclear pore |
Glycosylphosphatidylinositol ( GPI ) anchors attach nearly 150 proteins to the cell membrane . Patients with pathogenic variants in GPI biosynthesis genes develop diverse phenotypes including seizures , dysmorphic facial features and cleft palate through an unknown mechanism . We identified a novel mouse mutant ( cleft lip/palate , edema and exencephaly; Clpex ) with a hypo-morphic mutation in Post-Glycophosphatidylinositol Attachment to Proteins-2 ( Pgap2 ) , a component of the GPI biosynthesis pathway . The Clpex mutation decreases surface GPI expression . Surprisingly , Pgap2 showed tissue-specific expression with enrichment in the brain and face . We found the Clpex phenotype is due to apoptosis of neural crest cells ( NCCs ) and the cranial neuroepithelium . We showed folinic acid supplementation in utero can partially rescue the cleft lip phenotype . Finally , we generated a novel mouse model of NCC-specific total GPI deficiency . These mutants developed median cleft lip and palate demonstrating a previously undocumented cell autonomous role for GPI biosynthesis in NCC development .
Inherited glycophosphatidylinositol deficiency ( IGD ) disorders are a class of congenital disorders of glycosylation that affect the biosynthesis of the glycosylphosphatidylinositol ( GPI ) anchor . The clinical spectrum of IGDs is broad and includes epilepsy , developmental delay , structural brain malformations , cleft lip/palate , skeletal hypoplasias , deafness , ophthalmological abnormalities , gastrointestinal defects , genitourinary defects , heart defects , Hirschsprung’s disease , hyperphosphatasia , and nephrogenic defects ( Kinoshita , 2014; Knaus et al . , 2018; Tarailo-Graovac et al . , 2015; Bellai-Dussault et al . , 2019 ) . However , across all GPI deficiency disorders , the most penetrant defects affect the central nervous system and the craniofacial complex ( Kinoshita , 2014; Knaus et al . , 2018; Tarailo-Graovac et al . , 2015; Bellai-Dussault et al . , 2019 ) . Indeed , automated image analysis was able to predict the IGD gene mutated in each patient from facial gestalt ( Knaus et al . , 2018 ) . Interestingly , facial gestalt was a better predictor of patient mutation than analysis of the degree of GPI biosynthesis by flow cytometry . Little is known about the mechanism ( s ) that causes these phenotypes or why disparate tissues are differentially affected . We sought to determine the mechanism responsible for these phenotypes using a novel mouse model of reduced enzymatic function within the GPI biosynthesis pathway . The GPI anchor is a glycolipid added post-translationally to nearly 150 proteins which anchors them to the outer leaflet of the plasma membrane and traffics them to lipid rafts ( Kinoshita , 2014 ) . The biosynthesis and remodeling of the GPI anchor is extensive and requires nearly 30 genes ( Kinoshita , 2014 ) . Once the glycolipid is formed and transferred to the C-terminus of the target protein by a variety of Phosphatidylinositol Glycan Biosynthesis Class ( PIG proteins ) , it is transferred to the Golgi Apparatus for remodeling by Post-GPI Attachment to Proteins ( PGAP proteins ) . One of these PGAP proteins involved in remodeling the GPI anchor is Post-Glycosylphosphatidylinositol Attachment to Proteins 2 ( PGAP2 ) . PGAP2 is a transmembrane protein that catalyzes the addition of stearic acid to the lipid portion of the GPI anchor and cells deficient in Pgap2 lack stable surface expression of a variety of GPI-anchored proteins ( GPI-APs ) ( Kinoshita , 2014; Hansen et al . , 2013 ) . Autosomal recessive mutations in PGAP2 cause Hyperphosphatasia with Mental Retardation 3 ( HPMRS3 OMIM # 614207 ) , an IGD that presents with variably penetrant hyperphosphatasia , developmental delay , seizures , microcephaly , heart defects , and a variety of neurocristopathies including Hirschsprung’s disease , cleft lip , cleft palate , and facial dysmorphia ( Hansen et al . , 2013; Jezela-Stanek et al . , 2016; Krawitz et al . , 2013; Naseer et al . , 2016 ) . Currently , there is no known molecular mechanism to explain the cause of these phenotypes or therapies for these patients . In a forward genetic ENU mutagenesis screen , we previously identified the Clpex mouse mutant with Cleft Lip , Cleft Palate , Edema , and Exencephaly ( Clpex ) ( Stottmann et al . , 2011 ) . Here , we present evidence that this mutant phenotype is caused by a hypo-morphic allele of Pgap2 . To date , embryonic phenotypes of GPI biosynthesis mutants have been difficult to study due to the early lethal phenotypes associated with germline knockout of GPI biosynthesis genes ( McKean and Niswander , 2012; Nozaki et al . , 1999; Mohun et al . , 2013; Zoltewicz et al . , 1999 ) . In this study , we took advantage of the Clpex hypo-morphic mutant and a conditional knockout of GPI biosynthesis to determine the mechanism of the various phenotypes and tested the hypothesis that GPI-anchored Folate Receptor 1 ( FOLR1 ) is responsible for the phenotypes observed .
We previously identified the Clpex ( cleft lip and palate , edema , and exencephaly ) mutant in a mouse N-ethyl-N-nitrosourea ( ENU ) mutagenesis screen for recessive alleles leading to organogenesis phenotypes ( Stottmann et al . , 2011 ) . Clpex homozygous mutants displayed multiple partially penetrant phenotypes . In a subset of 70 mutants from late organogenesis stages ( ~E16 . 5-E18 . 5 ) , we noted cranial neural tube defects ( exencephaly ) in 61 ( 87% ) , cleft lip in 22 ( 31% ) , cleft palate in 13 ( 19% ) , and edema in six embryos ( 9% ) ( Figure 1A–H ) . Skeletal preparations of Clpex mutants identified a defect in frontal bone ossification ( Figure 1I–L , n = 5/5 mutants ) and a statistically significant decrease in limb length ( Figure 1M–P ) . We previously reported a genetic mapping strategy with the Mouse Universal Genotyping Array which identified a 44 Mb region of homozygosity for the mutagenized A/J genome on chromosome 7 ( Figure 1Q ) ( Stottmann et al . , 2011 ) . We then took a whole exome sequencing approach and sequenced 3 Clpex homozygous mutants . Analysis of single base pair variants which were homozygous in all three mutants with predicted high impact as determined by the sequence analysis pipeline ( e . g . missense variants in conserved residue , premature stop codons , etc . ) and not already known strain polymorphisms in dbSNP left only one candidate variant ( Table 1 ) . This was a homozygous missense mutation in the initiating methionine ( c . A1G , p . M1V ) in exon 3 of post-GPI attachment to proteins 2 ( Pgap2 ) . We confirmed the whole exome sequencing result by Sanger Sequencing ( Figure 1R ) . This mutation abolishes the canonical translation start codon for Pgap2 . However , there are multiple alternatively spliced transcripts that may lead to production of variant forms of Pgap2 . To determine whether the Clpex phenotype was caused by the missense mutation in Pgap2 , we performed a genetic complementation test using the Pgap2tm1a ( EUCOMM ) Wtsi ( hereafter referred to as Pgap2null ) conditional gene trap allele . We crossed Pgap2Clpex/+ heterozygotes with Pgap2null/+ heterozygotes to generate Pgap2Clpex/null embryos . Pgap2Clpex/null embryos displayed neural tube defects , bilateral cleft lip , and edema similar to the Pgap2Clpex/Clpex embryos at E13 . 5 ( Figure 2A–H ) . Pgap2Clpex/null embryos also displayed micro-opthalmia ( Figure 2F ) and more penetrant cleft lip and edema phenotypes than observed in Pgap2Clpex/Clpex homozygotes ( Figure 2I ) . Pgap2Clpex/null embryo viability was decreased with lethality at approximately E13 . 5-E14 . 5 , precluding analysis of palatal development in these mutants . Histological analysis of the heart in Pgap2Clpex/null E13 . 5 embryos showed pericardial effusion , a reduction in thickness of the myocardium , and an underdeveloped ventricular septum and valves ( Figure 2J–Q ) . As the Clpex allele failed to complement a null allele of Pgap2 , we concluded the Clpex phenotype is caused by a hypo-morphic allele of Pgap2 . Pgap2null/null embryos are resorbed before E9 . 0 due to early embryonic lethality , whereas Pgap2Clpex/Clpex homozygotes survive to E18 . 5 confirming the Clpex allele is indeed a hypo-morphic allele of Pgap2 which preserves some function as compared to a true null ( Mohun et al . , 2013 ) . To determine the possible protein expression of alternatively spliced forms of PGAP2 in the Clpex mutant , we performed western immunoblotting with a commercially available antibody against PGAP2 but were unable to detect endogenous PGAP2 in cells or mouse tissues ( data not shown ) . Therefore , we cannot definitively address the presence of alternatively spliced variants of PGAP2 in the Clpex mutant . However , a search for Pgap2 alternatively spliced transcripts identifies 25 alternatively spliced transcripts , 13 of which are protein coding ( Supplementary file 1 ) ( Hunt et al . , 2018 ) . While the majority of alternatively spliced transcripts utilize the start codon mutated in the Clpex allele , variant Pgap2-203 utilizes an alternative start codon which significantly alters the C-terminal domain of the protein when compared to the canonical transcript Pgap2-225 ( Figure 2—figure supplement 1 ) . The differences in the C-terminal domain between these transcript variants include the cytoplasmic tail and the first helical domain that is predicted to traverse the Golgi membrane . However , the variants are very similar beyond these domains and it is possible the hypo-morphic phenotype of Clpex mutants compared to Pgap2null/null mutants may be due to some residual function of transcript variant Pgap2-203 . Based on the tissues affected in the Clpex mutant , we hypothesized Pgap2 is expressed in the neural folds and facial primordia of the developing mouse embryo at early stages . We used the lacZ expression cassette within the Pgap2null allele to perform detailed expression analysis of Pgap2 throughout development ( n = 21 litters at multiple developmental stages ) . We performed RNA in situ hybridization in parallel for some stages to test the fidelity of the lacZ expression and found high concordance . Pgap2 was expressed relatively uniformly and ubiquitously at neurulation stages in the mouse from E7 . 5-E8 . 5 ( Figure 3A–C ) . We also noted extraembryonic expression at E7 . 5 , consistent with the abnormal placental development observed in Pgap2null/null embryos ( Figure 3A , B ) ( Mohun et al . , 2013 ) . At E9 . 0–10 . 5 , there was clear enrichment of Pgap2 expression in the first branchial arch ( Figure 3D–L ) . Pgap2 RNA in situ hybridization identified a similar pattern of expression as observed in the Pgap2 LacZ reporter allele at E9 . 5 ( Figure 3F ) . At E9 . 5 and E10 . 5 , expression was enriched in the limb bud , somites , first branchial arch , eye , forebrain and midbrain ( Figure 3E–L ) . There was also increased expression at the medial aspects of both medial and lateral nasal processes at E10 . 5 ( Figure 3K–L ) , and strong expression in the heart starting at E11 . 5 ( Figure 3N ) . At later organogenesis stages , Pgap2 showed more regionalized and enriched expression , including in the ganglion cell layer of the retina at E11 . 5 and E14 . 5 ( Figure 3M , S2F ) . At E16 . 5 Pgap2 was expressed in the salivary gland , epidermis , stomach , nasal conchae , myocardium , bronchi , kidney , uroepithelium , lung parenchyma , a specific layer of the cortex , and ear ( Figure 3—figure supplement 1 ) . Interestingly , Pgap2 showed lower expression in the liver ( Figure 3—figure supplement 1H ) and most of the brain except for a thin layer of the cortex and the choroid plexus at E16 . 5 and P0 ( Figure 3—figure supplement 1J–M ) . We also noted expression in the genital tubercle ( Figure 3—figure supplement 1L ) . We conclude Pgap2 shows tissue specific regions of increased expression which may help to explain why certain tissues such as the craniofacial complex , central nervous system , and heart are preferentially affected in GPI biosynthesis mutants . These data are in contrast to previous reports in which some GPI biosynthesis genes are shown to be ubiquitously and uniformly expressed , including Pign in the mouse and pigu in zebrafish ( McKean and Niswander , 2012; Nakano et al . , 2010 ) . Our Pgap2 expression is more consistent with the expression of Pigv which is enriched in C . elegans epidermal tissues ( Budirahardja et al . , 2015 ) . PGAP2 is the final protein in the GPI biosynthesis pathway and catalyzes the addition of stearic acid to the GPI anchor ( Tashima et al . , 2006 ) . In the absence of Pgap2 , cells lack a variety of GPI-APs on the cell surface leading to a functional GPI deficiency ( Hansen et al . , 2013; Jezela-Stanek et al . , 2016; Krawitz et al . , 2013; Naseer et al . , 2016; Tashima et al . , 2006 ) . To determine the effect of the Clpex mutation on Pgap2 function , we performed Fluorescein-labeled proaerolysin ( FLAER ) flow cytometry staining to quantify the overall amount of the GPI anchor on the cell surface . FLAER is a bacterial toxin conjugated to fluorescein that binds directly to the GPI anchor in the plasma membrane ( Brodsky et al . , 2000; Sutherland et al . , 2007 ) . We hypothesized Pgap2 function is impaired in Clpex mutants due to the ENU mutation in the initiating methionine . We found mouse embryonic fibroblasts ( MEFs ) from Clpex mutants displayed a significantly decreased FLAER staining compared to wildtype MEFs , consistent with a defect in GPI biosynthesis ( n = 3 separate experiments with 4 WT and 4 Clpex cell lines; Figure 4A , B; p=0 . 0407 ) . Our genetic complementation analysis results suggested the Clpex allele might be a hypo-morphic allele of Pgap2 . To test this hypothesis , we generated human embryonic kidney ( HEK ) 293T clones with a 121 bp deletion in exon 3 of PGAP2 with CRISPR/Cas9 ( termed PGAPnull/null cells; Figure 4—figure supplement 1 ) . In parallel , we recapitulated the Clpex mutation in three independent clones of HEK293T cells by CRISPR/Cas9 mediated homologous directed repair ( termed Clpex KI Clones 1 , 4 , and 7; Figure 4—figure supplement 1 ) . We found there was a statistically significant decrease in FLAER staining between WT and 3/3 KI clones and the PGAP2null/null cells . However , we observed a smaller difference in FLAER staining in PGAP2null/null cells when compared to Clpex KI cells ( Figure 4C , D ) . Therefore , we conclude the Clpex missense mutation severely affects PGAP2 function similar to the effect seen upon total depletion of PGAP2 . Our in vivo findings suggest the Clpex mutation produces a hypo-morphic allele of Pgap2 but our in vitro FLAER staining shows functional equivalence between PGAP2null/null cells and the Clpex KI cells . This may reflect subtle differences in the function of PGAP2 in vitro versus in vivo or reflect a difference between mouse and human PGAP2 ( FLAER was performed in human HEK293T cells ) . As a positive control , we used CRISPR/Cas9 to delete phosphatidylinositol glycan anchor biosynthesis , class A ( Piga; Figure 4—figure supplement 1 ) . Piga is the first gene in the GPI biosynthesis pathway and is absolutely required for GPI biosynthesis ( Jezela-Stanek et al . , 2016; Johnston et al . , 2012 ) . We utilized CRISPR/Cas9 to generate a 29 bp deletion in exon 3 of PIGA . While not statistically significant , these PIGAnull/null cells trend toward a further decrease in FLAER staining compared to PGAP2null/null cells , confirming our staining accurately reflects GPI anchor levels ( n = 4 separate experiments; Figure 4C , D ) . Current estimates suggest nearly 150 genes encode proteins which are GPI anchored ( UniProt Consortium , 2018 ) . Our manual review of the MGI database found 102 GPI-APs have been genetically manipulated and phenotyped in mice ( Smith et al . , 2018 ) . Of these , the null allele of Folr1 has a phenotype most similar to the Clpex mutant with cranial neural tube defects , cleft lip/palate , and heart outflow tract phenotypes ( Piedrahita et al . , 1999 ) . Tashima et . al . previously showed PGAP2 is required for stable cell surface expression of FOLR1 in CHO cells ( Tashima et al . , 2006 ) . To confirm this finding , we overexpressed a myc-tagged FOLR1 construct in WT and PGAP2null/null 293 T cells and assessed the presentation on the plasma membrane as marked by immunocytochemistry for wheat germ agglutinin ( WGA ) . We observed a decrease in co-localization of FOLR1 with WGA in PGAP2null/null cells compared to controls ( Figure 5A–C , G–I ) . In the absence of PIGA , cells lack the surface expression of any GPI-APs ( Kinoshita , 2014 ) . We found PIGAnull/null cells showed decreased co-localization of FOLR1 with WGA to a greater degree than that observed in PGAP2null/null cells ( representative images from n = 3 technical replicates , Figure 5D–F , p<0 . 0001 ) . We found PGAP2null/null cells show a severe defect in FOLR1-myc/WGA co-localization but retain more surface staining than PIGAnull/null cells ( Figure 5J , p<0 . 0001 ) . Therefore , we concluded PGAP2null/null cells have an intermediate defect in FOLR1 membrane trafficking as compared to PIGAnull/null cells . However , both PIGAnull/null and PGAP2null/null cells produced similar amounts of FOLR1 protein by western blot , indicating that the defect is in membrane trafficking , and not protein production ( Figure 5K , L ) . A number of GPI-APs are critical for cranial neural crest cell ( cNCC ) migration and survival; including ephrins and FOLR1 ( Santiago and Erickson , 2002; Holmberg et al . , 2000; Li et al . , 2011; Rosenquist , 2013; Rosenquist et al . , 2010; Spiegelstein et al . , 2004; Tang et al . , 2004; Zhu et al . , 2007; Wahl et al . , 2015 ) . This led us to the hypothesis that cNCC migration may be impaired in Clpex mutants ultimately causing the cleft lip and palate phenotype ( Holmberg et al . , 2000 ) . To test whether cNCC migration was impaired in the Clpex mutant , we performed a NCC lineage trace using the Wnt1-Cre mouse ( B6 . Cg-H2afvTg ( Wnt1-cre ) 11RthTg ( Wnt1-GAL4 ) 11Rth/J ) in combination with the R26R LacZ reporter ( B6 . 129S4-Gt ( ROSA ) 26Sortm1Sor/J ( R26RTg ) to create Wnt1-Cre;R26RTg/wt;Pgap2clpex/clpex mutants in which the NCCs are indelibly labeled with LacZ expression at E9 . 5 and E11 . 5 ( Brewer et al . , 2004; Chai et al . , 2000; Soriano , 1999; Brault et al . , 2001; Danielian et al . , 1998 ) . We observed no significant deficit in cNCC migration in the mutant embryos as compared to littermate controls at either stage ( representative images of n = 2 E9 . 5 mutants and n = 5 E11 . 5 mutants; Figure 6A–F ) . However , we observed hypoplasia of the medial and lateral nasal processes at E11 . 5 , suggesting the Clpex phenotype is due to earlier defects in NCC survival ( Figure 6E–F ) . As Pgap2 was highly expressed in the epithelium and epithelial barrier defects are a known cause of cleft palate , we next sought to determine whether the epidermis was compromised in the Clpex mutant ( Ingraham et al . , 2006 ) . We performed a Toluidine Blue exclusion assay but found no significant defects in barrier formation in the mutant ( Figure 6—figure supplement 1 ) . Folr1-/- mice and zebrafish Folr1 morphants display increased cell death and decreased proliferation in the facial primordia ( Spiegelstein et al . , 2004; Zhu et al . , 2007; Wahl et al . , 2015; Tang and Finnell , 2003 ) . We hypothesized a similar mechanism may be responsible for the cleft lip and palate in the Clpex mutant embryos . To test this hypothesis , we performed immunohistochemistry for αAP2 to mark NCCs and the apoptosis marker Cleaved Caspase 3 ( CC3 ) . We found the cNCCs of the first arch and a specific population of cells within the neuroepithelium were undergoing apoptosis significantly more frequently in Clpex homozygous mutants ( Figure 6G–N ) . The ratio of CC3-positive to AP2-positive spots revealed a highly significant increase in the percentage of CC3-positive spots in the first arch of Clpex mutants ( n = 2 or more sections from 3 WT and 6 Clpex mutants; Figure 6Q; p=0 . 0045 ) We also observed apoptosis in the cranial neuroepithelium bilaterally at the dorsolateral hinge points in mutant sections ( Figure 6O , P ) . The dorsolateral hingepoints in the anterior neural tube are crucial for proper closure of the neural tube . This apoptosis was exclusively confined to the cranial aspects of the neural tube at the midbrain-hindbrain boundary . Dietary folinic acid supplementation has been shown to rescue the early embryonic lethal phenotype of Folr1-/- mice and these mice can then survive to adulthood ( Spiegelstein et al . , 2004; Zhu et al . , 2007; Tang et al . , 2005 ) . Our data suggest FOLR1 receptor trafficking is impaired in the Clpex mutant ( Figure 5 ) , leading us to hypothesize folinic acid supplementation in utero may rescue the Clpex phenotype . We further hypothesized the folinic acid diet would have a greater beneficial effect than folic acid as folinic acid ( reduced folate ) has a higher affinity for other folate receptors including solute carrier family 19 ( folate transporter ) , member 1 ( Slc19a1 ) and solute carrier family 46 , member1 ( Slc46a1 ) , which are not GPI anchored ( Zhao et al . , 2001 ) . In comparison , folic acid has a higher affinity for the GPI anchored folate receptors FOLR1 and FOLR2 ( Jansen and Peters , 2015; Jakubowski et al . , 2009 ) . We supplemented pregnant Clpex dams from E0-E9 . 5 or E16 . 5 with a 25 parts per million ( ppm ) folinic acid diet , 25 ppm folic acid diet or control diet , and collected Clpex mutants for phenotypic analysis and CC3 staining for apoptosis ( Figure 7A ) . The folinic-acid-treated group had a significantly smaller proportion of mutants with cleft lip ( p=0 . 02 ) , but there was no effect on the incidence of NTD or cleft palate compared to control diet ( Figure 7B , C; p=0 . 87 ) . We did note a mild decrease in the number of mutants with edema ( Figure 7B , C; p=0 . 06 ) . Consistent with our hypothesis , we found the folinic acid diet reduced the number of mutants with cleft lip by 23% ( 2/25 mutant vs . 22/70 control ) , to a greater degree than the 10% reduction observed in folic acid treated embyros ( 3/14 , n . s . , Figure 7B , C ) . Therefore , we conclude folinic acid treatment was more effective than folic acid treatment at reducing the cleft lip phenotype in Clpex mutants . We hypothesized this would be via a decrease in NCC apoptosis in folinic acid treated Clpex mutants compared to control . However , CC3 staining in control diet treated Clpex mutants compared to folinic acid treated Clpex mutants showed no statistical difference in the number of apoptotic spots in folinic acid treated embryos at E9 . 5 ( Figure 7—figure supplement 1; p=0 . 3796 ) . These data argue the partial rescue of the cleft lip phenotype in folinic-acid-treated Clpex mutants is not due to a decrease in apoptosis at E9 . 5 in the Clpex mutant . This failure of folinic acid treatment to decrease the degree of apoptosis at E9 . 5 may explain why folinic acid treatment could not rescue the cleft palate or NTD phenotype . As folinic acid supplementation could not rescue all phenotypes observed in the Clpex mutant , we took an unbiased transcriptomic approach to determine the major signaling pathway ( s ) affected upon reduced Pgap2 function . RNA sequencing was performed on pooled RNA samples from wild-type and Clpex homozygous mutant embryos at E9 . 5 ( 5 of each in each RNA pool ) . Sorting differentially expressed genes in ToppGene showed that the most differentially regulated pathways were sequence-specific DNA binding genes including 39 genes ( p=1 . 79×10−7; Table 2; Chen et al . , 2009 ) . Among the sequence-specific DNA binding genes , the majority ( 23/39 genes in the category ) were transcription factors which have been implicated in anterior/posterior ( A/P ) patterning including Cdx2 , Cdx4 , Tbxt , Hmx1 , Lhx2 , and Lhx8 ( Savory et al . , 2009; Chawengsaksophak et al . , 2004; van Nes et al . , 2006; Abe et al . , 2000; Shedlovsky et al . , 1988; Munroe et al . , 2009; Porter et al . , 1997; Ando et al . , 2005 ) ( Table 2 ) . Anterior patterning genes were statistically significantly downregulated and posterior patterning genes were statistically significantly upregulated ( Table 3 ) . We confirmed changes in expression of three of these A/P patterning defects by RNA in situ hybridization at E9 . 5 ( Supplementary file 3 ) . We identified a decrease in Alx3 in Clpex mutants which is both an anterior patterning gene with a prominent role in frontonasal development and a genetic target of folate signaling ( Lakhwani et al . , 2010 ) . We investigated Lhx8 because it is expressed in the head at E9 . 5 and Lhx8-/- mice develop cleft palate ( Zhao et al . , 1999 ) . We found Lhx8 was decreased in Clpex mutant heads . Finally , the master posterior patterning gene Tbxt ( brachyury ) is critical for determining tail length and posterior somite identity ( Abe et al . , 2000; Shedlovsky et al . , 1988 ) . We found the Tbxt expression domain was shifted anteriorly in Clpex mutants compared controls at E8 . 5 ( Supplementary file 3 ) . The second and third most altered pathways identified by ToppGene were cholesterol transporter activity ( p=1 . 1×10−6 ) , and apolipoprotein binding ( p=2 . 34×10−6 ) , respectively . Upon closer inspection , the genes in these categories were largely genes expressed in the mesendoderm including Alpha fetal protein and the Apolipoprotein gene family . We concluded the decreased expression of these genes in the Clpex mutant embryos is consistent with a defect in mesendoderm induction , rather than specific cholesterol and apolipoprotein activities . Collectively , these findings from our transcriptomic analysis suggest other GPI-Aps involved in A/P patterning and mesendoderm development may be affected in Clpex homozygous mutant embryos . Multiple other mutations in GPI biosynthesis genes including Pgap1 and Pign lead to defective CRIPTO mediated NODAL/BMP signaling which affects the formation of the A/P axis in the early gastrulating embryo ( McKean and Niswander , 2012; Zoltewicz et al . , 1999; Chen et al . , 2008; Zoltewicz et al . , 2009 ) . CRIPTO is a co-receptor for NODAL and necessary for the induction of the anterior visceral endoderm and subsequent forebrain and mesendoderm formation ( Thomas and Beddington , 1996; Rossant and Tam , 2009 ) . Mckean et al found CRIPTO signaling was impaired in the PignGonzo and Pgap1Beaker GPI biosynthesis mutants ( McKean and Niswander , 2012 ) . Furthermore , stem cells from GPI-deficient clones are unable to respond to TGFβ superfamily members due to defects in GPI anchored co-receptor anchoring ( McKean and Niswander , 2012; Chen et al . , 2008 ) . Zoltewicz et al found mutations in Pgap1 lead to defective A/P patterning by affecting other major signaling pathways including Wnt ( Zoltewicz et al . , 2009 ) . Our RNA-Seq results are consistent with the existing literature which has established a critical role for GPI biosynthesis in generating the A/P axis ( Lee et al . , 2016 ) . While this role is well established , few groups have investigated the tissue-specific role of GPI biosynthesis after the A-P axis has been established . Interestingly , two studies have found GPI-APs have cell autonomous roles separately in skin and limb development ( Ahrens et al . , 2009; Tarutani et al . , 1997 ) . As we found tissue-specific defects in the NCC population in the Clpex mutant , we sought to address a larger question and determine the cell autonomous role for GPI-APs in NCC development . We observed cell-type-specific apoptosis in the cNCCs in the Clpex mutant and to further our understanding we sought to determine the cell autonomous role of GPI biosynthesis more generally in these cells . Phosphatidylinositol glycan anchor biosynthesis , class A ( Piga ) is part of the GPI-N-acetylglucosaminyltransferase complex that initiates GPI biosynthesis from phosphatidylinositol and N-acetylglucosamine ( Kinoshita , 2014 ) . Piga is totally required for the biosynthesis of all GPI anchors and Piga deletion totally abolishes GPI biosynthesis ( Nozaki et al . , 1999; Miyata et al . , 1993; Watanabe et al . , 1996 ) . We first performed RNA whole mount in situ hybridization for Piga and showed it has a similar regionalized expression as we observed in the Pgap2 expression experiments . Piga expression at E11 . 5 is enriched in the first branchial arch , heart , limb , and CNS ( representative images from n = 8 antisense and two sense controls over three separate experiments , Figure 8A–F ) . However , Piga showed a unique enrichment in the medial aspect of both medial nasal processes as opposed to the Pgap2 expression which appeared to line the nasal pit epithelium ( Figure 8C ) . Other GPI biosynthesis genes showed a similar regionalization pattern of expression ( Figure 8—figure supplement 1 ) . To determine the NCC specific role for GPI biosynthesis , we generated a novel model of tissue-specific GPI deficiency in the neural crest cell lineage with Pigaflox/X; Wnt1-Cre mosaic conditional KO ( cKO ) mutants and Pigaflox/Y; Wnt1-Cre hemizygous cKO mutants . As Piga is located on the X chromosome , females with genotype Pigaflox/X; Wnt1-Cre+/- will develop genetic mosaicism due to random X inactivation whereas males with genotype Pigaflox/Y; Wnt1-Cre+/- will develop total Piga deficiency in the Wnt1-Cre lineage . To confirm the loss of GPI biosynthesis in these mutants , we cultured Mouse Palatal and Nasal Mesenchymal Cells ( MPNMCs ) from WT and mutant palates and performed FLAER staining as above . We found mutant MPMNCs lack virtually all GPI anchors on the cell surface ( n = 4 WT and three mutant cell lines stained in two separate experiments ) ( Figure 8G , H , p=0 . 0036 ) . Analysis of mosaic cKOs at E15 . 5–16 . 5 showed mild median cleft lip and cleft palate in all mutants examined ( n = 6 mutants; Figure 9A–F ) . Hemizygous cKOs showed a more severe median cleft lip and cleft palate in all mutants examined ( n = 7 mutants; Figure 9G–L ) . Skeletal preparations to highlight bone and cartilage demonstrates hypoplasia of the craniofacial skeleton and cleft palate ( n = 5/5 mutants examined; Figure 9M–R ) . These data confirm for the first time a cell autonomous role for GPI biosynthesis in cNCCs during development . As we reduced the amount of Piga from the mosaic cKO to the hemizgous cKO , we observed a worsening of the cleft lip/palate phenotype including more severe hypoplasia of the palatal shelves and widening of the median cleft lip . These data are consistent with the hypothesis that the dosage of GPI biosynthesis is related to the severity of the phenotype with mutants with less residual GPI anchor expression showing more severe phenotypes . Surprisingly , we found hemizygous cKO mutants are capable of forming all the bones and cartilage of the craniofacial skeleton , although they are all hypoplastic . These data are consistent with the hypothesis that GPI biosynthesis is involved in the survival of early cNCCs as we observed in the Clpex homozygous mutants and not in the later patterning or differentiation of cNCCs . The critical requirement for GPI biosynthesis appears to be at early stages of cNCC survival just after they have migrated from the dorsal neural tube , and before they have committed to differentiation into bone or cartilage .
In this study , we aimed to determine the role of GPI biosynthesis in craniofacial development with two novel models of GPI deficiency . First , we characterized the phenotype of the ENU-induced Clpex allele which shows partially penetrant cranial neural tube defects , bilateral cleft lip/palate , and edema . We found by mapping and whole exome sequencing the Clpex mutation is a missense allele in the initiating methionine of Pgap2 , the final enzyme in the GPI biosynthesis pathway . The Clpex allele failed to complement a null allele of Pgap2 , confirming the Clpex mutant is caused by a hypo-morphic mutation in Pgap2 . By expression analysis with a Pgap2tm1a ( lacZ ) reporter allele , we found Pgap2 is enriched in the first branchial arch , limb bud , neuroepthelium and around the interior aspect of the nasal pits during lip closure at E9 . 5-E11 . 5 . In later stages of organogenesis , Pgap2 is widely expressed and enriched in epithelia . These data argue expression of GPI biosynthesis genes is dynamic during development and not simply uniform and ubiquitous . FLAER flow cytometry and expression of a tagged FOLR1 showed that reduced levels of Pgap2 affected GPI biosynthesis , although not as severely as a total knockout for the GPI biosynthesis pathway , PIGAnull/null . Molecular analysis showed the Clpex mutants have increased apoptosis in cNCCs and cranial neuroepithelium . Folinic acid diet supplementation in utero partially rescued the cleft lip in Clpex mutants . Finally , we generated a NCC tissue-specific GPI-deficient model to determine the cell autonomous role of GPI biosynthesis . Pigaflox/X; Wnt1-Cre mosaic cKO mutants and Pigaflox/Y; Wnt1-Cre hemizygous cKO mutants displayed fully penetrant median cleft lip/cleft palate and craniofacial hypoplasia similar to our germline Clpex mutant , confirming a cell autonomous role for GPI biosynthesis in craniofacial development . Contrary to previous studies of other GPI biosynthesis pathway genes , we found Pgap2 clearly shows enriched expression in certain tissues during certain stages of development . We observed a similar pattern in Piga RNA expression suggesting GPI biosynthesis genes share similar gene enrichment domains . These tissues are the most affected in GPI biosynthesis mouse mutants and include the craniofacial complex , CNS , limb , and heart . This may mean Pgap2 and other GPI biosynthesis genes are required in certain tissues for anchoring GPI-APs critical to that tissue . Alternatively , these areas may be particularly ‘GPI-rich . ’ A variety of mutants have been described in the GPI biosynthesis pathway with a wide array of phenotypes ( Kinoshita , 2014; Bellai-Dussault et al . , 2019 ) . While germline mutants in this pathway remain poorly understood , recent research in Paroxysmal Nocturnal Hemoglobinuria ( PNH ) caused by somatic mutations in PIGA has revolutionized our understanding of GPI deficiency related pathology . In PNH , clones of GPI deficient hematopoietic stem cells proliferate in the bone marrow and give rise to blood cells that lack GPI-anchored CD55/59 which are required to prevent complement-mediated lysis of red blood cells . PNH patients suffer from episodes of hemolysis and thrombosis which can be deadly ( Hill et al . , 2017 ) . Blockade of complement in these patients via eculizumab , a monoclonal antibody that inhibits the conversion of C5 to C5a and C5b , has been shown to greatly improve survival ( Brodsky et al . , 2008; Rother et al . , 2007; Hillmen et al . , 2007; Hillmen et al . , 2006 ) . Thus , a single GPI-AP seems to be largely responsible for the disease observed in these patients . In this study , we aimed to identify a single GPI-AP that could be responsible for all the phenotypes observed in our germline GPI biosynthesis Clpex mutant . Of the known GPI-AP knockout models , Clpex shares the most phenotypic overlap with the Folr1null/null mouse . We directly tested the hypothesis that FOLR1 deficiency is solely responsible for the Clpex phenotype by dietary supplementation of folinic acid during embryonic development . To our surprise , folinic acid supplementation could partially rescue the cleft lip phenotype , but not the NTD or cleft palate ( Figure 6B , C ) . Folinic acid treatment could not rescue the apoptosis we observed in control diet treated Clpex mutants suggesting folinic acid is necessary for some other aspect of lip development in Clpex mutants ( Figure 7—figure supplement 1 ) . The failure of folinic acid to rescue the apoptosis in the cNCCs and in the neuroepithelium of Clpex mutants suggests that either , other GPI-APs are responsible for this apoptosis , or a combination of GPI-APs are required to prevent this apoptosis and folinic acid alone is not sufficient to prevent the apoptosis . Alternatively , the mistrafficking of GPI-APs may result in a cellular response such as the unfolded protein response which may trigger apoptosis in these cells . Further research is required to determine whether this may be the case . The partial rescue of cleft lip in Clpex mutants with high doses of folinic acid in utero suggests folinic acid may be a possible therapeutic for some phenotypes in patients with GPI biosynthesis variants . Further research is required to test whether the positive effect of folinic acid on the Clpex mutants could be observed in other GPI biosynthesis mutants . These data also argue the phenotypes observed in germline Clpex mutants do not share a single mechanism and are not due to the loss of a single GPI-AP given the varied response in different tissues to the rescue regimens utilized here . Many GPI-APs could be responsible for the phenotypes we observe in the Clpex mutant but were not tested explicitly in this work . Notably , the two receptors for Glial Derived Neurotrophic Factor ( GDNF ) are GPI- anchored ( GFRA1 , GFRA2 ) . GFRA1 and GFRA2 are known to be critical for the survival and development of NCCs in the gut during enteric nervous system development ( Tomac et al . , 2000; Enomoto et al . , 1998; Rossi et al . , 1999 ) . Interestingly , Gfra1 and Gfra2 are expressed in the craniofacial complex during development ( Golden et al . , 1998; Visel , 2004 ) . Whether GDNF plays a crucial in cNCC survival remains to be explored . Other candidate GPI-APs that may be affected by loss of Pgap2 including one form of Neural Cell Adhesion Molecule ( NCAM ) , a critical neural cell adhesion molecule . Ncam1null/null mice display defects in neural tube development including kinking and delayed closure ( Rabinowitz et al . , 1996 ) . A third candidate includes the glypican family members which are GPI anchored heparin sulfate proteoglycans that play critical roles in cell-cell signaling and have been shown to modulate critical patterning gradients in the neural tube and face including Sonic Hedgehog and Wnt ( Bassuk et al . , 2013; Capurro et al . , 2008; Galli et al . , 2003; Song and Filmus , 2002 ) . Other GPI-AP knockout models display NTDs including Repulsive guidance molecule A/B ( Rgma ) and Ephrin A5 ( Efna5 ) . However , Rgmanull/null mice do not develop increased apoptosis in the neuroepithelium as we observed in Clpex mutants ( Niederkofler et al . , 2004 ) . Efna5-/- mice appear to form DLHPs , though the neural folds do not fuse in the midline which is less severe than the defect we observe in Clpex mutants ( Holmberg et al . , 2000 ) . Therefore , we find it unlikely the loss of these GPI-APs are primarily responsible for the defects observed in the Clpex mutant , although contributions to the phenotype may come from abnormal presentation of one or several of these GPI-APs on the cellular membranes . It has been known for decades that treatment of embryos with phospholipase C to release GPI-APs from the cell surface causes NTD in utero ( O'Shea and Kaufman , 1980 ) . To investigate the cause of the NTD in Clpex mutants , we performed histological and immunohistochemical analysis of the mutant at neurulation stages . We found the Clpex mutant fails to form dorsolateral hinge points and the cranial neuroepithelium is apoptotic in the region of the developing DLHP . Neuroepithelial apoptosis was restricted to the midbrain/hindbrain boundary and likely explains why Clpex mutants develop cranial NTDs as opposed to caudal NTDs such as spina bifida . These cellular defects likely underlie the NTD but the cause of the neuroepithelial apoptosis remains unclear as the NTD did not respond to folinic acid supplementation . It remains controversial , but the NTD in Folr1null/null mice may be related to an expansion of the Shh signaling domain that patterns the neural tube ( Tang and Finnell , 2003; Murdoch and Copp , 2010 ) . Indeed , many Shh gain-of-function mutants develop NTD as Shh expansion impairs the formation of DLHPs and closure of the neural tube ( Murdoch and Copp , 2010 ) . Our RNA sequencing analysis did not identify a dysregulation in the Shh signaling pathway so there are likely differences in the mechanism responsible for the NTD in Folr1null/null mice and Clpex mutants . To determine alternative mechanisms responsible for the Clpex phenotype , we performed RNA sequencing from E9 . 5 WT and Clpex mutants . We found the largest differences in gene expression were in A/P patterning genes and mesendoderm induction genes . The A/P axis and induction of mesendoderm has been shown to require GPI-anchored CRIPTO , a Tgfβ superfamily member co-receptor of NODAL . A variety of studies have shown CRIPTO/Tgfβ super family members pathway function is impaired in GPI biosynthesis mutants because CRIPTO is GPI-anchored and cleavage of the anchor affects CRIPTO function ( McKean and Niswander , 2012; Chen et al . , 2008; Lee et al . , 2016 ) . While GPI deficiency has been studied in the context of A/P patterning , this is the first study to implicate GPI biosynthesis in the survival of neural crest cells in a cell autonomous fashion . Indeed , the enrichment of Piga in the developing medial nasal process and the median cleft lip/cleft palate and craniofacial hypoplasia in our Piga cKOs confirms a unique cell autonomous role for GPI biosynthesis in these structures . Interestingly , these mutants do not show a complete loss of the craniofacial skeleton , rather a general , mild hypoplasia consistent with a role for GPI biosynthesis in early NCC survival , but not later patterning or differentiation . Our study provides potential mechanistic explanations for the developmental defects observed in a GPI biosynthesis mutant model . We propose GPI biosynthesis is involved in anchoring critical survival factors for NCCs and the neuroepithelium . In GPI-deficient states , NCCs undergo apoptosis leading to hypoplastic nasal processes and palatal shelves . As we reduced the degree of GPI biosynthesis from the germline Clpex mutant hypomorph to our totally GPI-deficient NCC cKO model , we observed a worsening of the craniofacial phenotype as witnessed by the fully penetrant cleft lip/cleft palate and craniofacial hypoplasia . These data argue the degree of GPI deficiency correlates with the severity of the phenotype . In the neuroepithelium , loss of neuroepithelial cells at the DLHPs result in failure to bend and close the neural tube . Conditional ablations of critical GPI biosynthesis genes in other affected tissues including the CNS and heart will likely lead to new understandings of the diverse pathology of inherited glycophosphatidylinositol deficiency .
All animals were maintained through a protocol approved by the Cincinnati Children’s Hospital Medical Center IACUC committee ( IACUC2016-0098 ) . Mice were housed in a vivarium with a 12 hr light cycle with food and water ad libitum . The Clpex line was previously published by Stottmann et al . ( 2011 ) . Pigaflox ( B6 . 129-Pigatm1 ) mice were obtained from RIKEN and were previously generated by Taroh Kinoshita and Junji Takeda ( Nozaki et al . , 1999 ) . Wnt1-Cre ( B6 . Cg-H2afvTg ( Wnt1-cre ) 11RthTg ( Wnt1-GAL4 ) 11Rth/J ) mice and R26R LacZ reporter ( B6 . 129S4 Gt ( ROSA ) 26Sortm1Sor/J; R26RTg ) mice were purchased from Jackson Laboratories and previously published . Pgap2null ( Pgap2tm1a ( EUCOMM ) Wtsi ) mice were obtained from EUCOMM and genotyped using their suggested primers . Primers used to genotype all animals are listed in Supplementary file 2 . Sample2SNP custom Taqman probes were designed by Thermo-Fisher and used to genotype the point mutation in the Clpex line . Mapping of the Clpex mutation was previously described ( Stottmann et al . , 2011 ) . Whole exome sequencing was done at the CCHMC DNA Sequencing and Genotyping Core . The Pgap2 exon three variant was Sanger sequenced after PCR amplification and purification using the Zymo DNA clean and Concentrator kit ( Zymo Research Corporation , Irvine , CA ) . Alternative transcripts of Pgap2 were identified using Ensembl Genome browser and the UCSC genome browser ( Hunt et al . , 2018; Kent et al . , 2002 ) . Transcripts were aligned using ExPasy software ( Swiss Institute of Bioinformatics , Switzerland ) ( Gasteiger et al . , 2003 ) . RNA in situ hybridization was performed as previously described ( Belo et al . , 1997 ) . Briefly , whole E8-E11 . 5 embryos were fixed overnight in 4% PFA at 4°C and dehydrated through a methanol series . Samples were treated with 4 . 5 μg/mL Proteinase K for 7–13 min at room temperature , post-fixed in 4% PFA/0 . 2% glutaraldehyde and blocked with hybridization buffer prior to hybridization overnight at 65°C with constant agitation . The samples were washed and incubated with an anti-Digoxigenin antibody ( Roche #11093274910 ) o/n at 4°C . Embryos were washed and incubated with NBT/BCIP ( SIGMA ) or BM Purple ( Roche #11442074001 ) from 4 hr at room temperature to o/n at 4°C . Piga ( #MR222212 ) , Pgap2 ( #MR2031890 ) Pigp ( #MR216742 ) , Pigu ( #MR223670 ) , Pigx ( #MR201059 ) , Lhx8 ( #MR226908 ) , and Tbxt ( #MR223752 ) plasmids were obtained from Origene ( Rockville , MD ) . Antisense probes were generated from PCR products containing T3 polymerase overhangs . Piga , Pgap2 , Pigp , Pigu , Pigx , Lhx8 , and Tbxt antisense probes were generated from 910 , 750 , 556 , 952 , and 416 , 519 , and 665 base pair products , respectively . The PCR products were purified , in vitro transcription was performed with digoxigenin-labeled dUTP ( Roche #11277073910 ) , and the probe was purified with the MEGAclear Transcription Clean-up kit ( Thermo #AM1908 ) per the manufacturer’s instructions . For sense probes , the plasmids were cut with XhoI restriction enzyme after the coding sequence and T7 RNA polymerase was used for in vitro transcription . The Alx3 probe was generated by in vitro transcription of a 790 bp PCR product from Alx3 plasmid ( DNASU #MmCD00081160 ) containing T3 polymerase overhangs . MEFs were generated from E13 . 5 embryos . Embryos were dissected in PBS , decapitated , and eviscerated . The remaining tissue was incubated in trypsin o/n at 4°C to allow for enzymatic action on the tissue and remaining fibroblasts were passaged in complete DMEM containing 10%FBS and penicillin/streptomycin . MEFs were stained within three passages of their isolation . MEFs and 293 T cells were stained with 5 μL of Alexafluor-488 proaerolysin ( FLAER ) /1 × 106 cells ( CedarLane Labs , Burlington , Ontario , Canada ) and flow cytometry was performed on Becton-Dickinson FACSCanto II flow cytometer in the CCHMC Research flow cytometry core . Mouse Palatal Nasal Mesenchymal Cells ( MPNMCs ) were generated from E13 . 5-E14 . 5 microdissected embryo heads in a protocol similar to that used for MEPMS ( Fantauzzo and Soriano , 2017 ) . The lower jaw , eyes and brain were removed and the remaining upper jaw and nasal mesenchyme were lysed in 0 . 25% trypsin for 10 min at 37°C , passaged through a P1000 pipette several times to create a single-cell suspension , and cultured in 12 well plates . These cells displayed a stellate mesenchymal cell appearance after culture overnight . They were then stained 72 hr after isolation with FLAER . We utilized a double guide approach to generate knockout clones with deletions in PGAP2 and PIGA in HEK293T cells . Two small guide RNAs targeting exon 3 of either PGAP2 or PIGA were designed using Benchling software ( Benchling , San Francisco , CA ) and 5’ overhangs were added for cloning into CRISPR/Cas9 PX459M2 puromycin-resistance vector ( Ran et al . , 2013 ) . We also generated a single gRNA and donor oligonucleotide for homologous recombination to recapitulate the Clpex mutation in 293 T cells ( Integrated DNA Technologies ultramer ) . We cloned these guides into the PX459M2 plasmid using the one-step digestion-ligation with BbsI enzyme as described by Ran et . al . ( Ran et al . , 2013 ) . For knockout line , two guides per gene were transfected in WT 293 T cells using Lipofectamine 3000 . For Clpex knock-in lines , one sgRNA and the donor oligonucleotide containing 5’ and 3’ phosphorothiolate bonds were transfected into WT 293 T cells using Lipofectamine 3000 . Cells were selected for transfection by 3 days of culture in 10 μg/mL puromycin . Transfected cells were plated at clonal density into a 96-well plate and single clones were scored approximately one week post seeding . Single clones were Sanger sequenced to confirm deletion of the target exon 3 sequence of either PIGA or PGAP2 . PIGA clones carry a 50 bp out-of-frame deletion in PIGA and lack virtually all GPI expression on the cell surface by FLAER flow cytometry staining . PGAP2 clones carry a 121 bp out-of-frame deletion in PGAP2 . Clpex Knock-in clones were Sanger sequenced to identify clones carrying the desired knock-in mutation and clones with indels were discarded . Primers for sgRNA cloning and PCR amplification of targeted regions can be found in Supplementary file 2 . Sequencing of clones is presented in Figure 4—figure supplement 1 . HEK293T cells were purchased from ATCC and STR profiling at Genetica LabCorp . showed a HEK293 match . Cells were found to be mycoplasma negative via a LookOut Mycoplasma PCR Detection Kit ( SIGMA ) . 293 T cells were transfected with FOLR1-myc constructs ( Origene #RC212291 ) using Lipofectamine 3000 , incubated for 48 hr , then fixed for 15 min in 4% PFA , and blocked in 4% normal goat serum in PBS . They were stained o/n at 4°C with 1:1000 rabbit anti-myc ( Abcam ab9106 ) , washed the next day and stained with 1:500 488-congugated goat anti-rabbit ( Thermo #A11008 ) . They were then stained for 5 min with 5 μg/mL wheat germ agglutinin ( Thermo Fischer #W21405 ) and counter-stained with DAPI . They were visualized on Nikon C2 confocal microscope and Pearson co-efficient analysis was performed by Imaris software utilizing 8–15 60x z-stack confocal images/genotype . E9 . 5 embryos were dissected , fixed in 4% PFA o/n , equilibrated in 30% sucrose o/n , cryo-embedded in OCT , and sectioned from 10 to 20 μM by cryostat . Sections were subjected to antigen retrieval by citrate retrieval buffer , blocked in 4% normal goat serum , incubated in primary antibody 1:20 mouse anti-AP2 ( Developmental Studies Hybridoma Bank , University of Iowa , 3B5 supernatant ) and 1:300 rabbit anti-Cleaved Caspase 3 ( Cell Signaling Technology , Danvers , MA #9661 ) o/n in humid chamber . Sections were incubated with secondary antibody 1:500 Alexafluor 488-congugated goat anti-rabbit ( Thermo #A11008 ) and 1:500 Alexafluor 594 conjugated goat anti-mouse ( Thermo A11008 ) and counterstained with DAPI . Sections were imaged on Nikon C2 confocal microscope and CC3+ cells and AP2+ cells were quantified with Nikon Elements software brightspot analysis . 293 T cells were transfected with FOLR1-myc constructs using Lipofectamine 3000 , incubated for 48 hr and lysed in Pierce RIPA buffer ( Thermo #89901 ) containing Protease Inhibitor cocktail ( Roche #11697498001 ) . Lysate protein concentration was determined by BCA assay and electrophoresis was performed on a 10% Tris-glycine gel . Protein was transferred to a PVDF membrane , blocked in Odyssey blocking buffer and incubated o/n at 4°C with 1:1000 Rabbit anti-myc ( Abcam ab9106 ) and 1:1000 Mouse anti-Tubulin ( Sigma #T6199 ) antibodies ) . Membranes were washed and incubated for 1 hr in 1:15000 goat anti-rabbit IRDye 800CW ( LICOR # 926–32211 ) and 1:15000 goat anti-mouse IRDye 680Rd ( LICOR , #926–68070 ) and visualized on LICOR Odyssey imaging system . Whole embryos E8-E16 . 5 were fixed in formalin and embedded in paraffin for coronal sectioning and stained with hematoxylin and eosin using standard methods . Clpex heterozygous females were crossed to Wnt1-Cre R26R transgenic mice as described in results . Whole embryos were fixed in 4% PFA for 15 min at RT , washed in lacZ buffer , and stained in a solution containing 1 mg/mL X-gal ( Sigma #B4252 ) ( Behringer , 2014 ) . They were washed three times in PBS-T and imaged after several hours in X-gal stain at room temperature . Clpex pregnant dams were treated with either control chow , chow +25 ppm folic acid , or chow +25 ppm folinic acid generated by Envigo ( Indianapolis , Indiana ) from E0-E16 . 5 ad libitum ( Control diet #TD . 160112 , Folic acid diet #TD . 160472 , Folinic acid diet #TD . 160746 ) . They were euthanized at either E9 . 5 or E16 . 5 to assess phenotype . 5 WT and 5 Clpex mutant E9 . 5 embryos were dissected from the yolk sac and snap frozen on dry ice . RNA was isolated and pooled samples of each genotype were used for paired-end bulk-RNA sequencing ( BGI-Americas , Cambridge , MA ) . RNA-Seq analysis pipeline steps were performed using CSBB [Computational Suite for Bioinformaticians and Biologists: https://github . com/csbbcompbio/CSBB-v3 . 0] . CSBB has multiple modules , RNA-Seq module is focused on carrying out analysis steps on sequencing data , which comprises of quality check , alignment , quantification and generating mapped read visualization files . Quality check of the sequencing reads was performed using FASTQC ( http://www . bioinformatics . bbsrc . ac . uk/projects/fastqc ) . RNA-Seq reads for the mutant and wildtype were paired-end and had ~43 and~31 million reads respectively . Reads were mapped ( to mm10 version of Mouse genome ) and quantified using RSEM-v1 . 3 . 0 ( Li and Dewey , 2011 ) . Differential expression analysis was carried out by EBSeq [https://www . biostat . wisc . edu/~kendzior/EBSEQ/] ( Leng et al . , 2013 ) . Differential transcripts are filtered based on LogFC and p-value . Filtered DE transcripts are used for functional and pathway enrichment using toppgene [https://toppgene . cchmc . org/] ( Chen et al . , 2009 ) . For skeletal preparation , E16 . 5-E18 . 5 embryos were eviscerated and fixed for 2 days in 95% ethanol . They were stained overnight at room temperature in Alcian blue solution ( Sigma #A3157 ) containing 20% glacial acetic acid . They were destained for 24 hr in 95% ethanol and slightly cleared in a 1% KOH solution o/n at room temperature . They were then stained o/n in Alazarin red solution ( Sigma #A5533 ) containing 1% KOH . They were then cleared for 24 hr in 20% glycerol/1%KOH solution . Finally , they were transferred to 50% glycerol/50% ethanol for photographing . E18 . 5 embryos were dehydrated through a methanol series and then rehydrated . Next , they were placed in 0 . 1% Toludine Blue ( Sigma #89640 ) in water for 2 min on ice . They were destained in PBS on ice and imaged . Statistical analysis was performed using Graphpad Prism ( GraphPad Software , San Diego , CA ) . Tests between two groups were carried out using unpaired , two-tailed student’s t-test . Diet , comparisons between groups for FLAER staining , and subcellular localization ( Pearson Correlation co-efficient ) results were analyzed with one-way ANOVA on Graphpad Prism ( Graphpad Software , San Diego , CA ) . For statistical analysis of phenotypes observed for embryos under varying diet conditions , z-test of proportions was used . Significance was labeled with one asterisk = p < 0 . 05 , two asterisks = p < 0 . 01 , three asterisks = p < 0 . 001 , and four asterisks = p < 0 . 0001 . | Many of the proteins that cells produce have sugar molecules attached to them . These additions , called glycosylations , often help to deliver proteins to the parts of the cell where they are needed . In some genetic disorders , individuals have gene mutations that prevent glycosylation from occurring properly . This can lead to a variety of symptoms including seizures , cleft palates and heart defects . It was not clear how changes in glycosylation cause these symptoms . A GPI anchor is a specific glycosylation that helps to attach many different proteins to the outer membrane of cells . Lukacs et al . created mouse models with genetic mutations that prevent GPI anchors from forming correctly , and studied the effects these had in mouse embryos . This revealed that a loss of GPI anchors early in embryonic development causes the cells that produce the face to die , as they are very sensitive to an early loss of glycosylation . Because too few face cells survive , embryos develop cleft palate , and other reductions in facial tissues . However , giving the embryos supplements of folinic acid in the womb reduced these effects . In the future , further experiments using the genetically altered mice generated by Lukacs et al . could explore how glycosylation affects the development of other tissues and organs , like the heart and liver . This could ultimately help researchers to predict the effects of certain genetic conditions and to develop new treatments for them . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology"
] | 2019 | Glycosylphosphatidylinositol biosynthesis and remodeling are required for neural tube closure, heart development, and cranial neural crest cell survival |
While most animals thermotax only to regulate their temperature , female mosquitoes are attracted to human body heat during pursuit of a blood meal . Here we elucidate the basic rules of Aedes aegypti thermotaxis and test the function of candidate thermoreceptors in this important behavior . We show that host-seeking mosquitoes are maximally attracted to thermal stimuli approximating host body temperatures , seeking relative warmth while avoiding both relative cool and stimuli exceeding host body temperature . We found that the cation channel TRPA1 , in addition to playing a conserved role in thermoregulation and chemosensation , is required for this specialized host-selective thermotaxis in mosquitoes . During host-seeking , AaegTRPA1-/- mutants failed to avoid stimuli exceeding host temperature , and were unable to discriminate between host-temperature and high-temperature stimuli . TRPA1-dependent tuning of thermotaxis is likely critical for mosquitoes host-seeking in a complex thermal environment in which humans are warmer than ambient air , but cooler than surrounding sun-warmed surfaces .
Thermotaxis is a sensory-motor behavior that guides animals toward a preferred temperature . This type of sensory navigation allows animals to avoid environments of noxious cold and heat , with the goal of remaining in physiologically suitable ambient temperatures . For ectotherms , such as most insects , thermotaxis behavior is the primary method of thermoregulation . Terrestrial invertebrates are vulnerable to temperature extremes , facing the risk of desiccation at elevated temperatures , and rapid hypothermia at low temperatures . Therefore , mechanisms to detect environmental temperatures and trigger appropriate approach or avoidance behaviors are extremely important for their survival . For instance , adult Caenorhabditis elegans worms migrate preferentially toward a specific thermal environment determined by the temperature of their cultivation ( Hedgecock and Russell , 1975; Mori and Ohshima , 1995 ) . Adult Drosophila melanogaster flies prefer a narrow range of air temperatures around 24–25°C ( Sayeed and Benzer , 1996; Hamada et al . , 2008 ) and rapidly avoid air temperatures of ~31°C ( Ni et al . , 2013 ) . Interestingly , some hematophagous ( blood-feeding ) arthropods have evolved a specialized mode of thermotaxis to locate endothermic ( warm-blooded ) hosts . Such thermophilic behavior is seen in kissing bugs [Triatoma infestans ( Flores and Lazzari , 1996 ) and Rhodnius prolixus ( Schmitz et al . , 2000 ) ] , the bedbug [Cimex lectularius ( Rivnai , 1931 ) ] , the tick [Ixodes ricinus ( Lees , 1948 ) ] , and many species of mosquito ( Clements , 1999 ) including Ae . aegypti , a major tropical disease-vector ( Bhatt et al . , 2013 ) . Female Ae . aegypti require a vertebrate blood meal for the production of eggs , and finding a suitable warm-blooded host is therefore an essential component of reproduction . Mosquitoes use a variety of physical and chemical senses to locate hosts in their environment ( Cardé , 2015 ) . When host-seeking , these animals become strongly attracted to inanimate warm objects , eagerly probing at them as if they were hosts ( Howlett , 1910 ) . In nature , mosquitoes thermotax in a complex thermal landscape in which ambient air temperature , host body temperature , and surrounding surface temperatures can vary widely . For mosquitoes such as Ae . aegypti , host-seeking behavior can be activated by an increase in ambient carbon dioxide ( CO2 ) ( Majeed et al . , 2014 ) . This activation elicits flight activity ( Eiras and Jepson , 1991; McMeniman et al . , 2014 ) and results in an array of behaviors including attraction to visual stimuli ( van Breugel et al . , 2015 ) and host olfactory cues ( Dekker et al . , 2005; McMeniman et al . , 2014 ) , and landing on warm objects ( Burgess , 1959; Eiras and Jepson , 1994; Kröber et al . , 2010; Maekawa et al . , 2011; McMeniman et al . , 2014; van Breugel et al . , 2015 ) . Ae . aegypti flying in a wind tunnel can detect a warmed stimulus from a distance , eliciting attraction and thermotaxis ( van Breugel et al . , 2015 ) . What are the mechanisms by which animals detect thermal stimuli , and how might these be adapted for the specialized needs of heat-seeking female mosquitoes ? Thermotaxis is typically initiated by thermosensitive neurons that sample environmental temperature to inform navigational decision-making . Such neurons must be equipped with molecular thermosensors capable of detecting and transducing thermal stimuli . Diverse molecular thermoreceptors have been identified in the animal kingdom , many of which are members of the transient receptor potential ( TRP ) superfamily of ion channels ( Barbagallo and Garrity , 2015; Palkar et al . , 2015 ) . Different thermosensitive TRPs show distinct tuning spanning the thermal spectrum from noxious cold to noxious heat . Among these is TRPA1 , which is a heat sensor in multiple insects , including the vinegar fly D . melanogaster and the malaria mosquito Anopheles gambiae ( Hamada et al . , 2008; Wang et al . , 2009 ) . Neurons in thermosensitive sensilla ( Gingl et al . , 2005 ) of An . gambiae female antennae express TRPA1 ( Wang et al . , 2009 ) . In D . melanogaster , TRPA1 is expressed in internal thermosensors located in the brain , and DmelTRPA1-/- mutants fail to avoid high air temperature in a thermal gradient ( Hamada et al . , 2008 ) . Interestingly , some snakes and vampire bats express thermosensitive TRP channels in organs used to sense infrared radiation from warm-blooded prey ( Gracheva et al . , 2010; 2011 ) . This raises the possibility that AaegTRPA1 may be used by mosquitoes to find hosts . Recently , a structurally distinct insect thermosensor , Gr28b , was identified in D . melanogaster ( Ni et al . , 2013 ) . Gr28b , a gustatory receptor paralog , is expressed in heat-sensitive neurons of D . melanogaster aristae and is an important component of thermotaxis during rapid avoidance of heat ( Ni et al . , 2013 ) . It is also highly conserved among Drosophila species ( McBride et al . , 2007 ) , and has a clear ortholog in Ae . aegypti , AaegGr19 ( Ni et al . , 2013 ) . A functional role for these thermosensors has never been investigated in the mosquito . Here , we use high-resolution quantitative assays to examine the behavioral strategies underlying mosquito heat-seeking behavior . Our results show that by seeking relative warmth and avoiding both relative cool and high temperatures , female mosquitoes selectively localized to thermal stimuli that approximate warm-blooded hosts . Using genome editing , we generated mutations in the candidate thermoreceptors , AaegTRPA1 and AaegGr19 . We found that TRPA1 is required for tuning mosquito thermotaxis during host-seeking . AaegTRPA1-/- mutants lacked normal avoidance of thermal stimuli exceeding host body temperatures , resulting in a loss of preference for biologically relevant thermal stimuli that resemble hosts . This work is important because it identifies a key mechanism by which mosquitoes tune their thermosensory systems toward human body temperatures .
We previously described an assay to model heat-seeking behavior in the laboratory by monitoring mosquitoes landing on a warmed Peltier element in the context of a cage supplemented with CO2 ( Figure 1A , B ) ( McMeniman et al . , 2014 ) . This assay has the advantages that it is simple in design , produces robust behaviors , and enables the collection of data from large numbers of animals in a short experimental timeframe . Using this system , we can examine mosquito responses to diverse thermal stimuli and measure thermotaxis in different ambient temperature environments . We first needed to determine whether heat-seeking behavior habituates over multiple thermal stimulations . In our heat-seeking assay , Ae . aegypti mosquitoes reliably responded to 12 serial presentations of a 3-minute long 40°C stimulus over the course of more than 2 . 5 hr , with no evidence of habituation ( Figure 1—figure supplement 1 ) . 10 . 7554/eLife . 11750 . 003Figure 1 . Mosquitoes thermotax to stimuli approximating host body temperature . ( A ) Schematic of heat-seeking assay enclosure ( 30 × 30 × 30 cm ) . ( B ) Representative experimental image showing mosquitoes detected on and near the Peltier ( red square ) . ( C ) Typical skin and core temperatures of Ae . aegypti hosts , humans and chickens ( Richards , 1971; Yao , et al . , 2008 . ( D–F ) Heat-seeking behavior measured for a range of stimuli from 26 to 60°C ( n = 6 trials per condition ) . Peltier temperature measured by thermocouple ( D , top trace , mean in red , s . e . m . in gray ) and percent of mosquitoes on Peltier ( D , bottom trace , mean in black , s . e . m . in gray ) . We note that variance in both traces is low , making s . e . m . traces difficult to see . ( E ) Percent mosquitoes on Peltier during seconds 90–180 of each stimulus period in ( D ) . Each replicate is indicated by a dot , and mean by a line . Arrowheads indicate significant differences ( p < 0 . 05 ) from the second presentation of the 40°C stimulus or from 26°C ( repeated measures one-way ANOVA with Bonferroni correction ) . ( F ) Heat maps showing mean mosquito occupancy on the Peltier ( red square ) and surrounding area , during seconds 90–180 of each stimulus period in ( D ) . Bold borders indicate stimuli with responses significantly different from 26°C stimulus ( top row ) or 40°C stimulus ( bottom row ) in ( E ) ( p < 0 . 05; repeated-measures ANOVA with Bonferroni correction ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11750 . 00310 . 7554/eLife . 11750 . 004Figure 1—figure supplement 1 . Mosquitoes consistently thermotax to repeated 40°C stimuli . ( A ) Peltier temperature measured by thermocouple ( top trace , mean in red , s . e . m . in gray ) and percent of mosquitoes on Peltier ( bottom trace , mean in black , s . e . m . in gray ) . n = 10 trials . We note that variance in both traces is low , making s . e . m . traces difficult to see . ( B ) Percent mosquitoes on Peltier during seconds 90–180 of each stimulus period in ( A ) . Each replicate is indicated by a dot , and the mean by a line . There is no significant difference ( p > 0 . 05 ) in Peltier occupancy between the first and last stimulus ( repeated measures one-way ANOVA with Bonferroni correction ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11750 . 004 Ae . aegypti can feed on a variety of hosts ( Clements , 1999; Tandon and Ray , 2000 ) with core body temperatures ranging from ~37°C ( humans ) to ~40–43°C ( chickens ) ( Richards , 1971 ) ( Figure 1C ) . It is unknown whether there are minimal or maximal temperature thresholds constraining mosquito heat-seeking , and whether responses to thermal stimuli depend on the background ambient temperature . To investigate these questions , we measured attraction to thermal stimuli produced by heating the Peltier to temperatures ranging from ambient ( set to 26°C in these experiments ) to 60°C ( Figure 1D , E ) . We found that mosquitoes were highly sensitive to thermal contrast and were attracted to stimuli 2 . 5°C above ambient ( Figure 1D–F ) . Furthermore , mosquito occupancy on the Peltier increased monotonically with stimulus temperatures up to 40°C . However , for higher temperature stimuli , we observed a dramatic reduction in Peltier occupancy . A 50°C stimulus resulted in approximately half as many animals on the Peltier compared to a 40°C stimulus ( Figure 1D–F ) . Stimuli of 55°C or greater resulted in occupancy rates indistinguishable from an ambient thermal stimulus ( 26°C ) ( Figure 1D–F ) . Spatial analysis of mosquito occupancy on or near the Peltier revealed that while mosquitoes were still attracted to high-temperature stimuli , they populated the area peripheral to the Peltier , and strongly avoided the Peltier itself for stimuli ≥ 55°C ( Figure 1F ) . Female mosquitoes searching for a warm-blooded host may be responding to the absolute temperature of a stimulus or may instead be evaluating relative warmth , defined as the differential between a stimulus and background ambient temperature . To investigate the thermotaxis strategies constituting mosquito heat-seeking behavior , we conducted experiments at three ambient temperatures: 21 , 26 , and 31°C ( Figure 2A–F ) . We found that Peltier occupancy for stimuli 21–40°C depended on the differential between the Peltier and ambient temperature ( Figures 2B , C ) , rather than the absolute temperature of the Peltier ( Figure 2A ) . For example , at all ambient temperatures tested , a stimulus 5°C above ambient was sufficient to elicit significant heat-seeking , and elicited approximately half as much Peltier occupancy as a stimulus 10°C above ambient . On the other hand , heat-seeking to targets 50–55°C was inhibited at all ambient temperatures tested ( Figures 2D , F ) , despite the fact that the temperature differential varied widely in these situations ( Figure 2E ) . 10 . 7554/eLife . 11750 . 005Figure 2 . Mosquitoes thermotax to relative warmth and avoid both relative cooling and stimuli exceeding host body temperature . ( A–F ) Heat-seeking at different ambient temperatures ( n = 5–6 trials per condition ) : 21°C ( blue ) , 26°C ( gray ) , 31°C ( orange ) . Data in A , B , D , and E are plotted as mean ± s . e . m . ( A , D ) Percent of mosquitoes on Peltier during seconds 90–180 of stimuli of indicated temperature , normalized to stimulus 10°C above ambient ( A , open circle ) or 40°C stimulus ( D , open circle ) . ( B , E ) Same data as in ( A ) and ( D ) , respectively , plotted using differential between ambient and Peltier temperature . For each ambient temperature , arrowheads indicate the lowest temperature stimulus found to elicit a significant increase in heat-seeking compared to an ambient temperature stimulus ( A , B ) or a reduction in heat-seeking compared to a 40°C stimulus ( D , E ) ( p < 0 . 05; repeated-measures ANOVA with Bonferroni correction ) . For each ambient temperature , linear regressions ( A , B , 21°C: 10 . 6/°C , R2 = 0 . 98 , 26°C: 12/°C , R2 = 0 . 99 , 31°C: 9 . 5/°C , R2 = 0 . 97 ) or variable slope sigmoidal dose–response curves ( D , E , 21°C: IC50 = 55 . 4°C , R2 = 0 . 87 , 26°C: IC50 = 52 . 5°C , R2 = 0 . 92 , 31°C: IC50 = 50 . 5°C , R2 = 0 . 91 ) are plotted . ( C , F ) Heat maps showing mean mosquito occupancy on the Peltier ( red square ) and surrounding area , during seconds 90–180 of each stimulus period . Bold borders indicate stimuli with responses significantly different from an ambient-temperature stimulus ( C ) in ( A , B ) , or significantly different from a 40°C stimulus ( F ) in ( D , E ) ( p < 0 . 05; repeated-measures ANOVA with Bonferroni correction ) . ( G , H ) Analysis of mosquito responses to cooling from data in ( Figure 1D ) . ( G ) Mean percent of mosquitoes on Peltier during thermal stimuli 31–40°C . Dashed line indicates the end of the stimulus period . ( H ) Post-stimulus time at which the percent of mosquitoes on Peltier has decayed to one half of the mean during seconds 90–180 of the stimulus period from ( Figure 1E ) . Each replicate is indicated by a dot , mean ± s . e . m . by lines ( NS , not significant; one-way ANOVA with Bonferroni correction ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11750 . 00510 . 7554/eLife . 11750 . 006Figure 2—figure supplement 1 . Dynamics of Peltier temperature during stimulus periods . Mean Peltier temperature measured by thermocouple during presentation of thermal stimuli ( 31–40°C ) . Dashed line indicates the end of the stimulus period . DOI: http://dx . doi . org/10 . 7554/eLife . 11750 . 006 These results show that Ae . aegypti thermotaxis is driven by seeking relative warmth , but restricted by an absolute upper threshold of ~50–55°C . Because female mosquitoes are attracted to relative warmth , we hypothesized that they may also avoid relative cool . This complementary behavior would serve to improve host-seeking thermotaxis . We examined mosquito responses to cooling by analyzing the rate at which animals left the Peltier when it cooled at the conclusion of a stimulus period . We found that mosquitoes left the Peltier at similar rates regardless of the absolute temperature of the stimulus ( Figure 2G , H , Figure 2—figure supplement 1; based on analysis of data in Figure 1D , E ) , demonstrating that mosquitoes avoid relative cool during heat-seeking . Our characterization of Ae . aegypti heat-seeking revealed multiple behavioral components contributing to selective thermotaxis during host-seeking: 1 ) the seeking of relative warmth; 2 ) the avoidance of relative cool; and 3 ) the avoidance of thermal stimuli exceeding host temperature . Each of these sensory-motor functions may rely on the same molecular thermosensors , or may instead use distinct thermosensors . We considered the possibility that thermoreceptors ordinarily dedicated to the behavioral thermoregulation typical of most ectotherms such as D . melanogaster , may have evolved a function in host-seeking by mosquitoes and other hematophagous arthropods . Using this reasoning , we generated AaegTRPA1-/- mutants using zinc-finger nuclease-mediated genome editing ( Figure 3A , Figure 3—figure supplement 1A ) . 10 . 7554/eLife . 11750 . 007Figure 3 . AaegTRPA1-/- mutants fail to avoid a chemical irritant and high-temperature stimuli . ( A ) Representative bright field ( left ) and fluorescence ( right ) images of wild-type and AaegTRPA1-/- female pupae marked by ubiquitous expression of enhanced cyan fluorescent protein ( ECFP ) . Scale bars: 0 . 5 mm . ( B ) Schematic of capillary feeding ( CAFE ) assay . ( C , D ) Sucrose preference over sucrose containing the indicated concentration of N-methylmaleimide ( C , n = 10–12 trials per condition ) or denatonium benzoate ( D , n = 7 trials per condition ) for mosquitoes of the indicated genotypes ( NS , not significant; *p < 0 . 05 , **p < 0 . 01; one-way ANOVA with Bonferroni correction compared to wild-type ) . ( E ) Percent of mosquitoes of indicated genotypes on Peltier during seconds 90–180 of stimuli of indicated temperature ( mean ± s . e . m . , n = 6–9 trials per genotype; ***p < 0 . 001; repeated measures one-way ANOVA with Bonferroni correction ) . ( F ) Heat maps showing mean mosquito occupancy for the indicated genotypes on the Peltier ( red square ) and surrounding area , during seconds 90–180 of each stimulus period . Bold borders indicate stimuli with responses significantly different from wild-type in ( E ) ( p < 0 . 05; repeated-measures ANOVA with Bonferroni correction ) . ( G ) Mean percent of mosquitoes of indicated genotypes on Peltier during seconds 0–180 of thermal stimuli 40–60°C and during subsequent re-presentation of 40°C . Timespans with statistically significant increases in AaegTRPA1-/- mutant Peltier occupancy compared to wild-type are indicated by purple lines ( calculated from 15 second bins; p < 0 . 05; one-way ANOVA with Bonferroni correction ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11750 . 00710 . 7554/eLife . 11750 . 008Figure 3—figure supplement 1 . AaegTRPA1-/- mutants fail to avoid noxious heat in a thermal gradient . ( A ) Genomic organization and targeted mutagenesis of AaegTRPA1 . Non-coding ( white ) and coding ( black ) exons are shown to scale ( vectorbase . org ) . Introns are denoted by connecting lines ( not to scale ) . The AaegTRPA1 ZFN targets exon 12 ( purple arrow ) . An insertion cassette ( shown to scale ) , including the Ae . aegypti polyubiquitin ( polyUB ) promoter driving expression of enhanced cyan fluorescent protein ( ECFP ) and an SV40 polyadenylation signal , is integrated via homology-directed repair . ( B ) Schematic of the thermal gradient assay ( top , side view ) . Representative experimental image ( bottom , top view ) showing mosquitoes outlined in red detected across one lane of the thermal gradient assay , with instantaneous air temperature reported for each of 8 analysis sectors . ( C , D ) Percent of mosquitoes of indicated genotypes detected in each sector and mean sector air temperature ( rounded to nearest °C ) in the absence ( C ) or presence ( D ) of a thermal gradient ( n = 6 trials per genotype ) . Data are plotted as mean ± s . e . m . ( ***p < 0 . 001; two-way ANOVA with Bonferroni correction compared to wild-type ) . ( E ) Heat maps showing mean air temperature ( left ) and percent of mosquitoes of the indicated genotypes detected ( right ) in each sector over 90 min from the onset of a thermal gradient . ( F ) Same data as in ( E ) , showing total percent of mosquitoes of the indicated genotypes detected in sectors 1–3 ( cold side , solid line , mean; s . e . m . , shading ) and sectors 6–8 ( hot side , dotted line , mean; s . e . m . , shading ) during onset and maintenance of a thermal gradient . G , Percent of total mosquitoes of the indicated genotypes found dead in sectors 6–8 at the conclusion of the experiment . Each replicate is indicated by a dot , and mean ± s . e . m . by lines . Genotypes with different letters are significantly different ( p < 0 . 01 , one-way ANOVA with Bonferroni correction ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11750 . 00810 . 7554/eLife . 11750 . 009Figure 3—figure supplement 2 . AaegGr19-/- mutants show normal thermotaxis . ( A ) Genomic organization and targeted mutagenesis of AaegGr19 . Coding ( black ) exons are shown to scale , introns are denoted by connecting lines ( not to scale ) , and alternative splicing is denoted by dashed lines ( vectorbase . org ) . The AaegGr19 ZFN targets exon 3 ( green arrow ) . An insertion cassette ( shown to scale ) , including the Ae . aegypti polyubiquitin ( polyUB ) promoter driving expression of Discosoma sp . red fluorescent protein ( DsRed ) and an SV40 polyadenylation signal , is integrated via homology-directed repair . ( B ) Representative bright field ( left ) and fluorescence ( right ) images of wild-type and AaegGr19-/- female pupae marked with ubiquitous expression of DsRed Scale bars: 0 . 5 mm . The wild-type bright-field image is duplicated from ( Figure 3A ) . ( C ) Percent of mosquitoes of indicated genotypes on Peltier during seconds 90–180 of stimuli of indicated temperature ( mean ± s . e . m . , n = 6–9 trials per genotype ) . Neither AaegGr19-/- nor AaegGr19+/- were significantly different from wild-type at any stimulus temperature ( repeated measures one-way ANOVA with Bonferroni correction ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11750 . 009 In addition to its function as a thermoreceptor , TRPA1 is a highly conserved chemosensor of electrophile irritants such as N-methylmaleimide ( Macpherson et al . , 2007; Kang et al . , 2010 ) . Using a modified capillary feeding ( CAFE ) assay ( Ja et al . , 2007 ) ( Figure 3B ) , we found that wild-type Ae . aegypti mosquitoes strongly avoided consumption of N-methylmaleimide ( Figure 3C ) , as well as the bitter compound denatonium benzoate ( Figure 3D ) . AaegTRPA1-/- mutants rejected denatonium benzoate ( Figure 3D ) but did not avoid consumption of N-methylmaleimide ( Figure 3C ) . We interpret this result as a loss of N-methylmaleimide detection in AaegTRPA1-/- mutants , leading to no preference between sucrose and sucrose containing N-methylmaleimide . We note that this simple CAFE assay could be used to discover additional mosquito anti-feedants to repel Ae . aegypti , beyond the two chemicals identified here . Because TRPA1 is important in insect thermoregulation ( Hamada et al . , 2008 ) , we used a modified thermal gradient assay ( Sayeed and Benzer , 1996; Hamada et al . , 2008 ) to assess thermal preference in wild-type and AaegTRPA1-/- mutant mosquitoes ( Figure 3—figure supplement 1B ) . AaegTRPA1-/- mutants were impaired in avoidance of high air temperature , leading to significant mortality ( Figure 3—figure supplement 1C–G ) . Together , these data indicate that TRPA1 has a conserved chemosensory and thermosensory function in Ae . aegypti . We next asked if AaegTRPA1 is required for mosquito heat-seeking behavior . AaegTRPA1-/- mutants showed normal attraction to stimuli at or below 45°C , but strikingly lacked normal avoidance of higher temperature stimuli ( 50°C and 55°C ) ( Figures 3E , F ) . A detailed analysis of Peltier occupancy over time revealed that AaegTRPA1-/- mutants persisted on the Peltier during 50 , 55 , and 60°C stimulus presentations , whereas control animals rapidly left these high-temperature stimuli ( Figure 3G ) . Therefore , AaegTRPA1 is not required for initial attraction to warmth but is required for normal avoidance of high-temperature stimuli that exceed host body temperature . Because AaegTRPA1-/- mutants retained attraction to warm stimuli , we used targeted mutagenesis to test a requirement for AaegGr19 , the Ae . aegypti ortholog of Gr28b , in heat-seeking ( Figure 3—figure supplement 2A , B ) . Although this thermosensor is required for mediating rapid avoidance of warmth in D . melanogaster ( Ni et al . , 2013 ) , AaegGr19-/- mutant mosquitoes showed no thermotaxis defects ( Figure 3—figure supplement 2C ) . The different phenotypes of these mutations in Ae . aegypti and D . melanogaster may reflect differences in expression patterns of these genes . DmelTRPA1 is expressed in internal thermosensors of the brain ( Hamada et al . , 2008 ) , while thermosensitive isoforms of DmelGr28b are expressed in peripheral heat-sensors ( Ni et al . , 2013 ) . In Ae . aegypti , TRPA1 RNA is expressed in numerous tissues including the antennae ( Matthews et al . , 2015 ) , where it may be associated with peripheral thermosensitive sensilla responding to rapid thermal fluctuations ( Gingl et al . , 2005 ) , as it is in An . gambiae ( Wang et al . , 2009 ) . The cellular expression pattern of Gr19 in Ae . aegypti is not known , but its transcript is broadly expressed ( Matthews et al . , 2015 ) . Although AaegTRPA1-/- mutants did not show normal avoidance of high-temperature stimuli , they may still prefer host-temperature stimuli if presented with a choice . In a heat-seeking choice assay with two independently controlled Peltiers , we examined the importance of AaegTRPA1 in guiding mosquito thermotaxis in a more complex thermal landscape ( Figure 4A ) . In this assay , mosquitoes were simultaneously presented with two thermal stimuli . When presented with two 40°C stimuli , both wild-type and mutant mosquitoes distributed equally between the Peltiers ( Figure 4B ) , but in a choice between a 40°C and 50°C stimulus , wild-type mosquitoes strongly preferred the 40°C Peltier ( Figure 4C ) and avoided the 50°C Peltier ( Figure 4D ) . Remarkably , in this choice scenario , AaegTRPA1-/- mutants failed to avoid the 50°C Peltier , resulting in no preference for the 40°C stimulus ( Figure 4 C , D , Video 1 ) . 10 . 7554/eLife . 11750 . 010Figure 4 . AaegTRPA1-/- mutants fail to discriminate between host-temperature and higher-temperature targets . ( A ) Schematic of heat-seeking choice assay . ( B , C ) Preference for 40°C versus 40°C ( B ) or 50°C versus 40°C ( C ) Peltiers for indicated genotypes ( n = 6 trials per genotype; mean ± s . e . m . , with each replicate indicated by a dot; NS , not significant; ***p < 0 . 001; one sample t-test versus zero preference ) . In ( C ) AaegTRPA1-/- mutants are significantly different from wild-type and heterozygous mutants ( p < 0 . 05 , one-way ANOVA with Bonferroni correction ) . ( D ) Heat maps showing mean mosquito occupancy for the indicated genotypes on Peltiers ( red squares ) of the indicated temperatures and surrounding area , during seconds 60–240 of each stimulus period . ( E ) Model of mosquito thermotaxis . ( F ) Thermal image of a person ( arrow ) standing on a sunlit patch of grass in Central Park in New York City . DOI: http://dx . doi . org/10 . 7554/eLife . 11750 . 01010 . 7554/eLife . 11750 . 011Video 1 . AaegTRPA1 is required for tuning avoidance of high-temperature stimuli during heat-seeking . AaegTRPA1-/- mutants presented with a choice between two Peltiers , one at 40°C and one at 50°C . Video is sped up 10-fold ( images acquired at 1 Hz and reproduced at 10 frames/s ) , and shows seconds 60–240 of the stimulus period . DOI: http://dx . doi . org/10 . 7554/eLife . 11750 . 011
We have elucidated the basic thermotaxis strategies used by mosquitoes , and revealed an important role for TRPA1 in regulating this behavior ( Figure 4E ) . Using a quantitative thermotaxis assay , we modelled Ae . aegypti heat-seeking behavior in the laboratory . We found that mosquitoes can search for hosts in a wide range of ambient temperatures by seeking relative warmth and avoiding relative cool . Remarkably , these animals can detect a stimulus with thermal contrast as small as 2 . 5°C . In an outdoor environment , however , hosts are often warmer than the surrounding air but cooler than sun-warmed soil , rocks , trees , and human-made objects ( Figure 4F ) . For this reason , diurnal mosquitoes such as Ae . aegypti are poorly served by merely thermotaxing to the hottest object available . A more optimal strategy is to search specifically for biologically relevant stimuli , and to avoid thermal stimuli exceeding host temperature , as we have observed in our laboratory models of heat-seeking . Acquiring a blood meal is an essential component of reproduction for a female Ae . aegypti mosquito . To maximize her chances of success , a female mosquito should reject ‘distracting’ stimuli that exceed host temperatures . Our results demonstrate that AaegTRPA1 is critical for this selective thermotaxis . Mosquito heat-seeking behavior represents an excellent model system for further study of the genetics ( Kang et al . , 2012; Zhong et al . , 2012 ) , neuroscience ( Frank et al . , 2015; Liu et al . , 2015 ) , and decision-making ( Luo et al . , 2010 ) underlying thermosensation and thermotaxis . Until now , mechanistic studies of thermosensation have been largely restricted to traditional laboratory model organisms , such as domestic mice and Drosophila melanogaster flies , whose thermotaxis consists mainly of moving away from suboptimal thermal environments . Mosquitoes too , must undergo such behavioral thermoregulation , as we have found in our thermal gradient assay . However , their repertoire of thermotactic behaviors is expanded by the evolution of a specialized and highly tuned mode of thermotaxis to locate warm-blooded hosts . It may also be that these two thermotactic drives—host-seeking and thermoregulation—interact in mosquitoes . For instance , the avoidance of high-temperature stimuli during heat-seeking may be influenced by a nociceptive or thermoregulatory response independent of host-seeking behavior . It will be interesting to investigate the neural mechanisms that regulate divergent behavioral choices of thermoregulation and heat-seeking , and whether these systems are in behavioral conflict during mosquito host-seeking . Our work identifies TRPA1 as a gene regulating mosquito avoidance of high-temperature stimuli , which we have shown to be a major behavioral component of heat-seeking . However , because both AaegTRPA1-/- and AaegGr19-/- mutants retain normal attraction to warmth , this aspect of heat-seeking must rely on other thermoreceptors , still to be identified . Our study shows that this attraction must be mediated either by a single thermosensor that adapts to background temperature , or multiple thermosensors each tuned to a distinct absolute threshold . During interactions with a warm-blooded host or a 50°C Peltier , mosquitoes are likely experiencing a wide range of thermal fluctuations . For instance , there can be a 10°C thermal gradient in the air within 5 mm of a 37°C stimulus or human arm ( van Breugel et al . , 2015 ) . This suggests that for a mosquito standing on a warmed Peltier , the antennae , brain , thorax , forelegs , and proboscis may all be experiencing different temperatures . The temperature of mosquito tissues will also be greatly influenced by their material and geometric properties , as well as thermal conduction due to contact with the Peltier . Furthermore , convective plumes forming in the air near a vertical heated plate can be highly dynamic and turbulent in their structure , with thermal air gradients differing considerably at the bottom and top of the plate ( Bejan , 2013 ) . These features of thermal stimuli may explain why mosquitoes in our assay often appear to be differentially attracted to the bottom and top of the Peltier . Future studies characterizing this complex thermal microenvironment , and identifying the relevant thermosensory neurons and receptors will be required to define the thermal fluctuations experienced by mosquitoes during heat-seeking . Understanding the behavioral and molecular basis of thermotaxis in mosquitoes and other disease vectors ( Flores and Lazzari , 1996; Schmitz et al . , 2000 ) is of great biomedical importance . Ae . aegypti mosquitoes are potent vectors of yellow fever , chikungunya , and dengue arboviruses , resulting annually in hundreds of millions of infections ( Bhatt et al . , 2013 ) . Further study of mosquito heat-seeking behavior may aid in the design of next-generation traps , repellents , and control strategies .
Ae . aegypti wild-type ( Orlando ) , AaegGr19 , and AaegTRPA1 mutant strains were maintained and reared at 25–28°C , 70–80% relative humidity with a photoperiod of 14 hr light:10 hr dark ( lights on at 8 a . m . ) as previously described ( DeGennaro et al . , 2013 ) . Adult mosquitoes were provided constant access to 10% sucrose solution for feeding , and females were provided with a blood source for egg production , either live mice or human volunteers . Blood-feeding procedures were approved and monitored by The Rockefeller University Institutional Animal Care and Use Committee and Institutional Review Board , protocols 14756 and LV-0652 respectively . Human volunteers gave their informed written consent to participate in mosquito blood-feeding procedures . Before behavioral assays , mosquitoes were sexed and sorted under cold anesthesia ( 4°C ) and fasted for 15–24 hr in the presence of a water source . All assays were carried out between ZT2-ZT12 at 26°C and 70–80% relative humidity unless stated otherwise . Whenever possible , time of day was randomized across conditions . All mosquitoes used were 10–21 day-old females , age-matched across conditions and genotypes . All thermal images were acquired with an infrared camera ( E60 , FLIR Systems , Wilsonville , OR ) . All statistical analyses were performed using Prism 5 software ( GraphPad Software , Inc . , La Jolla , CA ) . | Temperature can vary considerably in an environment . Living organisms have evolved sensory systems to detect and avoid excessive heat or cold: a behavior that is termed ‘thermotaxis’ . In rare cases , animals use this ability to locate food sources in their environment . One example of such an adaptation is the female mosquito of the species Aedes aegypti . When a mosquito needs blood to produce her eggs , she becomes attracted to the body heat of warm-blooded hosts . But the range of temperatures that these mosquitoes prefer and the genes required for this behavior had not been been defined . Now , Corfas and Vosshall have found that female Aedes aegypti are highly sensitive to differences in temperature , and are capable of heat-seeking in a range of environmental temperatures . Furthermore , by seeking out things that are warmer than their surroundings , while avoiding those that are cooler or much hotter than their host’s body temperatures , these mosquitoes tune their thermotaxis toward targets that resemble a human to feed upon . Corfas and Vosshall also discovered that a protein called TRPA1 is required for this tuning of Aedes aegypti’s heat-seeking behavior . This protein is known to allow insects to detect chemical signals and regulate their own temperature , but it was not previously known that this protein was involved in mosquito thermotaxis . Mutant mosquitoes without the gene for TRPA1 failed to avoid high temperatures , which meant that they could no longer tell the difference between an overly hot target and a warm one that resembled their hosts . Following on from this work , the next challenge will be to characterize all the genes , sensory organs , and neural circuits that drive mosquito heat-seeking behavior . These findings may in the future inform the design of the next generation of repellents and traps for the control of mosquito-borne diseases , such as dengue and yellow fever . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials"
] | [
"neuroscience"
] | 2015 | The cation channel TRPA1 tunes mosquito thermotaxis to host temperatures |
In the Drosophila circadian clock , Period ( PER ) and Timeless ( TIM ) proteins inhibit Clock-mediated transcription of per and tim genes until PER is degraded by Doubletime/CK1 ( DBT ) -mediated phosphorylation , establishing a negative feedback loop . Multiple regulatory delays within this feedback loop ensure ~24 hr periodicity . Of these delays , the mechanisms that regulate delayed PER degradation ( and Clock reactivation ) remain unclear . Here we show that phosphorylation of certain DBT target sites within a central region of PER affect PER inhibition of Clock and the stability of the PER/TIM complex . Our results indicate that phosphorylation of PER residue S589 stabilizes and activates PER inhibitory function in the presence of TIM , but promotes PER degradation in its absence . The role of DBT in regulating PER activity , stabilization and degradation ensures that these events are chronologically and biochemically linked , and contributes to the timing of an essential delay that influences the period of the circadian clock .
The proteins that comprise circadian clocks are largely conserved across the animal kingdom ( Crane and Young , 2014; Young and Kay , 2001; Zheng and Sehgal , 2012 ) . These clocks , which promote daily rhythms in behavior and physiology , involve cell-autonomous , transcription-translation feedback loops that oscillate with ~24 hr periodicity . In Drosophila , the activator complex consists of dClock/Cycle ( dCLK/CYC ) , which initiates transcription of period ( per ) and timeless ( tim ) . PER and TIM proteins assemble to form an inhibitory complex in the cytoplasm that , after a delay , enters the nucleus to inhibit dCLK/CYC activity . After a second delay , the inhibitory complex is degraded , freeing the activator complex to reinitiate transcription , closing the feedback loop . Although the delay in transcriptional activation is critical for maintaining ~24 hr rhythmicity , its regulatory mechanisms are not well understood . A Casein Kinase I/Doubletime ( DBT ) phosphorylation cascade on PER regulates PER protein stability . Current models of this phosphorylation cascade infer that each DBT phosphorylation event gates later phosphorylation events on PER , thus delaying protein degradation and transcriptional activation . The cascade begins with the PER Short Downstream region ( PER-SD , amino acids 604–629 ) , which promotes S47 phosphorylation ( Garbe et al . , 2013 ) . Phosphorylation of S47 , is a necessary modification to trigger ubiquitination and subsequent degradation of PER by the E3-ligase SLIMB ( Chiu et al . , 2008; Garbe et al . , 2013; Kivimäe et al . , 2008 ) . PER-SD also blocks NEMO kinase from phosphorylating S596 . By phosphorylating S596 , NEMO promotes S589 phosphorylation by DBT , which blocks S47 phosphorylation ( Chiu et al . , 2011; Garbe et al . , 2013 ) . In other words , PER-SD phosphorylation removes the S589-mediated negative regulation of S47 phosphorylation while independently promoting S47 phosphorylation ( Garbe et al . , 2013 ) . The purpose of the second regulatory route involving negative regulation of degradation by S589 is unclear . Mutation of the S589 site in flies leads to a short behavioral period ( Baylies et al . , 1992; Konopka and Benzer , 1971; Rutila et al . , 1992 ) . Given that S589 is a DBT phosphorylation site ( Chiu et al . , 2008; Kivimäe et al . , 2008 ) , it is surprising that mutations that both block ( Gly ) and substitute for phosphorylation ( Asp ) lead to a short behavioral rhythm ( Rutila et al . , 1992 ) . Although this may suggest that the aspartate residue substitution does not serve as an effective phosphomimetic , the fact that the behavioral period of the mimetic is longer than a mutant that prevents phosphorylation ( Gly ) suggests a more complex regulatory mechanism . TIM binding to PER adds another layer of complexity to the model by protecting PER from hyperphosphorylation and degradation ( Kloss et al . , 2001 ) . This suggests that the DBT phosphorylation cascade on PER is influenced by TIM . The sites that comprise the cascade were identified using either purified protein ( Kivimäe et al . , 2008 ) or PER protein expressed alone , in cultured Drosophila embryonic ( S2 ) cells ( Chiu et al . , 2008 ) , concealing the identity of the sites that would have been protected by TIM , or overlooking sites that would have been phosphorylated in the presence of TIM . Thus , how the DBT phosphorylation cascade operates in the context of a complex is largely undetermined . It is generally assumed that inhibition of dCLK is gated by nuclear accumulation of PER , and release of dCLK inhibition occurs passively , through PER degradation . PER phosphorylation , PER degradation and PER inhibitory activity coincide closely in time and space ( Abruzzi et al . , 2011; Edery et al . , 1994 ) , which has led to a model in which PER stability and inhibitory activity are regulated by the same regulatory mechanism ( Kivimäe et al . , 2008 ) . An alternative possibility is that the different functions of PER ( transcription inhibition and degradation ) , are independently regulated by different programs of phosphorylation that occur concurrently or nearly so . This possibility has been difficult to test , due to the challenges of monitoring PER phosphorylation , transcriptional activity and protein stability simultaneously . In this study , we propose a new model for the regulation of PER by DBT , which includes separable features affecting PER stability and activity . Our findings suggest that the DBT/PER/TIM complex is interdependent in the initial stages of dCLK/CYC inhibition . We show that the initial sites of DBT phosphorylation stabilize the PER/TIM inhibitor complex and activate transcriptional inhibition . The classic per short site S589 serves two functions: first to stabilize the activated PER inhibitor in the presence of TIM and then to mediate its degradation after TIM is degraded . We propose a dual role for DBT in regulating the circadian clock . ( 1 ) In the nucleus DBT activates function of the PER/TIM complex as a transcriptional inhibitor and stabilizes it , ( 2 ) subsequently promoting PER degradation by midday . The dual function of DBT links PER/TIM inhibitory activity and stability , biochemically and chronologically , and underlies a key delay in the circadian clock that helps establish 24 hr behavioral rhythmicity .
PER S589 is involved in a phosphorylation cascade that regulates PER ubiquitination and degradation ( Chiu et al . , 2008; 2011; Garbe et al . , 2013; Kivimäe et al . , 2008 ) . Paradoxically , residue substitutions of S589 that either block or mimic phosphorylation both yield the same phenotype: short-period rhythmic behavior in transgenic flies ( Baylies et al . , 1992; Rutila et al . , 1992 ) . To confirm this , we generated transgenic flies in which we rescued the per0 mutation with per variants containing substitutions of the DBT-targeted perS site S589 , using targeted genomic integration to avoid positional effects on protein expression . As previously reported ( Baylies et al . , 1992; Rutila et al . , 1992 ) , a phosphonull alanine substitution ( S589A ) caused an advance in evening anticipation but not morning anticipation relative to controls in a 12:12 light/dark ( LD ) light regiment ( Figure 1A ) . In constant darkness ( DD ) , S589A mutants exhibited a ~ 6 hr shorter period in activity rhythm relative to controls ( Figure 1B , Supplementary file 1; ~19 . 5 hr S589A vs ~25 . 5 hr wild type ) . To test the effects of constitutive phosphorylation at residue S589 on period length , we generated a PER variant with the phosphomimetic aspartate substitution at site S589 ( S589D ) . Similar to phosphonull S589A mutants , we found that phosphomimetic S589D mutants exhibit a modest advance in evening anticipation and no change in morning anticipation relative to controls ( Figure 1A ) . Consistent with this result , the activity period of S589D in DD was also shortened , leading to a ~ 2 hr shorter subjective day ( Figure 1B , Supplementary file 1; 23 . 5 h S589D vs ~25 . 5 hr wild type ) . Thus , we confirmed that mutations that mimic or block phosphorylation both have the effect of shortening behavioral period by advancing evening anticipation . Paradoxically , these results suggest that both phosphorylation and loss of phosphorylation at S589 normally contribute to the post-transcriptional delays that regulate circadian period length . It was not clear whether these effects on period length by this residue is solely due to phosphorylation by DBT kinase or whether there may be a second kinase involved . We can distinguish between these two possibilities by testing genetic interactions between specific phosphomutations and kinase overexpression or kinase mutant expression . For example , mutation of specific target residues of the TIM protein blocks the effect of CK2 kinase overexpression on rhythmic behavior ( Top et al . , 2016 ) . DBT overexpression in transgenic flies expressing wild type PER caused a modest increase in period length of ~1 hr , as expected ( Muskus et al . , 2007; Venkatesan et al . , 2015 ) . We found that both S589 mutations blocked this effect of DBT overexpression on period length ( Figure 1C ) . Since exogenous DBT expression has a phenotypic effect on period length in wild type flies , endogenous DBT levels are rate limiting for mechanisms affecting behavior . The ability of S589 to block the effect of DBT overexpression on behavior suggests that its modification occurs before other functions of DBT that regulate behavior . The ability of S589 to curb the effect of DBT overexpression also provides evidence that S589 and DBT genetically interact . The activity of O-GlcNAc transferase ( OGT ) has been shown to influence behavioral rhythmicity in mammals and flies , either by competing with CKIδ for modification of mPER2 ( orthologues of DBT and PER , respectively ) or blocking BMAL1/CLOCK ( orthologues of CYC/dCLK in flies ) ubiquitination in mice , or prolonging PER stability in flies ( Kaasik et al . , 2013; Kim et al . , 2012; Li et al . , 2013 ) . Since we were able to successfully block the phenotypic effect of DBT overexpression in the S589A and S589D fly mutant backgrounds , we decided to test whether the S589 residue genetically interacts with OGT . We found that OGT overexpression in wild type flies lengthened rhythmic behavior of wild type flies , as expected ( Kaasik et al . , 2013; Kim et al . , 2012 ) . The same experiment conducted in the S589A and S589D mutant backgrounds similarly lengthened rhythmic behavior of the mutant flies ( Figure 1—figure supplement 1 ) . We therefore conclude that S589 does not genetically interact with OGT . Mutant forms of dbt that reduce DBT kinase activity produce long ( dbtL ) and short ( dbtS ) behavioral rhythms ( Kivimäe et al . , 2008; Preuss et al . , 2004; Price et al . , 1998 ) , suggesting that DBT may play two roles in regulating PER . Indeed , in cultured cells dbtL is defective in regulating PER stability , whereas dbtS is not associated with regulating PER stability ( Syed et al . , 2011 ) . To test whether S589 mutants genetically interact with DBT function affecting PER stability , we assayed the effects of combining S589 mutations with either dbtL and dbtS . We found that expression of either phosphonull ( S589A ) or phosphomimetic ( S589D ) per variants in a dbtL genetic background leads to longer behavioral rhythms ( Figure 1D ) . In contrast , expression of wild type per in a dbtS genetic background leads to a shorter behavioral rhythm of wt flies , but not of the S589 mutants . Thus , mutations of S589 block the period changes caused by dbtS but not dbtL , suggesting that the S589 mutations and dbtS lie within the same regulatory pathway . We set out to identify the mechanism by which PER phosphomutants interact with dbtS to delay the circadian clock . Delays in the circadian clock are regulated by post-translational modifications that govern PER nuclear entry , PER inhibitory activity , and PER/TIM degradation . dbtS , which has reduced kinase activity , delays PER nuclear accumulation in vivo ( Bao et al . , 2001; Kivimäe et al . , 2008; Rothenfluh et al . , 2000 ) . Although the original PERS mutant protein ( S589N ) exhibits kinetics of nuclear entry similar to wild type ( Curtin et al . , 1995 ) , because S589 phosphomutants and dbtS genetically interact , we first tested whether S589 phosphomutations alter the kinetics of PER/TIM nuclear accumulation . We quantified PER protein expression using immunofluorescence microscopy analysis across the circadian cycle , with focus on the master pacemaker neurons ( the ventral lateral neurons; LNvs ) that drive behavioral rhythms in constant darkness ( Stoleru et al . , 2005 ) ( Figure 2A ) . Similar to the original PERS mutant , we found that S589 mutants did not alter the kinetics of nuclear accumulation of PER or TIM in the small LNvs ( Figure 2B ) or the large LNvs ( Figure 2—figure supplement 1 ) relative to controls . Thus , residue substitutions of S589 do not shorten period length by changing the rate of nuclear accumulation of the PER/TIM inhibitory complex . This result further suggests that S589 phosphorylation regulates PER function after nuclear entry . We next set out to determine the role of S589 in PER and TIM protein stability after nuclear entry . Using in vivo fluorescence microscopy , we examined PER/TIM protein stability in the master pacemaker small LNvs , from ZT20 through CT08 , when the complex is predominantly nuclear in wild type flies ( Figure 2C ) . The phosphonull S589A substitution does not affect PER stability between ZT20 and CT02 , but destabilizes PER after CT02 , leading to a gradual loss of PER protein relative to controls ( Figure 2D ) . TIM is similarly destabilized in the S589A genetic background compared to wild type ( Figure 2E ) . In contrast , the phosphomimetic S589D mutant behaves like wild-type PER until TIM is degraded at CT06 , at which time the PER S589D protein is rapidly degraded compared to wild type ( Figure 2D and E ) . The changes in the stability of the PER variants correlate well with the differences in rhythmic behavior exhibited by the relevant fly mutants ( Figure 1B ) . These data suggest that S589 phosphorylation is required to stabilize the PER/TIM inhibitory complex after CT02 in the small LNvs . However , upon TIM degradation , the behavior of S589D indicates that phosphorylation at S589 now promotes PER degradation , ending the transcriptional inhibitory phase of the circadian clock . That is , phosphorylation at S589 both stabilizes and destabilizes PER at different points in the circadian cycle , depending on the presence or absence of TIM protein . Thus , S589 phosphorylation has a stabilizing effect when TIM is present . DBT activity regulates PER stability by targeting S589 and other sites on PER , directing DBT to block or promote S47 phosphorylation mediated PER degradation ( Chiu et al . , 2008; Garbe et al . , 2013; Kivimäe et al . , 2008 ) . We wanted to determine if mutating S589 affects the stability of the PER/TIM complex coupled to DBT activity . PER/TIM complex stability can be directly investigated using S2 cells , which offer a minimal system uncomplicated by an endogenous clock that could alter protein expression through transcriptional regulation . Analysis of the S589N variant of PER ( PERS ) in a similar study that included TIM ( but not DBT ) in cultured cells suggests that it is equivalent in stability to wild type PER ( Li and Rosbash , 2013 ) . However the side chain of asparagine has a dipole that has the potential to act as a phosphomimetic while simultaneously blocking phosphorylation by DBT . Therefore to further characterize the effect of the S589 substitutions on the stability of the nuclear PER/TIM complex , we performed pulse-chase co-expression experiments in which per or per variants were expressed by a heat-shock promoter and tim and dbt were co-expressed using a constitutively active actin promoter . In this experimental design , PER and associated proteins are predominantly nuclear because driving tim expression by an actin promoter bypasses the delay in nuclear accumulation ( Figure 3—figure supplement 1A ) observed when both tim and per are driven by heat-shock promoter ( Meyer et al . , 2006; Saez et al . , 2011; Top et al . , 2016 ) . S2 cells were lysed at 1 hr time resolution and analyzed by quantitative western blot ( Figure 3 ) . The analysis was plotted as ratios to normalize against different factors that affect protein stability . Because pulse-chase experiments involving cycloheximide display gradual loss of all proteins , as a control we normalized PER and TIM protein decay to DBT , whose stability is not under circadian clock regulation ( Figure 3—figure supplement 1B–C ) . By using DBT as a proxy for background degradation , we control for the effect of cycloheximide on the kinetics of general protein degradation . The resulting plots reveal a stable S589D variant compared to wild type , and an unstable S589A variant in the presence of wild type DBT ( Figure 3—figure supplement 1C ) . However , PER interactions with a DBT variant lacking kinase activity ( DBT-K38R ) stabilizes PER ( Muskus et al . , 2007 ) , suggesting that competitive binding of PER with DBT-K38R versus endogenous DBT should have a stabilizing effect on PER . To isolate DBT activity as the regulating factor in PER stability , we normalized PER and TIM levels co-expressed with DBT to PER and TIM levels co-expressed with DBT-K38R ( Figure 3 ) . Thus , our approach normalizes PER and TIM degradation to both background kinetics of protein degradation , accounts for loss of DBT and for the stabilizing effect of DBT-PER protein interactions , effectively isolating the phosphorylation effect of DBT on PER ( and TIM ) stability . Using this assay , we found that the phosphonull S589A mutation destabilizes PER relative to wild-type PER in the presence of DBT ( Figure 3A ) . This result suggests that phosphorylation of S589 on PER normally protects PER from DBT-mediated degradation in a complex with TIM . This result also suggests that S589 phosphorylation is not required to complete the PER phosphorylation cascade to promote PER degradation . Consistent with this finding , PER protein containing the phosphomimetic S589D mutation exhibits stability similar to wild type in the presence of DBT activity ( Figure 3A ) . The S589D result also suggests that DBT activity inhibits PER degradation at this stage . TIM stability in this assay follows the same pattern as the PER variants; destabilized by the PER phosphonull variant ( S589A ) but not the PER phosphomimetic ( S589D ) , compared to controls ( Figure 3B ) . This assay repeated in the absence of TIM reveals rapid PER degradation that is not quantifiable over time ( Figure 3—figure supplement 1D ) . These data therefore suggest that S589 phosphorylation stabilizes PER in the presence of TIM from DBT-mediated degradation , either by blocking phosphorylation of other PER destabilizing sites , or in spite of phosphorylation of these other sites . In contrast , blocking S589 phosphorylation redirects DBT to promote the degradation of the PER/TIM complex , likely by phosphorylating other PER sites . Therefore S589 acts to direct DBT to degrade PER or to stabilize PER within the PER/TIM complex . DBT-mediated degradation of PER occurs in both the nucleus and cytoplasm ( Price et al . , 1998; Venkatesan et al . , 2015 ) . The experimental system above investigates PER stability mainly in the nucleus . To determine if S589 phosphorylation affects PER stability in the cytoplasm , we repeated the pulse-chase experiments in conditions that restricted PER to the cytoplasm using a TIM mutant lacking its nuclear localization signal ( TIM-ΔNLS ) . This TIM variant blocks nuclear accumulation of both TIM and PER in S2 cells in the time scales used in our assay ( Saez et al . , 2011 ) . When PER and TIM are restricted to the cytoplasm , wild type PER and PER S589A are similar in their stability ( Figure 3C ) . Thus , lack of phosphorylation at S589 destabilizes PER specifically in the nucleus . As a side note , S589D is more stable than wild type PER in the cytoplasm . These results , coupled with our in vivo data , suggest that phosphorylation of S589 is required to protect PER protein from DBT-mediated degradation in the nucleus . Our data further suggest that phosphorylation of S589 is required for stabilization of PER in the presence of TIM . The role of S589 in protein stability is somewhat unexpected considering our genetic data showing an interaction between S589 mutants with the dbtS but not dbtL mutation ( Figure 1D ) . While the dbtL mutant is defective in regulating PER stability , dbtS is not ( Kloss et al . , 1998; Price et al . , 1998; Syed et al . , 2011 ) . We hypothesized that this apparent paradox could be resolved if the same residue ( S589 ) was involved with a second , unknown function of DBT in regulating PER , in addition to its function in regulating PER stability; DBT may also affect PER inhibitory activity . To test whether DBT sites on PER affect transcriptional inhibitory activity , we used an in vivo luciferase reporter assay for PER activity ( Stanewsky et al . , 1997 ) . Flies expressing the luciferase reporter driven by the per minimal promoter and the indicated per transgene in a per0 genetic background were monitored for bioluminescence ( Figure 4A ) . Both S589A and S589D mutants have shorter total periods than control flies expressing the wild type per transgene . Thus , the period of bioluminescence oscillation correlates well with behavioral period ( Figures 4B and 1C ) . To determine whether phosphomutations of the S589 residue in PER affect PER-mediated inhibition of dCLK transcription , we analyzed the two phases of bioluminescence oscillation . A single period of bioluminescence oscillation can be separated into a dCLK active phase ( crest ) or a dCLK inactive phase ( trough ) . This analysis revealed that the dCLK active phase of S589A mutants is shorter than both wild type and S589D , which are equivalent ( Figure 4A and B ) . On the other hand , the dCLK inactive phase of S589A and S589D mutants are equivalent to each other and shorter than wild type ( Figure 4A and B ) . These data can also be visualized when each cycle of bioluminescence oscillation is aligned with wild type ( Figure 4—figure supplement 1 ) . These data suggest that the dCLK inactive phase is shortened in both S589A and S589D mutants , consistent with the role of this residue in regulating PER stability , since early PER degradation allows early dCLK activation . These data further demonstrate that the dCLK active phase is shortened only in the S589A mutant , suggesting that inhibition of phosphorylation at S589 specifically accelerates PER activity ( dCLK inhibition ) . In other words , these data suggest that while both S589A and S589D mutants exhibit shorter circadian periods because of their effects on PER stability , S589A mutants exhibit a much shorter circadian period than S589D partly because inhibition of phosphorylation at this residue increases or accelerates PER-mediated inhibition of dCLK transcription . The above data suggest that inhibitory activity of PER is increased in S589A mutants while stability is decreased . On the other hand , phosphorylation of PER is correlated with increased PER-mediated inhibition and decreased PER stability , but the two activities are difficult to assess independently ( Edery et al . , 1994; Kivimäe et al . , 2008 ) . To distinguish between PER activity and stability as a function of phosphorylation state , we developed a novel flow cytometry based transcription assay in S2 cells to monitor both functions independently using per mutants that represent transitional isoforms informed by the phosphorylation cascade model ( Figure 5 ) . To monitor PER inhibition of dCLK activity , we measured expression of yfp fused to a PEST degradation sequence driven by a per promoter , which is activated by dCLK; clk itself was CFP tagged and built into the same vector so we could monitor dCLK expression . To monitor PER expression , we co-transfected this dual-expression vector with a vector that constitutively expresses per fused to mCherry . Using this system , we can measure PER stability ( mCherry fluorescence ) simultaneously with PER inhibitory activity ( YFP fluorescence ) ( Figure 5A ) . Co-transfection of the two vectors allowed for variability between dCLK and PER expression , but not dCLK and YFP expression . Thus , a low YFP signal can be definitively attributed to PER-mediated inhibition of transcription . Cells were imaged to verify expression and transcriptional inhibition of dCLK in PER-expressing cells ( Figure 5B ) . dCLK expressing cells were identified and quantified by flow cytometry for CFP , YFP and mCherry expression ( Figure 5—figure supplement 1A ) . Cells were then binned across the axis representing the mCherry fluorescence signal . Within each bin , YFP signal was normalized to dCLK ( YFP/CFP ) and plotted as a function of PER signal normalized to dCLK ( mCherry/CFP ) to measure PER activity and PER expression . The average fluorescence signal across all cells from the flow cytometry data correlates well with PER levels measured by immunoblotting ( Figure 5—figure supplement 1B–C ) . Using this assay , we confirmed the expected inverse correlation between YFP expression and wild type PER signals ( Figure 5C , open black circles ) . To assay the effects of S589 mutation on PER activity and stability , we compared PER variants containing S589A or S589D with wild-type PER . We found that PER protein containing the S589A phosphonull mutation exhibits a steeper negative slope than wt PER , suggesting that S589A mutant PER is a more effective inhibitor ( Figure 5C , red circles ) . Consistent with our in vivo data , S589D PER exhibits a slope statistically indistinguishable from wild type ( Figure 5C , blue circles ) . The extension of the curve across the x-axis is a measure of PER stability , demonstrating higher concentration of PER in individual cells , which can be quantified by plotting the percentage of cells as a function of PER ( mCherry ) signal ( Figure 5—figure supplement 1B ) . The destabilizing effect of mutating S589 , and the relative stability of S589D over S589A is in agreement with the stability exhibited by these mutants in vivo ( Figure 2C and D , Figure 5—figure supplement 1B–1C ) . Our data also demonstrate that S589A and S589D mutants inhibit dCLK-mediated transcription , with stronger transcriptional inhibition demonstrated by S589A . The relative instability of S589D compared to wild type PER seen here but not in vivo may be due to the inability of dCLK-mediated endogenous tim expression to stabilize high levels of actin-driven exogenous per expression . Differences in inhibitory activity or stability are not due to poor nuclear accumulation ( Figure 5—figure supplement 1D ) , which is facilitated by sufficient TIM expression mediated by dCLK activity in S2 cells ( Saez et al . , 2007 ) . This transcriptional inhibition assay therefore allows us to distinguish between mutations that affect transcription , protein stability or both . Our data demonstrate that both S589A and S589D mutants exhibit decreased stability relative to wt PER , while the S589A mutant exhibits increased PER activity relative to both wt PER and S589D . Thus , the phosphorylation state of S589 delays the circadian transcriptional feedback loop in two ways: by stabilizing PER and delaying its degradation and by regulating PER-mediated inhibition of dCLK transcriptional activity . PER stability is proposed to be regulated by the DBT phosphorylation of PER in a cascade that culminates in PER S47 phosphorylation , ubiquitination and degradation , which is antagonized by S589 phosphorylation ( Figure 6A ) ( Chiu et al . , 2011; Garbe et al . , 2013; Kivimäe et al . , 2008 ) . Blocking S589 phosphorylation promotes PER degradation ( Figure 3 ) , suggesting that a second phosphorylation route that bypasses S589 triggers PER degradation ( Routes I and II ) . PER-SD phosphorylation that promotes S589 phosphorylation can independently promote S47 phosphorylation and PER degradation ( Chiu et al . , 2011; Garbe et al . , 2013 ) . Thus , PER degradation through S47 can be regulated through two routes of phosphorylation , diverging with S589 phosphorylation . Why two routes exist to promote PER degradation through the same mechanism ( S47 phosphorylation ) has not been addressed . It is possible that selection of either of the two routes of PER depends on the timed regulation of PER inhibitory activity or PER degradation . We therefore set out to determine the effect of promoting the activation of each phosphorylation route and determining its effect on PER stability and activity . To test the functional relevance of the predicted phospho-states of PER regulated by DBT , we first tested the function of the PER-SD sites ( S604 , S607 , T610 , S613 ) that precede S589 and S47 phosphorylation ( Garbe et al . , 2013; Kivimäe et al . , 2008 ) . When these sites are substituted with alanines to block phosphorylation ( PER 4A ) , transcription is weakly inhibited ( Figure 6B ) , consistent with the long-period behavioral phenotype seen in similarly mutated flies ( Garbe et al . , 2013 ) . This protein also exhibits increased stability in comparison to wild type , producing relatively large amounts of PER ( Figure 6B , Figure 5—figure supplement 1B ) , again consistent with delayed PER degradation in the s-LNvs ( master pacemaker neurons ) of similarly mutated flies ( Garbe et al . , 2013 ) . Phosphomimetic substitutions at the same sites ( PER 4D ) exhibit wild type-like inhibitory activity , consistent with wild type behavioral rhythmicity seen in similarly mutated flies ( D . Garbe , personal communication , December 2013 ) , but also exhibit decreased stability in our transcription assay ( Figure 6B , Figure 5—figure supplement 1B ) . Thus in our assay , blocking phosphorylation of the PER-SD delays PER degradation , while mimicking phosphorylation in this region reduces PER stability . Functionally , mutations that block phosphorylation of the PER-SD exhibit lower transcriptional inhibitory activity , while phosphomimetic mutations are indistinguishable from wild type . The PER-SD region can also regulate PER stability and activity by blocking S589 phosphorylation ( Figure 6A ) . To determine how the PER-SD and S589 cooperate to regulate PER stability , we generated mutants that stabilize predicted transitional phospho-states of PER and that avoid endogenous phosphorylation events , and monitored their activity and stability ( Figure 6C ) . The phosphorylation-blocked mutant form ( S589A/4A ) results in a stable protein that is a poor inhibitor of dCLK-mediated transcription , similar to what is observed when a small DBT binding site on PER is deleted ( Kim et al . , 2007 ) . Thus the increased inhibitory activity of S589A is muted when combined with PER 4A . In S2 cells , the PER-SD region is phosphorylated without exogenous DBT or TIM expression ( Chiu et al . , 2008 ) , suggesting that PER-SD occurs before S589 phosphorylation , and is independent of TIM binding . The first stage of phosphorylation represented by S589A/4D reveals a strong , unstable inhibitor that resembles S589A . Thus , PER phosphorylation blocked at S589 is likely re-directed through Route I for degradation ( Figure 6A ) , while still maintaining its inhibitory activity established by a phosphorylated PER-SD . To determine the consequence of permitting S589 phosphorylation , we substituted both S589 and the PER-SD serines with aspartates ( S589D/4D ) to reveal a PER variant that is a strong and stable inhibitor . Analysis of purified PER fragments by size exclusion chromatography multi-angle light scattering ( SEC-MALS ) and small-angle X-ray scattering ( SAXS ) revealed no change in the global structure of either protein , with overall shapes of the proteins approximately the same , and with little difference in their flexibility as determined by trypsin protection assay . These data suggest that phosphorylation does not affect the global conformation of PER but may alter local structure in a manner that influences its function ( Figure 6—figure supplement 1 ) . Thus , phosphorylation of S589 commits PER to Route II , leading to a PER isoform that is stable and active through subtle changes to protein conformation . S589 phosphorylation is likely not blocked by PER-SD phosphorylation , but functions as a switch in delayed transcription in the circadian clock , committing PER to degradation or stable inhibition . A PER variant not predicted by the phosphorylation model , S589D/4A , is an unstable protein with indeterminate inhibitory activity that resembles S589D ( Figures 5C and 6C ) . This transition state can occur if S589 phosphorylation blocks PER-SD phosphorylation ( Fu , 2008; Kivimäe et al . , 2008 ) and may represent ( i ) a later stage isoform of PER at the end of transcriptional inhibition , ( ii ) an isoform that commits PER to degradation prematurely , or ( iii ) may not be an isoform relevant to the PER degradation pathway .
Delays in the transcriptional negative feedback loop of the circadian clock ensure that it oscillates with ~24 hr periodicity . Delayed transcription ( i . e . delayed PER degradation and transcription re-activation ) is one such critical delay . In this study , we present a model in which DBT stabilizes and activates PER inhibition to delay transcription in the circadian clock ( Figure 7 ) . DBT begins phosphorylation of PER in the PER-SD ( Garbe et al . , 2013; Kivimäe et al . , 2008 ) ( Step 1 ) , initiating inhibition of dCLK-mediated transcription by PER . S589 is subsequently phosphorylated to stabilize the PER/TIM inhibitory complex in its inhibition-active form ( Step 2 ) . Although protein extracts from whole heads suggest that S589 is phosphorylated at ZT20 , after nuclear entry ( Chiu et al . , 2011 ) , the stabilization effect S589 phosphorylation has on PER is apparent in the s-LNvs after CT02 ( Figure 2 ) , and regulates the temporal delay of transcription reactivation . Upon TIM degradation , PER is released from TIM protection and is degraded after CT06 ( Step 3 ) . Thus , the steps that delay dCLK-mediated transcription activity rely on the activation and stabilization of PER by DBT phosphorylation , before TIM degradation ensues . DBT is likely to regulate the stability and inhibitory activity of PER through separate mechanisms . Such functional uncoupling has also been suggested for the Neurospora clock ( Larrondo et al . , 2015 ) . Regions near the N- and C-termini of PER regulate PER stability and inhibitory activity , respectively . Phosphorylation of S47 at the N-terminus of PER regulates PER-SLIMB binding that leads to the ubiquitination and degradation of PER ( Chiu et al . , 2008 ) . The dCLK/CYC Inhibitory Domain ( CCID ) at the C-terminus of PER physically interacts with dCLK to block transcriptional activity ( Chang and Reppert , 2003 ) . These two functional regions of PER correlate with the two predicted DBT binding sites that reside at the PER N-terminus ( residues 1–365 ) and a third of the way from the C-terminus ( residues 755–809 ) ( Kim et al . , 2007; Kloss et al . , 1998 ) . The two target regions of DBT also provide an explanation for the function of the long- and short-period mutant forms of DBT . Both long-period ( dbtL ) and short-period ( dbtS ) behavioral phenotypes carry mutations that reduce the kinase activity of DBT ( Kivimäe et al . , 2008; Preuss et al . , 2004; Price et al . , 1998 ) . Lowered kinase activity that produces divergent behavioral phenotypes are explained by likely changes in DBT activity on specific PER target sequences . Indeed , we show that the S589 mutant flies resistant to the effects of a dbtS mutant background have periods that are lengthened in a dbtL mutant background ( Figure 1D ) . These data suggest that dbtS genetically interacts with S589 , as inferred by others ( Rothenfluh et al . , 2000 ) , while dbtL does not . Since dbtL and not dbtS delays PER degradation in cultured S2 cells ( Syed et al . , 2011 ) , dbtS may affect PER inhibition function through its ( in ) activity in the S589/PER-SD region . Distinct phospho-programs also regulate human PER2 stability and inhibitor activity ( Xu et al . , 2007 ) . The human study focuses on the hPER2 variants S662G and S662D , analogous to the PER S589A and S589D variants in flies , targeted by the mammalian orthologue of DBT , CKIδ . Unlike S589D , S662D exhibits a long behavioral period . This difference may be due to the differences in how these sites regulate PER . Although both S589A and S662G lead to a reduction of PER or hPER2 respectively , the decreased levels of the hPER2 S662G mutant is primarily due to increased hPER2 inhibitory activity at the hper2 locus ( decrease in transcription ) , rather than a change in protein stability . On the other hand , S589 phosphorylation ( S589D ) stabilizes PER in complex with TIM and helps activate PER inhibitory activity . Despite these differences , both S589A and S662G advance evening anticipation behavior , suggesting that phospho-regulation of these sites underlie this aspect of rhythmic behavior . Also , a reduction of DBT activity in mutant flies ( dbtL ) or reduced CKIδ expression in mice ( heterozygous expression ) both lead to longer period behavioral rhythms that are independent of residue substitution in positions S589/S662 . Thus , despite differences in the PER function that sites S662 and S589 regulate , both mammals and flies utilize distinct mechanisms that employ DBT/CKIδ to separately regulate the activity and stability of Period protein , delaying transcription reactivation in the circadian clock . The relationship between the DBT phosphorylation cascade , PER inhibitory activity and PER stability is not well understood . In vivo , it is difficult to determine if a mutation in per results in lower PER protein levels caused by destabilized protein or overactive transcriptional inhibition . To distinguish between the two possibilities , we must simultaneously measure transcription factor expression and activity . We therefore developed a fluorescence-based transcription assay to probe the relationship between S589 and the PER-SD in regulating stability and inhibitor activity . The previous PER phosphorylation model begins with phosphorylation of PER-SD that blocks PER S589 phosphorylation , and promotes PER S47 phosphorylation ( Chiu et al . , 2011; Garbe et al . , 2013 ) ( Figure 6A ) . Residue substitutions in PER that block and mimic phosphorylation reveal a step-wise activation and stabilization of PER in our transcription assay ( Figure 6 ) . S589A/4D represents the initiation of the phosphorylation cascade that activates PER transcriptional inhibitory activity . This isoform is unstable however , stabilized in inhibitor-active form with the phosphorylation of S589 , in the form of S589D/4D . This suggests no antagonistic relationship between the two regions . The previous model also predicts that S589 cannot be phosphorylated when the PER-SD region is ( Chiu et al . , 2011; Garbe et al . , 2013 ) . If this were the case , S589A/4D and PER 4D would behave similarly . Instead , S589A/4D appears more like S589A and PER4D more like S589D/4D in our assay , again suggesting no antagonistic relationship between the two regions . In vivo , S589 is phosphorylated at ~ZT20 , and appears to remain phosphorylated as PER is degraded ( Chiu et al . , 2011 ) . Other data suggest that the PER-SD region is phosphorylated before S589 , based on mass spectrometry of PER isolated from cultured cells ( Chiu et al . , 2008 ) , or epigenetic data ( Garbe et al . , 2013 ) . Thus , it is not clear that the isoform represented by S589A/4D is a long-lived isoform of PER , assuming that the PER-SD region is indeed phosphorylated prior to S589 . If this were true , it would suggest that this unstable early isoform of PER ( mimicked by S589A/4D ) requires rapid stabilization . There is no apparent need for an unstable PER protein early in transcriptional inhibition and it is more likely that both the S589 and PER-SD sites cooperate to regulate PER stability and function . Therefore , we suggest that PER-SD promotes S589 phosphorylation to stabilize and activate PER inhibition , and delay transcription by inhibiting dCLK , which peaks around CT02 . Indeed , S589 phosphorylation is relevant after CT02 to stabilize the PER/TIM complex ( Figure 2D and E ) . Phosphatases play a role in maintaining 24 hr behavioral rhythms ( Chen et al . , 2007; Fang et al . , 2007 ) . If phosphatases act on the stable inhibitor represented by S589D/4D , they can convert this PER variant to either an isoform represented by S589A/4D , or through multiple actions , to the S589D/4A isoform . In such a scenario , the re-formed S589A/4D representative isoform can be re-phosphorylated , pushing the clock forward . On the other hand , the S589D/4A representative isoform would be rapidly degraded , disallowing the clock from shifting backward . The small number of S589D/4A variants that escape degradation and become dephosphorylated to form S589A/4A can be re-phosphorylated through the same phosphorylation cascade . Thus the ordered phosphorylation of PER-SD and then S589 ensures that the circadian clock ticks forward . Mutations that block regulatory mechanisms of the circadian clock can cause significant health problems . We have shown that DBT-mediated activation , stabilization and degradation of PER regulates delayed transcription during the day . Mutations that interfere with this delay mechanism shorten perceived day and hasten evening anticipation ( Figure 1A ) . Thus , we demonstrate a biochemical mechanism that influences one aspect of circadian behavior ( evening anticipation ) . Similarly , a single mutation that alters splicing of cryptochrome ( transcriptional inhibitor in humans ) alters its binding affinity to Clock , also changing the delay of transcription inhibition in humans to cause a familial form of delayed sleep phase disorder ( Patke et al . , 2017 ) .
DNA encoding tim , per , dbt , clk ( and the indicated corresponding mutant ) was subcloned into the HS-Casper ( Saez and Young , 1996 ) or pAc5 . 1/V5-HisA plasmids ( Invitrogen , Norwalk , CT ) . Mutant per genes were subcloned into each vector using standard cloning techniques . CFP , YFP , mCherry , 3xHA , 3xmyc , 3xFLAG , 2xV5 were fused to the C-termini of the indicated genes , separated by a triple-glycine linker . Plasmid expressing yfp fused to a PEST sequence by per promoter ( per > yfp-PEST ) was a gift from Sheyum Syed . A HindIII site 3’ to the YFP-PEST cassette was used to insert a pAc > clk-cfp cassette to generate the reporter plasmid . S2 cells were obtained from ATCC ( Manassas , VA ) maintained in Schneider’s medium ( Invitrogen ) supplemented with 10% FBS ( Sigma-Aldrich , St . Louis , MO ) and tested negative for mycoplasma contamination . S2 cells were transfected using Effectene ( Qiagen , Hilden , Germany ) as per manufacturer’s protocols . Cells were resuspended in fresh medium one day after transfection and seeded onto Lab-Tek II chamber slides ( Nunc , Rochester , NY ) for imaging . The per mini-gene ( Bargiello et al . , 1984 ) was mutated using standard cloning techniques , and sub-cloned into an attB plasmid ( Bischof et al . , 2007 ) with a C-terminal 3xmyc tag . Plasmids were injected into attp40 Drosophila lines to generate transgenic lines ( BestGene , Chino Hills , CA ) . Individual flies were analyzed for locomotor activity in 12 hr: 12 hr light/dark light regiments ( LD ) or in constant darkness ( DD ) after at least three days entrainment , using the Drosophila Activity Monitor System IV ( Trikinetics , Waltham , MA ) . Period was determined using ClockLab Software ( Actimetrics , Wilmette , IL ) . UAS-DBT transgenic flies were a gift from Jeffrey Price ( University of Missouri , Kansas City ) . UAS-OGT transgenic flies were a gift from Louis Ptáček ( University of California , San Francisco ) . Flies overexpressing OGT were generated by recombining the second chromosome carrying the tim-UAS-Gal4 and UAS-OGT transgenes . Flies expressing the luciferase gene by per promoter ( plo ) were crossed to flies expressing the indicated per transgenes in a per0 background to generate a y , w , per0;plo/[per];+genotype for analysis . Two sets of 80 males were monitored in 3 cm plates with standard fly food supplemented with luciferin ( Cayman Chemicals , Ann Arbor , MI ) , as previously described ( Stanewsky et al . , 1997 ) in a scintillation counter ( Hamamatsu Photonics , Model LM2400 , Hamamatsu , Japan ) . Bioluminescence was measured in photons/10 min . Fly brains were collected , fixed , mounted , and imaged using Leica confocal microscopy as previously described ( Saez et al . , 2011; Top et al . , 2016 ) . Briefly , fly heads were fixed in PBS with 4% paraformaldehyde and 0 . 5% Triton X-100 . Brains were dissected and washed in PBS with 0 . 5% Triton X-100 and blocked in the same solution supplemented with 5% donkey serum . Brains were then probed with 1:200 dilution of PER antibody , or 1:1000 dilution of PDF antibody ( DSHB , Iowa City , IA ) . Washed brains were re-probed using 1:200 diluted secondary antibodies conjugated to Alexa-488 ( PER ) or Alexa-647 ( PDF ) . Brains were mounted using Fluoromount ( Beckman Coulter , Brea , CA ) and imaged using Leica confocal microscopy at 40X magnification . Fluorescence intensity was quantified using ImageJ , with bright and dark reference points to maintain consistent relative quantification across all images . In quantifying protein expression , entire cells were analyzed with no distinction between nuclear and cytoplasmic signal . To measure nuclear accumulation , S2 cells in chamber slides ( see above ) were heat shock induced to exogenously express TIM ( YFP ) or PER ( mCherry ) alongside actin-driven PER ( mCherry ) or TIM ( YFP ) , respectively , and actin-driven CFP in an air incubator at 37°C for 30 min . Cells were imaged using a DeltaVision system ( Applied Precision , Issaquah , WA ) equipped with an inverted Olympus IX70 microscope ( 60X oil objective , 1 . 42 N . A . ) , a CFP/YFP/mCherry filter set and dichroic mirror ( Chroma , Foothill Ranch , CA ) , a CCD camera ( Photometrics , Tucson , AZ ) , and an XYZ piezoelectric stage for locating and revisiting multiple cells . 50–70 cells were selected based on their CFP expression and CFP , YFP , mCherry channels recorded . Imaged cells were analyzed using a locally written algorithm in Matlab ( Mathworks Inc , Natick , MA ) , as previously described ( Top et al . , 2016 ) . To measure PER inhibition of YFP reporter , S2 cells expressing the dCLK/YFP reporter plasmid and PER were seeded on chamber slides and imaged similarly , with no heat shock treatment . 10–20 cells were selected based on their CFP ( dCLK ) expression and CFP , YFP , mCherry channels recorded and typical results shown . S2 cells were transfected in 10 cm dishes with plasmids expressing the indicated genes as described above . At the end of two days , cells were resuspended in 11 ml of medium and placed in 15 ml conical tubes . The tubes were heat shocked in a 37°C on a nutator for 30 min , and recovered for another 30 min at room temperature . Samples ( 1 ml ) were taken at the indicated time points ( once an hour ) , with cycloheximide ( 20 μg/ml ) added at hour 1 ( 1 hour post heat shock recovery ) . Cells were washed once with PBS and lysed using 200 μl modified RIPA ( 50 mM Tris pH 8 . 0 , 150 mM NaCl , 1 mM EDTA , 1% Triton X-100 , 0 . 5% Na-Deoxycholate , 25 mM NaF , 1 mM DTT , 20% Glycerol , 0 . 01% NaN3 ) . Equal volumes ( 15 μl , ~5–10 ug total protein ) of lysate were used in analysis by immunoblot using 6% or 10% SDS-PAGE gels . Blots were probed with 1:1000 anti-TIM antibody ( Myers et al . , 1996 ) , 1:1000 anti-PER antibody ( Top et al . , 2016 ) , 1:10 , 000 anti-V5 antibody ( Sigma-Aldrich ) , 1:3000 anti-HA antibody ( Roche , Basil , Switzerland ) , 1:10 , 000 anti-myc antibody ( Sigma-Aldrich ) , 1:10 , 000 anti-tubulin antibody ( Sigma-Aldrich ) . Protein bands were quantified using imageJ freeware . S2 cells expressing the reporter/clk-cfp plasmid described above were analyzed on a BD LSR-II cell sorter ( BD Biosciences , Franklin Lakes , NJ ) using FACSDiva v . 6 . 1 . 1 software . Fluorescence signal was filtered to monitor CFP ( 405 nm ) , YFP ( 488 nm ) and mCherry ( 561 nm ) . CFP positive cells ( N =~10 , 000 ) were binned into groups of increasing mCherry signal and plotted as a function of YFP signal . Variants of per ( amino acids 1–700 ) were sub-cloned into pGEX-6–1 p plasmid using standard cloning techniques . Bacteria were induced to express GST-fused PER fragment at 17°C at OD600 0 . 6 and grown overnight . Bacteria were lysed chemically in lysis buffer ( 500 mM NaCl , 50 mM Tris pH 7 . 5 , 5 mM DTT , Bug Buster ( EMD Millipore , Billerica , MA ) , RNase , and benozonase nuclease ) supplemented with protease inhibitors . PER fragments were purified using glutathione beads affinity chromatography , the GST tag was cleaved by prescission protease ( GE Lifesciences , Pittsburgh , PA ) , followed by size exclusion chromatography purification , and confirmed by mass spectrometry . SEC-MALS was conducted at room temperature on a Wyatt–WTC050N5 SEC column with buffer containing 500 mM NaCl , 50 mM Tris pH 7 . 5 , and 3 mM DTT . SAXS data was collected in a similar fashion as previously described ( Skou et al . , 2014 ) , but with an SEC component upstream from SAXS data collection . Briefly , SEC-SAXS data was collected on a size-exclusion column pre-equilibrated with 500 mM NaCl , 50 mM Tris pH 7 . 5 , and 3 mM DTT at the G1 line at the Cornell High Energy Synchrotron Source ( CHESS ) on a dual 100K PILATUS detector system . Approximately 10 mg/ml PER variants were injected onto the SEC column at a continuous flow rate of 0 . 2 ml/min . Data was collected with exposures of 2 s per frame ( 5 × 1011 photons/sec ) . Blank buffer with no protein was exposed and collected as background for subtraction . Lysozyme and glucose isomerase were used as standards for Rg and molecular weight calculations . RAW ( Nielsen et al . , 2009 ) and Primus ( Konarev et al . , 2003 ) were used to generate plots in Figure 6—figure supplement 1 . ‘q’ is calculated as q = 4πsin ( θ ) /λ , where θ is half of the angle between the incident X-ray beam and the scattered beam , and λ is the wavelength of X-rays . PER fragments ( 2 μl; 80 μM ) were incubated with trypsin ( 80 μM ) for the indicated times , and compared to undigested controls . Digestion reactions were stopped using 2 μl of 160 μM trypsin inhibitor and 10 μl SDS gel loading buffer . Digested products were analyzed by 4–12% NuPAGE BisTris gel , Coomassie stain , and quantified using ImageJ . Protein fragments were confirmed as PER fragments by mass spectrometry . | Many behaviors , such as when we fall asleep or wake up , follow the rhythm of day and night . This is regulated in part by our ‘circadian clock’ , which controls biological processes through the timed activation of hundreds of genes over the 24-hour day . In fruit flies , the proteins that form the core of the circadian clock activate and repress each other in such a way that their expression oscillates over a 24-hour cycle . During the late afternoon and early evening , the Clock protein initiates the production of proteins Period and Timeless: these two molecules then accumulate in the cell , and after binding to each other , they are transported into the nucleus . During the late night and early morning , this Period/Timeless complex inhibits the activity of Clock . After a delay , Period and Timeless are degraded . This allows Clock to be reactivated , restarting the cycle for the next day . Period is critical to help maintain the 24-hour oscillation shown by these proteins . A protein called Doubletime is responsible for making a number of chemical modifications on Period . It is unclear how these changes interact with each other , and how they influence the stability and function of Period when it is associated with Timeless . Here , Top et al . generate mutations in the fruit fly gene period to study these processes , and develop a new biomolecular technique to monitor the stability and activity of Period protein in insect cells grown in the laboratory . The experiments reveal new roles for the chemical changes made by Doubletime to Period . First , after Period associates with Timeless , Doubletime triggers certain modifications that lead to Period being able to inactivate Clock . Second , Doubletime makes another change in a nearby region of Period that results in the Period/Timeless complex being stabilized . Both sets of modifications help the complex to stay active and keep inhibiting Clock for long enough such that a 24-hour rhythm can be maintained . Finally , when Timeless is degraded , Period is released from the complex . At this time , the modifications made by Doubletime promote the degradation of Period , resetting the clock . Fruit flies with mutations that block this mechanism perceive the day as shorter . This shows that the smallest change to clock genes can disorganize behavior . Indeed in humans , health problems such as sleep or mental health disorders are associated with irregular circadian clocks . Understanding the biochemical mechanisms that keep the body clocks ticking could help to find new therapeutic targets for these conditions . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"neuroscience"
] | 2018 | CK1/Doubletime activity delays transcription activation in the circadian clock |
Plant trait diversity is known to influence population yield , but the scale at which this happens remains unknown: divergent individuals might change yields of immediate neighbors ( neighbor scale ) or of plants across a population ( population scale ) . We use Nicotiana attenuata plants silenced in mitogen-activated protein kinase 4 ( irMPK4 ) – with low water-use efficiency ( WUE ) – to study the scale at which water-use traits alter intraspecific population yields . In the field and glasshouse , we observed overyielding in populations with low percentages of irMPK4 plants , unrelated to water-use phenotypes . Paired-plant experiments excluded the occurrence of overyielding effects at the neighbor scale . Experimentally altering field arbuscular mycorrhizal fungal associations by silencing the Sym-pathway gene NaCCaMK did not affect reproductive overyielding , implicating an effect independent of belowground AMF interactions . Additionally , micro-grafting experiments revealed dependence on shoot-expressed MPK4 for N . attenuata to vary its yield per neighbor presence . We find that variation in a single gene , MPK4 , is responsible for population overyielding through a mechanism , independent of irMPK4’s WUE phenotype , at the aboveground , population scale .
Plant trait diversity is known to increase the productivity and stability of plant populations ( Cardinale et al . , 2012; Isbell et al . , 2017; Isbell et al . , 2015; Liang et al . , 2016; Loreau and de Mazancourt , 2013 ) . These effects may be produced from single gene variation among individuals: previously , wheat populations comprising of two lines varied only in their resistance to powdery mildew through single-gene modifications outperformed monoculture populations , even when the monocultures were comprised of resistant individuals ( Zeller et al . , 2012 ) . Additionally , forward genetics research recently identified a genetic locus associated with this diversity-productivity relationship ( Wuest and Niklaus , 2018 ) . However , establishing the mechanism that confers population productivity from a single genetic basis is difficult due to the complexity of plant population experiments . When individuals varying in traits or in loci of interest are planted in pairs , it is possible to conclude that their responses are due to the neighbor ( Gibson et al . , 1999; Harper , 1977 ) . When planted in populations , however , it is unclear whether a plant responds only to properties of its direct neighbors or of the entire population ( Gibson et al . , 1999; Radosevich , 1987 ) . The scale , or the hierarchical level in biological organization ( Allen and Starr , 1982 ) , at which diversity-productivity effects are constrained has not yet been determined , despite their importance in the study of complex population interactions ( Schneider , 2001 ) . Here , we define and investigate two spatial scales within populations that can be responsible for changes in total yield ( Figure 1 ) : the neighbor scale , where responding individuals ( RIs ) are constrained to change their growth and yield ( quantified in biomass and fitness correlates , that is flowers and seed capsules ) only in response to direct neighbors that differ in trait expression ( divergent plants ) ; and the population scale , where RIs can respond to the total composition of divergent plants in the entire population , creating a change in total population yield in direct proportion to RI abundance ( Crawford and Rudgers , 2012; Hughes et al . , 2008; Smith and Knapp , 2003 ) . Water-use traits are known to result in changes in total population yield ( Caldeira et al . , 2001; Comas et al . , 2013; Forrester , 2015; Kimball et al . , 2014; Marguerit et al . , 2014; Wang et al . , 2016; Wu et al . , 2016 ) . However , the scale at which this occurs ( neighbor or population ) remains unknown . WUE naturally varies among individuals , both within and among species ( Anderegg , 2015; Donovan et al . , 2007; Heschel et al . , 2002; Tortosa et al . , 2016; Yoo et al . , 2009 ) and intraspecific variation in WUE traits can be as great as interspecific variation ( Messier et al . , 2010 ) . Yield effects resulting from intraspecific WUE trait variation are of considerable agricultural interest ( Dutra et al . , 2018; Sreeman et al . , 2018 ) . Interestingly , WUE traits of some trees species alter the photosynthetic parameters and survival of neighboring trees ( Bunce et al . , 1977 ) , suggesting potential neighbor-scale responses that can dramatically influence the yield of populations . However , studies have not pursued how WUE trait variation cause either population-scale or neighbor-scale responses that are responsible for changes in population growth and yield due to the complications that emerge in studying variations in WUE phenotypes . To adequately study the scale at which RIs respond to variation in WUE of neighbors , one needs to anticipate several factors that would confound the analysis . WUE is calculated as the ratio of carbon assimilation to transpirational water loss , and WUE phenotypes typically result from altered stomatal function that increases plant transpiration . As the frequency of plants with low WUE ( high transpiration ) increases in a population , the availability of soil water to the population is known to decrease proportionally ( Zea-Cabrera et al . , 2006 ) . RIs may change their growth and yield in response to differences in soil water availability , rather than to the abundance of low-WUE plants in a population . Therefore , controlling for soil water availability independently of the frequency of plants with different WUE traits is essential for the analysis . The ecological relevance of variation in WUE traits is best evaluated in field populations , but standardizing water availability across populations in the field is rarely possible and thus combining inferences from field and glasshouse experiments , where soil water availability can be controlled using gravimetrically controlled watering , provides a useful way forward . Plants that vary in WUE as a result of single-gene manipulations greatly facilitate investigations into the scales at which yield responses are realized in populations . Here , we use isogenic plants , silenced in the expression of a single gene that profoundly influences stomatal behavior , to explore the scale at which WUE variation influences population yields . Mitogen-activated protein kinases ( MAPKs ) are part of a conserved signaling cascade essential in eukaryotes . The downstream targets of this phosphorylation cascade , such as transcription factors , enable specific plant responses through changes in plant growth and development ( Xu and Zhang , 2015 ) . Mitogen-activated protein kinase 4 ( MPK4 ) in Nicotiana attenuata and its homologues , MPK12 in Arabidopsis thaliana and MPK4/MPK4L in N . tabacum , have been implicated in responses to herbivore damage ( Gomi et al . , 2005; Hettenhausen et al . , 2013; Yanagawa et al . , 2016 ) , bacterial inoculation ( Hettenhausen et al . , 2012 ) , changes in exogenous and endogenous abscisic acid ( ABA ) and hydrogen peroxide levels ( Des Marais et al . , 2014; Hettenhausen et al . , 2012; Jammes et al . , 2009 ) , vapor pressure deficits ( Des Marais et al . , 2014 ) and ozone levels ( Gomi et al . , 2005; Yanagawa et al . , 2016 ) . Most of these responses involve the regulation of stomatal structure and function: silencing NaMPK4 or NtMPK4/L by RNA interference ( Na-irMPK4 and Nt-MPK4/L-IR , respectively ) or knocking out AtMPK12 ( At-mpk12 ) results in plants with larger stomata and stomatal apertures , and varying disruptions in stomatal closure ( Des Marais et al . , 2014; Gomi et al . , 2005; Hettenhausen et al . , 2012; Marten et al . , 2008; Yanagawa et al . , 2016 ) . The alteration of stomatal phenotypes by MPK4/12 expression strongly influences WUE . Na-irMPK4 , Nt-MPK4/L-IR and At-mpk12 all have increased transpiration rates which can be attributed to increased stomatal conductance ( Des Marais et al . , 2014; Gomi et al . , 2005; Hettenhausen et al . , 2012; Yanagawa et al . , 2016 ) . For Na-irMPK4 and At-mpk12 , this increase in transpiration rates has been shown to dwarf the associated increases in assimilation rates , resulting in low WUE ( Des Marais et al . , 2014; Hettenhausen et al . , 2012 ) . However , previous glasshouse studies that tested whether the presence of MPK4/12-derived WUE phenotypes results in individual growth and yield effects in paired-plant-in-a-pot interactions did not control for soil water availability ( Des Marais et al . , 2014; Hettenhausen et al . , 2012 ) . To our knowledge , no study has investigated whether variation in the abundance of a low WUE trait , generated from the silencing of a single gene , affects population yield; similarly unstudied is the scale at which this might occur . Here , we conduct such a study using experimental N . attenuata populations in both the glasshouse and field . The wild tobacco N . attenuata grows in xeric habitats in the western United States , where water limitation and WUE are selective factors throughout the growing season . N . attenuata typically grows in genetically diverse populations , in near-monocultures ( Baldwin and Morse , 1994; Baldwin et al . , 1994 ) , and is known to respond differently when paired with genetically varied intraspecific neighbors: a N . attenuata accession collected in Utah ( UT ) sharing a pot with an accession collected in Arizona ( AZ ) produces significantly smaller stalks than when sharing a pot with another UT plant ( Glawe et al . , 2003 ) . N . attenuata plants also naturally interact with AMF in the field , establishing networks of connected plants that have the potential to significantly change individuals’ responses to neighbors or populations ( reviewed in Oelmüller , 2019 ) . AMF are known to change soil water availability and transport among individuals in populations based on each individual’s ability to interact with AMF ( Egerton-Warburton et al . , 2007; Reynolds et al . , 2003; Yang et al . , 2013 ) . Additionally , AMF have been shown to significantly affect plant-plant interactions in populations , as they can transfer nutrients , defense signals and allelopathic chemicals ( Ferlian et al . , 2018; Gorzelak et al . , 2015; Song et al . , 2019 ) . Manipulating the ability of a population to interact through an AMF network can provide a means of dramatically altering belowground interactions and narrow the potential causes of population yield changes due to within- ( or between- ) species plant diversity ( Figure 1 ) . Calcium and calmodulin-dependent kinase ( CCaMK ) is required for successful plant symbiosis with AMF ( Lévy et al . , 2004 ) , and the abrogation of CCaMK expression provides a valuable tool to disconnect plants from AMF networks in the field ( Groten et al . , 2015 ) . Field plantations of transgenic N . attenuata crossed with CCaMK-deficient transgenic lines ( irCCaMK ) in the plant’s native habitat , the Great Basin Desert , allow for the study of population growth and yield effects resulting from trait variation , as well as the scales and tissues in which this variation occurs . Here , we use a single gene manipulation to create variation in WUE in the background of irCCaMK lines to separate the effects of AMF-mediated interactions in our analysis of the spatial scales at which variation in WUE traits influence population yields . We investigated the spatial scales at which variation in abundance of low WUE N . attenuata plants , generated by the abrogation of MPK4 expression , change total population growth and yield . We used a previously characterized irMPK4 line ( Hettenhausen et al . , 2012 ) and varied the percentages of this line in field populations with empty-vector ( EV ) control plants , both crossed with either irCCaMK and EV lines , to manipulate connectivity to the AMF network . We observed increased yields , referred to as overyielding , in populations with low percentages of MPK4-deficient plants ( ‘low-irMPK4’ ) , due primarily to increases in EV plant yield . To exclude soil water availability effects , we grew homozygous irMPK4 and EV lines in glasshouse populations under equal water availability and again observed overyielding in low-irMPK4 populations due to increases in EV yield . We further tested responses at the neighbor scale by growing mono- and mixed-genotype pairs of EV and irMPK4 under conditions of equal water availability and found no changes in growth or yield between pair types . We analyzed the yield of individuals with different configurations of immediate neighbor genotypes in our glasshouse populations , but these also did not explain changes in individual yields . From these results , we conclude that neighbor-scale responses are unlikely to be responsible for the overyielding phenomena . In the glasshouse , changes in EV plants’ photosynthetic parameters did not explain the yield increases in low-irMPK4 populations and importantly , EV and irMPK4 plants did not differ in their WUE phenotypes in field populations . Therefore we inferred that irMPK4’s WUE phenotype was not responsible for the observed overyielding at the population scale . In neighbor absence-presence paired tests , we observed that EV plants change their growth and yield when planted with or without a neighbor , while interestingly , irMPK4 plants do not . In similar experiments with EV shoots micro-grafted to irMPK4 roots , MPK4 expression in the shoot could remediate this effect , demonstrating a shoot-localized role for MPK4 in N . attenuata’s ability to alter yield in the presence of a neighbor . Additionally , manipulating field population’s AMF connectivity did not change observed reproductive overyielding , denoting a lack of belowground influence on this effect . From these results , we suggest a novel function of shoot MPK4 in mediating N . attenuata’s yield response to neighbors , unrelated to WUE , which in low-irMPK4 populations may result in reproductive overyielding .
N . attenuata plants silenced in the expression of MPK4 ( irMPK4 ) have a low water-use efficiency ( WUE ) phenotype in comparison to empty-vector ( EV ) control plants in the glasshouse ( Figure 2A ) . The loss of stomatal control increases transpiration rates to levels that surpass the increases in assimilation rates , consequently decreasing WUE , calculated as the ratio of assimilation:transpiration rates ( Hettenhausen et al . , 2012 ) . Populations of plants growing in the field are commonly interconnected by arbuscular mycorrhizal fungal ( AMF ) networks that are known to influence access to water and nutrients in the plant rhizosphere ( Egerton-Warburton et al . , 2007; Reynolds et al . , 2003; Yang et al . , 2013 ) , as well as within-population plant neighbor responses ( Ferlian et al . , 2018; Gorzelak et al . , 2015; Song et al . , 2019 ) . As silencing the expression of NaCCaMK disconnects plants from AMF networks ( Groten et al . , 2015 ) , we crossed isogenetic , homozygous irCCaMK plants with homozygous EV and irMPK4 lines to generate hemizygous EV x irCCaMK ( EVxCC ) and irMPK4 x irCCaMK ( MPxCC ) lines ( Figure 2B ) , which were used for field experiments . The hemizygous crosses retained the levels of MPK4 silencing of the homozygous irMPK4 lines: MPxCC showed an 87% reduction of MPK4 transcript accumulation relative to EVxCC in the field ( Figure 2C ) , whereas irMPK4 had 83% silencing efficiency relative to EV in the glasshouse ( Figure 2—figure supplement 1 ) . To evaluate the abrogation of AMF associations under controlled conditions , we grew the EVxCC and MPxCC crosses in the glasshouse with and without live AMF inoculum ( Rhizophagus irregularis ) and compared their AMF colonization characteristics to that of EV and a hemizygous irMPK4xEV ( MPxEV ) control cross . While EV and MPxEV were highly colonized in comparison to non-inoculated controls ( Figure 2C , LM , emmeans ( EV , AMF- ( n = 8 ) to AMF+ ( n = 8 ) ) , t = −8 . 894 , p = <0 . 0001; emmeans ( MPxEV , AMF- ( n = 7 ) to AMF+ ( n = 8 ) ) , t = −6 . 253 , p = <0 . 0001 ) , both EVxCC ( emmeans ( EVxCC , AMF- ( n = 7 ) to AMF+ ( n = 7 ) ) , t = −2 . 105 , p=0 . 4251 ) and MPxCC ( emmeans ( MPxCC , AMF- ( n = 8 ) to AMF+ ( n = 8 ) ) , t = −1 . 417 , p=0 . 8453 ) did not differ from un-inoculated controls in root length colonization ( RLC ) . Trypan blue-staining of roots showed the establishment of vesicles and hyphae in EV , but not in EVxCC and MPxCC plants ( Figure 2C ) . From these results , we conclude that the hemizygous crosses retain their MPK4 silencing and do not associate with AMF . In order to evaluate if the percentage of MPK4-deficient plants influences population yield under field conditions , growth and yield of EVxCC and MPxCC individuals in populations with varying percentages of MPxCCs ( 0 , 25 , 75 , 100%; Figure 3A; Figure 3—figure supplement 1 ) were measured and analyzed using de Wit replacement diagrams ( Figure 3B–G; Figure 3—figure supplement 2; de Wit , 1960; Harper , 1977 ) . Increases in yield , referred to as overyielding , were observed in the relative yield totals ( RYTs ) of 25% MPxCC populations in stalk height ( Figure 3C ) , shoot and root biomass ( Figure 3D–E ) , and unripe and ripe seed capsule values ( Figure 3F–G ) . This overyielding was due only to increases in EVxCC plants: in 25% irMPK4 populations , cumulative EVxCC plant trait values exceeded their predicted values based on their performance in monoculture . MPxCC plant trait values did not differ from their monoculture values in any population type . However , the increase in the cumulative EVxCC trait values in the replacement diagram was not reflected in significant differences between means of EVxCC individuals in 25% MPxCC populations versus in other population types ( Figure 3—figure supplement 2 ) , emphasizing the role of incremental benefits observable at the population-scale rather than in the performance characteristics of each individual in the population . Plants with low WUE are thought to increase the flow of water-soluble nutrients to the immediate area around their roots as a consequence of excessive transpiration rates ( del Amor and Marcelis , 2005; Zea-Cabrera et al . , 2006 ) . Therefore , we collected soil cores at 5 , 15 and 30 cm below the center of each population type and quantified total carbon ( Ctotal ) , nitrogen ( N ) , inorganic carbon ( Cinorg ) , organic carbon ( Corg ) , copper ( Cu ) , iron ( Fe ) , potassium ( K ) , phosphorus ( P ) and zinc ( Zn ) concentration ( Figure 3—figure supplement 3 ) . At each depth , there were no significant differences among populations for any nutrient except for Cinorg which was slightly increased at 5 cm depth in the 0% populations ( EVxCC monoculture ) , at 15 cm in 75% MPxCC populations , and at 30 cm in 100% MPxCC populations ( Figure 3—figure supplement 3C ) . We observed no increases in any inorganic nutrient at any soil depth in the 25% MPxCC populations ( Figure 3—figure supplement 3 ) . Furthermore , the percentage of MPxCC plants in populations did not significantly predict soil moisture at any sampling depth ( Figure 3—figure supplement 4A ) . From these results , we conclude that increasing the percentage of MPxCC plants in populations under field conditions leads to a non-additive trend in population yield , unrelated to soil moisture and inorganic nutrient availability , with overyielding occurring in 25% irMPK4 populations . To further evaluate whether the water-use phenotype of irMPK4 plants contributed to differences in water and nutrient availability for populations in ways that were undetectable in the field , we created populations in the glasshouse with increasing percentages of irMPK4 ( 0 , 17 , 50 , 83% and 100%; Figure 3H; Figure 3—figure supplement 5A ) in which we experimentally controlled for water availability among populations ( Figure 3—figure supplement 5B; see Water treatments in Materials and methods ) . Replacement diagrams were again used to analyze cumulative growth and yield of EV and irMPK4 plants in varying population types ( Figure 3I–M; individual means in Figure 3—figure supplement 6 ) . The analysis revealed overyielding in shoot biomass and total fitness correlates ( reproductive yield ) of low-irMPK4 populations ( 17%; Figure 3K , M ) , consistent with the field results . Due to the controlled watering schema of the glasshouse experiment , we conclude that this overyielding effect is independent of population water availability . To test if overyielding in low-irMPK4 field and glasshouse populations ( Figure 3 ) resulted from neighbor interactions of EV and irMPK4 plants , we investigated the growth and yield of EV and irMPK4 in monoculture and mixed pairs ( Figure 4A ) , again under conditions of equal water availability ( Figure 4B ) . Replacement diagrams revealed no evidence of overyielding in any of the measured growth and yield parameters for the mixed pairs ( Figure 4G–J ) . We conclude that for EV plants having one irMPK4 neighbor was not sufficient to produce the overyielding response we observed in low-irMPK4 populations . Varying local configurations of irMPK4 neighbors could also cause neighbor-scale overyielding in EV plants , a property we would not observe in our paired plant experiment . Therefore , in the glasshouse population experiment ( Figure 3H ) , we analyzed growth and fitness measurements of centrally located EV individuals with four direct neighbors . In 0% irMPK4 populations , all four neighbors were EV plants , in 17% irMPK4 populations , two were EV and two were irMPK4 plants , and in 50% and 83% irMPK4 populations , all four were irMPK4 plants . We observed that only in 50% irMPK4 populations , EV plants with four irMPK4 neighbors produced significantly higher growth and yield in comparison to EV plants grown in 0% irMPK4 populations ( Figure 3—figure supplement 6; Supplementary file 1 ) . However , 50% irMPK4 populations did not show overyielding ( Figure 3I–M ) , likely because irMPK4 plants simultaneously had significantly smaller rosettes , water contents , and yields compared with 100% irMPK4 monocultures ( Figure 3—figure supplement 6; Supplementary file 1 ) . Importantly , EV plants grown in 50% and 83% irMPK4 populations , with the same irMPK4 direct neighbor configuration , did not show consistent changes in growth and yield compared to monocultures . These results are consistent with the inference that overyielding does not occur at the neighbor scale . While an immediate neighbor response is not mediating the population overyielding response , we observed strong changes mainly in EV plants across our different population experiments . To test whether EV and irMPK4 plants respond differently to the presence of a neighbor , we included EV and irMPK4 planted as singles in our paired-pot experiment ( Figure 4A ) . We compared the growth and yield to single plants with individuals in mono- and mixed-culture pairs . Water contents of EV and irMPK4 plants did not differ , whether planted alone or in pairs ( Figure 4E; Table 1 ) , indicating equal water availability in the two potting types . EV plants with an EV or irMPK4 neighbor had smaller rosettes , shoot biomass , and reproductive yield than when planted alone ( Figure 4C–D and F; Table 1 ) . However , this reduction was independent of the neighbor’s genotype . In contrast , irMPK4 plants showed no differences in their rosette growth , shoot biomass , or yield when planted in pairs as compared to being grown alone ( Figure 4C–F ) . From these results , we conclude that MPK4 is required for N . attenuata’s growth and yield responses to a neighbor . To determine if the WUE phenotypes of EV and irMPK4 plants in glasshouse and field populations change with the percentage of MPK4-deficient plants , potentially causing overyielding in low-irMPK4 populations ( Figure 3 ) , we measured leaf photosynthetic parameters ( assimilation rate , transpiration rate , stomatal conductance ) and calculated the WUE of all individuals in both glasshouse and field experiments . In the glasshouse paired experiment , all measured leaf photosynthetic parameters of EV and irMPK4 plants in single pots were as previously reported ( Figure 2A ) , with irMPK4 plants having significantly higher assimilation rates , transpiration rates , and stomatal conductance than EV plants , and significantly lower WUE ( Figure 5A; Table 2 ) . When planted in pairs , EV and irMPK4 plants’ assimilation rates , transpiration rates , and stomatal conductance were not significantly different in monoculture ( red and blue shadings , respectively ) versus in mixed culture ( purple shading ) . EV plants had significantly lower WUE in mixed versus monoculture ( Figure 5A; Table 2 ) , whereas irMPK4 plants showed no significant change in WUE across the planting types . In the glasshouse population experiment , irMPK4 plants in 100% ( blue shading ) versus 50% ( purple shading ) irMPK4 populations showed no significant differences in any photosynthetic parameter ( Figure 5B ) , which was consistent with the glasshouse paired experiment . EV plants in 0% ( red shading ) versus 50% ( purple shading ) irMPK4 populations were not significantly different from each other in any parameter except for significantly higher transpiration rates of EV plants in 50% irMPK4 populations compared with those in 0% irMPK4 populations ( Figure 5B; LMER , EV: emmeans 0% ( n = 32 ) -50% ( n = 16 ) , t = −3 . 744 , p=0 . 0082 ) . While the results of statistical comparisons of EV responses ( 0% versus 50% irMPK4 ) differ between the pair ( Figure 5A ) and population ( Figure 5B ) experiments , the effects on the means for the two experiments remained consistent: with EV transpiration rates increasing ( Pairs0% to 50%: +0 . 49; Populations0% to 50%: +1 . 22 ) and WUE decreasing ( Pairs0% to 50%: −3 . 56; Populations0% to 50%: −13 . 78 ) . In the glasshouse , EV plants in low-irMPK4 populations did not have significantly different photosynthetic parameter values compared with other population types ( Figure 5B ) . In addition , photosynthetic parameters were measured at a pre-dawn ( AM; Figure 5—figure supplement 1 ) , which included dark-adapted chlorophyll fluorescence measurements ( Fv/Fm ) reflecting the maximum yield of the photosynthetic system ( Signarbieux and Feller , 2011 ) . The AM photosynthetic parameter values of EV plants in low-irMPK4 ( 17% ) populations also did not significantly differ from EV plants in any other population type ( Figure 5—figure supplement 1 ) . In the field experiment , EVxCC and MPxCC plants that lacked the ability to associate with arbuscular mycorrhizal networks ( AMF- ) , did not differ in any photosynthetic parameters , whether these were compared between genotypes or across population types ( Figure 5C ) . To test if the ability to interact with an AMF network changes patterns of photosynthetic performance , we additionally analyzed photosynthetic parameters of EV and irMPK4xEV ( MPxEV ) plants that could interact with AMF networks ( AMF+ ) . Similar to the irCCaMK crosses ( -AMF ) , EV and irMPK4 plants capable of associating with AMF did not differ across population types and the two genotypes ( Figure 5C ) . We further tested whether the AMF association could change photosynthetic parameters within a planting type . Only irMPK4 plants in 75% irMPK4 populations had marginally higher assimilation rates , and EV plants in 75% irMPK4 populations had a higher WUE ( Figure 5C , GLS , irMPK4 in 75% irMPK4: emmeans AMF- ( n = 3 ) - AMF+ ( n = 3 ) , t = −2 . 511 , p=0 . 0363; GLS , EV in 75% irMPK4: emmeans AMF- ( n = 3 ) - AMF+ ( n = 3 ) , t = −8 . 148 , p=0 . 0144 ) . From these field and the previous glasshouse results , we conclude that the WUE phenotype is not likely to have accounted for the greater growth and yield of plants in low-irMPK4 populations . We tested the effect of tissue-specific MPK4 expression on plant yield responses to a neighbor . To separate the role of irMPK4 expression in shoots from those in roots , we created chimeric plants by micro-grafting EV shoots to irMPK4 roots ( heterografts ) , EV shoots to EV roots ( EV homografts ) and irMPK4 shoots to irMPK4 roots ( irMPK4 homografts; Figure 6A ) . Because the RNAi silencing signals travel from shoots-to-roots but not vice versa in N . attenuata , micrografting of RNAi lines such as irMPK4 does not permit the investigation of shoot-only MPK4 knockdowns ( Fragoso et al . , 2011 ) . Hetero- and homo-irMPK4 grafts retained similar levels of MPK4 silencing in roots or roots and shoots , respectively ( Figure 2—figure supplement 1 ) . We grew the grafts under conditions of equal water availability , with or without an ungrafted EV neighbor . Photosynthetic parameter profiling of these grafted plants revealed that the heterografts were similar to EV homografts in assimilation , transpiration , stomatal conductance , and WUE ( Figure 6B ) , while the irMPK4 homografts showed significantly higher transpiration rates , and stomatal conductance and lower WUE ( Figure 6B; Table 3 ) . All graft types had shoot biomasses that were significantly reduced when plants were grown in pairs versus planted alone ( Figure 6C; Table 4 ) . The total root biomass per pot represented the roots of one plant for single pots , and two plants together for the paired pots . We compared the observed paired-pot root biomasses to a linear prediction of the paired-pot root biomass based on the addition of single-pot root biomasses of the respective graft types in the pair . Root biomasses of EV homograft pairs were equal to two times the root biomass of an EV homograft in a single pot . In contrast , the paired heterografts and irMPK4 homografts had smaller root biomasses than were predicted from individually grown plants ( Figure 6D ) . While both the EV homografts and heterografts displayed significant reductions in reproductive yield in response to an EV neighbor , irMPK4 homografts did not show a significant difference in reproductive yield between single and paired pl ants ( Figure 6E; Table 4 ) . From these results , we conclude that silencing MPK4 in the roots changes the neighbor-related root biomass production of N . attenuata , but MPK4 in the shoots is required to alter reproductive yield in response to neighbors . In order to evaluate if altering belowground interactions affects overyielding in 25% irMPK4 field populations , we compared the growth and yield of EV and MPxEV crosses , having the ability to interact with an AMF network , in field populations with varying percentages of MPxEV ( 0 , 25 , 75 , 100% ) with responses observed in populations with abrogated AMF interactions ( Figure 3 ) . Overyielding was observed in unripe and ripe seed capsule counts in 25% MPxEV populations ( Figure 7D–E ) , but not in the shoot and root biomasses RYTs for these same populations ( Figure 7B–C ) . The overyielding in capsules , similar to the response in populations without AMF network associations , occurred as a result of increases in the number of capsules in EV plants , relative to the predicted yield based on their productions in monoculture . To compare the biomass-to-reproductive-yield associations across populations with and without AMF association , we analyzed the data as allometric trajectories ( Weiner , 2004; Wu et al . , 2003 ) . The presence of the AMF network significantly changed the allometric trajectories of EV individuals in 0% irMPK4 populations: EV plants had a significantly larger allocation to seed capsules per unit biomass than did EVxCC plants ( Figure 7F , slopes: EVxCC ( n = 10 ) =0 . 73 , EV ( n = 7 ) =3 . 3 ) , and their trajectories started at a higher biomass threshold . In addition to the slope of the allometric trajectory , which indicates plasticity in resource allocation , the R2 value , which indicates the extent to which a plant’s trajectory is close to its reproductive potential ( Weiner , 2004 ) , also increased from EVxCC to EV plants in 0% irMPK4 populations ( Figure 7F; R2: EVxCC ( n = 10 ) =0 . 51 , EV ( n = 7 ) =0 . 82 ) . However , the allometric allocations of EV and EVxCC plants did not differ in 25% and 75% irMPK4 populations . MPxEV and MPxCC allometric trends did not differ in the 25% and 100% irMPK4 population type ( Figure 7G ) . The 75% populations were excluded due to a lack of replication at the end of the field season . We conclude that the loss of biomass overyielding in low-irMPK4 populations with AMF association is due to a change in the allometric trajectory of EV plants to higher biomass levels , which dwarfed EV biomass production in all other populations , without altering seed capsule production relative to the other populations . These results indicate that the overyielding does not require AMF-mediated belowground interactions .
Individual-level variation in resource-use traits , such as water-use efficiency ( WUE ) , can change overall population yields ( Campitelli et al . , 2016; Kenney et al . , 2014; Montazeaud et al . , 2017 ) . This occurs when plants respond to neighbors which are divergent in WUE with changes in growth and reproductive yield . However , the scale within a population’s hierarchical organization ( Allen and Starr , 1982 ) at which this phenomenon occurs is not yet known: yield changes in responding individuals ( RIs ) may be triggered only in immediate neighbors ( neighbor scale; Figure 1 ) or in individuals across a population ( population scale ) . Additionally , interactions may occur at these scales above- or belowground . Our analyses revealed that low abundances of irMPK4 plants intermixed with EV plants result in higher yields for N . attenuata populations , both in the glasshouse and the field ( Figure 3 ) . This overyielding effect was not caused by differences in soil water availability , which was controlled for in the glasshouse ( Figure 3H–M; Figure 3—figure supplement 5B and Figure 3—figure supplement 8; Figure 4 ) , nor irMPK4’s WUE phenotype ( Figure 2A ) , which was not different from the WUE of EV plants in the field ( Figure 5 ) . Interestingly , we find that yield-increasing responses in low-irMPK4 populations likely occurred aboveground , at the population scale ( Figures 4 , 5 and 7 ) . Given that manipulating mixing proportions ( Weiner , 1980 ) and total density ( He et al . , 2005; Stachová et al . , 2013 ) are known to change overyielding results in substitutive experiments , it was important to consider that our glasshouse populations ( 12 plants , 5 cm apart ) differed in plant density from our field populations ( four plants , 10 cm apart ) . However , given that the overyielding results were consistent between the glasshouse and field , we infer that in our case the overyielding observed in mixtures with low proportions of irMPK4 was not due to differences in total plant density of mixtures . Based on previous neighbor-effect studies using low-WUE phenotypes , we initially hypothesized that the neighbor scale ( Figure 1 , black arrow ) would be critical for producing individual changes in yield . For example , in a paired-plant competition experiment , an A . thaliana CVI mutant ( low WUE ) produced more seeds than Ler ( high WUE ) when both were planted with a control neighbor ( Campitelli et al . , 2016 ) . When extrapolated to a population level , this result could predict a neighbor-scale change on population yield . Specifically , if one plant in each paired interaction were an RI , changes in populations yields would primarily occur when abundant RIs exist in the population in close proximity to few differentiated plants ( i . e . populations with low percentages of well-spread differentiated individuals ) , creating many paired interactions without reducing RI numbers . Alternatively , if both plants were RIs , changed yields would be observed in populations with the either the highest abundance of interaction fronts between the two genotypes if the responses were in the same direction , or the lowest if they were in opposite directions . However , these possibilities rely on the assumption that a yield response is observed in response to trait variation in pairs . We therefore tested this pre-condition with EV and irMPK4 plants planted in monoculture and mixed culture pairs . Despite previously observed changes in neighbor responses to plants altered in their WUE ( Campitelli et al . , 2016 ) , we did not see differences in any growth or yield parameters in mixed EV and irMPK4 pairs in comparison to monoculture pairs ( Figure 4 ) . This discrepancy may be due to the difference between our methods and those of the earlier study: we controlled for water availability in our glasshouse experiments , to exclude that differences in water availability are driving the observed responses . Plants with low WUE can change the soil water availability in certain microenvironments and thus responses of neighbors due to the decreasing water table ( Zea-Cabrera et al . , 2006 ) . However , we did not observe differences in soil moisture among field populations with different irMPK4 abundances ( Figure 3—figure supplement 4 ) . Even an additional watering treatment on a section of the field plot did not cause differentiated rates of drying-down among the different population types ( Figure 3—figure supplement 4 ) . We directly tested the influence of water availability in the glasshouse by controlling for soil water availability among populations ( Figure 3H–M ) . We hypothesized that , with equalized soil water availability , we would not observe the overyielding response in low-irMPK4 populations . Interestingly , we again observed overyielding in low-irMPK4 glasshouse populations and therefore inferred the effect to be independent of soil water availability . Additionally , since controlling for soil water availability seemed to manipulate immediate neighbor plant interactions ( i . e . differences between our results and Campitelli et al . , 2016 ) , the overyielding in controlled water glasshouse populations suggested that the effect was likely not mediated at the neighbor scale . Paired-plant-in-a-pot experiments have only limited potential to study neighbor-scale interactions in population as individuals can only be observed in 0% , 50% and 100% irMPK4 mixtures ( Figure 4 ) . However , the growth and yield responses of plants at the neighbor scale could also depend on the identities of several immediate neighbors , or different percentages of immediate neighbors . Therefore , using a glasshouse population experiment , we compared the growth and yield of EV plants among a range of immediate irMPK4 neighbors ( 0 irMPK4 in 0% irMPK4 populations , 2 in 17% populations , and 4 in both 50% and 83% populations ) . We did not observe a correlation between the growth and yield of EV plants and the number of immediate irMPK4 neighbor plants ( Figure 3—figure supplement 6 ) . Based on these results and the lack of yield effects from our paired-pot experiments , we reject the hypothesis that the overyielding effects occurred at the neighbor scale , which suggest that the responses resulting in the observed population overyielding likely appear at the population scale . At the population scale , we initially investigated known mechanisms through which individual plants could affect the yield generation of other individuals in the population , independent of changes in soil water availability . Trees with differing WUE influence photosynthetic parameters of neighboring trees ( Bunce et al . , 1977 ) and altered photosynthetic parameters of plants , including WUE , have been shown to cause significant yield changes in plants ( Hatfield and Dold , 2019; South et al . , 2019 ) . Therefore , we examined the photosynthetic parameters of plants in the field and glasshouse populations . In the field , EV and irMPK4 plants did not differ in their photosynthetic parameters , including WUE ( Figure 5C ) . In the glasshouse populations , irMPK4 showed a decreased WUE phenotype , but the photosynthetic parameters of EV and irMPK4 individuals in low-irMPK4 populations were not different from those in their respective monocultures ( Figure 5B ) . From these results , we infer that the population-scale factor responsible for the low-irMPK4 population overyielding is independent not only of WUE , but also of the other previously observed photosynthetic phenotypes of irMPK4 plants , such as higher stomatal conductance , transpiration , and photosynthesis ( Hettenhausen et al . , 2012; Hettenhausen et al . , 2013 , Figure 5B ) . Other factors that could cause the observed population-scale overyielding in low-irMPK4 populations include niche complementarity and the exchange of chemical signals . Hydrological niche partitioning could explain the increased population yield through a more diverse , and therefore more efficient , use of local water resources , either spatially or temporally ( reviewed in Barry et al . , 2019 ) . In these instances , complementary plants may vary in above- or belowground water use through differences in water loss ( e . g . transpiration , WUE; reviewed in Silvertown et al . , 2015 ) or water acquisition ( e . g . through root spatial distribution , Dimitrakopoulos and Schmid , 2004 ) . However , the lack of differences in the WUE of irMPK4 plants and in the soil moisture among the different populations in the field , where overyielding effects were strongest ( Figure 3—figure supplement 4 ) , renders hydrological niche partitioning an unlikely explanation . Even so , other types of complementarity could have occurred , for example through one neighbor type ameliorating the habitat of its neighbors . This typically occurs through abiotic means ( i . e . changes in water or nutrient access; Bertness and Callaway , 1994; Wright et al . , 2017 ) but again , the field results showed no differences in soil moisture or nutrient content among the soils of different population types ( Figure 3—figure supplements 3 and 4 ) . However , the habitat could also be improved biotically: it is known that the presence of divergent individuals in a population that are resistant to a particular plant disease can reduce the spread of the disease , therefore allowing control plants in the population to produce more yield than when in monoculture ( Schmid , 1994; Zeller et al . , 2012 ) . Although irMPK4 plants were shown to be more susceptible to the bacterial pathogen Pseudomonas syringae pv tomato ( Hettenhausen et al . , 2012 ) , having the presence of susceptible individuals could still confer benefits to a population by serving as disease sinks ( see e . g . Keesing et al . , 2006 ) . Alternatively , aboveground spatial complementarity could have occurred , where irMPK4 plants’ smaller size ( Hettenhausen et al . , 2012 ) may have led to less competition for aerial space with EV plants , leading to their increased growth and population yield benefits ( see e . g . Lorentzen et al . , 2008; Williams et al . , 2017 ) . Although possible biotic factors could not be tested in the glasshouse , the role of spatial complementarity could be evaluated through a better understanding of the neighbor responses of EV and irMPK4 plants . The response of a plant to a neighbor is commonly evaluated through comparisons of the plant’s growth and yield when planted alone and when planted in a pair ( Díaz-Sierra et al . , 2017 ) . We therefore included single plants in our paired plant experiments ( Figures 4 and 6 ) . Interestingly , EV plants were the only individuals that responded with decreases in growth and yield to the presence of a neighbor ( Figure 4 ) . Although the yield in paired-plant experiments cannot be directly compared to the yield of EV plants in various population types , it suggests that EV plants are RIs: they change their growth and yield in response to neighbors , unlike irMPK4 plants ( Figure 4 ) . It is conceivable that EV plants could benefit from only certain percentages of irMPK4 plant neighbors to cause an emergent effect at the population scale if irMPK4’s consistent yield with or without a neighbor indicated a non-competitive response , though this cannot be distinguished in this study . However , complementarity effects would generally be expected to be strongest in equal mixtures , decreasing the likelihood of this possibility . Volatiles can accumulate differently in the headspaces of various plant populations ( Schuman et al . , 2015 ) and have been shown to be a mechanism by which plants detect and respond to potentially competitive neighbors ( Engelberth and Engelberth , 2019; Ninkovic et al . , 2016; Pierik et al . , 2013 ) . irMPK4 and EV plants are known to emit distinct volatile profiles in response to herbivory: irMPK4 plants release 5x higher levels of trans-α-bergamotene than EV plants ( Hettenhausen et al . , 2012 ) . As irMPK4 plants are inhibited in stomatal closure ( Hettenhausen et al . , 2012 ) , they may continuously release a volatile or volatile blend that causes a population-scale effect only at a certain concentration ( e . g . one that is generated in low-irMPK4 populations ) . This effect may be caused by a direct response of RIs , or might occur through interactions of the plant with various trophic levels , of which the latter has been observed to be volatile-mediated under field conditions ( Joo et al . , 2018 ) . In N . attenuata , a population yield effect was previously shown to be affected by single gene variation through a tri-trophic interaction: plants silenced in one component of the jasmonic-acid defense pathway ( i . e . defense-deficient plants ) intermixed with wild-type controls increased the susceptibility of entire populations to a generalist herbivore , which then reduced total canopy damage by a successive specialist herbivore , and increased wild-type yield in mixed versus monoculture populations ( Adam et al . , 2018 ) . NaMPK4 is known to silence part of a jasmonic acid defense pathway; the uninhibited pathway in irMPK4 plants reduces the mass of a specialist caterpillar feeding on this line ( Hettenhausen et al . , 2013 ) . This reduction of mass is often associated with higher mortality in young larvae due to disrupted feeding ( Cambron et al . , 2019; McGale et al . , 2018 ) , which could then reduce the presence of this specialist herbivore in the population , therefore decreasing canopy damage , and increasing the yield of control plants . Different volatiles released from irMPK4 plants could enhance this sort of effect ( i . e . increase oviposition of this herbivore on irMPK4 plants due to increased attractive volatile emissions; Kessler et al . , 2008 ) or otherwise might facilitate alternate multi-trophic interactions . Belowground , root exudates can accumulate in populations and provide information for plants about their neighbors’ identity and performance , which may result in plant growth responses ( Semchenko et al . , 2018 ) . Root exudates can also promote certain microbial community in the rhizosphere that can confer benefits to the entire population ( Berg and Smalla , 2009; Huang et al . , 2014; Fincheira and Quiroz , 2018 ) . However , irMPK4 root exudates have not yet been characterized . Further research on the chemical profiles and multi-trophic interactions of irMPK4 and EV plants in field populations is needed to identify potential above- or belowground factors which could cause responses at the population scale , but perhaps first narrowing the effect to either the above- or belowground scale would be helpful to study these potential mechanisms . We sought to distinguish the influence of the above- or belowground scale on the observed overyielding by experimentally altering arbuscular mycorrhizal fungal ( AMF ) associations in field populations with varied amounts of MPK4-deficient plants . Although AMF associations have a strong impact on plants’ access to soil water and nutrients ( Egerton-Warburton et al . , 2007; Reynolds et al . , 2003; Yang et al . , 2013 ) , as well as on belowground plant-plant interactions ( Ferlian et al . , 2018; Gorzelak et al . , 2015; Song et al . , 2019 ) , we consistently observed reproductive overyielding in low-irMPK4 populations regardless of the AMF association ( Figure 3; Figure 7 ) . The lack of change in the reproductive overyielding pattern supports further attention to aboveground tissues for mechanistic determination , although it does not exclude other belowground interactions as potential causative factors . To investigate the tissue dependence of the EV and irMPK4 plant neighbor responses we had previously studied in our paired experiment , we again used a single versus paired experimental set-up with grafted plants . The grafting experiment revealed that MPK4 transcript accumulation , or an MPK4-dependent downstream factor , is required in aboveground tissues for N . attenuata plants to be able to change their yield in response to neighbor presence ( Figures 4F and 6E ) . Additionally , the presence of MPK4 transcripts in the shoot alone is sufficient to facilitate the yield-altering neighbor response , specifically in reproductive correlate production ( Figure 7E ) . This tissue-specific role of MPK4 abundance in the shoot could be important in the mediation of reproductive overyielding in populations , but would need to be tested in a population experiment . Because the RNAi silencing signals travel from shoots to roots in N . attenuata ( Fragoso et al . , 2011 ) , we were unable to investigate the function of shoot-only MPK4 knockdowns . The use of plants with mutant or natural alleles of MPK4 or homologs , such as those identified in A . thaliana ( Des Marais et al . , 2014 ) , would allow for the analysis of reciprocal mpk12/wt and wt/mpk12 grafts . Together with a metabolic characterization ( e . g . volatiles ) and evaluations of multi-trophic interactions , these grafting experiments could identify the mechanisms mediating the observed population scale overyielding . The results of our study are consistent with the well-established phenomenon that divergent individuals in a community can increase community productivity ( Chapin III , 1997; Chapin , et al . , 1998; Crutsinger , 2016; Hooper et al . , 2005; Naeem et al . , 1994; Schulze and Mooney , 1994 ) and advance our understanding by identifying the spatial scale ( neighbors vs . population , above vs . belowground ) at which this yield response occurs . Additionally , we identify a single gene , MPK4 , which when silenced in low abundances in a population , is responsible for population overyielding . By excluding neighbor-scale effects and controlling resource availability , we demonstrate that MPK4 influences population yield through RIs at the population scale independent of water availability , but further experiments are needed to identify the specific mechanisms mediating this effect . This work contributes to our understanding of how populations may become more productive as a result of greater genetic and functional diversity and suggests that experiments exploring the scales at which these effects occur can identify novel means to increase the productivity of agronomic monocultures .
Characterization of the empty-vector ( EV ) Nicotiana attenuata control line ( pSOL3NC , line number A-04-266-3 ) is described in Bubner et al . ( 2006 ) . The irMPK4 line ( pRESC5MPK4 , line number A-7–163 ) , silenced in the production of MITOGEN-ACTIVATED PROTEIN KINASE 4 ( MPK4 ) through RNAi targeting MPK4 transcripts , is characterized in Hettenhausen et al . ( 2012 ) ; Hettenhausen et al . ( 2013 ) . The irCCaMK line ( pSOL8CCAMK , line number A-09-1212-1-4 ) , silenced in the production of CALCIUM AND CALMODULIN-DEPENDENT PROTEIN KINASE ( CCaMK ) through RNAi targeting CCaMK transcripts , is characterized in Groten et al . ( 2015 ) . EVxirCCaMK ( pSOL3NCxpSOL8CCAMK , ‘EVxCC’ ) and irMPK4xirCCaMK ( pRESC5MPK4xpSOL8CCAMK , ‘MPxCC’ ) crosses were generated by growing homozygous EV ( second generation , T2 ) , irMPK4 ( T2 ) and irCCaMK ( third generation , T3 ) in the glasshouse and hand pollinating the styles of EV and irMPK4 emasculated flowers with pollen from the anthers of irCCaMK flowers . Control crosses EVxEV ( pSOL3NCxpSOL3NC , ‘EVxEV’ ) and irMPK4xEV ( pRESC5MPK4xpSOL3NC , ‘MPxEV’ ) with the same paternal genotypes were created by pollination with pollen from EV . Hand-pollinated flowers were tagged with string and resulting seed capsules were collected . The ripe seeds from these crosses provided the seed source for the field population experiment ( Figure 3 ) . A characterization experiment in the glasshouse revealed that EVxEV and EVxCC , as well as MPxEV and MPxCC , were not significantly different in water loss rates per day ( Figure 3—figure supplement 7A , Supplementary file 1 ) , shoot and root biomass ( Figure 3—figure supplement 7B–C ) . For all other experiments in the glasshouse , T3 generation EV and irMPK4 homozygous lines were used . Importation and release of transgenic crosses in the field station ( Lytle Ranch , UT ) was carried out under Animal and Plant Health Inspection Service ( APHIS ) import permit numbers 07-341-101n ( EV ) and 10-349-101m ( EVxirCCaMK , irMPK4xirCCaMK , irMPK4xEV ) , and release 16-013-102r . Glasshouse and field germination and growth were described previously ( McGale et al . , 2018 ) , with modifications only in planting design . Field plants were planted in four-plant populations in a square design ( Figure 3; Figure 3—figure supplement 1 ) , with 10 cm between each adjacent neighbor . Plants of the glasshouse population experiment were potted in 12-plant populations ( Figure 3; Figure 3—figure supplement 5 ) , with 5 cm between each adjacent neighbor . Glasshouse plants in both of the paired experiments ( grafted and ungrafted ) were also planted 5 cm from their neighbor plants . The planting substrate consisted of a bottom layer of large clay aggregate ( Lecaton , 8–16 mm diameter , approximately 10% of pot volume ) , a central layer of small clay aggregate ( Lecaton , 2–4 mm diameter , approximately 80% of pot volume ) and a top layer of fine sand ( approximately 10% of pot volume ) . This substrate provides optimal drainage in the pots for the purposes of water control , and conditions similar to the sandy , clay soil of the natural habitat of N . attenuata . For the field experiment ( Figure 3A–G ) , rosette diameter measurements were extracted from photos taken between 19:00 and 20:00 , in which each individual plant was pictured next to a standard metal square ( 5 × 5 cm ) for scale . Plant stalk height measurements were recorded as the height from the base of the stalk at the ground level to the highest point of the topmost inflorescence . Plant shoot and root dry biomass were measured by placing respective biological matter in a paper bag inside of a plastic box with ventilation holes of 1 cm diameter drilled through the lid and left to dry for 15 days in the sun , before being removed from the bag and weighed . Unripe seed capsules were counted simultaneously for all plants , immediately before harvesting for shoot and root biomass . Due to APHIS regulations , ripening seed capsules were counted and subsequently removed to prevent opening and releasing seeds into the field; the total ripe capsules collected is presented ( Figure 3G ) . For all glasshouse experiments ( Figures 3H–M , 4 and 6 ) , rosette diameter was measured directly on the plant . Plant stalk height was measured as in the field . Shoot biomass consisted of all aboveground matter ( severed below the rosette ) , placed inside a bag for drying at 80°C for 2 hr , after which the plant matter was removed from the bag and weighed . The shoot biomass was also weighed for fresh mass , and the water content of the plant at harvest was reported as the difference between the fresh and dry shoot biomasses . All fitness correlates were counted at harvest , including flowers ( counted as flowers when the corolla became visible by pushing through the sepals ) , unripe and ripe seed capsules , and the total of all of these together was reported ( Figure 3M ) . Soil cores were taken from the field by driving a split tube core borer ( 53 mm , Eijkelkamp , Giesbeek , Netherlands ) 30 cm into the ground , and carefully removing it with the core intact . 5 cm pieces of field soil were cut from the core from 0 to 5 , 10 to 15 , and 25 to 30 cm below ground . Each of these 5-cm-thick sections were weighed , left to dry in the sun in UV-excluding boxes similar to those used for the drying of shoot biomass ( see Plant growth and yield measurements ) , and weighed again when dry ( determined to be when the mass fluctuated <0 . 1 g between days ) . Soil moisture was calculated for each sample ( % soil moisture = ( fresh soil mass - dry soil mass/fresh soil mass ) * 100 ) , taken from 21 to 30 dpp in the different population types ( Figure 3—figure supplement 4 , n = 1 per population ) . Soil cores were obtained using the same method at 54 dpp with replication ( n = 2–9 ) to determine the soil content of total , inorganic and organic carbon ( Ctotal , Cinorg , Corg , respectively ) , nitrogen ( N ) , copper ( Cu ) , iron ( Fe ) , potassium ( K ) , phosphorus ( P ) , and zinc ( Zn ) in each type of population at the end of the season ( Figure 3—figure supplement 3 ) . Samples were dried at 80°C for 6 hr in a drying oven , sieved and milled for Ctotal determination ( elemental analyzer; High TOC , Elementar , Hanau , Germany ) , Cinorg ( loss-on-ignition from elemental analyzer ) , Corg ( Ctotal - Cinorg ) , and N ( elemental analyzer ) at the Max Planck for Biogeochemistry in Jena , Germany . Cu/Fe/K/P/Zn concentration were determined by microwave digestion and atomic absorption spectroscopy ( Karpiuk et al . , 2016 ) . Field populations were watered every week for 1 hr at dusk ( 20:00 to 21:00 ) from a central water dripper ( 2 L/h drip rate ) present in each population . After 34 dpp , one section of the plot was no longer watered until the final harvest ( Dry ) , while a small subsection was watered two more times ( Wet ) in order to obtain gas exchange measurements on sections with varying water treatments at 48 dpp . Soil moistures at 21–30 dpp in these two parts of the plot ( see Soil moisture and element content ) were analyzed by regression to test if results from both of these parts could be summarized together in Figure 3 ( Figure 3—figure supplement 4 ) . Watering treatment and the interaction with depth or day did not significantly predict soil moisture ( Figure 3—figure supplement 4B , Wet subsection: ‘Part2’; day: ‘variable’; model fit: R2 = 0 . 406 , F ( 7 , 147 ) =16 . 04 , p-value=1 . 474e-15 ) . Therefore shoot and root biomass , as well as unripe and ripe seed capsule data collected from full populations in both sections were reported together as one mean ( Figure 3 ) . In the glasshouse , all populations and pairs ( grafted and ungrafted ) underwent the following regimented watering to control for water availability: after potting , pot were given establishment watering ( soil moisture maintained around 20% ) , allowing root development to the bottom of the pot for a transition from top watering to bottom watering . After 3 weeks , pots with population types began to show detectable differences in water loss and consumption-based watering began at ecologically relevant soil moistures ( Valim et al . , 2019 ) . This reflected the known decrease in soil moisture throughout the life cycle of N . attenuata in the field ( Zavala and Baldwin , 2004 ) . For the population and pair experiment , ecologically relevant soil moisture was achieved by daily watering of individual pots to a 2-day water supply , calculated as:WM=2∗mean ( WL−1 , WL−2 ) +DPWM=potmass ( g ) towhichthepotneededtobewateredWL−1=waterloss ( g ) fromtheprevioustothecurrentdayWL−2=waterloss ( g ) fromtwodaystoonedaypriorDP=drypotmass ( gofpotwithdrysubstrate , beforeplanting ) The 2-day water supply is illustrated for our glasshouse paired experiment ( Figure 4B ) . To allow larger growth and thus accentuate growth differences in plants in the grafted pair experiment ( Figure 6 ) , the water supply was raised to 5 days ( WM = 5*mean ( WL-1 , WL-2 ) + DP ) , bringing soil moisture percentages up to 20–30% . The higher soil moisture did not affect the differences in photosynthetic parameters of EV and irMPK4 homografts compared to those reported for the homozygous EV and irMPK4 plants in the paired experiment ( Figures 5A and 6B ) . There was no significant correlation between the amount of water added in our watering regimes and the amount of water lost ( demonstrated two times during watering regime of the grafted experiment , Figure 6—figure supplement 1B–C ) . Yara ZIM-probes were placed on 2–3 replicates of EV or irMPK4 plants in all glasshouse population types ( Figure 3H ) . The probes consist of two magnets clipped onto both sides of a leaf , of which the lower magnet includes a pressure sensor . All probes are initialized at a clamping pressure between 10 and 30 kPa on a turgescent leaf . The probes allow continuous measurement of leaf turgor pressure throughout an experiment , and we present 48 hr of continuous monitoring from 00:00 December 4th to 00:00 December 7th ( Figure 3—figure supplement 8 ) . Absolute pressure values could not be compared quantitatively; in contrast to previously reported measurements ( Zimmermann et al . , 2008 ) our recordings initialized at different clamping pressures . A comparison of three replicates of irMPK4 in 100% irMPK4 populations showed that variation in peak-trough amplitudes within one genotype/population type ( irMPK4 in 100% irMPK4 populations; Figure 3—figure supplement 8C ) exceeds variation between EV and irMPK4 in all other populations ( Figure 3—figure supplement 8A–B ) . Therefore , we compared the time of day at which changes in turgor pressure values occurred . For all EV and irMPK4 plants in all population types , the diurnal turgor pressure changes ( peaks and troughs ) do not occur at different times ( Figure 3—figure supplement 8A–B ) . Additionally , we observed if our genotypes experience differential diurnal dry-downs or unexpected drought events that may not be captured by our daily pot weighing and watering for our controlled water treatment . This may be reflected in ‘noisier’ curves ( increased oscillations within the peaks or troughs of the diurnal leaf turgor changes ) or inverted leaf turgor pressure curves around noon ( Martínez-Gimeno et al . , 2017 ) , however , we did not observe any of these qualities across our measured plants . We therefore inferred that our controlled watering treatment was not causing unknown diurnal drying differences among individuals in our glasshouse population types and proceeded with applying it to all glasshouse experiments ( Figure 3H–M , Figure 4 , Figure 5A–B , Figure 6 ) . Gas exchange measurements including photosynthesis and transpiration rates , and stomatal conductance ( via calculation ) , were performed using a LI-COR 6400XT infrared gas analyzer ( Lincoln , NE ) , both in the field and the glasshouse between 12:00 and 14:00 ( Figure 5 ) . The LI-6400XT was combined with a Leaf Chamber Fluorometer in the glasshouse to additionally obtain chlorophyll fluorescence measurements after 6 hr of dark adaptation ( lights off at 22:00 , measurements from 4:00 to 6:00; Figure 5—figure supplement 1A ) . A saturating pulse of light was applied to the dark-adapted leaves to ensure that all photosystem II ( PII ) energy was released as fluorescence and detected as the Fm value . Fv was calculated from Fm minus F0 ( F0 being the base level of fluorescence emitted without the saturating pulse ) . Fv/Fm represents the maximum quantum yield of PII , which was used as a measure of photosynthesis limitations unrelated to stomata ( Signarbieux and Feller , 2011 ) . Photosynthesis and transpiration rates were also acquired concomitantly with Fv/Fm values during this pre-dawn sampling , and stomatal conductance and WUE were calculated ( Figure 5—figure supplement 1B–E ) . Water-use efficiency ( WUE ) was calculated as the ratio of photosynthetic rate ( µmol CO2/m2s ) to transpiration rate ( mmol H2O/m2s ) , thus resulting in units of carbon dioxide molecules used per 1000 water molecules ( Figures 5A , B , C and 6B; Figure 5—figure supplement 1E ) . Seven-day-old seedlings were micro-grafted as described previously ( Fragoso et al . , 2011 ) , with EV scions grafted to both EV ( EV/EV ) and irMPK4 ( EV/irMPK4 ) rootstocks , and irMPK4 scions grafted only to irMPK4 ( irMPK4/irMPK4 ) rootstocks ( Figure 6 ) . The average grafting success was 90% ( p>0 . 05 between genotypes , ANOVA , Tukey HSD post hoc ) . RNA was extracted with TRIzol reagent ( Invitrogen ) according to the manufacturer’s instructions . cDNA was synthesized from 500 ng of total RNA using RevertAid H Minus reverse transcriptase ( Fermentas ) and oligo ( dT ) primer ( Fermentas ) . qPCR was performed in a Mx3005P PCR cycler ( Stratagene ) using 5X Takyon for Probe Assay ( No ROX ) Kit ( Eurogentec ) , TaqMan primer pairs and double fluorescent dye-labeled probe . N . attenuata Sulfite Reductase ( ECI ) was used as a standard housekeeping gene , and its primer sequences and probe , as well as the MPK4 primer sequences and probes , are as published previously ( Wu et al . , 2007 ) . MPK4 transcript levels were quantified relative to the housekeeping gene as described in Wu et al ( Figure 2C and Figure 2—figure supplement 1; Wu et al . , 2007 ) . All data were analyzed using R version 3 . 4 . 2 ( R Development Core Team , 2017 ) and RStudio version 1 . 0 . 153 ( Rstudio Team , 2016 ) . Replication for experiments is indicated in the figure captions . The replacement diagrams in Figures 3 , 4 and 7 do not display statistical significance , but facilitate the visualization of cumulative population overyielding ( de Wit , 1960 ) . Statistical means of the data used to produce these diagrams are presented in Figure 3—figure supplements 2 and 6 . Some pseudoreplication resulted from plants being measured from within the same population or pot throughout our experiments ( Figures 3–4 , 7 and Figure 3—figure supplements 2 and 6 ) . We evaluated whether this effect contributed significantly to changes in our dependent variable using ANOVA comparisons of nested linear mixed effects models ( i . e . LME/R models with and without the pseudoreplication as a random effect ) as described by Zuur et al . ( R packages lme4 , nlme; Bates et al . , 2015; Pinheiro and Bates , 2019; Zuur et al . , 2009 ) . Pseudoreplication was only included as a random effect in the respective LME/R model if it was significant; the model was then fitted for its fixed effects , and was checked for outliers ( through Cook’s distance and leverage plots ) , homoscedasticity and normality ( through graphical analysis of residuals; Zuur et al . , 2009 ) . Pairwise post hoc comparisons of fixed effects were extracted from the model using the R package emmeans ( Lenth et al . , 2019 ) , following their significance in an ANOVA . ANCOVA analyses were not used as the variable representing pseudoreplication ( i . e . population or pot number ) is inherently non-independent , which violates the assumptions for testing the significance of a covariate with ANCOVA , but does not violate the assumptions of a random effect in a mixed effects model . Datasets without significant pseudoreplication were fit to the best suited of either a linear model ( LM; RC Team Package stats ) , a generalized least squares model ( GLS , R package nlme ) or an LME model , and were checked for outliers , homoscedasticity and normality ( as above; Zuur et al . , 2009 ) . Pairwise post hoc comparisons for fixed effects were extracted as above , or with Tukey HSD tests ( R-core Team , 2015 ) following their significance in ANOVA . Regression analyses ( Figure 7F , G ) were performed using the lstrends function ( R package emmeans ) and statistical significances were extracted using pairs ( R-core Team , 2012 ) . | Whether on farmland or in a forest , plants do not grow in isolation . Plants compete with their neighbors over limited space and resources , and individual plants respond to this competition in different ways by changing how much they grow and how they use resources . The efficiency with which crop plants use water , for example , is one trait that is dramatically influenced by neighboring plants and is of increasing concern given the warming climate . Understanding the effects of interactions between individual plants in a population as a whole is complicated , especially in natural plant communities where neighbors are often from different species . For this reason , McGale et al . took a different approach and looked at neighbors that were all from the same species and differed only in the activity of a single gene . The species in question was coyote tobacco , a plant that is native to western North America . McGale et al . used genetic engineering to silence a gene called MPK4 , which was known from previous studies to have the effect of reducing water-use efficiency . Some of these ‘water-inefficient’ plants were then grown in mixed populations with plants that had normal levels of MPK4 . In experiments conducted both in a glasshouse and at a field station in the Utah desert , McGale et al . found that populations with a low percentage of the MPK4-silenced plants were actually more productive than ‘monocultures’ that were all one type or the other . Further analysis showed that the increase in productivity did not depend on the different soil nutrient or water use of the different populations , or even the density of the plants in the populations . Pairs of plants grown in single pots essentially ruled out any interactions between immediate neighbors being responsible for the increased productivity , suggesting that that effect must instead emerge at the level of the population . Perhaps unexpectedly , McGale et al . also found that the MPK4-silenced plants and control plants did not actually differ in how they used water when grown in the field ( previous studies had all been conducted in glasshouses ) , indicating that this trait also could not explain the observed population-level effect . Finally , experiments that involved grafting the shoots of one plant onto the roots of another suggested that the effect most likely comes from the aboveground parts of the plant . Ecologists have previously noted that more diverse populations typically have higher productivity . This new finding that a small percentage of slightly different plants in an otherwise uniform population can increase overall productivity will likely to be of special interest to researchers looking to boost the efficiency of agricultural ecosystems . Also , since MPK4 is highly conserved , and thus likely to be found in many plant species , this could be an interesting trait with which to study the interactions of natural plant communities . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"ecology",
"plant",
"biology"
] | 2020 | Determining the scale at which variation in a single gene changes population yields |
Diet plays a significant role in maintaining lifelong health . In particular , lowering the dietary protein: carbohydrate ratio can improve lifespan . This has been interpreted as a direct effect of these macronutrients on physiology . Using Drosophila melanogaster , we show that the role of protein and carbohydrate on lifespan is indirect , acting by altering the partitioning of limiting amounts of dietary sterols between reproduction and lifespan . Shorter lifespans in flies fed on high protein: carbohydrate diets can be rescued by supplementing their food with cholesterol . Not only does this fundamentally alter the way we interpret the mechanisms of lifespan extension by dietary restriction , these data highlight the important principle that life histories can be affected by nutrient-dependent trade-offs that are indirect and independent of the nutrients ( often macronutrients ) that are the focus of study . This brings us closer to understanding the mechanistic basis of dietary restriction .
Dietary restriction , also called calorie restriction , is a moderate reduction in food intake that extends healthy lifespan across a broad range of taxa , from yeast to primates ( Chapman and Partridge , 1996; Colman et al . , 2009; Lin et al . , 2002; McCay et al . , 1935 ) . The generality of this observation has inspired confidence that the health benefits of dietary restriction might also be employed to improve human ageing ( Campisi et al . , 2019 ) . In an attempt to harness its benefits , a great deal of current research is focused on discovering the nutritional components and the molecular mechanisms that underpin the lifespan benefits of dietary restriction ( López-Otín et al . , 2013; Simpson et al . , 2017 ) . Our current understanding of how diet modifies lifespan has grown out of evolutionary theory and experiments using model organisms . The most prominent theoretical explanation has been the disposable soma theory , which employs resource-based trade-offs to explain how dietary restriction can benefit lifespan ( Kirkwood , 1977; Shanley and Kirkwood , 2000 ) . This theory postulates that organisms will maximise fitness by strategically partitioning limiting dietary energy either to reproduction or somatic maintenance , the latter determining lifespan . This means that longer lifespan is inevitably coupled with reduced reproduction because both traits compete for the same limiting resource . Recent experimental work across a broad range of taxa has challenged the disposable soma theory by demonstrating that reproduction and lifespan respond predominantly to the balance of dietary macronutrients , not the overall energy content of the diet ( Mair et al . , 2005; Lee et al . , 2008; Skorupa et al . , 2008; Grandison et al . , 2009; Solon-Biet et al . , 2014; Solon-Biet et al . , 2015; Simpson et al . , 2017; Regan et al . , 2020 ) . Specifically , high protein , low carbohydrate diets are consistently associated with high reproduction and short lifespan , while low-protein , high-carbohydrate diets are associated with longer lifespan and lower levels of reproduction ( Piper et al . , 2011; Simpson et al . , 2017 ) . These data indicate that lifespan and reproduction are not in competition for limiting energy derived from the diet , but instead are optimised at different dietary protein: carbohydrate ratios . In response to these findings , an enormous effort is now focused on uncovering how macronutrient rebalancing , in particular protein dilution , acts to improve lifespan ( Blagosklonny , 2006; Blagosklonny , 2010; Moatt et al . , 2020; Regan et al . , 2020; Speakman , 2020 ) . Accumulating evidence indicates that the effect is mediated by reducing signalling through the amino acid sensitive Target Of Rapamycin ( TOR ) pathway to enhance cellular proteostasis ( Sanz et al . , 2004; Ayala et al . , 2007; Raubenheimer and Simpson , 2009; Simpson and Raubenheimer , 2009; Taylor and Dillin , 2011; Fanson et al . , 2012; Sabatini , 2017 ) . Although detrimental for lifespan , relatively high protein , low carbohydrate diets are beneficial for female reproduction ( Chong et al . , 2004; Solon-Biet et al . , 2015 ) . We have studied this closely in the fruit fly Drosophila melanogaster , where the principle driver of egg production is dietary protein ( Min and Tatar , 2006; Grandison et al . , 2009; Piper et al . , 2017 ) . Although protein is key , females must transfer dozens of nutrients into eggs for future embryo formation and not all of these components contribute to the flies’ decision to produce eggs ( Piper et al . , 2014; Mirth et al . , 2019; Wu et al . , 2020 ) . This means that high protein diets might drive mothers to produce eggs at a faster rate than they can support if the diet contains insufficient levels of the other components that are required to make eggs . In this scenario , the macronutrients would have an indirect effect on lifespan by changing the availability of another limiting nutrient that is required for somatic maintenance . If true , this would move the focus of mechanistic studies away from the direct effects of protein , TOR , and proteostasis , towards some other component of nutritional physiology . Distinguishing between these possible causes of death is important since it would fundamentally change our understanding of the way diet alters lifespan . It also has the important knock-on effect that we could change the way we design diets for longer life . For instance , supplementing high protein diets with key limiting nutrients would be as beneficial as restricting dietary protein or treating with pharmacological suppressors of TOR . Of the many studies that have examined the effects of dietary protein and carbohydrate on lifespan and reproduction in Drosophila , most have done so by varying dietary yeast and sugar proportions , where yeast is the flies’ natural source of protein ( Mair et al . , 2005; Lee et al . , 2008; Skorupa et al . , 2008 ) . However , yeast also contains all of the flies’ other essential macro and micronutrients whose relative proportions can change , and thus possibly interact with protein and carbohydrates to dictate life-history outcomes . We have previously found that depriving adult female flies of a source of sterols , an essential micronutrient for insects , imposes a minor cost on reproduction , but a substantial ( >50% ) cost to lifespan ( Piper et al . , 2014; Wu et al . , 2020 ) . These data indicate that yeast sterol levels may contribute to the effects on lifespan of protein and carbohydrate . To investigate the interactions between dietary protein , carbohydrate , and sterols systematically , we have used the design principles of the geometric framework for nutrition ( Simpson and Raubenheimer , 2012; Simpson and Raubenheimer , 1993 ) and a completely defined ( holidic ) diet that allows us to control the levels of each nutrient independently of all others ( Piper et al . , 2014; Piper et al . , 2017 ) . These data point to an important role for sterols in determining Drosophila lifespan , which we verified to be relevant in two yeast-based media that are often used in Drosophila lifespan studies . This work is critical to identifying how diet modifies lifespan at the molecular level , and highlights a new way to think about diet design to improve healthy ageing .
To examine the interactive effects of dietary protein , carbohydrate , and cholesterol on Drosophila lifespan and fecundity , we used our completely defined ( holidic ) diet ( Piper et al . , 2014 ) to manipulate each nutrient independently of all others . We selected dietary protein and carbohydrate concentrations that we know to elicit the full range of lifespan and fecundity responses to dietary restriction ( Lee et al . , 2008; Piper et al . , 2014 , Piper et al . , 2017; Ma et al . , 2020 ) . Similar to what we and others have found previously ( Mair et al . , 2005; Lee et al . , 2008; Grandison et al . , 2009; Piper et al . , 2014 , Piper et al . , 2017; Katewa et al . , 2016 ) , lifespan and reproduction were modified by dietary protein manipulations ( Figure 1 ) . Specifically , egg production showed a linear , positive correlation with dietary protein content ( Figure 1 , Supplementary file 1 ) , while lifespan showed a peak at intermediate protein ( 66 d median at 10 . 7 g/l ) , and fell away at both higher ( 49 d median at 33 . 1 g/l ) and lower ( 43 d median at 5 . 2 g/l ) concentrations ( Figure 1a–b , Supplementary file 2 ) . Thus , as is typical for dietary restriction experiments , restricting dietary protein from high to intermediate levels increased lifespan and decreased reproduction ( Lee et al . , 2008; Skorupa et al . , 2008; Grandison et al . , 2009; Katewa et al . , 2016; Le Couteur et al . , 2016 ) . When increasing dietary carbohydrate against an otherwise fixed nutritional background , egg laying was suppressed in a dose-dependent fashion , but lifespan remained at its maximum level and was unchanged across all carbohydrate doses ( ~66 d median , Figure 1c–d ) . The diet with the lowest concentration of carbohydrate ( 5 . 7 g/l ) , which also contained the intermediate protein level ( 10 . 7 g/l ) , supported both maximum lifespan ( Figure 1d; 66 d median ) and the highest level of egg laying ( 75 eggs/female ) of any diet in our experiment . Thus , as we have previously shown ( Piper et al . , 2017 ) , balancing the dietary protein and carbohydrate concentrations can reveal a single dietary optimum for both traits , showing that lifespan shortening is not necessarily caused by high egg laying alone . Most dietary restriction studies on Drosophila vary dietary protein by modifying the yeast levels in food ( Chapman and Partridge , 1996; Mair et al . , 2005; Lee et al . , 2008; Skorupa et al . , 2008 ) . While yeast is the flies’ major source of protein , it is also their only source of dozens of other nutrients , including sterols , which are essential micronutrients for insects ( Carvalho et al . , 2010 ) . To quantify the effects of varying dietary sterol levels on fly lifespan and egg laying , we maintained flies on the same set of diets as above , varying in protein and carbohydrate concentrations , while also varying cholesterol across four different levels: 0 g/l , 0 . 15 g/l ( low ) , 0 . 3 g/l ( medium; also our standard level ) , and 0 . 6 g/l ( high ) . Initial analysis of the data showed that diet type , when defined by variation of macronutrient composition , as well as variation in cholesterol concentration both significantly modified egg laying and lifespan ( Figure 2 , Supplementary files 3 and 4 ) . By contrast , we saw no main effect of dietary energy density ( calories ) on either trait , which is consistent with previous findings showing that these traits are driven by the relative proportion of protein and carbohydrate in the diet independently of caloric value ( Lee et al . , 2008; Mair et al . , 2005; Skorupa et al . , 2008 ) . We therefore proceeded in our analysis to assess how cholesterol modified the effects of protein and carbohydrates on these traits . We first compared the flies’ responses to variation in both protein and cholesterol ( Figure 2—figure supplement 1a–b ) . In general , lifespan was optimised at our intermediate dose of protein , while increasing cholesterol was beneficial , but with diminishing effect as its concentration was increased ( Figure 2 , Supplementary file 5 ) . Interestingly , changing cholesterol modified the flies’ lifespan response to protein , an effect that can be seen when the data are separated by level of cholesterol addition ( Figure 2—figure supplement 1a–b ) . At 0 g/l cholesterol ( Figure 2a ) increasing protein concentration in the diet decreased lifespan . However , at 0 . 15 g/l cholesterol , the shape of the response changed such that only the highest protein concentration decreased lifespan ( 35 d median; Figure 2c ) when compared with intermediate ( 9 . 7 g/l; 55 d median ) and low-protein ( 4 . 7 g/l; 52 d median ) diets . At 0 . 3 g/l of cholesterol , lifespan was highest on the diet with intermediate protein concentration ( 66d median ) and flies on the high protein diet were longer lived ( 49 d median ) than the flies on the lowest protein diet ( 43 d median ) . Finally , increasing cholesterol from 0 . 3 g/l to 0 . 6 g/l ( Figure 2g – Figure 2—figure supplement 1a–b ) did not change the way that lifespan responded to protein . Thus , lowering dietary cholesterol was detrimental for lifespan and it intensified the negative effects of increasing dietary protein concentrations . Across the same set of diets , we observed a generally beneficial effect on egg laying of increasing dietary protein and cholesterol , and both had diminishing benefits as their concentrations increased ( Figure 2—figure supplement 1d–e , Supplementary file 6 ) . Cholesterol also modified the way egg laying was affected by dietary protein ( Figure 2 ) . Increasing cholesterol from 0 g/l ( Figure 2a ) to 0 . 15 g/l ( Figure 2c ) amplified the positive effect on egg laying of increasing dietary protein . Further increasing cholesterol to 0 . 3 g/l had an additional benefit for egg laying ( Figure 2e ) , but only for flies on the highest protein diet ( compare Figure 2c with Figure 2e ) , while increasing cholesterol even further , to 0 . 6 g/l ( Figure 2g ) , did not change egg laying from that seen on 0 . 3 g/l . Thus , the response of egg laying to increasing protein was only compromised when cholesterol was completely removed from the diet , or when cholesterol was low ( 0 . 15 g/l ) and protein was high ( 30 g/l ) ( Figure 2c ) . Together , these data show that reducing cholesterol had negative effects on both lifespan and egg laying , and that these negative effects became more pronounced with increasing dietary protein . Furthermore , the negative interaction between lowering cholesterol and increasing protein was more severe and occurred at a lower protein concentration for lifespan than it did for egg laying . Next , we looked to see if changing dietary cholesterol modified the responses of lifespan and egg laying to variation in carbohydrate concentration ( Figure 2b , d , f , h - Figure 2—figure supplement 1c and f Supplementary files 5 and 6 ) . At 0 g/l cholesterol , lifespan was generally short ( 31d median ) but positively affected by increasing dietary carbohydrate ( up to 40 d median ) ( Figure 2b ) . As dietary cholesterol was increased to 0 . 15 g/l , lifespan on all diets was higher and the positive effect of increasing carbohydrate was preserved ( Figure 2d ) . However , when cholesterol reached 0 . 3 g/l , the flies were constantly long-lived , and lifespan was unaffected by dietary carbohydrate level ( 66 d median ) ( Figure 2f ) . This pattern was not changed by increasing cholesterol further to 0 . 6 g/l ( Figure 2h ) . Thus , each of our dietary carbohydrate levels could support maximal fly lifespan , but the lower carbohydrate diets were more susceptible to the detrimental effects of dietary cholesterol dilution . Increasing dietary carbohydrate had a generally negative impact on egg laying , and this effect was modified by the benefits of increasing dietary cholesterol ( Figure 2—figure supplement 1f , Supplementary file 6 ) . Without any cholesterol in the food , egg laying was consistently low and was negatively affected by increasing dietary carbohydrate ( Figure 2b ) . This negative effect of carbohydrate on egg laying became stronger as cholesterol was increased to 0 . 15 g/l ( Figure 2d ) and 0 . 3 g/l ( Figure 2f ) , with no further change as cholesterol increased from 0 . 3 g/l to 0 . 6 g/l ( Figure 2h ) . This increasingly negative relationship between carbohydrate and egg laying was caused because increasing cholesterol benefited egg laying more at lower dietary carbohydrate levels – the opposite of what we observed for lifespan . Thus , once again fly lifespan and egg laying worsened as cholesterol was diluted , but unlike its negative interaction with increasing dietary protein , the detrimental effects of lowering cholesterol became stronger as carbohydrate levels decreased . This indicates that the negative impact of lowering cholesterol is not a specific interaction with either high protein or low carbohydrate levels in the diet . We also found that the benefits of cholesterol addition were not related to the caloric content of the diet because cholesterol improved fecundity and extended lifespan of flies on almost all diets , including those with the lowest , intermediate , and highest caloric densities ( Figure 2—figure supplement 2 ) . Instead , lowering cholesterol produces worse outcomes as the dietary protein: carbohydrate ratio increases . This is the same change in macronutrient balance that promotes increasing egg laying . We saw that flies produce more eggs in response to increasing dietary protein: carbohydrate ratio and that these positive effects persist even as dietary cholesterol falls to a level that cannot fully support lifespan ( less than 0 . 3 g/l cholesterol ) . Thus , the protein: carbohydrate ratio appears to take precedence over dietary sterol levels in the decision to commit to reproduction . If this is the case , we expect to see a positive relationship between the dietary protein: carbohydrate ratio and egg laying across all levels of dietary cholesterol . This is indeed what we found , although the positive relationship was modified by cholesterol level ( Figure 3a , c , e , g , Supplementary file 7 ) , starting with a weak positive effect on 0 g/l cholesterol ( Figure 3a ) and becoming increasingly positive as cholesterol was increased to 0 . 15 g/l ( Figure 3c ) and 0 . 3 g/l ( Figure 3e ) . Once again , increasing cholesterol from 0 . 3 g/l to 0 . 6 g/l promoted no further benefit ( Figure 3g ) . Reproduction can impose a cost on future survival if resources that are essential for somatic maintenance are preferentially committed to making eggs . Since increasing protein: carbohydrate levels drove increasing egg laying , even when the adults were completely deprived of sterols , it is possible that females are committing sterols to egg production at a rate faster than they can replenish it from the diet . If true , mothers on low cholesterol diets would become shorter lived as egg laying increases , but when cholesterol is sufficient , the relationship between egg production and lifespan should become less negative . To test this , we plotted egg laying against lifespan for all replicates across all diets . This showed that egg laying was a significant predictor of lifespan , and that this relationship was modified by dietary cholesterol ( Figure 3b , d , f , h , Supplementary file 8 ) . When the data are grouped by dietary cholesterol level ( Figure 3 ) , we see that when cholesterol was at 0 g/l ( Figure 3b ) , there was a negative relationship between egg laying and lifespan , but as the cholesterol level increased , the correlation flattened , such that the slope was no longer negative for each level of cholesterol supplementation ( Figure 3d , f , h , Supplementary file 8 ) . Thus , when dietary cholesterol was insufficient , increasing dietary protein: carbohydrate drove higher egg laying ( Figure 3a ) and this predicted lifespan shortening ( Figure 3b ) – a scenario that exemplifies the negative relationship between reproduction and lifespan in response to increasing protein: carbohydrate levels that is regularly reported in the dietary restriction literature ( Mair et al . , 2005; Lee et al . , 2008; Skorupa et al . , 2008; Solon-Biet et al . , 2014; Solon-Biet et al . , 2015; Simpson et al . , 2017 ) . However , when cholesterol was increased , the negative relationship was reduced such that egg laying was either completely independent of lifespan ( Figure 3d ) or became slightly positively correlated , indicating that the dietary conditions , which promote egg laying , are the same as those that promote longer lifespan ( Figure 3f , h ) . Thus , higher egg laying in response to increasing protein: carbohydrate levels only shortens lifespan when cholesterol is insufficient to support egg production . TOR signalling is a key amino acid signalling pathway that is critical for growth , reproduction , and lifespan . Because TOR activity increases with dietary protein levels , it has been implicated as mediating the detrimental effects on lifespan of high protein diets ( Liu and Sabatini , 2020 ) . This is supported by the fact that rapamycin , a pharmacological suppressor of TOR , has been shown to extend lifespan across numerous species , including Drosophila where it also suppresses egg production across different caloric densities ( Bjedov et al . , 2010; Schinaman et al . , 2019; Scialò et al . , 2015 ) . If sterol limitation is the reason why high egg production on high protein: carbohydrate diets causes reduced lifespan , rapamycin might extend lifespan because it reduces egg production and therefore rescues females from sterol depletion . If true , rapamycin should extend life only when the flies on high protein: carbohydrate diets are sterol limited . As before , when we maintained flies on a high protein: carbohydrate diet , increasing dietary cholesterol from 0 . 1 to 0 . 3 g/l increased lifespan ( 62 d median v 69 d median ) ( Figure 4a ) . Egg laying was also slightly ( 34% ) , but significantly , elevated by cholesterol supplementation ( Figure 4b ) indicating that 0 . 1 g/l cholesterol was limiting for both lifespan and reproduction . When rapamycin was added to both foods , egg laying was almost completely suppressed ( Figure 4b ) . Rapamycin also extended fly lifespan , but only for flies on low dietary cholesterol ( 0 . 1 g/l ) ( Figure 4a ) , bringing their lifespan up to the same level as flies on higher cholesterol food ( 0 . 3 g/l; 69 d median ) . Adding rapamycin to the food with higher cholesterol did not result in any additional lifespan improvement over what was already achieved by increasing cholesterol alone ( 69 d median; Figure 4a ) . These data show that lifespan extension by rapamycin administration is conditional on the flies being on a low cholesterol diet . Together , our data are consistent with the flies’ lifespan being determined by having access to sufficiently high levels of dietary sterols that they have enough left over after reproduction to meet their needs for somatic maintenance . This can be achieved either by enriching the amount of cholesterol in the diet , or by reducing the flies’ expenditure on egg production , which can be achieved by reducing the dietary protein: carbohydrate ratio or by suppressing egg production pharmacologically . The experiments above were all performed using synthetic diets in which our ability to vary the absolute and relative concentrations of protein , carbohydrate , and sterol are limited only by solubility . However , most laboratories maintain fly populations on a diet that consists of yeast and sugar as well as variable numbers of other ingredients ( Piper , 2017 ) . Although the relative concentration of each nutrient in yeast is more constrained than on our synthetic diet , systematic studies have shown that the type and commercial source of yeast can have significant effects on overall dietary composition ( Lesperance and Broderick , 2020 ) and the relationship between lifespan and egg laying ( Bass et al . , 2007 ) . In Bass et al . , 2007 , the most dramatic lifespan reduction for increasing yeast was found when the fly food was made with a water-soluble extract of yeast that would contain very little , if any , sterols . Thus , similar to what we demonstrated on the synthetic diet , the shortening of fly lifespan when increasing the yeast content ( protein: carbohydrate ratio ) in lab foods may be caused by an insufficiency of dietary sterols . We tested the effects of supplementing cholesterol into two sugar/yeast recipes that have been commonly used to study the effects of dietary restriction on lifespan ( Mair et al . , 2005; Bass et al . , 2007; Katewa et al . , 2016 ) . These diets differ in both the number of ingredients used and the type of yeast; while both are Saccharomyces cerevisiae , one is a whole cell autolysate , while the other is a water-soluble extract . Adding 0 . 3 g/l cholesterol to both the low yeast ( dietary restriction ) and high yeast foods of both yeast types had a significant positive effect on lifespan ( Figure 5a , c ) and egg laying ( Figure 5b , d ) when compared to diets without cholesterol supplementation . The magnitude of this benefit to lifespan was greater for flies on the high yeast foods than on the low yeast foods , meaning that cholesterol supplementation narrowed the difference between the dietary restriction vs high yeast diet from 9 to 4% for flies on the autolysed yeast diets ( Figure 5a ) and from 81 to 25% lifespan extension for flies on the yeast extract diets ( Figure 5c ) . We note that even with cholesterol supplementation , the flies on the high yeast diet were still significantly shorter lived than those on the cholesterol supplemented low yeast food . This small additional cost of the high yeast food could reflect a detrimental ( toxic ) effect on lifespan of very high dietary protein , similar to what we observed in our highest protein diets on the synthetic foods ( Figure 2e , g ) . This is not rescuable by cholesterol supplementation and is not related to the number of eggs that females produce .
In the lab , flies can be successfully reared and maintained on a mixture of just sugar and yeast ( Pearl and Parker , 1921 ) . This diet is thought to reflect their natural diet of rotting fruit and the microbial community – principally the yeasts – that cause the fruit to decay ( Markow et al . , 2015; Piper , 2017 ) . Yeast contains all of the nutrients that flies require , including protein ( ~45% ) , carbohydrate ( ~40% ) , a small amount of fat ( ~7% ) , nucleic acids ( ~7% ) , and micronutrients , such as sterols , metal ions and vitamins , which are essential for flies . Drosophila rely heavily on protein from yeast , as well as carbohydrate from both yeast and plant sources , to guide their feeding behaviour . They select amongst foods containing the appropriate protein and carbohydrate concentrations to enhance their fitness ( Ribeiro and Dickson , 2010; Vargas et al . , 2010; Walker et al . , 2017 ) . Many of the other nutrients from their diet , including sterols , do not affect feeding behaviour , presumably because they are normally acquired in adequate quantities as part of a diet that is sufficient in macronutrients ( Walker et al . , 2015; Leitão-Gonçalves et al . , 2017; Münch et al . , 2020 ) . While the proportion of protein and carbohydrate in yeast remains relatively constant across growth conditions , the abundance of sterols can vary over a 10-fold range in response to changes in oxygen availability , which is essential for sterol biosynthesis ( Starr and Parks , 1962; Wilson and McLeod , 1976 ) . Thus , because fly feeding behaviour and egg production are almost entirely shaped by the macronutrients , fly lifespan is susceptible to reductions in the sterol: protein content of dietary yeast . Our data indicate that this is because protein drives sterols to be preferentially partitioned towards reproduction at the expense of maintaining the adult soma . While we have found this to be the case for flies feeding on lab based foods , it is also reasonable to expect it for flies feeding on rotting fruit , where microbial growth is largely fermentative ( driven by high sugar levels and limiting oxygen ) , producing ethanol and short chain acids to which Drosophila has evolved a healthy tolerance ( Geer et al . , 1993 ) . There have been several theoretical attempts to describe the mechanistic basis for the lifespan benefits of dietary restriction ( Blagosklonny , 2006; Blagosklonny , 2010; Kirkwood and Rose , 1991; Moatt et al . , 2020; Regan et al . , 2020; Speakman , 2020 ) . In particular , the disposable soma theory proposes that organisms will strategically reallocate nutrients towards somatic maintenance at the cost of reproduction when nutrients are scarce and that this enhances lifespan ( Kirkwood and Rose , 1991 ) . Our data indicate that this trade-off can exist for flies feeding on yeast , but only when dietary sterols are limiting . However , when dietary sterols are not limiting , this trade-off does not need to exist and a single nutritional optimum for both lifespan and reproduction can be found . Thus , the macronutrient balance that drives higher egg laying does not necessarily inflict a direct cost on lifespan . In mechanistic work , the increased lifespan under dietary restriction has been attributed to the benefits of reduced dietary protein , which enhances proteome maintenance via reduced TOR signalling ( Harrison et al . , 2009; Partridge et al . , 2011; Kapahi et al . , 2017; Piper et al . , 2017; Sabatini , 2017; Dobson et al . , 2018; Liu and Sabatini , 2020 ) . Interestingly , lysosomal cholesterol levels have recently been found to be a potent modifier of mTORC1 activity ( Castellano et al . , 2017; Zhang et al . , 2020 ) , which raises the possibility that both protein depletion and cholesterol addition modify ageing by reducing TOR signalling . However , the published data shows that cholesterol is an activator of TOR and cholesterol depletion inhibits its activity . Thus , we expect adding cholesterol to the diet would not reduce TOR signalling , but instead optimise conditions for maximal TOR signalling – especially on the high protein: carbohydrate diets in which we find cholesterol addition to be most effective for prolonging lifespan . These data indicate , therefore , that long life is possible when TOR signalling is high as long as the flies have sufficient sterols in their diets . Alternatively , longevity can still be assured under sterol-limiting conditions by reducing the cost of reproduction , either by reducing dietary protein , by adding rapamycin which suppresses reproduction or by making flies infertile ( Wu et al . , 2020 ) . These data indicate that longevity assurance in D . melanogaster is not the result of enhanced proteostasis triggered by lowered TOR , but is instead , a side effect of avoiding sterol depletion caused by an over-investment in egg production . Rapamycin is known to extend the lifespan of various organisms including C . elegans , yeast and mammals ( Harrison et al . , 2009; Kapahi et al . , 2010; Powers et al . , 2006; Robida-Stubbs et al . , 2012 ) . Because C . elegans cannot synthesise its own sterols , rapamycin might increase lifespan by preventing sterol depletion in a manner similar to what we have observed in Drosophila . However , sterols may not be lifespan limiting in other organisms such as yeast and mammals that have the ability to synthesise their own sterols . One explanation is that the administration of rapamycin prevents other micronutrient deficiencies caused by over-investment in growth and/or reproduction in response to high levels of dietary protein . For instance , rodents will export calcium from their own bones and teeth to meet the demands of pregnancy and lactation ( Miller and Bowman , 2004; Ozbek et al . , 2004; Speakman , 2008 ) . For this reason , it would be interesting to see if providing additional micronutrients to the diets of ad libitum-fed mice can mimic the benefits of dietary restriction , similar to what we see for sterol supplementation in flies . Another possibility is that rapamycin rescues animals from the effects of protein toxicity , which can occur at concentrations of protein higher than what we used in this study . In our experiments , we limit the upper range of dietary protein concentrations so as not to exceed those that are beneficial to reproduction . This practice is informed by the desire to study how dietary restriction enhances somatic maintenance to extend life in already healthy individuals , rather than studying the reduced risk of dying that occurs when flies are prevented from over-consuming . To test this , it would be interesting to study the effects of rapamycin addition over a broader range of protein concentrations than what we have used . If true , this has the important implication that rapamycin , and indeed different diet compositions , may prolong animal lifespan by more than one molecular mechanism . These are important implications to explore since the majority of work studying ageing in lab organisms is based on the assumption that the mechanisms are evolutionarily conserved , in the hope that they will benefit humans . Our data show that the detrimental effects of a high protein: carbohydrate diet on lifespan in female Drosophila melanogaster are , to a significant extent , driven by an indirect nutrient trade-off , in which the macronutrients drive maternal sterol depletion by enhancing egg laying . This is a fundamentally different mechanism from the predominant view that reducing protein: carbohydrate balance in diets improves lifespan by a direct action to reduce TOR signalling and enhance proteostasis . Because of our discovery , we show that the shortened lifespan of flies on a high protein: carbohydrate diet can be improved by supplementing their diet with cholesterol , as well as by reducing egg production by lowering the dietary protein: carbohydrate ratio or by administering rapamycin . Further work is now needed to discover the mechanisms through which cholesterol works to modify lifespan in Drosophila melanogaster , and the role of other important micronutrients in healthy ageing across taxa .
All experiments were conducted using a wild type Drosophila melanogaster strain called Dahomey ( Mair et al . , 2005 ) . These flies have been maintained in large numbers with overlapping generations to maintain genetic diversity . Upon removal from their population cages , flies were reared for two generations at a controlled density before use in experiments , to control for possible parental effects . Eggs for age-synchronised flies were collected over 18 hr , and the resulting adult flies emerged during a 12 hr window . They were then allowed to mate for 48 hr before being anaesthetised with CO2 , at which point females were separated and allocated into experimental vials . Stocks were maintained and experiments were conducted at 25 °C on a 12 hr: 12 hr light:dark cycle at 65% humidity ( Bass et al . , 2007 ) . For all lifespan assays , flies were placed into vials ( FS32 , Pathtech ) containing 3 ml of treatment food at a density of ten flies per vial , with ten replicate vials per treatment . Flies were transferred to fresh vials every two to three days at which point deaths and censors were recorded and saved using the software package Dlife ( Linford et al . , 2013; Piper and Partridge , 2016 ) . Fecundity was measured as the sum of the mean number of eggs laid per female once per week over four weeks ( commencing on approximately day 8 of the experiment ) , except for the sugar yeast ( SY ) medium experiment , for which egg counts were recorded in weeks one , two and three . These timepoints were selected because measuring reproductive output during the first weeks of egg laying has shown to be representative of life-long fecundity in flies ( Chapman and Partridge , 1996 ) . The eggs laid on the food surfaces of all vials were imaged using a web camera mounted on a Zeiss dissecting microscope and eggs were counted both manually and using Quantifly ( Waithe et al . , 2015 ) . Quantifly was trained using five images for each cholesterol concentration due to variance in food opacity . The same methods for making the holidic medium described above were used to make all diets used in the rapamycin experiment . In this case however 18 . 9 g/l protein: 17 . 1 g/l carbohydrate were used . Cholesterol was added to the diet at a concentration of either 0 . 1 g/l or 0 . 3 g/l ( cholesterol supplemented ) and rapamycin was added to a final concentration in the diet of 10 μM . Diets were either un-supplemented , supplemented with cholesterol , rapamycin , or both . Four sugar/yeast ( SY ) diets were created using sucrose ( Bundaberg Sugar , Melbourne Distributors ) and either whole yeast autolysate ( MP Biomedicals , LLC , #903312 ) or yeast extract ( Bacto Yeast Extract , #212750 ) . These diets correspond to previously published conditions for high protein ( fully fed ) and low protein ( dietary restriction ) conditions ( Bass et al . , 2007; Katewa et al . , 2016; Mair et al . , 2005 ) . The high protein diets contained , per litre 50 g sucrose and 200 g autolysed yeast , or 50 g sucrose , 50 g yeast extract plus 86 g of cornmeal ( The Full Pantry , Victoria , Australia ) . The low- protein diets contained , per litre 50 g sucrose and 100 g autolysed yeast or 50 g sucrose , 5 g yeast extract plus 86 g cornmeal . To each of these diets , we added cholesterol ( Glentham Life Sciences , GEO100 , #100IEZ ) at a concentration of either 0 or 0 . 3 g/l . Cholesterol was added to all diets as a powder which was mixed in with all other dry ingredients prior to cooking . This gave us a total of four experimental diets per yeast . All statistical analyses were performed using R ( version 3 . 3 . 0 , available from http://www . R-project . org/ ) . One outlier was removed from the data set as the total number of eggs laid for that particular vial was more than two standard deviations from the mean . Omitting this point did not modify the significance of any of the statistical analyses or change any conclusions . For each experimental vial the median lifespan and mean number of eggs laid were obtained prior to analysis . Linear mixed effect models were used to analyse all data obtained using the holidic media . For the analysis of data obtained using the holidic media , a model reduction was performed by stepwise removal of the most complex non-significant term until any further removal significantly reduced the model fit . Log rank tests were used to compare the survival curves in the rapamycin experiment and yeast based dietary experiments . Finally , two-way ANOVAs were used to analyse egg laying results for the yeast based experiments and rapamycin experiment . Plots were made in Graphpad Prism ( version 8 . 4 . 2 ) . | For the past fifteen years , animal studies have consistently shown that a low-protein , high-carbohydrate ( ‘carbs’ ) diet can extend the lifespan of many organisms , but at the cost of the number of offspring an individual can produce . Yet , it is still unclear what the best dietary balance is , and how these effects arise . One potential explanation could be that reproduction damages the body: low levels of proteins would therefore prolong life by lowering the reproductive output . Here , Zanco et al . examined the possibility that protein intake in fruit flies could instead be acting indirectly by changing the levels of a fat-like molecule called cholesterol , which is used to maintain the body and to support reproduction . To test this idea , groups of fruit flies were fed high levels of proteins . This led to increased reproduction rates , in turn depleting the mothers’ reserves of cholesterol . Without enough of the molecule in their diet , the insects were less able to maintain their bodies , which reduced their lifespan . When Zanco et al . added cholesterol to a high-protein diet , the flies lived for the normal length of time . Longer lifespan therefore did not require restriction of the diet or any of its components . In fact , the flies that lived the longest ate protein rich diets , and reproduced the most . This study helps to better understand why changes in diet can influence how long an organism lives for , highlighting that the abundance of certain key molecules may be more important than restricting the levels of proteins , carbs or calories actually consumed . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"evolutionary",
"biology",
"medicine"
] | 2021 | A dietary sterol trade-off determines lifespan responses to dietary restriction in Drosophila melanogaster females |
SCP1 as a nuclear transcriptional regulator acts globally to silence neuronal genes and to affect the dephosphorylation of RNA Pol ll . However , we report the first finding and description of SCP1 as a plasma membrane-localized protein in various cancer cells using EGFP- or other epitope-fused SCP1 . Membrane-located SCP1 dephosphorylates AKT at serine 473 , leading to the abolishment of serine 473 phosphorylation that results in suppressed angiogenesis and a decreased risk of tumorigenesis . Consistently , we observed increased AKT phosphorylation and angiogenesis followed by enhanced tumorigenesis in Ctdsp1 ( which encodes SCP1 ) gene - knockout mice . Importantly , we discovered that the membrane localization of SCP1 is crucial for impeding angiogenesis and tumor growth , and this localization depends on palmitoylation of a conserved cysteine motif within its NH2 terminus . Thus , our study discovers a novel mechanism underlying SCP1 shuttling between the plasma membrane and nucleus , which constitutes a unique pathway in transducing AKT signaling that is closely linked to angiogenesis and tumorigenesis .
Tumor angiogenesis is required for tumor growth and metastasis since tumors cannot grow without nutrients and oxygen when their diameters are beyond 1–2 mm , which is considered a distinct characteristic of cancer progression from early to terminal stages ( Folkman , 1971; Nicholson and Theodorescu , 2004 ) . Angiogenesis inhibition approaches have emerged as attractive and promising strategies in anti-cancer treatment . AKT is a central protein kinase in various cell activities , especially for angiogenesis , tumor growth , and progression ( Vivanco and Sawyers , 2002; Manning and Cantley , 2007 ) . Therefore , therapeutic strategies targeting the inhibition of AKT activity in cancer treatments have drawn great public attention during recent decades . AKT is activated by different extracellular signals such as GPCRs ( G protein-coupled receptor ) , growth factors , and integrins ( Hemmings and Restuccia , 2015 , 2012 ) . The activation of AKT is initiated by its recruitment to the cell membrane through the interaction of its PH ( Pleckstrin homology ) domain with PIP3 on the plasma membrane ( Franke et al . , 1997 ) . A large body of evidence shows that AKT activation is closely correlated with its phosphorylation at amino acid serine 473 ( S473 ) , while full activation of AKT requires its phosphorylation at S473 and threonine 308 ( T308 ) ( Ramaswamy et al . , 1999; Kawase et al . , 2009 ) . Activated AKT consequently regulates various downstream targets such as GSK3 and the transcription factor Foxos by direct phosphorylation ( Hemmings and Restuccia , 2012; Cohen and Frame , 2001 ) . Interestingly , accumulating data have shown that AKT can be activated through its S473 phosphorylation by a large variety of molecules and relay a stimulus to downstream oncogenic processes ( Ju et al . , 2014; Morrison Joly et al . , 2016; Meric-Bernstam et al . , 2014 ) . In addition , recent discoveries implicate that phosphorylated AKT on S473 may function as an important parameter of oncogenesis and cancer therapy . Likewise , AKT was found to relocate to the nucleus of resistant cells , where it was phosphorylated at S473 by DNA-dependent protein kinase , and this activation inhibited cisplatin-mediated apoptosis in cervical cancers ( Stronach et al . , 2011 ) . Furthermore , cholesterol increased AKT S473 phosphorylation , leading to enhanced tumor growth and a greater number of spontaneous metastases to the lungs in Apoe–/– mice , whereas cholesterol depletion in the cell membrane abrogated AKT S473 phosphorylation , suggesting AKT S473 phosphorylation/dephosphorylation may take place at the plasma membrane ( Alikhani et al . , 2013 ) . Protein phosphorylation is a reversible process that is mediated by kinases and phosphatases . Compartmentalized phosphorylation/dephosphorylation is a key switch for controlling protein activation and inactivation and complex signal transduction in various biological processes . It has been reported that AKT activity is negatively regulated by protein phosphatases . For example , ubiquitously expressed protein phosphatase 1 ( PP1 ) and PP2A can suppress AKT activity by direct dephosphorylation of AKT at T308 ( Ivaska et al . , 2002; Resjö et al . , 2002; Yellaturu et al . , 2002; Xu et al . , 2003 ) . PH domain-containing proteins such as the phosphatases PHLPP1 and PHLPP2 are able to dephosphorylate AKT at both S473 and T308 ( the other site required for full AKT activation ) in tumors ( Brognard et al . , 2007 ) . Unfortunately , so far , the known phosphatases involved in AKT inhibition are ubiquitously expressed in cells , which make those phosphatases difficult to be quantified and analyzed under living conditions . Therefore , these phosphatases are not applicable as observational tools for screening potential drugs and novel therapeutic targets in cancer treatments . Moreover , intracellular AKT activity is actually the cause of compromised signals that are orchestrated by various signaling pathways and imbalanced with the direct input signals of drug loading . In pursuit of optimal outcomes of cancer therapies , it is important to identify a unique membrane-localized phosphatase for AKT S473 inactivation , which is the upstream target relaying AKT signals directly from outside the drug stimulus and is the prerequisite of blocking AKT oncogenic signaling from the initial step . Small CTD phosphatases ( SCPs ) belong to a family of metal-dependent serine/threonine phosphatases containing a Mg2+ binding DXDX ( T/V ) motif ( Kamenski et al . , 2004 ) . SCPs were originally identified as small CTD containing Pol II phosphatases that shared a similar phosphatase domain with the Pol II CTD phosphatase FCP1 ( Yeo et al . , 2003 ) . SCPs are evolutionarily conserved transcriptional co-repressors for silencing neuronal gene expression via their interaction with the REST/NRSF ( RE1 Silencing Transcription Factor or Neural Restrictive Silencing Factor ) complex ( Yeo et al . , 2005 ) . SCPs , identified as SCP1 ( protein-encoding gene Ctdsp1 ) , SCP2 , and SCP3 , are involved in the TGF-β pathway and dephosphorylate distinct oncoproteins such as Snail , promyelocytic leukemia , and c-Myc ( Knockaert et al . , 2006; Wrighton et al . , 2006; Wu et al . , 2009; Lin et al . , 2014 ) . SCP1 and SCP2 are found to be localized in the nucleus , whereas SCP3 is found in both the cytosol and intracellular membranes ( Yeo et al . , 2005; Wu et al . , 2009; Siniossoglou et al . , 2000; Visvanathan et al . , 2007 ) . Despite a large body of evidence indicating that the proper subcellular location of signal molecules is crucial for the accurate signal transduction that accounts for normal biological functions ( Hung and Link , 2011 ) , the mechanisms underlying the trafficking diversion of SCPs are still unknown . Protein palmitoylation through the thioester linkage of 16-carbon fatty acids to cysteine residues is unique in that it is the only reversible lipid modification ( Bijlmakers and Marsh , 2003; Smotrys and Linder , 2004; Aicart-Ramos et al . , 2011 ) . Palmitoylation governs protein function in many ways: palmitoylation can promote membrane translocation of protein and it can facilitate vesicle trafficking and thus contribute to cellular signaling transduction . In neurons and cardiomyocytes , palmitoylation plays an important role in organ development , synaptic plasticity , and the establishment of membrane excitation platforms by clustering various ion channels and transporters ( Kang et al . , 2008; Fukata and Fukata , 2010; Fujiwara et al . , 2016 ) . Moreover , palmitoylation is intimately involved in signaling networks by interacting with various protein molecules such as insulin in the control of phenotypic and functional changes of endothelial cells ( Wei et al . , 2014 ) . A previous study has shown that SCP1 is a palmitoylation substrate by mass spectrometry , whereas direct evidence to prove SCP1 palmitoylation is missing ( Martin and Cravatt , 2009 ) . In this study , we find that SCP1 , a known nuclear phosphatase , can dephosphorylate AKT by screening a phosphatase library to search for a potential phosphatase that is capable of inactivating AKT . To our surprise , SCP1 is found to be largely located at the plasma membrane by tracing its cellular localization using EGFP-fused SCP1 and SCP1 proteins with different epitope tags . Such membrane-bound SCP1 specifically dephosphorylates AKT at S473 and suppresses angiogenesis , thereby decreasing tumorigenic risk and subsequent tumor growth of lung carcinoma cell-inoculated nude mice . In parallel , increased AKT phosphorylation and promoted angiogenesis , together with a notable risk of tumorigenesis , were observed in SCP1-knockout mice . Moreover , the membrane localization of SCP1 is majorly dependent on the palmitoylation of a conserved cysteine motif within its NH2 terminus , which has a prominent role in SCP1 shuttling between the plasma membrane and nucleus , and thus halting angiogenesis and tumor growth . Collectively , our findings reveal the distinct role of SCP1 as a palmitoylation-dependent phosphatase that negative regulates AKT-mediated angiogenesis and tumorigenesis .
To explore the functional role of SCP1 in the inhibition of AKT activity , we surprisingly found that both GFP-SCP1 and Flag-SCP1 were mainly localized at the plasma membrane in HeLa cells , where they were well co-localized with the plasma membrane marker GFP-PLCδ-PH ( Figure 1A ) . Similar results were also obtained in various cell lines , including MDCK , COS7 , MCF7 , HEK293T , and DLD1 cells ( Figure 1—figure supplement 1A ) . To rule out the artificial intervention from the fused GFP protein , we compared the subcellular localizations of untagged and GFP-fused SCP1 at the N terminus or C terminus , respectively . Our data demonstrated that both untagged and GFP-fused SCP1 showed similar membrane localizations ( Figure 1—figure supplement 1B ) , suggesting that SCP1 is a membrane-localized protein . SCP2 and SCP3 , which are structurally similar to SCP1 , appeared to possess an identical distribution pattern at the plasma membrane when they were transiently expressed in 293 T cells , whereas CTDSPL2 ( SCP4 ) was mainly localized in the nucleus ( Figure 1—figure supplement 1C ) . Cell fractionation experiments confirmed that most part of the SCP1 was in the membrane fraction , with a small proportion in the nuclear fraction , but none of the SCP1 was found in the cytosolic fraction ( Figure 1B ) . As expected , endogenous SCP1 was also easily detected at the plasma membrane using cell membrane fraction assays ( Figure 1C ) . The membrane localization of SCP1 was further confirmed using FRAP ( The Ferric Reducing Ability of Plasma ) assays and time-lapse microscopy ( Figure 1—figure supplement 1D ) . Interestingly , treatment with brefeldin A ( BFA ) , a blocker of protein trafficking from the Golgi to the plasma membrane , had little effect on the membrane localization of SCP1 ( Figure 1—figure supplement 1E ) , suggesting that the membrane-localized SCP1 proteins are not newly synthesized through Golgi/endosome trafficking . Taken together , our data clearly demonstrate that SCP1 is a plasma membrane-associated phosphatase . 10 . 7554/eLife . 22058 . 003Figure 1 . Membrane localization of SCP1 ( A ) SCP1 was co-localized with PLCδ-PH on the cell membrane . FLAG-SCP1 was transfected with or without PLCδ-PH-GFP in HeLa . The subcellular localization of SCP1 and PLCδ-PH-GFP was analyzed using immunofluorescence assay , and both the horizontal section ( X–Y ) and vertical section ( X–Z ) were photographed . ( B ) and ( C ) Subcellular localization of SCP1 in cells . HEK-293T cells were transfected and the subcellular localizations of transfected SCP1 ( B ) or endogenous SCP1 ( C ) were analyzed using western blotting . ( D ) Cartoon of different deletion mutations of SCP1 . Yes ( Y ) and no ( N ) represent SCP1 or truncated mutant membrane localizations , respectively . ( E ) HeLa cells were transfected with GFP-SCP1 or its mutants for 24 h and then analyzed for their subcellular localization using immunofluorescence assays . DOI: http://dx . doi . org/10 . 7554/eLife . 22058 . 00310 . 7554/eLife . 22058 . 004Figure 1—figure supplement 1 . SCP1 is membrane localized . ( A ) SCP1 was localized on the cell membrane in various cells . SCP1 was expressed in DLD1 , MCF7 , HeLa , HEK293T , or MCDK cells and the subcellular localization of SCP1 was analyzed using immunofluorescence assay . ( B ) HeLa cells were transfected and the subcellular localization of SCP1 was analyzed using immunofluorescence assay . ( C ) SCP1 , SCP2 , and SCP3 , but not SCP4 , were localized on the plasma membrane . HeLa cells were transfected with GFP-SCP1/SCP2/SCP3/SCP4 , respectively . ( D ) Membrane-localized SCP1 was photobleached ( as showed in the red frame ) , and the fluorescence was recovered for 3 min as indicated . ( E ) The SCP1 membrane distribution was not affected by the disruption of the Golgi . HeLa cells were transfected with GFP-Golgi and mCherry-SCP1 , respectively , and treated with DMSO or brefeldin A ( BFA; 5 μg/ml ) for 6 h . DOI: http://dx . doi . org/10 . 7554/eLife . 22058 . 004 To further dissect the molecular basis of the membrane-dependent localization of SCP1 , serial deletion mutants ( Figure 1D ) were generated and their subcellular localizations were detected by immunostaining . As shown in Figure 1E , the SCP1 mutants containing the residues from 31 to 55 were able to target GFP to the plasma membrane , suggesting that the protein region of SCP1 from residues 31 to 55 is essential for targeting SCP1 to the plasma membrane . We next determined the underlying mechanism of SCP1 membrane localization . Because we could not find any transmembrane region within SCP1 , we examined whether its membrane localization is mediated by lipid modifications . To this end , we examined the effects of various lipid modification inhibitors , including the farnesyltransferase inhibitor FTI-227 , the prenyl transferase inhibitor GGTI , and the palmitoylation inhibitor 2-bromopalmitate ( 2BP ) on SCP1 membrane localization ( Rowinsky et al . , 1999 ) . We did not detect any effect of FTI-227 and GGTI on SCP1 membrane localization ( Figure 2—figure supplement 1A ) . However , FTI-227 and GGTI treatments dramatically blocked the membrane localization of H-Ras in control experiments ( Figure 2—figure supplement 1A ) . In sharp contrast , treatment with 2BP for 4 h resulted in a remarkable reduction of membrane-localized SCP1 and significantly increased cytosolic and nuclear SCP1 ( Figure 2A and Figure 2—figure supplement 1B ) . This result was confirmed by cell fractionation assay , which showed that the membrane-associated SCP1 level was markedly reduced upon 2BP treatment . In parallel , the cytosolic and nuclear distribution of SCP1 increased ( Figure 2B ) . Similar results were obtained from SCP2- and SCP3-transfected cells ( Figure 2—figure supplement 1C ) . Interestingly , a small amount of SCP1 was also found on the Golgi membrane after 8 h of 2BP treatment ( Figure 2—figure supplement 1D ) . These data indicate that membrane recruitment of SCP1 is mediated by palmitoylation , but not farnesylation or prenylation , which offers direct evidence for the previous report that identified SCP1 as a palmitoylation substrate ( Martin and Cravatt , 2009 ) . 10 . 7554/eLife . 22058 . 005Figure 2 . SCP1 was palmitoylated ( A ) Palmitoylation inhibitor 2-bromopalmitate ( 2BP ) blocked the SCP1 membrane localization . HeLa cells were transfected and treated with 2BP ( 10 μM ) for 4 , 8 , or 16 h or DMSO as a control . The subcellular localization of SCP1 was detected using immunofluorescence assay . ( B ) HEK293T cells were treated with DMSO or 2BP ( 10 μM ) for 6 h and the subcellular location of SCP1 was detected using western blotting . ( C ) Potential palmitolylation sites Cys44 and Cys45 of SCP1 were evolutionarily conserved . Amino acids 33–55 of SCP1 in different species , ranging from Caenorhabditis elegans to Homo sapiens , 30–53 of SCP2 in H . sapiens , and 33–59 of SCP3 in H . sapiens are shown . ( D ) HeLa cells were transfected with WT-SCP1 , C44S-SCP1 , C45S-SCP1 , C47S-SCP1 , and C44/45S ( 2S ) -SCP1 for 24 h . The subcellular localizations of WT-SCP1 and its mutants were detected using immunofluorescence assay . ( E ) HEK293T cells were transfected with WT-SCP1 , C44S-SCP1 , C45S-SCP1 , and C44/45S ( 2S ) -SCP1 for 24 h and cell fractions of were analyzed using western blotting . ( F ) FLAG-SCP1 was expressed in HEK293T cells , immunoprecipitated , and palmitoylation was detected using the acyl–biotin exchange ( ABE ) assay . ( G ) Palmitoylation of endogenous SCP1 in HEK293T cells was detected using the ABE assay . ( H ) FLAG-SCP1 was expressed in HEK293T cells for 24 h and treated with 2BP ( 10 μM ) or palmostatin B ( 50 μM ) for 12 h . Palmitoylation of SCP1 was detected using pan-palmitoylation antibody . ( I ) and ( J ) Identification of palmitoylation sites using the ABE assay ( I ) and the [3H] palmitate incorporation assay ( J ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22058 . 00510 . 7554/eLife . 22058 . 006Figure 2—figure supplement 1 . SCP1 membrane localization depends on its palmitoylation . ( A ) The membrane localization of SCP1 was not affected by farnesyltransferase or prenyltransferase inhibitor . The transfected HeLa cells were treated with DMSO , FTI-277 ( 10 μM ) , or GGTI ( 15 μM ) for 8 h . ( B ) The membrane localization of SCP1 was blocked by palmitoyltransferase inhibitor . GFP-Ras and/or GFP-SCP1 were co-expressed in HeLa cells for 24 h . The transfected cells were treated with DMSO or 2BP ( 10 μM ) for 8 h . ( C ) The membrane localizations of SCP2 and SCP3 were blocked by palmitoyltransferase inhibitor . HeLa cells were transfected with GFP-SCP2/SCP3 for 24 h . ( D ) The newly synthesized SCP1 was transported to the Golgi without palmitoylation and then translocated to the plasma membrane by palmitoylation . HeLa cells were transfected with GFP-Golgi and mCherry-SCP1 for 24 h and treated with 2BP ( 10 μM ) or cycloheximide ( 15 μg/ml ) for 8 h . ( E ) The working model for SCP1 palmitoylation and cell membrane location is shown . ( F ) Amino acid residues from 31 to 55 are important for SCP1 palmitoylation and cell membrane localization . HeLa cells were transfected with the truncated mutant of GFP-SCP1 31–55 for 24 h and treated with DMSO or 2BP ( 10 μM ) for 8 h . DOI: http://dx . doi . org/10 . 7554/eLife . 22058 . 006 It has been reported that palmitoylated proteins can be recycled from the plasma membrane to the Golgi ( Resh , 2006 ) . Therefore , we tested whether the nucleus- or Golgi-localized SCP1 was newly synthesized and recycled to the nucleus or Golgi from the plasma membrane . SCP1 localization was monitored in transfected HeLa cells treated with cycloheximide ( CHX ) for 8 h to block new protein synthesis in the presence or absence of 2BP , which demonstrated that CHX had little effect on the membrane localization of SCP1 ( Figure 2—figure supplement 1D ) . However , in the presence of 2BP , CHX treatment significantly reduced the distribution of SCP1 on the Golgi membrane , but minimally affected its nuclear accumulation ( Figure 2—figure supplement 1D ) . These data indicate that the newly synthesized SCP1 protein is translocated to the Golgi without being palmitoylated and then targeted to the plasma membrane by palmitoylation . The pool of SCP1 at the plasma membrane was constitutively depalmitoylated and the 2BP treatment prevented its re-palmitoylation , leading to the redistribution of depalmitoylated SCP1 in cytosolic and nuclear compartments ( Figure 2—figure supplement 1E ) . These data suggest that palmitoylation is required for SCP1 localization at the plasma membrane , while depalmitoylation allows SCP1 to be recycled from the cell surface to the nucleus . As mentioned above , this is a reversible modification as palmitoylation involves the attachment of a thioester chain to cysteine residues , and this controls the transient membrane targeting of peripheral membrane proteins ( so-called S-palmitoylation ) ( el-Husseini et al . , 2002 ) . We found that SCP1 contains an evolutionarily conserved di-cysteine motif at its membrane-associated region , which constitutes a potential palmitoylation site predicted by ExPASy or CSS-Palm software ( Figure 2C ) . The membrane-localized N-terminal fragment of SCP1 containing putative S-palmitoylation sites is sensitive to 2BP treatment ( Figure 2—figure supplement 1F ) . Therefore , we examined the functional role of these conserved cysteines for targeting SCP1 to the membrane by mutating these cysteines into serines . As shown in Figure 2D , the individual mutations of Cys44 ( C44 ) , Cys45 ( C45 ) , or Cys47 ( C47 ) to Ser had minimal effects on the membrane distribution of SCP1 . However , the simultaneous mutation of both Cys44 and Cys45 into serines ( 2S ) resulted in a dramatic reduction of membrane-associated SCP1 by 90% ( Figure 2D ) . The C45S and C44/45S mutations also significantly increased the levels of SCP1 in the cytosol and nucleus ( Figure 2E ) . These data indicate that SCP1 binds to the plasma membrane through the conserved cysteines of Cys44 and Cys45 . To test whether SCP1 was palmitoylated in vivo , a cysteine accessibility assay ( also named the acyl–biotin exchange [ABE] assay ) was used ( Noritake et al . , 2009 ) . We found that both overexpressed and endogenous SCP1 were palmitoylated in HEK293T cells ( Figure 2F and G ) . It is well established that a distinguishing feature of S-palmitoylation is its reversibility . Since both the forward and reverse reactions take place in a reversible fashion , we also validated the dynamic palmitoylation of SCP1 using both the palmitoylation inhibitor ( 2BP ) and the depalmitoylation inhibitor ( palmostatin B ) ( Dekker et al . , 2010 ) . As expected , the palmitoylation of SCP1 can be regulated by both 2BP and palmostatin B in an opposing fashion , as shown by the decreased palmitoylation of SCP1 by 2BP versus the increased palmitoylation of SCP1 by palmostatin B via using pan Anti-palmitoylation Antibody ( Fang et al . , 2016 ) ( Figure 2H ) , suggesting a dynamic balance of SCP1 inside and outside of the plasma membrane . The residues required for palmitoylation were also examined . As shown in Figure 2I , all of the C44S , C45S , and C47S mutations led to a reduction of palmitoylation to different extents , while the SCP1 C44/45S double mutant almost completely abolished palmitoylation . The palmitoylation level of the C45S mutant was much less than that of the C44S or C47S mutants , which matched these mutants to their role in SCP1 membrane association . The palmitoylation of wild-type ( WT ) SCP1 was also confirmed by a [3H] palmitic acid incorporation assay . The 2S mutant showed a remarkably reduced level of [3H] palmitic acid incorporation ( Figure 2J ) , suggesting that such cysteines are the major palmitoylation sites of SCP1 . Taken together , these results indicate that SCP1 is palmitoylated at its N-terminus , which accounts for the targeting SCP1 of to the plasma membrane . To further clarify the function of SCP1 , we generated SCP1-knockout ( KO ) mice ( Figure 3—figure supplement 1A and B ) . The KO efficiency of the Ctdsp1 gene was confirmed by gene sequencing , polymerase chain reaction ( PCR ) , and western blotting ( Figure 3—figure supplement 1C–E ) . The SCP1-KO mice gave birth at a typical median ratio and developed normally ( data not shown ) . Because palmitoylation plays different roles in neurogenesis and is tightly controlled by angiogenesis in that growth factors such as VEGF perform multiple functions ( Sun et al . , 2003; Lee et al . , 2007 ) , we examined whether SCP1 is involved in embryonic vasculogenesis using SCP1-KO mice . To this end , we analyzed in vivo retinal vasculogenesis at postnatal day 5 ( P5 ) using whole-mount and imaging techniques . We found that the vascular areas were larger in the SCP1-KO mice than those of WT mice ( Figure 3A ) . We also found significantly increased vascular sprouts and branch points in SCP1-KO mice ( Figure 3B ) . We further examined the effect of SCP1 on postnatal angiogenesis using a hind limb ischemia model and found a significantly promoted hind limb flow recovery in SCP1-KO mice ( Figure 3C ) . Because angiogenesis is essential for tumorigenesis ( Testa and Bellacosa , 2001 ) , we also explored the role of SCP1 on tumor growth by injecting Lewis lung carcinoma cells ( LLCs ) into WT or SCP1-KO mice ( O'Reilly et al . , 1994 ) . Our data showed that SCP1 KO significantly promoted tumor growth ( Figure 3D and Figure 3—figure supplement 1F ) . Accordingly , we found that SCP1 deficiency significantly enhanced angiogenesis in tumors ( Figure 3E ) . Thus , our data indicate that SCP1 is a negative regulator of angiogenesis . 10 . 7554/eLife . 22058 . 007Figure 3 . SCP1 knockout promoted angiogenesis ( A ) SCP1 deletion impaired the development of retinal angiogenesis . Retinas of postnatal day 5 were isolated from littermates of wild-type ( WT; n = 8 ) or SCP1-knockout ( KO ) mice ( n = 7 ) and stained with Isolectin B4 . Quantification of vessel length was measured , and the rate of vessel length/retina length was calculated . **p<0 . 01 . ( B ) Loss of SCP1 reduced the branching of the vessels . The branching of vessels was counted . ***p<0 . 001 . ( C ) SCP1 deficiency promoted the recovery of hind limb ischemia . The laser Doppler blood flowmetry ratio was significantly higher in SCP1-KO mice ( n = 6 ) than in WT mice ( n = 8 ) . ***p<0 . 001 , **p<0 . 01 . ( D ) SCP1 KO promoted Lewis lung carcinoma cell ( LLC ) tumor bearing in C57 mouse . 2 × 105 LLC cells were injected into SCP1-WT or -knockdown littermates . The diameter of the tumor was measured every 2 days , and the volume of the tumor was calculated . ***p<0 . 001 , **p<0 . 01 , *p<0 . 05 . ( E ) Angiogenesis was promoted in SCP1-KO mice . The angiogenesis in tumors was analyzed using immunohistochemistry by CD31 staining . ( F ) SCP1 deficiency promoted angiogenesis in an AKT-dependent manner . Segments ( 1 mm in length ) of the aorta from SCP1-WT ( n = 8 ) or SCP1-KO ( n = 6 ) mice were embedded in Matrigel and treated with DMSO or AKT inhibitor ( MK2206 , 2 nM ) for 6 days . Sprouting was observed and photographed by microscopy . The vascular area of each group was measured using Image J . ***p<0 . 001 , **p<0 . 01 , *p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 22058 . 00710 . 7554/eLife . 22058 . 008Figure 3—source data 1 . SCP1 knockout promoted angiogenesis ( A ) Retinas of postnatal day 5 mice were isolated from littermates of wild-type ( n = 8 ) or SCP1-knockout mice ( n = 7 ) and stained with Isolectin B4 . Quantification of vessel length was obtained , and the ratio of vessel length/-retina length was calculated . ( B ) The branching of vessels was counted . ( C ) The laser Doppler blood flowmetry ratio was significantly higher in SCP1-knockout mice ( n = 6 ) than in wild-type mice ( n = 8 ) . ( D ) 2 × 105 Lewis lung carcinoma cells were injected into SCP1-wild-type or -knockdown littermates . The diameter of the tumor was measured every 2 days , and the volume of the tumor was calculated . ( E ) The angiogenesis in tumors was analyzed using immunohistochemistry by CD31 staining . ( F ) Segments ( 1 mm in length ) of the aorta from SCP1-wild-type ( n = 8 ) or -knockout ( n = 6 ) mice were embedded in Matrigel and treated with DMSO or AKT inhibitor ( MK2206 , 2 nM ) for 6 days . Sprouting was observed and photographed by microscopy . The vascular area of each group was measured using Image J . DOI: http://dx . doi . org/10 . 7554/eLife . 22058 . 00810 . 7554/eLife . 22058 . 009Figure 3—figure supplement 1 . Generation and validation of SCP1-knockout mice . ( A ) Constructs of the Cas 9/RNA system: DR , direct repeat to separate signal; NLS , nuclear localization signal . ( B ) Schematic overview of the strategy used to generate the SCP1-knockout mice . The sgRNA coding sequence is underlined , capitalized , and labeled in red . The protospacer-adjacent motif ( PAM ) sequence is labeled in green . ( C ) The DNA sequences of the Ctdsp1 genomic loci in the founders . ( D ) The genotyping of Ctdsp1+/+ , Ctdsp1+/– , and Ctdsp1–/– mice . The ratio of Ctdsp1+/– offspring is listed . ( E ) SCP1 was analyzed by IP ( Immunoprecipitation ) and western blotting . ( F ) The Lewis lung carcinoma celltumor-bearing SCP1 wild-type and -knockout mice in Figure 3D are photographed . DOI: http://dx . doi . org/10 . 7554/eLife . 22058 . 009 AKT is intimately involved in the regulation of angiogenesis . VEGF can regulate angiogenesis through the PI3K/AKT pathway ( Shiojima and Walsh , 2002 ) . Next , we examined whether the effect of SCP1 on angiogenesis depends on AKT activation . To this end , we performed an aortic ring assay by in vitro culturing thoracic aorta from WT or SCP1-KO littermates ( Kitamura et al . , 2008 ) . We found that the aortic rings from SCP1-KO mice showed an increased capability of angiogenesis , evidenced by enlarged areas of capillary sprouting ( Figure 3F ) . MK2206 was reported as an allosteric AKT inhibitor that enhanced the antitumor efficacy of standard chemotherapeutic agents in vitro and in vivo ( Hirai et al . , 2010 ) . Treatment with MK2206 reversed the above-mentioned effects in SCP1-KO mice ( Figure 3F ) . These results indicate that SCP1 negatively regulates angiogenesis in an AKT-dependent manner . Because the endothelial cell is the major cell type involved in angiogenesis , we investigated whether SCP1 deficiency impairs the function of endothelial cells by examining the functional role of SCP1 in HUVECs ( a human endothelial cell line from umbilical veins ) . To this end , we first examined whether SCP1 regulates AKT phosphorylation in HUVECs . Our data showed that SCP1 depletion significantly promoted VEGF-induced AKT activation in HUVECs ( Figure 4A ) . Moreover , we found that SCP1 overexpression markedly inhibited tubule formation and cell migration ( Figure 4B and Figure 4—figure supplement 1A ) . Importantly , the inhibitory effect of SCP1 was rescued by the coexpression of the AKT-S473D mutant ( Figure 4B and Figure 4—figure supplement 1A ) . In parallel , SCP1 depletion markedly increased angiogenesis and endothelial cell migration , which could be reversed by AKT inhibitors ( Figure 4C and D ) . These data indicate that SCP1 suppresses angiogenesis in an AKT phosphorylation-dependent fashion . 10 . 7554/eLife . 22058 . 010Figure 4 . SCP1 inhibited AKT-mediated angiogenesis ( A ) SCP1 deletion promoted VEGF-induced AKT activation in HUVECs . HUVECs were transfected with siNC ( Small Interfering RNA for Normal Control ) or siSCP1 ( Small Interfering RNA for SCP1 ) for 72 h and stimulated with VEGF ( 100 ng/ml ) as indicated after starvation for 8 h . ( B ) SCP1 impaired HUVEC tube formation through AKT . HUVECs were overexpressed with SCP1 with or without AKT-S473D . The cells were placed in plates coated with Matrigel and tubular structures were photographed after 6 h . The tube lengths were measured in each field . *p<0 . 05 . ( C ) SCP1 depletion inhibited the tube formation of HUVECs through AKT . HUVECs were transfected with siSCP1 and treated with or without AKT inhibitor ( AKTi; MK2206 , 2 nM ) for 5 days as indicated . The tube lengths were measured in each field . **p<0 . 01 , *p<0 . 05 . ( D ) SCP1 deletion promoted HUVEC migration through AKT . Cell migration was detected using a wound healing assay . HUVECs were transfected and treated with or without AKTi ( MK2206 , 2 nM ) . The migration cell number in each field was calculated . **p<0 . 01 , *p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 22058 . 01010 . 7554/eLife . 22058 . 011Figure 4—source data 1 . SCP1 inhibited AKT-mediated angiogenesis ( A ) HUVECs were overexpressed with SCP1 with or without AKT-S473D . The cells were placed in plates coated with Matrigel and tubular structures were photographed after 6 h . The tube lengths were measured in each field . ( B ) HUVECs were transfected with siSCP1 and treated with or without AKT inhibitor ( AKTi; MK2206 , 2 nM ) for 5 days as indicated . The tube lengths were measured in each field . ( C ) Cell migration was detected using a wound healing assay . HUVECs were transfected and treated with or without AKTi ( MK2206 , 2 nM ) . The migration cell number in each field was calculated . DOI: http://dx . doi . org/10 . 7554/eLife . 22058 . 01110 . 7554/eLife . 22058 . 012Figure 4—figure supplement 1 . SCP1 inhibits HUVEC migration . ( A ) SCP1 inhibited HUVEC migration . Cell migration was detected using a wound healing assay . Values represent mean ± SD ( n = 3 ) . **p<0 . 01 , *p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 22058 . 012 To test whether SCP1 can directly regulate AKT , we expressed a constitutively active form of AKT , herein referred as Myr-AKT , in which the N-terminus is fused with a myristoylation signal and localized at the plasma membrane ( del Peso et al . , 1997 ) . We co-expressed SCP1 with Myr-AKT in HEK293T cells and found that the co-expression of SCP1 , but not its catalytically inactive mutant DN-SCP1 ( in which Asp96 is substituted to Glu ) ( Wrighton et al . , 2006 ) , reduced the phosphorylation level of Myr-AKT at Ser473 by 80% ( Figure 5A ) . To examine whether SCP1 can directly dephosphorylate AKT , Myr-AKT was incubated with GST , GST-WT-SCP1 , or GST-DN-SCP1 purified from Escherichia coli . We found that WT-SCP1 , but not DN-SCP1 , markedly dephosphorylated AKT at Ser473 , accompanied by a lesser extent of AKT dephosphorylation at Thr308 ( Figure 5B ) . 10 . 7554/eLife . 22058 . 013Figure 5 . SCP1 dephosphorylated AKT ( A ) Wild-type ( WT ) -SCP1 dephosphorylated AKT Ser473 . Myr-AKT was co-expressed with vector , WT-SCP1 , or DN-SCP1 . The phosphorylations of p-Ser473-AKT , p-Thr308-AKT , and p-Thr450-AKT was analyzed using western blotting . The relative phosphorylations of p-Thr308-AKT and p-Thr450-AKT are displayed in the form of a histogram . ( B ) WT-SCP1 dephosphorylated AKT in vitro . HA-Myr-AKT was immunoprecipitated from HEK293T cells and incubated with purified GST , GST-WT-SCP1 , or GST-DN-SCP1 for 30 min . The phosphorylations of p-Ser473-AKT and p-Thr308-AKT were analyzed using western blotting . ( C ) WT-SCP1 dephosphorylated AKT in HeLa cells . HeLa cells were transfected with GFP-SCP1 for 24 h . The phosphorylation of p-Ser473-AKT and total AKT was detected using immunofluorescence assay . ( D ) SCP1 knockdown promoted AKT Ser473 phosphorylation in HeLa cells . Control or Ctdsp1 shRNA was transfected into HeLa cells for 72 h . The phosphorylation of p-Ser473-AKT and total AKT was detected using immunofluorescence assay . ( E ) SCP1 knockdown promoted EGF-induced AKT activity . H1299 cells were transfected with control or Ctdsp1 siRNA for 72 h . The cells were stimulated with EGF ( 100 ng/ml ) as indicated after 8 h of starvation , and phosphorylation of AKT was detected using immunofluorescence assay . ( F ) SCP1 depletion promoted insulin-stimulated AKT activation . Ctdsp1+/+ or Ctdsp1–/–MEFs ( mouse embryonic fibroblast ) were stimulated with insulin ( 1 mM ) as indicated after 6 h of starvation . ( G ) WT-SCP1 decreased the AKT kinase activity . AKT was transfected into HEK293T cells with vector , WT-SCP1 , or DN-SCP1 , immunoprecipitated , and incubated with GST-GSK3β . The phosphorylation of GSK3β was measured using western blotting . ( H ) Endogenous AKT interacted with endogenous SCP1 . Endogenous SCP1 was immunoprecipitated using an anti-SCP1 antibody and the associated AKT was detected using an anti-AKT antibody . ( I ) The interaction of SCP1 with WT or myristoylated AKT1 is independent of its phosphatase activity . ( J ) Purified GST , GST-WT-SCP1 , and GST-DN-SCP1 were incubated with cell lysates overexpressing AKT . The interaction was detected using western blotting . DOI: http://dx . doi . org/10 . 7554/eLife . 22058 . 01310 . 7554/eLife . 22058 . 014Figure 5—figure supplement 1 . SCP1 negatively regulates AKT activation on membrane . ( A ) SCP1 overexpression strongly suppressed EGF-induced AKT membrane activation . ( B ) SCP1 knockdown promoted insulin-stimulated AKT activation in H1299 cells . H1299 cells were transfected with control or Ctdsp1 siRNA for 72 h . The cells were stimulated with insulin ( 1 mM ) as indicated after 6 h of starvation , and phosphorylation of AKT was detected using western blotting . ( C ) The knockdown efficiency of SCP1 was measured at the mRNA or protein level . ( D ) SCP1 or PHLPP knockdown promoted EGF-stimulated AKT activation in MEF cells . MEF cells were transfected with control or Ctdsp1 siRNA or PHLPP siRNA or Ctdsp1/PHLPP siRNA for 72 h . The cells were stimulated with EGF ( 100 ng/ml ) as indicated after 8 h of starvation , and phosphorylation of AKT was detected using western blotting . ( E ) The knockdown efficiencies of SCP1 and PHLPP were measured at the mRNA level . ( F ) PHLPP rescued the EGF-stimulated AKT activation after SCP1 knockdown in HeLa cells . ( G ) SCP1 interacted with endogenous AKT . ( H ) SCP1 interacted with wild-type AKT and its mutants . DOI: http://dx . doi . org/10 . 7554/eLife . 22058 . 014 We next examined whether SCP1 overexpression could result in AKT dephosphorylation by immunostaining assay . We found that overexpression of SCP1 significantly diminished the phosphorylation of endogenous AKT at Ser473 without affecting total AKT levels ( Figure 5C ) . In addition , SCP1 knockdown by shRNA markedly increased AKT phosphorylation at Ser473 ( Figure 5D ) . AKT can be activated by various upstream signals , such as growth factors , hormones , and cytokines ( Rodon et al . , 2013 ) . Therefore , we tested whether SCP1 could block AKT activation under such conditions . We found that EGF treatment markedly induced the membrane distribution of AKT and AKT phosphorylation at Ser473 in HeLa cells . In addition , SCP1 expression suppressed EGF-induced AKT phosphorylation at Ser473 without affecting AKT distribution at the plasma membrane ( Figure 5—figure supplement 1A ) . Meanwhile , both EGF- and insulin-induced AKT phosphorylation at Ser473 in H1299 cells can be significantly elevated by SCP1 depletion ( Figure 5E and Figure 5—figure supplement 1B ) . The knockdown efficiency of SCP1 was measured at the mRNA and protein level ( Figure 5—figure supplement 1C ) . We also found that SCP1 depletion significantly enhanced the expression levels of AKT-phosphate substrates as examined by anti-AKT-phosphate substrate antibody ( Figure 5E ) . Similar results were obtained using MEF cells derived from SCP1-WT and -KO littermates under the insulin treatment ( Figure 5F ) . Phosphatase PHLPP has been implicated in cleaving pS473 from AKT ( Cailliau et al . , 2015; Qiao et al . , 2007 ) . We found that SCP1 or PHLPP depletion could promote AKT phosphorylation at Ser473 . Furthermore , double knockdown of SCP1 and PHLPP could exaggerate this effect ( Figure 5—figure supplement 1D and E ) . We also overexpressed PHLPP1 when SCP1 was disabled and found that PHLPP1 can still decrease AKT phosphorylation at Ser473 ( Figure 5—figure supplement 1F ) . Based on these data , we conclude that the phosphatase function of SCP1 that dephosphorylates AKT is independent of PHLPP activity . Taken together , our results indicate that SCP1 is a phosphatase that dephosphorylates AKT at Ser473 . The phosphorylation of Ser473 is critical for AKT activation . Therefore , we examined whether SCP1 could regulate AKT kinase activity using an in vitro kinase assay . We found that SCP1 overexpression markedly reduced AKT activity ( Figure 5G ) . Thus , our data indicate that dephosphorylation of AKT by SCP1 significantly inhibits AKT kinase activity . Next , we investigated whether SCP1 could interact with AKT in vivo . To this end , SCP1 was immunoprecipitated from HEK293T cells and the association of endogenous AKT and SCP1 was detected by immunoblotting . We found that endogenous AKT could be co-immunoprecipitated by both overexpressed and endogenous SCP1 ( Figure 5H and Figure 5—figure supplement 1G ) , indicating that SCP1 associates with AKT in cells . In addition , we found that membrane-localized Myr-AKT had a stronger affinity to SCP1 than WT AKT ( Figure 5I ) . The interaction between SCP1 and AKT was independent of SCP1 phosphatase activity , since dominant-negative SCP1 still showed an association with AKT ( Figure 5I ) . Consistently , AKT could directly bind to the purified SCP1 in vitro , while mutations of AKT at Thr308 , Thr450 , or Ser473 had little effect on their interaction , suggesting that the direct interaction between SCP1 and AKT is independent of AKT phosphorylation ( Figure 5J and Figure 5—figure supplement 1H ) . Then , we examined whether SCP1 palmitoylation is involved in its inhibitory effect on AKT signaling . As shown in Figure 6A , the mutant ( SCP1 2S ) that cannot be palmitoylated significantly reduced the capacity for dephosphorylating AKT at Ser473 in cells . Accordingly , the 2S mutant abolished the inhibitory capability of SCP1 on cell growth ( Figure 6B and Figure 6—figure supplement 1A ) . Meanwhile , 2BP treatment inhibited the ability of SCP1 to dephosphorylate AKT and increased the endogenous level of phosphorylated AKT Ser473 ( Figure 6C ) . These data indicate that palmitoylation is pivotal for the suppressive effect of SCP1 on cell growth . 10 . 7554/eLife . 22058 . 015Figure 6 . Palmitoylation was required for SCP1-mediated AKT inhibition ( A ) Palmitoylation was required for SCP1 to dephosphrylate AKT Ser473 . HEK293T cells were transfected with WT-SCP1 , DN-SCP1 , and 2S-SCP1 for 24 h and treated with DMSO or 2-bromopalmitate ( 2BP; 10 μM ) for 6 h . The phosphorylations of p-Ser473-AKT and p-Thr450-AKT were detected using western blotting . ( B ) Palmitoylation of SCP1 at C44 and C45 was required for its suppression of cell proliferation . Values represent mean ± SD ( n = 3 ) ( C ) Palmitoylation inhibition increased the phosphorylation levels of endogenous AKT Ser473 . HEK293T cells and H1299 cells were treated with DMSO or 2BP ( 10 μM ) for 6 h . The phosphorylation of p-Ser473-AKT was detected using western blotting . ( D ) Depalmitoylation blocked the interaction between SCP1 and AKT . HEK293T cells were transfected with Myr-AKT-HA and FLAG-SCP1 or FLAG-SCP1-2S for 24 h and treated with DMSO or 2BP ( 10 μM ) for 6 h . The interaction between AKT and SCP1 or SCP1-2S was detected using western blotting . ( E ) SCP1 blocked HUVEC migration in a palmitoylation-dependent manner . HUVECs were transfected with vector , WT-SCP1 , DN-SCP1 , and 2S-SCP1 , respectively . Cell migration was detected using a wound healing assay . ( F ) SCP1 blocked HUVEC tube formation in a palmitoylation-dependent manner . HUVECs were transfected with vector , WT-SCP1 , DN-SCP1 , and 2S-SCP1 , respectively . Tube formation was detected using a tube formation assay . ( G ) The working model for SCP1 palmitoylation and cell membrane localization is shown . DOI: http://dx . doi . org/10 . 7554/eLife . 22058 . 01510 . 7554/eLife . 22058 . 016Figure 6—figure supplement 1 . SCP1 suppresses AKT-mediated biological functions . ( A ) The relative cell survival in Figure 6B was quantified . **p<0 . 01 , *p<0 . 05 . ( B ) The phosphatase activity of SCP1 was not affected by its palmitoylation status . The phosphatase activity was measured using a pNPP ( p-nitrophenyl-phosphate ) assay . ( C ) The migration was calculated , as shown in Figure 6E . **p<0 . 01 , *p<0 . 05 . ( D ) The tube length in Figure 6F was measured . **p<0 . 01 , *p<0 . 05 . ( E ) SCP1 impaired H1299 tumor bearing in nude mice in a palmitoylation-dependent manner . H1299 cells were infected with lentivirus with vector , WT-SCP1 , DN-SCP1 , or 2S-SCP1 , and they were subcutaneously injected into nude mice ( 1 × 107 per mouse ) . The tumor was photographed after 42 days . WT: wild-type . DOI: http://dx . doi . org/10 . 7554/eLife . 22058 . 016 Next , we examined whether palmitoylation is required for the phosphatase activity of SCP1 . To this end , HA-tagged WT-SCP1 , DN-SCP1 , and 2S-SCP1 mutants were expressed , respectively , in HEK293T cells and immunoprecipitated with an anti-HA antibody . The phosphatase activities of the above-mentioned mutants were measured by a pNPP phosphatase ( p-nitrophenyl-phosphate ) assay ( Wang et al . , 2016 ) . We found that mutations at SCP1 palmitoylation sites did not affect the phosphatase activity of SCP1 ( Figure 6—figure supplement 1B ) . The effect of palmitoylation on the interaction between SCP1 and AKT was also examined . As shown in Figure 6D , both treatment with 2BP and mutations at palmitoylation sites ( F-SCP1-2S ) decreased the interaction between SCP1 and the constitutively active form of AKT ( Myr-AKT-HA ) . Moreover , the mutant 2S-SCP1 was unable to block the migration of HUVECs in a wound healing assay ( Figure 6E and Figure 6—figure supplement 1C ) , suggesting that palmitoylation of SCP1 affects the functional role of endothelial cells that accounts for angiogenesis . Similar results were obtained for tubule formation using a Matrigel angiogenesis assay ( Figure 6F and Figure 6—figure supplement 1D ) . We also found that WT-SCP1 , but not DN-SCP1 or 2S-SCP1 , suppressed tumorigenesis ( Figure 6—figure supplement 1E ) . In summary , these data indicate that palmitoylation is essential for SCP1 to dephosphorylate AKT via their interactions , which plays a crucial role in angiogenesis and tumor growth .
SCP1 was initially identified as a nuclear phosphatase that dephosphorylates RNA pol ll ( Yeo et al . , 2003; Zhang et al . , 2006 ) . In this study , we unexpectedly found that SCP1 , SCP2 , and SCP3 are mainly localized at the plasma membrane in different cell types in a steady state ( Figure 1 and Figure 1—figure supplement 1 ) . Such a novel phenotype observed from SCP1 raises much interest because only a few serine/threonine phosphatases are localized at the plasma membranes , although approximately 30 serine/threonine and 107 tyrosine protein phosphatases have been reported ( Shi , 2009; Alonso et al . , 2004 ) . Importantly , all of these membrane-located phosphatases have been demonstrated to play crucial roles in various biological processes . For instance , PTEN can function as a protein phosphatase that localizes at the plasma membrane to dephosphorylate membrane proteins and is intimately involved in tumor progression by inhibiting oncogenic signaling ( Wu et al . , 2000 ) . PH domain-containing family members such as PHLPP1 and PHLPP2 can mediate AKT signaling by recruiting proteins to the plasma membrane through its interaction with PIP2 ( Brognard et al . , 2007 ) . Another study indicated that PHLPP1 interacts with scribble via its PDZ ( Post synaptic density protein ( PSD95 ) , Drosophila disc large tumor suppressor ( Dlg1 ) , and Zonula occludens-1 protein ( zo-1 ) ) domain and thereby mediates its localization at the plasma membrane , which negatively regulates AKT-mediated tumorigenic signals ( Hung and Link , 2011 ) . In order to explore the mechanisms underlying the membrane distribution of SCP1 , we used both in vitro and in vivo animal models and consequently found that SCPs were palmitoylated at a conserved cysteine motif within its N-terminus ( Figure 2 and Figure 2—figure supplement 1 ) . S-palmitoylation is a lipid post-translational modification involving the covalent attachment of fatty acid palmitate to cysteine residues via a thioester bond . So far , it is the only reversible lipid modification that has been identified ( Mumby , 1997 ) . Importantly , palmitoylated proteins were found to play very important roles in multiple biological processes , such as synaptic plasticity during neuronal development ( Fukata and Fukata , 2010; El-Husseini et al . , 2002; Kutzleb et al . , 1998; Arstikaitis et al . , 2008 ) . In addition , various membrane-localized proteins that are crucial in the maintenance of protein structure and function are palmitoylated ( Rocks et al . , 2010 ) . Unlike other lipid modifications , protein palmitoylation is highly dynamic , and cycles of palmitoylation and depalmitoylation can regulate protein function and subcellular localization ( Kang et al . , 2008; Fukata and Fukata , 2010 ) . Due to its special nature , palmitoylation may be particularly important for modulating protein function during cycles of activation and deactivation through its effect on a protein's affinity for membranes , subcellular localization , and interactions with other proteins . Moreover , reversible palmitoylation of signaling proteins allows proteins to rapidly shuttle between intracellular membrane compartments , since this palmitate cycling can be regulated by different physiological stimuli , which contributes to cellular homeostasis . Thus , the identification of a novel palmitoylated protein like SCP1 could be critical for clarifying special points of cross-talk between molecular signals that are implicated in diverse cellular functions . Based upon our findings , it would be reasonable to assume that the distribution of SCP1 at the plasma membrane via its N-terminal palmitoylation might intensively deal with some uncovered functions of SCP1 . We first looked for direct evidence of SCP1 palmitoylation . From a variety of pharmacological inhibitor tests , substantial evidence confirmed the existence of SCP1 palmitoylation and further clarified a distinct role of palmitoylation , but not of farnesylation nor prenylatin , in the control of the membrane recruitment of SCP1 ( Figure 2 and Figure 2—figure supplement 1 ) . Our results bridge the gap between the SCP1 post-translational modification and biological function observed in a previous study by Martin and Cravatt , who identified SCP1 as a palmitoylation substrate in their search for palmitoylated proteins ( Martin and Cravatt , 2009 ) . Thus , we conclude that SCP1 is a palmitoylated phosphatase . From our data , SCP1 palmitoylation is indeed a reversible process that is dynamically regulated by palmitoylation/depalmitoylation ( Figure 2H ) . Palmitoylation of SCP1 stabilizes the association of SCP1 with membranes , thereby facilitating its vesicular trafficking to the plasma membrane and the subsequent dephosphorylation of AKT by their direct interaction ( Figure 5H and J ) . Meanwhile , depalmitoylation releases depalmitoylated SCP1 into the cytoplasm , allowing its return to the Golgi for another round of palmitoylation or its direct nuclear entry , where it activates different downstream signaling cascades ( Figure 6G ) . Thus , such regulation of the subcellular localization of SCP1 by reversible palmitoylation/depalmitoylation may intimately link it to multiple intracellular signaling pathways . Accordingly , our data prove that the blockade of palmitoylation significantly induces the nuclear localization of SCP1 , indicating that SCP1 can shuttle between the plasma membrane and nucleus , which is consistent with previous reports that SCP1 can dephosphorylate nuclear oncoproteins such as Snail , c-Myc , and Smad ( Knockaert et al . , 2006; Wu et al . , 2009; Wang et al . , 2016 ) , suggesting that such a novel phenotype of SCP1 shuttling observed in our study may be closely correlated with oncogenesis and tumor progression . In addition , palmitoylation has been shown to be regulated by different signaling molecules such as nitric oxide or Src kinases ( Aicart-Ramos et al . , 2011 ) . It will be instructive to investigate whether the SCP1 subcellular localization is regulated under certain physiological or pathological conditions and whether such conditions could also regulate SCP1 palmitoylation-mediated signaling . In order to explore the role of SCP1 palmitoylation in the regulation of biological processes , we conducted a variety of experiments including both in vitro and in vivo studies . Our data indicate that the palmitoylation of SCP1 is essential for AKT dephosphorylation at the plasma membrane , whereas such palmitoylation is not required for its phosphatase activity ( Figure 6 and Figure 6—figure supplement 1 ) . Meanwhile , AKT dephosphorylation by SCP1 requires a direct interaction between them at the plasma membrane ( Figure 6 ) . This is consistent with several lines of evidence showing a tight correlation between AKT activity and its interaction with some specific phosphatases . For example , AKT binds to APPL1 in the endosome , where it regulates the specificity for its substrates ( Schenck et al . , 2008; Saito et al . , 2007 ) . Another AKT phosphatase , PHLPP1 , is also membrane associated via binding to scribble , which is necessary for its dephosphorylation of AKT ( Hung and Link , 2011 ) . Furthermore , we discovered that C44/C45 cysteines are indispensable for SCP1 palmitoylation and subsequent direct interaction with AKT ( Figures 5 and 6 ) . Such an intimate link between membrane-localized SCP1 by palmitoylation on C44/C45 cysteines and the dephosphorylation of AKT is highlighted by embryonic vasculogenesis tests and hind limb ischemia models in SCP1-KO mice with markedly enhanced angiogenesis ( Figure 3 ) . In addition , SCP1 deficiency-promoted angiogenesis is also intimately involved in tumorigenesis ( Figure 3 ) . Using EGF/insulin-induced MEF cells derived from WT and SCP1-KO littermates , our data illustrate a direct interaction and consequent dephosphorylation of AKT by SCP1 ( Figure 5 and Figure 5—figure supplement 1 ) . Since it is relatively easy and straight-forward to examine or quantify membrane-localized SCP1 together with its interaction with AKT , which is the prerequisite for screening potential inhibitors of AKT activity and consequently fine-tuning AKT signaling during tumorigenesis , our study may open up a brand-new avenue for the exquisite regulation of AKT activity that is anticipated to play a critical role in angiogenesis and tumorigenesis . To further clarify the underlying mechanisms of SCP1-mediated AKT inhibition and the resultant suppression of angiogenesis and tumor growth , we point out that AKT Ser473 is the key site for SCP1-mediated dephosphorylation ( Figures 4 , 5 and 6 ) . AKT signaling is regulated through its dephosphorylation on serine/threonine residues in both the cytosol and nucleus ( Stronach et al . , 2011; Testa and Bellacosa , 2001; Lee et al . , 2015 ) . Among these phosphorylation/dephosphorylation sites , Ser473 has drawn much attention recently , as it is assumed to participate in oncogenesis , drug resistance and the anti-apoptosis competence of various types of tumor cells ( Stronach et al . , 2011; Wendel et al . , 2004 ) . The unique phenotype of AKT Ser473 dephosphorylation endowed by its direct interaction with SCP1 at the plasma membrane sheds light on novel therapy strategies for the modulation of aberrant AKT signaling and consequent tumorigenesis ( Figure 3 and Figure 6—figure supplement 1 ) . In this context , we propose a de novo signaling mechanism such that SCP1 is recruited to the membrane by its palmitoylation at the N-terminus , where SCP1 negatively regulates AKT kinase activity , followed by impaired angiogenesis and tumorigenesis . Considering the importance of AKT signaling in angiogenesis and oncogenic development , we present substantial evidence for the identification of SCP1 as a new membrane-localized phosphatase for AKT that carries obvious elements of novelty and interest to the cancer community .
The expression plasmids SCP1 , SCP2 , SCP3 , and SCP4 were the gift from Professor Xinhua Feng ( Zhejiang University ) . SCP1 , SCP2 , SCP3 , and SCP4 were then sub-cloned into pcDNA3 . 1 vectors with a FLAG , HA , or GFP tag . The deletion mutants of SCP1 were cloned into pcDNA3 . 1 vectors with a FLAG or GFP tag . All constructs were confirmed by DNA sequencing . HEK293T ( American Type Culture Collection , RRID: CVCL-0063 ) , HeLa ( American Type Culture Collection , RRID: CVCL-0030 ) , MDCK ( RRID: CVCL-0422 ) , COS7 ( RRID: CVCL-0224 ) , MCF7 ( RRID: CVCL-0031 ) , and DLD1 cells ( RRID: CVCL-0248 ) were cultured in Dulbecco's modified Eagle's medium ( DMEM ) ( Gibco ) . H1299 cells ( RRID: CVCL-0060 ) were cultured in 1640 medium ( Gibco ) . PC-3 cells ( American Type Culture Collection , RRID: CVCL-0035 ) were cultured in F-12K medium ( Gibco ) , supplemented with 10% heat-inactivated fetal bovine serum ( FBS ) at 37°C in 5% CO2 . SaOS-2 ( American Type Culture Collection , RRID: CVCL-0548 ) cells were cultured in McCoy's 5a Medium , supplemented with 15% FBS at 37°C in 5% CO2 . The identification of all of the cell lines has been authenticated by the American Type Culture Collection through STR ( Short Tandem Repeat ) profiling . No mycoplasma contamination was detected in the cultured cells . Transfections were performed using calcium phosphate-DNA coprecipitation for HEK293T cells and SunbioTrans-EZ for HeLa cells ( Shanghai Sunbio Medical Biotechnology Co . , Ltd ) . H1299 or HUVEC cells ( RRID: CVCL-2959 ) were transfected with siRNA oligonucleotides using Lipofectamine 2000 . Immunoprecipitation and western blotting were performed as in our previous report ( Liu et al . , 2010 ) . Briefly , cells were transfected and lysed using 2× RIPA ( Radio Immunoprecipitation Assay ) buffer ( Tris-HCl , pH 7 . 4 [100 mM]; NaCl [300 mM]; 1% NP-40; 2% sodium deoxycholate; 10 mM NaF; and 10 mM Na vanadate ) . The cell lysates were cleared by centrifugation and incubated with 1 μg antibody for 1 h at 4°C followed by incubation with 15 μl protein A and G beads ( Santa Cruz ) for 2 h at 4°C . Immunoprecipitates were subjected to western blot . For western blot analysis , cells were scraped from the dishes into the lysis buffer . A total of 25 mg of total protein was separated by SDS-PAGE and blotted with Pan Anti-palmitoylation ( from Haojie Lu ) , anti-β-actin ( Santa Cruz 47778 , RRID: AB-626632 ) , anti-AKT ( Abcam 1085–1 , RRID: AB-562034 ) , anti-p473-AKT ( CST 4060S , RRID: AB-2315049 ) , anti-p308-AKT ( CST 2965S , RRID: AB-2255933 ) , anti-p450-AKT ( CST 9267S , RRID: AB-823676 ) , anti-FLAG ( Sigma F1804 , RRID: AB-262044 ) , anti-HA ( Santa Cruz 805 , RRID: AB-631618 ) , anti-GFP ( Santa Cruz 9996 , RRID: AB-627695 ) , anti-GSK3 ( Abcam 2199–1 , RRID: AB-991733 ) , anti-pS9-GSK3 ( Abcam 2435–1 , RRID: AB-1267179 ) , and anti-AKT-substrate ( CST 9611S , RRID: AB-330302 ) . 2-BP ( Sigma ) , palmostatin B ( Merck ) , EGF ( Sigma ) , insulin ( ThermoFisher ) , CHX ( Sigma ) , FTI-277 ( Selleckchem ) , GGTI-298 ( Selleckchem ) , BFA ( Sigma ) , and AKT inhibitor angiogenesis MK2206 ( Selleckchem ) were also used in experiments . . Negative control siRNA: 5'-UUCUCCGAACGUGUCACGUTT-3'; hCtdsp1 siRNA: 5’-GCCGGUUGGGUCGAGACCUTT-3’; hPHLPP siRNA: 5’-GGAATCAACTGGTCACATT-3’; Scramble shRNA: 5‘-TTCTCCGAACGTGTCACGTTT-3’;Ctdsp1 shRNA: 5’- AGCGACGTCCTCACGTGGATGAGTTCTAGTGAAGCCA CAGATGTAGAACTCATCCACGTGAGGACGC-3’ . SCP1 was immunoprecipitated from cell lysate after transfection for 24 h . After three washes with reaction buffer ( Wang et al . , 2016 ) , the purified SCP1 was incubated with 5 μl pNPP ( 50 mM , NEB ) in a 50-μl reaction volume adjusted for the conditions of the reaction buffer for 30 min at 30°C . The reactions were then terminated using 1 ml of 1 M NaOH and the absorbance at 405 nm was measured . AKT was immunoprecipitated from HEK293T cells . The immunoprecipitates were washed three times with lysis buffer and once with phosphatase reaction buffer ( 50 mM Tris-HCl , pH 6 . 8 , 150 mM NaCl , 10 mM MgCl2 , pH 8 . 0 , 10 mM DTT ) without phosphatase inhibitors . The immunoprecipitates were then resuspended in the phosphatase reaction buffer and divided into four equal aliquots , three of which were incubated with GST , GST-WT-SCP1 , or GST-DN-SCP1 . After 10 min at 37°C , the dephosphorylation reactions were terminated and the samples were analyzed by western blotting ( Kops et al . , 2002; Peng Liao , 2017 ) . A recombinant GST-GSK3β protein was purified from E . coli using standard protocols . Flag-AKT1 or its mutant forms were expressed in HEK293T cells and purified using anti-Flag M2 beads ( Sigma ) for immunoprecipitation . The kinase assay was performed as described in our previous study ( Liu et al . , 2010 ) . The ABE method was performed as previously described ( Noritake et al . , 2009 ) . Briefly , cells were washed with phosphate-buffered saline containing 10 mM N-ethyl-maleimide ( NEM ) twice and solubilized with 0 . 3 ml of lysis buffer ( LB; 50 mM Tris-HCl , pH 7 . 5 , 5 mM EDTA , and 50 mM NaCl ) containing 1% SDS and 10 mM NEM before harvest . After 30 min of extraction , LB with 1% Triton X-100 and 10 mM NEM was added to a final volume of 1 ml and incubated for 30 min at 4°C . After centrifugation at 12 , 000 g for 10 min , the supernatants were precipitated by the chloroform–methanol ( CM ) method . Precipitated protein was solubilized in 0 . 2 ml SB ( 50 mM Tris-HCl , pH 7 . 5 , 5 mM EDTA , and 4% SDS ) containing 10 mM NEM at 37°C for 10 min . The protein was diluted into 0 . 8 ml LB with 0 . 2% Triton X-100 and 1 mM NEM and incubated overnight at 4°C . NEM was removed by three sequential CM precipitations . Precipitated protein was solubilized in 0 . 2 ml of buffer SB , and then 0 . 8 ml HB ( 1 M hydroxylamine , pH 7 . 5 , 150 mM NaCl , 0 . 2% Triton X-100 , and 1 mM biotin-HPDP ) or buffer TB ( 1 M Tris-HCl , pH 7 . 5 , 150 mM NaCl , 0 . 2% Triton X-100 , and 1 mM biotin-HPDP ) was added . The mixture was incubated for 1 h at room temperature and subjected to CM precipitation . The precipitated protein was dissolved in 0 . 2 ml SB , diluted into 0 . 8 ml LB containing 150 mM NaCl , 0 . 2% Triton X-100 , and 200 µM biotin-HPDP , and incubated for 1 h at room temperature . Confluent HUVECs grown in 12-well plates were treated with MMC ( 10 mg/ml ) for 6 h in order to inactivate cell proliferation . The cells were wounded and images were captured after 12 h . Mice were caged in groups of five in a laminar airflow cabinet under specific pathogen-free conditions , fed with sterilized food and water , and kept on a 12-h light–dark cycle . We checked the bodyweight of the mice every day and observed their drinking and eating conditions , as well as their activity , in order to monitor their health before they were sacrificed . No mice was observed to be ill or dead during the experimental term . No early euthanasia/humane endpoints for animals were performed since none of animals became severely ill/moribund during the experiment ( s ) . Mice were sacrificed by CO2 euthanasia to minimize the suffering of the mice . All treatments were administered according to the Guide for the Care and Use of Laboratory Animals ( Eighth Edition ) . All of the animals were handled according to approved Institutional Animal Care and Use Committee protocols ( AR20130902 ) of the East China Normal University . SCP1-KO mice were generated using CRISPR-CAS9 methods ( Qiu et al . , 2013 ) . Briefly , guide RNA ( gRNA ) expression vectors were constructed for pGS3-T7-gRNA . The sequence of gRNA is CCCTCTTCTGCTGTGTCTGC . The pGS3-T7-gRNA vector and the Cas9-encoding plasmids were linearized using DraI and NotI , respectively . The linearized templates were transcribed in vitro via run-off reactions using T7 RNA polymerase , the In vitro Transcription T7 Kit ( Takara ) , and the Sp6 mMESSAGE mMACHINE Kit ( Ambion ) . TE solution containing 25 ng/μl gRNA and 50 ng/l Cas9 mRNA was injected into the cytoplasm of one cell-stage embryos . A mismatch-sensitive T7E1 assay was used to identify the founders . To confirm the modifications in the founders , the PCR products from each founder were generated using the TA cloning kit ( Takara ) according to the manufacturer’s instructions . PCR was used to identify the genotype of the offspring from the intercrossed Ctdsp1+/- mice . The SCP1-KO MEF cells were generated as described below . Briefly , mice homozygous for Ctdsp1 were intercrossed . The pregnant female mice were sacrificed at day 13 post-coitum . The individual embryos were collected , and any extra-embryonic tissue was removed . Then , the embryos were dispersed using scissors , and the dispersed tissues were trypsinized at 37°C for 30 min . Trypsin was inactivated by adding DMEM . The cells were isolated via centrifugation at 1000 rpm in a microcentrifuge for 5 min at room temperature . Then , the cells were resuspended in DMEM and were seeded on 10-mm dishes . Retinas collected from Ctdsp1–/– mice and control littermates at age P5 were dissected , fixed , and permeabilized in Tris-buffered saline , 1% bovine serum albumin , and 0 . 5% Triton X-100 at 4°C overnight . They were then incubated in Alexa 488-conjugated isolectin B4 ( Bandeiraea simplicifolia; Invitrogen ) at 4°C overnight . After five washes , the flat-mounted retinas were analyzed by fluorescence microscopy ( Lee et al . , 2014 ) . The study protocols were approved by the Institutional Animal Care and Use Committee . We used a mouse model of angiogenesis , in which the entire left femoral artery and vein were excised surgically . When hind limb ischemia was induced , new blood vessels grew into the ischemic limb . We prepared this model in Ctdsp1–/– mice and WT mice to determine whether ischemia-induced angiogenesis was affected by the deficiency of SCP1 . In brief , mice were subjected to unilateral hind limb ischemia under anesthesia with sodium pentobarbital ( 50 mg/kg intraperitoneally ) . Before surgery , bodyweight and systemic arterial blood pressure ( SBP ) were determined . SBP was determined using a tail-cuff pressure analysis system ( TK370C , Unicom ) in the conscious state . Capillary angiogenesis and hind limb blood flow were examined by the methods below ( Egami et al . , 2006 ) . We measured the ratio of the ischemic ( left ) /normal ( right ) hind limb blood flow by laser Doppler blood flowmetry ( moorLDI , Moor Instruments ) . At seven predetermined time-points ( before surgery and at postoperative days 1 , 3 , 7 , 14 , 21 , and 28 ) , we performed laser-beam scanning over the legs and feet . The average laser Doppler blood flowmetry of the ischemic and non-ischemic hind limbs was then computed . To minimize variations as a result of ambient light , blood flow was expressed as the ischemic ( left ) /normal ( right ) hind limb laser Doppler blood flowmetry ratio ( Nilsson et al . , 1980 ) . 2 × 105 LLCs were injected into Ctdsp1-WT or -knockdown littermates . The diameter of the tumor was measured every 2 days , and the volume of the tumor was calculated . At 14–16 days after injection , the tumors were permitted to grow to 15 mm in any direction by the Institutional Animal Care and Use Committee at Massachusetts General Hospital and collected , fixed overnight with 4% paraformaldehyde , embedded in paraffin , and sectioned . Tumor volume was measured by digital caliper every other day , and was calculated using the following formula: tumor volume = length × width2 × 0 . 52 . Sections of the tumors were stained with hematoxylin and eosin and analyzed histologically . The angiogenesis in the tumors was analyzed using immunohistochemistry by CD31 staining ( Standiford et al . , 2011 ) . Aortae from 2-month-old WT and Ctdsp1–/–-deficient mice were dissected and cut into 1-mm long pieces . Aortic rings were placed in growth factor-reduced Matrigel ( BD Biosciences ) and cultured for 5 days in EBM-2 medium ( Lonza ) . Images of individual aortic explants were taken and the microvascular sprouting areas were quantified by measuring the area covered by outgrowth of the vascular sprouts with Image J ( RRID: SCR-003070 ) ( Wang et al . , 2013 ) . HUVECs ( RRID: CVCL-2959 ) were cultured in 2% FBS/DMEM cultured on 24-well plates coated with growth factor-reduced Matrigel ( BD Biosciences ) at 1 . 5 × 105 cells per well and were stimulated with VEGF ( 100 ng/ml ) . The capillary tube length was measured 16 h after the stimulation ( Wang et al . , 2013 ) . The significance of differences was determined using the Student t-test . All quantitative data are expressed as means ± SD . p<0 . 05 was regarded as a significant difference . | Cancerous tumors are the leading cause of death worldwide . Tumors cannot grow beyond a couple of millimeters in diameter unless they are supplied with nutrients and oxygen . To receive these , tumors connect to the body’s blood supply by stimulating the growth of new blood vessels . Drugs that reduce the ability of new blood vessels to form have therefore been investigated as possible anti-cancer treatments . New blood vessels emerge from pre-existing ones in a process called angiogenesis . The first stage of angiogenesis involves the endothelial cells that line the inner wall of the blood vessels moving outwards to form new ‘sprouts’ . Within the endothelial cells , a signaling protein called AKT drives angiogenesis by moving to the cell membrane , where it is activated and triggers further signaling events . The activation of AKT occurs via a phosphate group being attached to a particular site on the protein . Enzymes called phosphatases remove phosphate groups from proteins and so can inactivate AKT , hence preventing angiogenesis . Although some phosphatases are known to inactivate AKT , they cannot easily be counted or analyzed . This means that they cannot be used to develop new cancer treatments . In addition , for the phosphatase to best prevent tumor growth , it should inactivate AKT at the cell membrane . Liao , Wang , Li , Wang , Jin et al . now show that a phosphatase called SCP1 can localize to the cell membrane and inactivate AKT there . SCP1 was not previously known to anchor to the cell membrane . Liao et al . found that this anchoring occurs via a modification that attaches a fatty acid molecule to SCP1 . Further experiments showed that mice that lacked SCP1 had increased levels of AKT phosphorylation in their endothelial cells , more new blood vessel growth and , consequently , had tumors that grew faster . Further research is now needed to investigate exactly how SCP1 moves to the cell membrane from elsewhere in the cell . Ultimately , this knowledge could play an important role in identifying potential drugs that prevent or reduce the growth of tumors . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology",
"cancer",
"biology"
] | 2017 | Palmitoylated SCP1 is targeted to the plasma membrane and negatively regulates angiogenesis |
The risk of developing cancer is correlated with body size and lifespan within species . Between species , however , there is no correlation between cancer and either body size or lifespan , indicating that large , long-lived species have evolved enhanced cancer protection mechanisms . Elephants and their relatives ( Proboscideans ) are a particularly interesting lineage for the exploration of mechanisms underlying the evolution of augmented cancer resistance because they evolved large bodies recently within a clade of smaller-bodied species ( Afrotherians ) . Here , we explore the contribution of gene duplication to body size and cancer risk in Afrotherians . Unexpectedly , we found that tumor suppressor duplication was pervasive in Afrotherian genomes , rather than restricted to Proboscideans . Proboscideans , however , have duplicates in unique pathways that may underlie some aspects of their remarkable anti-cancer cell biology . These data suggest that duplication of tumor suppressor genes facilitated the evolution of increased body size by compensating for decreasing intrinsic cancer risk .
Among the constraints on the evolution of large bodies and long lifespans in animals is an increased risk of developing cancer . If all cells in all organisms have a similar risk of malignant transformation and equivalent cancer suppression mechanisms , then organisms with many cells should have a higher prevalence of cancer than organisms with fewer cells , particularly because large and small animals have similar cell sizes ( Savage et al . , 2007 ) . Consistent with this expectation there is a strong positive correlation between body size and cancer incidence within species; for example , cancer incidence increases with increasing adult height in humans ( Million Women Study collaborators et al . , 2011; Nunney , 2018 ) and with increasing body size in dogs , cats , and cattle ( Dobson , 2013; Dorn et al . , 1968; Lucena et al . , 2011 ) . There is no correlation , however , between body size and cancer risk between species; this lack of correlation is often referred to as ‘Peto’s Paradox’ ( Caulin and Maley , 2011; Leroi et al . , 2003; Peto et al . , 1975 ) . Indeed , cancer prevalence is relatively stable at ~5% across species with diverse body sizes ranging from the minuscule 51 g grass mouse to the gargantuan 4800 kg African elephant ( Abegglen et al . , 2015; Boddy et al . , 2020; Tollis et al . , 2020 ) . The ultimate resolution to Peto’s Paradox is trivial , large-bodied and long-lived species evolved enhanced cancer protection mechanisms , but identifying and characterizing the mechanisms that underlie the evolution of augmented cancer protection has proven difficult ( Ashur-Fabian et al . , 2004; Seluanov et al . , 2008; Gorbunova et al . , 2012; Tian et al . , 2013; Sulak et al . , 2016 ) . One of the challenges for discovering how animals evolved enhanced cancer protection mechanisms is identifying lineages in which large-bodied species are nested within species with small body sizes . Afrotherian mammals are generally small-bodied , but also include the largest extant land mammals . For example , maximum adult weights are ~70 g in golden moles , ~120 g in tenrecs , ~170 g in elephant shrews , ~3 kg in hyraxes , and ~60 kg in aardvarks ( Tacutu et al . , 2013 ) . In contrast , while extant hyraxes are relatively small , the extinct Titanohyrax is estimated to have weighed ~1300 kg ( Schwartz et al . , 1995 ) . The largest living Afrotheria are also dwarfed by the size of their recent extinct relatives: extant sea cows such as manatees are large bodied ( ~322–480 kg ) but are relatively small compared to the extinct Stellar’s sea cow which is estimated to have weighed ~8000–10 , 000 kg ( Scheffer , 1972 ) . Similarly African Savannah ( 4800 kg ) and Asian elephants ( 3200 kg ) are large , but are dwarfed by the truly gigantic extinct Proboscideans such as Deinotherium ( ~12 , 000 kg ) , Mammut borsoni ( 16 , 000 kg ) , and the straight-tusked elephant ( ~14 , 000 kg ) ( Larramendi , 2015 ) . Remarkably , these large-bodied Afrotherian lineages are nested deeply within small-bodied species ( Figure 1; O Leary et al . , 2013a; Springer et al . , 2013; O Leary et al . , 2013b; Puttick and Thomas , 2015 ) , indicating that gigantism independently evolved in hyraxes , sea cows , and elephants ( Paenungulata ) . Thus , Paenungulates are an excellent model system in which to explore the mechanisms that underlie the evolution of large body sizes and augmented cancer resistance . Many mechanisms have been suggested to resolve Peto’s paradox , including a decrease in the copy number of oncogenes , an increase in the copy number of tumor suppressor genes ( Caulin and Maley , 2011; Leroi et al . , 2003; Nunney , 1999 ) , reduced metabolic rates , reduced retroviral activity and load ( Katzourakis et al . , 2014 ) , and selection for ‘cheater’ tumors that parasitize the growth of other tumors ( Nagy et al . , 2007 ) , greater sensitivity of cells to DNA damage ( Abegglen et al . , 2015; Sulak et al . , 2016 ) , enhanced recognition of neoantigens by T cells , among many others . Among the most parsimonious routes to enhanced cancer resistance may be through an increased copy number of tumor suppressors . For example , transgenic mice with additional copies of TP53 have reduced cancer rates and extended lifespans ( García-Cao et al . , 2002 ) , suggesting that changes in the copy number of tumor suppressors can affect cancer rates . Indeed , candidate genes studies have found that elephant genomes encode duplicate tumor suppressors such as TP53 and LIF ( Abegglen et al . , 2015; Sulak et al . , 2016; Vazquez et al . , 2018 ) as well as other genes with putative tumor suppressive functions ( Caulin et al . , 2015; Doherty and de Magalhães , 2016 ) . These studies , however , focused on a priori candidate genes; thus it is unclear whether duplication of tumor suppressor genes is a general phenomenon in the elephant lineage or reflects an ascertainment bias . Here we trace the evolution of body mass , cancer risk , and gene copy number variation across Afrotherian genomes , including multiple living and extinct Proboscideans ( Figure 1 ) , to investigate whether duplications of tumor suppressors coincided with the evolution of large body sizes . Our estimates of the evolution of body mass across Afrotheria show that large body masses evolved in a stepwise manner , similar to previous studies ( O Leary et al . , 2013a; Springer et al . , 2013; O Leary et al . , 2013b; Puttick and Thomas , 2015 ) and coincident with dramatic reductions in intrinsic cancer risk . To explore whether duplication of tumor suppressors occurred coincident with the evolution of large body sizes , we used a genome-wide Reciprocal Best BLAT Hit ( RBBH ) strategy to identify gene duplications and used maximum likelihood to infer the lineages in which those duplications occurred . Unexpectedly , we found that duplication of tumor suppressor genes was common in Afrotherians , both large and small . Gene duplications in the Proboscidean lineage , however , were uniquely enriched in pathways that may explain some of the unique cancer protection mechanisms observed in elephant cells . These data suggest that duplication of tumor suppressor genes is pervasive in Afrotherians and preceded the evolution of species with exceptionally large body sizes .
Similar to previous studies of Afrotherian body size ( Puttick and Thomas , 2015; Elliot and Mooers , 2014 ) , we found that the body mass of the Afrotherian ancestor was inferred to be small ( 0 . 26 kg , 95% CI: 0 . 31–3 . 01 kg ) and that substantial accelerations in the rate of body mass evolution occurred coincident with a 67 . 36× increase in body mass in the stem-lineage of Pseudoungulata ( 17 . 33 kg ) ; a 1 . 45× increase in body mass in the stem-lineage of Paenungulata ( 25 . 08 kg ) ; a 11 . 82× increase in body mass in the stem-lineage of Tehthytheria ( 296 . 56 kg ) ; a 1 . 39× increase in body mass in the stem-lineage of Proboscidea ( 412 . 5 kg ) ; and a 2 . 69× increase in body mass in the stem-lineage of Elephantimorpha ( 4114 . 39 kg ) , which is the last common ancestor of elephants and mastodons using the fossil record ( Figure 2A , B ) . The ancestral Hyracoidea was inferred to be relatively small ( 2 . 86–118 . 18kg ) , and rate accelerations were coincident with independent body mass increases in large hyraxes such as Titanohyrax andrewsi ( 429 . 34 kg , 67 . 36× increase ) ( Figure 2A , B ) . While the body mass of the ancestral Sirenian was inferred to be large ( 61 . 7–955 . 51 kg ) , a rate acceleration occurred coincident with a 10 . 59× increase in body mass in Stellar’s sea cow ( Figure 2A , B ) . Rate accelerations also occurred coincident with dramatic reductions in body mass ( 36 . 6× decrease ) in the stem-lineage of the dwarf elephants Elephas ( Palaeoloxodon ) antiquus falconeri and Elephas cypriotes ( Figure 2A , B ) . These data indicate that gigantism in Afrotherians evolved step-wise , from small to medium bodies in the Pseudoungulata stem-lineage , medium to large bodies in the Tehthytherian stem-lineage and extinct hyraxes , and from large to exceptionally large bodies independently in the Proboscidean stem-lineage and Stellar’s sea cow ( Figure 2A , B ) . In order to account for a relatively stable cancer rate across species ( Abegglen et al . , 2015; Boddy et al . , 2020; Tollis et al . , 2020 ) , intrinsic cancer risk must also evolve with changes in body size and lifespan across species . We used empirical body size and lifespan data from extant species and empirical body size and estimated lifespan data from extinct species to estimate intrinsic cancer risk ( K ) with the simplified multistage cancer risk model K≈Dt6 , where D is the maximum body size and t is the maximum lifespan ( Peto et al . , 1975: Peto , 2015; Armitage , 1985; Armitage and Doll , 2004 ) . As expected , intrinsic cancer risk in Afrotheria also varies with changes in body size and longevity ( Figure 2A , B ) , with a 6 . 41-log2 decreases in the stem-lineage of Xenarthra , followed by a 13 . 37-log2 decrease in Pseudoungulata , and a 1 . 49-log2 decrease in Aardvarks ( Figure 2A ) . In contrast to the Paenungulate stem-lineage , there is a 7 . 84-log2 decrease in cancer risk in Tethytheria , a 0 . 67-log2 decrease in Manatee , a 3 . 14-log2 decrease in Elephantimorpha , and a 1 . 05-log2 decrease in Proboscidea . Relatively minor decreases occurred within Proboscidea including a 0 . 83-log2 decrease in Elephantidae and a 0 . 57-log2 decrease in the American Mastodon . Within the Elephantidae , Elephantina and Loxodontini have a 0 . 06-log2 decrease in cancer susceptibility , while susceptibility is relatively stable in Mammoths . The three extant Proboscideans , Asian Elephant , African Savana Elephant , and the African Forest Elephant , meanwhile , have similar decreases in body size , with slight increases in cancer susceptibility ( Figure 2A , B ) . Our hypothesis was that genes which duplicated coincident with the evolution of increased body mass ( IBM ) and reduced intrinsic cancer risk ( RICR ) would be uniquely enriched in tumor suppressor pathways compared to genes that duplicated in other lineages . Therefore , we identified duplicated genes in each Afrotherian lineage ( Table 1 and Figure 3A ) and tested if they were enriched in Reactome pathways related to cancer biology ( Figure 3B , Table 2 ) . No pathways related to cancer biology were enriched in either the Pseudoungulata ( 67 . 36-fold IBM , 13 . 37-log2 RICR ) , but few genes were inferred to be duplicated in this lineage reducing power to detect enriched pathways . Consistent with our hypothesis , 18 . 18% of the pathways that were enriched in the Paenungulate stem-lineage ( 1 . 45-fold IBM , 1 . 17-log2 RICR ) , 63% of the pathways that were enriched in the Tethytherian stem-lineage ( 11 . 82-fold IBM , 7 . 84-log2 RICR ) , and 38 . 81% of the pathways that were enriched in the Proboscidean stem-lineage ( 1 . 06-fold IBM , 3 . 14-log2 RICR ) were related to tumor suppression ( Figure 3B , Table 2 ) . Similarly , 21 . 28% and 38 . 00% of the pathways that were enriched in manatee ( 1 . 11-fold IBM , 0 . 89-log2 RICR ) and aardvark ( 67 . 36-fold IBM , 1 . 49-log2 RICR ) , respectively , were related to tumor suppression . In contrast , only 2 . 86% of the pathways that were enriched in hyrax ( 1 . 6-fold IBM , 1 . 49-log2 RICR ) were related to tumor suppression ( Figure 3B , Table 2 ) . Unexpectedly , however , lineages without major increases in body size or lifespan , or decreases in intrinsic cancer risk , were also enriched for tumor suppressor pathways . For example , 13 . 85% , 37 . 04% , and 22 . 00% of the pathways that were enriched in the stem-lineages of Afroinsectivoa and Afrosoricida , and in E . telfairi , respectively , were related to cancer biology ( Figure 3B , Table 2 ) . Our observation that gene duplicates in most lineages are enriched in cancer pathways suggest either that duplication of genes in cancer pathways is common in Afrotherians , or that there may be a systemic bias in the pathway enrichment analyses . For example , random gene sets may be generally enriched in pathway terms related to cancer biology . To explore this latter possibility , we generated 5000 randomly sampled gene sets of between 10 and 5000 genes , and tested for enriched Reactome pathways using ORA . We found that no cancer pathways were enriched ( median hypergeometric p-value ≤0 . 05 ) among gene sets tested greater than 157 genes; however , in these smaller gene sets , 12–18% of enriched pathways were classified as cancer pathways . Without considering p-value thresholds , the percentage of enriched cancer pathways approaches ~15% ( 213/1381 ) in simulated sets . Thus , for larger gene sets , we used a simulated threshold of ~15% to determine if pathways related to cancer biology were enriched more than one would expect from sampling bias ( Table 2 ) . We directly compared our simulated and observed enrichment results by lineage and gene set size , and found that Afrosoricida , Cape golden mole , tenrec , Elephantidae , elephant shrew , Asian elephant , African Savannah elephant , African Forest elephant , Columbian mammoth , aardvark , Paenungulata , Proboscidea , Tethytheria , and manatee had enriched cancer pathway percentages above background with respect to their gene set sizes , that is expected enrichments based on random sampling of small gene sets ( Table 2 ) . Thus , we conclude that duplication of genes in cancer pathways is common in many Afrotherians but that the inference of enriched cancer pathway duplication is not different from background in some lineages , particularly in ancestral nodes with a small number of estimated duplicates . While duplication of cancer associated genes is common in Afrotheria , the 157 genes that duplicated in the Proboscidean stem-lineage ( Figure 3A ) were uniquely enriched in 12 pathways related to cancer biology ( Figure 3B ) . Among these uniquely enriched pathways ( Figure 3C ) were pathways related to the cell cycle , including ‘G0 and Early G1’ , ‘G2/M Checkpoints’ , and ‘Phosphorylation of the APC/C’ , pathways related to DNA damage repair including ‘Global Genome Nucleotide Excision Repair ( GG-NER ) ’ , ‘HDR through Single Strand Annealing ( SSA ) ’ , ‘Gap-filling DNA repair synthesis and ligation in GG-NER’ , ‘Recognition of DNA damage by PCNA-containing replication complex’ , and ‘DNA Damage Recognition in GG-NER’ , pathways related to telomere biology including ‘Extension of Telomeres’ and ‘Telomere Maintenance’ , pathways related to the apoptosome including ‘Activation of caspases through apoptosome-mediated cleavage’ , and pathways related to ‘mTORC1-mediated signaling’ and ‘mTOR signaling’ , which play important roles in the biology of aging . Thus , duplication of genes with tumor suppressor functions is pervasive in Afrotherians , but genes in some pathways related to cancer biology and tumor suppression are uniquely duplicated in large-bodied ( long-lived ) Proboscideans ( Figure 4A , B ) . Among the genes uniquely duplicated within Proboscideans are TP53 , COX20 , LAMTOR5 , PRDX1 , STK11 , BRD7 , MAD2L1 , BUB3 , UBE2D1 , SOD1 , LIF , MAPRE1 , CNOT11 , CASP9 , CD14 , and HMGB2 ( Figure 4C ) . Two of these , TP53 and LIF , have been previously described ( Abegglen et al . , 2015; Sulak et al . , 2016; Vazquez et al . , 2018 ) . These genes are significantly enriched in pathways involved in apoptosis , cell cycle regulation , and both upstream and downstream pathways involving TP53 . The majority of these genes are expressed in African Elephant transcriptome data ( Figure 4D ) , suggesting that they maintained functionality after duplication . Prior studies found that the ‘master’ tumor suppressor TP53 duplicated multiple times in elephants ( Abegglen et al . , 2015; Sulak et al . , 2016 ) , motivating us to further study duplication of genes involved in TP53-related pathways in Proboscidea . We traced the evolution of genes in the TP53 pathway that appeared in one or more Reactome pathway enrichments for genes duplicated recently in the African Elephant , which has the most complete genome among Proboscideans and for which several RNA-Seq data sets are available . We found that the initial duplication of TP53 in Tethytheria , where body size expanded , was preceded by the duplication of GTF2F1 and STK11 in Paenungulata and was coincident with the duplication of BRD7 . These three genes are involved in regulating the transcription of TP53 ( Liang and Mills , 2013; Launonen , 2005; Drost et al . , 2010; Burrows et al . , 2010 ) , and their duplication prior to that of TP53 may have facilitated re-functionalization of TP53 retroduplicates . Interestingly , STK11 is also tumor suppressor that mediates tumor suppression via p21-induced senescence ( Launonen , 2005 ) . The other genes that are duplicated in the pathway are downstream of TP53; these genes duplicated either coincident with TP53 , as in the case of SIAH1 , or subsequently in Proboscidea , Elephantidae , or extant elephants ( Figure 4 ) . These genes are expressed in RNA-Seq data ( Figure 4D ) , suggesting that they are functional . While transcript abundance estimates inferred from RNA-Seq data can suggest that genes are functional , recent non-functional duplicates can still be transcribed . Therefore we inferred if each duplicate shown in Figure 4C/D encoded a putatively function protein by manually curation , specifically to identify premature stop codons and overall sequence conservation . Most genes in Figure 4C/D , such as STK11 , CD14 , SOD1 , and BRD7 , were well conserved and lacked premature stop codons . We also find that the STK11 , CD14 , and BRD7 genes in the manatee were also well conserved , suggesting that extant manatees may also have enhanced tumor suppression and an augmented stress response . However , some of the duplicate genes in the mantatee genome have premature stop codons suggesting they are not translated into functional proteins , including the additional copies of MAPRE1 , BUB3 , and COX20 as well as at least one of the duplicate copies of CNOT11 , HMGB2 , MAD2L1 , LIF , and TP53 . For TP53 , we have previously shown that duplicate copies of genes containing premature stop codons may still serve a functional role in regulating its progenitor's function . Thus , some of the genes with premature stop codons , such as duplicate COX20 and MAD2L1 which are expressed in RNA-Seq data , may encode functional lncRNA transcripts or truncated proteins . Some copies , including for CASP9 and PRDX1 , contained partial RBBH hits with no premature stop codons; however , they also lacked the totality of the coding sequence and thus may represent cases of pseudogenization , subfunctionalization , or neofunctionalization .
The hundred- to hundred-million-fold reductions in intrinsic cancer risk associated with the evolution of large body sizes in some Afrotherian lineages , in particular Elephantimorphs such as elephants and mastodons , suggests that these lineages must have also evolved remarkable mechanisms to suppress cancer . While our initial hypothesis was that large-bodied lineages would be uniquely enriched in duplicate tumor suppressor genes compared to other smaller-bodied lineages , we unexpectedly found that the duplication of genes in tumor suppressor pathways occurred at various points throughout the evolution of Afrotheria , regardless of body size . These data suggest that this abundance of tumor suppressors may have contributed to the evolution of large bodies and reduced cancer risk , but that these processes were not necessarily coincident . Interestingly , pervasive duplication of tumor suppressors may also have contributed to the repeated evolution of large bodies in hyraxes and sea cows , because at least some of the genetic changes that underlie the evolution of reduced cancer risk were common in this group . It remains to be determined whether our observation of pervasive duplication of tumor suppressors also occurs in other multicellular lineages . Using a similar reciprocal best BLAST/BLAT approach that focused on estimating copy number of known tumor suppressors in mammalian genomes , for example , Caulin et al . , 2015 found no correlation between copy number or tumor suppressors with either body mass or longevity , whereas Tollis et al . , 2020 found a correlation between copy number and longevity ( but not body size ) ( Tollis et al . , 2020; Caulin et al . , 2015 ) . These opposing conclusions may result from differences in the number of genes ( 81 vs 548 ) and genomes ( 8 vs 63 ) analyzed , highlighting the need for genome-wide analyses of many species that vary in body size and longevity . While we observed that duplication of genes in cancer related pathways – including genes with known tumor suppressor functions – is pervasive in Afrotheria , the number of duplicate tumor suppressor genes was relatively small , which may reflect a trade-off between the protective effects of increased tumor suppressor number on cancer risk and potentially deleterious consequences of increased tumor suppressor copy number . Overexpression of TP53 in mice , for example , is protective against cancer but associated with progeria , premature reproductive senescence , and early death; however , transgenic mice with a duplication of the TP53 locus that includes native regulatory elements are healthy and experience normal aging , while also demonstrating an enhanced response to cellular stress and lower rates of cancer ( García-Cao et al . , 2002; Tyner et al . , 2002 ) . These data suggest that duplication of tumor suppressors can contribute to augmented cancer resistance , if the duplication includes sufficient regulatory architecture to direct spatially and temporally appropriate gene expression . Thus , it is interesting that duplication of genes that regulate TP53 function , such as STK11 , SIAH1 , and BRD7 , preceded the retroduplication TP53 in the Proboscidean stem-lineage , which may have mitigated toxicity arising from dosage imbalances . Similar co-duplication events may have alleviated the negative pleiotropy of tumor suppressor gene duplications to enable their persistence and allow for subsequent co-option during the evolution of cancer resistance . Our genome-wide results suggest that duplication of tumor suppressors is pervasive in Afrotherians and may have enabled the evolution of larger body sizes in multiple lineages by lowering intrinsic cancer risk either prior to or coincident with increasing body size . However , our study has several inherent limitations . For example , we have shown that genome quality plays an important role in our ability to identify duplicate genes , and several species have poor quality genomes ( and thus were excluded from further analyses ) . While several efforts have been established with the goal of generating high quality ( chromosome length ) reference genomes for mammals , such as DNAZoo , The Zoonomia Project , the Vertebrate Genomes Project , and Genome 10K , Atlantogenatans represent a minority of available genome projects . And while a few high quality Atlantogenatan genomes are available , they lack reference gene and transcriptome annotations , and genome browser graphical user interfaces that allow for easy access to genome data for the broader community , limiting their usefulness . Similarly , without comprehensive gene expression data we cannot be certain that duplicate genes are actually expressed , and thus functional . Our results on genome quality suggest several research priorities for these less well-studies species , including generating chromosome length reference genomes and genome annotations , and incorporating these species into existing genome browsers ( such as UCSC Genome Browser ) . We also assume that gene duplicates either maintain ancestral tumor suppressor functions and increase cancer resistance through dosage effects or provide redundancy to loss of function mutations thereby increasing robustness of tumor suppression . Many processes , such as developmental systems drift , neofunctionalization , and sub-functionalization , can cause divergence in gene functions and invalidate the assumption of conservation of gene function ( Rastogi and Liberles , 2005; Qian and Zhang , 2014; Stoltzfus , 1999 ) , leading to inaccurate inferences in gene and pathway functions which is a common problem in comparative genomic studies using pathway and gene ontologies to categorize gene function . In addition , we assume that most duplicate genes are functional but it is likely that some of the duplicates were identify are non-functional pseudogenes . Differentiating between functional and non-functional genes using comparative genomics can be challenging . For example , non-functional pseudogenes often accumulate non-synonymous amino acid substitutions and premature stop codons but these same changes can also occur in functional genes . For example , we have found that the elephant genome encodes TP53 retogenes ( TP53RTGs ) all of which encode premature stop codons suggesting they are pseudogenes , but these TP53TRGs are expressed , encode functional separation of function mutants of the ancestral TP53 gene , and contribute to enhanced DNA damage sensitivity in elephant cells . Similarly , we have characterized duplicate LIF gene in elephants ( LIF6 ) that lacks the start codon and exon 1 of the parent LIF gene . LIF6 is expressed , encodes a functional protein with translation initiated at an alternative downstream start site , and also contributes to enhanced DNA damage sensitivity in elephant cells . In addition , duplicate genes that lack coding potential , such as PTENP1 , can also be expressed and while not translated function as LINC RNAs ( in this case acting as a sponge for microRNAs that target the parent PTEN transcript ) . In each case classifying duplicates into putatively functional and non-functional categories based on sequence characteristic would misclassify TP53RTGs , LIF6 , and PTENP1 . Thus , sequence features of pseudogenes may maintain function , as a consequence of not excluding putative pseudogenes some of the genes we include in downstream analyses may be non-functional . Further experimental studies are needed to determine which duplicates are expressed and functional . The focus of this study , motivated by our previous identification of TP53 and LIF duplicates , was on the role gene duplication in general may have played in the resolution of Peto’s paradox in large-bodied Afrotherians , particularly Proboscidea . Duplication of tumor suppressor genes , however , is unlikely to be the sole mechanism responsible for the evolution of large body sizes , long lifespans , and reduced cancer risk . The evolution of regulatory elements , coding genes , genes with non-canonical tumor suppressor functions , and immune cell recognition of cancerous cells are also likely important for reducing the risk of cancer . While we found that duplication of tumor suppressor genes is common in Afrotheria , genes that duplicated in the Proboscidean stem-lineage ( Figure 3A , B ) were uniquely enriched in functions and pathways that may be related to the evolution of unique anti-cancer cellular phenotypes in the elephant lineage ( Figure 3C ) . Elephant cells , for example , cannot be experimentally immortalized ( Fukuda et al . , 2016; Gomes et al . , 2011 ) , rapidly repair DNA damage ( Sulak et al . , 2016; Hart and Setlow , 1974; Francis et al . , 1981 ) , are extremely resistant to oxidative stress ( Gomes et al . , 2011 ) , and yet are also extremely sensitive to DNA damage ( Abegglen et al . , 2015; Sulak et al . , 2016; Vazquez et al . , 2018 ) . Several pathways related to DNA damage repair , in particular nucleotide excision repair ( NER ) , were uniquely enriched among genes that duplicated in the Proboscidean stem-lineage , suggesting a connection between duplication of genes involved in NER and rapid DNA damage repair ( Hart and Setlow , 1974; Francis et al . , 1981 ) . Similarly , we identified a duplicate SOD1 gene in Proboscideans that may confer the resistance of elephant cells to oxidative stress ( Gomes et al . , 2011 ) . Pathways related to the cell cycle were also enriched among genes that duplicated in Proboscideans , and cell cycle dynamics are different in elephants compared to other species; population doubling ( PD ) times for African and Asian elephant cells are 13–16 days , while PD times are 21–28 days in other Afrotherians ( Gomes et al . , 2011 ) . Finally , the role of ‘mTOR signaling’ in the biology of aging is well known . Collectively these data suggest that gene duplications in Proboscideans may underlie some of their cellular phenotypes that contribute to cancer resistance .
We first assembled a time-calibrated supertree of Eutherian mammals by combining the time-calibrated molecular phylogeny of Bininda-Emonds et al . , 2007; Bininda-Emonds et al . , 2008 with the time-calibrated total evidence Afrotherian phylogeny from Puttick and Thomas , 2015 . While the Bininda-Emonds et al . , 2007; Bininda-Emonds et al . , 2008 phylogeny includes 1679 species , only 34 are Afrotherian , and no fossil data are included . The inclusion of fossil data from extinct species is essential to ensure that ancestral state reconstructions of body mass are not biased by only including extant species . This can lead to inaccurate reconstructions , for example , if lineages convergently evolved large body masses from a small-bodied ancestor . In contrast , the total evidence Afrotherian phylogeny of Puttick and Thomas , 2015 includes 77 extant species and fossil data from 39 extinct species . Therefore , we replaced the Afrotherian clade in the Bininda-Emonds et al . , 2008 phylogeny with the Afrotherian phylogeny of Puttick and Thomas , 2015 using Mesquite . Next , we jointly estimated rates of body mass evolution and reconstructed ancestral states using a generalization of the Brownian motion model that relaxes assumptions of neutrality and gradualism by considering increments to evolving characters to be drawn from a heavy-tailed stable distribution ( the ‘Stable Model’ ) implemented in StableTraits ( Elliot and Mooers , 2014 ) . The stable model allows for large jumps in traits and has previously been shown to outperform other models of body mass evolution , including standard Brownian motion models , Ornstein–Uhlenbeck models , early burst maximum likelihood models , and heterogeneous multi-rate models ( Elliot and Mooers , 2014 ) . We developed a reciprocal best hit BLAT ( RBHB ) pipeline to identify putative homologs and estimate gene copy number across species . The Reciprocal Best Hit ( RBH ) search strategy is conceptually straightforward: ( 1 ) Given a gene of interest GA in a query genome A , one searches a target genome B for all possible matches to GA; ( 2 ) For each of these hits , one then performs the reciprocal search in the original query genome to identify the highest-scoring hit; ( 3 ) A hit in genome B is defined as a homolog of gene GA if and only if the original gene GA is the top reciprocal search hit in genome A . We selected BLAT ( Kent , 2002 ) as our algorithm of choice , as this algorithm is sensitive to highly similar ( >90% identity ) sequences , thus identifying the highest-confidence homologs while minimizing many-to-one mapping problems when searching for multiple genes . RBH performs similar to other more complex methods of orthology prediction and is particularly good at identifying incomplete genes that may be fragmented in low quality/poorly assembled regions of the genome ( Altenhoff and Dessimoz , 2009; Salichos and Rokas , 2011 ) . In low-quality genomes , many genes are fragmented across multiple scaffolds , which results in BLA ( S ) T-like methods calling multiple hits when in reality there is only one gene . To compensate for this , we developed a novel statistic , Estimated Copy Number by Coverage ( ECNC ) , which averages the number of times we hit each nucleotide of a query sequence in a target genome over the total number of nucleotides of the query sequence found overall in each target genome ( Figure 3—figure supplement 1 ) . This allows us to correct for genes that have been fragmented across incomplete genomes , while accounting for missing sequences from the human query in the target genome . Mathematically , this can be written as: ( 1 ) ECNC=∑n=1lCn∑n=1lbool ( Cn ) where n is the given nucleotide in the query , l is the total length of the query , Cn is the number of instances that n is present within a reciprocal best hit , and bool ( Cn ) is 1 if Cn >1 Cn>0 or 0 if Cn =1 Cn=0 . We created a custom Python pipeline for automating RBHB searches between a single reference genome and multiple target genomes using a list of query sequences from the reference genome . For the query sequences in our search , we used the hg38 UniProt proteome ( The UniProt Consortium , 2017 ) , which is a comprehensive set of protein sequences curated from a combination of predicted and validated protein sequences generated by the UniProt Consortium . Next , we excluded genes from downstream analyses for which assignment of homology was uncertain , including uncharacterized ORFs ( 991 genes ) , LOC ( 63 genes ) , HLA genes ( 402 genes ) , replication dependent histones ( 72 genes ) , odorant receptors ( 499 genes ) , ribosomal proteins ( 410 genes ) , zinc finger transcription factors ( 1983 genes ) , viral and repetitive-element-associated proteins ( 82 genes ) , and ‘Uncharacterized’ , ‘Putative’ , or ‘Fragment’ proteins ( 30 , 724 genes ) , leaving a final set of 37 , 582 query protein isoforms , corresponding to 18 , 011 genes . We then searched for all copies of 18 , 011 query genes in publicly available Afrotherian genomes ( Dobson , 2013 ) , including African savannah elephant ( Loxodonta africana: loxAfr3 , loxAfr4 , loxAfrC ) , African forest elephant ( Loxodonta cyclotis: loxCycF ) , Asian Elephant ( Elephas maximus: eleMaxD ) , Woolly Mammoth ( Mammuthus primigenius: mamPriV ) , Colombian mammoth ( Mammuthus columbi: mamColU ) , American mastodon ( Mammut americanum: mamAmeI ) , Rock Hyrax ( Procavia capensis: proCap1 , proCap2 , proCap2HiC ) , West Indian Manatee ( Trichechus manatus latirostris: triManLat1 , triManLat1HiC ) , Aardvark ( Orycteropus afer: oryAfe1 , oryAfe1HiC ) , Lesser Hedgehog Tenrec ( Echinops telfairi: echTel2 ) , Nine-banded armadillo ( Dasypus novemcinctus: dasNov3 ) , Hoffman’s two-toed sloth ( Choloepus hoffmannii: choHof1 , choHof2 , choHof2HiC ) , Cape golden mole ( Chrysochloris asiatica: chrAsi1 ) , and Cape elephant shrew ( Elephantulus edwardii: eleEdw1 ) ( Dudchenko et al . , 2017; Palkopoulou et al . , 2015; Palkopoulou et al . , 2018; Foote et al . , 2015 ) . A summary of gene duplications in each species is available in Supplementary file 1 . In order to condense transcript-level hits into single gene loci , and to resolve many-to-one genome mappings , we removed exons where transcripts from different genes overlapped , and merged overlapping transcripts of the same gene into a single gene locus call . The resulting gene-level copy number table was then combined with the maximum ECNC values observed for each gene in order to call gene duplications . We called a gene duplicated if its copy number was two or more , and if the maximum ECNC value of all the gene transcripts searched was 1 . 5 or greater; previous studies have shown that incomplete duplications can encode functional genes ( Sulak et al . , 2016; Vazquez et al . , 2018 ) , therefore partial gene duplications were included provided they passed additional inclusion criteria ( see below ) . The ECNC cut-off of 1 . 5 was selected empirically , as this value minimized the number of false positives seen in a test set of genes and genomes . The results of our initial search are summarized in Figure 3A . Overall , we identified 13 , 880 genes across all species , or 77 . 1% of our starting query genes . In order to determine the effect of genome quality on our results , we used the gVolante webserver and CEGMA to assess the quality and completeness of the genome ( Nishimura et al . , 2017; Parra et al . , 2009 ) . CEGMA was run using the default settings for mammals ( ‘Cut-off length for sequence statistics and composition’=1; ‘CEGMA max intron length’=100 , 000; ‘CEGMA gene flanks’=10 , 000 , ‘Selected reference gene set’ = CVG ) . For each genome , we generated a correlation matrix using the aforementioned genome quality scores , and either the mean copy number or mean ECNC for all hits in the genome . We observed that the percentage of duplicated genes in non-Pseudoungulatan genomes was higher ( 12 . 94–23 . 66% ) than Pseudoungulatan genomes ( 3 . 26–7 . 80% ) . Mean copy number , mean ECNC , and mean CN ( the lesser of copy number and ECNC per gene ) moderately or strongly correlated with genomic quality , such as LD50 , the number of scaffolds , and contigs with a length above either 100K or 1M ( Figure 3—figure supplement 2 ) . The Afrosoricidians had the greatest correlation between poor genome quality and high gene duplication rates , including larger numbers of private duplications . The correlations between genome quality metric and number of gene duplications were particularly high for Cape golden mole ( Chrysochloris asiatica: chrAsi1 ) and Cape elephant shrew ( Elephantulus edwardii: eleEdw1 ) ; therefore we excluded these species from downstream pathway enrichment analyses . In order to ascertain the functional status of duplicated genes , we generated de novo transcriptomes using publicly available RNA-sequencing data for African savanna elephant , West Indian manatee , and nine-banded armadillo ( Supplementary file 2 ) . We mapped reads to the highest quality genome available for each species , and assembled transcripts using HISAT2 and StringTie ( Kim et al . , 2015; Pertea et al . , 2015; Pertea et al . , 2016 ) . We found that many of our identified duplicates had transcripts mapping to them above a Transcripts Per Million ( TPM ) score of 2 , suggesting that many of these duplications are functional . RNA-sequencing data was not available for Cape golden mole , Cape elephant shrew , rock hyrax , aardvark , or the lesser hedgehog tenrec . We encoded the copy number of each gene for each species as a discrete trait ranging from 0 ( one gene copy ) to 31 ( for 32+ gene copies ) and used IQ-TREE to select the best-fitting model of character evolution ( Minh et al . , 2020; Hoang et al . , 2018; Kalyaanamoorthy et al . , 2017; Wang et al . , 2018; Schrempf et al . , 2019 ) , which was inferred to be a Jukes-Cantor type model for morphological data ( MK ) with equal character state frequencies ( FQ ) and rate heterogeneity across sites approximated by including a class of invariable sites ( I ) plus a discrete Gamma model with four rate categories ( G4 ) . Next we inferred gene duplication and loss events with the empirical Bayesian ancestral state reconstruction ( ASR ) method implemented in IQ-TREE ( Minh et al . , 2020; Hoang et al . , 2018; Kalyaanamoorthy et al . , 2017; Wang et al . , 2018; Schrempf et al . , 2019 ) , the best fitting model of character evolution ( MK+FQ+GR+I ) ( Soubrier et al . , 2012; Yang et al . , 1995 ) , and the unrooted species tree for Atlantogenata . We considered ancestral state reconstructions to be reliable if they had Bayesian Posterior Probability ( BPP ) ≥ 0 . 80; less reliable reconstructions were excluded from pathway analyses . We note that there may be 'ghost' duplication events , that is genes that duplicated in , for example , the Tethytherian stem-lineage that are maintained in the Stellar’s sea cow genome and lost in the manatee genome . These genes will be reconstructed as a Proboscidean-specific duplication events because we cannot determine copy number in extinct species that lack genomes . To determine if gene duplications were enriched in particular biological pathways , we used the WEB-based Gene SeT AnaLysis Toolkit ( WebGestalt ) ( Liao et al . , 2019 ) to perform Over-Representation Analysis ( ORA ) using the Reactome database ( Jassal et al . , 2020 ) . Gene duplicates in each lineage were used as the foreground gene set , and the initial query set was used as the background gene set . WebGestalt uses a hypergeometric test for statistical significance of pathway over-representation , which we refined using two methods: a False Discovery Rate ( FDR ) -based approach and an empirical p-value approach ( Chen et al . , 2013 ) . The Benjamini–Hochberg FDR multiple-testing correction was generated by WebGestalt . In order to correct p-values based on an empirical distribution , we modified the approach used by Chen et al . in Enrichr ( Chen et al . , 2013 ) to generate a 'combined score' for each pathway based on the hypergeometric p-value from WebGestalt , and a correction for expected rank for each pathway . In order to generate the table of expected ranks and variances for this approach , we randomly sampled foreground sets of 10–5000 genes from our background set 5000 times , and used WebGestalt ORA to obtain a list of enriched terms and P-values for each run; we then compiled a table of Reactome terms with their expected frequencies and standard deviation . These data were used to calculate a Z-score for terms in an ORA run , and the combined score was calculated using the formula C=logp⋅z . The dramatic increase in body mass and lifespan in some Afrotherian lineages , and the relatively constant rate of cancer across species of diverse body sizes ( Abegglen et al . , 2015 ) , indicates that those lineages must have also evolved reduced cancer risk . To infer the magnitude of these reductions we estimated differences in intrinsic cancer risk across extant and ancestral Afrotherians . Following Peto , 2015 , we estimate the intrinsic cancer risk ( K ) as the product of risk associated with body mass and lifespan . In order to determine ( K ) across species and at ancestral nodes ( see below ) , we first estimated ancestral lifespans at each node . We used Phylogenetic Generalized Least-Square Regression ( PGLS ) ( Felsenstein , 1985; Martins and Hansen , 1997 ) , using a Brownian covariance matrix as implemented in the R package ape ( Paradis and Schliep , 2019 ) , to calculate estimated ancestral lifespans across Atlantogenata using our estimates for body size at each node . In order to estimate the intrinsic cancer risk of a species , we first inferred lifespans at ancestral nodes using PGLS ( Supplementary file 3 ) and the model . Next , we calculated K1 at all nodes , and then estimated the fold-change in cancer susceptibility between ancestral and descendant nodes ( Figure 2 ) . Next , in order to calculate K1 at all nodes , we used a simplified multistage cancer risk model for body size D and lifespan t: K≈Dt6 ( Peto et al . , 1975: Peto , 2015; Armitage , 1985; Armitage and Doll , 2004 ) . The fold change in cancer risk between a node and its ancestor was then defined as log2 ( K2K1 ) . All data analysis was performed using Python version 3 . 8 and R version 4 . 0 . 2 ( 2020-06-22 ) , and the complete reproducible manuscript , along with code and data generation pipeline , can be found on our GitHub page at https://github . com/docmanny/atlantogenataGeneDuplication ( Vazquez and Lynch , 2021; copy archived at swh:1:rev:6bc68ac31ef148131480710e50b0b75d06077db2; Paradis and Schliep , 2019; Paradis et al . , 2020; R Development Core Team , 2019; Xie , 2020; Bolker and Robinson , 2020; Dowle and Srinivasan , 2019; Wickham et al . , 2020a; Wickham , 2020; Harmon et al . , 2020; Yutani , 2020; Yu , 2020a; Campitelli , 2020; Wickham et al . , 2020b; Yu , 2020b; Kassambara , 2020; Slowikowski , 2020; Xiao , 2018; Yu and Lam , 2020c; Zhu , 2019; Ooms , 2020; Bache and Wickham , 2014; Pinheiro and Bates , 2020; Sievert et al . , 2020; Henry and Wickham , 2020; Wickham et al . , 2018; Hlavac , 2018; Wickham , 2019a; Müller and Wickham , 2020; Wickham and Henry , 2020; Yu , 2020d; Wickham , 2019b; Yu , 2020e; Gehlenborg , 2019; Xie , 2016; Alfaro et al . , 2009; Eastman et al . , 2011; Slater et al . , 2012; Harmon et al . , 2008; Pennell et al . , 2014; Wickham , 2016; Yu , 2020f; Yu et al . , 2018; Yu et al . , 2017; Sievert , 2020; Wickham et al . , 2019; Wang et al . , 2020 ) . All files necessary to reproduce the data in this manuscript are provided in Source data 1 . We manually verified the coding potential of the 16 genes shown in Figure 4 by first identifying the reciprocal best ( DNA sequence ) BLAT hits in the elephant and manatee genomes , which allowed us to determine conservation and presence of premature stop codons in the each open reading frame ( ORF ) . We translated the ORF for each hit into amino acid sequences and grouped up hits for each gene into one FASTA file along with the UniProt protein sequences for the human , dog , cat , and cow orthologs . Using a pipeline hosted at NGPhlyogeny . fr ( Lemoine et al . , 2019 ) , the homologs were aligned using MAFFT Katoh and Standley , 2013; the aligned sequences were cleaned using BMGE ( Criscuolo and Gribaldo , 2010 ) . Finally we used FastME ( Lefort et al . , 2015 ) to infer a gene tree for each duplicate . Alignments were then visually inspected for conservation and presence of premature stop codons . | From the gigantic blue whale to the minuscule bumblebee bat , animals come in all shapes and sizes . Any species can develop cancer , but some are more at risk than others . In theory , if every cell has the same probability of becoming cancerous , then bigger animals should get cancer more often since they have more cells than smaller ones . Amongst the same species , this relationship is true: taller people and bigger dogs have a greater cancer risk than their smaller counterparts . Yet this correlation does not hold when comparing between species: remarkably large creatures , like elephants and whales , are not more likely to have cancer than any other animal . But how have these gigantic animals evolved to be at lower risk for the disease ? To investigate , Vazquez and Lynch compared the cancer risk and the genetic information of a diverse group of closely related animals with different body sizes . This included elephants , woolly mammoths and mastodons as well as their small relatives , the manatees , armadillos , and marmot-sized hyraxes . Examining these species’ genomes revealed that , during evolution , elephants had acquired extra copies of ‘tumour suppressor genes’ which can sense and repair the genetic and cellular damages that turn healthy cells into tumours . This allowed the species to evolve large bodies while lowering their risk of cancer . Further studies could investigate whether other gigantic animals evolved similar ways to shield themselves from cancer; these could also examine precisely how having additional copies of cancer-protecting genes helps reduce cancer risk , potentially paving the way for new approaches to treat or prevent the disease . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"evolutionary",
"biology",
"cancer",
"biology"
] | 2021 | Pervasive duplication of tumor suppressors in Afrotherians during the evolution of large bodies and reduced cancer risk |
Many neurodegenerative diseases are linked to amyloid aggregation . In Huntington’s disease ( HD ) , neurotoxicity correlates with an increased aggregation propensity of a polyglutamine ( polyQ ) expansion in exon 1 of mutant huntingtin protein ( mHtt ) . Here we establish how the domains flanking the polyQ tract shape the mHtt conformational landscape in vitro and in neurons . In vitro , the flanking domains have opposing effects on the conformation and stabilities of oligomers and amyloid fibrils . The N-terminal N17 promotes amyloid fibril formation , while the C-terminal Proline Rich Domain destabilizes fibrils and enhances oligomer formation . However , in neurons both domains act synergistically to engage protective chaperone and degradation pathways promoting mHtt proteostasis . Surprisingly , when proteotoxicity was assessed in rat corticostriatal brain slices , either flanking region alone sufficed to generate a neurotoxic conformation , while the polyQ tract alone exhibited minimal toxicity . Linking mHtt structural properties to its neuronal proteostasis should inform new strategies for neuroprotection in polyQ-expansion diseases .
Huntington’s disease ( HD ) is an inherited neurodegenerative disease characterized by movement disorders , behavioral abnormalities , and brain atrophy ( Orr and Zoghbi , 2007; Ross et al . , 2014; Vonsattel and DiFiglia , 1998 ) . HD arises from mutations in Huntingtin ( Htt ) that expand a polyglutamine ( polyQ ) -encoding CAG repeat in exon 1 above a threshold length of 35 Qs ( MacDonald , 1993 ) . The length of the polyQ tract correlates with both disease severity and propensity to form amyloid aggregates ( Scherzinger et al . , 1999 ) . A link between neuronal toxicity and amyloid aggregation is further supported by post-mortem analyses of HD brains , which contain amyloid aggregates formed primarily by N-terminal exon 1 truncations of Htt ( Difiglia , 1997; Landles et al . , 2010; Mangiarini et al . , 1996; Ross and Poirier , 2004 ) . Such N-terminal fragments may arise from aberrant splicing at the Htt exon 1 junction or from caspase cleavage ( Sathasivam et al . , 2013; Wellington et al . , 2002 ) . Importantly , mutant Htt exon 1 carrying a polyQ expansion ( herein 'mHtt-Ex1' ) suffices to cause HD-like disease in animal models ( Goldberg et al . , 1996; Sathasivam et al . , 2013; Wellington et al . , 2002 ) and is thus widely used as a relevant model for HD biology and pathology . Despite the link between aggregation and neurodegeneration , the underlying neurotoxic species in HD and other amyloid-linked neurodegenerative diseases such as Alzheimer’s disease , Parkinson’s disease , and amyotrophic lateral sclerosis remains elusive . One model proposes that amyloid aggregates are toxic because they sequester and deplete essential cellular proteins such as transcription factors or molecular chaperones ( Kirstein-Miles et al . , 2013; Olzscha et al . , 2011 ) . However , a number of findings question the causal relationship between mHtt amyloid aggregates and toxicity . For instance , the medium spiny striatal neurons more vulnerable to HD toxicity show few to no aggregates in HD patient brains . In contrast , less affected neuronal cell-types contain many aggregates ( Kuemmerle et al . , 1999 ) . In addition , while transgenic mHtt-Ex1 is neurotoxic in mice , a longer N-terminal mHtt fragment encompassing exons 1 and 2 forms many aggregates in transgenic mice but exhibits no neuronal dysfunction ( Slow et al . , 2005 ) . Finally , longitudinal survival studies of primary neurons expressing fluorescently-tagged mHtt-Ex1 demonstrated that the formation of an amyloid inclusion correlated with neuronal survival ( Arrasate et al . , 2004 ) . Since fluorescence and EM imaging studies have suggested that mHtt can be sequestered in different types of cellular inclusions ( Caron et al . , 2014; Lu et al . , 2013; Nekooki-Machida et al . , 2009; Sahl et al . , 2016 ) one possible explanation for these observations is that only some inclusions are toxic , while others are protective . An alternative hypothesis proposes that toxicity resides in soluble oligomeric mHtt conformations ( Behrends et al . , 2006; Campioni et al . , 2010; Kim et al . , 2016; Miller et al . , 2011; Sun et al . , 2015 ) . Biophysical studies indicate that amyloidogenic proteins , including mHtt , not only form amyloid fibrils but also populate an array of ill-defined soluble , oligomeric conformations . Studies with conformationally-sensitive antibodies indicate that these states are conformationally highly heterogeneous ( Duim et al . , 2014; Kayed and Glabe , 2006; Nucifora et al . , 2012; Sontag et al . , 2012 ) . The transient and structurally diverse nature of these soluble species has hindered their characterization; accordingly , the nature and determinants involved in the formation of these oligomeric mHtt species remain elusive . It remains imperative to clarify the mHtt conformational landscape and link the various species – aggregates , oligomers , or even aberrant monomers – to proteotoxicity . We also must understand how various mHtt species engage the cellular protein homeostasis ( or 'proteostasis' ) pathways that clear aberrant and aggregation-prone conformations ( Martinez-Vicente et al . , 2010; Ravikumar et al . , 2004; Rubinsztein et al . , 2012; Rui et al . , 2015; Tam et al . , 2006; Tsvetkov et al . , 2013 ) . mHtt interactions with the chaperones and degradation pathways that handle misfolded proteins in the cell ( Balch et al . , 2008; Hartl et al . , 2011 ) are likely key determinants of the cellular balance of toxic and non-toxic conformers . Recent studies indicate that two domains flanking the expanded polyQ tract greatly influence mHtt aggregation and biology . N17 , the 17 amino acid N-terminal flanking domain , has been shown to enhance mHtt aggregation ( Tam et al . , 2009; Thakur et al . , 2009 ) . N17 also mediates mHtt interaction with chaperones , hosts many post-translational modifications regulating mHtt toxicity ( Gu et al . , 2009; Steffan , 2004; Thompson et al . , 2009 ) , and harbors a functional nuclear export sequence ( Maiuri et al . , 2013; Rockabrand et al . , 2007; Zheng et al . , 2013 ) . The C-terminal proline-rich flanking domain ( herein 'PRD' ) also binds cellular factors and influences mHtt toxicity in yeast ( Duennwald et al . , 2006; Gao et al . , 2014 ) . The PRD has also been shown to slow aggregation ( Bhattacharyya et al . , 2006; Tam et al . , 2009 ) . Importantly , an integrated understanding of the interplay between the biophysical and cellular modulation of mHtt by these flanking regions should provide insights into the nature of proteotoxicity . Here we combine biophysical and cell biological approaches to define how the domains flanking the polyQ tract modulate the ensemble of mHtt conformations in vitro and in vivo . Importantly , we link the mHtt conformational ensemble to mHtt proteostasis and toxicity in cultured neurons and brain slices . Biophysical and structural analyses demonstrate that N17 and PRD have opposing effects on the energetic barriers dictating the formation of aggregates and oligomers by mHtt both in vivo and in vitro . This interplay between N17 and PRD determines the formation of toxic mHtt conformations and their interaction with cellular proteostasis pathways . One corollary of our data is that neuronal mHtt toxicity cannot be explained by a simple model whereby amyloid fibrils or oligomers are toxic , but rather one that points to specific toxic conformational sub-populations . Our work linking the mHtt conformational landscape with neuronal proteostasis and toxicity informs rational avenues to leverage the roles of the polyQ flanking regions for HD therapeutics .
To evaluate the impact of N17 and PRD on the expanded polyQ tract , we created a set of mHtt-Ex1 deletion variants containing a pathogenic-length polyQ tract ( Q51 ) and lacking the N17 ( ∆N ) , PRD ( ∆P ) , or both N17 and PRD domains ( ∆N∆P ) . These variants were compared to an otherwise identical mHtt-Ex1 with the same pathogenic polyQ tract but containing both flanking domains ( Ex1 ) ( Figure 1A ) . The mHtt-Ex1 variants were recombinantly expressed and purified as soluble N-terminal GST fusion proteins ( Figure 1—figure supplement 1 ) ; aggregation is initiated by cleavage of the GST moiety as previously described ( Scherzinger et al . , 1999; Tam et al . , 2006 ) . 10 . 7554/eLife . 18065 . 003Figure 1 . Flanking regions impact pathogenic mHtt aggregation propensity in vitro and in striatal neurons . ( A ) Schematic representation of polyQ expanded mutant Htt-Ex1 variants ( mHtt-Ex1 ) used in this study . mHtt-Ex1 contains N17 ( red ) , an expanded polyQ tract ( grey ) , the proline-rich domain ( 'PRD' , blue ) , as well as a short , 10-amino acid C terminal tail ( 'C' , white ) . Variants were generated by deleting the regions flanking the polyQ domain . An additional C-terminal S tag ( not shown ) was used for immunoblot detection of recombinantly produced mHtt . For recombinant expression of all mHtt variants , a 51-mer polyQ tract was used . ( B ) Kinetics of formation of SDS-insoluble , heat-stable aggregates for the mHtt-Ex1 variants as measured by the filter trap assay . Aggregation of purified recombinant mHtt-Ex1 variants from ( A ) was initiated by cleavage of a solubilizing N-terminal GST tag by TEV protease . All aggregation reactions were performed at a concentration of 3 µM . Data is representative of at least three independent experiments . ( C ) Rate of accumulation of amyloid aggregates for the mHtt variants measured by the ThioflavinT fluorescence assay . All variants were aggregated at a concentration of 3 µM . ( D ) Normalized curves of ThioflavinT amyloid aggregation kinetics from ( C ) used to compare kinetic rates . Variants without the N17 region ( ∆N , ∆N∆P ) form amyloids much more slowly than variants with the N17 region ( Ex1 , ∆P ) . ( E ) Fluorescence images of each of the mHtt mutants transfected into the ST14a striatal neuron-derived cell line . mHtt variants were constructed similar as in ( A ) with a 51-mer polyQ length . Instead of a N-terminal GST and C-terminal S-tag , constructs had only a C-terminal GFP tag . Images were taken 48 hr post transfection . Scale bar is 20 µm . ( F ) Percentage of transfected cells containing aggregates for each of the Htt mutants as shown in ( E ) . Similar to as seen in vitro deletion of the N17 region leads to overall less aggregation while deletion of the PRD leads to overall more aggregation . Data are mean ± SEM of three independent experiments counting at least 150 cells . *p <0 . 05 , **p<0 . 005 , ***p<0 . 001 . ( G ) Summary model for how the N17 and PRD regions contribute to mHtt aggregation propensity . DOI: http://dx . doi . org/10 . 7554/eLife . 18065 . 00310 . 7554/eLife . 18065 . 004Figure 1—figure supplement 1 . Additional data and modeling of aggregation kinetics . DOI: http://dx . doi . org/10 . 7554/eLife . 18065 . 004 We initially characterized how these flanking regions contribute to the aggregation kinetics of purified mHtt through two complementary approaches . The filter trap assay detects large , SDS-insoluble aggregates through filtration through a 20 µm cellulose acetate membrane , ( Wanker et al . , 1999 ) . Formation of β-sheet rich amyloid structures was detected using the fluorescent dye ThioflavinT ( LeVine , 1999 ) . Consistent with previous reports , deletion of N17 reduces the rate and yield of amyloid formation and aggregation , while deletion of the PRD domain enhances formation of amyloids and aggregates ( Figure 1B–C , Figure 1—figure supplement 1B–C ) ( Crick et al . , 2013; Tam et al . , 2009; Thakur et al . , 2009 ) . ∆N∆P exhibited a complex , combined behavior of both N17 and PRD deletions . Similar to ∆P , the ∆N∆P mutant displayed enhanced aggregation rates measured in the filter trap assay ( Figure 1B ) . On the other hand , similar to ∆N , ∆N∆P had very slow kinetics of amyloid formation as measured by the ThioflavinT assay ( Figure 1C ) . Indeed , fitting the normalized ThioflavinT amyloid formation kinetics to the Finke-Watzky kinetic model ( Alvarez et al . , 2013; Morris et al . , 2009 ) showed that the elongation rate v for amyloid formation and half-time to saturation of amyloid formation t1/2 for mHtt variants lacking N17 ( ∆N and ∆N∆P ) were much slower than those for mHtt variants containing N17 ( Ex1 and ∆P ) ( Figure 1D , Figure 1—figure supplement 1D ) . We conclude that the presence of a PRD disfavors formation of large , SDS-insoluble aggregates while N17 exerts a dominant effect to promote the ThioflavinT-reactive , amyloid conformation . To relate these biophysical observations to mHtt-Ex1 behavior in a neuronal cellular environment , the equivalent mHtt-Ex1 variants were fused C-terminally to GFP and expressed in striatal neuron-derived ST14a cells ( Cattaneo and Conti , 1998 ) . Formation of GFP-inclusions provided a read-out for the aggregation propensity of the mHtt variants in vivo . As observed in vitro , deleting N17 reduced the formation of visible inclusions in vivo , while deleting PRD enhanced the formation of aggregates ( Figure 1E–F ) . Notably , few aggregates were visible in the ∆N∆P expressing cells , despite rapid formation of insoluble aggregates in vitro . Given the slow kinetics of amyloid aggregation by ∆N∆P in vitro , it is possible that in the absence of the N17 and PRD flanking regions , the polyQ tract does not efficiently generate amyloidogenic fibrils but instead forms non-amyloidogenic aggregates that are less stable in vivo ( Crick et al . , 2013 ) . We conclude that N17 and PRD have opposing effects of on amyloid formation and aggregation in vitro and in vivo ( Figure 1G ) and further suggest that the cellular environment destabilizes the non-amyloid aggregates generated by the polyQ tract in ∆N∆P . Next , we used cryo-electron microscopy ( cryo-EM ) to gain a structural understanding of how N17 and PRD impact the formation of mHtt amyloid fibrils . mHtt-Ex1 fibrils have a characteristic architecture , in which frayed fibril ends branch out from a bundled central core ( Figure 2A , Figure 2—figure supplement 1A ) ( Bugg et al . , 2012; Darrow et al . , 2015; Shahmoradian et al . , 2013 ) . For the ∆N mHtt variant , we observed dramatically fewer fibrils , consistent with its lower amyloid aggregation propensity ( Figure 1 ) . In addition , the fibrils formed by ∆N had a strikingly distinct morphology , which lacked the bundled architecture of Ex1 fibrils and were much thinner and straighter ( Figure 2—figure supplement 2 ) . Allowing ∆N aggregation to reach saturation by prolonged incubation increased the number of fibrils but did not change their thin morphology ( Figure 1C , Figure 2—figure supplement 1B ) . Thus , the thin fibril structure of ∆N aggregates is intrinsic to the mutation . In contrast , ∆P formed many large , densely packed aggregates with individual fibrils arranged in parallel bundles ( Figure 2A , Figure 2—figure supplement 2 ) , consistent with its increased aggregation propensity . As observed for kinetic measurements , the morphology of ∆N∆P aggregates combined properties from both the ∆N and ∆P fibrils . Similar to ∆N fibrils , the ∆N∆P fibrils were shorter , thinner and lacked the frayed fibril ends observed for Ex1 ( Figure 2—figure supplement 2 ) ; similar to ∆P fibrils , ∆N∆P aggregates consisted of more densely packed fibrils ( Figure 2A ) . Quantification of at least 10 individual micrographs for each fibril variant supported these observations , indicating that ∆N fibrils were only several nanometers in width , whereas Ex1 and ∆P fibrils were on average almost a micron wide and over a micron long ( Figure 2—figure supplement 2 ) . We conclude that N17 and PRD have independent and dramatic effects on the amyloid formation propensity of the polyQ tract and its fibrillar structure . N17 promotes amyloid formation but also enhances interfibrillar contacts , thereby driving the bundling of individual fibrils observed for Ex1 and ∆P ( Figure 2B ) . In contrast , PRD appears to destabilize lateral contacts among fibrils and thus prevents their dense packing ( Figure 2B ) . The combined effects of N17 and PRD in Ex1 lead to the characteristic mHtt structure consisting of fibril bundles with frayed ends ( Figure 2C ) . When N17 is absent , PRD leads to thin sparse fibrils observed for ∆N; while when PRD is absent , N17 leads to thick , bundled aggregates as observed for ∆P ( Figure 2C ) . 10 . 7554/eLife . 18065 . 005Figure 2 . Morphology of mHtt aggregation highlight how N17 and PRD contribute to mHtt aggregation . ( A ) Morphology of fiber aggregates from each of the mHtt mutants by cryo-EM . mHtt fibers were imaged 24 hr post initiation of aggregation . Scale bar is 500 nm . Top row is the cryo-EM micrographs . The second and third rows are traced annotations of the EM micrographs and zoom-ins of individual areas , respectively . Bottom row is the original cryo-EM micrograph of the zoomed-in area . ( B ) Summary of the N17 and PRD contributions to aggregation morphology and propensity . ( C ) Summary model for how the N17 and PRD regions direct morphology of mHtt aggregates . Morphology differences between ∆N∆P and Ex1 showcase how the flanking regions impact mHtt aggregation in a combinatorial manner . DOI: http://dx . doi . org/10 . 7554/eLife . 18065 . 00510 . 7554/eLife . 18065 . 006Figure 2—figure supplement 1 . Additional cryo-EM images of mHtt variant aggregate morphology . DOI: http://dx . doi . org/10 . 7554/eLife . 18065 . 00610 . 7554/eLife . 18065 . 007Figure 2—figure supplement 2 . Quantification of mHtt variant aggregate morphology . DOI: http://dx . doi . org/10 . 7554/eLife . 18065 . 007 Our model predicts that PRD destabilizes the interfibrillar contacts within an aggregate . This effect may be of importance for HD in light of the prion hypothesis that postulates that intercellular transmission of aggregate 'seeds' can nucleate aggregation in naïve cells ( Pearce et al . , 2015 ) . For prions , distinct fibrillar amyloid stability is a hallmark of different prion strains ( Tanaka et al . , 2006 ) as their reduced stability ( or increased 'frangibility' ) is directly linked to their propensity for intercellular transmission ( Cushman et al . , 2010 ) . We thus tested if the polyQ flanking regions change the mechanical stability of the mHtt amyloid aggregates . Amyloid aggregates of all mHtt variants were gently isolated by centrifugation and subjected to different sonication conditions to test their stability to mechanical disruption ( Figure 3A , Figure 3—figure supplement 1A ) . The size and morphology of the resulting species were measured by Dynamic Light Scattering ( DLS ) and cryoEM , respectively ( Figure 3B–C , Figure 3—figure supplement 1B ) . 10 . 7554/eLife . 18065 . 008Figure 3 . The PRD region mechanically destabilizes mHtt aggregates . ( A ) mHtt fibers were isolated by centrifugation and sonicated to test their mechanical stability . Sonicated species were analyzed by cryo-EM imaging and Dynamic Light Scattering ( DLS ) . ( B ) Size comparisons between sonicated mHtt aggregates lacking PRD ( ∆N∆P , ∆P ) and mHtt variants with the PRD ( Ex1 , ∆N ) as measured by DLS . Sonication of mHtt variants containing the PRD results in much smaller fiber fragments than those lacking PRD . Data are mean ± SEM . ( C ) Cryo-EM images of sonicated fibers for mHtt variants lacking PRD ( ∆N∆P , ∆P ) or mHtt variants with the PRD ( Ex1 , ∆N ) . Scale bar is 100 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 18065 . 00810 . 7554/eLife . 18065 . 009Figure 3—figure supplement 1 . DLS data of sonicated mHtt variants at harsher conditions . DOI: http://dx . doi . org/10 . 7554/eLife . 18065 . 00910 . 7554/eLife . 18065 . 010Figure 3—figure supplement 2 . Comparison of Ex1 and ∆P susceptibility to urea and formic acid denaturation . DOI: http://dx . doi . org/10 . 7554/eLife . 18065 . 010 DLS analyses confirmed that PRD significantly reduced the mechanical stability of mHtt aggregates . mHtt fibrils formed by variants containing a PRD , namely Ex1 and ∆N , were readily fragmented into small , relatively homogeneous seeds of 30–50 nm in size ( Figure 3B–C ) . In contrast , aggregates formed by mHtt variants lacking a PRD , namely ∆P and ∆N∆P , resisted disruption and remained in large and highly bundled fibrillar structures ranging from 0 . 3–2 µm even at the harshest sonication conditions ( Figure 3B–C , Figure 3—figure supplement 1 ) . Cryo-EM imaging confirmed and extended these observations ( Figure 3C ) . The sonicated seeds of Ex1 and ∆N were small structures only one or two fibrils in width , while the sonication products of ∆P and ∆N∆P were large , thick , and densely-packed bundles of fibers . Of note , the presence or absence of N17 had no significant impact on the size of the fragmented species , indicating that while N17 contributes to the formation of these fibrils , it does not contribute to their mechanical stability . We conclude that the PRD mechanically destabilizes mHtt aggregates by disfavoring inter-fibrillar contacts ( Figure 2A ) , thus providing a molecular basis for reduced aggregation propensity of mHtt variants with a PRD ( Figure 1 ) . We next examined whether the PRD domain also contributed to the chemical stability of mHtt fibrils to protein denaturating agents , such as urea or formic acid . The isolated amyloid aggregates of Ex1 and ∆P were initially treated with 8M urea , which was unable to disaggregate either fibril variant ( Figure 3—figure supplement 2A–B ) . We then subjected urea-treated fibrils with increasing concentrations of the much harsher denaturant formic acid ( Figure 3—figure supplement 2A–B ) . Both Ex1 and ∆P fibrils were similarly resistant to intermediate concentrations of formic acid and only following incubation with 100% formic acid , which can dissolve mHtt aggregates formed in vivo ( Hazeki et al . , 2000 ) , did we observe significant denaturation of the mHtt fibrils of both Ex1 and ∆P ( Figure 3—figure supplement 2B–C ) . These experiments suggest that the mechanical stability differences for fibrils with or without the PRD derives from their structural architecture rather than intrinsic differences in their chemical stability . We next examined how the polyQ flanking regions impact the conformational ensemble of soluble , oligomeric mHtt species by monitoring the kinetics of formation and stability of soluble , oligomeric species by mHtt flanking region variants ( Haass and Selkoe , 2007; Nucifora et al . , 2012; Sontag et al . , 2012 ) ( Figure 4A ) . At different times following TEV cleavage of the GST moiety , agarose gels ( AGEs ) were used to examine the formation of oligomeric species larger than 400 kDa ( Figure 4B–C ) ( Sontag et al . , 2012; Weiss et al . , 2008 ) while Blue Native acrylamide gels ( Blue-Native PAGE ) were used to monitor species smaller than 400 kDa ( Figure 4D ) . Oligomeric species larger than 400 kDa were resolved under native ( Figure 4B , Figure 4—figure supplement 1A ) or mildly denaturing ( 0 . 1% SDS ) conditions ( Figure 4C , Figure 4—figure supplement 1B ) . At each time-point , the filter trap assay was used to assess formation of large SDS-insoluble , amyloid aggregates ( Figure 4C , filter trap ) . 10 . 7554/eLife . 18065 . 011Figure 4 . N17 and PRD direct the formation and stabilization of oligomer populations . ( A ) Schematic of oligomer and aggregate populations generated during the mHtt aggregation pathway . We characterized the oligomer populations using Agarose Gel Electrophoresis ( AGE ) under either native or mildly denaturing , 0 . 1% SDS conditions or Blue-Native PAGE under native conditions . Oligomers were isolated by taking time-points of the in vitro mHtt aggregation reaction , conducted at the same conditions as in Figure 1 . ( B ) Native-AGE showing native , large oligomers for the mHtt variants . ∆N generated a large population of mHtt oligomeric species that persisted through the aggregation reaction , whereas ∆P generated very few oligomeric species that disappeared quickly from the gel . Blot was immunoprobed for the C-terminal S-tag on the mHtt variants . ( C ) 0 . 1% SDS-AGE gel showing partially SDS-soluble , oligomers over 400 kDa . Blot was immunoprobed for the C-terminal S-tag on the mHtt variants . Large , SDS-insoluble mHtt aggregates do not enter the SDS-AGE gel , as shown by the reference filter trap ( top row , filter trap ) . ∆N generates few aggregates because the protein remains trapped in the oligomeric range . ( D ) Blue-Native PAGE gel showing native , small oligomers below 400 kDa . Aggregation time-points were run in a Blue-Native PAGE gel and immunoprobed for the S-tag ( i ) and the 3B5H10 conformational antibody ( ii ) . 3B5H10 only recognizes the smaller oligomers of the mHtt variants Ex1 , ∆N , and ∆P . ( E ) Energy landscape model of the different conformational species in the mHtt-Ex1 aggregation pathway ( left ) . Aggregation through the ∆N or ∆P pathways changes the energetic barriers between the oligomeric and fibrillar states of mHtt ( right ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18065 . 01110 . 7554/eLife . 18065 . 012Figure 4—figure supplement 1 . Additional 0 . 1% SDS-AGE and Native-AGE assays of mHtt-Ex1 variants . DOI: http://dx . doi . org/10 . 7554/eLife . 18065 . 012 Native-AGE gel analyses revealed that native oligomers of Ex1 appeared early during aggregation kinetics and gradually decreased as aggregation proceeded to form the aggregates retained in the filter trap ( Figure 4B–C ) ( Sontag et al . , 2012 ) . In contrast , large SDS-soluble oligomers appeared to increase over time in the SDS-AGE gel ( Figure 4C ) . By contrast , ∆P had much fewer oligomers in both native- and SDS-AGE gels . Furthermore , the native oligomers disappeared much more rapidly during the time-course of aggregation that those observed for Ex1 , consistent with the overall faster aggregation of ∆P . Thus , the absence of PRD alters the balance between oligomer and fibril formation , perhaps by allowing faster progression of these oligomeric species to aggregates ( Figure 1B-C , 4D ) . Surprisingly , very large ∆N oligomers persisted through the aggregation reaction in both the native-AGE and SDS-AGE gels ( Figure 4B–C ) . Even after 60 hr , when ∆N amyloid aggregation was saturated , the population of ∆N oligomers remained under both native and mildly denaturing conditions ( Figure 4—figure supplement 1B ) . Consistent with our previous data , the increased presence of ∆N oligomers resulted in few filter-trap retained aggregates ( Figure 1B–C , 4C ) . We wondered if these stable , large ∆N oligomers corresponded to the small , thin ∆N amyloid fibrils observed by cryo-EM ( Figure 2A ) . However , while large ∆N oligomers were already abundant as early as 3 hr post-aggregation ( Figure 4B–C ) , cryo-EM imaging of ∆N at 3 hr of aggregation showed very few fibrils and no additional identifiable densities or structures ( not shown ) . Furthermore the amyloidogenic , ThioflavinT-reactive species formed by ∆N only accumulated significantly after 30 hr of aggregation ( Figure 1C ) . Together , these data indicate that ∆N becomes trapped forming stable soluble oligomers that may be mildly SDS-resistant but are neither fibrillar nor amyloid . These ∆N oligomers , despite being very large in size , appear to be too structurally heterogeneous to yield identifiable diffraction signatures by cryoEM . We next examined the impact of the polyQ flanking regions on the formation of soluble oligomer populations smaller than 400 kDa ( Nucifora et al . , 2012; Schagger and von Jagow , 1991 ) ( Figure 4D ) . The aggregation reaction was initiated by TEV cleavage of the GST-moieties and time-points analyzed by Blue-Native PAGE ( Figure 4Di ) . Interestingly , all mHtt variants formed several small oligomer species very early in the aggregation reaction . For Ex1 and ∆P , these small oligomers disappeared as aggregation progressed , while the species of ∆N and ∆N∆P were very stable , similar to that observed for large oligomers in SDS-AGE gels ( Figure 4C ) . This analysis supports the view that N17 promotes the formation of fibrils ( Figure 2 ) at the expense of oligomer formation , whereas the PRD contributes to the accumulation of oligomers . Interestingly , N17 and PRD seem to act independently in oligomer formation since the oligomers formed by ∆N∆P exhibited a behavior intermediate between the ∆N and ∆P phenotypes in the AGE oligomer assays . While a significant amount of ∆N∆P species progressed to the insoluble aggregates ( Figure 4C , filter trap ) , a population of trapped oligomers persisted in the SDS-AGE gel and in the Blue-Native PAGE gel . It thus appears that N17 and PRD have independent and antagonistic effects on the conformational landscape of the polyQ ( Figure 4E ) . N17 promotes amyloid formation and fibril bundling and prevents the accumulation of non-amyloidogenic oligomeric species . On the other hand , PRD disfavors aggregation by enhancing the levels of soluble oligomeric species while structurally destabilizing the amyloid fibril association . We next considered whether the flanking regions also influence the structural properties of the oligomers , as observed for the fibrils . Indeed , their distinct migration patterns in Native-AGE and Blue-Native PAGE , which separate oligomers based on a number of shape , size and exposed charge characteristics , seems to support this hypothesis . Unfortunately , characterizing these oligomeric species by conventional structural approaches such as electron microscopy , crystallography or NMR is technically challenging given that these oligomeric species are structurally heterogeneous and chemically unstable . Conformational antibodies have provided a useful approach to distinguish specific conformations of amyloidogenic proteins , albeit the characteristics of these reactive species are not often defined ( Brooks et al . , 2004; Kayed and Glabe , 2006; Kayed et al . , 2003 ) . These conformational antibodies have also been proposed to recognize more toxic conformations of amyloidogenic species ( Kayed et al . , 2003; Miller et al . , 2011 ) . We tried a panel of conformationally-sensitive antibodies to identify conformational subsets in the oligomer populations separated in our AGE and PAGE analyses . We failed to observe significant reactivity of the oligomers of any flanking domain variant with A11 or OC conformationally-sensitive antibodies ( not shown ) . Interestingly , we did observe reactivity with the 3B5H10 antibody , previously proposed to recognize a toxic polyQ conformation ( Miller et al . , 2011 ) . The 3B5H10 reactive species were small oligomers resolved by the Blue-Native PAGE gels , while none of the larger oligomers in the SDS-AGE gel were 3B5H10 reactive ( Figure 4—figure supplement 1C ) ( Miller et al . , 2011 ) . Interestingly , 3B5H10 reactivity did support the hypothesis that the flanking regions influence the structural conformation of the oligomers . We observed that Ex1 , ∆N and ∆P all formed small 3B5H10-reactive oligomers , albeit of different sizes and mobilities . In contrast , the small oligomers formed by ∆N∆P , which had similar mobility to those formed by ∆N , were not 3B5H10 reactive ( Figure 4Dii ) . This suggests that the conformation adopted by the polyQ tract within oligomers is influenced by the flanking regions . Thus , the flanking regions influence structural differences in not only the fibrillar aggregate species but also in soluble species affecting both the stabilities and conformations of the heterogeneous species formed along the mHtt aggregation pathway . In considering the relationship between the polyQ flanking regions and oligomer populations , the above results suggest that N17 plays a role in promoting the oligomer to fibril transition ( Figure 5A ) . Indeed , previous studies showed that trans addition of an N17 peptide to ∆N or even to Ex1 enhanced formation of filter-trappable aggregates ( Tam et al . , 2009 ) . We thus tested if N17 acts by converting 'kinetically trapped' ∆N oligomers to an amyloid-competent conformation that proceeds to fibrillar aggregates . To this end , we added the N17 peptide in trans to a ∆N aggregation reaction and examined the aggregates by cryoEM ( Figure 5A–B ) . The ∆N fibers formed without N17 peptide were sparse , thin , and short , as expected ( Figure 5B ) and no aggregates were observed for N17 alone ( not shown ) . Strikingly , we observed that trans addition of the N17 peptide led to a remarkable increase in the amount of ∆N fibrils ( Figure 5B ) . Furthermore , the N17-treated ∆N fibrils exhibited the distinctive bundled morphology observed for Ex1 ( Figure 5B , Figure 5—figure supplement 1A , Figure 2A ) . 10 . 7554/eLife . 18065 . 013Figure 5 . N17 is essential to promote transition of oligomers to amyloid aggregates . ( A ) General schematic of trans N17 addition experiments to ∆N aggregation reaction to determine how N17 impacts the balance between stable , ∆N oligomers and amyloid fibrillar aggregates . ( B ) 12 . 5x excess of N17 peptide was added when ∆N aggregation was initiated . Then 60 hr after initiation of aggregation , the fibers were analyzed by cryo-EM . N17-seeded ∆N aggregates become more bundled and more resemble the morphology of Ex1 aggregates ( Figure 2A ) . ( C ) 12 . 5x or 2 . 5x excess of N17 peptide was added in trans when ∆N aggregation was initiated . The impact of N17 peptide on ∆N oligomers was analyzed by 0 . 1% SDS-AGE gels and immunoprobed for the C-terminal S-tag . Trans addition of N17 promotes disappearance of these stable ∆N oligomers from the AGE gel . ( D ) 12 . 5x excess of N17 peptide was added in trans 7 hr after initiation of ∆N aggregation , allowing for the formation of ∆N oligomers before N17 addition . Impact of N17 peptide on ∆N oligomers was analyzed by 0 . 1% SDS-AGE gels and immunoprobed for the C-terminal S-tag . ( E ) Schematic of in vivo N17 addition experiment: ST14a striatal-derived neurons were transfected with the C-terminally GFP tagged mHtt-Ex1 ( mHtt-Ex1-GFP ) . 10 hr after transfection , when expressed mHtt protein is still completely soluble , cells are protein transfected using the Xfect kit ( Clontech ) with N17 or a mutant N17 peptide ( NA ) that inhibits aggregation . After 10 hr , resulting cells containing GFP fluorescent puncta are counted . ( F ) Fluorescence microscopy of ST14a striatal neurons transfected with mHtt-Ex1-GFP and N17 variant peptides ( left ) and quantification of cells containing puncta ( right ) . Data are mean ± SEM of three independent experiments with at least 200 cells counted in each condition . Scale bar is 20 µm . *p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 18065 . 01310 . 7554/eLife . 18065 . 014Figure 5—figure supplement 1 . Additional cryo-EM images of ∆N aggregates with N17 peptide added in trans and transfection of N17-peptide into neurons expressing Htt-Q25 . DOI: http://dx . doi . org/10 . 7554/eLife . 18065 . 014 The effect of trans addition of N17 on the trapped ∆N oligomer populations was next examined by the addition of two different N17 peptide concentrations to a ∆N aggregation reaction . Oligomer formation was assessed by SDS-AGE and immunoblotting against ∆N ( Figure 5C ) . The addition of N17 at the start of the aggregation reaction led to the disappearance of ∆N oligomers ( Figure 5C ) , consistent with the increase in aggregates observed by EM and filter trap ( Figure 5B ) ( Tam et al . , 2009 ) . The rate of oligomer disappearance was enhanced at higher N17 concentrations ( Figure 5C ) . Thus , it seems that N17 triggers a conformational switch that disfavors the accumulation of trapped , oligomeric species and allows the formation of amyloid fibrils . Such a role explains the slow ∆N amyloid formation observed in the kinetic and cryo-EM analyses ( Figure 1B–C , Figure 2A ) . We next asked whether N17 could impact pre-formed , stable ∆N oligomers . We incubated ∆N for 7 hr to generate a large population of oligomers but no fibrils , then added N17 in trans ( Figure 5D ) . Strikingly , N17 also promoted disappearance of these pre-formed , non-amyloidogenic oligomers ( Figure 5D ) , suggesting that N17 can act on the kinetically trapped ∆N oligomers themselves to promote formation of amyloid fibrils ( Figure 5B ) . The ability of the N17 peptide to modulate aggregation in vivo was next examined following mHtt-ex1-GFP transfection in ST14a striatal-like neurons ( Figure 5E ) . At 10 hr post-mHtt transfection , when mHtt-Ex1 is still diffusely localized in all cells , we transfected the cells with either the N17 peptide or an N17 peptide mutant that cannot promote aggregation in vitro ( 'NA-N17' ) ( Tam et al . , 2009 ) ( Figure 5E ) . We next followed the formation of GFP-positive aggregates in these cells versus a buffer control . While no significant differences were observed between Ctrl-treated and NA-N17 treated cells , we observed a significant increase in the number of cells with mHtt-GFP inclusions in the N17-transfected cells ( Figure 5F ) . Of note , N17 did not promote inclusion formation in ST14a cells expressing the non-pathogenic Q-length Htt-Ex1-Q25-YFP ( Figure 5—figure supplement 1B ) . These results indicate that , N17 trans addition can increase mHtt aggregation in vivo . Since N17 and PRD exert opposing effects on the mHtt conformational landscape , they present an opportunity to test the competing models that neuronal toxicity is caused by the formation of amyloid aggregates or the formation of oligomers . Specifically , ∆N forms many oligomers and few fibrillar aggregates; ∆P forms many fibrillar aggregates and few oligomers; and ∆N∆P , which is essentially a pathogenic length Q-tract , forms both oligomers and fibrils ( Figure 6A ) . To evaluate the toxicity of these polyQ expanded mHtt variants we looked at the striatal neurons of rat corticostriatal brain slices , which provide a more disease relevant , 'tissue contextual' model of HD when compared to traditional models of neuronal cell cultures ( Khoshnan et al . , 2004; Reinhart et al . , 2011 ) . In corticostriatal brain slices , the striatal medium spiny neurons ( MSNs ) , which are preferentially affected in HD , remain within their intact local tissue environment and maintain interactions among multiple brain cell-types . mHtt toxicity was measured by co-transfecting corticostriatal brain slices with a plasmid encoding untagged mHtt-Ex1 variants and another plasmid encoding YFP , which serves as an independent morphological marker for the transfected neurons ( Crittenden et al . , 2010 ) ( Figure 6B ) . MSN viability is measured by examining the cellular and dendritic morphology by YFP expression ( Figure 6C–E ) . Importantly , when corticostriatal brain slices are transfected with polyQ-expanded mHtt there is a progressive degeneration of MSNs over 3–5 days ( Reinhart et al . , 2011 ) . No degeneration is observed when YFP is co-transfected with Htt carrying non-pathogenic polyQ repeats ( Q8 , Q23 ) or with a control vector ( Reinhart et al . , 2011 ) . 10 . 7554/eLife . 18065 . 015Figure 6 . Assessing proteotoxicity of mHtt variants in corticostriatal brain slices . ( A ) Compared to Ex1 , ∆N forms many more oligomers and much fewer fibrillar aggregates , whereas ∆P forms many more fibrillar aggregates and much fewer oligomers . ( B ) Schematic of toxicity assay: Corticostriatal slices were prepared from rat brains and biolistically co-transfected with mHtt-Ex1 variants or an Empty Vector control ( EV ) and an independent YFP plasmid as a marker of cell viability . Viability of medium spiny neurons ( MSNs ) was visually assessed by cell morphology and YFP expression after 3–4 days post-transfection . ( C ) Fluorescence images of MSNs from the brain slices co-transfected with the mHtt variants and YFP . Images were taken 1 day ( top row ) and 4 days ( bottom row ) after transfection . Data are representative of at least three independent experiments . Scale bar is 20 µm . ( D ) Relative viability of MSNs transfected with mHtt variants 3 days after transfection . Data are mean ± SEM . Data are representative of at least three independent experiments . **p<0 . 0001 ( E ) Relative viability of MSNs transfected with mHtt variants 4 days after transfection . Data are mean ± SEM . Data are representative of at least three independent experiments . *p<0 . 01 , **p<0 . 0001DOI: http://dx . doi . org/10 . 7554/eLife . 18065 . 01510 . 7554/eLife . 18065 . 016Figure 6—figure supplement 1 . Transfected Htt-Ex1 is only toxic in transfected MSNs at pathogenic Q-lengths . DOI: http://dx . doi . org/10 . 7554/eLife . 18065 . 016 Surprisingly , when the mHtt-Ex1 variants were transfected into brain slices , there were striking discrepancies between their toxicity and their structural properties . Although ∆N and ∆P have dramatically different aggregation and oligomerization propensities in vitro and in vivo , both were as toxic as Ex1 as early as 3 days after transfection ( Figure 6C–E ) . Even more surprisingly , ∆N∆P , which generates both aggregates and oligomers and has an expanded polyQ tract of identical length , exhibited only modest toxicity even by day 4 ( Figure 6E ) . These results negate simple models of aggregate-only or oligomer-only toxicity and instead call for a more nuanced view of proteotoxicity . Since both ∆N , which forms few amyloid fibrils , and ∆P , which forms few oligomers are both highly toxic , our data cannot be explained by either fibrillar aggregates or , oligomeric species per se as being toxic . Remarkably , since the expanded polyQ tract itself is only mildly toxic , the action of the flanking regions must be key to induce the toxic polyQ conformation ( s ) . Our unexpected toxicity findings suggest a number of hypotheses . First , it is possible that there is no universal proteotoxic species , and either specific fibrillar or oligomer states mediate toxicity , most likely through different mechanisms given their radically different properties . However , it was intriguing that mHtt toxicity in the brain slices was observed for those variants that gave rise to 3B5H10-reactive small oligomers , namely Ex1 , ∆N and ∆P but not ∆N∆P . Thus , an alternative is that toxicity resides in a subclass of structural conformers , such as oligomers linked to 3B5H10 reactivity , rather than a particular size of species . Yet another possibility is that mHtt interconverts between several conformations when expressed in vivo and that toxicity arises when those species cannot be efficiently cleared by the cellular machinery . We thus extended our analysis to the soluble mHtt conformers formed in striatal neurons , as well as their interactions with the proteostasis machinery . To understand how the neuronal cellular environment impacts the mHtt conformational ensemble , we tested whether in vivo the 3B5H10-reactive conformation was also specifically observed for the toxic variants Ex1 , ∆N and ∆P , but not for the less toxic ∆N∆P ( Figure 7A–B ) ( Brooks et al . , 2004; Kayed and Glabe , 2006; Kayed et al . , 2003 ) . Indeed , immunofluorescence analysis showed that striatal-derived neurons expressing the toxic variants Ex1 , ∆N and ∆P contained a diffusely localized 3B5H10-reactive species that was absent in cells expressing the less toxic ∆N∆P ( Figure 7B ) . Little 3B5H10 reactivity was observed in neurons expressing Htt-Ex1 construct with a non-pathogenic polyQ length ( Figure 7—figure supplement 1A ) , similar to a previous report ( Miller et al . , 2011 ) . Of note , 3B5H10 did not stain large mHtt aggregates or inclusions ( Figure 7C ) , suggesting that in vivo the 3B5H10-reactive conformation is excluded from large aggregates . 10 . 7554/eLife . 18065 . 017Figure 7 . Impact of PolyQ flanking domains of the formation of soluble oligomers in vivo . ( A ) We investigate the impact of the neuronal environment on the mHtt conformational landscape . ( B ) Confocal imaging of ST14a striatal-derived neurons transfected with mHtt-Ex1 variants C-terminally tagged with GFP . Cells were imaged in the GFP channel well as immunostained with the 3B5H10 polyQ conformational antibody . Scale bar is 20 μm . ( C ) Confocal images of Ex1 and ∆P-transfected ST14a cells containing puncta . Cells were imaged in the GFP channel as well as immunostained with the 3B5H10 polyQ conformational antibody . The 3B5H10 antibody shows no specific recognition of the GFP puncta . Scale bar is 10 μm . ( D ) Immunoblots of lysates from ST14a striatal neurons transfected with mHtt-Ex1 variants C-terminally tagged with GFP and run in Blue-Native PAGE gels . Blots were probed for GFP ( left ) and the 3B5H10 conformational antibody ( right ) . ( E ) Schematic for ApiCCT1 addition experiment: Corticostriatal brain slices ( i ) and ST14a striatal-derived neurons ( ii ) were transfected with mHtt-Ex1 simultaneously with exogenous addition of ApiCCT1 to media . Then , viability of MSNs in the corticostriatal brain slices was measured ( i ) and 3B5H10-positive small oligomers in the striatal neuron extracts were characterized ( ii ) . Data are mean ± SEM . **p<0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 18065 . 01710 . 7554/eLife . 18065 . 018Figure 7—figure supplement 1 . Immunofluorescence images and 0 . 1% SDS-AGE of ST14a cells transfected with Htt-Ex1 variants and immunoprobed with 3B5H10 . DOI: http://dx . doi . org/10 . 7554/eLife . 18065 . 018 We further examined the soluble mHtt conformations generated in vivo by subjecting the ST14a neuronal extracts to SDS-AGE and Blue-Native PAGE analyses , followed by GFP and 3B5H10 immunoblotting ( Figure 7D ) . GFP immunoblotting of the Blue-Native PAGE revealed variant-specific single bands , suggesting that , unlike what we observed in vitro each variant forms a predominant small oligomeric species in vivo ( Figure 7D ) . In contrast , the larger mHtt oligomers observed in vivo displayed a heterogeneous spectrum of sizes as observed with purified mHtt . Of note , the only 3B5H10-reactive species formed in vivo were the small mHtt oligomers detected in the Blue-Native PAGE ( Figure 7D ) . None of the larger oligomeric species detected in SDS-AGE reacted with 3B5H10 ( Figure 7—figure supplement 1B ) . Notably , as seen in vitro , 3B5H10 only recognized the small oligomers formed by the toxic variants Ex1 , ∆N , and ∆P but not the oligomers formed by ∆N∆P ( Figure 7D ) . We conclude that N17 and PRD influence the kinetics and stabilities of oligomers during aggregation and also modulate the toxic conformations adopted by the polyQ stretch in these oligomers . The polyQ tract itself can clearly generate different oligomer populations , but formation of oligomer conformations reactive to 3B5H10 requires either the N17 or PRD domain . The similar 3B5H10 reactivity of Ex1 , ∆N , ∆P and ∆N∆P in vitro and in vivo suggests that the conformational differences among these mHtt variants is intrinsic to the structural influence of the flanking regions on the polyQ region . While it is tempting to speculate that these soluble , 3B5H10-reactive small oligomers are by themselves the toxic Htt species , it is possible that additional toxic species exist but are not recognized by any existing conformational antibody . Nonetheless , these experiments demonstrate that mHtt variants that produce neurotoxic species share specific conformers that are disfavored in the ensembles formed by the less toxic , expanded polyQ tract alone . We analyzed how known protective proteostasis mechanisms impact the presence of small oligomers in vivo . One established mechanism is the chaperonin TRiC/CCT , which suppresses mHtt aggregation and toxicity in many HD models through recognition of N17 by the CCT1 subunit ( Behrends et al . , 2006; Kitamura et al . , 2006; Tam et al . , 2006; Tam et al . , 2009 ) . In neuronal cell culture , overexpression of CCT1 or exogenous addition of the purified substrate-binding domain of CCT1 ( ApiCCT1 ) to the media suffices to suppress mHtt toxicity ( Kitamura et al . , 2006; Sontag et al . , 2013; Tam et al . , 2009 ) . We thus examined if ApiCCT1 could also be neuroprotective in the corticostriatal brain slice toxicity model of HD ( Figure 7E ) . Indeed , exogenous administration of purified recombinant ApiCCT1 significantly protected MSNs from mHtt neurotoxicity in a concentration-dependent manner ( Figure 7Ei ) . Strikingly , Blue-Native PAGE revealed that this exogenous ApiCCT1 addition caused a concentration-dependent reduction in the levels of small 3B5H10-reactive oligomers implicated in toxicity ( Figure 7Eii ) . These experiments support the notion that these 3B5H10-reactive small oligomers are diagnostic of a conformational pathway linked to toxicity and that chaperonin protection against HD toxicity decreases the levels of these soluble species . The toxicity of mHtt populations depends not only on their biophysical properties but also their interactions with cellular proteostasis pathways ( Kim et al . , 2016 ) , which may also depend on the influence of the polyQ flanking domains ( Figure 8A ) . We first used corticostriatal brain slices to investigate the impact of the different polyQ flanking domain variants on TRiC-mediated neuroprotection . CCT1 was cotransfected with the toxic mHtt-Ex1 variants into the corticostriatal brain slices and MSN viability assessed as above ( Figure 8B , Figure 8—figure supplement 1A ) . As expected , overexpression of CCT1 dramatically protected MSNs from mHtt-Ex1 toxicity but did not protect MSNs from ∆N-induced toxicity . These results are consistent with the notion that N17 hosts the CCT1 binding site in mHtt ( Tam et al . , 2009 ) . Surprisingly , despite the presence of N17 , CCT1 afforded only a modest degree of protection for ∆P-induced toxicity . The minimal rescue of ∆P by CCT1 may be due to the aggressive neurotoxicity of ∆P ( Figure 6C–E ) , which may impair neurons too quickly to be effectively rescued by CCT1 . Alternatively , we considered that the absence of the PRD domain may reduce the availability of N17 to engage CCT1 . To further probe the interplay between N17 and PRD in vivo , we examined two additional cellular processes regulated by N17 . First , N17 contains a nuclear export sequence ( Gu et al . , 2015; Maiuri et al . , 2013; Zheng et al . , 2013 ) , leading ∆N variants to accumulate in the nucleus . In striatal-derived neurons , we confirmed that mHtt-Ex1 was both cytoplasmic and nuclear while ∆N was enriched in the nucleus ( Figure 8C ) . Surprisingly , ∆P was also enriched in the nucleus , even though it contains the N17 domain ( Figure 8C ) , further suggesting that the PRD region may regulate the exposure of N17 . Second , N17 is also proposed to promote mHtt degradation ( Thompson et al . , 2009 ) , which occurs via the ubiquitin-proteasome and autophagy pathways ( Thompson et al . , 2009; Tsvetkov et al . , 2013 ) . A pulse of 35S–methionine labeling followed by a non-radioactive chase measured the impact of N17 and the PRD on the clearance of soluble mHtt variants , which were detected by immunoprecipitation and autoradiography ( Figure 8D , Figure 8—figure supplement 1B–E ) . The pulse began 14 hr post-transfection when cellular mHtt was still soluble ( not shown ) to ensure labeling of only soluble mHtt populations . Under these conditions , the disappearance of mHtt-Ex1 is abrogated by inhibition of the proteasome and lysosomal pathways , as expected ( not shown ) . Notably , both ∆N and ∆P had significantly slower protein clearance rates compared to Ex1 ( Figure 8—figure supplement 1A–B ) . This suggests that efficient mHtt recognition by the cellular degradation pathways requires the presence of both polyQ-flanking domains . To clarify the synergy between the PRD and N17 , we considered the structural and energetic contributions of the two domains to mHtt conformation . As N17 is critical to drive the initial stages of amyloidogenesis and aggregation and the absence of the PRD promotes rapid aggregation , N17 may become more buried in the faster-forming ∆P oligomer species . We thus assessed the relative exposure of N17 in purified mHtt-Ex1 variants with and without PRD by exploiting the enhanced protease susceptibility of exposed unstructured protein regions . Oligomers formed by Ex1 and ∆P were generated in vitro and the degree of exposure of N17 was tested by a short incubation with low concentrations of Proteinase K , a broad specificity protease ( Figure 8Ei ) . N17 was then detected by SDS-PAGE immunoblotting with an anti-N17 antibody ( Figure 8Ei ) . We found that N17 was significantly more protease-sensitive in Ex1 oligomers containing the PRD than in ∆P oligomers ( Figure 8Eii–iii ) . As an alternative assay , we exploited the fact that N17 is the only region in mHtt-Ex1 that contains lysine residues , which are specifically cleaved by trypsin . Since Ex1 and ∆P contain the same N17 sequence , any differences in trypsin digestion will arise from the differential exposure of N17 . Indeed , N17 in Ex1 oligomers was much more sensitive to trypsin protease cleavage than in ∆P oligomers ( Figure 8—figure supplement 1F ) . Thus , these experiments show that the PRD increases exposure of the N17 domain , which explains the similar responses of ∆N and ∆P to CCT1 protection , their enhanced nuclear localization and their slower protein degradation kinetics , all of which were previously shown to depend on N17 ( Figure 8B–D ) . Taken together , these analyses indicate that , despite their opposing effects on the energetics of aggregation or oligomer formation , the two polyQ flanking regions also act synergistically in vivo to determine the cellular fate of mHtt ( Figure 8E ) . 10 . 7554/eLife . 18065 . 019Figure 8 . Flanking domains act synergistically in vivo to determine mHtt fate . ( A ) Htt flanking domains engage various cellular proteostasis pathways ( B ) Relative viability of MSNs from rat brain slices biolistically co-transfected with the toxic mHtt variants , YFP , and the CCT1 subunit of the TRiC chaperonin . Viability was quantified 4 days after transfection . Data are mean ± SEM . Data are representative of at least three independent experiments . ****p<0 . 0001 ( C ) Nuclear and cytosolic distribution of GFP-tagged mHtt variants transfected into ST14a neurons . Distribution was quantified by measuring the ratio of mean intensity between nuclear and cytosolic fluorescent signals of individual cells . Data are mean ± SEM with at least 20 cells counted per condition . ****p<0 . 0001 , **p<0 . 01 ( D ) 35S pulse-chase measuring soluble protein degradation of mHtt-Ex1 variants in transfected ST14a cells . After transfection , cells were pulsed with 35S methionine for 3 hr and chased with complete media . Respective time points were lysed and subject to a clearing spin to only characterize soluble mHtt . Soluble mHtt was immunoprecipitated using the GFP tag . ( E ) ( i ) Schematic of ProteinaseK ( PK ) digestion experiment: Ex1 and ∆P oligomers were generated in vitro , then digested with increasing concentrations of ProteinaseK . Protease-digested reactions were run in a SDS PAGE gel and probled for N17 . ( ii ) SDS-PAGE gel of ProteinaseK-digested Ex1 and ∆P oligomers immunoprobled for N17 . ( iii ) Quantification of the N17 signal from the SDS-PAGE gel in ( ii ) . Data are mean ± SEM . *p<0 . 05 . ( F ) Model for how the polyQ flanking regions contribute to mHtt protoestasis and cellular mechanisms for clearance of toxic species . DOI: http://dx . doi . org/10 . 7554/eLife . 18065 . 01910 . 7554/eLife . 18065 . 020Figure 8—figure supplement 1 . Trypsin sensitivity of Ex1 and ∆P oligomers and 35S pulse-chase measuring protein degradation for mHtt-Ex1 variants . DOI: http://dx . doi . org/10 . 7554/eLife . 18065 . 020
While an expanded polyQ tract is intrinsically aggregation-prone , we find that by itself it inefficiently generates toxic conformations when expressed in MSNs of corticostriatal brain circuits . In vitro the ∆N∆P variant , encompassing essentially only the polyQ tract , accumulates filter-trappable SDS-insoluble species at similar rates as mHtt-Ex1 and also forms large and small oligomers . However , the conformations populated by ∆N∆P are clearly different from those of Ex1: its rate of amyloid formation measured by ThioflavinT is much slower , similar to the toxic ∆N variant and its fibrils are more tightly bundled , similar to the toxic ∆P ( Figure 1B–C , Figure 2A ) . Supporting the role of flanking domains in generating specific polyQ conformers , ∆N∆P does not give rise to 3B5H10 reactive species , unlike the toxic Ex1 variants , even though this antibody is known to recognize a polyQ-containing β-hairpin conformation ( Brooks et al . , 2004; Peters-Libeu et al . , 2012 ) . These data suggest that in the absence of flanking regions , the polyQ tract quickly forms large , non-amyloidogenic oligomers and SDS-insoluble aggregates but takes longer to develop an amyloid conformation ( Binette et al . , 2016; Crick et al . , 2013 ) and is less likely to populate toxic conformation ( s ) . N17 and PRD affect the energetic barriers leading to formation of diverse polyQ oligomers and aggregates , as well as their structural characteristics . We propose that N17-N17 and N17-polyQ interactions lower the kinetic barrier between non-amyloidogenic oligomeric species and amyloid fibrils to disfavor accumulation of oligomeric species and promote amyloid aggregation ( Figure 4E ) . Without the N17 domain , ∆N becomes kinetically trapped in large and small oligomer states ( Figure 4B–D , Figure 4—figure supplement 1B , Figure 4E ) . Trans addition of the N17 peptide can convert these trapped oligomers to fibrillar aggregates in vitro ( Figure 5B ) and accelerate mHtt-Ex1 aggregation in striatal-derived neurons ( Figure 5E–F ) . N17 may act by increasing local mHtt concentration through inter-molecular interactions with the polyQ and N17 domains ( Atwal et al . , 2011; Jayaraman et al . , 2012a; Tam et al . , 2009 ) , thereby nucleating amyloid formation ( Fiumara et al . , 2010; Jayaraman et al . , 2012b ) and promoting fibril bundling ( Figure 5B ) . In addition N17 may induce a conformational switch in the polyQ tract , through direct intrachain interaction to induce its transition to β-sheet structures ( Kokona et al . , 2014; Williamson et al . , 2010 ) , that may be important for on-pathway aggregation of polyQ stretches ( Nagai et al . , 2007 ) . The importance of the non-polar residues of N17 in promoting aggregation ( Figure 5E–F ) ( Tam et al . , 2009 ) could reflect the N17-N17 interaction interface , or a model where nonpolar N17 residues impact the conformation of the polyQ region . In contrast to N17 , the PRD destabilizes fibrils and stabilizes oligomeric species ( Figure 3 , 4 ) and disfavors aggregation both in vitro and in vivo ( Figure 1 ) ( Bhattacharyya et al . , 2005; Tam et al . , 2009 ) . Thus , removing the PRD in ∆P enhances the progression to amyloid fibrils and makes oligomer states more transient ( Figure 4B–D ) . ∆P amyloid fibrils are also more densely packed and more mechanically stable than amyloid fibrils containing a PRD region ( Figure 2A , 3B–C ) . Several models may explain these findings . The PRD could form a PolyProline-II ( PPII ) helix that propagates into the polyQ region to promote a more helical conformation in the polyQ rather than a β-sheet structure to disfavor aggregation ( Binette et al . , 2016; Darnell et al . , 2007 , 2009 ) . A dynamic PPII-helical PRD within the fibril could also interfere with fibril packing interactions ( Bugg et al . , 2012 ) , perhaps by adopting a 'bottle brush' like architecture protruding from the amyloidogenic core of the fibril ( Isas et al . , 2015 ) . Differences in the mechanical stabilities of fibrillar aggregates ( Figure 3 ) may have implications for their frangibility in vivo and the generation of mHtt seeds competent for intercellular transmission and spread . One perplexing aspect of CAG-repeat disorders like HD is why an expansion of the polyQ length beyond a certain threshold leads to neuronal toxicity ( Orr and Zoghbi , 2007 ) . One proposed theory is that toxicity arises from a feature in the repetitive ligand-binding units of a linear polyQ lattice ( Bennett et al . , 2002; Owens et al . , 2015 ) . In this sense , the longer the polyQ repeat , the more of these repetitive units may confer aberrant cellular interactions and cause pathogenesis . Alternatively , the 'emergent conformation' model proposes that a unique , toxic polyQ conformation appears at only pathogenic , expanded polyQ lengths ( Miller et al . , 2011 ) . Since mHtt constructs of identical polyQ length , but different flanking regions , have different toxicities ( Figure 6 ) , our results support the 'emergent conformation' model . The reduced toxicity of ∆N∆P implies the flanking regions contribute to generating an 'emergent' toxic conformation ( s ) . Interestingly , all the toxic variants share a subset of conformers , as evident from their shared 3B5H10-reactivity . Even if the 3B5H10-reactive oligomer itself is not the toxic species , it appears diagnostic of a conformational pathway linked to toxicity . Overall , these results have two major implications . Firstly , mHtt structural conformers and their cellular interactions are more relevant to pathogenesis than polyQ aggregation propensity . The Ex1 , ∆N and ∆P variants are all highly toxic ( Figure 6C–E ) despite their different propensities to form insoluble aggregates or soluble oligomers both in vitro and in vivo ( Figure 1B-C , 4B-E , 7D , Figure 7—figure supplement 1B ) . This suggests that only certain conformational sub-populations are neurotoxic ( Figure 7C–D ) . Defining these toxic conformers is hindered by the limitations of current structural approaches , highlighting the need for improved methodologies to structurally define soluble mHtt states . Secondly , since polyQ alone generates toxic conformations inefficiently , the sequence context is key to the neurotoxic structural transition . Thus , the other polyQ-expansion disease causing proteins , such as ataxins ( Almeida et al . , 2013 ) must contain additional sequence elements that are functionally equivalent to N17 or the PRD . This finding highlights the need for understanding polyQ dynamics in the relevant protein context . Unexpectedly , we find that despite their distinct intrinsic effects in stabilizing either the oligomer or fibrillar states , N17 and PRD cooperate in vivo to determine mHtt fate ( Figure 9A ) . mHtt may be prevented from exerting cellular damage through interaction with chaperones , clearance pathways as well as through sequestration into protective inclusion bodies , such as the aggresomes or IPOD ( Figure 9A ) ( Arrasate et al . , 2004; Kaganovich et al . , 2008; Kopito , 2000 ) . However , the synergistic action of the N17 and PRD regions is essential to engage these proteostasis pathways that promote mHtt clearance and/or prevent formation of toxic species ( Figure 9 ) . While N17 contains the recognition site for CCT1 and is known to promote mHtt degradation , both N17 and PRD are required for efficient CCT1-mediated neuroprotection ( Figure 8B , Figure 8—figure supplement 1A ) and clearance of soluble mHtt species ( Figure 8D ) . N17 is also essential to prevent the accumulation of mHtt in the nucleus ( Gu et al . , 2015 ) , but deletion of PRD also causes nuclear-enrichment of mHtt toxic species , which may play an enhanced role in Htt pathogenesis ( Difiglia , 1997 ) . We propose the observed synergy of the flanking regions in vivo is rooted in the conformational modulation exerted by the PRD to promote exposure of N17 ( Figure 8E ) . This explains the paradoxical observations that , despite having an N17 domain , ∆P shows enhanced nuclear localization and poor neuroprotection from toxicity by TRiC/CCT1 ( Figure 7C , Figure 8A–B ) . One model for this is the PRD may directly enhance N17 availability through transient interactions with each other ( Behrends et al . , 2006; Caron et al . , 2013 ) . Alternatively , the structural hindrance imposed by the PRD may prevent N17 from becoming buried in the early stages of oligomerization ( Figure 8D , Figure 8—figure supplement 1C ) . The key role of different proteostasis pathways in suppressing mHtt toxicity also suggests a model for the age-association of HD pathogenesis . Both proteostasis ( Brehme et al . , 2014; Taylor and Dillin , 2011 ) and nucleo-cytoplasmic trafficking ( D'Angelo et al . , 2009 ) decrease during aging . As a result , the N17-driven aggregation of mHtt into large , inert , amyloidogenic aggregates ( Figure 1B-F , 5B–D ) remains the predominant available route for protection , causing neurons to become saturated with aggregates ( Waelter et al . , 2001 ) . As a neuron’s ability to clear mHtt declines , the ensuing accumulation of soluble species may exert further toxicity ( Figure 9B ) . With age or the absence of N17 , toxic oligomeric species accumulate , thereby exerting cellular damage . Of note , toxic , off-pathway oligomers are also reported for α-synuclein , emphasizing the potential toxicity of soluble species that cannot be sequestered into larger aggregates ( Chen et al . , 2015 ) . Our study suggests that effective therapeutics for HD should impact the conformers associated with toxicity . Better structural methods to study conformationally heterogeneous samples will be key to define these toxic species . The cellular proteostasis machinery , which can clearly identify and neutralize these toxic species , also offers therapeutic avenues . We show that TRiC , previously shown to interact with mHtt monomers ( Tam et al . , 2009 ) , large oligomers ( Sontag et al . , 2013 ) and fibrillar aggregates ( Shahmoradian et al . , 2013 ) , promotes clearance of small 3B5H10-reactive mHtt oligomers ( Figure 7E ) confirming its crucial protective role in HD . In addition , mHtt clearance rate in neurons is linked to their likelihood of survival ( Tsvetkov et al . , 2013 ) . Thus , enhancing the ability of mHtt flanking regions to engage proteostasis pathways could provide highly specific and potent HD therapeutics . One interesting question raised by our studies is whether ∆N and ∆P variants are generated during mHtt biosynthesis . Htt contains a methionine at residue 8 which could be used as an alternative translation initiation site , leading to a partial truncation of N17 . Likewise , it has long been observed that proline-repeat stretches can stall translation ( Gutierrez et al . , 2013; Ude et al . , 2013 ) . For mHtt , this proline-dependent stalling may generate a partially truncated '∆P-like' mHtt-Ex1 variant . Future studies should test if these alternative translation mechanisms contribute to the generation of toxic mHtt fragments and HD pathogenesis .
pGEX-mHtt-Ex1-Q51 plasmid and mutants were constructed as previously described ( Tam et al . , 2006 ) . Plasmids were expressed in Rosetta 2 ( DE3 ) pLysS competent cells ( Agilent Technologies , Santa Clara , CA , USA ) in LB media supplemented with carbenicilin and chloramphenicol . Cultures were induced with 1 mM IPTG for 2 . 5 hr at 16°C . For purification , pellets were resuspended in 50 mM sodium phosphate , pH 8 . 0; 150 mM NaCl; 1 mM EDTA and lysed using an Emulsiflex ( Avestin , Ottowa , Canada ) . Lysate was incubated with GSH-Sepharose resin ( GE Healthcare , Pittsburgh , PA , USA ) and washed with 0 . 1% Triton , 500 mM NaCl , and 5 mM Mg-ATP before eluting protein with 15 mM Glutathione . Protein was concentrated and buffer exchanged with 50 mM Tris-HCl , pH 8 . 0; 100 mM NaCl; 5% glycerol . Concentrated protein was 0 . 2 μm filtered before storage at −80°C . Aggregation reactions were performed at concentration 3 μM of mHtt and 0 . 044 Units/μl acTEV ( TEV ) protease ( Invitrogen , Carlsbad , CA , USA ) . Aggregation was conducted in TEV reaction buffer ( Invitrogen ) and incubated at 30°C . Time-points were taken accordingly . Time-points were combined in a 1:1 ratio with a 4% SDS , 100 mM DTT solution , boiled for 5 min at 95°C , and stored at −20°C . Samples were then filtered through a 0 . 22 μm cellulose acetate membrane ( Whatman , Maidstone , United Kingdom ) and washed with 0 . 1% SDS . Membrane was probed using an S-tag antibody ( Abcam , Cambridge , United Kingdom ) . Aggregation reaction was prepared as above and combined with 12 . 5 μM ThioflavinT dye ( Sigma-Aldrich , St . Louis , MO , USA ) . Reactions were transferred to a 3904 Corning plate and read with an Infinite M1000 plate reader ( Tecan Systems , San Jose , CA , USA ) . Plate reader conditions were 30°C incubation , 446 nm excitation , 490 nm emission , reading every 15 min . The F-W model predicts a sigmoidal function of the fractional concentration of a product as a function of time , θ ( t ) . Such equation can be described by two independent parameters: ( 1 ) the time required to produce half of the fractional concentration ( t1/2 ) , and ( 2 ) the rate of growth at that time ( v=θ ( t1/2 ) ) . Therefore our F-W fitting function results: ( 1 ) θ ( t ) =11+e−4v ( t−t1/2 ) Each curve was baseline corrected and normalized by its final value and then fitted to Equation 1 by least-square fitting in Matlab . Live imaging was carried out on a Zeiss LSM 700 microscope ( Carl Zeiss , Oberkochen , Germany ) using epifluorescence . Cells were imaged 48 hr after transfection with 1 μg/ml Hoechst stain ( Invitrogen ) for 5 min prior to imaging . At least 150 cells were counted for each mHtt mutant . The percentage of cells with puncta was normalized with a number of transfected cells . Time-points were flash frozen at LN2 and kept at −80°C for storage . Quantifoil 1 . 2/1 . 3 grids ( Quantifoil 2/2 grids for ∆P fibers ) ( Electron Microscopy Sciences , Hatfield , PA , USA ) were glow-discharged for 30 s . Each sample was thawed and deposited on EM grids . Grids were plunge-frozen using a Vitrobot Mark III ( FEI , Hillsboro , Oregon , USA ) with a '4 s , 1' blot setting in LN2-cooled liquid ethane . Grids were stored at −80°C . Grids were loaded into JEOL JEM2010F ( JEOL , Peabody , MA , USA ) electron microscope . For each sample , approximately 15–25 images were collected on a Gatan 4k CCD camera at a defocus of −5 μm and 27 , 624x magnification with a sampling value of 5 . 43 Å per pixel . Representative 2D micrographs were converted from . DM3 to . PNG file type using Fiji software , then manually traced using a Wacom tablet on a separate invisible layer made atop each micrograph using Adobe Photoshop software . Detailed traces corresponding to zoomed-in regions were performed via Adobe Photoshop software by linear thresholding followed by selective manual removal of background using a Wacom tablet . Aggregation reactions were prepared at 20 μM for Ex1 , ∆P and ∆N∆P and 50 μM for ∆N . After 60 hr , aggregation reactions were spun at 16 , 000 ×g for 1 hr at room temperature to pellet fibers . The fiber pellet was resuspended in 1x TEV buffer ( Invitrogen ) and probe sonicated for various time and power ratings with 1 s on/1 s off pulsing using a Fisher Scientific 120 W Sonic Dismembrator ( Thermo Fisher Scientific , Waltham , MA , USA ) . The size of the sonicated fibers was measured by DLS using a Zetasizer Nano ZS ( Malvern , Worcestershire , UK ) . Aggregation reactions were prepared at 20 μM for Ex1 and ∆P . After 48h , aggregation reactions were spun at 16 , 000 ×g for 30 min at room temperature to pellet fibers . Supernatant sample was saved , and pellets were resuspended with 8M urea and incubated for 30 min at 37C , 1000 rpm shaking in a thermomixer . The samples were spun again with the same conditions . Supernatant sample was saved , and the pellets were resuspended with various concentrations of formic acid and incubated for 30 min at 37°C , 1000 rpm shaking in a thermomixer . Formic acid samples were dried by SpeedVac and resuspended in 8M urea . Samples were combined with 4x Laemlli buffer and 4M urea for running in SDS-PAGE gel and immunoblotting against the S-tag . Time-points were combined in a 1:1 ratio with sample buffer ( 150 mM Tris , pH 6 . 8 , 33% glycerol , bromophenol blue , and 1 . 2% SDS only for the SDS-AGE samples ) and stored at −80°C . Samples were then loaded on a 1% agarose gel with running buffer 25 mM Tris , 192 mM glycine , and 0 . 1% SDS for SDS-AGE gel . Gels were run at 125V until the dye front had migrated approximately 13 cm . The gel was then semi-dry blotted for 1 hr onto a PVDF membrane and probed with a S-tag antibody ( Abcam ) , GFP antibody ( Clontech , Mountain View , CA , USA ) , or 3B5H10 antibody ( Sigma-Aldrich ) . Lyophilized N17 peptide ( Genscript , Piscataway , NJ ) was solubilized with PBS and sonicated for 30 min in an ice-bath before transfection . Peptide transfections were performed using the Protein Xfect transfection reagent ( Clontech ) according to the manufacturer’s instructions . ST14a cells were cultured at 32°C , 5% CO2 in DMEM media ( Invitrogen ) supplanted with 10% FBS and Pen/Strep . Cells were transfected using Lipofectamine 2000 ( Invitrogen ) . Media was refreshed immediately before and within 24 hr post transfection . Cells were harvested 48 hr after transfection and lysed with 10 mM Tris , pH 7 . 5; 150 mM NaCl; 1 mM EDTA; 0 . 5% NP-40; 1 mM PMSF; 1x Protease Inhibitor ( Roche , Basel , Switzerland ) . Protein concentration of lysate was measured by BCA assay ( Thermo Fisher Scientific ) . Hemi-coronal brain slices were prepared accordingly to established procedures ( Reinhart et al . , 2011 ) . Briefly , postnatal day 10 Sprague–Dawley rat pups ( Charles River , Wilmington , MA , USA ) were used to prepare 250 μm thick hemi-coronal brain slices using vibratomes ( Vibratome; Bannockburn , IL , USA ) . Animals were killed in accordance with NIH guidelines and under Duke IACUC approval and oversight . Slices were placed into 12-well plates on a semi-solid support consisting of culture medium ( Neurobasal A medium supplemented with 15% heat-inactivated horse serum , 10 mM KCl , 10 mM HEPES , 100 U/mL penicillin/streptomycin , 1 mM sodium pyruvate , 1 mM L-glutamine , and 1 μM MK-801 ) set in 0 . 5% agarose . Slice cultures were maintained at 32°C in humidified incubators under 5% CO2 . A modified Helios Gene Gun ( Helios Gene Gun , Bio-Rad , Hercules , California ) was used for particle-mediated gene transfer , or biolistics , of all mHtt plasmids into brain slices as described previously ( Lo et al . , 1994 ) . Brain slices were co-transfected with a yellow fluorescent protein ( YFP ) expression plasmid in all cases , to image medium spiny neurons ( MSNs ) . MSNs were identified by their location in the striatum and characteristic dendritic aborization . Brain slices were assessed for viability on day 3 or 4 after transfection by counting the number of healthy MSNs , as previously described ( Crittenden et al . , 2010 ) . Briefly brain slices were imaged on a fluorescence stereomicroscope ( SteREO Lumar . V12 , Carl Zeiss , Thornwood , NY ) and qualified MSNs as 'healthy' if presenting a normal cell body diameter , a continuous expression of YFP within all cell compartments , and more than two discernable primary dendrites were scored as viable ( Crittenden et al . , 2010 ) . Average numbers of healthy MSNs per brain slice explant ( N=12 brain slices per condition ) are shown as means ± SEM and statistical significance was assessed by an unpaired t test with Welch’s correction using the GraphPad software . Transfected ST14a cells were immunostained with the 3B5H10 antibody . Briefly , cells were fixed with 4% PFA for 15 min at 4°C , then quenched with 10 mM NH4Cl for 5 min at room temperature . Cells were permeabilized with 0 . 2% Triton , then blocked for 1 hr with 5% Normal Donkey Serum ( Jackson Immunoresearch , West Grove , PA ) . Cells were stained with 3B5H10 in 1:10 , 000 dilution for 2 hr , then Alexa Fluor 546 conjugate secondary antibody ( Life Technologies , Carlsbad , CA ) for 1 hr . Time points were combined with 4x Native PAGE Sample Buffer ( Invitrogen ) and stored at −80°C . Samples were run on a 4–16% Bis-Tris gel ( Invitrogen ) at 150V for 150 min and transferred for 16 hr onto a PVDF membrane using the NuPAGE transfer buffer ( Invitrogen ) . ST14a cells were transfected with mHtt-Ex1-GFP constructs . 14 hr after transfection , cells were methionine-starved for 1 hr with 'methionine starvation media' ( DMEM , high glucose , no methionine , dialyzed FBS , 1 mM L-glutamine , 1 mM L-cystine ) . Cells were then pulse labeled for 3 hr with methionine starvation media with 50 mCi/ml 35S-methionine ( Perkin Elmer , Santa Clara , CA , USA ) . Then , cell media was replaced with rich DMEM high glucose media for the chase , which was in total 18 hr post transfection . Cells were harvested at respective time points and lysed with 50 mM Tris-HCl/7 . 5 , 150 mM NaCl , 0 . 5% NP-40 , 1x Roche Protease Inhibitor , 1 mM DTT , 1 mM PMSF . Lysate concentration was measured by BCA assay . mHtt protein was immunoprecipitated with a GFP nanobody . Protein mean lifetimes were calculated by normalizing radiogram signal intensity for each time point , then fitting the data to a nonlinear regression curve using the GraphPad software . Aggregation reactions were removed at the 3 hr time point and incubated with Proteinase K or Trypsin ( Sigma-Aldrich ) for 30 min on ice . Digestion was stopped with 1 mM PMSF and 1x Laemmli Sample Buffer . N17 antibody ( Ab1 ) was kindly provided by Dr . Marian DiFiglia . | Huntington’s disease is a neurodegenerative disorder in which misshapen proteins accumulate in the brain and kill neurons . The misshapen proteins form as a result of specific mutations in the gene that encodes a protein called huntingtin . These mutations result in a region of the protein called the polyQ tract being longer than normal . Other regions of huntingtin that are near to the polyQ tract can dramatically change the behavior of the mutant protein . Shen et al . investigated how these regions control the shape of mutant huntingtin and how this affects the toxicity of the mutant protein in neurons . The experiments found that the two regions on either side of the polyQ tract dramatically change the shape of mutant huntingtin proteins . In the absence of these flanking regions , the extended polyQ region is not very toxic , demonstrating that the flanking sequences play important roles in generating the toxic protein shapes . These flanking regions help mutant huntingtin to form a particular shape that was strongly linked with the death of neurons in rat brain slices . The flanking regions also change the way that the cellular machinery in neurons recognizes mutated huntingtin proteins and acts to prevent them from causing harm . Misshapen forms of other proteins are responsible for causing other neurodegenerative diseases , including Alzheimer’s and Parkinson’s diseases . Therefore , the findings of Shen et al . may help researchers to develop new drugs for these conditions , as well as for Huntingdon’s disease . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"cell",
"biology"
] | 2016 | Control of the structural landscape and neuronal proteotoxicity of mutant Huntingtin by domains flanking the polyQ tract |
The dynamics of local climates make development of agricultural strategies challenging . Yield improvement has progressed slowly , especially in drought-prone regions where annual crop production suffers from episodic aridity . Underlying drought responses are circadian and diel control of gene expression that regulate daily variations in metabolic and physiological pathways . To identify transcriptomic changes that occur in the crop Brassica rapa during initial perception of drought , we applied a co-expression network approach to associate rhythmic gene expression changes with physiological responses . Coupled analysis of transcriptome and physiological parameters over a two-day time course in control and drought-stressed plants provided temporal resolution necessary for correlation of network modules with dynamic changes in stomatal conductance , photosynthetic rate , and photosystem II efficiency . This approach enabled the identification of drought-responsive genes based on their differential rhythmic expression profiles in well-watered versus droughted networks and provided new insights into the dynamic physiological changes that occur during drought .
Projected impacts of climate change on crop yields vary widely depending on crop type and location; however , rising temperatures , with attendant increases in drought as well as insect and disease outbreaks , are predicted to result in net losses in yield of North American crops by the end of the 21st century ( Settele et al . , 2014 ) . Water stress accounts for the largest proportion of crop loss in the U . S . , and an estimated 45% of U . S . land surface suffers from low water availability ( DeLucia et al . , 2014 ) . To mitigate the predicted increase in water stress on plants ( Blum , 2005; Jones and Corlett , 1992; Anderegg et al . , 2012 ) and achieve maximal crop yield potential , locally adapted stress tolerance traits are needed . In response to soil water deficit , plants can exhibit either drought escape or drought resistance mechanisms ( Levitt , 1980; Harb et al . , 2010 ) . Under drought escape , plants complete their life cycle before the onset of stress . Drought resistance can occur through dehydration avoidance or through tolerance ( Levitt , 1980 ) . With dehydration avoidance , plants maintain high cellular water potential by lowering stomatal conductance and/or reducing water loss through changes in leaf area or orientation and by increasing resource allocation to roots . Drought tolerant plants conserve cell turgor through osmotic adjustments to survive the drought stress ( Levitt , 1980 ) and may also tolerate lower cell water potentials through anisohydric water potential regulation ( Franks et al . , 2007 ) while maintaining cellular metabolism ( Sade et al . , 2012 ) . Depending on the plant species and genotype , a combination of avoidance and tolerance traits may be utilized ( Chaves and Oliveira , 2004 ) . The potential allocation changes in drought-stressed plants also depend on selection by herbivores and the true cost of defensive molecules as leaf carbon fixation is reduced by drought ( Züst and Agrawal , 2017; Hamilton et al . , 2001 ) . Achieving maximum yield while breeding for drought stress responses will likely rely on a balance between avoidance and tolerance strategies ( Tuberosa , 2012 ) . A commonly used measure for assessing drought resistance is the intrinsic water use efficiency ( WUE ) , which is defined as the ratio between stomatal conductance ( gs ) and CO2 assimilation ( A ) ( Condon et al . , 2002 ) . WUE is often used as a proxy for drought resistance , but it is not always an accurate predictor of yield capacity under drought ( Medrano et al . , 2012 ) especially when biomass allocation to roots increases in response to drought ( Edwards et al . , 2016 ) or when yield is tightly correlated with water use ( Blum , 2009 ) . Smaller plants that limit water use and have moderate growth or short growing seasons often have higher WUE but low yield potential ( Blum , 2005 ) . This argues for using the individual gs and A measures to separately assess the impact of CO2 supply and demand effects on yield . Studies in numerous plant species have explored transcript level changes following various degrees of drought stress ( Zhang et al . , 2012; Yamaguchi-Shinozaki and Shinozaki , 2006; Smita et al . , 2013; Seki et al . , 2002; Shinozaki and Yamaguchi-Shinozaki , 2007; Degenkolbe et al . , 2009 ) and the plant responses to drought at physiological and molecular levels ( Sakuma et al . , 2006; Yu et al . , 2008; Ning et al . , 2010 ) . Few studies have evaluated physiological and molecular responses in plants experiencing mild drought ( Watkinson et al . , 2003; Vásquez-Robinet et al . , 2010 ) , although mild drought is more relevant to intensive agricultural settings than severe drought . Many drought-responsive genes are under circadian regulation ( Covington et al . , 2008 ) resulting in specific time-of-day responses to drought ( Wilkins et al . , 2010; Dubois et al . , 2017 ) . To associate the relevant transcriptomic changes with physiology , these time-of-day effects must be considered . Temporal changes complicate the assessment of differential gene expression in response to an abiotic stress due to the differences in the phasing of maximal and minimal expression levels for transcripts under circadian or diel control ( Greenham and McClung , 2015 ) . Because time-of-day changes in transcript phasing are dominant relative to the responses to drought status , comparisons of gene expression levels at any single time point will chiefly capture time-of-day expression differences rather than drought responses . Network analysis is a powerful way to extract meaningful differences across treatments , development , or time by providing pathway structure ( Priest et al . , 2014; Gehan et al . , 2015; Righetti et al . , 2015 ) . In addition , a network approach facilitates the integration of multiple datasets that can provide support to the network structure and can be used to generate predictive regulatory networks ( Langfelder et al . , 2011; Krouk et al . , 2013 ) . Here , we applied a co-expression network approach to analyze both transcriptome and physiological parameters over a two-day time course in drought-stressed and control plants , providing temporal resolution necessary for correlation of network modules with dynamic changes in drought response . We performed these studies in the crop species Brassica rapa . The genus Brassica includes the closest crop relatives of Arabidopsis and therefore is an excellent crop model for comparative studies , including analyses of adaptive drought responses . There is tremendous morphological diversity within Brassica species with important vegetable , oilseed , and forage crops that have acquired a range of stress response traits ( Ashraf and Mehmood , 1990 ) . Rapeseed ( B . napus ) , an allopolyploid derived via hybridization between B . rapa and B . oleracea , is the second largest oil crop after soybean with an annual global production of 70 million tons ( http://faostat . fao . org , 2014 ) . The majority of Brassica crops are grown in arid and semi-arid regions making drought stress a major determinant of yield . B . rapa shows a wide spectrum of drought responses ( Yarkhunova et al . , 2016; Edwards et al . , 2012 ) , suggesting that there is extensive genetic variation to explore . Further , quantitative genetic analyses of B . rapa under well-watered and drought conditions found opposite correlations between WUE and shoot biomass: plants with low WUE had higher biomass under well-watered conditions , whereas those with high WUE were larger under drought conditions ( Edwards et al . , 2012; El-Soda et al . , 2014 ) . Subsequent studies revealed quantitative trait locus ( QTL ) allele contributions to the association between gs and total biomass ( Edwards et al . , 2016 ) . Here , our objective was to associate the earliest transcriptomic responses to a water deficit with the dynamic changes in physiology throughout the day . To clarify the gene reprogramming under mild drought we focused our attention on the early response to drought stress . We measured several physiological traits and transcript abundances in B . rapa over 48 hr of controlled mild drought . To identify important regulatory pathways contributing to drought responses we applied a circadian guided co-expression network approach to correlate changes in temporally regulated transcripts with photosynthetic rate ( A ) , stomatal conductance ( gs; a measure of CO2 supply ) , and maximum efficiency of photosystem II ( PSII ) in light conditions ( Fv’/Fm’; a measure of available energy for CO2 demand ) . Gene reprogramming was altered over the time course of drought treatment , and significant changes in temporal dynamics of gs and Fv’/Fm’ reveal them to be reliable indicators of early drought perception .
We assessed the early stages of mild drought , completely withholding water for the droughted cohort of B . rapa ( Yellow Sarson ) R500 beginning at 16 days after sowing ( DAS ) . Tissue sampling and physiological measurements were conducted on Day 3 and 4 of drought , 18 and 19 DAS , respectively ( Figure 1A , Figure 1—source data 1 ) . The experiment was performed twice under similar temperature , photoperiod , and soil moisture conditions ( Figure 1A , Figure 1—source data 1 ) . In order to assess the reproducibility of our conditions , Fv’/Fm’ and above-ground biomass were monitored in both experiments , and there were no significant differences between the temporal replicates . After four days of drought , soil water potential ( Ψs ) had declined progressively to −1 . 5 MPa , whereas Ψs was relatively constant between 0 and −0 . 5 MPa for the well-watered soil ( Figure 1B , Figure 1—source data 1 ) . The droughted plants showed a significant decrease in dry above-ground biomass by the end of Day 4 ( Figure 2A ) ; however , there was no wilting during the experiment ( Figure 2B ) . In a previous experiment , prolonged progressive drought resulted in Ψs equal to approximately −5 MPa , yet some R500 plants were able to recover upon re-watering and maintained their gas exchange ( Guadagno et al . , 2017 ) . Therefore , the drought conditions applied in this study capture the early perception of drought stress ( Harb et al . , 2010 ) . We observed diurnal changes in gas exchange in the well-watered plants , expressed as WUE and as its components A ( Figure 3A , Figure 3—figure supplement 1 ) and gs ( Figure 3B , Figure 3—source data 1 ) . As expected based on a previous analysis ( Medrano et al . , 2012 ) , WUE did not provide an accurate measure of the plant response to drought stress due to limited drought severity and duration ( Figure 3A , Figure 3—figure supplement 1 ) . On both Day 3 and Day 4 of drought , A peaked at Zeitgeber Time ( ZT ) five where ZT0 corresponded to lights on ( Figure 1A ) . At this time point , we recorded a net CO2 uptake of 12 ± 2 μmol m−2s−1 ( Figure 3A ) . Droughted plants had photosynthetic capacity similar to well-watered plants for the duration of the experiment ( Figure 3A ) , in agreement with previous studies of anisohydric plants where overall photosynthetic capacity was not disturbed by mild drought stress ( Chaves et al . , 2009; Cornic and Massacci , 1996; Flexas and Medrano , 2002 ) . In contrast , there were significant differences in gs between droughted versus well-watered plants with a reduction of 50% late in the day ( ZT13 ) and 30% at night ( between ZT17 and ZT21 ) on both Day 3 and Day 4 ( Figure 3B ) . Thus , in late afternoon and at night gs responds to small changes in soil water potential and seems to play an important role in the early response to drought . Our results show that B . rapa , like many crops , can reduce CO2 supply before A is impacted ( Edwards et al . , 2016 ) . Our findings are consistent with studies showing that plants can lose as much as 30% of the daily water budget overnight ( Dawson et al . , 2007; Caird et al . , 2007 ) . Night transpiration is hypothesized to occur to enhance nutrient uptake ( Matimati et al . , 2014 ) and responds quickly to atmospheric and soil drought ( Neumann et al . , 2014; Schoppach et al . , 2014 ) as shown here . It is likely that the low nighttime gs observed in R500 plants ( Figure 3B ) contributed to the maintenance of turgor throughout the four days of drought ( Figure 2B ) . Plants were still far from the wilting point ( between −1 . 7 and −2 MPa ) for R500 ( Guadagno et al . , 2017 ) . Although signaling mechanisms are not fully understood , diurnal patterns of gs are sensitive to rapid changes in leaf water potential , causing both gs and leaf hydraulic function to decline under stress ( Brodribb and Cochard , 2009; Domec et al . , 2009 ) with ABA synthesis as a major control over anisohydric responses ( Brodribb and McAdam , 2013 ) . Although an understanding of the relationships between the circadian clock , night transpiration , and nutrient uptake would dramatically improve predictive understanding of drought , information is scarce on how anisohydric plants behave at night in drought conditions ( Rogiers et al . , 2009; Klein , 2014; Martínez-Vilalta et al . , 2014; Attia et al . , 2015 ) . The maintenance of photosynthetic capacity in droughted plants despite the significant decrease in gs may be partly explained by Fv’/Fm’ , which was significantly greater for droughted than well-watered plants on both Days 3 and 4 of drought during the middle of the light period ( Figure 3C ) . Fv’/Fm’ presented a diurnal pattern with the highest values early in the day ( ZT1 and ZT5 ) in both droughted and well-watered plants . Elevated Fv’/Fm’ fully compensated for reduced gas exchange under mild drought conditions . Our results are consistent with recent work showing that the circadian clock optimizes photosynthetic capacity by modulating temporal dynamics of Fv’/Fm’ ( García-Plazaola et al . , 2017 ) . As expected , non-structural carbohydrates ( NSC ) accumulated during the day and decreased during the night in well-watered conditions ( Figure 3D ) . In droughted plants , NSC levels were elevated throughout the night compared to well-watered controls . The presence of above-ground NSC accumulation suggested that the reduction in biomass observed ( Figure 2A ) was not due to a reduction in carbon availability but rather to the decreased nighttime conductance that preserves leaf turgor and high water potential ( Chaves , 1991 ) at the cost of sugar translocation to growing tissues ( Sevanto , 2014 ) . That NSC levels were elevated at night in droughted plants highlights the close association between water use and carbon dynamics . Specifically , early perception of drought will influence carbon allocation by lowering gas exchange and respiration rate as we observe ( Figure 3A ) . As previously reported , the lower level of respiration led to a decrease in biomass accumulation , and carbon remains in the chloroplasts because of slower transport of sugars out of the leaves ( McDowell , 2011 ) . Our results are supported by previous studies under fluctuating environmental conditions in which sugars such as glucose , fructose , and sucrose play a crucial role in maintaining cell turgor and vascular integrity in more extreme drought conditions than those studied here ( Volaire , 1995; Sala et al . , 2012 ) . Our time course analysis revealed physiological drought responses between ZT13 and ZT21 of each day , with higher magnitude on Day 4 than on Day 3 . Early in drought , plants had lower gs and higher levels of NSC in the above-ground tissues with respect to well-watered plants ( Figure 3B , D ) . We found gs to be the best physiological indicator of the early perception of drought stress in the plant , consistent with the view that A and gs are regulated separately ( Dodd et al . , 2004; von Caemmerer et al . , 2004 ) and that small decreases in gs do not lead directly to reductions in A under mild drought . In parallel with the leaf physiological measurements , transcriptomic analysis ( RNA-seq ) was performed on leaf tissue to capture the temporal changes in transcript levels during the initial stages of drought . The breadth of circadian and diel regulation of gene expression results in time-of-day-dependent changes in the transcriptome ( Harmer et al . , 2000; Covington et al . , 2008; Michael et al . , 2008 ) . Consequently , the response to abiotic stress , and in particular drought , has been shown to be dependent on the time of day in Arabidopsis and poplar , with the maximal transcriptome response occurring late in the day ( Dubois et al . , 2017; Wilkins et al . , 2009; Wilkins et al . , 2010 ) as was found in the physiological traits ( Figure 3 ) . To capture the diel transcriptome changes in the early stages of drought we applied a weighted gene co-expression network analysis ( WGCNA , Langfelder and Horvath , 2008 , Langfelder and Horvath , 2012Langfelder and Horvath , 2012 ) approach to classify genes based on their expression patterns throughout the day . We first generated well-watered and droughted networks and examined the module eigengenes , or principal components , of the gene profiles for each module in the two networks . Not surprisingly , the top eight modules , containing 80–85% of the genes in the network analysis ( see Materials and methods , Supplementary file 1 ) , showed strong temporal expression patterns across the two-day time course ( Figure 4 ) . Previous studies have shown that time-of-day effects on the transcriptome are often greater than the effect of stress treatment ( Wilkins et al . , 2010 ) . We performed hierarchical clustering of Day 4 samples from droughted and well-watered plants . In agreement with time-course transcriptome studies in Arabidopsis , poplar , and soybean ( Wilkins et al . , 2009; Wilkins et al . , 2010; Rodrigues et al . , 2015; Dubois et al . , 2017 ) samples clustered based on time of day , rather than treatment , revealing that the transcriptome varied more with time of day than due to drought ( Figure 5A ) . To examine the conservation in network topology between the droughted and well-watered transcriptomes , a consensus network to identify modules shared between the two networks was generated as previously described ( Langfelder and Horvath , 2008 ) . The consensus network contained significant overlaps in module classifications between the droughted and well-watered networks , consistent with the strong diurnal effects on the transcriptome ( Figure 5B ) . The well-watered and droughted networks contained 17 and 20 modules , respectively , suggesting that there are additional expression patterns in the droughted network due to rearrangement of the transcriptome in response to the drought treatment . For example , module 5 from the well-watered network ( wM5 ) contained genes with expression patterns that produce a peak in transcript levels at ZT5 ( Figure 6A ) . Roughly 95% of the genes in the wM5 module resolved into three distinct droughted modules , highlighted by the different color nodes in the network view ( Figure 6B ) . The mean expression of the genes in the droughted modules revealed a change in expression pattern upon drought treatment ( Figure 6B ) . The droughted module 5 ( dM5 ) appeared to be most similar to the wM5 profile , whereas dM1 showed a shift in the phase of the time of lowest transcript level and dM10 shows a bi-phasic expression peak in both days ( Figure 6B ) . Extensive rearrangement of the transcriptomic network , shown graphically in Figure 4 , occurred as expected for anisohydric plants adjusting to a mild drought ( Dal Santo et al . , 2016 ) . Similarly colored modules between the well-watered and droughted networks contained a significant overlap of genes with a common consensus network module ( Figure 5B ) , consistent with their similar eigengene profiles ( Figure 4C , D ) . As demonstrated by the network views ( Figure 4A , B ) , there was visible rearrangement of the genes within the overlapping modules ( wM2-4 compared to dM2-5 ) . The co-expression network approach successfully incorporated time-of-day information to group genes based on their diurnal patterns of expression providing a more integrated view of the well-watered and drought transcriptomes . To relate the gene expression modules to the physiology time-course data , we used WGCNA to correlate the module eigengenes with the mean values of each individual physiology measurement ( A , gs , NSC , and Fv’/Fm’ ) at each time point . Gene significance measures were calculated as the absolute value of the correlation with the physiological data ( Supplementary file 1 , Horvath et al . , 2006; Fuller et al . , 2007 ) . Several modules in both networks were positively or negatively correlated with various physiological measurements ( Figure 7A , B ) . Modules in both networks with similar phasing had similar trait correlations . For example , the wM5 and dM5 modules with peak expression between ZT1-5 were positively correlated with the A , gs , and Fv’/Fm’ , which had similarly phased peaks ( Figure 7A , B ) . Conversely , wM7 and dM8 , which peak around ZT17 , were negatively correlated with A , gs , and Fv’/Fm’ . Both sets of modules had a significant overlap of genes with consensus module 21 ( Figure 5B ) . wM11 , wM16 , and dM10 were positively correlated with and wM7 , wM10 , wM12 , dM6 , and dM8 were negatively correlated with A , gs , NSC , and Fv’/Fm’ ( Figure 7A , B ) . Within the modules there were genes that had high gene significance measures with the physiology and high module membership with the module eigengenes ( Figure 7C ) . The similar correlations observed for both well-watered and droughted networks are to be expected with a treatment that causes mild changes to physiology; however , we did observe significant differences in Fv’/Fm’ , gs , and NSC measurements in response to drought , suggesting that these traits are valid predictors of the early perception of drought stress in the plant when sampled throughout the day . We focused on these traits and selected the modules in the droughted network that were significantly correlated with these measures . Many of the genes within the modules in the droughted network that were significantly correlated with the physiology data were also correlated in the well-watered network making it difficult to identify drought-specific changes . The rhythmic patterns of gene expression and physiology inherent in the data make it amenable to circadian data analyses . In order to identify genes that are differentially expressed in response to drought we applied a circadian transcript analysis program , JTK-CYCLE , to compare the rhythmic profiles of the genes within the modules of interest between the two networks . JTK-CYCLE is a non-parametric statistical algorithm designed to identify circadian regulated transcripts and estimates period , phase , and amplitude ( Hughes et al . , 2010 ) . The genes within the droughted network modules that were positively ( dM1 , dM5 , dM10 , dM16 , and dM19 ) or negatively ( dM2 , dM6 , dM7 , dM8 , dM11 , dM14 , dM17 , and dM20 ) correlated with gs and Fv’/Fm’ ( Figure 7B ) with p<0 . 01 ( Supplementary file 1 ) were selected for analysis . The expression levels from both the well-watered and droughted datasets were used for JTK-CYCLE with period parameters set at 24 hr since our data was collected under 24 hr light/dark cycles . Genes were classified as rhythmic using a cut-off q-value <0 . 01 ( Supplementary file 2 ) . Drought-responsive candidate genes were identified based on two criteria . First , we selected transcripts that were not rhythmic in the well-watered dataset but which became rhythmic upon the imposition of drought . Second , among the transcripts that were rhythmic under both conditions , we were interested in transcripts that changed ( either increased or decreased ) in amplitude of expression upon the imposition of drought . To identify these transcripts we calculated the difference in amplitude for each transcript between the droughted and well-watered datasets and chose transcripts with an amplitude difference greater than 10 for further analysis ( Supplementary file 2 ) . To examine the expression change for the selected genes we re-grouped them based on their modules in the droughted network and plotted the mean expression profiles of these genes for each module . We first examined the positively correlated modules ( Figure 8A ) . The log2 mean expression profiles of dM1 and dM5 genes exhibit peak expression levels at ZT5 as do the gs data , consistent with the positive correlation of these modules with gs . In both modules , genes appeared to be down regulated at the end of the light period and into the night for dM1 and down regulated early in the night for dM5 . The dM10 module , which was correlated with Fv’/Fm’ , showed an elevated level of expression on Day 4 relative to Day 3 , consistent with the elevated Fv’/Fm’ in droughted plants on Day 4 ( Figure 3C ) . To validate the biological relevance of the selected genes from these modules , we analyzed the top 10 enriched GO categories for the positively and negatively correlated module lists containing at least five genes with Arabidopsis syntenic orthologs . The dM1 module was enriched for primary metabolism and response to abiotic stimulus ( Figure 8B , C ) . dM5 was enriched for photosynthesis , response to light , and abiotic stress stimulus ( Figure 8B , C ) . Abiotic stress response is expected under mild drought in anisohydric plants because the mesophyll cells are exposed to lower water potentials earlier in the drought than in isohydric plants ( Dal Santo et al . , 2016 ) . Interestingly , the dM10 module with the bi-phasic peaks contained genes involved in glucosinolate biosynthesis and metabolism ( Figure 8B , C ) . Previous work has shown that abiotic stress leads to an increase in secondary metabolism that is likely the result of carbon reallocation ( Del Carmen Martínez-Ballesta et al . , 2013 ) . At Day 4 , the stage in the mild drought treatment at which Fv’/Fm’ was beginning to decrease , the transcript data suggested that the plant is altering glucosinolate production . Although the exact purpose of this response is unclear ( Del Carmen Martínez-Ballesta et al . , 2013 ) , growth-defense tradeoffs are expected when stress reduces growth ( Züst and Agrawal , 2017 ) and secondary metabolism alterations can change circadian clock outputs ( Kerwin et al . , 2011 ) that potentially include drought responses . We next examined the modules that were negatively correlated with gs and Fv’/Fm’ . Consistent with the significant decrease in gs on Day 4 in droughted plants compared to well-watered plants , the genes in these modules showed a decrease in expression on Day 4 and in the case of dM6 an increase in expression on Day 3 as well ( Figure 9A ) . Interestingly , dM6 and dM8 displayed slight phase shifts in expression pattern with an earlier peak in expression on Day 3 compared to Day 4 suggesting that these genes contribute to the initial stages of the drought response . The genes in these modules are related to photosystem efficiency and light response pathways ( Figure 9B , C ) , consistent with the decrease in Fv’/Fm’ observed on Day 4 . dM7 , dM11 , and dM17 show dramatic decreases in expression on Day 4 relative to well-watered plants and contain genes involved in nitrogen metabolism , amino acid biosynthesis , and phosphatase activity ( Figure 9B , C ) . Comparing circadian features proved to be an effective way of identifying genes with altered patterns in the droughted relative to the well-watered network as seen by the significant GO enrichment of the selected genes ( Figures 8 and 9 ) that not only validates the biological relevance of the module structure but also the potential importance of the selected genes within these pathways . For this analysis , we chose an amplitude change of 10 as a cutoff based on an initial screen of rhythmic gene expression profiles but there are likely to be genes outside of this cutoff that exhibit a biologically meaningful change and genes within the list that do not . To validate some of the identified genes , we compared the expression levels in five biological replicate plants for each treatment , harvested during the drought experiment , without pooling of tissue from multiple plants as was done with the RNA-seq experiment . One of the limitations of time-course experiments is the cost associated with sequencing each time point at high replication . Using the JTK-CYCLE filtered gene list , we ranked the genes based on their module membership and selected the top three genes from the modules correlated with the physiology data ( Figures 8 and 9 ) . In addition , we selected genes from the list with GO ontologies associated with abiotic stress response and light harvesting processes . We identified a list of 36 genes for validation using the NanoString PlexSet technology . The NanoString data supported the trends observed in the RNA-seq dataset . The diel expression patterns seen in the well-watered and droughted plants and specific time-of-day responses to drought were recapitulated for the genes evaluated ( Figure 10 , Figure 10—figure supplements 1 and 2 , Figure 10—source data 1 ) . The expression of two members of the C-repeat-binding factor ( CBF ) regulon COR15B and COR47 ( dM1 , Figure 8C ) showed increased and shifted peak expression on Day 4 of droughted plants relative to well-watered plants ( Figure 10 ) consistent with their known roles in abiotic stress response ( Novillo et al . , 2007 ) . Consistent with the increase in Fv’/Fm’ , several genes related to light harvesting and photosystem regulation showed elevated expression levels during the day in droughted plants . The EARLY LIGHT-INDUCIBLE PROTEIN 1 and 2 ( ELIP1/2 ) genes ( Figure 8C ) , members of chlorophyll a/b – binding ( CAB ) protein superfamily and postulated to be photoprotectants for PSII under various stress conditions ( Hayami et al . , 2015 ) were both elevated in expression level and showed phase-delayed expression profiles ( Figure 10 ) . Components of PSII , LIGHT-HARVESTING CHLOROPHYLL B-BINDING 2 ( LHCB2 . 2 ) and PHOTOSYSTEM II BY ( PSBY ) , also exhibited elevated expression during the day ( Figure 8C , Figure 10 ) . As with all studies that use a correlation in time to study gene expression to trait relationships , we could not address gene expression to trait relationships that take longer than 4 hr , our sampling frequency . Similarly , we note that changes in transcript abundance do not inevitably result in changes in protein abundance or activity and will not identify meaningful changes resulting from post-translational regulation ( Graf et al . , 2017 ) . In addition to changes associated with light responses , altered expression for genes involved in nitrogen metabolism was confirmed . The GLUTAMINE SYNTHETASE 2 ( GS2 ) gene , encoding the light- and CO2-induced chloroplastic glutamine synthetase GS2 that assimilates ammonium produced during photorespiration and nitrite reduction ( Taira et al . , 2004 ) was elevated late in the day in droughted plants ( Figure 10 , Figure 10—figure supplement 2 ) . An overall reduction in expression of the mRNA encoding an integral membrane HPP family protein predicted to transport nitrite into plastids ( Maeda et al . , 2014 ) was observed in droughted plants ( Figure 10—figure supplement 1 ) . The decrease in nitrate transport is consistent with a decrease in nutrient uptake , and the increase in GS2 levels may be a response to ammonium produced from an increase in photorespiration . The drop in gs during the night and the accumulation of NSC in droughted plants on Day 4 of the treatment coincided with decreased expression of the gene encoding CYP79F2 , which metabolizes long-chain aliphatic glucosinolates ( Figure 8C , Figure 10 ) . The nitrogen metabolism changes are consistent with the role of night transpiration in nitrogen uptake ( Matimati et al . , 2014 ) and the constitutive nature of nitrogen uptake and assimilation compared to other nutrients ( Hole et al . , 1990 ) and suggest a fruitful line of research on interactions among drought , nitrogen uptake and assimilation , and the circadian clock . Knockdown of CYP79F2 using RNAi in Arabidopsis led to a drop in aliphatic glucosinolates and an increase in indole glucosinolates as well as storage carbohydrates such as fructose and galactose in addition to changes in several hormone levels ( Chen et al . , 2012 ) . The significant drops that we observed in CYP79F2 expression occurred at the ZT9 and ZT13 time points on Day 4 of the treatment when sugar accumulation was observed ( Figures 3D and 10 ) . These temporal changes in gene expression are examples of the rearrangements seen in the drought network and offer new insights into the dynamic transcriptome level changes occurring following early drought perception in B . rapa . In this study we characterized the onset of drought response by using temporal changes in physiology to support the biological significance of transcriptome changes . This approach validated the need for time-of-day resolution to observe the dynamic changes in physiology and to filter out the diel changes that cause transcript abundance variations independent of treatment . Integrating these dynamic changes in physiology with the transcriptome data using a circadian-guided network approach uncovered changes in expression of several photosynthetic and metabolic genes , suggesting an early sensing of the drought treatment at the molecular level . Future work is needed to compare the time-of-day dependent drought response of these genes in genetically and phenotypically diverse plants in order to associate the unique transcript dynamics with specific physiological responses to drought .
Seeds of Brassica rapa subsp . trilocularis ( Yellow Sarson ) R500 were planted in pots ( 156 cm3 ) filled with a soil mix ( Miracle-Gro Moisture control Potting Mix ( 20% v/v ) , Marysville , OH , and Profile Porous Ceramic ( PPC ) Greens Grade ( 80% v/v ) , Buffalo Grove , IL ) amended with 2 ml of Osmocote 18-6-12 fertilizer ( Scotts , Marysville , OH ) per pot . Plants were randomized per treatment into four growth chamber compartments ( PGC-9/2 Percival Scientific , Perry , IA ) . Chambers were set to a 14 hr/10 hr ( day/night ) photoperiod with a photosynthetic photon flux density ( PPFD ) at the plant height of ~130 μmol photons m−2 s−1 . Temperature was set to 21°C ( ±2 ) /18°C ( day/night ) cycle with relative humidity maintained between 28–33% . Plants were watered daily to maintain moist soil conditions until 16 days after sowing ( DAS ) when water was withheld from half the plants ( Droughted; Figure 1A ) . Sampling began two days after drought onset ( 18 DAS ) , with samples collected every 4 hr over 48 hr beginning 1 hr after lights on ( ZT1 ) on the third day of drought ( Figure 1A ) . The well-watered soils maintained soil water potential ( Ψs ) between 0 and −0 . 5 MPa throughout the experiment ( Figure 1B ) . Ψs declined progressively to −1 . 5 MPa over the 48 hr for the droughted plants ( Figure 1B ) . Physiological data and leaf tissue for RNA-seq were collected in separate experiments performed under identical conditions ( Figure 1B ) in order to minimize duration of sampling and to avoid potential alterations of gene expression in response to perturbations associated with the physiological measurements . To assess whether the two experiments elicited similar physiological responses to drought , Fv’/Fm’ was measured at 4 hr intervals during the day and above-ground biomass was determined at ZT17 on Day 3 and Day 4 for each experiment; neither showed any significant difference between the two experiments ( Supplementary file 1 ) . Accordingly , for these two traits , we pooled data from the replicate experiments ( Figure 3—source data 1 ) . Photosynthetic rate ( A ) and stomatal conductance ( gs ) were measured on the youngest fully expanded leaves according to the protocol described by Long and Bernacchi ( Long and Bernacchi , 2003 ) using three portable gas exchange systems provided with a 2 cm2 leaf chamber fluorimeter ( LI-COR-6400XT; LI- COR Biosciences Inc . , Lincoln , NE , USA ) . All spot measurements were taken in the same growth chamber compartment where plants were growing , and environmental conditions in the cuvette matched those in the growth chamber . The following conditions were set for the LiCOR measurements: flow rate , 300 μmol s−1; CO2 concentration , 400 μmol mol−1; VPD , 1 . 3–1 . 9 kPa; PPFD , 150 μmol photons m−2 s−1; leaf temperature , 22°C; and the cuvette fan was set to fast . Measurements in the dark ( ZT14 through ZT24 on Day 3 and Day 4 ) were taken with the same cuvette settings except that a dim green light ( ~1 μmol photons m−2 s−1 ) was used . For each replicate , gas exchange values were recorded after stabilization of the readings ( max 4 min ) . The intrinsic WUE was calculated as A/gs according to Seibt et al . ( Seibt et al . , 2008 ) . Chlorophyll a fluorescence ( Humplík et al . , 2015 ) was measured using a hand-held fluorimeter ( Fluopen FP100 , PSI , Brno , Czech Republic ) as Fv’/Fm’ , maximum efficiency of PSII in light conditions . The actinic light source of the FluorPen was maintained at ~200 μmol photons m−2 s−1 . Fv’/Fm’ was measured using a saturation pulse ( 0 . 800 s;~2200 photons μmol m−2 s−1 ) . Calculations of Fo' used the following equation from Oxborough and Baker ( Oxborough and Baker , 1997 ) where Fo’=Fo/ ( FvFm +Fo/Fm' ) . For the nighttime samples ( ZT14 through ZT24 on Day 3 and Day 4 ) , Fv/Fm , maximum efficiency of PSII in dark-adapted conditions , was measured as described previously ( Murchie and Lawson , 2013 ) ; the measuring light of the FluorPen was set at ~1 , 500 μmol photons m−2 s−1 with a saturation pulse at ~2200 photons μmol m−2 s−1 . All dark measurements were taken using a dim green light ( ~1 μmol photons m−2 s−1 ) . At ZT17 on Day 3 and Day 4 , six replicate plants from each treatment were harvested for fresh and dry biomass measures . Above-ground tissue was cut at the soil level with a razor blade , weighed , oven-dried for 10 days at 65°C and weighed again for dry biomass . NSC were measured using the anthrone method ( Seifter et al . , 1950 ) . Above-ground plant tissue ( leaves and cotyledons ) was collected , flash-frozen , and ground . The powder ( ~0 . 1 g ) , after air-drying , was extracted in 10 ml of 80% ethanol , incubated at 80°C for 30 min , and centrifuged for 5 min . The pellets were extracted two more times with 80% ethanol . An aliquot of the extract was hydrolyzed in 5 ml anthrone solution ( 4 g anthrone in 1000 ml 95% H2SO4; Sciencelab . com , Houston , TX ) in a boiling water bath for 15 min . After cooling , the sugar concentration was determined spectrophotometrically at 620 nm using glucose as a standard . We averaged all replicate samples for each physiological trait and calculated standard errors for each time point . The two treatments were compared at every time point using a one-tailed unpaired Student’s t-test . For RNA-seq , ~1 cm2 sections were cut from the youngest fully developed leaf and immediately flash frozen in liquid nitrogen . Preserved tissue was placed in long-term storage at −80°C until RNA extraction . At each time point , tissue from 10 plants in the same treatment was collected and five plants were pooled for each biological replicate , resulting in two biological replicates per treatment at each time point . We used a modified mRNA isolation protocol ( Supplementary file 3 ) to isolate mRNA directly from B . rapa R500 leaf tissue . The mRNA was used to make strand specific libraries according to the low-cost library protocol from Wang et al . ( Wang et al . , 2011a ) . Library quality and size were verified using a 2100-bioanalyzer ( Agilent Technologies , Santa Clara , CA ) . Libraries were pooled into 12 sample sets and sequenced across 4 lanes ( 12 time points/time course +2 replicates of each treatment = 48 libraries ) as 101 bp paired-end reads using Illumina HiSeq2500 ( Illumina , San Diego , CA ) . Raw data has been submitted to GEO ( http://www . ncbi . nlm . nih . gov/geo ) under accession number GSE90841 . The raw fasta reads were filtered using trimmomatic ( RRID:SCR_011848; http://www . usadellab . org/cms/index . php ? page=trimmomatic ) with mostly default settings ( ILLUMINACLIP: . /TruSeq3-PE-2 . fa:2:30:10 LEADING:3 TRAILING:3 SLIDINGWINDOW:4:25 HEADCROP:14 MINLEN:50 ) . Prior to aligning to the Chiifu reference genome ( Wang et al . , 2011b ) , existing R500 DNA-seq data ( https://www . ncbi . nlm . nih . gov/sra; SRR065676 ) were used to call SNPs in the Chiifu genome using GATK ( RRID:SCR_001876; https://www . broadinstitute . org/gatk/ ) with default settings . The vcf file generated by GATK was filtered to remove any SNPs with a quality score below 30 , coverage below 10 and all heterozygous SNPs . The remaining SNPs were used to replace the Chiifu reference genome using the vcf-consensus tool in the VCFtools package ( RRID:SCR_001235; http://vcftools . sourceforge . net/perl_module . html ) . Tophat2 ( RRID:SCR_013035 ) was used to align the RNA-seq reads to the modified Chiifu genome file using default settings and first-strand library type . Transcripts were assembled using cufflinks with the Chiifu reference Brassica_rapa . IVFCAASv1 . 19 . gtf annotation file . FPKM values were generated using cuffdiff ( RRID:SCR_001647 ) with –-no-diff –-no-js-tests options ( Trapnell et al . , 2010; 2012 ) . The well-watered and drought time course datasets were filtered to remove any genes that did not reach an FPKM value of 10 in at least one time point in order to remove non-varying or low-abundance genes that introduce noise into the network analysis . Log2 normalized FPKM values were used to generate the co-expression networks using the WGCNA ( RRID:SCR_003302 ) package in R ( Team RC , 2016; Langfelder and Horvath , 2008; Langfelder and Horvath , 2012 ) . Independent signed networks were constructed from the well-watered and drought time-course samples . An adjacency matrix was constructed using a soft threshold power of 16 . Network interconnectedness was measured by calculating the topological overlap using the TOMdist function with a signed TOMType . Average hierarchical clustering using the hclust function was performed to group the genes based on the topological overlap dissimilarity measure ( 1-TOM ) of their connection strengths . Network modules were identified using a dynamic tree cut algorithm with minimum cluster size of 30 and merging threshold function at 0 . 25 . To visualize the expression profiles of the modules , the eigengene ( first principal component ) for each module was plotted using ggplot2 in R . To identify hub genes within the modules , the module membership ( MM ) for each gene was calculated based on the Pearson correlation between the expression level and the module eigengene . Genes within the module with the highest MM are highly connected within that module . To relate the physiology measurements with the network , the module eigengenes were correlated with the physiology data . Correlations were performed for each physiology trait separately using the mean values at each time point to associate the diel patterns between the physiology and eigengenes . To associate individual genes with the physiology we calculated Gene Significance ( GS ) as described in the WGCNA package as the absolute value of the correlation between gene expression and physiology across the time series . To validate a subset of genes identified in the WGCNA and JTK-CYCLE analysis , five individual plants from the well-watered and drought conditions were collected during the time course experiment alongside the plants harvested for RNA-seq . Leaf tissue was ground in Lysis Binding Buffer ( LBB ) as described in the mRNA extraction protocol ( Supplementary file 3 ) . Following the centrifugation in LBB , 400 μl of lysate was used for RNA extraction using the Zymo Research Plant RNA MiniPrep kit ( Zymo , Irvine , CA ) . RNA purity was assessed with a NanoDrop spectrophotometer ( Thermo Fisher Scientific , Waltham MA ) , and concentration was determined using the Qubit broad range RNA assay kit according to the manufacturer’s instructions ( Thermo Fisher Scientific , Waltham MA ) . An initial RNA titration test was performed for each probeset with 50 ng , 100 ng , 150 ng and 200 ng probe to optimize the concentration . We chose 150 ng for the full time course assay . All five replicate samples for each time point and treatment were randomly arranged across 96-well plates with a random set of technical replicates . The NanoString PlexSet assay was performed according to the manufacturer’s instructions ( NanoString Technologies , Seattle , WA ) at the Molecular Biology Core Facility at Dartmouth College . Normalization was performed using the NanoString nSolver Analysis Software 3 . 0 with default settings . The housekeeping genes selected for Content Normalization were Bra021411 , Bra014841 , and Bra020305 . These genes were selected based on criteria including absence from the rhythmic modules and JTK-CYCLE list of cycling genes and low level of overall change in FPKM across the 2 day time course in both the well-watered and droughted samples . These genes also represent low , medium , and high expression levels . For the CodeSet normalization , a row of the plate containing technical replicates of two pooled RNA samples ( droughted samples Day 3 ZT5 and Day4 ZT39 ) were used . Normalized data were exported and further analyzed in R . Based on the technical replicate comparisons it was evident that there are occasional spurious probe counts for a single gene within a sample that were not reproduced in the technical replicate indicating a technical problem rather than biological . To remove these probe counts we calculated the modified Z-score for each probe across all samples and removed probes above 3 . For all samples with technical replicates we selected the sample with the lowest maximum modified Z-score . The five biological replicate samples were averaged , and standard errors calculated for each time point and a one-tailed unpaired Student’s t-test was performed to compared data from the well-watered and droughted samples at every time point . | Around 60% of the food produced worldwide relies entirely on rain for its water supply . However , in the decades ahead global climate change is predicted to cause droughts to happen more often and become more severe in many regions . Therefore , in order to sustain our food supply we need to better understand how plants respond to drought and then use that knowledge to improve the ability of crops to cope with it . Unlike animals , plants cannot move away from drought or other stressful situations so they must face these difficulties ‘head on’ . For example , when water is in short supply , plants close pores known as stomata on the surface of their leaves to reduce water loss . However , these pores need to be open to allow carbon dioxide gas , which plants use to make sugars in a process called photosynthesis , to enter the plant . Their response to drought must therefore be carefully controlled to make sure that the plant is still capable of performing photosynthesis . Turnip , napa cabbage , bok choy and field mustard are all varieties of a crop species known as Brassica rapa . These crops are grown in relatively dry regions such as the Canadian prairies and northern China , making drought stress a major threat to production . Previous studies had shown that drought stress causes changes in the activities of genes at certain times of day . To investigate this further , Greenham , Guadagno et al . studied how young B . rapa plants grown in a controlled environment with a steady supply of water responded when watering stopped . The experiments show that , even before the plants show obvious signs of drought stress such as wilting , there are extensive changes in the activity of many genes and processes inside plant cells that vary according to the time of day . Greenham , Guadagno et al . used an analysis technique to bring together all of the data into a network based on similar patterns of changes over time . This identified groups of genes whose changes in activity match the timing of the observed changes in the opening and closing of stomata , photosynthesis and other processes . These represent very early responses to drought stress in the plant . This work emphasizes the importance of time of day on plant stress responses . Changes that occurred only in the morning could not have been detected by measurements taken in the afternoon , and vice versa . The next step is to find out which of the changes observed in this work are most important in making plants resistant to drought . In the future , these findings may help researchers to develop strategies that would improve drought resistance in crop plants . | [
"Abstract",
"Introduction",
"Results",
"and",
"discussion",
"Materials",
"and",
"methods"
] | [
"plant",
"biology"
] | 2017 | Temporal network analysis identifies early physiological and transcriptomic indicators of mild drought in Brassica rapa |
Evaluation of preclinical evidence prior to initiating early-phase clinical studies has typically been performed by selecting individual studies in a non-systematic process that may introduce bias . Thus , in preparation for a first-in-human trial of mesenchymal stromal cells ( MSCs ) for septic shock , we applied systematic review methodology to evaluate all published preclinical evidence . We identified 20 controlled comparison experiments ( 980 animals from 18 publications ) of in vivo sepsis models . Meta-analysis demonstrated that MSC treatment of preclinical sepsis significantly reduced mortality over a range of experimental conditions ( odds ratio 0 . 27 , 95% confidence interval 0 . 18–0 . 40 , latest timepoint reported for each study ) . Risk of bias was unclear as few studies described elements such as randomization and no studies included an appropriately calculated sample size . Moreover , the presence of publication bias resulted in a ~30% overestimate of effect and threats to validity limit the strength of our conclusions . This novel prospective application of systematic review methodology serves as a template to evaluate preclinical evidence prior to initiating first-in-human clinical studies .
The decision to initiate an early phase clinical trial requires careful evaluation of the benefits and risks of a novel intervention . However , for first-in-human studies for which there is no prior clinical experience , the assessment of potential therapeutic efficacy must rely solely on the preclinical investigations . Although regulatory guidance exists for the conduct of preclinical evaluation of novel therapies ( U . S . Department of Health and Human services , 2013 ) , there is little guidance to help stakeholders summarize and assess the benefit and risks of novel therapies prior to first-in-human studies . As a result , the evidence from individual preclinical studies is often summarized and described in a non-systematic and potentially biased manner ( Food and Drug Administration , 2015 ) . Here , we present an approach to transparently evaluate preclinical evidence of a therapy prior to its potential clinical translation . Our exemplar is mesenchymal stem cell ( MSC ) therapy for sepsis . A selective narrative summary of preclinical evidence has significant limitations because the methods used to identify studies are neither comprehensive nor transparent ( Sena et al . , 2014 ) . This is of particular concern given that studies replicating high profile experiments fail in up to 50–90% of attempts ( Begley and Ellis , 2012; Scott et al . , 2008; Steward et al . , 2012 ) and significant publication bias results in a skewed representation of effects ( Sena et al . , 2010 ) . Further , fewer than 5% of high impact preclinical reports are clinically translated ( Contopoulos-Ioannidis et al . , 2003 ) and only 11% of clinically tested agents receive licensing ( Kola and Landis , 2004 ) . Thus trialists have based predictions of clinical success of novel therapies on flawed data and an inappropriately highly selected and positive preclinical evidence base ( Grankvist and Kimmelman , 2016 ) . Systematic reviews and meta-analyses have become very popular because they can overcome many of these challenges by promoting the transparent evaluation of therapies . Systematic reviews are guided by a protocol with explicit methods to identify , synthesize ( which may include meta-analysis ) , and appraise all investigations pertinent to a particular research question . Similarly , meta-analysis enables pooling of effect sizes across studies and increases statistical power by reducing standard error around the average effect size , providing a more precise estimate of an overall treatment effect ( Sena et al . , 2014; Cohn and Becker , 2003 ) . Systematic reviews and meta-analyses have long been regarded as essential tools to summarize and evaluate clinical research ( Higgins and Green , 2009 ) and have become a requisite component of grant applications for clinical trials ( Canadian Institutes of Health Research , 2016 ) ; however , the application of these tools to preclinical studies has been limited . Preclinical systematic reviews may help predict the magnitude and direction of novel therapeutic effects in high stakes first-in-human trials . For example , preclinical systematic reviews of stroke ( Horn et al . , 2001 ) and heart failure ( Lee et al . , 2003 ) therapies demonstrated that the resulting negative clinical trials could have been predicted had available preclinical evidence been analyzed in a rigorous manner . Thus , thousands of patients may have avoided exposure to potential risk without any benefit ( Kalra et al . , 2002; Shuaib et al . , 2007 ) . Similarly , previous preclinical systematic reviews have demonstrated that failure to report threats to methodological quality ( i . e . internal validity , risk of bias ) and construct validity ( i . e . extent a model corresponds to the human condition it is intended to represent [Henderson et al . , 2013] ) influence treatment effect sizes ( Crossley et al . , 2008; Hirst et al . , 2014; Macleod et al . , 2008 , 2015; Rooke et al . , 2011 ) . Unlike this ‘retrospective’ approach that has been described in previous studies , a prospective application of preclinical systematic review methodology may help delineate the limits of a therapy prior to first-in-human application . Our preclinical systematic review was conducted prior to the initiation of a Phase 1/2 clinical trial of immunomodulatory cell therapy ( mesenchymal stromal cells , mesenchymal stem cells [MSCs] , “adult stem cells” ) for septic shock ( NCT02421484 ) . The specific question addressed was: In preclinical in-vivo animal models of sepsis , what is the effect of MSC administration ( compared to control treatment ) on death ? Septic shock is the result of an overwhelming systemic infection; it is one of the most common and acutely devastating health problems in the intensive care unit with a 90-day mortality rate of approximately 20–30% despite modern therapy ( Peake et al . , 2014; Mouncey et al . , 2015; Stevenson et al . , 2014 ) . It is caused by a maladaptive mismatch between host inflammatory response and pathogenic stimuli which leads to organ failure and death . MSCs are ubiquitous cells ( da Silva Meirelles et al . , 2006 ) that support tissue repair and are mobilized under inflammatory conditions ( Hannoush et al . , 2011; Rochefort et al . , 2006 ) . Exogenously administered MSCs represent an especially attractive therapeutic for sepsis because they have antibacterial and organ protective effects , in addition to their immune modulatory functions ( Walter et al . , 2014 ) . We quantitatively summarized the results of all preclinical studies of MSC therapy for in vivo animal models of sepsis to predict effect size and establish an ethical basis for exposing high-risk patients to this novel therapy . This is the first systematic evaluation of a novel biologic therapy prior to initiating a first-in-human trial . We believe our approach serves as a roadmap to transparently evaluate a preclinical therapy prior to its potential clinical translation . This study has been written in an explicatory manner so that other preclinical and translational researchers not familiar with systematic review methodology may replicate our approach . Readers wishing to replicate our approach for their research agendas are directed to the methods section where explanations are provided in greater depth , and encouraged to contact the authors for further guidance .
Our systematic search of MEDLINE , Embase , BIOSIS , and Web of Science yielded 3114 records . Following deduplication and screening , 18 studies were included in the review ( Figure 1 ) . These studies were published over a six year period ( 2009 to 2015 ) and corresponded to 20 unique experiments and involved a total of 980 animals ( Table 1 ) ( Bi et al . , 2010; Chang et al . , 2012; Chao et al . , 2014; Gonzalez-Rey et al . , 2009; Hall et al . , 2013; Kim et al . , 2014; Krasnodembskaya et al . , 2012; Li et al . , 2012; Liang et al . , 2011; Luo et al . , 2014; Mei et al . , 2010; Nemeth et al . , 2009; Pedrazza et al . , 2014; Sepúlveda et al . , 2014; Yang et al . , 2015; Zhao et al . , 2013 , 2014; Zhou et al . , 2014 ) . Six authors were contacted for additional information and all replied . 10 . 7554/eLife . 17850 . 003Figure 1 . Preferred reporting items for systematic reviews and meta-analysis ( PRISMA ) flow diagram for study selection . DOI: http://dx . doi . org/10 . 7554/eLife . 17850 . 00310 . 7554/eLife . 17850 . 004Table 1 . General characteristics of preclinical studies investigating the efficacy of mesenchymal stromal cells in models of sepsis . DOI: http://dx . doi . org/10 . 7554/eLife . 17850 . 004Author year CountrySpecies , Strain , GenderSepsis modelResuscitationMSC source , CompatibilityMSC DoseTime ( hours ) *MSC routeControlGonzalez-Rey et al . ( 2009 ) A SpainMouse BALB/c , NRCLP ( 1 × 22 G ) NoneAdipose Xenogenic or Allogeneic1 . 0 × 1064IPDMEMGonzalez-Rey et al . ( 2009 ) B SpainMouse BALB/c , NRLPS ( i . p . ) NoneAdipose Xenogenic1 . 0 x 106 or3 . 0 x 1050 . 5IPDMEMNemeth et al . ( 2009 ) United StatesMouse C57BL/6 , MCLP ( 2 × 21 G ) Fluid and antibioticsBone marrow Allogeneic1 . 0 × 1060 or 1IVPBS or FibroblastBi et al . ( 2010 ) ChinaMouse C57BL/6 , NRCLP ( 2 × 21 G ) NoneBone marrow Xenogenic1 . 0 × 1061 1IVPBSMei et al . ( 2010 ) A CanadaMouse C57BL/6J , FCLP ( 1 × 22 G ) FluidBone marrow Syngeneic2 . 5 × 1056IVNSMei et al . ( 2010 ) B CanadaMouse C57BL/6J , FCLP ( 1 × 18 G ) Fluid and antibioticsBone marrow Syngeneic2 . 5 × 1056IVNSLiang et al . ( 2011 ) ChinaRat Wistar , FLPS ( i . v . ) NoneBone marrow Syngeneic1 . 0 × 1062IVNSChang et al . ( 2012 ) ChinaRat SPD , MCLP ( 2 × 18 G ) NoneAdipose Autologous3 × 1 . 2 × 1060 . 5 , 6 then 18IPNSKrasnodembskaya et al . ( 2012 ) , USAMouse C57BL/6J , MP . aeruginosa ( i . p . ) NoneBone marrow Xenogenic1 . 0 × 1061IVPBS FibroblastLi et al . ( 2012 ) ChinaRat SPD , MLPS ( i . p . ) NoneUmbilical cord Xenogenic5 . 0 × 1051IVNS or FibroblastHall et al . ( 2013 ) USAMouse BALB/c , MCLP ( 2 × 21 G ) NoneBone marrow Syngeneic1 × 5 . 0× 105 + 2× 2 . 5 × 1052 then 24 then 48IVPBS or FibroblastZhao et al . ( 2013 ) ChinaRat SPD , FLPS ( i . v . ) NoneBone marrow Syngeneic2 . 5 ×1062IVNSChao et al . ( 2014 ) TaiwanRat Wistar , MCLP ( 1 × 18 G ) NoneBone Marrow or Umbilical Cord Xenogenic5 . 0 × 1064IVPBSKim et al . ( 2014 ) CanadaMouse C57BL/6 , MSEB+ ( i . p ) NoneBone marrow Syngeneic2 . 5 × 1053IVPBSLuo et al . ( 2014 ) ChinaMouse C57Bl/6 , MCLP ( 2 × 21 G ) FluidBone marrow Syngeneic1 . 0 × 1063IVNSPedrazza et al . ( 2014 ) BrazilMouse C57BL/6 , ME . coli ( i . p . ) NoneAdipose Syngeneic1 . 0 × 1060IVPBSSepulveda et al . ( 2014 ) SpainMouse BALB/c , MLPS ( i . p . ) NoneBone Marrow Xenogenic1 . 0 × 1060 . 5IPPBSZhao et al . ( 2014 ) ChinaMouse C57BL/6 , MCLP ( NR ) NoneUmbilical cord Xenogenic1 . 0 × 1061IVNSZhou et al . ( 2014 ) ChinaMouse NOD SCID , MLPS+ ( i . p . ) NoneUmbilical Cord Xenogenic2 . 0 × 1066IVNo treatmentYang et al . ( 2015 ) ChinaMouse NOD SCID , MLPS+ ( i . p . ) NoneUmbilical cord Xenogenic5 . 0 × 1050IVDMEMLegend: * = Time of delivery post-sepsis induction , + = Models also administered D-galactosamine , CLP = Cecal ligation and puncture , DMEM = Dulbecco's modified Eagle's medium , i . p . = Intraperitoneal , i . v . = Intravenous , LPS = Lipopolysaccharide , NR = Not reported , NOD SCID = NOD . Cg-Prkdcscid Il2rgtm1Wjl/SzJ ( immunodeficient ) , NS = Normal saline , PBS = Phosphate buffered saline , SEB = Staphylococcal enterotoxin B , SPD = Sprague-Dawley . All experiments used rodents , and most were mice ( 80% ) . Several methods were used to establish sepsis or sepsis-like pathophysiology , including cecal-ligation and puncture ( 50% ) , live bacterial injection ( 10% ) , and bacterial component injection ( 40% ) . Tissue sources of MSCs included bone marrow ( 60% ) , adipose tissue ( 20% ) , and umbilical cord ( 20% ) . Similarly , immunological compatibility between donor MSCs and recipients varied between xenogenic ( 50% ) , syngeneic ( 40% ) , allogeneic ( 5% ) and autologous ( 5% ) . Two of ten experiments with xenogenic cells used immunocompromised mice , while the remainder used immunocompetent mice . Total doses of MSCs ranged from 2 . 5 × 105 to 5 . 0 × 106 and most studies administered cells as a single dose ( 90% ) either intravenously ( 80% ) or intraperitoneally ( 20% ) . MSC therapy was initiated between 0 to 6 hr after experimental induction of the disease state . MSC therapy in preclinical models of sepsis significantly reduced the overall odds of death ( odds ratio ( OR ) 0 . 27 , 95% confidence interval ( CI ) 0 . 18–0 . 40 ( Figure 2 ) . Since it is important to consider the consistency of results between studies , we calculated the I2 test , which demonstrated a low degree of heterogeneity across studies ( I2 = 33% ) . The reduction in mortality was maintained regardless of when death occurred , whether considering deaths before two days after induction of sepsis ( OR 0 . 31 , 95% CI 0 . 21–0 . 46 ) , between two and four days ( OR 0 . 20 , 95% CI 0 . 11–0 . 38 ) , or more than four days ( OR 0 . 18 , 95% CI 0 . 11–0 . 32 ) ( Figure 3 ) . 10 . 7554/eLife . 17850 . 005Figure 2 . Forest plot summarizing effects of mesenchymal stromal cell ( MSC ) therapy on mortality in preclinical models of sepsis and endotoxemia . Point estimates ( odds ratio ) and 95% confidence intervals ( CI ) are depicted for individual studies; size of point estimate depicts relative contribution to pooled effect . A pooled meta-analytic summary ( random effects model ) of overall effect of MSC therapy on mortality is depicted by the diamond at the bottom of the plot ( vertical points represent odds ratio point estimate and horizontal points represent 95% CIs ) . Heterogeneity is represented with the I2 statistic . Data from Pedrazza et al . ( 2014 ) was included in total counts but not included in meta-analysis due to 100% mortality in both study arms . DOI: http://dx . doi . org/10 . 7554/eLife . 17850 . 00510 . 7554/eLife . 17850 . 006Figure 2—figure supplement 1 . Forest plot summarizing relationship of compatibility of donor mesenchymal stromal cell ( MSC ) with recipient animal ( xenogenic vs syngeneic vs allogeneic vs autologous ) on mortality in preclinical models of sepsis and endotoxemia . Point estimates ( odds ratio ) and 95% confidence intervals ( CI ) are depicted for individual studies; size of point estimate depicts relative contribution to pooled effect . A pooled meta-analytic summary ( random effects model ) of overall effect is depicted by the diamond at the bottom of each subgroup ( vertical points represent odds ratio point estimate and horizontal points represent 95% CIs ) . Heterogeneity is represented with the I2 statistic . Data from Pedrazza et al 2014 was included in total counts but not included in meta-analysis due to 100% mortality in both study arms . DOI: http://dx . doi . org/10 . 7554/eLife . 17850 . 00610 . 7554/eLife . 17850 . 007Figure 2—figure supplement 2 . Forest plot summarizing relationship of mesenchymal stromal cell ( MSC ) dose on mortality in preclinical models of sepsis and endotoxemia . Point estimates ( odds ratio ) and 95% confidence intervals ( CI ) are depicted for individual studies; size of point estimate depicts relative contribution to pooled effect . A pooled meta-analytic summary ( random effects model ) of overall effect is depicted by the diamond at the bottom of each subgroup ( vertical points represent odds ratio point estimate and horizontal points represent 95% CIs ) . Heterogeneity is represented with the I2 statistic . Data from Pedrazza et al 2014 was included in total counts but not included in meta-analysis due to 100% mortality in both study arms . DOI: http://dx . doi . org/10 . 7554/eLife . 17850 . 00710 . 7554/eLife . 17850 . 008Figure 2—figure supplement 3 . Forest plot summarizing relationship of mesenchymal stromal cell ( MSC ) therapy timing of administration on mortality in preclinical models of sepsis and endotoxemia . Point estimates ( odds ratio ) and 95% confidence intervals ( CI ) are depicted for individual studies; size of point estimate depicts relative contribution to pooled effect . A pooled meta-analytic summary ( random effects model ) of overall effect is depicted by the diamond at the bottom of each subgroup ( vertical points represent odds ratio point estimate and horizontal points represent 95% CIs ) . Heterogeneity is represented with the I2 statistic . Data from Pedrazza et al 2014 was included in total counts but not included in meta-analysis due to 100% mortality in both study arms . DOI: http://dx . doi . org/10 . 7554/eLife . 17850 . 00810 . 7554/eLife . 17850 . 009Figure 2—figure supplement 4 . Forest plot summarizing relationship of mesenchymal stromal cell ( MSC ) administration route ( intravenous vs intraperitoneal ) on mortality in preclinical models of sepsis and endotoxemia . Point estimates ( odds ratio ) and 95% confidence intervals ( CI ) are depicted for individual studies; size of point estimate depicts relative contribution to pooled effect . A pooled meta-analytic summary ( random effects model ) of overall effect is depicted by the diamond at the bottom of each subgroup ( vertical points represent odds ratio point estimate and horizontal points represent 95% CIs ) . Heterogeneity is represented with the I2 statistic . Data from Pedrazza et al 2014 was included in total counts but not included in meta-analysis due to 100% mortality in both study arms . DOI: http://dx . doi . org/10 . 7554/eLife . 17850 . 00910 . 7554/eLife . 17850 . 010Figure 2—figure supplement 5 . Forest plot summarizing relationship of mesenchymal stromal cell ( MSC ) tissue source ( adipose vs bone marrow vs umbilical cord tissue ) on mortality in preclinical models of sepsis and endotoxemia . Point estimates ( odds ratio ) and 95% confidence intervals ( CI ) are depicted for individual studies; size of point estimate depicts relative contribution to pooled effect . A pooled meta-analytic summary ( random effects model ) of overall effect is depicted by the diamond at the bottom of each subgroup ( vertical points represent odds ratio point estimate and horizontal points represent 95% CIs ) . Heterogeneity is represented with the I2 statistic . Data from Pedrazza et al 2014 was included in total counts but not included in meta-analysis due to 100% mortality in both study arms . DOI: http://dx . doi . org/10 . 7554/eLife . 17850 . 01010 . 7554/eLife . 17850 . 011Figure 2—figure supplement 6 . Forest plot summarizing relationship of animal species ( rat vs mouse ) on mortality in preclinical models of sepsis and endotoxemia treated with mesenchymal stromal cells ( MSCs ) . Point estimates ( odds ratio ) and 95% confidence intervals ( CI ) are depicted for individual studies; size of point estimate depicts relative contribution to pooled effect . A pooled meta-analytic summary ( random effects model ) of overall effect is depicted by the diamond at the bottom of each subgroup ( vertical points represent odds ratio point estimate and horizontal points represent 95% CIs ) . Heterogeneity is represented with the I2 statistic . Data from Pedrazza et al 2014 was included in total counts but not included in meta-analysis due to 100% mortality in both study arms . DOI: http://dx . doi . org/10 . 7554/eLife . 17850 . 01110 . 7554/eLife . 17850 . 012Figure 2—figure supplement 7 . Forest plot summarizing relationship of animal sex ( male vs female vs unreported ) on mortality in preclinical models of sepsis and endotoxemia treated with mesenchymal stromal cells ( MSCs ) . Point estimates ( odds ratio ) and 95% confidence intervals ( CI ) are depicted for individual studies; size of point estimate depicts relative contribution to pooled effect . A pooled meta-analytic summary ( random effects model ) of overall effect is depicted by the diamond at the bottom of each subgroup ( vertical points represent odds ratio point estimate and horizontal points represent 95% CIs ) . Heterogeneity is represented with the I2 statistic . Data from Pedrazza et al 2014 was included in total counts but not included in meta-analysis due to 100% mortality in both study arms . DOI: http://dx . doi . org/10 . 7554/eLife . 17850 . 01210 . 7554/eLife . 17850 . 013Figure 2—figure supplement 8 . Forest plot summarizing relationship of preclinical models of sepsis and endotoxemia ( cecal ligation and puncture vs live bacteria administration vs bacterial product such as lipopolysaccharide ) on mortality following treatment with mesenchymal stromal cells ( MSCs ) . Point estimates ( odds ratio ) and 95% confidence intervals ( CI ) are depicted for individual studies; size of point estimate depicts relative contribution to pooled effect . A pooled meta-analytic summary ( random effects model ) of overall effect is depicted by the diamond at the bottom of each subgroup ( vertical points represent odds ratio point estimate and horizontal points represent 95% CIs ) . Heterogeneity is represented with the I2 statistic . Data from Pedrazza et al 2014 was included in total counts but not included in meta-analysis due to 100% mortality in both study arms . DOI: http://dx . doi . org/10 . 7554/eLife . 17850 . 01310 . 7554/eLife . 17850 . 014Figure 2—figure supplement 9 . Forest plot summarizing relationship of resuscitation ( fluids +/- antibiotics vs no resuscitation ) on mortality in preclinical models of sepsis and endotoxemia treated with mesenchymal stromal cells ( MSCs ) . Point estimates ( odds ratio ) and 95% confidence intervals ( CI ) are depicted for individual studies; size of point estimate depicts relative contribution to pooled effect . A pooled meta-analytic summary ( random effects model ) of overall effect is depicted by the diamond at the bottom of each subgroup ( vertical points represent odds ratio point estimate and horizontal points represent 95% CIs ) . Heterogeneity is represented with the I2 statistic . Data from Pedrazza et al 2014 was included in total counts but not included in meta-analysis due to 100% mortality in both study arms . DOI: http://dx . doi . org/10 . 7554/eLife . 17850 . 01410 . 7554/eLife . 17850 . 015Figure 2—figure supplement 10 . Forest plot summarizing relationship of comparison ( control ) treatment on mortality in preclinical models of sepsis and endotoxemia treated with mesenchymal stromal cells ( MSCs ) . Point estimates ( odds ratio ) and 95% confidence intervals ( CI ) are depicted for individual studies; size of point estimate depicts relative contribution to pooled effect . A pooled meta-analytic summary ( random effects model ) of overall effect is depicted by the diamond at the bottom of each subgroup ( vertical points represent odds ratio point estimate and horizontal points represent 95% CIs ) . Heterogeneity is represented with the I2 statistic . Data from Pedrazza et al 2014 was included in total counts but not included in meta-analysis due to 100% mortality in both study arms . DOI: http://dx . doi . org/10 . 7554/eLife . 17850 . 01510 . 7554/eLife . 17850 . 016Figure 2—figure supplement 11 . Forest plot summarizing relationship of adherence to elements of construct validity on mortality in preclinical models of sepsis and endotoxemia treated with mesenchymal stromal cells ( MSCs ) . Subgroups are studies that adhered to a majority of elements suggested to increase construct validity ( ≥5 of 8; see text for details of elements ) vs those that did not adhere to majority . Point estimates ( odds ratio ) and 95% confidence intervals ( CI ) are depicted for individual studies; size of point estimate depicts relative contribution to pooled effect . A pooled meta-analytic summary ( random effects model ) of overall effect is depicted by the diamond at the bottom of each subgroup ( vertical points represent odds ratio point estimate and horizontal points represent 95% CIs ) . Heterogeneity is represented with the I2 statistic . Data from Pedrazza et al 2014 was included in total counts but not included in meta-analysis due to 100% mortality in both study arms . DOI: http://dx . doi . org/10 . 7554/eLife . 17850 . 01610 . 7554/eLife . 17850 . 017Figure 3 . Forest plot summarizing relationship of mesenchymal stromal cell ( MSC ) therapy on mortality over time in preclinical models of sepsis and endotoxemia ( outcome windows: ≤2 days , >2 to ≤ 4 days , > 4 days ) . Point estimates ( odds ratio ) and 95% confidence intervals ( CI ) are depicted for individual studies; size of point estimate depicts relative contribution to pooled effect . A pooled meta-analytic summary ( random effects model ) of overall effect of MSC therapy on mortality is depicted by the diamond at the bottom of each time interval ( vertical points represent odds ratio point estimate and horizontal points represent 95% CIs ) . Heterogeneity is represented with the I2 statistic . Data from Pedrazza et al . ( 2014 ) was included in total counts but not included in meta-analysis due to 100% mortality in both study arms . DOI: http://dx . doi . org/10 . 7554/eLife . 17850 . 017 The effects of therapies may not be sustained under varied experimental conditions , so we evaluated the generalizability and replicability of results by analyzing efficacy in pre-specified sub-groups . Heterogeneity ( i . e . I2 statistic ) was low to moderate unless otherwise stated . Similar efficacy was noted regardless of the compatibility of donor MSCs with recipient animal ( syngeneic vs . allogeneic vs . xenogenic , Figure 2—figure supplement 1 ) , dose of MSC ( <1 . 0 × 106 cells vs . ≥1 . 0 × 106 cells , Figure 2—figure supplement 2 ) , and timing of a single dose of MSCs ( less than or equal to 1 hr vs . 1–6 hr after disease induction , Figure 2—figure supplement 3 ) . Intravenous administration of MSCs demonstrated efficacy ( OR 0 . 28 , 95% CI 0 . 20–0 . 40 ) ; whereas intraperitoneal administration of MSCs did not have a statistically significant effect ( OR 0 . 21 , 95% CI 0 . 02–1 . 89; Figure 2—figure supplement 4 ) and had high heterogeneity ( I2 = 78% ) , suggesting a high degree of inter-study variability . Significant effects were seen using MSCs derived from bone marrow ( OR 0 . 13 , 95% CI 0 . 05–0 . 35 ) and umbilical cord ( OR 0 . 30 , 95% CI 0 . 21–0 . 43; Figure 2—figure supplement 5 ) , but the MSCs derived from adipose tissue did not demonstrate statistically significant efficacy ( OR 0 . 35 , 95% CI 0 . 03–4 . 39 , I2 = 79% ) . Two studies administered multiple doses of MSCs , with one demonstrating benefit and the other having no statistically significant effect . The multiple dose study with no effect was also the only investigation of autologous cells ( Chang et al . , 2012 ) . MSCs administered to mice were effective ( OR 0 . 23 , 95% CI 0 . 15–0 . 36 ) however MSC administration to rats did not produce a statistically significant effect ( OR 0 . 47 , 95% CI 0 . 18–1 . 21; Figure 2—figure supplement 6 ) . Neither the sex of the diseased animal nor the model used ( cecal ligation and puncture vs . live bacterial injection vs . lipopolysaccharide or other bacterial product ) influenced efficacy ( Figure 2—figure supplements 7 and 8 ) . The addition of resuscitation ( fluids or antibiotics , which are current clinical standards of therapy ) did not influence the protective effect of MSCs ( Figure 2—figure supplement 9 ) . The comparator control group ( phosphate buffered saline vs . fibroblast vs . normal saline vs . medium ) had no effect; but , the one study that did not administer vehicle to the control animals did not demonstrate a statistically significant effect of MSC therapy ( Zhou et al . , 2014 ) ( Figure 2—figure supplement 10 ) . Practices such as blinding and randomization are known to affect the magnitude of effect in both clinical and preclinical studies . To determine if these threats to internal validity influenced our findings , we evaluated the risk of bias of included studies ( Table 2 ) . None of the experiments were considered low risk of bias across all six domains of methodological quality . Forty-five percent of experiments reported that the animals were randomized , none described methods of sequence generation or how allocation concealment was achieved . Similarly , no studies described blinding of personnel performing the experiments . One study did not blind assessors for the outcome of mortality , which may be of concern given that surrogate endpoints ( i . e . not true death due to animal welfare concerns ) were assessed ( Kim et al . , 2014 ) ; the remaining studies were assessed as ‘unclear’ as insufficient details of outcome assessment were reported . An assessment of high risk of bias for incomplete outcome data occurred in 10% of studies ( examined as consistent n values reported from methods to results ) ; in 65% of experiments the numbers ( n ) were not presented in both the methods and results in sufficient detail to permit judgment . No studies reported an appropriate rationale for selection of study sample size ( where appropriate rationale included a correctly calculated sample size , Table 3 ) . Given the paucity of studies that adequately implemented and reported internal validity practices , an analysis to determine the effects of high vs . low risk of bias on the effect size was not feasible . 10 . 7554/eLife . 17850 . 018Table 2 . Risk of bias assessment of preclinical studies investigating the efficacy of mesenchymal stromal cells in models of sepsis . DOI: http://dx . doi . org/10 . 7554/eLife . 17850 . 018StudyRandomizationAllocation concealmentBlinding of personnelBlinding of outcome assessmentIncomplete outcome dataSelective outcome reportingGonzalez-Rey et al . ( 2009 ) UUUULLNemeth et al . ( 2009 ) UUUULLBi et al . ( 2010 ) UUUUHLMei et al . ( 2010 ) UUUULLLiang et al . ( 2011 ) UUUUULChang et al . ( 2012 ) UUUUULKrasnodembskaya et al . ( 2012 ) UUUUULLi et al . ( 2012 ) UUUUULHall et al . ( 2013 ) UUUUULZhao et al . ( 2013 ) UUUUULChao et al . ( 2014 ) UUUUULKim et al . ( 2014 ) UUUHULLuo et al . ( 2014 ) UUUUULPedrazza et al . ( 2014 ) UUUUULSepulveda 2014UUUUULZhao et al . ( 2014 ) UUUUULZhou et al . ( 2014 ) UUUUHLYang et al . ( 2015 ) UUUUULLegend: H = High risk of bias , L = Low risk of bias , U = Unclear risk of biasBlinding of Outcome Assessment for Mortality: Low risk = Outcome assessors were blinded to the study groups when assessing mortality through surrogate endpoints or animals were allowed to die . Unclear = Insufficient information to determine if outcome assessors were blinded during assessment or if animals were allowed to die . High Risk = Outcome assessors not blinded to the study groups and death was defined according to surrogate endpoints . Incomplete Outcome Data: Low risk = N values were consistent between methods and results for the mortality outcome . Unclear = N value was either not presented in the methods or in the results , and therefore there is insufficient information to permit judgement . High risk = N values were not consistent between methods and results for the mortality outcome . Selective Reporting: Low risk = The methods section indicated mortality as a pre-specified outcome measure . High risk = The mortality outcome was presented in the results but not pre-specified in the methods section . 10 . 7554/eLife . 17850 . 019Table 3 . Risk of bias assessment ( other domains ) of preclinical studies investigating the efficacy of mesenchymal stromal cells in models of sepsis . DOI: http://dx . doi . org/10 . 7554/eLife . 17850 . 019StudyBaseline characteristics*Random housing*Source of fundingConflict of interestSample size calculationGonzalez-Rey et al . ( 2009 ) UUHHUNemeth et al . ( 2009 ) UULUUBi et al . ( 2010 ) UULUUMei ( 2010 ) UUHHULiang et al . ( 2011 ) UULUUChang et al . ( 2012 ) UULLHKrasnodembskaya et al . ( 2012 ) UULLULi et al . ( 2012 ) UULLUHall et al . ( 2013 ) UULLUZhao et al . ( 2013 ) UULLUChao et al . ( 2014 ) UULLUKim et al . ( 2014 ) UULLULuo et al . ( 2014 ) UULLUPedrazza et al . ( 2014 ) UULLUSepulveda 2014UULLUZhao et al . ( 2014 ) UULLUZhou et al . ( 2014 ) UULLUYang et al . ( 2015 ) UULLULegend: * = Items modified from SYRCLE risk of bias tool , H = High risk of bias , L = Low risk of bias , U = Unclear risk of biasBaseline Characteristics: Low risk = Baseline severity of disease equal between experimental groups , Unclear = Baseline severity of disease unreported , High risk = Baseline severity of disease unbalanced between experimental groups . Random Housing: Low risk = Animal cages were randomly placed within an animal room/facility , Unclear = Housing placement unreported , High risk = Animals place in non-random arrangement in animal room/facility . Other risk of bias was assessed according to source of funding , conflict of interest and pre-specified sample size calculations:Source of Funding: Low risk = Non-industry source of funding ( or no funding ) . Unclear = Funding source was not reported . High risk = Study was funded by industry . Conflict of Interest: Low risk = Authors reported on no conflict of interest . Unclear = Conflict of interest was not reported . High risk = Authors reported on potential conflict of interests . Sample Size Calculation: Low risk = Sample size calculations were correctly performed and followed . Unclear = Sample size calculations were not performed . High risk = Sample size calculations were incorrectly performed/followed . It has been suggested that failed preclinical to clinical translation may be related to a mismatch between experimental conditions and the clinical disease the model is intended to represent ( i . e . construct validity ) ( Henderson et al . , 2013; Kimmelman and Henderson , 2016 ) . To evaluate clinical generalizability of the experimental conditions used , we performed a formal evaluation of construct validity using an eight item index that had been developed in a systematic review of preclinical sepsis ( Table 4 ) ( Lamontagne et al . , 2010 ) . None of the experiments used large animal models . Two experiments ( 10% ) used animals with comorbidities ( both used immunodeficient mice ) , 40% of experiments used adult animal models ( 40% did not report animal age ) , and 50% used infectious models of sepsis . 90% of studies initiated MSC therapy after the induction of the disease ( as opposed to at the time of disease induction ) but none documented severity of the disease state prior to initiating MSC therapy . Four studies used fluid resuscitation while two of these studies also administered antibiotics . Two studies incorporated a majority of construct validity elements ( i . e . at least five of eight elements ) ; there was no difference in effect size between these studies ( OR 0 . 18 , 95% CI 0 . 08–0 . 42 ) and those studies that incorporated fewer elements ( OR 0 . 28 , 95% CI 0 . 17–0 . 44 ) ( Figure 2—figure supplement 11 ) . 10 . 7554/eLife . 17850 . 020Table 4 . Construct validity assessment of preclinical studies investigating the efficacy of mesenchymal stromal cells in models of sepsis . DOI: http://dx . doi . org/10 . 7554/eLife . 17850 . 020StudyLarge animal modelAdult animal modelComorbiditiesInfectious model of sepsisTherapy initiated after sepsis inductionDocumented sepsis severity prior to initiating treatmentResuscitation included antibioticsResuscitation included fluidsGonzalez-Rey et al . ( 2009 ) A NNNYYNNNGonzalez-Rey et al . ( 2009 ) BNNNNYNNNNemeth et al . ( 2009 ) NYNYYNYYBi et al . ( 2010 ) NUNYYNNNMei ( 2010 ) ANYNYYNNYMei ( 2010 ) BNYNYYNYYLiang et al . ( 2011 ) NUNNYNNNChang et al . ( 2012 ) NUNYYNNNKrasnodembskaya et al . ( 2012 ) NYNYYNNNLi et al . ( 2012 ) NUNNYNNNHall et al . ( 2013 ) NUNYYNNNZhao et al . ( 2013 ) NYNNYNNNChao et al . ( 2014 ) NUNYYNNNKim et al . ( 2014 ) NYNNYNNNLuo et al . ( 2014 ) NUNYYNNYPedrazza et al . ( 2014 ) NYNYNNNNSepulveda 2014 NYNNYNNNZhao et al . ( 2014 ) NUNYYNNNZhou et al . ( 2014 ) NNYNYNNNYang et al . ( 2015 ) NNYNNNNNLegend: N = No , U = Unclear , Y = Yes . Letters following author and year ( e . g . Mei 2010A ) indicate that more than one independent experiment was conducted in the same publication . Large Animal Model: Yes = Sheep , pig , dog , monkey . No = Mouse , ratAdult Animal Model: Yes = Rats ≥ 6 weeks old , mice ≥ 8 weeks old . No = Rats < 6 weeks old , mice < 8 weeks old . Unclear = No age statedComorbidities: Yes = e . g . Diabetes , obesity , immunodeficiency . No = No comorbidities . Infectious Model of Sepsis: Yes = Cecal-ligation and puncture , live bacterial administration . No = Bacterial product administration ( e . g . lipopolysaccharide ) . Therapy Initiated After Sepsis Induction: Yes = Mesenchymal stromal cells administered after sepsis model induced . No = Mesenchymal stromal cells administered at the time of sepsis induction . Documented Sepsis Severity Prior to Initiating Treatment: Yes = Mesenchymal stromal cells administered after marker of severity ( e . g . hypotension ) measured . No = Mesenchymal stromal cells administered without a marker of severity being measured . Resuscitation Included Fluids: Yes = Fluid therapy ( aside from vehicle for cell administration ) administered . No = Only vehicle for cell administration or no fluids administered . For the 20 experiments , 50% demonstrated statistically significant beneficial effects of MSCs with a median sample size of 19 animals per group . Visual inspection of a funnel plot analysis of all experiments suggested that publication bias exists ( Figure 4 ) , which was confirmed by Egger regression ( p=0 . 019 ) . Post-hoc trim and fill analysis suggested a relative overestimation of effect size of 27% , although MSCs remained associated with a statistically significant reduction in mortality after adjustment ( OR 0 . 34 , 95% CI 0 . 22–0 . 52 ) . 10 . 7554/eLife . 17850 . 021Figure 4 . Funnel plot to detect publication bias . Trim and fill analysis was performed on overall mortality . Open circles denote original data , black circles denote ‘filled’ studies . Open diamond denotes original pooled effect size ( log odds ratio ) and 95% confidence interval . Filled diamond represents adjusted effect size and 95% confidence interval . DOI: http://dx . doi . org/10 . 7554/eLife . 17850 . 021
Preclinical studies provide necessary justification to conduct a first-in-human clinical trial . Thus , a systematic review approach offers an attractive method to comprehensively synthesize the totality of available evidence . Our systematic review demonstrates that MSC therapy reduces the odds of death in preclinical animal sepsis models . This effect is maintained over a range of time periods ( less than two days , between two to four days , and longer than four days ) . These early outcome windows capture the majority of deaths that occur in these acute models . Moreover , the effect sizes are robustly maintained ( replicated ) over a variety of experimental conditions , varying models , and differing MSC immunologic compatibility ( e . g . allogeneic vs . syngeneic ) . It has been suggested that individual study findings have low probability of being ‘true’ ( Ioannidis , 2005 ) , however by aggregating results of similar experiments the positive predictive value of a finding dramatically increases ( Moonesinghe et al . , 2007 ) . Thus , the findings of this systematic review helped support our decision to initiate a Phase 1/2 trial to evaluate the safety of MSC therapy in human patients with septic shock ( NCT02421484 ) . We believe our approach of systematically reviewing preclinical evidence is widely applicable for researchers considering first-in-human studies . Although our synthesis suggests MSC treatment of sepsis may be beneficial these results are tempered by the presence of potential threats to validity . Our preclinical systematic review evaluated internal , external , and construct validity of the data . Methodological weaknesses ( i . e . poor internal validity ) in clinical trials are associated with an exaggeration of the treatment effect . Similarly , in preclinical studies , failure to address selection bias ( through methods such as randomization and allocation concealment ) and detection bias ( through blinded outcome assessment ) results in significantly increased effect sizes ( Crossley et al . , 2008; Hirst et al . , 2014; Rooke et al . , 2011 ) . The significance of selection and detection bias has been acknowledged by The National Institutes of Health’s recently issued guidelines for reporting preclinical research . These guidelines have specifically proposed randomization , blinding , and sample size calculations as key methodological information that must be described in preclinical reports ( National Institutes of Health , 2015; Landis et al . , 2012 ) . In our review , none of the included studies reported randomization or allocation concealment in a manner that could be considered at low risk of bias . Similarly , no studies reported appropriate a priori defined sample sizes . Most of these items were judged as ‘unclear’ in our risk of bias evaluation due to the convention to judge unreported items as ‘unclear’ rather than ‘high risk’ . We speculate that many of these ‘unclear’ items were not performed ( i . e . they were ‘high risk’ ) due to a general lack of training of basic scientists in methods to reduce risk of bias ( Landis et al . , 2012; Collins and Tabak , 2014 ) . This lack of reporting precluded an evaluation of their efforts and points to the need to improve the methodology used in preclinical investigations . To address external validity ( i . e . generalizability ) we performed a number of subgroup analyses . Overall , subgroup analyses suggested that MSC effects appeared to be robust over a number of varying experimental conditions and across a number of different laboratories . Results of specific subgroups ( e . g . autologous cells , multiple doses , intraperitoneal administration , and adipose tissue source ) should be interpreted cautiously as few studies were included in these groups , and the results of one study with differing results ( Chang et al . , 2012 ) may have skewed data . The ability of one study to heavily influence overall effect estimates is a short-coming of meta-analyses that include few studies . As such , these subgroup analyses should be treated as exploratory . Despite the large effect sizes noted , one must bear in mind the potential effect of publication bias ( i . e . bias due to the publication of only positive studies ) . Our funnel plot demonstrated a highly asymmetrical pattern and our trim and fill analysis indicated that a number of unpublished negative studies may exist . This is in keeping with previous analyses of preclinical stroke data that suggested up to one in six animal studies in that field were unreported and unpublished . ( Sena et al . , 2010 ) Our inability to analyze these potential studies may have led to an overstatement of effect size . To evaluate the potential clinical applicability of these results , we examined the construct validity of included studies . This was determined using recommendations that had been developed to improve the clinical generalizability of preclinical sepsis studies ( Lamontagne et al . , 2010 ) . Animal sepsis models may not be representative of human sepsis because of the timing and severity of sepsis induction , the dose and timing of the treatment in relation to sepsis induction , the use of small/young animals without comorbid illnesses , and lack of administration of standard of care co-interventions such as fluids and antibiotics during the study period . How well animal models of sepsis mimic the pathophysiology of human sepsis has also been a contentious issue ( Dyson and Singer , 2009; Osuchowski et al . , 2014; Seok et al . , 2013 ) . Only two studies incorporated a majority of elements addressing construct validity , thus the effect of construct validity on MSC therapy of sepsis remains to be determined . There are a number of other issues of note that may impact the translation of MSC therapy to the clinical setting . First , although we did not formally evaluate characterization of cell products , this was variably reported in the included studies . Differences in the quality of cell therapeutics may have accounted for some of the heterogeneity of results observed . Second , dosing of cell products was not equivalent between species , even after adjusting for total cells given . Equivalence dosing of drugs between species is a complex issue and the FDA has endorsed conversion based on body surface area , rather than a dose per weight basis . ( Reagan-Shaw et al . , 2008 ) Applying this guidance , 1 million cells in a mouse may be equivalent to 0 . 5 million in a rat; similarly , this dose in a human would be roughly equivalent to 3 million cells/kg . These equivalencies should be interpreted cautiously given the differences between typical drug therapies and the cellular therapy evaluated in our review . Third , the severity of disease in these animal models at the time of MSC administration is unclear . Based on our experience with endotoxemia and cecal-ligation and puncture models , at 1 hr after disease induction some symptoms may be apparent and after 6 hr most animals have both biochemical and physiological evidence of inflammation and organ-dysfunction . Thus , we performed a subgroup analysis based on timing of administration <1 hr , >1–6 hr , >6 hr as a rough correlate to early and more delayed ( intermediate and late ) administration of cells in an attempt to simulate the delays in treatment that may be seen in humans who present with severe infection . Of note , no study administered cells at a late time point . A clearer reporting of disease severity at time of cell administration may allow a more precise analysis of when these cells are more ( or less ) efficacious . A fourth issue is the lack of transparent reporting of risk of bias elements that minimize the ability to evaluate threats to validity in our systematic review . We would suggest that general poor understanding of these core methodological issues may underlie their incomplete reporting . In order to increase the robustness and interpretation of future preclinical systematic review results we submit that authors of primary studies and journal editors should ensure adherence to published reporting guidelines for pre-clinical research studies ( National Institutes of Health , 2015; Kilkenny et al . , 2010 ) . These guidelines not only detail items relating to risk of bias ( e . g . randomization and blinding ) but also touch on issues that are very important when primary studies are included in systematic reviews ( e . g . differentiating between biological and technical replicates , providing exact n numbers ) . The strengths of our systematic review are in the transparent and thorough literature search and an attempt to examine potential for translation by evaluating threats to validity . To date , three clinical trials have been initiated following a systematic review and meta-analysis of animal data ( McCann et al . , 2014; van der Worp et al . , 2007 ) ; all have repurposed currently used interventions for neurological conditions and are currently recruiting ( NCT01833312 , NCT01910259 , ISRCTN83290762 ) . To the best of our knowledge ours is the first preclinical systematic review that has evaluated a novel biological therapeutic in preparation for a high risk first-in-human clinical trial . The limitations of our review should be noted . First , we restricted our search to unmodified MSCs since our group was only considering a clinical trial of unmodified cells for sepsis . Although modified cells may be of clinical interest , there are a number of additional regulatory , ethical , and safety issues which significantly increase the complexity of clinical trials using these cells; this is an issue that members of our group have experienced first-hand ( NCT00936819 ) . Other limitations of our review relate to the potential methodological limitations of the included studies . None of the included studies were considered low risk of bias across all domains , and their construct validity was highly variable . It is unclear what the influence of these methodological limitations might be in this particular study due to our inability to perform meaningful subgroup analyses . Our evaluation of the methodological aspects of included studies also relied on what the authors reported , and this may have been incomplete in cases . We would suggest however , that similar to other fields , the failure to address threats to internal validity likely contributes to an exaggerated effect size . Despite the stated limitations of this review , the consistency of the results across the included studies and the large effect size suggest that MSCs reduce the odds of death in preclinical models of sepsis . Moreover , there are a number of studies that have demonstrated biological mechanisms that may underlie the benefits of MSCs in sepsis , including antibacterial , anti-inflammatory , and trophic effects ( Spees et al . , 2016 ) . These mechanisms do not require engraftment and have been demonstrated to work over thousands of molecular pathways that include improved cellular energetics and activation of macrophages ( dos Santos et al . , 2012 ) . Given the results of our review along with this biological plausibility , our group gained the support of regulatory agencies , ethics boards , and other stakeholders to proceed to a first-in-human clinical trial . Nonetheless , our efforts to translate this therapy into a clinical trial were tempered by the limitations of the preclinical studies performed to date . If this support was not provided , alternative methods to address efficacy of MSC therapy for sepsis could include conducting a low risk of bias ‘confirmatory’ preclinical study that was informed by the results of this systematic review ( Kleikers et al . , 2015 ) , or performing a multicenter preclinical study ( Llovera et al . , 2015 ) . Ultimately , ongoing and future clinical evaluations will determine whether the therapeutic effects of MSCs will translate to the human patient population .
The research question for this review was , “In preclinical in-vivo animal models of sepsis , what is the effect of MSC therapy ( compared to control treatment ) on death ? ” The protocol for this review was published on the Collaborative Approach to Meta Analysis and Review of Animal Data from Experimental Studies ( CAMARADES ) website ( http://www . dcn . ed . ac . uk/camarades/research . html#protocols ) and also the University of Ottawa’s Open Access Research Institutional Repository ( http://hdl . handle . net/10393/32833 ) . A priori publication of our protocol encourages transparency in the systematic review process and safeguards against reporting biases in the review . This review is reported in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analysis ( PRISMA ) Statement ( Moher et al . , 2009 ) . The PRISMA guidelines are an evidence-based minimum set of items that should be reported in a systematic review and meta-analysis . Similar to other reporting guidelines , PRISMA ensures complete and transparent reporting of a study . We included all pre-clinical in vivo studies of sepsis and endotoxemia that investigated treatment with mesenchymal stromal cells . MSCs must have been administered during or after experimental induction of sepsis . Since our group was considering a clinical trial of unmodified MSCs , studies were excluded if the MSCs were differentiated , altered , or engineered to over or under express particular genes . Neonatal animal models were excluded , as were models of acute lung injury . Finally , studies where MSCs were administered with another experimental therapy or cell type were excluded . To identify all relevant studies , we designed a search strategy in collaboration with a medical information specialist . We would suggest readers consult a medical librarian experienced in systematic searches if they wish to perform a literature search for a preclinical systematic review; this will ensure a comprehensive search is conducted . Although MSC terminology has been codified ( Dominici et al . , 2006 ) non-standard terms continue to be used in the literature , thus a number of MSC related terms were used in the search strategy . Validated animal filters were applied to increase relevancy ( de Vries et al . , 2014; Hooijmans et al . , 2010 ) ; post-hoc , an inadvertent truncation was noted in the application of these filters , thus an updated search was performed to include the complete filters . We searched Ovid MEDLINE In-Process and Other Non-Indexed Citations , Embase Classic+Embase , BIOSIS and Web of Science ( using Web of Knowledge ) from inception until May 2015 . The full search strategy is listed in the Appendix . Additional references were also sought through hand-searching the bibliographies of reviews and included primary studies . Studies were independently screened by two reviewers , with consensus required for articles to proceed to either the next screening stage or to the final analysis . Disagreements were resolved by discussion or by consultation with a senior team member when necessary . Data was extracted on the general characteristics of the study ( e . g . study design , country of origin , sample size ) , animal model ( e . g . disease induction method , use of resuscitation ) , and mesenchymal stromal cells ( e . g . condition and source of cells ) . Data was collected for the primary outcome of overall mortality . Mortality was further stratified by time: ≤ 2 days , > 2–≤ 4 days , and > 4 days . If multiple measurements were reported within a period , the latest measurement within the period was used . Data in graphical format was extracted using open source software ( Engauge Digitizer , github . com; http://markummitchell . github . io/engauge-digitizer/ ) . Extracted data were verified by a second reviewer with disagreements resolved by consultation with a third team member . Additionally , authors were contacted when further clarification was required . A priori determined subgroup analyses were conducted to determine the effects of important factors on the estimated treatment effect . These analyses were performed to assess generalizability of results over varying experimental conditions . Subgroups were analysed for the following: animal model ( e . g . mice , rat ) , gender , experimental model ( e . g . cecal ligation and puncture , endotoxemia ) , source of MSC ( e . g . autologous , xenogenic ) , route of MSC administration ( e . g . intravenous , intraperitoneal ) , dose of MSC ( less or greater than 1 . 0 × 106 cells ) , frequency of MSC dose , timing of MSC administration ( less than one hour , greater than 1 hr to less than or equal to 6 hr , greater than 6 hr , or multiple dosing ) , resuscitation used ( e . g . fluid , antibiotics ) , and control group ( phosphate buffered saline , fibroblasts , normal saline , medium , nothing administered ) . Given the number of analyses performed , the results were considered exploratory and hypothesis generating . Readers employing a similar analysis may consider adjusting the value of significance based on the number of comparisons ( e . g . for 11 analyses p<0 . 0045 would be considered significant ) . Risk of bias was assessed independently in duplicate as high , low , or unclear for the six domains of bias identified by the Cochrane Risk of Bias tool ( Higgins and Green , 2009 ) . Domains include: ( 1 ) sequence generation , ( 2 ) allocation concealment , ( 3 ) blinding of participants and personnel , ( 4 ) blinding of outcome assessors , ( 5 ) incomplete outcome data , and ( 6 ) selective outcome reporting; operational definitions can be found in the legend for Table 2 . Any disagreements were resolved through consultation with a senior member of the team . Other domains of risk of bias assessed were ( 1 ) source of funding , ( 2 ) conflict of interest , and ( 3 ) sample size calculations . Following reviewers’ suggestions we also included the SYRCLE Risk of Bias Tool , an alternative method of assessing risk of bias in preclinical animal studies ( Hooijmans et al . , 2014 ) . This tool is largely based on the Cochrane Risk of Bias Tool and includes several additional domains: ( 1 ) similarity of groups or adjustment for confounders at baseline , ( 2 ) random housing of animals , ( 3 ) animal selection at random for outcome assessment . The last domain was not evaluated given the outcome being assessed was death , and it was unclear for most studies whether true death or surrogate measures were being evaluated . In preclinical studies construct validity refers to the extent an animal model corresponds to the clinical entity it is intended to represent ( Henderson et al . , 2013 ) . We used a previously published framework to evaluate construct validity of the included studies ( Lamontagne et al . , 2010 ) . Items evaluated in each study included: ( 1 ) use of a large animal model ( e . g . pig , dog , sheep ) , ( 2 ) use of adult animals , ( 3 ) presence of co-morbid diseases , ( 4 ) use of an infectious model of sepsis , ( 5 ) documentation of severity of illness prior to initiating therapy , ( 6 ) follow-up duration ≥24 hr , ( 7 ) use of antibiotics , and ( 8 ) use of intravenous fluid resuscitation . Each item was assessed independently by two reviewers and assessed as either a ‘yes’ or a ‘no’ . Disagreements were resolved by consultation with a third team-member . Statistical analysis was performed in consultation with a statistician experienced in systematic reviews and meta-analysis . Readers seeking to replicate these methods for their own purposes are encouraged to similarly seek advice from an experienced statistician . Data from studies were pooled using meta-analysis that was performed with random effects modeling employing the DerSimonian and Laird random effects method ( Comprehensive Meta-Analysis 2 . 0 , Englewood , USA ) . Outcomes are expressed as odds ratios and 95% confidence intervals . There were completely independent control groups for the studies with more than one experiment extracted ( i . e . a control group was not shared between two experimental groups ) . Thus , no correction for the number of control animals was required for multiple comparisons within a single meta-analysis . Heterogeneity of effect sizes in the overall effect estimates was assessed using the I2 statistic . The following are suggested thresholds to interpret the I2 statistic: 0–40% may not be important , 30–60% moderate heterogeneity , 50–90% substantial heterogeneity , 75–100% considerable heterogeneity ( Higgins and Green , 2009 ) . Presence of publication bias was assessed using a funnel plot ( visually ) and Egger regression test ( statistically ) . The funnel plot is a scatterplot of the intervention effect of individual studies plotted against a measure of its precision or size . The characteristic ‘inverted funnel’ shape arises from the fact that precision of the effect estimate increases as the as the study size increases ( i . e . small studies will scatter more widely at the bottom of the funnel ) . A funnel plot would normally be expected to symmetrical , however the absence of symmetry can suggest publication bias ( Sterne et al . , 2011 ) . Duval and Tweedie’s trim and fill estimates were generated to estimate the number of missing studies and to estimate the adjusted effect size assuming the studies were present . | Most attempts to transform exciting findings from laboratories into clinical treatments are unsuccessful . One reason for this may be the failure to consider all of the laboratory work that has been performed before deciding to test a treatment on patients for the first time . In particular , negative findings ( that suggest that a potential new treatment is ineffective ) may be overlooked . Stem cells may help to treat life-threatening infections , but this has not been tested in human patients . However , the effectiveness of stem cell treatments has been tested in animals that act as models of human infection . Before deciding to begin a clinical trial of stem cell therapy for life-threatening infections , Lalu et al . performed an exhaustive search to find all the studies in which stem cells were used to treat animal models of infection . Combining the results of all of these studies using particular analysis techniques revealed that stem cell therapy increased the survival of these animals overall . These positive effects were seen over a range of different experimental conditions ( for example , when treating the animals with different doses of stem cells , or giving the doses at different times ) . Lalu et al . also identified some limitations with most of the laboratory studies that had tested stem cell therapy for infections . Many of the studies used animal models that may not be the best representations of humans with severe infection . In addition , many of the scientists did not report that they had used methods ( such as randomization ) that would generate the most confidence in their results . Despite these limitations , there was a lot of consistency in the reported results . Overall , the results support the decision to proceed to a clinical trial that tests the effectiveness of stem cells for treating human infections . More generally , Lalu et al . ’s analysis demonstrates a way of considering all laboratory evidence before deciding to proceed to a first clinical trial in humans . This may help researchers to identify promising therapies to further develop , and also to identify potential failures before they are tested in patients . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"epidemiology",
"and",
"global",
"health",
"medicine"
] | 2016 | Evaluating mesenchymal stem cell therapy for sepsis with preclinical meta-analyses prior to initiating a first-in-human trial |
Emiliania huxleyi is a model coccolithophore micro-alga that generates vast blooms in the ocean . Bacteria are not considered among the major factors influencing coccolithophore physiology . Here we show through a laboratory model system that the bacterium Phaeobacter inhibens , a well-studied member of the Roseobacter group , intimately interacts with E . huxleyi . While attached to the algal cell , bacteria initially promote algal growth but ultimately kill their algal host . Both algal growth enhancement and algal death are driven by the bacterially-produced phytohormone indole-3-acetic acid . Bacterial production of indole-3-acetic acid and attachment to algae are significantly increased by tryptophan , which is exuded from the algal cell . Algal death triggered by bacteria involves activation of pathways unique to oxidative stress response and programmed cell death . Our observations suggest that bacteria greatly influence the physiology and metabolism of E . huxleyi . Coccolithophore-bacteria interactions should be further studied in the environment to determine whether they impact micro-algal population dynamics on a global scale .
There are many microbes that have influenced Earth's biogeochemistry . Prime among these are the coccolithophores , a diverse group of unicellular marine algae of the haptophyte division . Because of their high abundance , these micro-algae are fundamental in the global oxygen , carbon , and sulfur cycles ( Balch et al . , 1991; Beaufort et al . , 2011; Simó , 2001; Field et al . , 1998 ) . As a consequence of their photosynthetic capacity these algae , together with other phytoplankton , are responsible for nearly half of our planet’s primary production ( Field et al . , 1998 ) . Coccolithophore cells are usually surrounded by elaborate platelets made of crystalline calcium carbonate ( calcite ) referred to as coccoliths . During coccolith production , carbon dioxide is released and can escape from the ocean to the atmosphere ( Marsh , 2003 ) . More importantly , coccoliths serve as a carbon sink as they accumulate on the bottom of the oceans ( Sabine et al . , 2004 ) . Coccolith production by E . huxleyi accounts for roughly 1/3 of the total marine calcium carbonate production ( Iglesias-Rodríguez et al . , 2002 ) . Hence , coccolithophores play a complex role in the global carbon cycle . Emiliania huxleyi is the most widespread coccolithophore in modern oceans , forming dense annual blooms ( Paasche , 2001 ) . The blooms can cover thousands of square kilometers of ocean surfaces and are easily detected by satellites due to the highly reflective nature of the coccoliths ( Balch et al . , 1991; Holligan et al . , 1983 ) . The blooms also exhibit unique dynamics; they form seasonally over several weeks and then suddenly collapse ( Behrenfeld and Boss , 2014; Lehahn et al . , 2014; Tyrrell and Merico , 2004 ) , a process that has been attributed to viral infection ( Bratbak et al . , 1993; Lehahn et al . , 2014; Vardi et al . , 2012 ) . Recent evidence suggests that environmental stresses and viral infection can trigger oxidative stress and a process similar to programmed cell death ( PCD ) in E . huxleyi ( Bidle et al . , 2007; Vardi et al . , 2009; Bidle , 2016 ) . The induction of PCD , which is an autocatalytic process , has been shown to occur in various widespread species of phytoplankton including E . huxleyi , and functional links have been demonstrated between viral infection , PCD , and algal bloom collapse ( Bidle , 2015 , 2016; Bidle and Vardi , 2011; Fulton et al . , 2014; Vardi et al . , 2009 , 2012; Rohwer and Thurber , 2009 ) . Interestingly , although E . huxleyi blooms harbor a rich community of bacteria , at times dominated by the Roseobacter group ( González et al . , 2000; Green et al . , 2015 ) , bacteria are not generally considered to be a factor influencing coccolithophore physiology and bloom dynamics . Various types of phytoplankton were shown to have both mutualistic and antagonistic interactions with bacteria ( Amin et al . , 2015; Miller and Belas , 2004; Miller et al . , 2004; Wang et al . , 2014; Durham et al . , 2015 ) . In addition , the possible role of algicidal bacteria in the ocean has been examined and discussed ( Mayali and Azam , 2004; Harvey et al . , 2016 ) . It has been previously suggested by our laboratories that bacteria might interact with E . huxleyi ( Seyedsayamdost et al . , 2011 ) . However , coccolithophore-bacteria interactions have not yet been unambiguously demonstrated . This gap is curious because E . huxleyi's important role in the global sulfur cycle is in part a consequence of an algal-bacterial interaction . E . huxleyi produces the osmolyte and antioxidant dimethylsulfoniopropionate ( DMSP ) ( Sunda et al . , 2002 ) . This molecule , when released into the water by leakage or cell lysis , can be used by some bacteria as a source of sulfur and carbon ( Curson et al . , 2011; González et al . , 1999 ) . During DMSP catabolism , bacteria such as Roseobacters produce the volatile by-product dimethyl sulfide ( DMS ) . E . huxleyi is also a producer of DMS , which is a bioactive gas with possible roles in climate regulation ( Charlson et al . , 1987; Alcolombri et al . , 2015 ) . When DMS enters the atmosphere it is oxidized and serves to form cloud condensation nuclei ( Curson et al . , 2011; González et al . , 1999 ) . While the DMSP flux from algae to bacteria , and the production of DMS gas by both algae and bacteria have been clearly demonstrated , the role of DMS in climate regulation has been questioned ( Quinn and Bates , 2011 ) . Accumulating evidence suggests that there may be widespread interactions between E . huxleyi and Roseobacters . Phaeobacter inhibens ( Buddruhs et al . , 2013 ) , a well-studied member of the Roseobacter group , was shown to produce molecules that specifically affect E . huxleyi ( Seyedsayamdost et al . , 2011 ) . This bacterium , when grown in a pure culture in the presence of p-coumaric acid , a product released by aging algae , produced novel compounds able to lyse E . huxleyi . The compounds were named roseobacticides and their discovery pointed towards a possible interaction between P . inhibens and E . huxleyi ( Seyedsayamdost et al . , 2011 ) . Furthermore , we recently showed that lipid metabolism in E . huxleyi is altered in the presence of P . inhibens ( Segev et al . , 2016 ) . However , a direct physical interaction between these algae and bacteria had not been previously described and no other details of their interaction were known . Here we describe the establishment of a co-culture model system between E . huxleyi and P . inhibens that allows the examination of the spatiotemporal dynamics of their interaction . We provide evidence that E . huxleyi and P . inhibens associate intimately when co-cultured . We show that bacteria promote algal growth but eventually kill their aging algal hosts . The same bacterial compound , indole-3-acetic acid , mediates stimulation of algal growth as well as algal death . Finally , algal death in the co-culture seems to involve an apoptotic process . Similar E . huxleyi - bacteria interactions might occur in the ocean and could thus affect algal physiology , bloom dynamics and biogeochemical cycles .
E . huxleyi and P . inhibens are two well-studied marine microbes . To determine if they co-occur in algal blooms we analyzed the bacterial community associated with E . huxleyi blooms using a culture-independent metagenomic approach . Two independent blooms were sampled in the Gulf of Maine during the summer of 2015 . The results shown in Figure 1 indicate that P . inhibens was indeed found co-occurring with E . huxleyi in algal blooms ( Figure 1 ) . Thus , the suggested interaction between these microorganisms might be ecologically significant . To study the interactions of E . huxleyi and P . inhibens , it was necessary to establish conditions to co-culture these two species . We started by examining pure cultures of each microorganism . Coccolith-forming ( i . e . calcifying ) E . huxleyi ( strain CCMP3266 ) were inoculated into L1-Si , a seawater based medium supplemented with additional sources of phosphorus ( 0 . 04 mM PO4 ) , nitrogen ( 0 . 9 mM NO3 ) and sulfur ( 0 . 08 µM SO4 ) , along with vitamins and trace metals ( Guillard and Hargraves , 1993 ) ( see Materials and methods ) . In this medium , E . huxleyi grows to 3×105 cell/ml . Under these conditions E . huxleyi produces calcium carbonate coccoliths that surround the algal cell ( Figure 2a ) . P . inhibens DSM17395 is normally grown in the rich medium 1/2YTSS ( Seyedsayamdost et al . , 2011 ) ( see Materials and methods ) where it easily aggregates; it often forms ‘rosette’ structures through a polysaccharide-containing pole ( Figure 2b , c ) ( Segev et al . , 2015 ) . Of note , alone these bacteria do not grow in the L1-Si medium ( Figure 2d , grey bars ) . However , we found that bacteria do grow in co-culture with E . huxleyi . To grow a co-culture , we inoculated algae into L1-Si medium and , after four days , introduced bacteria into the algal culture . In these co-cultures , bacterial numbers increased nearly five orders of magnitude over a period of 14 days ( Figure 2d , green bars ) . Microscopic examination of the co-culture revealed that some algae were no longer surrounded by coccoliths ( Figure 2e ) . Rather , naked algal cells were now covered by bacteria attached via their poles . This attachment was evident in both fixed ( Figure 2e ) and live ( Figure 2f ) samples . Of note , attachment of P . inhibens to other micro-algae as well as macro-algae has been previously demonstrated ( Frank et al . , 2015 ) . Using a specific fluorescent lectin to detect the polar polysaccharide ( see Materials and methods ) , it appeared that bacterial attachment onto the algal cell also involves the polar bacterial polysaccharide ( Figure 2f ) . Examination of co-cultures revealed that over time more algae have attached bacteria ( Figure 2g ) and each algal cell is associated with increasing numbers of bacteria as the co-culture ages ( Figure 2f , h ) . 10 . 7554/eLife . 17473 . 003Figure 1 . Metagenomic analysis of Roseobacters associated with E . huxleyi blooms reveals co-occurrence of P . inhibens . Two E . huxleyi blooms were sampled in the Gulf of Maine during the summer of 2015 and metagenomic analysis of the bacterial population was performed ( see Materials and methods ) . Shown is the relative abundance of members of the Rhodobacteraceae family , which accounted for 6% of bacteria . The same members of the Rhodobacteraceae family were detected in both blooms and their abundance changed ±2% between replicates and between the two blooms . P . inhibens was present in both blooms and is indicated by an asterisk . Shown are the results for the July 2015 bloom ( see Materials and methods ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17473 . 00310 . 7554/eLife . 17473 . 004Figure 2 . Algal-bacterial co-cultures . ( a ) Scanning electron microscopy ( SEM ) image of E . huxleyi ( CCMP3266 ) pure algal culture . ( b ) SEM image of P . inhibens ( DMS17395 ) pure bacterial culture . ( c ) Overlay image of a pure culture of P . inhibens bacteria ( phase contrast microscopy , grey ) stained with a fluorescent lectin ( Alexa Fluor 488 conjugated lectin , green ) . ( d ) Bacteria grown in L1-Si medium in the absence ( grey bars ) and presence ( green bars ) of algae over 20 days . Error bars represent the standard deviation of two biological replicates . ( e ) SEM image of cells from an algal-bacterial co-culture . ( f ) Phase contrast microscopy imaging of live co-culture samples ( grey ) overlaid with images of the fluorescent lectin ( Alexa Fluor 488 conjugated lectin , green ) showing increasing numbers of bacteria attaching onto algal cells over time . ( g ) Quantification of algal cells with attached bacteria as a function of time , n > 300 . Error bars represent the standard deviation between the multiple examined fields . ( h ) Quantification of the number of attached bacteria per algal cell as a function of time , n > 300 . Error bars represent the standard deviation between the multiple examined fields . All scale bars in the figure correspond to 1 µm . Statistical significance was calculated using a Student’s T-test and p values are presented above datasets . DOI: http://dx . doi . org/10 . 7554/eLife . 17473 . 004 Bacteria clearly benefit from interacting with the algal host as their growth is enabled by the algae in an otherwise non-permissive medium ( Figure 2d ) . What do bacteria receive from algae that allows them to grow ? Given that L1-Si does not contain significant amounts of organic carbon to permit robust growth of the heterotrophic bacteria , it stands to reason that the key nutrient that algae provide is fixed carbon . If indeed fixed carbon were to be the sole nutrient needed by the bacteria , addition of a utilizable form of carbon to L1-Si should enable bacterial growth . However , addition of 5 . 5 mM glucose did not lead to significant bacterial growth ( Figure 3a ) . This was an unexpected result because , as mentioned above , L1-Si in addition to seawater also contains added phosphorus ( 0 . 04 mM PO4 ) , nitrogen ( 0 . 9 mM NO3 ) and sulfur ( 0 . 08 μM SO4 ) . In fact , even addition of higher nutrient concentrations in forms shown to be utilizable by P . inhibens ( Zech et al . , 2009 ) as individual supplements ( nitrogen 5 mM NH4 , phosphorous 2 mM PO4 , and sulfur 33 mM SO4 ) or in various combinations of two or three of them did not lead to robust bacterial growth ( Figure 3a ) . Only addition of all four essential nutrients resulted in bacterial growth to a density of 5×108 CFU/ml , which we normalized to 100% in Figure 3a . Thus , E . huxleyi can provide all four essential nutrients ( C , N , P and S ) in suitable forms and concentrations to enable growth of the heterotrophic bacterium P . inhibens . 10 . 7554/eLife . 17473 . 005Figure 3 . Bacteria require essential nutrients to grow in L1-Si . ( a ) Bacterial growth in L1-Si medium supplemented with various essential nutrients ( C-glucose , N-nitrogen , P-phosphorus , S-sulfur ) was monitored over eight days ( see Materials and methods ) . Presented are the maximal growth values that were obtained after three days of incubation . The initial bacterial inoculum was 1×105 CFU/ ml . Growth in CNPS reached 5×108 and was normalized to 100% . ( b ) Bacteria consume externally added DMSP . ( c ) DMSP production by E . huxleyi in pure culture ( black bars ) and in co-culture ( grey bars ) . Error bars represent the standard deviation between two biological replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 17473 . 005 The sulfur flux from algae to bacteria is of special interest . Because of its ecological importance , we wanted to investigate whether DMSP plays a role in the interaction between E . huxleyi and P . inhibens . First , we examined the ability of bacteria to grow with DMSP as a sole source of sulfur or carbon . As shown in Figure 3a , DMSP can serve as a sulfur source ( Figure 3a , 'CNP + DMSP 30 μM' ) . In contrast , DMSP does not supply sufficient carbon to support robust bacterial growth . ( Figure 3a 'NP + DMSP 30 μM' ) . Even when DMSP was added in higher concentrations to supply carbon in a comparable amount to the carbon supplied by the 5 . 5 mM glucose in the parallel experiments , bacterial growth was not evident ( Figure 3a ‘NP + DMSP 6mM’ ) . Next , we directly monitored DMSP consumption in a growing bacterial culture . Our measurements indicate that all of the added DMSP is rapidly utilized by the growing bacteria whereas in un-inoculated medium the DMSP levels remain relatively stable over time ( Figure 3b ) . Based on these observations , we then proceeded to determine whether DMSP is produced and exuded by algae in pure culture and in co-cultures . Indeed , over a period of 17 days , DMSP concentration in the medium of a pure algal culture increased , reaching a concentration of nearly 6 µM ( Figure 3c , black bars ) . In contrast , in a co-culture of algae and bacteria , DMSP levels in the medium were nearly undetectable , presumably due to its rapid metabolism by the bacteria ( Figure 3c , grey bars ) . Since bacteria are directly attached to their algal host it is possible that they experience a considerably higher local concentration of DMSP than the concentration measured in the bulk medium . It has been reported that Roseobacters and other bacteria can chemotax towards DMSP and catabolize it and various metabolic pathways for the bacterial use of the DMSP sulfur have been proposed and tracked ( Miller and Belas , 2004; Miller et al . , 2004; Seymour et al . , 2010; Brock et al . , 2013; Wang et al . , 2016 ) . Thus , it is possible that DMSP serves as a chemical cue attracting bacteria to colonize the E . huxleyi host cell . Additional experiments are clearly needed to determine if indeed DMSP serves as an infochemical promoting bacterial colonization . Yet , it has been previously shown that algal DMSP and exudates serve as a strong cue to attract bacteria ( Seymour et al . , 2010; Smriga et al . , 2016 ) . To further characterize the algal-bacterial interaction we explored the bacterial effect on algal growth . Flow cytometry is commonly used to monitor algal growth . However , due to bacterial attachment and algal clumping , our co-cultures consist of aggregates of cells of varying sizes . Thus , it is challenging to accurately interpret the results of flow cytometry analyses . Therefore , our analyses included traditional flow cytometry as well as validation of our results using imaging cytometry in order to characterize the different particles ( see Materials and methods ) . We found that during the initial 10 days of culturing , there were greater numbers of algae in the co-cultures compared to pure algal cultures ( Figure 4a ) . In the co-culture , algae reach approximately 25% higher numbers in comparison with the maximum reached in the absence of bacteria . In addition , in an algal pure culture the death phase starts after day seven while in a co-culture a significant decrease in population is evident only at day 17 . Interestingly , the death phase in an algal pure culture seems more gradual than the rapid demise observed in a co-culture ( Figure 4a ) . Thus , it seems that the algal-bacterial interaction is dynamic . Initially the interaction is mutualistic , however over time bacteria may become harmful for their algal hosts . 10 . 7554/eLife . 17473 . 006Figure 4 . Algal growth in the presence of bacteria . ( a ) Algal growth was monitored over 17 days in the presence of bacteria ( green bars ) and in pure culture ( black bars ) . ( b ) Indole-3-acetic acid ( IAA ) production was observed in pure bacterial cultures grown in L1-Si supplemented with essential nutrients . Shown is an LC-MS extracted ion chromatogram ( EIC , m/z 176 . 0706 ± 10 ppm ) of an IAA standard of 500 nM ( black ) and two biological replicates ( red ) ( see Materials and methods ) . AU=Arbitrary Units . ( c ) Algal growth was examined upon addition of the auxin IAA . Percent algal growth is relative to a culture with no IAA added . Note that at a concentration of 1000 μM IAA , cell numbers dropped to less than 10% . ( d ) Following treatment with 1 mM IAA , the autofluorescence signal of dead algal cells ( the lower cell in this image ) appears similar to the signal observed in bacterially induced death ( compare with Figure 6l ) , indicative of chloroplast deformation but partially intact cell membrane . Scale bar corresponds to 4 μm . Error bars in a and c represent the standard deviation between two biological replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 17473 . 00610 . 7554/eLife . 17473 . 007Figure 4—figure supplement 1 . Validating the detection of bacterially produced IAA . ( a ) The accurate mass of the bacterial IAA was determined using High Resolution LC-MS . Identical m/z values were obtained for an IAA standard and bacterially produced IAA . ( b ) Both IAA standard and bacterial IAA exhibit characteristic MS/MS fragmentation patterns , such as the shown fragment m/z 130 . DOI: http://dx . doi . org/10 . 7554/eLife . 17473 . 007 Various bacteria that interact with plants are known to provide phytohormones ( auxins ) that promote plant growth ( Costacurta and Vanderleyden , 1995 ) . It was previously suggested that phenylacetic acid ( PAA ) produced by P . inhibens might serve as an auxin to enhance algal growth ( Seyedsayamdost et al . , 2011 ) . We could not detect PAA in bacterial cultures or algal-bacterial co-cultures . Intriguingly , the auxin indole-3-acetic acid ( IAA ) was previously demonstrated to play significant roles in numerous terrestrial plant-bacteria interactions and was recently shown to be key in a marine association between a diatom and a Roseobacter bacterium ( Amin et al . , 2015; Spaepen et al . , 2007 ) . As many members of the Roseobacter group have multiple metabolic pathways for IAA synthesis ( Moran et al . , 2007 ) , we posited that IAA might be produced by P . inhibens to promote the growth of E . huxleyi in co-culture . To test this , we examined whether IAA is produced by P . inhibens . We were able to detect IAA ( 0 . 4 nM ) in bacterial pure cultures ( see Materials and methods ) ( Figure 4b ) . A standard of IAA was analyzed by high resolution LC-MS and targeted MS/MS ( see Materials and methods ) . The retention time ( Figure 4b ) , exact mass ( Figure 4—figure supplement 1a ) , and fragmentation pattern ( Figure 4—figure supplement 1b ) of the standard matched with the IAA that was detected in bacterial cultures . Importantly , no IAA was detected in axenic algal cultures . Next we assessed algal growth upon addition of IAA . Similar to the increase we observed in co-culture , algal growth yield was improved by 20% upon addition of 1 µM and 100 µM IAA ( Figure 4c ) . Our attempts to monitor IAA levels in co-cultures revealed that in these conditions IAA was undetectable , suggesting rapid up-take by algae . This observation is in agreement with previous reports of significant binding of IAA and rapid signal transduction in plants , indicative of the high affinity towards this compound ( Kepinski and Leyser , 2005 ) . Interestingly , studies have described microbial signaling pathways taking place in the rhizosphere ( the soil immediately surrounding the roots of terrestrial plants ) ( Spaepen et al . , 2007 ) . It is tempting to hypothesize that similar short-circuit processes could take place in the phycosphere ( the immediate volume in proximity to the algal cell ) . Thus , it seems that in our experimental system bacteria inhabiting the algal phycosphere supply their algal host with growth promoting molecules . To further characterize bacterial IAA production in the algal phycosphere , we examined whether similarly to other IAA-producing bacteria , P . inhibens will alter IAA production in response to exogenous tryptophan ( Brandl and Lindow , 1996; Patten and Glick , 2002; Prinsen , 1993; Theunis et al . , 2004; Zimmer et al . , 1998 ) . Indeed , addition of 0 . 1 mM tryptophan resulted in the production of 10 . 8 nM IAA , approximately 25-fold increase in comparison to conditions with no added tryptophan ( Figure 5a ) . The increase in IAA production could be the result of two different mechanisms . Added tryptophan could enhance bacterial metabolism , thus resulting in general increase of bacterially-produced metabolites . Alternatively , exogenous tryptophan could be shunted primarily towards IAA production . To distinguish between these two possibilities , we supplemented a bacterial culture with uniformly labeled tryptophan ( 13C and 15N ) . If the labeled tryptophan were taken up and directly shuttled towards IAA production , all IAA should be fully labeled ( m/z 187 ) ( Figure 5b ) . However , if the imported tryptophan participates in other cellular processes , and atoms are exchanged with endogenous pools of carbon and nitrogen , then the resulting IAA will not be fully labeled , thus resulting in a lower mass . The lowest possible mass would be of a fully unlabeled IAA molecule ( m/z 176 ) . The results of our experiments indicate that all produced IAA is fully labeled ( Figure 5b ) , thus indicating that exogenous tryptophan is directly converted to IAA . Of note , a recent study reported on the production of IAA in the mM range in axenic E . huxleyi cultures supplemented with tryptophan ( Labeeuw et al . , 2016 ) . As all of our experiments and controls indicated no IAA production by algae , currently we do not understand the source of this apparent contradiction . 10 . 7554/eLife . 17473 . 008Figure 5 . Exogenous tryptophan promotes bacterial IAA biosynthesis . ( a ) Addition of 0 . 1 mM tryptophan to P . inhibens cultures results in approximately 25-fold increase in produced IAA . Shown is an LC-MS extracted ion chromatogram ( EIC , m/z 176 . 0706 ± 10 ppm ) of an IAA standard of 10 μM ( black ) , IAA detected in bacterial cultures supplemented with tryptophan ( blue ) , and IAA detected in untreated bacterial cultures ( red ) . ( b ) The addition of isotopically labeled tryptophan ( 13C and 15N ) leads to bacterial production of IAA with full isotopic incorporation , indicated by m/z 187 . 0011 . Inset shows the LC-MS chromatogram of an IAA standard ( black ) ( EIC , m/z 176 . 0706 ± 10 ppm ) and the labeled IAA detected in two biological replicates ( blue ) ( EIC , m/z 187 . 0011 ± 10 ppm ) . ( c ) Cell density ( OD600 ) after 16 hr at 30°C of a tryptophan auxotroph E . coli strain grown in known concentrations of tryptophan and in algal supernatant ( see Materials and methods ) . Dashed line indicates density of the initial inoculum . Error bars represent the standard deviation between four biological replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 17473 . 008 To elucidate the relevance of increased IAA production by bacteria in the presence of exogenous tryptophan , we wanted to examine whether E . huxleyi exudes tryptophan . If indeed tryptophan is exuded by algal cells , bacteria in the phycosphere could import it and convert it to IAA . Secreted bacterial IAA could be utilized by algae and lead to improved algal yields . Such chemical cross talk will feed into a positive algal-bacterial feedback loop . In line with this idea , in another algal-bacterial interaction it has been shown that tryptophan released by a diatom fuels IAA production by a bacterium ( Amin et al . , 2015 ) . To monitor bioavailable tryptophan in E . huxleyi exudates , we used an Escherichia coli strain that is a strict tryptophan auxotroph . This bacterial strain relies solely on exogenous tryptophan; in the presence of extracellular tryptophan the strain will grow and in the absence of tryptophan it will not grow . Cultivation of this strain in minimal medium supplemented with filtered algal supernatant resulted in marked bacterial growth ( Figure 5c ) . This indicates that algal exudates contain bioavailable tryptophan . Interestingly , we detected tryptophan ( 100 nM ) in filtered samples of E . huxleyi blooms ( see Materials and methods ) , suggesting that this metabolite may be relevant to environmental interactions . To better understand the temporal dynamics of the algal-bacterial interaction we examined co-cultures over a period of 17 days and compared them with algal pure cultures of the same age . After 10 days of incubation , no visible difference was apparent between algal pure cultures and co-cultures ( Figure 6a , b ) . However , after 17 days of incubation the color of the co-cultures rapidly changed from green ( the color of healthy algae ) to white , a process we refer to as ‘bleaching’ ( Figure 6c , d and Video 1 ) . To better understand these algal changes we examined the various cultures under the microscope . This revealed that the algal cells in a bleached co-culture exhibit more coccolith shedding ( Figure 6e , f , g , h , note arrow in Figure 6h ) . Furthermore , the typical fluorescent crescent shape that is made of the two chloroplasts in each cell ( Figure 6i , j , k ) is lost in the bleached co-culture and the fluorescent signal emanating from chlorophyll and accessory pigments fills the entire cell ( Figure 6l ) . In addition , using a viability stain ( Sytox green , see Materials and methods ) it is evident that the aging co-culture contains mostly dead algal cells ( Figure 6m , n , o , p ) . One possible explanation for the bleaching observed is that the bacteria simply degrade dying algal cells at day 17 while similarly dying algal cells in the pure culture at the same time point remain intact . However , comparison of the death rates in the algal population in pure and co-cultures reveals a significant difference . At this time point , the vast majority of algal cells in the co-culture are dead ( 94% as indicated by Sytox staining , see Figure 3p ) while in the pure culture only 21% are dead by day 17 . Thus , the presence of bacteria seems to play a key role in promoting algal death . Taken together , these results suggest that following the mutualistic phase in the algal-bacterial interaction , bacteria become pathogens that cause the bleaching and death of their algal partners . 10 . 7554/eLife . 17473 . 009Figure 6 . Bacteria induce a unique algal death in aging co-cultures . ( a–d ) Images of cultures demonstrating the change in the culture color over time . ( e–h ) Phase contrast microscopy images . Arrow points to shed coccoliths . ( i–l ) Fluorescent images of chlorophyll and accessory pigments autofluorescence . ( m–p ) Fluorescent images of dead cells stained with Sytox green . Percentages indicate the number of dead cells counted in each population . For each value n > 300 and the standard deviation between several analyzed fields was up to 20% of the indicated value . ( q–t ) Overlay of phase contrast microscopy images ( grey ) with fluorescent images of TUNEL assay ( green ) of cultures at day 20 ( see Materials and methods ) . ( q ) Co-culture , ( r ) Axenic algal culture , ( s ) Positive control , cells were pretreated with DNase I , ( t ) Negative control , the terminal deoxynucleotidyl transferase enzyme ( TdT ) was replaced with distilled water . Percentages indicate the number of positively stained cells counted in each population . For each value n > 300 and the standard deviation between several analyzed fields was up to 25% of the indicated value . Scale bar corresponds to 1 µm in e–p and 4 µm in q–t . DOI: http://dx . doi . org/10 . 7554/eLife . 17473 . 00910 . 7554/eLife . 17473 . 010Video 1 . Live imaging of the algal bleaching in co-culture . DOI: http://dx . doi . org/10 . 7554/eLife . 17473 . 010 To further explore the death process experienced by algae in co-culture , we examined the expression profile of a select group of algal genes . Out of 50 examined genes representing various functional groups ( Supplementary file 1 ) , 15 genes exhibited significant up-regulation in algae over time in co-culture ( Table 1 ) . To test whether this group of 15 genes exhibits expression levels that are significantly higher than the other 35 genes ( in Supplementary file 1 ) , we first calculated the variance in the two datasets at day 17 using an F-test . The variance was not found to be statistically different and thus a two-tailed Student’s T-test was performed assuming two samples with equal variance . The resulting probability of the T-test was 2 . 8×10−11 indicating that the difference between the two datasets is statistically significant . The highly up-regulated genes encode proteins that are presumed to be involved in various aspects of oxidative stress and programmed cell death ( PCD ) . PCD naturally occurs in an aging algal population and can be triggered by a variety of biotic stresses such as viral infection and abiotic environmental stresses such as nutrient limitation and various light regimes ( Bidle , 2015; 2016 ) . Our expression data suggest that PCD as well as responses to oxidative stress are more prevalent among algal cells in co-culture . Thus , PCD in E . huxleyi seems to be triggered by bacteria . Of note , the up-regulation of genes encoding proteins that are directly involved in the metabolism of reactive oxygen species ( ROS ) , such as glutathione ( Table 1 ) , was previously shown to be essential in viral infection of E . huxleyi ( Sheyn et al . , 2016 ) . To further investigate if algae in co-culture experience a bacterially-induced process similar to PCD , algal cells from axenic cultures and co-cultures were tested for the presence of degraded DNA , a physiological hallmarks of PCD ( Gavrieli et al . , 1992 ) . The fluorescent Terminal deoxynucleotidyl Transferase dUTP nick end-labeling assay ( TUNEL ) was previously established as a reliable method for in situ identification of DNA fragmentation as a PCD indicator in various phytoplankton species ( Johnson et al . , 2014; Segovia et al . , 2003; Berman-Frank et al . , 2004 ) . Using the TUNEL assay ( see Materials and methods ) we could detect that 90% of algal cells in 20-day old co-cultures contain highly fragmented DNA in comparison to 35% in axenic algal cultures of the same age ( Figure 6q–t ) . 10 . 7554/eLife . 17473 . 011Table 1 . Expression data ratios between algae in co-culture and in axenic culture . Select genes involved in stress response and programmed cell death show up-regulation over time . DOI: http://dx . doi . org/10 . 7554/eLife . 17473 . 011Functional groupGene annotationTarget transcriptDay 3Day 10Day 17Oxidative stress putative L-ascorbate peroxidase XM_005793355 1 . 36 3 . 19 6 . 81 Programmed cell death putative programmed cell death protein ( PDCD2 ) XM_005768970 1 . 31 2 . 13 5 . 29 Metacaspases putative metacaspase protein with Ca-binding EF hand domain XM_005784588 1 . 31 3 . 30 4 . 54 Metacaspases putative metacaspase protein XM_005763016 1 . 05 2 . 32 4 . 39 Oxidative stress ascorbate oxidase ( AO ) XM_005775302 1 . 69 2 . 82 4 . 09 Oxidative stress putative glutathione-S-transferase XM_005761417 1 . 31 2 . 20 4 . 07 Metacaspases putative metacaspase protein XM_005791576 1 . 99 2 . 65 4 . 03 Programmed cell death putative death-specific protein with Ca binding EF hand domain XM_005778875 1 . 48 2 . 40 3 . 88 Programmed cell death putative programmed cell death protein ( PDCD2 ) XM_005790372 1 . 21 2 . 12 3 . 84 Metacaspases putative metacaspase protein XM_005773908 1 . 46 2 . 28 3 . 83 Oxidative stress glutathione synthetase ( GSHS3 ) XM_005760150 1 . 32 2 . 14 3 . 56 Oxidative stress putative L-ascorbate peroxidase XM_005784352 1 . 17 2 . 19 3 . 33 Programmed cell death putative death-specific protein with Ca binding EF hand domain XM_005773034 1 . 41 2 . 00 3 . 22 Metacaspases putative metacaspase protein XM_005759676 1 . 17 1 . 90 3 . 06 Oxidative stress thioredoxin XM_005761968 1 . 03 2 . 01 2 . 75 To better understand how bacteria cause algal death in our model system , we explored the possible involvement of known algicidal compounds produced by P . inhibens . A previous study carried out in our laboratories reported on a novel group of compounds – roseobacticides – that are produced by P . inhibens and lyse E . huxleyi ( Seyedsayamdost et al . , 2011 ) . We conducted several experiments to examine whether roseobacticides are the killing agents of algae in aging co-cultures . We successfully reproduced the findings of the previous study ( Figure 7 ) . However , our results indicate that while roseobacticides from P . inhibens indeed kill the non-calcifying algal strain CCMP372 used in the previous study , they do not kill the calcifying algal strain CCMP3266 used in the current study ( Figure 7a ) . Importantly , in CCMP3266 cultures no bleaching was seen upon addition of roseobacticides . Moreover , there is no detectable production of roseobacticides in our co-culture conditions ( Figure 7b ) . Thus , roseobacticides do not appear to play a role in algal death in the co-culture experimental system described here . 10 . 7554/eLife . 17473 . 012Figure 7 . Roseobacticide-mediated algal lysis is strain specific . ( a ) Roseobacticides were introduced into 1 ml of algal cultures of E . huxleyi strains CCMP372 and CCMP3266 . After 12 hr , cell lysis accompanied by chloroplast discharge was evident only in strain CCMP372 ( black arrows ) . White arrow points to coccoliths in the image of the calcifying strain CCMP3266 . Control cultures treated with equivalent volumes of solvent ( methanol ) or medium , did not exhibit any change ( not shown ) . Scale bar corresponds to 1 µm . ( b ) Roseobacticides were extracted from various cultures ( see Materials and methods ) . Roseobacticides were detected in bacterial culture grown in the presence of para-coumaric acid ( ‘Bacteria + pCA’ , blue ) . Inset showing the characteristic absorbance peak of Roseobacticides at 430 nm ( Seyedsayamdost et al . , 2011 ) . No Roseobacticides were detected in a bleached co-culture ( red ) or in control pure cultures of algae ( green ) or bacteria ( purple ) grown without addition of pCA . DOI: http://dx . doi . org/10 . 7554/eLife . 17473 . 012 One small molecule released by bacteria that does appear to have a role in algal death is IAA . At concentrations between 1 and 100 µM IAA had a positive effect on algal growth , however 1000 µM proved to be harmful to algae ( Figure 4c ) . The hormetic effect of IAA has been previously described; IAA promotes plant growth at low concentration while acting as a growth inhibitor at high concentrations ( Persello-Cartieaux et al . , 2001 , 2003; Xie et al . , 1996 ) . Similar effects were observed when increasing concentrations of IAA were added to a culture of diatoms ( Amin et al . , 2015 ) . In addition , there are morphological similarities between algae that were killed by bacteria in co-culture and algae that were killed by high concentrations of IAA in pure culture ( Figure 4d ) . These observations drove us to hypothesize that IAA can mediate both mutualistic and pathogenic interactions between bacteria and algae . Initially , bacterially-produced IAA promotes algal growth while later on it might be the driving force underlying algal death . In various bacteria , attempts to perturb IAA synthesis were unsuccessful due to multiple redundant biosynthetic pathways ( Spaepen et al . , 2007 ) . We examined several mutants of P . inhibens that carry mutations in various pathways of IAA synthesis ( Table 2 and Figure 8a ) . In agreement with previous reports , these mutants were still capable of producing IAA ( Table 2 ) ( Spaepen et al . , 2007 ) and cause algal bleaching . Of note , a mutant in tryptophan biosynthesis ( Table 2 , transposon mutant 1630 ) was unable to grow in pure culture without addition of tryptophan ( data not shown ) but exhibited robust growth in co-culture with algae ( Figure 8b ) and was able to drive algal bleaching . These results further corroborate the presence of tryptophan in algal exudates . 10 . 7554/eLife . 17473 . 013Table 2 . P . inhibens mutant strains used in the current study . DOI: http://dx . doi . org/10 . 7554/eLife . 17473 . 013StrainGenotypeHypothetical functionIAA production in CNPS [nM]IAA production in CNPS + trp [nM]DSM 17395Wild type0 . 410 . 83756PGA1_c11890:: Tn5 ( KanR ) Indoleacetamide hydrolaseYesYes3796PGA1_c11890:: Tn5 ( KanR ) Indoleacetamide hydrolaseYesYes1630PGA1_c16910:: Tn5 ( KanR ) Indole-3-glycerol phosphate synthaseNA*Yes1397PGA1_c23870:: Tn5 ( KanR ) Nitrile hydratase subunit betaYesYes3422PGA1_c31390:: Tn5 ( KanR ) Aromatic-L-amino-acid decarboxylaseYesYes*NA - Not available since this mutant does not grow in these conditions without the addition of tryptophan10 . 7554/eLife . 17473 . 014Figure 8 . Examination of bacterial mutants . ( a ) Mapping the different mutants onto the various bacterial IAA biosynthetic pathways . Mutant strain numbers ( see Table 2 ) are indicated in red next to the pathway that was mutated . Known bacterial enzymes appear in italic . Trp - tryptophan , IAM - indole-3-acetamide , Ipy - indole-3-pyruvate , IAAId - indole-3-acetaldehyde . Image modified from Spaepen et al . ( 2007 ) . ( b ) Mutants in various biosynthetic pathways of indole-3-acetic acid ( IAA ) production were grown in co-culture with algae over a period of 17 days . As can be seen , all mutants exhibited growth dynamics similar to the wild type DSM 17395 strain . Of note , mutant 1630 that is a tryptophan auxotroph ( see Table 2 ) could not grow in pure culture in the absence of tryptophan but was able to grow in co-culture . Error bars represent the standard deviation of two biological replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 17473 . 01410 . 7554/eLife . 17473 . 015Figure 8—figure supplement 1 . Agarose gel of PCR amplifications confirming the transposon insertion site of select sequenced mutants in P . inhibens . Specific primers were designed to amplify the transposon insertion site ( Table 3 ) . As can be seen , amplicons from mutants result in a 2 kb increase in size due to the insertion of a transposon harboring the kanamycin resistance cassette . M- marker , WT- wild type DSM 17395 . DOI: http://dx . doi . org/10 . 7554/eLife . 17473 . 015 To test whether high concentration of bacterially-produced IAA could cause algal death , we wanted to promote the biosynthesis of IAA by bacteria in co-culture . Because P . inhibens exhibited a 25-fold increase in yields of IAA upon utilization of exogenous tryptophan ( Figure 5a and Table 2 ) , we reasoned that an increase in IAA production should occur when a co-culture is supplemented with exogenous tryptophan . Our results indicate , that a co-culture that has been supplemented with 0 . 1 mM tryptophan undergoes accelerated death with bleaching occurring one week earlier than in non-treated co-cultures ( Figure 9a ) . Moreover , examination under the microscope of co-cultures that have been treated with tryptophan revealed a remarkable change in bacterial behavior . In cultures supplemented with tryptophan , bacteria became hyper-colonizers and attached to algal cells in significantly higher rates ( Figure 9b ) . Taken together , our observations indicate that tryptophan serves both as a precursor for IAA biosynthesis and as a cue capable of altering bacterial behavior towards their algal host . 10 . 7554/eLife . 17473 . 016Figure 9 . In bacteria , exogenous tryptophan serves as both a precursor and a cue . ( a ) Images of co-cultures and algal cultures at day 10 . Upon inoculation of bacteria , 0 . 1 mM tryptophan ( trp ) was added . Seen is a co-culture that bleached a week earlier ( middle ) . An algal culture treated with the same concentration of tryptophan did not display bleaching ( right ) . ( b–c ) Phase contrast microscopy imaging of co-cultures at day 10 with ( +trp ) and without ( −trp ) addition of 0 . 1 mM tryptophan . Scale bar corresponds to 3 μm . Note that in the tryptophan treated co-culture ( +trp ) in ( c ) each algal cell has attached bacteria and the bacterial coverage is so dense that several covered algal cells cannot be seen . DOI: http://dx . doi . org/10 . 7554/eLife . 17473 . 016 The process in which algal biomass gradually increases to form a bloom , covering vast areas and then abruptly collapses , has attracted much attention from various disciplines . It is known that viral infection is key in the blooms’ demise ( Bratbak et al . , 1993; Lehahn et al . , 2014; Vardi et al . , 2012 ) . In our model system the presence of one species of bacteria resulted in the sharp demise of the algal population . While further studies are required in order to explore E . huxleyi-bacterial interactions in the ocean , it is possible that bacterial influences act in concert with viruses to drive the termination of the blooms . Environmental stresses such as nutrient depletion as well as viral infection have been shown to enhance PCD in E . huxleyi ( Bidle , 2015 , 2016 ) . Similarly , we have observed bacterially-induced PCD . While attached bacteria seem to be trapped on their dying host ( Figure 10 ) , their offspring get access to the nutrients from lysed algae and can then swim away and colonize a younger algal cell . Recently , Smirga and colleagues demonstrated experimentally how planktonic bacteria crowd around a lysing microalga , feeding off the released cellular content ( Smriga et al . , 2016 ) . Our discovery of chemical crosstalk in the phycosphere and the bacterially-mediated algal death were made possible due to the co-culture experimental system that enables investigation over time . 10 . 7554/eLife . 17473 . 017Figure 10 . Bacteria attached to remains of dead algal cells . ( a–c ) SEM images of co-cultures at day 17 of incubation reveal bacteria that are attached to algal debris . The vast majority of imaged algal cells were intact ( see for example the lower cell in panel c ) indicating that algal debris were not generated as a result of sample preparation . Arrow pointing at a bacterial cell attached to algal remains that are adjacent to an intact algal cell . Scale bar corresponds to 1 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 17473 . 017 In the current study tryptophan was identified as a central compound in the E . huxleyi – P . inhibens interaction . This led us to explore its abundance in E . huxleyi natural blooms , where it indeed was found . Previously , tryptophan was identified as a key metabolite in the interaction between a diatom and a Roseobacter ( Amin et al . , 2015 ) . The use of both targeted and untargeted metabolomics in the lab can greatly aid in identifying metabolites with ecological relevance ( Fiore et al . , 2015; Johnson et al . , 2016; Durham et al . , 2015; Amin et al . , 2015 ) . As revealed in the current study , when microorganisms are in close proximity , molecules can be produced and rapidly consumed and thus remain undetectable . Local and short-lived signals that are significant cues in the phycosphere would be undetectable in the environment but can be deciphered using model systems similar to the one we have described . In our study , IAA was identified as a key component of the algal-bacterial chemical crosstalk . The bacterially-produced IAA initially increases algal yields . This growth enhancement is followed by an inevitable death . Clearly IAA exhibits a hormetic effect - it is beneficial in low concentrations and becomes harmful at higher doses . The hormetic nature of IAA has been previously demonstrated in various plant-bacteria interactions ( Persello-Cartieaux et al . , 2001 , 2003; Xie et al . , 1996 ) . In the association of the plant pathogen Agrobacterium tumefaciens with its hosts , IAA is central too ( Spaepen et al . , 2007; Subramoni et al . , 2014 ) . It is produced to enhance growth of plant tissue while creating tumors . Several similarities between A . tumefaciens and P . inhibens seem to exist; both attach through their pole to their host ( Heindl et al . , 2014; Xu et al . , 2013 ) , both express a polar polysaccharide involved in bacterial attachment ( Heindl et al . , 2014; Xu et al . , 2013 ) and both utilize IAA to manipulate the growth of their host ( Subramoni et al . , 2014 ) . In light of these resemblances , it is possible that additional aspects of the E . huxleyi – P . inhibens interaction will be similar to the mechanisms employed by A . tumefaciens and its interactions with host plants . We have shown that IAA produced by bacteria enhances algal growth . How is algal growth promoted ? One possibility is that algal growth augmentation is the result of changes in light harvesting efficiency . As reported by Falkowski and colleagues , changes in the capability of transducing light energy to chemical energy and eventually to biomass will result in changes in growth ( Falkowski et al . , 1985 ) . Since all our cultures have been cultivated under the same light regime , changes in light harvesting capabilities would be the result of changes in levels of light harvesting pigments . However , chlorophyll a measurements revealed similar concentrations in algal cells over a period of 17 days , whether in the presence or absence of bacteria ( Table 3 ) . Only after bleaching , when the majority of the algal population was dead , a marked decrease in chlorophyll a concentration was evident in agreement with the macroscopic phenotype of bleaching ( Figure 6d ) and microscopic phenotype of chloroplast deformation ( Figure 6l ) that are seen in the co-culture at the same time . Thus , enhanced algal growth does not seem to be the result of more efficient light harvesting . The mechanisms underlying increased algal growth in response to IAA remain unknown . 10 . 7554/eLife . 17473 . 018Table 3 . Chlorophyll a measurements in algal cultures . Chlorophyll a was measured spectroscopically in pure and co-cultures over time ( see Materials and methods ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17473 . 018Sample IDChlorophyll a per cell [pg]Algae day 35 . 96×10−3Co-culture day 36 . 18×10−3Algae day 101 . 43×10−2Co-culture day 101 . 67×10−2Algae day 171 . 50×10−2Co-culture day 176 . 55×10−3 IAA-induced growth in our system could represent the loss of growth control by E . huxleyi . In various microorganisms , regulatory circuits lead to growth cessation before resources have been completely exhausted . When such controls are bypassed , growth does not stop and higher yields are obtained . Two examples can be discussed in the context of elevated growth as a consequence of loss of growth control; E . coli mutants in rpoS grow to higher cell numbers than the wild type ( Vulic and Kolter , 2001 ) . While the wild type senses the near depletion of essential nutrients and prepares in advance by ceasing growth , the mutant lacks this regulatory ability ( Vulic and Kolter , 2001 ) . Another example of growth regulation by microorganisms is the ability of cells to sense their critical density , a process referred to as quorum sensing ( QS ) . This process requires the release and sensing of signal molecules that serve as reporters of the population density . The response to increased cell numbers is often a transition to decreased growth , which promotes survival of the cells under conditions of limited resources . In trypanosomes , perturbed ability to sense signal molecules results in failure to undergo growth arrest and leads to uncontrolled proliferatation ( Mony and Matthews , 2015 ) . Whether E . huxleyi possesses regulatory circuits to control growth cessation , and whether IAA can influence such regulation , is yet to be explored . Bacteria in our co-cultures were found attached solely to naked algal cells . We did not detect bacteria attached to calcified algae or shed coccoliths . What underlies the specific bacterial attachment to naked algal cells ? Roseobacters posses the ability to chemotax ( Miller and Belas , 2006; Miller et al . , 2004 ) , thus it is likely that P . inhibens swims in its planktonic stage specifically towards the naked cells . In turn , this might suggest that naked cells release increased levels of exudates . However , the question still remains whether the algal cells are naked irrespective of bacteria , or whether bacteria encourage coccoliths shedding thus increasing the amount of naked algal cells . In every population of calcified E . huxleyi , a sub-population of naked non-calcified cells is seen . It has been previously demonstrated that microzooplankton display higher growth rates when feeding on naked E . huxleyi cells in comparison to calcified cells ( Harvey et al . , 2015 ) . Thus , bacteria might initially target a pre-existing algal sub-population . Reversible modes of attachment were shown to take place when bacteria such as A . tumefaciens are assessing whether or not to attach to a substrate in a new environment ( Heindl et al . , 2014 ) . As P . inhibens shares several similarities with this bacterium regarding their mode of attachment ( Segev et al . , 2015 ) , it is possible that P . inhibens evaluates the algal cells through reversible interactions . In this regard , the physical properties of a naked non-calcified algal cell might be more suitable for bacterial attachment . Similar influences might promote the permanent attachment that persists even after the algal cell has died ( Figure 10 ) . In this study we have established a well-defined microbial model system and carried out a detailed characterization of micro-scale interactions over time . The discovery of a dynamic microbial interaction and the mechanisms underlying it was made possible due to the use of a simplified experimental system for coccolithophore-bacteria interactions . Exploration of similar robust model systems could further reveal chemical and molecular details that might be studied in the ocean . Attempts to link laboratory findings and environmental studies will both widen and deepen our understanding of microbial interactions , their ecophysiology and the extent to which these interactions influence the marine environment .
The bacterial strain of Phaeobacter inhibens was DSM 17395 purchased from the German collection of microorganisms and cell cultures ( DSMZ , Braunschweig , Germany ) . Bacterial cultures were grown in liquid 1/2YTSS medium containing 2 g yeast extract ( BD , NJ , USA ) , 1 . 25 g tryptone ( Sigma-Aldrich , MO , USA ) and 20 g sea salt ( Sigma-Aldrich ) per liter . Cultures were incubated at 30°C shaking at 130 rpm . The axenic algal strain of Emiliania huxleyi was CCMP3266 purchased from the National Center for Marine Algae and Microbiota ( Bigelow Laboratory for Ocean Sciences , Maine , USA ) . Algae were grown in L1 medium according to ( Guillard and Hargraves , 1993 ) , with the exception that Na2SiO3 was omitted following the cultivation recommendations for this algal strain , and the medium was referred to as L1-Si . Algae were grown in standing cultures in a Percival chamber ( Percival Scientific , IA , USA ) at 18°C under a light/dark cycle of 12/12 hr . Illumination intensity during the light period was 150 µmoles/m2/s . Cultures were maintained axenic using a mixture of penicillin and streptomycin at a final concentration of 0 . 1 mg/ml and 0 . 05 mg/ml respectively . The antibiotic-treated cultures were used to inoculate antibiotic-free cultures . At least two passages through antibiotic-free medium were carried out prior to inoculation of the experimental cultures . Absence of bacteria in axenic algal cultures was monitored periodically both by plating on 1/2YTSS plates and under the microscope . Co-cultures of E . huxleyi and P . inhibens were cultured as follows: 14-day-old E . huxleyi cultures were inoculated into L1-Si medium at a 1:100 dilution and incubated as described above . After four days of algal growth , a colony of P . inhibens was resuspended in 1 ml L1-Si and 100 µl were added to 10 ml of algal culture . The co-cultures were incubated in a Percival chamber under the conditions described above for algal cultures . In order to evaluate bacterial growth in co-cultures , samples were taken from co-cultures at different time points , as indicated . Samples were serially diluted and plated on 1/2YTSS plates . These plates facilitate only bacterial growth , and thus counts of colony forming units ( CFU ) were indicative of bacterial numbers in the co-culture . In order to eliminate rosettes and obtain CFU from single cells , we attempted sonication prior to plating , however we were unable to eliminate rosettes and sonication did not alter the CFUs that were obtained . Thus , samples were not sonicated and CFU may include rosettes as well as individual bacteria . Samples of 10 μl were placed on 1 cm2 cover slips coated with poly-L-lysine and kept in a humid environment at room temperature for 1 hr . Samples were then submerged sequentially for 10 min in 5 ethanol solutions of increasing concentrations from 30% to 100% absolute ethanol . Following critical point drying the samples were sputter coated with an Au/Pd alloy . Images were obtained with a Zeiss Supra55 Field Emission scanning electron microscope . Fluorescence and phase contrast images were obtained using a Nikon TE-2000U inverted microscope equipped with a 100× Plan Apo NA 1 . 4 objective lens . All samples were spotted on thin 1% agarose pads for visualization at room temperature . Images were acquired using a cooled Hamamatsu CCD camera controlled with MetaMorph seven software ( Molecular Devices , CA , USA ) . Algal autofluorescence was visualized using a Cy5 long-pass filter ( Chroma #41024 ) . Alexa Fluor 488 conjugated WGA , Sytox green and TUNEL signals ( see below ) were captured using a narrow band eGFP filter ( Chroma #41020 ) . Phase contrast images where green chlorophyll is shown ( Figure 7 ) were obtained with a Zeiss Axioskop two plus microscope equipped with a Zeiss Axiocam MRc camera and using a Zeiss plan-Apochromat 63x lens with a 1 . 4 NA . Images were analyzed using the MetaMorph seven software . Images were processed identically for compared image sets . Phase contrast images of co-cultures at the specified ages were obtained and bacterial cells were manually counted using the MetaMorph seven software . In the case of algal clumps , the total number of attached bacteria was divided by the number of algae in the clump . Given the three dimensionality of cells and clumps , bacterial numbers are most likely underestimated , as bacteria that attach under the observed specimen cannot be seen . For all time points n > 300 . To assess the identity of essential nutrients provided by algae to support bacterial growth , bacteria were grown in L1-Si medium ( in which they cannot grow ) and essential nutrients were externally added . All nutrients were added in forms that were previously shown to be utilized by P . inhibens ( Zech et al . , 2009 ) ; glucose 5 . 5 mM , Na2SO433 mM , NH4Cl 5 mM , KH2PO42 mM , all purchased from Sigma-Aldrich . The sulfur source was replaced with 30 µM or 6 mM dimethylsulfoniopropionate ( DMSP , Research Plus , NJ , USA ) where noted . Cultures were grown as previously described . Prior to analysis , 250 µl of each culture were supplemented with the viability dye Sytox blue as mentioned above ( Life Technologies ) and 50 µl of CountBright fluorescent beads ( Life Technologies ) . The fluorescent beads were used to calculate cell numbers according to the manufacturer's instructions . Samples were analyzed on an 18-channel FACSAria SORP flow cytometer ( BD Biosciences , MA , USA ) . For each sample 100 , 000 events were recorded . Data analysis was carried out using the DIVA software version 6 . 3 . 1 ( BD Biosciences ) . In order to validate the accurate interpretation of particles in the analyzed samples , select samples were additionally analyzed using the ImageStream 100 multispectral imaging flow cytometry instrument ( Amnis , WA , USA ) equipped with 405 , 488 and 658 nm laser sources with variable laser power , and a brightfield source . Prior to analysis , samples were transferred to 500 µl siliconized microcentrifuge tubes ( Sigma-Aldrich ) . Data of boiled and/or stained control algae were collected and used to compensate fluorescence spectral overlap between the different fluorescent channels , and set the optimal laser power to avoid saturation of the camera . Samples were gated for single cell population using the area and aspect ratio features as previously described ( Ponomarev et al . , 2011 ) . Analysis was carried out while minimizing the 658 nm laser power to decrease bleed through of chlorophyll and accessory pigments autofluorescence . Data files contained 10 , 000–20 , 000 cells and were analyzed using the Image Data Exploration and Analysis Software ( IDEAS ) ( Amnis ) . E . coli strain CGSC#6666 was obtained from the Coli Genetic Stock Center ( Yale University , New Haven , CT ) . The strain was grown overnight in M9 medium supplemented with 0 . 4% glucose and 10 mM tryptophan . In the following morning , cells were centrifuged , washed twice with M9 + 0 . 4% glucose and diluted to OD600 = 0 . 05 into a 96-well plate . Each well contained 100 μl culture and 100 μl L1-Si or algal supernatant where specified . Treatments included no addition of tryptophan , addition of tryptophan ( 30 or 300 μM ) , or addition of algal supernatant filtered from a 14-day old algal culture . Each treatment was tested in four replicates . The plate was incubated at 30°C in a SPECTRA max M2 plate reader ( Molecular Devices ) . Every 30 min the plate was shaken for 5 s and OD600 was measured , over a course of 16 hr . Preparation of electrocompetent cells from P . inhibens DSM 17395 was conducted as previously described ( Petersen et al . , 2011 ) . Transposon mutagenesis in P . inhibens DSM 17395 was performed with the EZ-Tn5 <R6Kγori/KAN-2>Tnp Transposome kit ( Epicentre , Illumina , CA , USA ) . Individual transposon mutants were cultured in MB medium ( BD ) with 120 μg ml−1 kanamycin . Total DNA was isolated with the DNeasy Blood and Tissue Kit ( Qiagen ) and the insertion sites of 4000 transposon mutants were determined via arbitrary PCR as previously described ( O'Toole and Kolter , 1998 ) . Select mutants ( Table 2 ) were streaked out three subsequent times on MB plates containing 120 μg ml−1 kanamycin in order to eliminate wild type cells that could survive in rosette structures . Subsequently , the precise integration site of each mutant strain was validated via PCR amplification with specific primers ( Supplementary file 2 and Figure 8—figure supplement 1 ) followed by sequencing . 500 ml Erlenmeyer flasks containing 50 ml of 1/2YTSS medium and 1 mM para-coumaric acid ( Sigma-Aldrich ) were inoculated with 0 . 5 ml of an overnight culture and incubated at 130 rpm at 30°C for three days . Then , cultures were extracted once with an equal volume of ethyl acetate , dried in vacuo , resuspended in methanol and analyzed by HPLC . HPLC analysis was performed on a Beckman Coulter System Gold HPLC equipped with a diode array detector using an analytical Phenomenex C18 column ( 5 μm , 4 . 6 mm x 100 mm ) . Flow rate was 0 . 7 ml/min with a gradient of 10% acetonitrile in water to 100% acetonitrile over 25 min . Roseobacticides were identified according to their characteristic absorbance peak at 430 nm ( Seyedsayamdost et al . , 2011 ) . Cultures of E . huxleyi strains CCMP3266 and CCMP372 were grown seven days in L1-Si medium and then diluted with fresh medium 1:1 ( v/v ) . Cultures were placed in 1 ml aliquots in a 48 well plate . Each well was supplemented with 10 µl of roseobacticide extract ( as described above ) , methanol or medium . Samples were visualized under the microscope 12 and 24 hr after treatment . Four samples of 1 liter were collected in two E . huxleyi blooms ( see section above for details about the sampling area ) . Samples were filtered through a 0 . 2 μm aPES membrane ( Thermo Fisher Scientific ) and filtrates were collected in sterile bottles and stored at 4°C in the dark . Filtrates were acidified with formic acid to pH = 3 . Acidified samples were loaded onto Oasis HLB cartridges ( Waters ) and eluted with methanol . Eluted samples were dried in vacuo and resuspended in 200 μl methanol . Samples were filtered prior to high resolution LC-MS analysis , which was preformed as described above . For Chlorophyll extraction , at the specified time points 5 ml of cultures were collected on ice and centrifuged for 20 min at 10 , 000 rpm at 4°C . Supernatant was discarded and pellets were resuspended in 1 ml of 90% acetone . Pellets were disturbed using vigorous pipettation and vortexing for 2 min . Samples were kept overnight at 4°C . Prior to Chlorophyll a measurements , samples were filtered through a 0 . 2 μm PVDF membrane ( Pall ) directly into cuvettes . Filtrate absorbance was measured on a Beckman DU640 spectrophotometer ( Beckman Coulter , CA , USA ) at the following wavelengths: 750 , 663 , 645 and 630 nm . Absorbance at 750 nm was subtracted from all other values to correct for turbidity . Chlorophyll a concentration was calculated using the following equation: Chlorophyll a [μg/L] = {[11 . 64 ( Abs663 ) – 2 . 16 ( Abs645 ) + 0 . 10 ( Abs630 ) ] E} / V ( L ) Where E = the volume of acetone solution used for extraction ( ml ) V = the volume of sample filtered ( L ) L = the cuvette path length ( cm ) The obtained values were divided by cell numbers determined using flow cytometry as described above . | Microscopic algae that live in the ocean release countless tons of oxygen into the atmosphere each year . Widespread algae – known as coccolithophores – surround their little plant-like body with a mineral shell made of a material similar to chalk . These microscopic algae form seasonal blooms . Over several weeks in early summer , the algae grow to enormous numbers and cover hundreds of thousands of square kilometers in the ocean . These blooms become so vast that satellites can detect them . However , suddenly the blooms collapse; the algae die and their chalky shells sink to the bottom of the ocean where they have been accumulating for millions of years . More and more evidence suggests that these tiny algae interact with bacteria in various ways . However , so far , no one had documented a direct interaction between bacteria and a member of this key group of algae . Now , in a controlled laboratory environment , Segev et al . show that marine bacteria from the Roseobacter group physically attach onto a tiny coccolithophore alga called Emiliania huxleyi . While the bacteria are attached to their algal host , they enjoy a supply of nutrients that trickles from the algal cell . Unexpectedly , Segev et al . also discovered that the algae grow better in the presence of the bacteria . It turns out that the bacteria use a molecule that they obtain from their algal hosts to produce a small hormone-like molecule that in turn enhances the growth of the algae . However , after three weeks of growing together , the bacteria produce so much of the growth-enhancing molecule – which is harmful in higher concentrations – that they actually kill their algal host . These findings suggest that the bacteria first promote the alga’s growth to boost their supply of nutrients . But as algae grow older , the bacteria harvest the algae to enjoy a last pulse of nutrients and allow their offspring to swim away and attach to younger algae . The next challenge will be to link these laboratory observations to the actual microbial interactions in the ocean . It will also be important to explore whether other algae and bacteria interact in similar ways and if bacteria contribute to the sudden collapse of algal blooms by killing the algae . | [
"Abstract",
"Introduction",
"Results",
"and",
"discussion",
"Materials",
"and",
"methods"
] | [
"ecology"
] | 2016 | Dynamic metabolic exchange governs a marine algal-bacterial interaction |
Birds display remarkable diversity in the distribution and morphology of scales and feathers on their feet , yet the genetic and developmental mechanisms governing this diversity remain unknown . Domestic pigeons have striking variation in foot feathering within a single species , providing a tractable model to investigate the molecular basis of skin appendage differences . We found that feathered feet in pigeons result from a partial transformation from hindlimb to forelimb identity mediated by cis-regulatory changes in the genes encoding the hindlimb-specific transcription factor Pitx1 and forelimb-specific transcription factor Tbx5 . We also found that ectopic expression of Tbx5 is associated with foot feathers in chickens , suggesting similar molecular pathways underlie phenotypic convergence between these two species . These results show how changes in expression of regional patterning genes can generate localized changes in organ fate and morphology , and provide viable molecular mechanisms for diversity in hindlimb scale and feather distribution .
In birds , the genetic and developmental mechanisms that control the decision between scale and feather development remain poorly understood . Most birds possess scales on the foot ( tarsometatarsus and toes ) and feathers elsewhere . Exceptions to this pattern can provide insights into the evolutionary and developmental basis of skin appendage diversity . Some raptors and boreal birds evolved foot feathers instead of scales ( 'ptilopody'; Danforth , 1919; Lucas and Stettenheim , 1972 ) , but the lack of appendage variation within these species precludes their use as genetic models . Likewise , paravians ( birds and their close non-avian theropod dinosaur relatives ) and other dinosaurs evolved diverse feather coverings on their legs and feet that sometimes resemble flight-like feathers ( Xu et al . , 2003; Hu et al . , 2009; Turner et al . , 2012; Zheng et al . , 2013; Foth et al . , 2014; Godefroit et al . , 2014 ) , but the absence of living specimens preclude mechanistic molecular studies . In contrast , domestic pigeons ( Columba livia ) exhibit stunning variation within a single extant species ( Shapiro and Domyan , 2013 ) . Most breeds have feet covered by scaled epidermis ( wild-type ) , but scales are replaced by small feathers in grouse ( gr ) mutants , and by larger feather 'muffs' in birds that also carry mutant alleles at the Slipper ( Sl ) locus ( Doncaster , 1912; Wexelsen , 1934; Hollander , 1937; Levi , 1986 ) ( Figure 1A ) . In muffed breeds , scutellate scales are generally absent or poorly developed on the feathered epidermis covering the tarsometatarsus and toes , and feathers are surrounded by soft integument . The molecular identities of both gr and Sl are unknown , and additional loci probably control quantitative variation in the muff phenotype . Because both scale-footed and feather-footed pigeon breeds belong to the same species , we can use traditional genetic crosses and whole-genome resequencing to map the genes that control this striking variation ( Shapiro et al . , 2013; Domyan et al . , 2014 ) . Therefore , we can study diversity of the magnitude usually observed among different species without the roadblock of hybrid incompatibility that often eliminates the possibility of genetic mapping studies . 10 . 7554/eLife . 12115 . 003Figure 1 . Two QTL differentiate scale- and feather-footed domestic pigeons . ( A ) Common phenotypes of domestic rock pigeon , in order of increasing foot feathering ( left to right ) : scaled , groused , small- and large-muffed . ( B-F ) QTL scans and effect plots: proportion of tarsus feathered ( B , C ) , number of toe feathers ( D , E ) , and length of toe feathers ( F , G ) . Mean phenotypes ± S . E . are plotted in ( C , E , G ) . For ( E ) and ( G ) , genotypes at the QTL with the higher LOD score are on the x-axis , and genotypes at the other QTL are inset . See Tables 1 and 2 for detailed QTL statistics . S , allele from scale-footed grandparent; F , allele from feather-footed grandparent . DOI: http://dx . doi . org/10 . 7554/eLife . 12115 . 00310 . 7554/eLife . 12115 . 004Figure 1—figure supplement 1 . Foot-feathering phenotypes of genetic cross . ( A ) Founder Pomeranian pouter male ( muffed ) and Scandaroon female ( scale-footed ) . ( B ) Representative F1 individual with moderate foot feathering . ( C ) Representative sample set of F2 individuals . DOI: http://dx . doi . org/10 . 7554/eLife . 12115 . 004 During development in vertebrates , skin appendages form through interactions between the ectoderm-derived epidermis and the mesoderm-derived dermis , and signals from the dermis determine epidermal appendage fate ( Dhouailly , 2009; Hughes et al . , 2011 ) . Previous analyses of mutants and gene misexpression in chickens suggest candidates for feathered feet in the Hedgehog ( Harris et al . , 2002 ) , BMP ( Zou and Niswander , 1996; Harris et al . , 2002; 2004 ) , Delta-Notch ( Crowe et al . , 1998 ) , and Wnt ( Chang et al . , 2004 ) pathways . Our study of genetic variation and embryonic development in pigeons , however , reveals a surprisingly different mechanism with broad implications for limb identity and patterning .
To identify chromosome regions that contribute to feathered feet , we generated an F2 intercross between muffed ( Pomeranian pouter ) and scaled ( Scandaroon ) breeds ( Figure 1—figure supplement 1 ) . F1 hybrids displayed intermediate foot feathering , demonstrating a semi-dominant inheritance pattern . Scaled , muffed , and intermediate phenotypes were recovered in the F2 population , confirming that a small number of major-effect loci contribute to this trait . Among F2 offspring , digit 3 bore the largest and greatest number of feathers ( digit 1 , 7 . 04 ± 5 . 08 feathers; digit 2 , 7 . 64 ± 7 . 54; digit 3 , 15 . 46 ± 10 . 43; digit 4 , 6 . 13 ± 5 . 70 ) . Using quantitative trait locus ( QTL ) mapping with 130 F2 offspring genotyped at 3803 polymorphic markers ( Broman et al . , 2003 ) , we identified two linkage groups ( LG11 and LG20 ) that had significant effects on three different aspects of foot feathering ( log10 odds ratio ( LOD ) > 4 . 6; Figure 1B–G , Table 1 ) . LG11 had the largest effects on the proportion of the tarsometatarsal epidermis that was transformed from scaled to feathered ( LOD = 7 . 69 , percent variance explained ( PVE ) = 28 . 4% ) and toe feather number ( LOD = 6 . 72 , PVE = 21 . 3% ) , and a smaller effect on toe feather length ( LOD = 8 . 51 , PVE = 15 . 8% ) . LG20 had the largest effect on toe feather length ( LOD = 20 . 9 , PVE = 52 . 2% ) , and a smaller effect on toe feather number ( LOD = 5 . 36 , PVE = 16 . 5% ) . When toe feather number was analyzed for each digit individually , the same two QTL were identified and had the most pronounced effects on digits 3 and 4 ( Table 2 ) . In summary , two major QTL have marked and separable effects on qualitative and quantitative variation in epidermal appendages . 10 . 7554/eLife . 12115 . 005Table 1 . Summary of foot feathering QTL . DOI: http://dx . doi . org/10 . 7554/eLife . 12115 . 005TraitLGLoc ( cM ) ScaffoldPositionLODPVEMean ± S . D . SSSFFFProportion tarsus feathered11117799 , 205 , 2867 . 6928 . 40 . 46 ± 0 . 040 . 58 ± 0 . 03**0 . 80 ± 0 . 04***Number of toe feathers111247912 , 325 , 9776 . 7321 . 343 . 6 ± 8 . 967 . 3 ± 6 . 5*105 . 3 ± 8 . 1***Number of toe feathers2015951 , 451 , 1275 . 3616 . 545 . 3 ± 8 . 178 . 2 ± 6 . 5**100 . 3 ± 9 . 9***Toe feather length ( mm ) 20070136 , 74620 . 8952 . 25 . 2 ± 2 . 319 . 9 ± 2 . 0***37 . 3 ± 2 . 5***Toe feather length ( mm ) 111247912 , 325 , 9778 . 5115 . 811 . 4 ± 3 . 318 . 7 ± 2 . 4*28 . 5 ± 3 . 0**LG , linkage group; Loc , genetic location of peak LOD score in centimorgans; S , allele from scaled parent; PVE , percent variance explained; F , allele from feathered parent; LOD , log10 odds ratio . ( Welch two sample t-test of means compared to SS homozygote; *p≤ 0 . 05 , **p≤0 . 005 , ***p≤0 . 0005 . ) 10 . 7554/eLife . 12115 . 006Table 2 . Summary of QTL for numbers of feathers on individual toes . DOI: http://dx . doi . org/10 . 7554/eLife . 12115 . 006DigitLGLoc ( cM ) ScaffoldPositionLODPVEMean ± S . D . SSSFFFDigit 2 , left foot111487911 , 624 , 7015 . 2020 . 243 . 75 ± 1 . 136 . 95 ± 1 . 12*12 . 19 ± 1 . 16***Digit 3 , right foot20070136 , 7467 . 7124 . 19 . 06 ± 1 . 7017 . 35 ± 1 . 48**19 . 86 ± 1 . 83***Digit 3 , right foot111487911 , 624 , 7016 . 4719 . 59 . 98 ± 1 . 6015 . 60 ± 1 . 62*21 . 02 ± 1 . 65***Digit 3 , left foot20070136 , 74610 . 7929 . 848 . 66 ± 1 . 5917 . 06 ± 1 . 41**21 . 09 ± 1 . 86***Digit 3 , left foot111487911 , 624 , 7019 . 4825 . 529 . 37 ± 1 . 5216 . 26 ± 1 . 52**21 . 30 ± 1 . 57***Digit 4 , right foot2032952 , 464 , 7887 . 3321 . 362 . 44 ± 0 . 907 . 24 ± 0 . 75**10 . 14 ± 1 . 10***Digit 4 , right foot11118795 , 475 , 4746 . 7819 . 53 . 17 ± 0 . 896 . 91 ± 0 . 92*9 . 95 ± 1 . 00***Digit 4 , left foot2030952 , 464 , 7889 . 3726 . 832 . 44 ± 0 . 796 . 63 ± 0 . 66**9 . 65 ± 0 . 96***Digit 4 , left foot111487911 , 624 , 7017 . 5820 . 833 . 75 ± 1 . 146 . 95 ± 1 . 13*12 . 19 ± 1 . 17***LG , linkage group; Loc , genetic location of peak LOD score in centimorgans; S , allele from scaled parent; PVE , percent variance explained; F , allele from feathered parent; LOD , log10 odds ratio . ( Welch two sample t-test of means compared to SS homozygote; *p≤0 . 05 , **p≤0 . 005 , ***p≤0 . 0005 . ) In parallel to our experimental cross , we performed probabilistic whole-genome scans of allele-frequency differentiation ( pFst; see Kronenberg et al . , 2014 ) across a genetically and phenotypically diverse panel of breeds by comparing 15 feather-footed birds ( 4 groused and 11 muffed ) to 28 scale-footed birds ( Shapiro et al . , 2013 ) . Using this independent approach across breeds , the two most-highly differentiated pFst signals implicate the same genomic regions as the QTL study: genomic scaffold 79 is located on LG11 ( p=4 . 44 x 10–16 , genome-wide significance threshold = 2 . 11 x 10–9 ) , and scaffolds 70 and 95 are adjacent to one another on LG20 ( p=9 . 81 x 10–13 ) ( Figure 2A , Figure 2—figure supplement 1A , B ) 10 . 7554/eLife . 12115 . 007Figure 2 . Two regions of genomic differentiation and H3K27ac enrichment distinguish scale- and feather-footed pigeons . ( A ) Whole-genome pFst comparisons between genomes of feather-footed and scale-footed pigeons . Scaffolds are ordered by genetic position in a linkage map from an F2 cross ( see Figure 1 ) . Dashed line , genome-wide significance threshold . ( B ) pFst and extended haplotype homozygosity ( EHH ) plots for region of high differentiation on scaffold 79 . Feather-footed birds ( n=10 , red in EHH plot ) homozygous for a 44-kb deletion are differentiated from scale-footed birds ( n=28 , black ) and show extended haplotype homozygosity in this region . Smoothed lines follow a generalized additive model ( Wickham , 2009 ) . ( C ) H3K27ac ChIP-seq enrichment differed significantly between embryonic wing and leg buds of the scale-footed racing homer ( RH ) in several regions ( blue shading ) , including within a 44-kb interval that is deleted in the muffed Indian fantail ( IF; blue arrowheads ) . This deleted region is orthologous to a known human limb enhancer ( hs1473 ) . ( D ) Selection scans show similar patterns of differentiation on scaffold 70 between muffed ( n=11 , red in EHH plot ) and scale-footed birds ( n=28 , black ) . ( E ) H3K27ac ChIP-seq enrichment differed significantly between leg buds of racing homer and Indian fantail embryos in regions immediately 5’ of Tbx5 ( blue shading ) . Foot and wing drawings modified after Levi ( 1986 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12115 . 00710 . 7554/eLife . 12115 . 008Figure 2—figure supplement 1 . Synteny and genomic differentiation of pigeon LG20 . ( A ) Lastz alignment of pigeon scaffolds corresponding to linkage group 20 with chicken chromosome 15 . ( B ) pFst scan of scaffolds from linkage group 20 , ordered based on genetic linkage and synteny with chicken chromosome 15 . green = scaffold 70 , aqua = scaffold 95 , blue = scaffold 737 , black = scaffold 143 , red = scaffold 20 . Red line = whole genome significance threshold ( 2 . 11 x 10–9 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12115 . 00810 . 7554/eLife . 12115 . 009Figure 2—figure supplement 2 . Genomic association scans . ( A ) pFst scans on scaffold 79 showing all feather-footed birds vs . all scale-footed birds , feather-footed birds homozygous for deletion vs . scale-footed birds , feather-footed birds lacking the deletion ( alt ) vs . scale-footed birds , and feather-footed birds lacking the deletion ( alt ) vs . feather-footed birds homozygous for the deletion . ( B ) Histogram of genotypes for scaffold 79 deletion , scale-footed vs . feather-footed phenotype . ( C ) pFst scans on scaffold 70 showing all feather-footed birds vs . scale-footed birds , muff birds vs . scale-footed birds , muff vs . grouse birds , and grouse vs . scale-footed birds . Range of –log10 ( pFst ) values on left side of each plot . DOI: http://dx . doi . org/10 . 7554/eLife . 12115 . 00910 . 7554/eLife . 12115 . 010Figure 2—figure supplement 3 . Haplotype diagram of scaffold 70 candidate interval . Haplotypes clustered based on binary distance from muff ( red ) , grouse ( blue ) , and scale-footed ( black ) birds . DOI: http://dx . doi . org/10 . 7554/eLife . 12115 . 010 The peak pFst region on scaffold 79 contained a 44-kb deletion ( from 6 . 719–6 . 763 Mb ) that was homozygous in 10 , and heterozygous in 2 of the 15 feather-footed birds ( Figure 2B , C; Figure 2—figure supplement 2A ) . Birds homozygous for the deletion showed elevated levels of haplotype homozygosity relative to scaled birds , a signature of positive selection on this region ( Figure 2B ) . This deletion spans an element orthologous to a known human limb enhancer , hs1473 ( Spielmann et al . , 2012 ) , which contains active chromatin marks ( Cotney et al . , 2012 ) and is bound by the hindlimb-specific transcription factor Pitx1 in the developing mouse hindlimb ( Infante et al . , 2013 ) ( Figure 2C ) . The locus was homozygous for the deletion in 35 of 54 additional feather-footed birds from 21 breeds , but was never homozygous in 96 scale-footed birds from 56 breeds ( Chi-square , p<0 . 0001; Figure 2—figure supplement 2B ) . The 3 feather-footed birds from our whole-genome panel that did not have this deletion ( including the male founder of the aforementioned genetic cross implicating this same region ) also showed allelic differentiation from scale-footed birds over this interval , suggesting that an additional feathered-foot allele may also exist at this locus ( Figure 2—figure supplement 2A ) . In contrast to the differentiation signal we observed between scale-footed and all feather-footed birds on scaffold 79 , only the muffed birds ( more heavily feathered ) showed signatures of selection and shared similar haplotypes on scaffold 70 ( higher pFst signal than the adjacent scaffold 95 ) ( Figure 2D , Figure 2—figure supplements 2C , 3 ) . Thus , both QTL analyses and whole-genome scans show that mutation on scaffold 79 alone is sufficient for the grouse phenotype ( gr locus ) , and point to scaffold 70 as the major-effect locus for longer toe feathers in birds with muffs ( Sl locus ) ( Figure 1F , G ) . Next , we examined scaffolds 79 and 70 for candidate genes that might control the scale-to-feather transition . The highest pFst peak on scaffold 79 – corresponding to the major-effect QTL on LG 11 for the proportion of tarsometatarsal feathering – was approximately 200 kb upstream of Pitx1 , a gene that encodes a homeobox-containing transcription factor that is normally expressed in the vertebrate hindlimb but not the forelimb ( Figure 2B ) . The highest pFst peak on scaffold 70 – corresponding to the major-effect QTL on LG 20 for toe feather length – was 40 kb upstream of Tbx5 , a gene that encodes a T-box transcription factor that is normally expressed in the vertebrate forelimb but not the hindlimb ( Figure 2D ) . These regions were especially intriguing because these two genes encode key transcriptional regulators of forelimb ( Tbx5 ) and hindlimb ( Pitx1 ) identity and development ( Logan et al . , 1998; Logan and Tabin , 1999; Rodriguez-Esteban et al . , 1999; Szeto et al . , 1999; Takeuchi et al . , 1999 ) . For example , misexpression of Pitx1 in the embryonic chick forelimb blocks feather development ( Logan and Tabin , 1999 ) , while misexpression of Tbx5 in the early hindlimb field of embryonic chickens is sufficient to induce a partial wing-like transformation , including feather formation on the feet ( Takeuchi et al . , 1999 ) . In mouse , Pitx1 ( but not Tbx5 ) plays a role in determining limb-type identity ( Szeto et al . , 1999; Minguillon et al . , 2005; DeLaurier et al . , 2006 ) , suggesting that the roles of Tbx5 in limb outgrowth and identity have diversified during amniote evolution ( Horton et al . , 2008 ) . We did not identify any fixed non-synonymous coding changes in Pitx1 or Tbx5 between scale-footed and feather-footed breeds of pigeon . However , we found striking differences in embryonic hindlimb expression of these genes among three different breeds – racing homer ( scale-footed ) , Indian fantail ( small-muffed ) , and English trumpeter ( large-muffed ) – at Hamburger-Hamilton stage 25 ( HH25; Hamburger and Hamilton , 1951 ) . Pitx1 expression was reduced in both muffed breeds ( expression relative to racing homer: Indian fantail 0 . 75 ± 0 . 06 , p=0 . 0007; English trumpeter 0 . 40 ± 0 . 05 , p=0 . 0007; n = 6 each ) and was more severely reduced in the large-muffed English trumpeter ( p=0 . 002 ) ( Figure 3A ) . Conversely , Tbx5 , the forelimb-specific transcription factor , was ectopically expressed in the hindlimbs of both muffed breeds ( hindlimb expression relative to racing homer forelimb: racing homer 0 . 001 ± 0 . 0004; Indian fantail 0 . 01 ± 0 . 008 , p=0 . 0007; English trumpeter 0 . 14 ± 0 . 05 , p=0 . 0007; n = 6 each ) , and was higher in the large-muffed English trumpeter ( p=0 . 002 ) ( Figure 3B ) . Forelimb expression of Tbx5 was indistinguishable among the three breeds , indicating that upregulation of Tbx5 in feather-footed breeds is restricted to the hindlimb ( Figure 3—figure supplement 1A ) . 10 . 7554/eLife . 12115 . 011Figure 3 . Limb-type gene expression varies among feathered and scaled pigeons and chickens . ( A , B , F , G ) qRT-PCR analyses of Pitx1 and Tbx5 expression in HH25 hindlimbs of pigeon ( A , B ) and chicken ( F , G ) . Boxes span 1st to 3rd quartiles , bars extend to minimum and maximum observed values if within 1 . 5 times the interquartile range of the box , circles indicate values outside of this range , black line indicates median . **=p<0 . 01 , ***=p<0 . 001 . ( C-E , H-J ) RNA in situ hybridization for Tbx5 expression in HH25 embryos of racing homer ( C ) , Indian fantail ( D ) , and English trumpeter ( E ) pigeons; and white leghorn ( H ) , Cochin ( I ) , and silkie ( J ) chickens . Arrowheads indicate sites of ectopic Tbx5 expression . Scale bar = 1 mm . DOI: http://dx . doi . org/10 . 7554/eLife . 12115 . 01110 . 7554/eLife . 12115 . 012Figure 3—source data 1 . Source data from quantitative RT-PCR experiments . Abbreviations: fl , embryonic forelimb; hl , embryonic hindlimb . DOI: http://dx . doi . org/10 . 7554/eLife . 12115 . 01210 . 7554/eLife . 12115 . 013Figure 3—figure supplement 1 . Quantitative RT-PCR expression analyses . ( A ) Expression levels of Tbx5 are similar among racing homer , Indian fantail , and English trumpeter HH25 forelimb buds ( n = 6 samples each for all pigeon and chicken comparisons ) . ( B ) Expression comparisons for additional genes within candidate intervals on scaffolds 70 and 79 . See main text for further discussion of Tbx3 results . ( C ) Expression levels of Tbx4 are similar among racing homer , Indian fantail , and English trumpeter hindlimbs . ( D ) Expression levels of Tbx3 are reduced in HH25 hindlimb buds of one ( but not both ) feather-footed chicken breeds . DOI: http://dx . doi . org/10 . 7554/eLife . 12115 . 01310 . 7554/eLife . 12115 . 014Figure 3—figure supplement 2 . Spatial expression pattern of Pitx1 is similar in hindlimb buds of scaled-foot and feathered-foot embryos . Whole-mount RNA in situ hybridization results for Pitx1 expression in racing homer ( scaled-foot , left ) and Indian fantail ( feathered-foot , right ) HH25 embryos . DOI: http://dx . doi . org/10 . 7554/eLife . 12115 . 01410 . 7554/eLife . 12115 . 015Figure 3—figure supplement 3 . Ectopic hindlimb expression of Tbx5 and epidermal transformations in embryos and adults . ( A-C’ ) Vibrotome sections through HH25 pigeon hindlimbs shown in Figure 3C–E . Tbx5 expression is absent from the hindlimb buds of the wild-type racing homer ( A ) , but is present in the mesenchyme ( most notably in the posterior-dorsal region ) of the muffed Indian fantail and English trumpeter breeds . Magnified images ( B’ , C’ ) show that staining is limited to the mesenchyme ( m ) and does not extend into the ectoderm ( e ) . Sections were cut in the zeugopod region; magnified sections include the fibula condensation ( f ) . ( D-G ) Embryonic hindlimb of an English trumpeter at approximately 15 days of incubation . Digits are numbered from medial ( 1 ) to lateral ( 4 ) . Feather primordia are visible as string-like structures throughout the limb in medial ( D ) and lateral ( E ) views . Feathers are longer on the lateral digits ( G ) than on the medial digits , and these differences persist into adulthood . ( H ) Comparison of adult feet of a racing homer ( scaled , wild-type , left ) and an English trumpeter ( feathered , right ) . All small feathers have been plucked from the English trumpeter foot , and large feathers have been trimmed down to their insertions in the skin . Note that digit 4 in the English trumpeter is not visible due to the expanded skin on digit 3 , and unlike the backward-facing digit 1 of the racing homer , digit 1 faces forward and medially . ( I-N ) Details of feather size and distribution in the foot of an English trumpeter . Distal ends of long feathers were clipped and therefore full adult feather size is not represented in these images . The toes are barely visible in a dorsal view of an intact foot ( I ) , and the insertion of large feathers on the lateral side of the foot is visible in ventral view ( J ) . Small feathers were plucked from digit 1 and the medial metatarsus to reveal the small feathers inserting on digit 2 in medial view ( K ) , and feathers were plucked from digit 2 to reveal small feathers inserting on dorsal digit 3 and metatarsus ( L ) . Removal of all small feathers shows the lateral insertion points of large feathers on digit 3 in dorsal view ( M ) . Substantial expansion of lateral skin accommodates the insertion of these large feathers , and digit 4 is hidden from view . Similar skin expansion and large feather insertions characterize digit 4 and the lateral metatarsus ( N , ventral view ) . Soft tissue webbing joins digits 3 and 4 ( dashed line ) , as described by Darwin , 1868 . DOI: http://dx . doi . org/10 . 7554/eLife . 12115 . 015 We examined expression of additional genes within the two candidate regions at HH25 , and found that the Tbx5 paralog Tbx3 was also differentially expressed in both feather-footed pigeon breeds relative to racing homer ( Figure 3—figure supplement 1B ) . This could be due to the fact that Tbx3 is a target of Tbx5 ( Mori et al . , 2006; Postma et al . , 2008 ) , and additional experiments confirm that cis-regulatory changes do not drive this expression difference ( see below ) . The hindlimb-specific transcription factor Tbx4 is not contained in the candidate regions defined by our QTL mapping and genome-wide association studies , but this gene is a downstream transcriptional target of Pitx1 ( Logan and Tabin , 1999; Takeuchi et al . , 1999; Duboc and Logan , 2011 ) . We therefore compared expression levels of Tbx4 among scaled and feathered breeds at HH25 , but found no significant differences at this stage ( racing homer 1 . 00 ± 0 . 21; Indian fantail 1 . 00 ± 0 . 17 , p=0 . 85; English trumpeter 1 . 05 ± 0 . 19 , p=0 . 66; n = 6 each ) ( Figure 3—figure supplement 1C ) . Thus , the embryonic hindlimbs of muffed pigeons show quantitative expression changes in transcription factors with reciprocal limb expression domains , including the striking downregulation of a key hindlimb identity gene ( Pitx1 ) , and the novel expression of a key forelimb-specific gene ( Tbx5 ) . We next analyzed the patterns of Tbx5 and Pitx1 expression in scaled and muffed pigeon embryos at HH25 . Interestingly , ectopic hindlimb expression of Tbx5 in muffed embryos was markedly different than its normal forelimb pattern in wild-type pigeons and other vertebrates ( Gibson-Brown et al . , 1998a; 1998b; Logan et al . , 1998; Tamura et al . , 1999; Ruvinsky et al . , 2000 ) . Tbx5 is typically expressed throughout the mesoderm of the forelimb , but ectopic Tbx5 expression was largely localized to the mesoderm of the proximal and posterior-dorsal hindlimb of the small-muffed Indian fantail ( Figure 3D , Figure 3—figure supplement 3B ) . This domain was further expanded in the large-muffed English trumpeter ( Figure 3E , Figure 3—figure supplement 3C ) , consistent with the quantitative differences in expression between the two breeds ( Figure 3B ) . This domain shows a striking correlation with regions of epidermal transformation , as foot feathers are usually longest and most numerous on the posterior digits ( Darwin , 1868; Levi , 1986 ) ( Figure 3—figure supplement 3D–N ) . In contrast , and despite quantitative expression differences among breeds , Pitx1 had a qualitatively similar expression domain in embryos of scale-footed and feather-footed breeds at this stage ( Figure 3—figure supplement 2 ) . Therefore , consistent with the critical role of mesoderm in determining ectodermal fate ( Hughes et al . , 2011 ) , regionalized ectopic expression of Tbx5 is correlated with enhanced local transformation of epidermal appendages . If cis-acting regulatory mutations are responsible for the differences in Pitx1 , Tbx5 , and/or Tbx3 expression between embryos of scale-footed and feather-footed pigeons , then differential expression of scaled-foot and feathered-foot alleles should persist in a common trans-acting cellular environment . To test this prediction , we generated F1 hybrid pigeon embryos by crossing an Old Dutch Capuchine ( scale-footed ) to a fairy swallow ( muffed ) , and measured parent-of-origin allele expression in the hybrid embryonic hindlimb at HH25 ( Figure 4A ) ( Domyan et al . , 2014 ) . Consistent with expression differences we observed among breeds ( Figure 3A , B ) , expression of the feathered-foot allele of Pitx1 was approximately 20% lower than the scaled-foot allele ( expression of feathered-foot relative to scaled-foot allele: 0 . 807 ± 0 . 039 , p=0 . 003 , n = 6 embryos ) , and expression of the feathered-foot allele of Tbx5 was nearly 1600% higher than the scaled-foot allele ( relative expression of feathered-foot allele: 15 . 75 ± 4 . 69 , p=0 . 002 , n = 7 embryos ) ( Figure 4B , C ) . In contrast , expression levels of feathered-foot and scaled-foot alleles of Tbx3 were indistinguishable ( relative expression of feathered-foot allele: 0 . 99 ± 0 . 05 , p=0 . 68 , n = 7 embryos ) ( Figure 4D ) . These results directly show that cis-acting genetic changes alter expression of feathered-foot alleles of Pitx1 and Tbx5 , but not Tbx3 , in the embryonic pigeon hindlimb . 10 . 7554/eLife . 12115 . 016Figure 4 . Allele-specific expression ( ASE ) assays demonstrate cis-regulatory changes in Pitx1 and Tbx5 . ( A ) Schematic of ASE assay using Tbx5 expression as an example . Differences in Tbx5 expression between scale-footed and feather-footed breeds could be due to trans- and/or cis-acting mutations . If expression differences between parental breeds are due to trans changes only ( stars ) , then expression of the two Tbx5 alleles in hybrid embryos will be the same ( top right ) . In contrast , if cis-regulatory changes underlie differences in Tbx5 expression between the parental breeds , then expression of the two Tbx5 alleles in hybrid embryos will be different ( bottom right ) . ( B-D ) ASE assay in hybrid hindlimb buds indicate cis-regulatory divergence between scale-footed ( Old Dutch Capuchine ) and muffed ( fairy swallow ) pigeon breeds in Pitx1 ( B ) and Tbx5 ( C ) , but not in Tbx3 ( D ) . Dashed blue line indicates null hypothesis of equal expression of alleles . ( E ) ASE assay in hybrid hindlimb buds indicate cis-regulatory divergence in Tbx5 between feather-footed ( silkie ) and scale-footed ( white leghorn ) chicken breeds . Boxes in ( B-E ) span 1st to 3rd quartiles , bars extend to minimum and maximum observed values if within 1 . 5 times the interquartile range of the box , circles indicate values outside of this range , black line indicates median . **p≤0 . 01 ***p≤0 . 005 . DOI: http://dx . doi . org/10 . 7554/eLife . 12115 . 01610 . 7554/eLife . 12115 . 017Figure 4—source data 1 . Source data from pyrosequencing ASE experiments . Abbreviations: fl , embryonic forelimb; hl , embryonic hindlimb . DOI: http://dx . doi . org/10 . 7554/eLife . 12115 . 017 We next performed genome-wide comparisons of differential enhancer activity in embryonic limbs of racing homers and Indian fantails , using H3K27ac as a marker for open chromatin ( Creyghton et al . , 2010 ) . Strikingly , two regions of significantly different enrichment in racing homer hindlimbs relative to forelimbs were within the 44-kb genomic region that is deleted in the Indian fantail , and one of these regions ( log10 likelihood ratio = 4 . 93 ) is adjacent to the hs1473 limb enhancer ( Figure 2C ) . Furthermore , the most significant differentially enriched region in Indian fantail hindlimbs relative to racing homer hindlimbs was directly upstream of Tbx5 ( log10 likelihood ratio = 10 . 3 ) ( Figure 2E ) . The overlapping patterns of enrichment in Indian fantail hindlimbs and wild-type forelimbs suggest that ectopic hindlimb expression of Tbx5 could be due to de-repression of forelimb-specific enhancers . In summary , the differential expression of Pitx1 and Tbx5 among pigeon breeds ( Figure 3 ) and between alleles ( Figure 4 ) is also reflected by differential chromatin activation at these genes . In mouse and chicken embryos , experimental manipulation of Pitx1 and Tbx5 expression can result in muscular and skeletal abnormalities . Experiments in both chick and mouse consistently demonstrate that Pitx1 plays a necessary ( but not sufficient ) role in determining hindlimb-type morphology of the skeleton , muscles , and tendons ( Logan and Tabin , 1999; Takeuchi et al . , 1999; Minguillon et al . , 2005; DeLaurier et al . , 2006; Duboc and Logan , 2011 ) . Experimentally induced ectopic expression of Tbx5 in the hindlimbs of chick embryos can also induce muscular and skeletal anomalies , although Tbx5 does not directly control limb skeletal patterning or determine forelimb-type morphology in mice ( Rodriguez-Esteban et al . , 1999; Takeuchi et al . , 1999; Minguillon et al . , 2005; Hasson et al . , 2007 ) However , normal patterning of limb muscles and tendons is dependent on Tbx5 and Tbx4 in mice ( Hasson et al . , 2010 ) . These apparent discrepancies between mammalian and avian systems point to subtle differences in limb development in different lineages ( Horton et al . , 2008 ) . Given the dramatic musculoskeletal defects observed in other organisms with experimentally altered Pitx1 and Tbx5 expression , we compared the hindlimb morphology of adult feral pigeons ( scale-footed , n=2 ) to that of the English trumpeter ( muffed , n=2 ) and the Pomeranian pouter ( muffed , n=1 ) . We found consistent soft-tissue patterning defects in both feather-footed pigeon breeds: the fibularis longus ( FL ) tendon inserts on the dorsal rather than ventral surface of the proximal tarsometatarsus , the flexor perforans et perforatus ( FPP3 ) muscle adopts a longitudinal rather than pennate fiber orientation , and a slip of the FPP3 fuses with the FL tendon ( Figure 5A ) . These changes are aberrations of normal patterning , although they are not necessarily clear transformations to a more forelimb-like configuration . We also found that the fibula , which is normally splint-like and shorter than the tibia in pigeons , was enlarged ( Figure 5D , E ) and two phalanges of digit 4 were fused in feather-footed breeds ( not shown ) . These are not necessarily limb-type transformations , either . However , experimental ectopic expression of Tbx5 in the hindlimbs of chick embryos produces an enlargement of the fibula reminiscent of extreme pigeon phenotypes , and Takeuchi et al . , 1999 compared this morphology to a forelimb-like condition ( the fibula “makes a joint at its distal end like a normal ulna [the corresponding postaxial zeugopod bone of the forelimb] , ” p . 810 ) . Notably , all of the modified structures of ptilopodous pigeons develop in the posterior ( lateral in the adult ) and dorsal hindlimb , which are the primary sites of ectopic Tbx5 expression . Thus , the morphological changes to the hindlimbs of feather-footed pigeon breeds are considerably more than skin deep . 10 . 7554/eLife . 12115 . 018Figure 5 . Muffed pigeons have re-patterned hindlimb musculoskeletal system and wing-like foot feathers . ( A , B , C ) Gross muscle morphology of scale-footed ( feral ) and muffed ( Pomeranian pouter ) left hindlimbs , dorsal view . ( A ) Skin and superficial muscles have been removed to reveal re-patterning of the fibularis longus ( FL , red ) . Dashed black line , approximate position of ankle joint axis . ( B , C ) The FPP3 is a pinnate muscle in scale-footed pigeons ( B ) , but a slip of fibers fuses with the adjacent FL in muffed pigeons ( arrowhead in C ) . ( D , E ) X-ray computed tomography images of scale-footed ( feral , right leg ) and muffed ( English trumpeter , left leg ) hindlimbs . Arrowheads mark the proximal and distal ends of the fibula . The wild-type pigeon fibula ( D ) is short and splint-like . In the muffed bird ( E ) , the fibula extends from the knee to the ankle . We observed distal elongation of the fibula in another muffed breed ( fairy swallow ) but the fibula did not completely extend to the ankle ( not shown ) . t , tibia; tmt , tarsometatarsus . ( F ) Toe and wing ( flight ) feathers of a muffed pigeon ( English trumpeter ) , highlighting vane width asymmetries . Blue bar , inner vane; red bar , outer vane . Scale bar = 2 cm . DOI: http://dx . doi . org/10 . 7554/eLife . 12115 . 018 Other bird species , including domestic chickens , independently evolved foot feathers . Similar to pigeons , Tbx5 was ectopically expressed at HH25 in hindlimb buds of two feather-footed chicken breeds , the Cochin and the silkie ( hindlimb expression relative to white leghorn forelimb: white leghorn 0 . 005 ± 0 . 004; Cochin 0 . 066 ± 0 . 050 , p=0 . 002; silkie 0 . 034 ± 0 . 014 , p=0 . 009; n = 6 white leghorns , 6 Cochins , 4 silkies ) ( Figure 3G ) . Ectopic Tbx5 expression in feathered-foot chicken embryos had a similar domain to that of feathered-foot pigeon embryos at HH25 ( Figure 3I , J ) and , as in pigeons , cis-acting changes contributed to this expression ( expression of feathered-foot allele relative to scaled-foot allele in HH25 silkie x white leghorn F1 hybrid hindlimbs: 1 . 80 ± 0 . 41 , p=2 . 55 x 10–5 , n = 11 hybrid embryos ) ( Figure 4E ) . Hence , Tbx5-related developmental mechanisms may , in part , underlie the evolution of foot feathering in two species that last shared a common ancestor over 80 million years ago ( Claramunt and Cracraft , 2015 ) . Classical genetic studies implicate at least two loci in heavy foot feathering in chickens ( Punnett and Bailey , 1918; Lambert and Knox , 1929; Warren , 1948; Somes , 1992 ) , although the molecular genetic origins of the trait remain unknown . Previously , a chromosome region containing Pitx1 was implicated in foot feathering in silkie chickens ( Dorshorst et al . , 2010 ) . However , we did not detect statistically significant changes in Pitx1 expression between scaled-foot ( white leghorn ) and feathered-foot ( silkie and Cochin ) chicken embryos at HH25 ( expression relative to white leghorn: Cochin 0 . 92 ± 0 . 24 , p=0 . 93; silkie 0 . 71 ± 0 . 18 , p=0 . 18; n = 6 each ) ( Figure 3F ) . This apparent conflict could be because the causative gene in silkies is not actually Pitx1 but rather a gene closely linked to it , or because Pitx1 expression differences are more pronounced and consistent at developmental stages that we did not assay . Furthermore , different populations of breeds such as silkies appear to have different constellations of ptilopody loci and alleles , and it is possible that we used strains that do not have Pitx1 mutations ( Wexelsen , 1934; Somes , 1992 ) . Also in contrast to our results in feather-footed pigeons , Tbx3 was not upregulated in ptilopodous chicken breeds ( white leghorn 1 ± 0 . 17 , silkie 0 . 48 ± 0 . 18 , p=0 . 004; Cochin 0 . 86 ± 0 . 44 , p=0 . 40; silkie vs . Cochin p=0 . 07; n = 6 samples each ) ( Figure 3—figure supplement 1D ) . In all , these results suggest that both shared and distinct mechanisms regulate foot feathering among avian species .
Extensive classical breeding experiments in pigeons demonstrate that complex derived traits can often be parsed into component parts ( Sell , 1994 , 2012 ) . Thus , while traits are not always simple , they are often genetically tractable when using an informed breeding strategy ( Domyan et al . , 2014 ) . Equivalent insights about the genetic architecture of phenotypic divergence between wild vertebrate species are often considerably more difficult to acquire . With pigeons , however , we have documentation for specific breed selection criteria and direct evidence for the resulting genetic architecture of derived traits ( Levi , 1965; 1986; Sell , 1994; National Pigeon Association , 2010; Sell , 2012 ) . This information offers a crucial advantage because it informs how we design genetic crosses and choose breeds for whole-genome resequencing to identify causal genes and mutations . Thus , we can combine classical breeding strategies and genomics to identify the molecular basis of both simple and oligogenic traits , as well as dissect different components of a complex phenotype , and define functional interactions among genes ( Shapiro and Domyan , 2013; Domyan et al . , 2014; Vickrey et al . , 2015 ) . Classical studies in pigeon suggest two major-effect loci – grouse ( gr ) and Slipper ( Sl ) – are responsible for most of the variation in foot feathering ( Doncaster , 1912; Wexelsen , 1934; Hollander , 1937; Levi , 1986 ) . Through a combination of genetic , genomic , and developmental approaches , our data implicate regulatory mutations in the limb outgrowth and identity genes Pitx1 and Tbx5 as the molecular identities of the gr and Sl locus , respectively ( Figure 6 ) . Each locus has significant and separable effects on qualitative and quantitative variation in epidermal appendages: derived alleles of Pitx1 increase the extent of foot feathering , while a derived allele of Tbx5 is associated with the more elaborate muffed phenotype ( Figure 1 ) . Notably , these feathers are most numerous on the central forward-facing toe ( digit 3; Table 2; Figure 3—figure supplement 3 ) , just as forelimb feathers are most numerous on the central forelimb digit in birds and their dinosaurian relatives ( Yalden et al . , 1985; Gishlick et al . , 2001; Hieronymus , 2015 ) . Further , we also find that muscular and skeletal morphology are altered in muffed pigeons . 10 . 7554/eLife . 12115 . 019Figure 6 . Model describing link between Pitx1 and Tbx5 expression levels and foot epidermal appendage morphology . Darker colors indicate higher expression levels . Decreased expression of Pitx1 and ectopic expression of Tbx5 are associated with foot feathering ( and other morphological transformations ) in domestic pigeons . DOI: http://dx . doi . org/10 . 7554/eLife . 12115 . 019 Collectively , these findings point to a partial alteration of the identity of the developing hindlimb , rather than localized changes to individual epidermal placodes . These alterations do not represent a complete transformation of limb type , as the hindlimbs of feather-footed pigeons are still readily recognizable as legs . This suggests that limb-type identity is not a simple binary choice between two global fates . For example , feather-footed pigeons have a radical transformation of the distal hindlimb dermis , yet changes to other hindlimb mesoderm derivatives ( muscle , skeleton ) are subtler and largely restricted to lateral structures in the adult . Therefore , we propose that different aspects of fore- and hindlimb morphology could have different dosage- and/or stage-dependent requirements for exposure to identity cues . Our ongoing analyses of musculoskeletal phenotypes in our F2 cross , which includes individuals with different combinations of feathered-foot alleles of Pitx1 and Tbx5 , will allow us to understand the separate and epistatic effects of these loci on musculoskeletal anatomy . We note that , although genetic manipulations indicate that Tbx5 does not specify forelimb identity in mice , the divergence time between mammals and birds is deep ( >300 million years ) and subtly different roles for this transcription factor in limb outgrowth and identity might have evolved in these lineages ( Minguillon et al . , 2005; Horton et al . , 2008 ) . Another important avenue of future research will be to determine the downstream molecular consequences of Pitx1 and Tbx5 misregulation , and how this ultimately results in the transformation of scaled into feathered epidermis . Mutations at these genes can cause congenital limb deformities in humans , including clubfoot and Liebenberg syndrome ( Pitx1; Gurnett et al . , 2008; Spielmann et al . , 2012 ) , and Holt-Oram syndrome ( Tbx5; Basson et al . , 1997 ) . Notably , haploinsufficiency causes these human syndromes , and clubfoot is partially penetrant in Pitx1+/- mice ( Alvarado et al . , 2011 ) , collectively pointing to an exquisite sensitivity of limb morphology to levels of Pitx1 and Tbx5 gene products . Pitx1 is also involved repeatedly in the evolution of adaptive pelvic fin loss in stickleback fish ( Cresko et al . , 2004; Shapiro et al . , 2004; 2006; Coyle et al . , 2007; Chan et al . , 2010; Shikano et al . , 2013 ) . Threespine sticklebacks that are homozygous for a Pitx1 pelvic enhancer deletion have severely reduced or absent pelvises , but heterozygotes also have smaller pelvises , thereby allowing natural selection to act on fish carrying one mutant allele ( Shapiro et al . , 2004; Chan et al . , 2010 ) . Similarly , we observed increased foot feathering in pigeons with just one derived allele of Pitx1 or Tbx5 ( Figure 1 ) . Ancient pigeon breeders could have rapidly selected for ptilopodous pigeon lines starting with birds that were heterozygous for mutations at either locus , and later generated the extreme muffed phenotype by hybridization . Together , studies of diversity and disease indicate that modest changes to the amount and location of Pitx1 and Tbx5 gene expression can cause dramatic alterations to limb development and morphology . In addition to implicating cis- acting mutations in Pitx1 and Tbx5 driving transformation of scales into feathers in domestic pigeon , our results suggest that additional , as yet unidentified , mutations contribute to the muff phenotype . Although all feathered-foot embryos examined in our gene expression experiments contained derived Pitx1 and Tbx5 haplotypes , misregulation of each gene was more severe in the large-muffed English trumpeter than in the small-muffed Indian fantail . The English trumpeter may therefore contain additional cis-acting mutations at one or both loci , and/or mutations in upstream regulators of Pitx1 and Tbx5 . Additional studies will be required to discriminate between these possibilities . Our findings suggest a quantitative link between transcription factor abundance and skin appendage fate and morphology , thereby highlighting foot-feathering in pigeons as a model for studying the regulatory interactions that govern expression of these two important determinants of limb morphology . How might pigeons help us understand the evolution of epidermal appendage distribution and limb morphology in other species ? Our findings suggest the mechanistic basis for the development of feathered feet in two distantly related domestic bird species is due to a partial transformation of limb identity , through cis-acting regulatory mutations in limb-type specific transcription factors . Most modern wild birds have a scaled metatarsus and toes , although some species ( e . g . , ptarmigan , snowy owl , and golden eagle ) have extensive foot feathering . However , recent paleontological evidence suggests that feathers – not scales – might be the ancestral hindlimb skin appendages in birds and some of their close non-avian dinosaur relatives ( Hu et al . , 2009; Zheng et al . , 2013 ) . Thus , the epidermis of feather-footed modern birds might actually represent a reversion to the ancestral avialan skin condition . In some cases , the large , asymmetric-vaned , pennaceous metatarsal feathers of basal birds and their non-avian dinosaur relatives are so extensive that their hindlimbs have been interpreted as 'hind wings' , although they clearly retain hindlimb skeletal identity ( Zheng et al . , 2013 ) . Furthermore , these long foot feathers are directed laterally from the foot , and they display vane width asymmetries reminiscent of flight feathers; we find a similar morphology in muffed pigeons ( Figure 5F , Figure 3—figure supplement 3 ) . Perhaps not coincidentally , Darwin , 1868 noted of the muffed English trumpeter pigeon , “Their feet are so heavily feathered , that they almost appear like little wings” ( p . 155 ) . Building on classical breeding experiments in both pigeons and chickens , we find that a relatively small number of genetic changes account for a large proportion of the variation in epidermal appendage morphology and distribution . Thus , major determinants of dramatic phenotypic variation can be mechanistically simple and therefore potentially evolve rapidly . In pigeons , these mechanisms can generate wing-like feathers on a hindlimb that is not used for powered flight or gliding . This , in turn , suggests that wing-like foot and leg feathers in other species , such as non-avian dinosaurs , might result from developmental constraints on the morphology of large limb feathers , rather than from functional adaptations for flight ( Gould and Lewontin , 1979; Foth et al . , 2014 ) .
Animals were housed in accordance with the University of Utah Institutional Animal Care and Use Committees of University of Utah ( protocols 10–05007 and 13–04012 ) . 130 F2 offspring were generated by mating a male Pomeranian pouter to two female Scandaroons , and DNA samples extracted ( DNeasy Blood and Tissue Kit , Qiagen , Valencia , CA ) . 114 F2 offspring survived to 6 months of age , at which time they were euthanized and phenotypic measurements taken . Proportion of the tarsus was measured by dividing the length of the dorsal tarsus that was feathered by the total length of the tarsus ( measured from the tibia-tarsometatarsus joint to the distal aspect of tarsometatarsal-phalangeal joint of digit 3 ) , and averaged between the two tarsi . Toe feathers were counted on each toe , and summed across all 8 toes . The length of each of the longest three toe feathers on digit 3 ( the central forward-directed toe ) , which bore the longest toe feathers on each foot , was measured to the nearest 1 mm and averaged for each bird . For genotyping , we used a previously published approach ( Elshire et al . , 2011 ) with minor modifications . Briefly , for each founder parent and 130 F2 offspring , 50 ng of DNA was digested with ApeKI , ligated to barcoded adapters , and then 10 ng of each barcoded sample was pooled in batches of 26 individuals and purified ( Qiagen PCR Purification Kit ) . DNA fragments 550–650 bp in size were selected using Pippin Prep ( Sage Science , Beverly , MA ) , and amplified by 10–12 cycles of PCR using custom indexed primers . Libraries were purified with Ampure beads ( Sigma-Aldrich , St . Louis , MO ) and sequenced using 100- or 125-bp , paired-end sequencing on the Illumina HiSeq2000 platform at the University of Utah Genomics Core Facility . Reads were trimmed to 90 bp , filtered for quality , and de-multiplexed using Stacks ( Catchen et al . , 2011 ) . Reads were mapped to the pigeon reference genome ( Shapiro and Domyan , 2013 ) using Bowtie2 ( Langmead and Salzberg , 2012 ) , filtering for MAPQ < 20 . The average number of mapped reads among F2 individuals was 3 , 397 , 598 , with a mean depth of 6 . 3x . Genotypes were called using Stacks ( Catchen et al . , 2011 ) , with a minimum read-depth cutoff of 5 . Markers that were genotyped in ≥ 70 of the 130 F2 individuals were retained . Genetic map construction and QTL mapping was performed using R/qtl ( www . rqtl . org ) ( Broman et al . , 2003 ) . Markers showing segregation distortion ( Chi-square , p<0 . 05 ) were removed . 3803 markers were assembled into linkage groups using the parameters ( max . rf = 0 . 15 , min . lod = 6 ) . Linkage groups were numbered in descending order , based on the number of markers . Linkage group 11 and 20 QTL were initially mapped using the scanone function using Haley-Knott regression . Probable false-homozygote genotyping errors resulting from the low read-depth cutoff used ( 5x ) , identified as closely-spaced double-crossover events , were manually corrected on these linkage groups . Subsequently , the stepwiseqtl function was used to identify additional QTL , and the fitqtl function used to account for the effect of one linkage group while calculating the LOD scores and percent variance explained ( PVE ) of the other . Significance thresholds of α = 0 . 05 were calculated with 1000 permutations of each phenotype across all linkage groups . The peak markers for each phenotype were used to test for the effect of each QTL . BAM files generated previously for a whole-genome resequencing panel ( Shapiro and Domyan , 2013 ) were combined with BAM files for two new Pomeranian pouter whole-genome sequences to call genomic variants ( SNVs and small indels ) using the Genome Analysis Toolkit ( Unified Genotyper and LeftAlignAndTrimVariants functions , default settings ) ( McKenna et al . , 2010 ) . We removed variant sites that were called in two or fewer genomes ( i . e . , all other genomes were no-calls ) or that had variant alleles on only two or fewer chromosomes , as these categories of low-frequency variants would be uninformative to our analyses . The resulting variant call format ( VCF ) file was used for subsequent analyses . Individual birds from different breeds were binned into the following phenotypic classifications: Groused: Berlin long-faced tumbler , Lahore , Oriental frill , Shaksharli . Muffed: English long-faced muffed tumbler , English pouter , English trumpeter , frillback , ice pigeon , Indian fantail ( 2 individuals ) , Pomeranian pouter ( 2 individuals ) , Saxon monk , Saxon pouter . The English pouter is an unusual breed that is sometimes classified by breeders as slipper only . Its foot feathering is far more extensive than groused breeds , which led us to include it in the muffed group for the purposes of the genomic analyses . Scale-footed: African owl , archangel , Birmingham roller , carneau , Chinese owl , cumulet , Egyptian swift , English carrier , fantail , feral ( 2 individuals ) , Iranian tumbler , Jacobin , king , Lebanon , Marchenero pouter , mookee , Oriental roller , parlor roller , runt , Scandaroon , Spanish barb , starling , Syrian dewlap , Thai laugher . pFst , a modified likelihood ratio test that accounts for genotype uncertainty , extended haplotype homozygosity ( EHH ) , and haplotype network analyses were implemented using the GPAT++ software library ( Kronenberg et al . , 2014; see https://github . com/vcflib/vcflib for software updates ) . Primers for genotyping the scaffold 79 deletion are listed in Supplementary file 1 . Breeds used for association testing were as follows: Feather-footed ( 21 breeds total ) : Berlin long-faced tumbler , Berlin short-faced tumbler , Bokhara trumpeter , classic Oriental frill , crested Saxon field color , English trumpeter , fairy swallow , frillback , German double-crested trumpeter , ice pigeon , Indian fantail , Lahore , Mindian fantail , Oriental frill , Persian roller , Pomeranian pouter , Russian tumbler , saint , Schmalkaldner moorhead , Uzbeck tumbler , West of England . Scale-footed ( 56 breeds total ) : African owl , Altenburg trumpeter , American flying tumbler , American giant homer , American mini crest , American show racer , archangel , Bohemian pouter , Brunner cropper , Budapest tumbler , Cauchois , Chinese owl , cumulet , Danzig highflier , domestic show flight , dragoon , English baldhead long-faced clean-legged tumbler , English carrier , English magpie , English short-faced tumbler , exhibition homer , fantail , Franconian trumpeter , French mondaine , giant runt , Holle cropper , horseman pouter , Italian owl , Jacobin , Jiennesse pouter , king , Lebanon , Spanish little friar tumbler , Maltese , medium-faced crested helmet , Modena , mookee , Norwich cropper , nun , Old Dutch Capuchine , Old German owl , Oriental roller , parlor roller , Portuguese tumbler , Scandaroon , showtype racing homer , Spanish barb , starling , Syrian Baghdad , Texas pioneer , Thai laugher , Thuringer clean leg , Vienna medium-faced tumbler , Voorburg shield cropper , zitterhals . Forelimb and hindlimb buds from HH25 racing homer and Indian fantail embryos were collected and placed in 1% formaldehyde for 20 min at room temperature , then washed 3x in ice-cold PBS and stored at -80°C until chromatin extraction . ChIP was performed on 200 micrograms of chromatin isolated from embryonic pigeon limbs . Control libraries were prepared using 100 ng of input chromatin . A total of 16 libraries were created ( 8 ChIP and 8 input controls for each breed and limb combination ) . A validated monoclonal antibody against H3K27ac ( Millipore #05–1334 , Billerica , MA ) was used to perform ChIP , and sequencing libraries were prepared using NEBNext Ultra DNA Library Prep Kit for Illumina with NEBNext Multiplex Oligos for Illumina ( Index Primers Set 1; New England BioLabs , Ipswich , MA ) . All libraries were size selected using SPRI magnetic beads to eliminate adapter dimers . All 8 ChIP libraries showed enrichment for a positive control site relative to input libraries ( tested by qPCR ) . Single-end , 50-bp read sequencing was performed on Illumina HiSeq2000 platform at the University of Utah Genomics Core Facility . Fold-enrichment plots were generated using MACS ( Zhang et al . , 2008; Feng et al . , 2012 ) and visualized in IGV ( Robinson et al . , 2011; Thorvaldsdóttir et al . , 2013 ) . Regions of differential enrichment between racing homer and Indian fantail hindlimbs were identified using function bdgdiff in MACS2 ( https://pypi . python . org/pypi/MACS2/2 . 0 . 10 . 20130522 ) . Regions with a log10 likelihood ratio ≥ 3 were considered to have differential enrichment between the two groups . To assay gene expression , limb buds from HH25 embryos were harvested and placed in RNAlater ( Qiagen , Valencia , CA ) at 4°C overnight . Total RNA was extracted , cleaned and DNase-treated ( Qiagen RNeasy Kit ) . mRNA was reverse-transcribed to cDNA using oligo-dT and M-MLV RT ( Invitrogen , Carlsbad , CA ) according to the manufacturer’s protocol . cDNA was amplified using intron-spanning primers for each target using a CFX96 qPCR instrument and iTaq Universal Sybr Green Supermix ( Bio-Rad , Hercules , CA ) . Results were compared by Mann-Whitney U test . Two technical replicates of each sample were performed , and the mean value determined . Differences were considered statistically significant if p< ( 0 . 05 / # genes assayed ) to control for multiple-testing . Each experiment was performed three times , and the results presented are from one representative experiment . Primers used for each assay are listed in Supplementary file 1 . SNPs in Pitx1 and Tbx5 transcripts were identified by Sanger sequencing in the parents of a cross between an Old Dutch Capuchine ( scale-footed ) and a fairy swallow ( muffed ) that was homozygous for the 44-kb deletion upstream of Pitx1 , and PyroMark Custom Assays ( Qiagen ) for each SNP were designed using the manufacturer’s software . Pyrosequencing was performed on cDNA and gDNA derived from HH25 limb buds using a PyroMark Q24 instrument ( Qiagen ) . The signal intensity ratio of feathered allele to scaled allele from cDNA samples was normalized to ratios obtained from gDNA samples from the same embryos to control for allele-specific amplification bias . Normalized ratios were analyzed by Mann-Whitney U test , and considered significant if p< ( 0 . 05 / # genes assayed ) to control for multiple-testing . Each experiment was performed twice , and the results presented are representative . Primers used for each assay are listed in Supplementary file 1 . Linear templates for probe synthesis were amplified from cDNA by PCR using primers listed in Supplementary file 1 . Binding sites for T3 and T7 polymerase were incorporated into the forward and reverse primers to facilitate subsequent transcription of sense and antisense probe , respectively . Embryos used for RNA in situ hybridization were dissected from eggs , and fixed overnight in 4% paraformaldehyde at 4°C on a shaking table , then dehydrated into 100% MeOH and stored at -20°C . RNA in situ hybridization was performed as described ( Abler et al . , 2011 ) . Hybridization with sense probe was performed as negative control . | Animals ranging from fish to birds display dramatic diversity within and among species; yet remarkably little is known about the genetic and developmental mechanisms that underlie this variation . In birds and their extinct dinosaur relatives , the distribution of scales and feathers on the feet is a highly variable trait . Different breeds of domestic pigeon all belong to the same species but have feet that can be feathery or scaly to different extents . Classical genetics experiments suggested that only a few genes are involved in the transition from scaled to feathered skin on the feet of pigeons . However , the molecular basis for this transition was unknown . Domyan et al . set out to identify the genes involved in the transition from scaled to feathered feet by mating different breeds of pigeon in the laboratory and then sequencing the birds’ DNA . They also surveyed the entire DNA sequences of many additional pigeon breeds with and without feathered feet . This combined approach showed that two regions of the pigeon genome have a profound effect on the number and size of feathers on the feet of domestic pigeons . These regions contain genes that are known to play key roles in controlling the growth of a limb and whether it develops into a leg or a wing . In developing pigeon embryos , Domyan et al . found that a gene called Pitx1 , which is typically considered a hindlimb gene , is expressed at lower levels in the developing legs of breeds with feathered feet than in a breed with scaled feet . The experiments also showed that Tbx5 , a gene that is expressed in the forelimbs of many animals , is expressed abnormally in the embryonic hindlimbs of breeds of pigeon and chicken with feathery feet . Together , these findings suggest that the hindlimbs of domestic birds with feathery feet are more like wings at the molecular level , which results in them being covered in feathers rather than scales . Future work will now aim to discover the specific DNA sequences that alter the expression of Pitx1 and Tbx5 in feather-footed breeds , and whether the same genes control the foot feathers of other species of birds . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology",
"genetics",
"and",
"genomics"
] | 2016 | Molecular shifts in limb identity underlie development of feathered feet in two domestic avian species |
Reduced cardiac contractility during heart failure ( HF ) is linked to impaired Ca2+ release from Ryanodine Receptors ( RyRs ) . We investigated whether this deficit can be traced to nanoscale RyR reorganization . Using super-resolution imaging , we observed dispersion of RyR clusters in cardiomyocytes from post-infarction HF rats , resulting in more numerous , smaller clusters . Functional groupings of RyR clusters which produce Ca2+ sparks ( Ca2+ release units , CRUs ) also became less solid . An increased fraction of small CRUs in HF was linked to augmented ‘silent’ Ca2+ leak , not visible as sparks . Larger multi-cluster CRUs common in HF also exhibited low fidelity spark generation . When successfully triggered , sparks in failing cells displayed slow kinetics as Ca2+ spread across dispersed CRUs . During the action potential , these slow sparks protracted and desynchronized the overall Ca2+ transient . Thus , nanoscale RyR reorganization during HF augments Ca2+ leak and slows Ca2+ release kinetics , leading to weakened contraction in this disease .
The basic processes for cardiac excitation-contraction coupling are well described . Depolarization of the sarcolemma triggers the opening of voltage-gated L-Type Ca2+ channels ( LTCCs ) , and the resulting Ca2+ influx elicits additional Ca2+ release via Ryanodine Receptors ( RyRs ) in the sarcoplasmic reticulum ( SR ) . This process of Ca2+-induced Ca2+ release leads to a sharp increase in cytosolic Ca2+ concentration which initiates cardiomyocyte contraction . In ventricular myocytes , Ca2+ release is tightly controlled by the arrangement of LTCCs and RyRs in dyads , with LTCCs present in t-tubules juxtaposed from RyRs across a narrow 12–15 nm dyadic cleft ( Bers , 2001 ) . The RyRs themselves are organized into clusters; an arrangement that couples their gating , promoting synchronized opening and closing of neighbouring channels ( Marx et al . , 2001; Sobie et al . , 2006 ) . Recent data have indicated that neighbouring clusters of RyRs can also act concertedly if the Ca2+ diffusion distance between them is sufficiently short ( Macquaide et al . , 2015 ) . Referred to as ‘superclusters’ or Ca2+ Release Units ( CRUs ) , these functional arrangements of RyR clusters generate Ca2+ sparks , the fundamental units of SR Ca2+ release in cardiomyocytes ( Cheng et al . , 1993 ) . Ca2+ sparks are not only elicited by LTCC opening , but also occur spontaneously during diastole , where spark frequency and geometry can be measured to assess CRU function . While Ca2+ sparks are an important source of RyR-mediated Ca2+ leak from the SR , ‘silent’ or ‘non-spark’ events also occur , and involve the opening of a subset of RyRs within a CRU; so-called ‘quarky’ release ( Brochet et al . , 2011 ) . Impaired cardiomyocyte Ca2+ homeostasis is believed to importantly contribute to reduced cardiac contractility and arrhythmogenesis in heart failure ( HF ) . SR Ca2+ release is reduced and slowed in this condition , and these changes have been linked to altered dyadic structure ( Louch et al . , 2010 ) . We and others have observed marked remodeling of the t-tubular system in failing cardiomyocytes , while RyRs remain predominantly distributed along z-lines ( Song et al . , 2006; Louch et al . , 2006; Heinzel et al . , 2008 ) . Thus , the coupling between LTCCs and RyRs is disrupted , with ‘orphaned’ CRUs exhibiting delayed Ca2+ release only after trigger Ca2+ diffuses from intact dyads . However , abnormal gaps occurring between t-tubules only account for a fraction of the overall de-synchronization of Ca2+ release in HF ( Louch et al . , 2006; Øyehaug et al . , 2013 ) . This suggests that other alterations might also occur , perhaps at the nanometer scale of CRU organization , which hinder efficient triggering of Ca2+ release . CRU reorganization could in principle contribute to increased Ca2+ leak , including silent leak , which is a hallmark of heart failure ( Zima et al . , 2010; Walker et al . , 2014 ) . Exaggerated Ca2+ leak in failing cells has been linked to reduced SR Ca2+ content and depressed contractile function , elevation of resting Ca2+ levels and impaired relaxation , pro-arrhythmic early and delayed afterdepolarizations , and energetic inefficiency as Ca2+ is redundantly cycled ( Bers , 2014 ) . Thus , a detailed understanding of CRU structure and function in failing cells is critical . The advent of super-resolution microscopy techniques has markedly improved our ability to visualize and quantify CRU organization ( Baddeley et al . , 2009; Macquaide et al . , 2015; Jayasinghe et al . , 2018 ) . However , these techniques have not previously been employed to examine RyR configuration in HF . Using direct stochastic optical reconstruction microscopy ( dSTORM ) , we presently report that CRUs become dispersed in failing myocytes , as RyR clusters are broken apart . With the aid of mathematical modeling , we directly link these changes in CRU structure to experimentally measured increases in RyR leak and slowed SR Ca2+ release , identifying a novel mechanism underlying pathological remodeling of Ca2+ homeostasis in this disease .
Imaging was performed on isolated , fixed cardiomyocytes with antibody labelling of RyR2 . Using diffraction-limited confocal imaging ( resolution ≈250 nm ) and Structured Illumination Microscopy ( SIM , resolution ≈120 nm ) , the localization of RyRs along z-Lines was clearly apparent , but organization of RyRs within CRUs was not discernable ( Figure 1A ) . With dSTORM imaging , spatial resolution was markedly improved ( mean localization precision = 21 ± 3 nm ) enabling detailed CRU geometry to be assessed . For analysis of RyR cluster and CRU configuration , acquired raw images were fitted to a 30 × 30 nm grid , corresponding to the quatrefoil structure of the RyR protein ( Baddeley et al . , 2009 ) . Thresholding was then performed to create binary images ( Figure 1B ) , enabling quantification of RyR clusters , with an RyR counted as present if >half the area of a 30 nm square was above threshold . RyR clusters were defined by occupied , neighbouring grid positions , and CRUs were delineated by collecting neighbouring RyR clusters located within 150 nm ( Macquaide et al . , 2015 ) ( red boundaries in Figure 1B ) or 100 nm ( Baddeley et al . , 2009; Hou et al . , 2015 ) ( Figure 2—source data 1 ) . RyR organization was compared in cardiomyocytes from rats with post-infarction HF and cells from Sham-operated controls . Overall RyR expression was similar in Sham and HF , as evidenced by Western blotting of ventricular homogenates ( Figure 2— figure supplement 1 ) , and equivalent RyR labeling density in cardiomyocytes ( 41 . 9 ± 1 . 4 RyR/µm , 40 . 4 ± 1 . 3 RyR/µm in Sham , HF respectively ) . In both groups , RyR staining showed a predominantly transverse , striated pattern ( Figure 2A ) . However , despite rather similar organization of RyRs at the macroscale , nanoscale dSTORM imaging revealed fragmentation of RyR clusters in failing cardiomyocytes ( see insets in Figure 2A ) . Cluster breakup resulted in a reduction in the number of RyRs per cluster , and a greater proportion of small clusters in HF ( Figure 2B ) . The overall number of clusters increased accordingly in failing cells ( Figure 2D ) , and inter-cluster distance was reduced ( Figure 2E; see Figure 2—source data 1 for mean data across animals ) . Consistent with fragmentation of clusters into smaller adjacent groupings , the number of clusters contained in a CRU increased in HF ( Figure 2F ) , although the number of RyRs per CRU decreased ( Figure 2C ) since RyR clusters were markedly reduced in size . Convex hull analysis ( see methods ) revealed a consequent decrease in CRU solidity in HF ( Figure 2G ) . Thus , RyR reorganization in failing cells resulted in CRUs with a more sparse , dispersed configuration of smaller sub-clusters . We examined the functional implications of altered nanoscale organization of RyRs , first hypothesizing that RyR dispersion would augment SR Ca2+ leak in failing cardiomyocytes . Total Ca2+ leak was assessed in SR microsomes obtained from the left ventricle of Sham and failing hearts . Following initiation of microsomal Ca2+ uptake by addition of ATP , SERCA function was halted by thapsigargin treatment to reveal RyR-mediated Ca2+ leak ( Figure 3A ) . The Ca2+ leak rate was markedly higher in HF compared to Sham ( inset in Figure 3A , mean data in Figure 3C ) . In agreement with previous work ( reviewed in Bers , 2006 ) , we additionally observed significantly slowed SR Ca2+ uptake in HF ( Figure 3B ) and lower Ca2+ content ( Figure 3D ) . Ca2+ spark-mediated RyR leak was assessed by confocal linescan imaging of freshly-isolated cardiomyocytes ( Figure 3E ) . While the average Ca2+ release per spark ( spark mass ) was significantly increased in HF cells , this effect was offset by a tendency toward lower spark frequency ( Figure 3F , G ) . Indeed , overall spark-mediated Ca2+ leak was similar in HF and Sham cells ( Figure 3H ) . Since we observed an increase in total RyR leak in HF , these results are consistent with augmented 'silent’ , non-spark mediated leak in failing cells . To investigate whether increased silent Ca2+ leak could be linked to RyR dispersion , we employed a mathematical model of the dyad ( illustrated schematically in Figure 4—figure supplement 1A ) that enabled simulation of Ca2+ sparks with varied placement of RyRs within the CRU . We first incorporated small idealized CRUs containing as few as 4 RyRs ( Figure 4A ) , as our dSTORM imaging indicated that HF cells contain an increased fraction of small CRUs ( Figure 2C ) . During repeated simulations , a single RyR was opened at a random position within the CRU , and subsequent triggered RyR openings were allowed to proceed stochastically . Simulated Ca2+ release events with amplitudes ΔF/F0 ≥0 . 4 were defined as sparks , based on the detection threshold determined experimentally ( see methods ) . Ca2+ release from the smallest CRUs was never detected , but a progressively greater proportion of events yielded visible sparks as the number of RyRs in these idealized dyad geometries was increased ( Figure 4A ) . These results support the assertion that an increased fraction of small CRUs in HF promotes undetectable , silent Ca2+ leak . We next examined whether dispersion of clusters in larger more realistic CRUs could similarly contribute to increased silent Ca2+ leak in failing cells . To this end we incorporated real CRU geometries obtained by dSTORM imaging into the model ( Figure 4B ) . Four CRUs were selected containing roughly the same number of RyRs , but with different numbers of RyR clusters ( 1 , 3 , 7 or 10 clusters ) . As in the simulations described above for idealized CRU geometries , a single , randomly chosen RyR was opened in each simulation , to determine the likelihood that such triggering would result in a detectable Ca2+ spark . While relatively high fidelity spark generation was observed for the single-cluster CRU , Ca2+ release was more rarely observed to propagate between clusters , and spark fidelity was significantly lower in multi-cluster CRUs ( Figure 4B ) . This reduced efficiency of Ca2+ spark triggering in dispersed CRUs partly resulted from greater Ca2+ diffusion distance between neighbouring clusters , as demonstrated by progressively increasing the distance between RyR clusters in an idealized dyad ( Figure 4—figure supplement 1B ) . Furthermore , released Ca2+ is less efficiently confined in the dyadic space when the junctional SR has a more distributed and irregular shape . This latter point was demonstrated in the model by altering the amount of junctional SR surrounding the CRU; increasing junctional SR ‘padding’ increased spark fidelity in both idealized dyads ( Figure 4—figure supplement 1B ) and dSTORM-based geometries ( Figure 4—figure supplement 1C ) . In summary , these results indicate that nanoscale reorganization of RyRs in HF promotes non-spark-mediated SR Ca2+ leak by two mechanisms: ( 1 ) by creating smaller CRUs which produce Ca2+ release events below the detection limit , and ( 2 ) by creating more distributed CRU configurations in which multiple RyR clusters are less likely to co-operatively generate sparks . We next hypothesized that CRU dispersion would slow cardiomyocyte Ca2+ release; a hallmark of HF . Representative confocal recordings of Ca2+ sparks and their temporal profiles are shown in Figure 5A . Spark kinetics in Sham cells generally exhibited rapid rising and declining phases . While some sparks also showed fast kinetics in HF cells , others were markedly slow to rise and decay ( Figure 5A ) . Indeed , measurements of spark rise time and duration exhibited broader distributions and were , on the average , prolonged in HF compared to Sham ( Figure 5B ) . To investigate whether CRU dispersion in HF could underlie slowing of Ca2+ spark kinetics , we again employed our mathematical model with dSTORM-based CRU configurations . During the simulations , the time to opening of each RyR was registered , and the time course of the overall Ca2+ spark determined . Representative simulations show that RyR opening times were delayed in the dispersed , multi-cluster CRUs compared to the solid , single-cluster CRU ( Figure 6A ) . Simulations of Ca2+ spark time courses further showed that the delayed RyR openings in multi-cluster CRUs resulted in more variable kinetics and overall slowing of spark rise time ( Figure 6B , mean data Figure 6C ) , reproducing experimental observations . Of note , although CRUs were observed to contain fewer RyRs in HF than Sham ( Figure 2C ) , simply reducing the RyR number to an equivalent degree in the mathematical model did not markedly alter Ca2+ spark kinetics ( Figure 6—source data 1 ) , further confirming a key role of CRU fragmentation in failing cells . Finally , we examined the consequences of increased variability in Ca2+ spark kinetics for the Ca2+ transient in failing cells . We observed that field-stimulated Ca2+ transients were significantly slower to rise in HF than Sham ( Figure 7A–C ) . This slowing of Ca2+ release was associated with marked de-synchronization of the Ca2+ transient , which we quantified by measuring the variability in time to reach half-maximal fluorescence ( TTF50 ) across the cell ( see lower panels in Figure 7A ) . This ‘dyssynchrony index’ ( Louch et al . , 2006 ) was significantly increased in HF compared to Sham , with a strongly right-shifted distribution of values ( Figure 7D ) . T-tubule disruption in failing cells ( Figure 7—figure supplement 1 ) has been previously established in this model of HF ( Frisk et al . , 2016 ) , and is a recognized cause of Ca2+ release dyssynchrony in this disease ( Song et al . , 2006; Louch et al . , 2006; Heinzel et al . , 2008 ) . We examined whether alterations in Ca2+ spark kinetics also promote dyssynchrony , by examining local Ca2+ transients within narrow , 2 µm regions of the line scan . These regions were centered at the locations of spontaneous Ca2+ sparks observed when electrical pacing was halted . We specifically distinguished between locations with ‘slow’ sparks , defined by a rise time >13 ms ( ie . 1 S . D . > mean rise time in Sham ) , and remaining ‘fast’ sparks . By this definition , 24% of sparks in HF cells were defined as slow , while only 13% of Sham sparks fit this definition . Representative examples of such sparks and their temporal profiles are shown in Figure 7E , with corresponding positions along the line scan indicated in Figure 7A . Local transients from slow spark locations in HF exhibited markedly slower rise times than those from fast spark locations in both HF and Sham ( Figure 7F ) . The association between slow sparks and slow local transients was also apparent in ‘heat map’ plots ( Figure 7G ) . These results show that by protracting Ca2+ sparks , CRU dispersion during HF slows and desynchronizes the overall Ca2+ transient .
In the present study , we have employed dSTORM imaging to reveal key changes in CRU morphology during heart failure . We specifically observed marked dispersion of RyRs , which resulted in a shift towards smaller RyR clusters and CRUs . Remaining larger CRUs became less solid , with more fragmented configurations . Experiments and mathematical modeling linked these changes in RyR arrangement to two central aspects of impaired Ca2+ homeostasis in failing cells: increased ‘silent’ RyR leak and slowing of Ca2+ release , which are believed causative for reduced contractility in this condition . In contrast to lower resolution optical imaging techniques such as confocal and SIM microscopy , the dSTORM technique allows quantification of RyR organization within CRUs . We presently employed this technique with a grid-based quantification method for RyR counting , as previously developed by the Soeller group ( Baddeley et al . , 2009; Hou et al . , 2015 ) . Using images collected close to the cell surface ( depth of 200 – 500 nm , see also Macquaide et al . [Macquaide et al . , 2015] ) , we calculated that an average cluster contains approximately 14 RyRs in our healthy , Sham-operated rat ventricular cells . This estimation is in close agreement with previous estimates made at the cell surface ( 14 RyRs/cluster ) ( Baddeley et al . , 2009 ) , but considerably lower than estimates made deep within the cell interior ( Hou et al . , 2015 ) , a discrepancy which may reflect regional differences in RyR organization across the cell . However , it should be noted that the grid-based method for RyR counting assumes that the channels lie parallel to the field of view , with a uniform , grid-like configuration; presumptions which are likely less valid when RyRs are visualized internally than at the cell surface . Due to superimposition of internal RyRs along the z-axis , it is therefore likely that present RyR cluster sizes are somewhat underestimated . Assuming that RyR clusters located within 150 nm cooperatively form a CRU ( Macquaide et al . , 2015 ) , we calculated that average CRUs contain roughly 3 – 4 clusters , and a total of ≈30 RyRs . Using a narrower CRU definition , with inclusion of neighbouring clusters within 100 nm ( Baddeley et al . , 2009; Hou et al . , 2015 ) , reduced average CRU size to ≈25 RyRs ( Figure 2—source data 1 ) . Despite possible underestimation of RyR numbers/CRU due to methodological issues noted above , these estimates are in relatively close agreement with an electron microscopy tomography study of mouse ventricular cardiomyocytes ( Hayashi et al . , 2009 ) . The authors reported that while there was great variability in CRU size and geometry , dyadic volume was an order of magnitude lower than previous estimates ( Franzini-Armstrong et al . , 1999; Scriven et al . , 2013 ) ; an average-sized dyad could only hold up to 43 RyRs , with RyRs occupying ≈78% of the dyadic space ( ≈34 RyRs/CRU ) . It is estimated that typical sparks result from the activation of somewhat fewer RyRs , ≈20–30 ( Shkryl et al . , 2012 ) . One possible explanation for this discrepancy is that not all RyRs within a CRU may contribute to every spark . Alternatively , previous work based on estimation of the dyadic size available for RyRs ( Hayashi et al . , 2009 ) may have overestimated RyR number if RyRs are not densely packed ( Asghari et al . , 2014 ) . Our own present estimates have assumed that RyRs located within a fixed distance share the same junctional SR ( jSR ) , which may also be an overestimation since true jSR geometry is unknown . Indeed , our modeling results showed that reducing the degree of jSR ‘padding’ around each RyR cluster markedly reduces the ability of neighbouring clusters to function as cooperative CRUs ( Figure 4—figure supplement 1B , C ) . Future work may address this important issue by simultaneously assessing jSR and RyR arrangement by multi-colour super-resolution imaging , or direct visualization of RyRs and jSR using electron microscopy . Previous work by Zima et al . has shown that silent , non-spark mediated leak is a significant contributor to overall RyR leak ( Zima et al . , 2010 ) . We presently show that silent leak can be partly traced to small CRUs , which produce Ca2+ release events that are not detectable experimentally ( see also Walker et al . , 2014 ) . Consistent with previous work ( Baddeley et al . , 2009; Hayashi et al . , 2009; Hou et al . , 2015 ) , we observed that many CRUs have very small geometry even in healthy cells; 42% of CRUs contained five or fewer RyRs in Sham ( Figure 2C ) . However , larger CRUs can also contribute to non-spark mediated RyR leak , when Ca2+ release from CRU sub-clusters does not propagate to remaining clusters; so-called ‘quarky’ Ca2+ release ( Brochet et al . , 2011 ) . Our results show that the decreased fidelity of spark triggering in these dispersed CRUs results from the spacing between neighbouring clusters , which inhibits cooperative , diffusion-based triggering . Furthermore , released Ca2+ more easily escapes from less densely packed CRUs , making it less likely to trigger additional RyRs . This finding is in agreement with previous modeling studies showing that spark fidelity declined when clusters deviated from circular and compact shapes ( Walker et al . , 2014 ) , and that RyR activation during triggered release is less likely when CRUs are broken into sub-clusters ( Cannell and Soeller , 1997 ) . We presently link augmented silent leak in HF to an increased fraction of both small CRUs and CRUs with larger , distributed configurations . These results have important implications . Increased RyR leak during heart failure is widely believed to promote arrhythmia via generation of both early and delayed afterdepolarizations . Furthermore , greater leak also promotes depression of contraction , via loss of SR Ca2+ content , and poor relaxation , due to elevation of resting Ca2+ ( reviewed in Bers , 2014 ) . The present results provide a structural basis for these maladaptive functional alterations , and suggest that the nanometer scale of these changes prevented their previous detection by lower resolution imaging techniques . Slowing of Ca2+ release is another key component of the failing phenotype . Prolonged rise time of the Ca2+ transient during HF ( Figure 7A–C ) has been previously linked to slowed isolated cardiomyocyte contraction in both animals ( Bokenes et al . , 2008; Mørk et al . , 2009 ) and patients ( Davies et al . , 1995 ) . In vivo systolic tissue velocity is also reduced in HF patients ( Vinereanu et al . , 2005 ) which decreases contractile power . Thus , understanding the mechanisms underlying slowed Ca2+ release in failing cells is critical . Mathematical modeling studies showed that while out-of-focus release events can theoretically reduce spark kinetics artefactually , these events rapidly become undetectable as they are shifted further from the focal plane ( Figure 6—figure supplement 1 ) . Furthermore , there is no clear basis for expecting that dispersion of RyR clusters in HF should change the proportion of out-of-focus sparks relative to control , and for this reason we do not expect that variation in the focal plane has systematically impacted our measured rise times . Our modeling results also suggest that an altered number of RyRs/CRU , due either to methodological under-estimation or loss of RyRs from CRUs in HF , will not in and of itself promote slowing of sparks kinetics ( Figure 6—source data 1 ) . Indeed , we observed that spark kinetics were relatively insensitive to changes in RyR number for medium-sized CRUs , in agreement with previous work ( Cannell et al . , 2013 ) . Rather , our results point to an important role of CRU dispersion in slowing Ca2+ sparks , as multi-cluster dyads exhibited progressive triggering of individual clusters by diffusion of released Ca2+ ( Figure 6 ) . Importantly , sites with slow spontaneous Ca2+ sparks were observed to also have slow local triggered Ca2+ release during the action potential ( Figure 7 ) . Thus , dispersed CRUs promote overall slowing and de-synchronization of the Ca2+ transient . While these results link visible spark-mediated leak to slowing of overall Ca2+ release , small , undetectable RyR openings ( silent leak ) might also play an important role . Previous work has shown that RyR clustering allows the channels to be functionally coupled , whereby they exhibit coordinated opening and closing ( Marx et al . , 2001; Wang et al . , 2004; Sobie et al . , 2006 ) . Wang and colleagues reported that this thermodynamic stability is lost when RyRs are present alone or with a small number of neighbouring channels , and slow Ca2+ release kinetics result ( Wang et al . , 2004 ) . An increased fraction of small CRUs in failing cells may therefore augment slow , but undetectable Ca2+ release events which nevertheless contribute to an overall Ca2+ transient which is slow and de-synchronized . Dyssynchronous Ca2+ release during heart failure has been previously linked to t-tubule disruption in a large number of studies ( reviewed in Louch et al . , 2010 ) , and we have similarly observed t-tubule disorganization in this post-infarction rat model ( Figure 7—figure supplements 1 and Frisk et al . , 2016 ) . An important question is therefore whether CRU morphology and slow Ca2+ sparks occur independently from t-tubule reorganization , or whether these two aspects of structural remodeling are related . In previous work , we ( Louch et al . , 2013 ) and others ( Meethal et al . , 2007 ) employed simultaneous imaging of t-tubules and Ca2+ and observed that sparks occurred almost exclusively at t-tubule sites in both healthy and failing cells . Similarly , de-tubulation experiments have been shown to dramatically reduce the occurrence of Ca2+ sparks in the cell interior , suggesting that SR-t-tubule junctions are important for spark initiation ( Brette et al . , 2005 ) . It may be postulated , therefore , that CRU dispersion resulting in slow sparks occurs at intact dyads in HF ( ie . not at sites with ‘orphaned RyRs’ ) , and that t-tubule and CRU remodeling may occur independently . Verification of this point will likely require simultaneous nanoscale imaging of t-tubules and RyRs , as small degrees of t-tubule drift out of dyads may be critical , and not detected by lower resolution imaging techniques . Regardless , we believe that t-tubule and CRU disruption have additive , detrimental effects , resulting in markedly de-synchronized and slowed SR Ca2+ release . What signals RyR dispersion in HF ? Emerging data indicate that RyRs are not firmly anchored within the CRU , but exhibit a highly malleable organization dependent on factors such as phosphorylation status and cytosolic Mg2+ levels ( Asghari et al . , 2014 ) . However , while changes in such conditions were shown to influence whether RyRs are positioned in a ‘checkerboard’ or side-by-side arrangement , it is unclear whether they can lead to reorganization of clusters and CRUs on the scale of changes presently observed in HF . Another important dyadic regulator is Junctophilin-2 ( JPH-2 ) , which anchors the t-tubule to the SR ( Takeshima et al . , 2000; Minamisawa et al . , 2004 ) , and interacts with the RyR ( Beavers et al . , 2013; Munro et al . , 2016 ) . Munro et al recently reported that JPH-2 levels regulate RyR clustering; however , while they observed that JPH-2 overexpression triggered the formation of larger RyR clusters , JPH-2 knockdown did not reduce cluster size ( Munro et al . , 2016 ) . Although others have observed JPH-2 loss during HF ( Minamisawa et al . , 2004; Wei et al . , 2010 ) , we did not presently observe reduced JP-2 protein levels in our rat HF model ( Figure 2—figure supplement 1 ) , suggesting that JP-2 downregulation is not a prerequisite for reorganization of CRUs ( Figure 2 ) or t-tubules ( Figure 7—figure supplement 1 ) . Another dyadic regulator , BIN1 , is a well-established regulator of t-tubule growth and structure ( Hong et al . , 2014 ) , and recent data have suggested that this protein may also attract phosphorylated RyRs to the dyad ( Fu et al . , 2016 ) . Although BIN1 loss has been reported in other HF models ( Lyon et al . , 2012; Caldwell et al . , 2014 ) , our data indicate that BIN1 expression is unaltered in our rat model ( Figure 2—figure supplement 1 ) . Thus , we do not believe that BIN1 changes are related to the CRU reorganization presently observed in failing myocytes , and the precise trigger for such changes remains unclear . In conclusion , our results contribute to an emerging understanding that cardiomyocyte dyads are highly plastic structures . While previous work has shown that t-tubule structure is impressively malleable , and degraded during heart failure , our present findings show that there is also detrimental reorganization of RyRs in this disease . Dispersion of RyRs within the CRU was linked to increased silent RyR leak , slowing of Ca2+ sparks , and de-synchronization of the overall Ca2+ transient , indicating a novel mechanism underlying impaired contractility in HF .
All experiments were approved by the Norwegian National Animal Research Authority ( project license no . FOTS 5982 , 7786 ) , and were performed in accordance with the National Institute of Health guidelines ( NIH publication No . 85 – 23 , revised 2011 ) and European Directive 2010/63/EU . Large anterolateral myocardial infarctions were induced in ~300 g male Wistar-Hannover rats , by ligation of the left coronary artery as previously described ( Lunde et al . , 2012 ) . Development of HF was verified six weeks later using a Vevo 2100 echocardiography imaging system ( VisualSonics , Toronto , Canada ) . Inclusion of failing animals was based on established criteria ( Sjaastad et al . , 2000 ) , including dilation of the left atrium ( diameter >5 mm ) and ventricle , and increased lung weight ( >2 . 5 g ) . Sham-operated rats served as controls . Experiments were performed over a two year period , using animals from 10 rounds of animal surgery . Sample sizes were determined by power analysis , assuming that only 50% of post-infarction animals would be included in the final data set , and based on a pilot project of variability in CRU morphology in healthy controls . Cardiac myocytes from failing and Sham-operated rats were isolated using a standard enzymatic dispersion technique ( Louch et al . , 2011 ) . Excised hearts were mounted on a Langendorff setup , and retrogradely perfused through the aorta with Ca2+-free solution containing ( in mmol/L ) : 130 NaCl , 25 Hepes , 5 . 4 KCl , 0 . 5 MgCl2 , 0 . 4 NaH2PO4 , 5 . 5 D-glucose , pH 7 . 4 . Once cleared of blood , hearts were then perfused with the above solution including collagenase ( 2 mg/mL , Worthington Biochemical Corp . , Lakewood , NJ , USA ) and low [Ca2+] ( 0 . 05 mmol/L ) . After 10 min of digestion , hearts were cut down , minced , and filtered , and isolated cardiomyocytes were allowed to sediment . Isolated cardiomyocytes were transferred to cell culture medium ( DMEM 1X , Life Technologies with 10% FBS , Biowest Nuaillé , France and 1% Penicillin-Streptomycin , Sigma ) , and plated on laminin-coated , glass bottom culture dishes ( MatTek corporation , Ashland MA ) . Staining was performed according to a described protocol ( Swift et al . , 2007 ) , with consecutive steps for chemical fixation ( 4% Formaldehyde in 1 mol/L HEPES buffer , 10 min ) , quenching ( PBS + 100 mmol/L Glycine , 10 min ) , permeabilization ( PBS + 0 , 03% Triton X-100 , 10 min ) , and blocking ( NaCl 150 mmol/L , Na3 citrate 17 . 5 mmol/L , 5% goat serum , 3% BSA , 0 . 02% NaN3 , 2 hr ) . PBS washing was performed in between each step . The cells were then incubated overnight with 1/100 diluted mouse-anti-RyR2 primary antibody ( ThermoFischer Scientific , MA3–916 ) in low blocking buffer , containing 150 mmol/L NaCl , 17 . 5 mmol/L Na3 citrate , 2% goat serum , 1% BSA , and 0 . 02% NaN3 at 4°C . This protocol has previously been reported to result in the binding of multiple primary antibodies to each RyR tetramer ( Baddeley et al . , 2009 ) . The following day , cells were washed with PBS and incubated with 1/200 diluted secondary antibody ( Alexa Fluo 647 conjugated goat-anti-mouse secondary Ab , Molecular Probes/Invitrogen ) in low blocking buffer for 2 hr . Cells were then washed and stored in PBS until image acquisition . Of note , the fab-fragment secondary antibody employed places the fluorescent label far closer to the epitope than traditional antibodies . Thus , under our experimental conditions , the steric error is generally <10 nm , and dwarfed by the localization of the dSTORM technique ( ≈20 nm , see below ) . dSTORM imaging was performed using an OMX V4 system ( Applied Precision , GE Healthcare ) with a 60 × 1 . 49 NA TIRF objective ( Olympus ) , a pco . edge sCMOS camera ( PCO ) , a 100 mW 642 nm laser , and a 683/40 emission filter . Focusing was performed with a 30V300nanoX CL focusing unit ( Piezosystem , Jena ) . Cells were placed in ‘switching buffer’ ( 0 . 5 mg/mL glucose oxidase , 40 μg/mL catalase , 10% wt/vol glucose , 50 mmol/L β-mercaptoethylamine in Tris-buffer , pH 8 . 0 , all Sigma–Aldrich ) , and fluorophores were pushed into the dark state by illumination with the 642 nm laser at a highly inclined , but sub-TIRF angle ( Highly Inclined and Laminated Optical sheet , HILO; Tokunaga et al . , 2008 ) . Spontaneous blinking occurred without the use of an activation laser , and was recorded at a depth of 200–500 nm during ten-thousand frames per field of view ( 20 . 48 × 20 . 48 µm ) , with a maximum of 350 , 000 blinks recorded . Data were processed with built-in software ( softWoRx , GE Healthcare ) using a Dense Stochastic Sampling Imaging ( DSSI ) algorithm and multiple Gaussian fits to localize events . Drift correction was performed with a model-based algorithm . Average localization precision was 21 . 6 nm for the events included in the final reconstructions . Images were further processed using a custom analysis program written in Python , which was similar to one previously employing scikits-image , scipy . spatial and Opencv ( http://opencv . willowgarage . com/ ) modules ( Macquaide et al . , 2015 ) . These algorithms are publicly available in an online repository , ( https://github . com/TerjePrivate/Ryanodine_Receptor_Dispersion_during_Heart_Failure ) ( Kolstad , 2018; copy archived at https://github . com/elifesciences-publications/Ryanodine_Receptor_Dispersion_during_Heart_Failure ) . Images with 10 × 10 nm pixels were convolved with a 2D Gaussian function equal to the calculated resolution of the image ( ~20 nm ) , and downscaled by a factor of 3 to produce a final pixel size of 30 × 30 nm . RyR locations were defined using a modified automated thresholding algorithm ( Kolstad , 2018 ) ( Otsu method ) , excluding the brightest 0 . 3% of the signal . This prevented skewing of the threshold by regions which were constantly in the active state . To minimize inclusion of autofluorescence artefacts , blinks appearing in 10 or more consecutive frames were excluded from the final reconstruction . Acquired images were fitted to a 30 × 30 nm grid corresponding to the quatrefoil structure of the RyR protein ( Baddeley et al . , 2009 ) . An RyR was counted as present if >half the area of a 30 nm square was above threshold , and RyR clusters were defined by occupied , neighbouring grid positions . Based on previous ( Macquaide et al . , 2015 ) and present ( Figure 4—figure supplement 1 ) calculations clusters with edge-to-edge distance <150 nm were assumed to cooperatively generate Ca2+ sparks , and grouped into CRUs accordingly . A stricter CRU definition , with edge-to-edge distances < 100 nm , ( Baddeley et al . , 2009; Hou et al . , 2015 ) was also examined . Inter-cluster distances were calculated from the centroid of each cluster . CRU solidity was calculated as the proportion of the bounding polygon ( convex hull method ) which was filled with RyRs; clusters containing less than 5 RyRs were excluded from this calculation . The solidity ratio was 1 if totally filled and 0 if completely empty; therefore , lower values indicate greater CRU fragmentation . Of note , all analyses of RyR localization were performed by automated protocols in a blinded manner . To address whether unspecific secondary antibody binding affected measurements of RyR configurations , dSTORM imaging of cardiomyocytes was performed in the absence of primary antibody . The obtained signal was then added to RyR-labeled images obtained by the standard protocol ( primary plus secondary antibodies ) , and RyR configuration was analyzed . Non-specific labeling was observed to only negligibly increase the number of RyRs/cluster and RyRs/CRU by 3 . 7% and 3 . 2% , respectively . Similarly , RyR density was increased by 0 . 5% , and inter-cluster distance was reduced by 0 . 2% , supporting that unspecific labelling had a minute influence on the dataset . Using an LSM 7Live confocal microscope ( Zeiss ) , Ca2+ sparks were recorded from quiescent cardiomyocytes loaded with fluo-4 AM ( 20 µmol/L , Molecular Probes , Eugene , OR ) and superfused with a HEPES Tyrode solution containing ( in mmol/L ) : 140 NaCl , 1 . 0 CaCl2 , 0 . 5 MgCl2 , 5 . 0 HEPES , 5 . 5 glucose , 0 . 4 NaH2PO4 , 5 . 4 KCl , pH 7 . 4 , 37°C . Scans were performed with a 1024 pixel line drawn along the longitudinal axis of the cell with a temporal resolution of 1 . 5 ms . Ca2+ sparks were analysed with a custom program ( CaSparks 1 . 01 , D . Ursu , 2003 ) , as previously ( Louch et al . , 2013 ) . Sparks were defined as local increases in fluorescence with a minimum amplitude ( ΔF/F0 ) of 0 . 4 , to minimise the inclusion of false positives . Linescan images of cells obtained during inhibition of Ca2+ sparks ( prolonged exposure to 10 mM caffeine ) confirmed the appropriateness of this detection threshold . Ca2+ spark frequency was normalized to cell length and recording time , and spark geometry was assessed by measurements of time to peak ( TTP ) , full width at half maximum ( FWHM ) , and full duration at half maximum ( FDHM ) . Spark-mediated Ca2+ leak was calculated as the product of spark mass ( amplitude X FWHM X FDHM ) and frequency . Ca2+ transients were elicited by field-stimulation through a pair of platinum wires ( 3 ms supra-threshold current pulses at 1 Hz ) , and recorded as confocal linescans under the same experimental conditions as Ca2+ spark measurements . Global Ca2+ transient characteristics were analyzed by averaging the Ca2+ signal along the linescan , with measurements of transient magnitude ( normalized to resting fluorescence , F/F0 ) , time to half maximal fluorescence ( TTF50 ) , and TTP . Local Ca2+ transients were averaged across narrow 2 µm bands of the linescan . Synchrony of Ca2+ release was assessed as previously described ( Louch et al . , 2006 ) , by plotting the profile of TTF50 measurements across the cell and measuring the standard deviation of these values ( the ‘dyssynchrony index’ ) . SR Ca2+ content was assessed by rapidly applying 10 mM caffeine and measuring the amplitude of the elicited Ca2+ transient . Ca2+ handling was additionally examined using crude homogenates from rat left ventricle , based on methods described by O’Brien and modified by Li et al . ( O'Brien , 1990; Li et al . , 2002 ) . Fresh ventricular tissue was weighed and homogenized in ice cold buffer ( 1:10 wet weight/vol , pH 7 . 9 ) containing ( in mmol/L ) : 300 sucrose , 5 NaN3 , 1 EDTA , 40 L-histidine , 40 Tris HCl and protease inhibitors . Homogenization was performed with a Polytron 1200 ( Kinematica AG , Luzern , Switzerland ) at 25000 rpm for 3 × 20 s , with a 20 s break between bursts . Homogenates were then aliquoted , frozen in liquid N2 , and stored at −80°C until use . Ca2+ uptake and release were measured in 2 . 2 ml of assay buffer , containing ( in mmol/L ) : 165 KCl , 22 Hepes , 7 . 5 oxalate , 11 NaN3 , 0 . 0055 TPEN , 4 . 5 MgCl2 , 9 Tris HCl and 0 . 002 fura-2 salt ( pH = 7 . 0 , 37°C ) . Ca2+ fluxes were monitored with an LS50B luminescence spectrometer ( Perkin Elmer Ltd , Beaconsfield , Buckinghamshire , United Kingdom ) after addition of 100 µl of freshly-thawed and vortexed homogenate . Ca2+ uptake by the vesicles was initiated by addition of Na4ATP ( 2 . 2 mmol/L ) , and then blocked by application of thapsigargin ( 1 . 5 µmol/L ) to assess RyR leak . Releasable SR Ca2+ content was estimated by measuring Ca2+ release induced by application of the RyR opener 4-chloro-m-cresol ( 4-CMC ) ( 5 . 5 mmol/L ) . The fluorescence ratio was calibrated to [Ca2+] using the following equation: [Ca2+]=Kd * ( ( R - Rmin ) / ( Rmax - R ) ) * ( Sf2/Sb2 ) , where R is the 340 nm/380 nm fluorescence ratio , Kd is the dissociation constant of fura-2 and Sf2/Sb2 is the ratio of measured fluorescence intensity at 380 nm when fura-2 is Ca2+ free or saturated , respectively . Rmin is the ratio at very low [Ca2+]i and Rmax is the ratio at saturating [Ca2+]i , obtained by adding 3 . 3 mmol/L EGTA and 4 . 8 mmol/L CaCl2 respectively to the cuvette at the end of each recording . Frozen tissue from rat left ventricles was homogenized in cold buffer ( 210 mM sucrose , 2 mM EGTA , 40 mM NaCl , 30 mM HEPES , 5 mM EDTA ) with the addition of a Complete EDTA free protease inhibitor cocktail tablet ( Roche Diagnostics , Oslo , Norway ) and a PhosSTOP tablet ( Roche ) . SDS was then added to the homogenates to a final concentration of 1% , and protein concentrations were quantified using a micro BCA protein assay kit ( Thermo Fischer Scientific Inc , Rockford , IL ) . Bovine serum albumin ( BSA ) was used as standard protein . The following primary antibodies were employed for immunoblotting: RyR ( 1:1000; MA3-916 , Thermo Scientific ) , BIN1 ( 1:500; sc23918 , Santa Cruz Biotechnology ) , junctophilin-2 ( 1:1000; sc-51313 , Santa Cruz Biotechnology ) and GAPDH ( 1:500; sc-20357 , Santa Cruz Biotechnology ) . Secondary antibodies were anti-rabbit ( NA934V , GE Healthcare ) , anti-mouse ( NA931V , GE Healthcare ) or anti-goat ( HAF109 , R and D Systems ) IgG-HRP linked whole antibody . Data were normalized to GAPDH and then to Sham values ( Figure 2—figure supplement 1 ) . Protein homogenates ( 5 or 15 µg/lane ) were size fractionated on 4–15% or 15% Criterion TGX gels ( Biorad Laboratories , Oslo , Norway ) and transferred to 0 . 45 μM PVDF-membranes ( GE Healthcare ) . The membranes were blocked in 5% non-fat milk or 5% Casein ( Roche Diagnostics ) in Tris-buffered saline with 0 . 1% Tween ( TBS-T ) for 1 hr at room temperature , and then incubated with primary antibody overnight at 4°C . Secondary antibodies were incubated for 1 hr at room temperature and blots were developed using Enhanced Chemiluminescence ( ECL prime , GE healthcare ) . Chemiluminiscense signals were detected by a LAS 4000 ( GE healthcare ) and protein levels were quantified using ImageQuant software ( GE Healthcare ) . A mathematical model was created to simulate the effects of varied RyR localization and CRU geometry on Ca2+ spark characteristics . We have made all simulation results , geometries , and code specific to this study available in an online repository ( Kolstad , 2018 ) , along with code for the full reaction-diffusion simulator . The model extended from the work of Hake et al . ( 2012 ) , with included RyR stochasticity developed from previous work by Cannell and colleagues ( Cannell et al . , 2013 ) . We have chosen this model for the relative simplicity of its gating ( no direct inter-RyR coupling , or explicit luminal Ca2+ regulation ) , and because it was built for a similarly constructed ( spatially discretized ) geometry , ( Cannell et al . , 2013 ) unlike most other recent RyR2 gating ( Williams et al . , 2011; Wescott et al . , 2016 ) . A set of coupled partial differential equations was employed to describe the temporal evolution of the free and bound [Ca2+] in the SR and cytosol:∂c∂t=Dc∇2c−∑i=14Ri ( c , bi ) , x∈ΩC∂bi∂t=Di∇2bi+Ri ( c , bi ) , i=1 , 2 , 3 , 4 x∈ΩC∂s∂t=Ds∇2s−R5 ( s , b5 ) , x∈ΩS∂b5∂t=R5 ( s , b5 ) , x∈ΩS Here ΩC is this cytosolic domain , including the cleft space , and ΩS is the SR , including both junctional and network SR components . Four buffers were included in ΩC: ATP , calmodulin , troponin and Fluo-4 , and one buffer , calsequestrin , was included in ΩS . These buffers are numbered from 1 to 5 and their corresponding concentrations are denoted bi . Troponin and calsequestrin were regarded to be stationary , and the corresponding diffusion coefficients ( σ ) were therefore set to zero ( see Supplementary file 1 ) . The calcium concentrations in ΩC and ΩS are denoted c and s respectively . The buffering reactions are of the formRc , bi= koncBtot-bi-koffbiwhere B tot is the total buffer concentration , and k on and k off are the on and off rates for the buffer , respectively . The two domains are coupled through a flux condition over the SR membrane:Dc∂c∂n=−Ds∂s∂n=J ( c , s ) whereJ ( c , s ) ={JRyRx∈ΓRyRJSercax∈ΓSerca0elsewhere The RyR flux is computed by:JRyR ( c , s ) =gRyRγ ( c−s ) where γ∈0 , 1 is a stochastic variable that switches between the conductive ( O ) and non-conductive ( C ) states according to:k+C⇄O . k- While we have chosen a different form for the equations expressing the default transition rates , they are equivalent to those in the original model of Cannell et al . ( 2013 ) with the exception that we have set limits to both k+ and k- at low dyadic calcium:k+c=fcK+n+ , kmin+ , kmax+k-c=fcK-n- , kmin- , kmax-where for a<b:f ( y , a , b ) ={a , if y<ay , if a≤y≤bb , if y>b The parameters are given in Supplementary file 1 . The SERCA formulation is taken from Tran et al . ( 2009 ) and is of the form:JSerca ( c , s ) =a1c2−a2s2a3c2+a4s2+a5 . The computational domain in our model ( ΩS∪Ωc ) was a ( 2 µm ) cube containing a single CRU ( Illustrated in Figure 4—figure supplement 1A ) . Unlike the original work of Cannell and colleagues , for which RyR locations were fixed for all simulations , our simulations involve algorithm-defined changes in the jSR geometry and location of RyRs to reflect the structural differences captured by the dSTORM recordings . In all geometries the domain consisted of a 12 nm wide cleft space sandwiched between the junctional SR ( jSR ) surface and the t-tubular surface . The latter was represented as a non-conductive slab inside Ωc , serving as a barrier to diffusion . RyRs were located on the opposing jSR surface , each occupying a space 36 × 36 nm , with neighbouring RyRs placed 36 nm apart ( centre-to-centre distance ) . The jSR was modeled as a physical extension ( ‘padding’ ) around each RyR by a defined distance equivalent to 1 RyR diameter ( 36 nm ) . As mentioned above , RyRs were arranged according to dSTORM-derived locations , and the jSR shape was adjusted according to the RyR locations . In some simulations ( Figure 4A , Figure 6—source data 1 ) the geometries of the jSR were fixed while idealized RyR lattice geometries were varied to explore the independent effect of modifying RyR number with fixed jSR volume and the locally releasable calcium pool ( Figure 4A ) . To specifically investigate the effects of RyR dispersal , example dSTORM-identified RyR patterns were selected with similar total RyR number but with arrangement into a varied number of sub-clusters ( 1 , 3 , 7 , or 10 ) . The ratio of RyR number to jSR volume for these example CRUs is presented in Figure 4—source data 1 . To limit the effect variation in jSR volume has on the releaseable Ca2+ store , we fixed the total SR volume ( and thus initial SR Ca2+ content ) across all geometries by modifying the non-junctional SR ( nSR ) volume as required . Of note , while nSR concentration was modeled as a continuum , global SR calcium concentration was effectively clamped at initial values as expected for the time-scale and spatial-scale simulated . Dyadic Ca2+ release was initiated by opening a single RyR in the CRU; this ‘trigger’ RyR was selected randomly and varied between consecutive simulations . Simulated triggered Ca2+ release from the CRU was then allowed to proceed , with the above equations discretized in space with a finite volume approach ( 12 nm edge length throughout the domain ) , and solved in time using explicit Euler time stepping . Specifically , we used operator splitting and solved each of the reaction and diffusion sub-problems with a fixed ∆t = 0 . 1µs , except when calculating the RyR release current . Due to the small element volumes and high fluxes this calculation is very stiff , so instead we solve it analytically . If we use x to denote the SR calcium concentration and y to denote the cleft calcium concentration , we can write this sub-problem as:x˙=K ( y−x ) y˙=K ( x−y ) where K is the channel conductance per element volume . The solution to this subsystem is given by:x ( t ) =S−De−2Ktyt=S+De-2Kt Where:S= ( y0+x ( 0 ) ) 2D= ( y0-x ( 0 ) ) 2 Using this scheme it is possible to take arbitrarily long time steps without introducing instabilities . For the RyR gating model half-maximal activation was achieved at 80 µM Ca2+ , which allowed cooperative opening of adjacent RyR clusters located up to ≈150 nm apart ( ie . 4 RyR lengths ) if the clusters shared jSR ( Figure 4—figure supplement 1B , C ) . Thus , we defined CRUs as groupings of RyR clusters with edge-to-edge distances < 150 nm , in agreement with recent work ( Macquaide et al . , 2015 ) , but also compared data with a stricter CRU definition ( cluster distances < 100 nm ) employed in other publications ( Baddeley et al . , 2009; Hou et al . , 2015 ) . RyR rates are shown in Supplementary file 2 . All results are expressed as mean values ± standard error of the mean . All statistical significance was calculated in SigmaPlot by Student’s t-test or ANOVA with Bonferroni post-hoc comparison for normally distributed data , as appropriate . Skewed distributions of experimental and modelled Ca2+ spark parameters were respectively assessed by the nonparametric Mann-Whitney Rank Sum Test and Kruskal-Wallis ANOVA with Dunn’s test for post-hoc comparisons . dSTORM-based measurements of RyR geometries were compared with averages taken both across cells and animals , with respective statistical testing by t-tests and linear mixed effects models ( Lindstrom and Bates , 1988 ) . Statistical significance was defined as p<0 . 05 . All raw data acquired and analyzed in this study are publicly available at https://github . com/TerjePrivate/Ryanodine_Receptor_Dispersion_during_Heart_Failure ( Kolstad , 2018; copy archived at https://github . com/elifesciences-publications/Ryanodine_Receptor_Dispersion_during_Heart_Failure ) . | The muscle cells of the heart coordinate how they contract and relax in order to produce the heartbeat . During heart failure , these cells become less able to contract . As a result the heart becomes inefficient , pumping less blood around the body . For the cardiac muscle cells to contract , the levels of calcium ions in the cells needs to rapidly increase . In failing hearts , these increases in calcium ion levels are smaller , slower and less well coordinated . It was not known what causes these changes , making it difficult to treat heart failure . Calcium ions are released in cardiac muscle cells through protein channels called ryanodine receptors . These receptors form clusters that allow them to synchronize when they open and close . Could the reorganization of ryanodine receptors account for the problems seen in failing hearts ? To investigate , Kolstad et al . examined rat hearts using a technique called super-resolution microscopy . This showed that the clusters of ryanodine receptors break apart during heart failure to form smaller clusters . Further experiments showed that calcium ions ‘leak’ from these smaller clusters , reducing the amount of calcium that can be released into cardiac muscle cells during each heartbeat . Released calcium also spreads between the dispersed clusters , resulting in a slower rise of the calcium levels in the cells . Both changes contribute to weakened contractions of cells in failing hearts . Therefore , heart failure can be traced back to very small rearrangements of the ryanodine receptors . This understanding will help researchers as they investigate new ways to treat heart failure . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology"
] | 2018 | Ryanodine receptor dispersion disrupts Ca2+ release in failing cardiac myocytes |
Apical constriction is a widely utilized cell shape change linked to folding , bending and invagination of polarized epithelia . It remains unclear how apical constriction is regulated spatiotemporally during tissue invagination and how this cellular process contributes to tube formation in different developmental contexts . Using Drosophila salivary gland ( SG ) invagination as a model , we show that regulation of folded gastrulation expression by the Fork head transcription factor is required for apicomedial accumulation of Rho kinase and non-muscle myosin II , which coordinate apical constriction . We demonstrate that neither loss of spatially coordinated apical constriction nor its complete blockage prevent internalization and tube formation , although such manipulations affect the geometry of invagination . When apical constriction is disrupted , compressing force generated by a tissue-level myosin cable contributes to SG invagination . We demonstrate that fully elongated polarized SGs can form outside the embryo , suggesting that tube formation and elongation are intrinsic properties of the SG .
Organs that transport gases and nutrients , as well as those producing and secreting vital hormones and enzymes , are organized as epithelial tubes , many of which arise from already polarized epithelial sheets ( Andrew and Ewald , 2010 ) . To enter the third dimension , a flat sheet of polarized epithelial cells bends or invaginates using either of two distinct processes–wrapping or budding ( Lubarsky and Krasnow , 2003 ) . During wrapping , the entire epithelial sheet folds in until its edges meet and seal to form an elongated tube , as occurs in vertebrate neural tube formation and in Drosophila gastrulation ( Andrew and Ewald , 2010; Massarwa et al . , 2014 ) . During budding , a subset of cells extend out of the plane of the epithelium in an orthogonal direction to form a tube; this process is observed during branching morphogenesis of many organs , including the mammalian lungs and kidney , and the primary branches of the Drosophila trachea ( Andrew and Ewald , 2010; Lubarsky and Krasnow , 2003 ) . A limited number of cellular processes are involved in creating three-dimensional structures , which include regulated changes in cell shape , arrangement and position , as well as oriented cell divisions and spatially restricted programmed cell death ( Andrew and Ewald , 2010 ) . One cell shape change associated with such tissue remodeling is apical constriction , wherein the nuclei move to a basal position in the cell and the apical domains constrict ( Martin and Goldstein , 2014; Sawyer et al . , 2010 ) . In polarized epithelial cells that maintain cell-cell adhesion , apical constriction is linked to tissue folding or invagination ( Alvarez and Navascués , 1990; Hardin and Keller , 1988; Kam et al . , 1991; Lewis , 1947; Sweeton et al . , 1991; Wallingford et al . , 2013 ) . Non-muscle myosin II-dependent contractility generates the force that drives this cellular process . Particularly , a pulsatile actomyosin complex in the apical medial region of the cell ( hereafter referred to as apicomedial myosin ) has been described in tissues that undergo apical constriction ( Blanchard et al . , 2010; Martin et al . , 2009 ) . Studies in early Drosophila embryos have identified the Folded gastrulation ( Fog ) pathway that regulates apical constriction and apicomedial myosin formation ( Manning and Rogers , 2014 ) . During gastrulation , mesodermal cells undergo apical constriction to form the ventral furrow along the anterior/posterior body axis . In those cells , the mesoderm-specific transcription factors Twist and Snail activate G protein-coupled receptor signaling and recruit RhoGEF2 to the apical surface , which , in turn , activates Rho1 ( Costa et al . , 1994; Kölsch et al . , 2007; Manning et al . , 2013; Parks and Wieschaus , 1991 ) . GTP-bound Rho1 then activates Rho-associated kinase ( Rok ) , which phosphorylates and activates non-muscle myosin II , which forms an actomyosin complex at the medial apical cortex ( Dawes-Hoang et al . , 2005 ) . This actomyosin complex causes asynchronous contractions that pull the adherens junctions ( AJs ) inward . Contractions are maintained between pulses by the actomyosin belt , which serves as a ‘ratchet’ to incrementally reduce apical area ( Martin et al . , 2009 ) . Although apical constriction and its associated forces are suggested to drive tissue invagination , the exact role of this cell shape change in tube formation remains controversial ( Llimargas and Casanova , 2010 ) . In Drosophila trachea defective for EGF receptor signaling , apical constriction is impaired , but most cells invaginate ( Brodu and Casanova , 2006; Nishimura et al . , 2007 ) . Similarly , in Drosophila embryos mutant for twist or fog , mesodermal cells with defective apical constriction still invaginate , although the process is both delayed and aberrant ( Leptin and Grunewald , 1990; Sweeton et al . , 1991 ) . In these mutants , however , apical constriction is not completely blocked; it is simply less extensive and more random ( Brodu and Casanova , 2006; Costa et al . , 1994; Nishimura et al . , 2007; Sweeton et al . , 1991 ) , making it difficult to draw any clear conclusions . A recent study , using an optogenetic method to deplete phosphatidylinositol-4 , 5 bisphosphate ( PI ( 4 , 5 ) P2 ) and actin from the cell cortex , showed that local inhibition of apical constriction is sufficient to cause global arrest of mesoderm invagination during Drosophila gastrulation ( Guglielmi et al . , 2015 ) . This finding suggests that apical constriction is essential for the invagination by wrapping that occurs during ventral furrow formation . It remains unclear , however , whether apical constriction is also critical for tissue invagination by budding . The Drosophila salivary gland ( SG ) is an excellent system to study the role of apical constriction during tissue invagination by budding ( Figure 1A–A’’ , B , B’ , C and C’ ) . The SG begins as a two-dimensional sheet of cells on the embryo surface that internalizes to form an elongated tube ( Chung et al . , 2014 ) . Since neither cell division nor cell death occurs once the SG has been specified , the entire morphogenetic process must be driven by changes in cell shape and rearrangement . Indeed , apical constriction has been observed in this tissue ( Myat and Andrew , 2000a ) , and an increase in apical myosin has been reported during SG invagination ( Escudero et al . , 2007; Nikolaidou and Barrett , 2004; Xu et al . , 2008 ) . More detailed analyses revealed several distinct myosin structures in the forming SG , including a supracellular myosin cable that surrounds the entire tissue and is thought to be involved in tissue invagination , as well as a web-like myosin structure in the apicomedial region of cells that colocalizes with actin ( Röper , 2012 ) . The latter shows pulsatile behavior , suggesting that contractile forces by the apicomedial actomyosin complex may drive apical constriction during SG invagination ( Booth et al . , 2014 ) . 10 . 7554/eLife . 22235 . 003Figure 1 . Clustered apical constriction does not occur in fkh mutant SGs . ( A–A’’ ) Epithelial invagination by budding . 3D reconstruction of Drosophila embryonic SGs stained with Crb ( green ) , an apically localized transmembrane protein , and CrebA ( magenta ) , an SG nuclear transcription factor , before ( A ) , at the beginning ( A’ ) and during ( A’’ ) invagination . White arrows in A’ indicate the budding epithelium . Insets , Crb signals only . ( B ) SEM images of ventral views of early and late stage 11 embryos show two SG placodes , with higher magnification to the right . An invagination pit is observed in the posterior dorsal region of each placode at late stage 11 in WT ( arrows in B’ ) , which is absent in fkh H99 mutants ( B’’ ) . ML , midline . ( C ) E-Cad staining of WT and fkh H99 mutant SGs of early and late stage 11 . Ventral views of a single SG placode are shown . Red lines denote the border of the SG placodes , based on a CrebA staining ( not shown ) . In this figure and later , anterior is to the left and dorsal is up . Robust apical constriction is observed in WT SGs of late stage 11 ( arrowhead in C’ ) , which is not detected in WT SG of early stage 11 ( C ) or in fkh H99 SG of late stage 11 ( C’’ ) . Asterisk , invagination pit . ( D–H ) Coordinated apical constriction is observed in WT SGs prior to and during invagination , which does not occur in fkh mutants . Representative SGs for the four distinct stages of invagination observed in WT ( D–G ) and late stage 11 fkh H99 embryos ( H ) and the corresponding heat maps of apical area are shown ( D’–H’ ) . Apical area of each cell was calculated by automated tracing of E-Cad along cell boundaries . Red and white lines denote the border of the SG placodes . Arrowheads , clustered apical constriction in the posterior ( white ) or anterior ( black ) region of the placode . Asterisks , invagination pit . ( I–K ) Percentage ( I ) and cumulative percentage ( J ) of WT SG cells in different apical area bins at each stage of invagination . Comparison of the cumulative percentage of cells in WT and fkh mutants ( K ) . P values are calculated using the Mann-Whitney U-test ( I ) and the Kolmogorov-Smirnov test ( J , K ) . See also Figure 1—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 22235 . 00310 . 7554/eLife . 22235 . 004Figure 1—source data 1 . SG cells quantified for apical area . DOI: http://dx . doi . org/10 . 7554/eLife . 22235 . 004 Here , we elucidate the mechanism by which apical constriction is regulated during SG budding and determine its role in tissue invagination . We show that the spatial and temporal pattern of apical constriction correlates with apicomedial myosin formation during SG invagination . We uncover the molecular pathway through which the FoxA transcription factor Fork head ( Fkh ) coordinates apical constriction by regulating Fog signaling in the SGs . Through genetic manipulations to completely block apical constriction and detailed quantitative analysis , we show that apical constriction is not required for SG invagination , but is required for proper tissue geometry . SG cells can internalize even without forming apicomedial myosin , suggesting a role for other additional forces . We provide evidence that the compressing force generated by the tissue-level supracellular myosin cable contributes to invagination in SGs with defective apical constriction . By analyzing the externalized SG phenotypes of fog mutants , we also reveal that tube formation can be decoupled from tissue invagination .
To gain a detailed view of temporal and spatial regulation of apical constriction during SG invagination , we analyzed apical area in wild type ( WT ) SGs of stage 11 embryos . Using labeling with E-Cadherin ( E-Cad ) , an AJ marker , and CrebA , an SG nuclear marker , we segmented apical cell outlines and calculated the apical area of individual cells . By analyzing >60 WT SGs for apical area and depth of invagination , we classified four distinct stages ( Figure 1D–G , D'–G' ) . At early stage 11 , all SG cells were on the embryo surface and cells with different apical area were distributed relatively randomly , although cells with small and large apical area tended to be toward the inside and at the periphery of the tissue , respectively ( Before invagination; Figure 1D and D’ ) . Shortly after , subsets of cells clustered in specific regions had smaller apical areas , most notably in a middle region close to the posterior edge of the SG where invagination begins ( Clustered apical constriction; Figure 1E and E’ ) . The less prominent anterior cluster corresponds to a small indentation that can be observed in lateral views of stage 12 SGs ( data not shown ) . When a few cells in the posterior cluster had internalized to form an invagination pit , many cells near the pit were apically constricted ( Beginning of invagination; Figure 1F and F’ ) . At late stage 11 , as more cells had invaginated , most cells anterior to the invagination pit were apically constricted ( Deep invagination; Figure 1G and G’ ) . At the same time , cells in the periphery of the SG were apically expanded , especially in the region ventral to the invagination pit . Quantification of the percentage and cumulative percentage of cells of different apical area showed a gradual and significant increase in the number of cells with smaller apical area as invagination progressed . Small increases in the number of cells with larger apical area were also observed ( Figure 1I and J ) . The SGs of fkh null mutants undergo early apoptosis and fail to internalize ( Myat and Andrew , 2000a ) . fkh mutant SGs rescued from cell death by deletion of the pro-apoptotic genes contained within the H99 deficiency also fail to internalize . SGs of H99 deficiency embryos develop completely normally during embryogenesis , indicating that blocking apoptosis does not affect SG morphogenesis ( Myat and Andrew , 2000a ) . Scanning electron microscope ( SEM ) and confocal images of the late stage 11 fkh H99 SGs did not reveal invagination pits ( Figure 1B’’ and C’’ ) , consistent with a complete failure of invagination . Analysis of apical areas of late stage 11 fkh H99 SGs revealed that fkh H99 SG cells had overall smaller apical area than WT cells; both the mean and median values for apical area were lower than those of WT ( Figure 1—source data 1 ) . Importantly , cells with the smallest apical areas were not clustered in fkh H99 mutant SGs and were instead randomly distributed ( Figure 1H and H’ ) . Moreover , the cumulative percentage of late stage 11 fkh H99 SG cells of different apical areas showed a similar distribution trend to that of WT SG cells before invagination ( Figure 1K ) . Indeed , median scaling normalization of the apical area of fkh H99 late stage 11 SG cells revealed a very similar distribution to the apical area of WT SGs before invagination . Particularly , the percentages of cells with the lower 70% apical area were indistinguishable ( Figure 1K ) . Taken together , our analysis reveals that apical constriction is both temporally and spatially coordinated during SG invagination and does not occur in fkh H99 mutants . Apical constriction of both vertebrate and invertebrate tissues is associated with contraction of the actomyosin cytoskeleton ( Dawes-Hoang et al . , 2005; Martin et al . , 2009; Nishimura et al . , 2007; Roh-Johnson et al . , 2012 ) . Specifically , pulsatile accumulation of apicomedial myosin has been linked to apical constriction ( Martin et al . , 2009; Blanchard et al . , 2010; Booth et al . , 2014 ) . Therefore , we investigated myosin accumulation in WT and fkh H99 SGs , using spaghetti squash ( sqh ) -GFP , a functional tagged version of the myosin regulatory light chain ( Royou et al . , 2004 ) . Consistent with results from ( Röper , 2012 ) , several myosin structures were observed , including a supracellular myosin cable that encircles the whole tissue , AJ-associated cortical myosin , as well as an apicomedial myosin web ( Figure 2A–C ) . Importantly , apicomedial myosin formed in a manner that recapitulates the observed temporal and spatial pattern of apical constriction . Before invagination , myosin predominantly localized at the cortical regions in all SG cells , partially colocalizing with E-Cad and often showing higher intensity at vertices ( hereafter referred to as junctional myosin; Figure 2A ) . When invagination began , moderate levels of apicomedial myosin were observed in cells near where invagination initiates ( Figure 2B ) . During deep invagination , apicomedial myosin was quite prominent in the cells anterior to the invagination pit ( Figure 2C ) . Indeed , the measured intensity ratios of junctional to apicomedial myosin decreased over time in the WT SG cells ( Figure 2E ) . In contrast , junctional myosin was predominant throughout stage 11 in fkh H99 SGs , and only dispersed , weak myosin signals were observed in the apical region ( Figure 2D ) . Correspondingly , the ratio of junctional to apicomedial myosin in fkh H99 SGs was comparable to that of WT before invagination ( Figure 2E ) . 10 . 7554/eLife . 22235 . 005Figure 2 . Apicomedial myosin accumulation and coordinated apical constriction are regulated both spatially and temporally during SG invagination . ( A–D ) Myosin accumulation at different stages of invagination of WT ( A–C ) and in the fkh mutant SGs ( D ) . Two focal planes ( apical and junctional ) for the posterior/dorsal region of each placode ( yellow box ) are shown in higher magnification . Z sections along the yellow lines are shown at the bottom . Before invagination ( A ) , most myosin is found along cell junctions , often with higher intensity at vertices ( yellow arrowheads ) . Only very weak and dispersed myosin is observed apically ( red arrowheads ) . When the first cells begin to invaginate ( B ) , web-like myosin structures become prominent in the apical region of cells ( red arrowheads ) . Strong myosin signals at cell junctions are still observed ( yellow arrowheads ) . Supracellular myosin cables along the anterior and posterior boundaries of the SG are also observed ( yellow arrows ) . During deep invagination ( C ) , high intensity apicomedial myosin web structures are observed in cells near the invagination pit ( red arrowheads ) . Note that the epithelial sheet is tilted a little basally toward the invagination pit ( asterisk ) in the magnified images . At this stage , large supracellular cables form at the dorsal boundary of the tissue ( yellow arrows ) to connect the lateral cables and surround the entire tissue . Short intercellular myosin cables across several cells are also occasionally observed ( red arrows ) . In fkh mutant SGs ( D ) , strong myosin signals are observed only along junctions , even at late stage 11 ( yellow arrowheads ) . Apical myosin is weak and dispersed ( red arrowheads ) . The supracellular myosin cables along the lateral boundaries of the tissue are still visible ( arrows ) , but a connected dorsal cable does not form . ( E ) Ratio of junctional to apicomedial myosin signals of the SG cells . Shown are mean ± SEM . P values are calculated using the two-tailed Student’s t-test . See Figure 2—source data 1 . ( F ) Time-lapse images of sqh-GFP in a single WT SG cell show pulsatile behavior of apicomedial myosin ( red arrowheads ) . Cell deformation is occasionally observed during the peak intensity period of apicomedial myosin ( yellow arrowheads ) . See Video 1 . ( G ) Whereas cells have a roughly hexagonal shape before invagination ( G ) , significant cell membrane distortion is observed during invagination ( G’ ) where apicomedial myosin contacts E-Cad ( arrowheads ) . ( H ) Weak actin signals colocalize with apicomedial myosin ( red arrowheads ) . Strong actin signals colocalize with myosin at cell junctions ( yellow arrowheads ) . ( I ) Heat maps corresponding to the late stage 11 WT SG shown in ( C ) . Higher intensity myosin signals are observed in cells near the invagination pit ( I ) , which have smaller apical area ( I’ ) . ( J ) Negative correlation between myosin intensity and apical area during deep invagination . Myosin intensity for each SG is re-scaled for 0 to 100 ( a . u . ) . Cells from six WT SGs are plotted with different colors . Trendlines are shown for each SG . R , Pearson correlation coefficient . p<0 . 0001 for all samples . Asterisks in B , C , H and I , invagination pit . DOI: http://dx . doi . org/10 . 7554/eLife . 22235 . 00510 . 7554/eLife . 22235 . 006Figure 2—source data 1 . Ratio of junctional to apicomedial myosin signals of the SG cells . DOI: http://dx . doi . org/10 . 7554/eLife . 22235 . 006 To investigate the dynamics of apicomedial myosin in the SG , we took time-lapse images of sqh-GFP . Apicomedial myosin in the SG was also pulsatile , with an average interval of 131 . 7s ± 42 . 8 s between pulses ( mean ± s . d . , n = 14 cells in 4 SGs , 20 pulses; Figure 2F; Video 1 ) . Pulses were mostly asynchronous between adjacent cells . The apicomedial myosin colocalized with F-actin , indicating formation of an actomyosin complex ( Figure 2H; Röper , 2012 ) . Importantly , we observed cell distortions during the peak intensity of pulses where the apicomedial myosin contacts E-Cad ( Figure 2F and G' ) , reminiscent of the transient bending of AJs in constricting mesodermal cells ( Martin et al . , 2009 ) , suggesting that actomyosin contraction generates a pulling force during SG invagination . Measurements of average myosin intensity in the apical domain of each cell showed a negative correlation with the apical area of cells in the entire SG ( Figure 2I , I’ and J ) . This correlation , combined with the relative increase in apicomedial myosin in cells anterior to the invagination pit , suggests a role for this myosin pool in the clustered apical constriction observed during SG invagination . 10 . 7554/eLife . 22235 . 007Video 1 . Pulsatile apicomedial myosin , Related to Figure 2 . Time-lapse movie of sqh-GFP in a WT SG during invagination; a single confocal section is shown . Signals in a single SG cell are shown in the right panel . Frames are 5 s apart . DOI: http://dx . doi . org/10 . 7554/eLife . 22235 . 007 Rho kinase ( Rok ) is a key regulator for myosin phosphorylation and activation ( Amano et al . , 1996 ) . During ventral furrow formation , Rok is polarized to an apicomedial domain of mesodermal cells , where it promotes assembly of apicomedial myosin ( Mason et al . , 2013 ) . Therefore , we analyzed Rok localization dynamics using a ubiquitously expressed GFP-tagged Rok transgene ( Rok-GFP; Abreu-Blanco et al . , 2014 ) . Rok localization recapitulated the temporal and spatial distribution of apicomedial myosin . Rok was initially detected as occasional small punctate structures in the apical domain of all SG cells ( Figure 3A ) , but as invagination proceeded , gradually larger apical puncta were observed in the posterior region of the SG ( Figure 3B ) . During deep invagination , huge globular structures of Rok occupied nearly the entire apicomedial domain ( Figure 3C ) . Live imaging revealed congregating and separating behaviors , suggesting that the large globular structures are dynamic ( Figure 3D; Video 2 ) . Importantly , only small punctate signals were detected in the apical region of fkh H99 SG cells throughout all of stage 11 ( Figure 3E ) , indicating that Fkh is required for apicomedial Rok accumulation . In embryos treated with the Rok inhibitor Y-27632 , myosin was detected only along the lateral membrane and in the basal region of the cells , suggesting that Rok activity is required for the formation of apicomedial myosin ( Figure 3F ) . Furthermore , the SGs of Y-27632-treated embryos did not invaginate , suggesting a critical role for Rok . 10 . 7554/eLife . 22235 . 008Figure 3 . Spatiotemporal regulation of Rok is critical for formation of apicomedial myosin during SG invagination . ( A–C ) Apicomedial Rok increases dramatically during SG invagination and forms huge globular structures . Two focal planes ( apical and junctional ) for the posterior/dorsal region of each placode ( yellow box ) are shown at higher magnification . Bottom panels for each time point are the Z sections along the yellow lines . Before invagination ( A ) , Rok is observed only as small puncta in the apical region ( arrowheads ) . Additional Rok signals are shown along the entire lateral membranes . When cells first begin to invaginate ( B ) , Rok is observed in large punctate structures in the apical region of the posterior/dorsal region of the placode ( arrowheads ) . During deep invagination ( C ) , huge globs of Rok accumulation are observed in cells near the invagination pit ( arrowheads ) . Asterisk , invagination pit . ( D ) Time-lapse images of Rok-GFP in a single WT SG cell show dynamic apicomedial Rok accumulation . See Video 2 . ( E ) In late stage 11 fkh mutants , apical Rok is present only in small punctate structures ( arrowheads ) . ( F ) Y-27632 inhibits formation of apicomedial myosin and SG invagination . Myosin is only observed along the lateral membrane , including the AJ domain ( yellow arrowheads ) and in the basal region of the cells ( red arrowheads ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22235 . 00810 . 7554/eLife . 22235 . 009Video 2 . Rok is observed as huge globular structures in the invaginating SGs , Related to Figure 3 . Time-lapse movie of Rok-GFP in a WT SG during invagination; a single confocal section is shown . Signals in a single SG cell are shown in the bottom panel . Frames are 20 s apart . DOI: http://dx . doi . org/10 . 7554/eLife . 22235 . 009 We next asked which effector molecule acts downstream of Fkh to regulate apicomedial accumulation of Rok and myosin in SG cells . A good candidate was Fog , a secreted ligand known to regulate apical constriction of cells in several tissues ( Costa et al . , 1994; Dawes-Hoang et al . , 2005; Nikolaidou and Barrett , 2004 ) . Fluorescent in situ hybridization experiments revealed that fog mRNA is upregulated in the SG prior to and during invagination ( Figure 4A; Nikolaidou and Barrett , 2004 ) and persists until at least stage 14 ( data not shown ) . Importantly , SG expression of fog requires Fkh; only background levels of fog mRNA were observed in fkh H99 SGs ( Figure 4A’ ) . 10 . 7554/eLife . 22235 . 010Figure 4 . Fog , a downstream effector of Fkh , is essential for proper Rok localization , apicomedial myosin formation and coordinated apical constriction . ( A ) fog mRNA ( magenta ) is expressed in the SG ( red arrowheads in A ) , overlapping with the SG-specific marker CrebA ( green ) . In fkh mutants , fog mRNA in the SG is at background levels ( red arrowheads in A’ ) whereas fog expression in the developing trachea is unaffected ( yellow arrowheads ) . ( B , C ) E-Cad staining of WT and fog SGs during invagination ( B , C ) and the corresponding heat maps of apical area ( B’ , C’ ) . ( D ) Percentage and cumulative percentage of cells of different apical area in WT and fog mutant SGs are indistinguishable during invagination . See Figure 4—source data 1 . ( E ) Scatter plots showing the position of cells relative to the invagination pit . X and Y axes represent the distance along the A/P axis and D/V axis from the pit , respectively . Cells of lower 10% , 20% , 30% of apical area ( blue ) and all cells ( red ) are plotted . Note that the cells of lower 30% of area are plotted on top of all cells in the merged plots ( right-most panels ) . The same cells quantified in D were analyzed in E and F . ( F ) Quantification of distribution of cells along the A/P and D/V axis . Compared to WT , cells with small apical area are more dispersed along the D/V axis in fog mutant SGs . See Figure 4—source data 2 . ( G ) Only weak apicomedial myosin structures form in fog mutant SGs during invagination ( red arrowheads ) . Yellow arrowheads , junctional myosin . ( H ) Rok is more dispersed at the apical region and is observed only in small punctate structures ( arrowheads ) in fog mutant SGs . ( I ) The ratio of junctional to apicomedial myosin is significantly higher in fog mutants during deep invagination . Shown are mean ± SEM . P values are calculated using the two-tailed Student’s t-test . See Figure 4—source data 3 . Asterisks in B , C , G and H , invagination pit . DOI: http://dx . doi . org/10 . 7554/eLife . 22235 . 01010 . 7554/eLife . 22235 . 011Figure 4—source data 1 . SG cells quantified for apical area . DOI: http://dx . doi . org/10 . 7554/eLife . 22235 . 01110 . 7554/eLife . 22235 . 012Figure 4—source data 2 . Distribution of SG cells along the A/P and D/V axis . DOI: http://dx . doi . org/10 . 7554/eLife . 22235 . 01210 . 7554/eLife . 22235 . 013Figure 4—source data 3 . Ratio of junctional to apicomedial myosin signals of the SG cells . DOI: http://dx . doi . org/10 . 7554/eLife . 22235 . 013 Unlike fkh mutants , fog mutant SGs invaginate . fog mutant SGs formed an invagination pit at a relatively normal position at about the same stage as WT , although the pit was often somewhat larger ( Figure 4C ) . Cell segmentation analyses revealed that the percentage and cumulative percentage of cells of different apical area did not show significant differences between fog mutant and WT SGs ( Figure 4D; Figure 4—source data 3 ) . Cells with smaller apical area , however , showed a less coordinated spatial distribution in fog mutants . Unlike in WT SGs , where apically constricted cells were tightly clustered anterior to the invagination pit ( Figure 4B and B’ ) , in fog mutant SGs , apically constricted cells were dispersed , with significantly increased dispersion along the dorsal/ventral ( D/V ) axis of the tissue ( Figure 4C , C’ , E and F ) . Consistent with the uncoordinated apical constriction in fog mutant SGs , analysis of myosin signals revealed less organized web-like structures with lower staining intensity than with age-matched WT samples ( Figure 4G; compare to Figure 2C ) . The ratio of intensity between junctional and apicomedial myosin in fog mutants during deep invagination was more like that of WT glands at earlier stages - between the value measured at the before invagination and at the beginning of invagination stages ( Figure 4I; compare to Figure 2E ) . We then asked if the apicomedial accumulation of Rok is affected by loss of fog . Although apicomedial Rok has been reported during ventral furrow formation ( Mason et al . , 2013 ) , its dependence on fog has not been addressed . In fog mutant SGs , we observed Rok in small punctate structures dispersed along the apical domain during deep invagination , rather than in the huge globular structures observed in WT SG cells at this stage ( Figure 4H; compare to Figure 3C ) . These data indicate that reduced Rok accumulation in the apicomedial region of fog mutant SG cells leads to reduced apicomedial myosin . Since the loss of fog only affected the pattern of apical constriction , we next sought to completely block apical constriction . Overexpression of Crb , which expands the apical domain when overexpressed ( Wodarz et al . , 1995 ) , or a constitutively-active form of Diaphanous ( Dia-CA; Somogyi and Rørth , 2004 ) , a fly formin protein that nucleates and facilitates the elongation of actin filaments ( Higgs and Peterson , 2005 ) , from the onset of SG specification throughout development efficiently blocked apical constriction; clustered apical constriction was not observed at any stage of invagination ( Figure 5A–H , A’−H’ ) . Moreover , analyses of the percentage and cumulative percentage of cells of different apical area showed that Crb or Dia-CA overexpression not only blocked apical constriction but also caused significant increase in apical area . Before invagination , Crb-overexpressing cells showed apical areas comparable to those of WT cells at early stage 11 , and Dia-CA-overexpressing cells showed a small ( but significant ) increase of apical area ( Figure 5I; Figure 5—source data 1 ) . At the beginning of invagination , cells showed a notable increase in apical area , with obvious shifts in the distribution of apical area when compared to WT ( Figure 5I; Figure 5—source data 1 ) . Importantly , the percentage of cells in different apical area bins of Crb- or Dia-CA-overexpressing SGs was comparable to that of WT SG cells before invagination , confirming that the cells do not undergo apical constriction ( Figure 5I; Figure 5—source data 1 ) . During deep invagination , Dia-CA-overexpressing SG cells still have apical areas comparable to those of WT SG cells before invagination , but overexpression of Crb caused a huge increase in apical area ( Figure 5I; Figure 5—source data 1 ) . Importantly , however , neither blocking apical constriction nor increasing apical area arrested SG invagination . SGs overexpressing Crb or Dia-CA formed invagination pits without any delays . Plotting of cells of lower 10% , 20% and 30% of apical area showed that cells with smaller apical area were not as clustered as in WT , with significantly increased dispersion along the anterior/posterior ( A/P ) axis of the tissue ( Figure 5—figure supplement 1A and B ) . These data indicate that overexpression of Crb or Dia-CA effectively blocks apical constriction and alters the distribution of cells with smaller apical areas . 10 . 7554/eLife . 22235 . 014Figure 5 . Blocking apical constriction does not prevent SG internalization . ( A–H ) Representative confocal images for Crb- ( A–D ) and Dia-CA-overexpressing SGs ( E–H ) at each stage of invagination and the corresponding heat maps for apical area ( A’–H’ ) . Red and white lines mark the SGs . Asterisks , invagination pit . ( I ) Percentage and cumulative percentages of cells with different apical area show a gradual and significant increase of apical area for Crb- and Dia-CA-overexpressing SG cells . P values are calculated using the Mann-Whitney U test ( percentage of cells ) and using the Kolmogorov-Smirnov test ( cumulative percentage of cells ) . See also Figure 5—source data 1 and Figure 5—figure supplement 1 . ( J , K ) In Crb-overexpressing cells , Rok is observed as small- to medium-sized puncta ( yellow arrowheads in J ) . Strong apicomedial myosin ( green arrowheads in J ) colocalizes with F-actin ( magenta arrowheads in K ) . Junctional myosin ( white arrowheads in K ) is also clearly shown . Occasional cell deformation is observed at the contact site of apicomedial actomyosin complex and the junction ( arrows ) . ( L , M ) Only small punctate Rok signals are observed in Dia-CA-overexpressing cells ( yellow arrowheads in L ) . Myosin is observed in punctate structures in a broad cortical area ( green arrowheads in M ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22235 . 01410 . 7554/eLife . 22235 . 015Figure 5—source data 1 . SG cells quantified for apical area . DOI: http://dx . doi . org/10 . 7554/eLife . 22235 . 01510 . 7554/eLife . 22235 . 016Figure 5—figure supplement 1 . Overexpression of Crb- and Dia-CA affects the distribution of cells with smaller apical areas . ( A ) Scatter plots showing the position of cells relative to the invagination pit . The same cells quantified in Figure 5I were analyzed . X and Y axes represent the distance along the A/P axis and D/V axis from the pit , respectively . Cells of lower 10% , 20% , 30% of apical area ( blue ) and all cells ( red ) are plotted . Note the apical area of cells in the 10% range with overexpression of Crb or Dia-CA is larger than in either WT or fog mutant SGs up through the 30% range ( Compare to Figure 4E ) . ( B ) Quantification of distribution of cells along the A/P and D/V axis ( the same cells analyzed in A ) . Compared to WT , cells overexpressing Crb or Dia-CA with relatively smaller apical area are more dispersed along the A/P axis . DOI: http://dx . doi . org/10 . 7554/eLife . 22235 . 016 To understand how apical constriction is blocked in Crb- and Dia-CA-overexpressing SGs , we asked if the cellular machinery required for apical constriction was affected . Even during deep invagination , Rok signals were detected only as small- to medium-sized puncta ( in Crb-overexpressing SG cells; Figure 5J ) or as very small puncta ( in Dia-CA-overexpressing SG cells; Figure 5L ) rather than the huge globular structures detected in WT cells ( Figure 3C ) . However , Crb-overexpressing SGs formed relatively normal pools of apicomedial myosin that are associated with actin and with membrane distortions ( Figure 5K ) , consistent with our findings that in fog mutants even dispersed Rok can still result in some apicomedial myosin formation ( Figure 4F and G ) . These data suggest that in Crb-overexpressing SG cells , the forces of apical expansion overwhelm those of apical constriction despite the actomyosin machinery being in place and functional . Interestingly , in Dia-CA-overexpressing SG cells , most myosin signals were observed in a broad cortical area partially overlapping with actin , but no web-like apicomedial myosin formed ( Figure 5M; compare to Figure 2C ) . A similar increase in cortical myosin was observed in Dia-CA-overexpressing amnioserosa cells , but interestingly , premature apical constriction was observed in those cells ( Homem and Peifer , 2008 ) , suggesting possible tissue-specific effects . Moreover , the significant ‘wiggliness’ of the AJs in Dia-CA-overexpressing SGs suggests a decrease of cortical tension ( Figure 5E , G , H and L; Blanchard et al . , 2010; Choi et al . , 2016; Lecuit and Lenne , 2007; Martin et al . , 2009 ) . Overall , since SG invagination occurs despite the failure in apical constriction with overexpression of either Crb or Dia-CA , we conclude that apical constriction is not required . Although invagination pits formed at approximately the right position and at the right time in the SGs with either uncoordinated ( in fog mutants ) or completely blocked apical constriction ( in Crb- or Dia-CA-overexpressing SGs ) , the pits were abnormal . Specifically , both the fog mutant and Crb-overexpressing SGs frequently formed a pit that was elongated along the D/V axis , rather than the smaller oval-shaped pit of WT SGs ( Figure 6A–C , A'–C' ) . Overexpression of Dia-CA resulted in a huge invagination pit , often larger than half the size of the tissue ( Figure 6D and D' ) . We asked how this aberrant invagination and the continuous increases in apical area affected tube structure . 3D reconstruction of invaginating SGs ( late stage 11 ) revealed a wider lumen in fog mutant and Crb- and Dia-CA-overexpressing SGs compared to WT , indicating that early tube architecture is affected by defective apical constriction and consequent abnormal invagination ( Figure 6E–6H ) . We next analyzed stage 14 embryos , when in WT , all SG cells have fully internalized ( Figure 6I ) . Interestingly , Dia-CA-overexpressing SGs showed a significant delay in cell internalization and migration . All Dia-CA-overexpressing cells eventually internalized , but formed a short gland with increased luminal diameter that remained close to the embryo surface ( Figure 6J ) . Overproduced Crb not only localized to the apical membranes , but also mislocalized to the basal membranes facing outside of the tube; nonetheless , SG cells still formed a single-layered fully internalized tube at stage 14 ( Figure 6K ) . Although Crb overexpression did not delay SG internalization or migration , partially or completely externalized SGs with no clear internal lumen were occasionally observed at stage 15 and 16 ( Figure 6L ) . Since all SGs were internalized at earlier stages , these findings suggest that the Crb-overexpressing SG cells subsequently evaginated from the inside to the outside of the embryo , likely because of continued apical expansion . Similar invagination-followed-by-evagination phenotypes were observed when a myristoylated , membrane-tethered form of Wiskott–Aldrich syndrome protein ( Myr-Wasp ) , an activator of the actin nucleating Arp2/3 complex , was overexpressed in the SGs ( Figure 6M; Video 3 ) . Overall , these data suggest that the major role of apical constriction is to ensure the proper tissue geometry during SG invagination . 10 . 7554/eLife . 22235 . 017Figure 6 . Apical constriction is essential for tissue geometry . ( A–D , A'–D' ) Representative images showing morphology and size of the invagination pit ( dotted lines ) in the different genotypes . ( E–H ) 3D reconstruction of late stage 11 SGs stained with Crb ( green ) and CrebA ( magenta ) . Compared to a narrow lumen in a WT SG ( E ) , fog mutant ( F ) , Crb-overexpressing ( G ) and Dia-CA-overexpressing ( H ) SGs form a wider lumen . Insets , Crb signals only . ( I–L ) Fully formed late stage SGs . Dia-CA-overexpressing SGs have wider lumens and are closer to the embryo surface ( J ) . Crb-overexpressing SGs show normal SG internalization until stage 14 ( K ) , but some cells occasionally evaginate at later stages ( arrowhead in L ) . White dashed lines , embryo boundary . ( M ) Time-lapse images of Myr-Wasp-overexpressing SG show evagination behavior of SG cells . The SG was re-centered at 38:30 . Cells that were completely internalized at the beginning of the movie ( bracket ) are shown on the embryo surface over time ( arrowheads ) . Red dashed lines , embryo boundary . See Video 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 22235 . 01710 . 7554/eLife . 22235 . 018Video 3 . Myristylated-Wasp-overexpressing SGs evaginate , Related to Figure 6 . Time-lapse movie for a late stage SG overexpressing a membrane-bound form of Wasp . The sample was re-centered at 38:30 . Stacks of two confocal sections are shown . The fully internalized SG at the beginning of the movie gradually evaginates over time . Frames are 5 s apart . DOI: http://dx . doi . org/10 . 7554/eLife . 22235 . 018 The supracellular myosin cable that surrounds the entire SG is under tension ( Röper , 2012 ) , suggesting that it generates a compression force . Therefore , we asked if this tissue-scale myosin cable is present in the SGs defective for apical constriction and could contribute to invagination . At late stage 11 , fog and Crb-overexpressing SGs formed a supracellular myosin cable almost indistinguishable from that of WT ( Figure 7A– , A'–C' ) . However , this cable was defective in both Dia-CA-overexpressing SGs and in fkh H99 mutants , where invagination is either delayed or completely fails , respectively . In Dia-CA-overexpressing SGs , myosin signals were observed in a broad cortical area of individual cells ( Figure 5M ) and no obvious cable structure surrounding the tissue was observed ( Figure 7D and D' ) . In fkh H99 mutants , myosin cables were observed along the lateral boundaries of the tissue , coinciding with the parasegmental boundaries , but the connecting dorsal cable was not observed ( Figure 7E and E' ) . Similar myosin cables are observed in the lateral boundaries of Scr mutants ( Röper , 2012 ) , which do not form SGs ( Andrew et al . , 1994 ) . 10 . 7554/eLife . 22235 . 019Figure 7 . The tissue-level supracellular cable is defective in the SGs with defective invagination . ( A–E ) 3D reconstruction of late stage 11 SGs stained with Crb ( green ) , CrebA ( magenta in A , B and D and green in C and E ) , and sqh-GFP ( white ) . A supracellular myosin cable ( red arrowheads ) that surrounds the entire SG placode is observed in the WT ( A ) , fog mutant ( B ) and Crb-overexpressing SGs ( C ) . The tissue-level myosin cable is not obvious in the Dia-CA-overexpressing SGs ( D ) . Only the supracellular cables along the lateral boundaries of the placode are shown in the fkh H99 mutant SGs ( cyan arrowheads in E ) . ( A’–E’ ) sqh-GFP signals only . ( F , G ) Analysis of the circularity of the SG placode boundary as a measure of smoothness and tension . The circularity of the placode boundary in the WT , fog , and Crb-overexpressing SGs , where the cable is located , is significantly greater than that of the boundaries shifted outside the placode by one cell ( outer boundary ) or inside the placode by one cell ( inner boundary ) . Numbers of the SGs measured are shown below each genotype . mean ± SEM; P values are calculated using the unpaired t-test . *p<0 . 05; **p<0 . 01; ***p<0 . 001 . ( G ) An example of a WT SG with the boundaries measured . ( H , I ) Compared to moderate levels at early stage 11 ( H , H’ ) , Crb levels ( magenta ) notably increase in the SG cells at late stage 11 ( bracket ) . Anisotropic Crb localization ( yellow arrowheads ) correlates with supracellular myosin cable at the SG boundary ( red arrowheads ) . ( J , K ) In fkh H99 mutant SGs , Crb levels in the SG are reduced throughout stage 11 . Weak Crb signals do not show anisotropic localization at the SG boundary ( yellow arrowheads ) , and the supracellular myosin cable does not form at the dorsal boundary of the SG ( red arrowheads ) . Cyan arrowheads , supracellular myosin cable along the lateral boundaries . DOI: http://dx . doi . org/10 . 7554/eLife . 22235 . 019 We calculated the circularity of the SG placode as a measure of smoothness and tension ( See Materials and methods for details ) . The circularity of the WT placode is slightly , but significantly increased from early stage 11 to late stage 11 , suggesting an increased tension over time ( in Figure 7F ) . At late stage 11 , the border of the WT placode where the supracellular cable was observed was very smooth , and circularity of the placode boundary was significantly greater than that of the boundaries shifted outside the placode by one cell ( outer boundary ) or inside the placode by one cell ( inner boundary ) , suggesting that this big cable is under tension ( Figure 7F and G ) . The circularity of the fog and Crb-overexpressing SG placode was comparable to that of WT , and was significantly greater than that of the outer and inner boundaries ( Figure 7F ) . Consistent with the lack of a supracellular cable , the circularity of the Dia-CA-overexpressing SG boundary was significantly lower than that of WT , and was not significantly different from the circularity of the inner boundary , suggesting disrupted tension at the SG boundary ( Figure 7F ) . The circularity of the fkh H99 mutant SG boundary was between that of early and late stage 11 WT , and was also comparable to the fkh H99 outer boundary , indicating loss of compressive forces on cells at and near the SG boundary ( Figure 7F ) . Anisotropic Crb distribution at the placode boundary has been suggested to drive supracellular myosin cable formation during SG formation ( Röper , 2012 ) . Consistent with this , Crb levels were higher in WT SGs than in surrounding tissues by early stage 11 and further increased by late stage 11 , with overt anisotropic localization of Crb in SG cells at the dorsal boundary ( Figure 7H , H' , I , I' ) . In these cells , myosin levels were very high at the boundary , where Crb levels were lowest ( inset ) . Importantly , Crb levels were lower in fkh H99 SGs than in WT throughout stage 11 and did not show anisotropy in boundary cells ( Figure 7J , J' , K , K' ) . Also , unlike in WT SG cells , myosin levels were never high at the dorsal boundary ( Figure 7H’–K’ ) . Taken together , we conclude that the compressing force generated by the tissue-level supracellular myosin cable , which fails to form in fkh mutants where SG levels of Crb are low , contributes to SG invagination . A range of SG defects was observed in late stage fog mutants , with 34% of SGs either partially or fully externalized ( Figure 8A–E , A'–E' and K ) . Given that loss of fog had relatively minor effects on apical constriction compared to overexpression of Crb or Dia-CA , and that fog mutants also form the circumferential cable surrounding the tissue , these severe tube internalization defects were unexpected . Therefore , we asked if SG expression of Fog in otherwise fog mutant embryos could rescue internalization using the UAS-Gal4 system , which has allowed the rescue of multiple SG mutant phenotypes , including loss of fkh and other transcription factors expressed in the early SG ( D . Johnson and D . J . Andrew , unpublished data ) . The range of defects observed with SG expression of Fog in otherwise fog mutant embryos was similar to that seen with complete loss of fog ( Figure 7K ) , indicating that the Fog requirement for internalization is not SG autonomous . Instead , analysis of early fog mutant embryos suggested that the separation of the SG from the surrounding epithelia causes these defects . In stage 11 fog mutants , deep ingression furrows often formed between the two tissues , and in more than 60% of embryos with such furrows ( 24/39 ) , some SG cells at the edge were tilted and even perpendicular to the rest of the cells in the SG epithelium , forming a lip-like structure ( Figure 8F–J and F'–J' ) . These cells maintained adhesion to their neighbors and were properly polarized , with their AJs connected to those of neighboring SG cells at the edge . Strong myosin signals at the level of AJs along the edge of the lip ( Figure 8H’ ) suggest that pulling or squeezing forces along the edge of the gland in combination with the pushing forces of the ingression furrow caused the SG cells to slip outside the embryo . 10 . 7554/eLife . 22235 . 020Figure 8 . Invagination decouples internalization of cells . ( A–E , A'-E' ) fog mutant SGs of late stages , from fully internalized ( A , A' ) , partially internalized ( B , B' ) to completely externalized glands , either folded ( C , C' ) or elongated ( D , D' , E , E' ) . The externalized SG that migrates along the embryo surface ( dashed bracket in E ) formed a longer tube than the internalized one ( solid bracket in E ) . All SGs have proper apicobasal polarity . ( F , G ) Whereas the surface epithelium is in the same plane as the SG in WT ( F ) , a deep ingression between the SG and the neighboring surface epithelium is often observed in fog mutants ( G ) , with some SG cells perpendicular to the rest of the SG on the apical side of the epithelial sheet ( arrowheads ) . Z sections along the lines ( F’ , G’ ) . Two SG cells are outlined with white dotted lines for each sample; one of them is perpendicularly positioned at the edge in the fog mutant ( arrowhead ) . ML , midline . Yellow dashed lines , the boundary of the neighboring surface epithelia . ( H ) Strong myosin signals are often observed along the edge of the fog mutant SG ( arrows ) . Arrowheads , nuclei of the slipped-out cells . ( I , J ) fog mutant SGs with slipped-out cells ( arrowheads ) in the anterior region of the placode form an externalized tube as cells invaginate ( I ) . fog mutant SGs with slipped-out cells in the posterior region of the placode form an externalized tube as cells invaginate ( J ) . Z sections along the lines ( I’ , J’ ) . White lines , the boundary of SGs . Green lines , embryo surface . Arrows , the direction of invagination . Asterisks , non-specific signals . ( K ) SG-specific expression of Fog in fog mutant embryos did not rescue the externalized SG phenotype . Numbers inside bars indicate the number of SGs counted . P value was calculated using the Chi-square test . ( L ) A model for SG invagination . Fkh-dependent high-level Crb expression in the SG versus low levels of Crb in the surrounding ectodermal tissues regulates formation of the circumferential myosin cable that surrounds the entire SG to provide a compressing force during invagination . Regulation of the SG expression of fog by Fkh controls apicomedial localization of Rok and formation of apicomedial myosin , which drives clustered apical constriction to ensure the proper tissue geometry . Fkh-dependent uncharacterized target genes are proposed to drive the cell rearrangement that also contributes to SG internalization . DOI: http://dx . doi . org/10 . 7554/eLife . 22235 . 020 We propose that if the ingression furrow formed at or near the posterior end of the primordia ( where invagination always initiated ) , then the resulting glands would slip outside and form fully externalized tubes ( Figure 8J and J' ) . However , if the ingression furrows formed elsewhere , the resulting glands would either fully or partially internalize , depending on the extent of ingression between the SG primordia and surface epithelium ( Figure 8I and I' ) . Importantly , all fog mutant SGs ( whether internalized , externalized , elongated , folded or twisted ) showed intact , closed and properly polarized tubes with the apical membrane facing an internal lumen ( Figure 8A–E , A'–E' ) . The completely externalized SGs that elongated along the embryo surface often formed longer and narrower tubes than internalized SGs ( Figure 8E and E' ) , suggesting that , although the SG does not require contact with internal tissues to elongate , both tube diameter and length may be regulated through contact with internal tissues . Altogether , these findings reveal that SG tube formation occurs in a tissue-autonomous manner and can be decoupled from internalization .
Fog has been implicated in invagination of several embryonic tissues , including the ventral furrow , the posterior midgut and the SG , and also functions in imaginal disc folding during larval development ( Manning and Rogers , 2014; this study ) . In the ventral furrow , Twist and Snail regulate expression of the Fog ligand and its receptor , respectively ( Manning et al . , 2013; Seher et al . , 2007 ) . Likewise , Caudal , through activation of additional downstream transcription factors , is required for fog expression in the posterior midgut ( Costa et al . , 1994; Wu and Lengyel , 1998 ) . Here , we demonstrate that Fkh regulates SG expression of fog to coordinate apical constriction of SG cells ( Figure 4 ) . Therefore , at least during embryogenesis , the tissue-specificity of Fog signaling is determined by the tissue-specific transcription factor ( s ) that activate expression of fog ( and its receptors ) . The downstream pathway components are both maternally deposited and widely expressed , and are thus likely to be shared in all cell/tissue types where Fog functions . Indeed , mutations in RhoGEF2 , a key component of the Fog signaling pathway in the mesoderm , cause SG invagination defects ( Nikolaidou and Barrett , 2004 ) . Due to their ubiquitous expression , localized activation of the downstream components , rather than their localized expression , is likely to determine their site of action . Consistent with this notion , we demonstrated that the robust apicomedial Rok accumulation in the SG cells requires both Fkh and Fog ( Figures 3 and 4 ) . An interesting question is what determines regional specificity for Fog signaling within a tissue . Levels of fog transcripts are relatively uniform across the primordia , yet both Rok accumulation and apicomedial myosin accumulation occur only in a temporally and spatially regulated subset of SG cells . We propose that production and apical release of Fog and/or the apical localization of its transmembrane receptors could be regulated post-translationally . Excellent candidate regulators include Huckebein ( Hkb ) , an SP1-like transcription factor , and its key target Klarsicht ( Klar ) , a putative regulator of dynein ATPase ( Mosley-Bishop et al . , 1999; Myat and Andrew , 2000b , 2002 ) . The expression of both hkb and klar is limited to the dorsal posterior domain of the SG , where we observe high Rok , high myosin , and apical constriction . Since Klar is required for the delivery of vesicles to the apical SG surface , which is enriched in minus-end microtubules ( MT ) ( Myat and Andrew , 2002 ) , a possible scenario is that during invagination , Klar-dependent apically-targeted vesicles contain Fog ligand and/or its receptor . Consistent with this model , a recent study has demonstrated a requirement for the MT cytoskeleton in the formation of apicomedial myosin ( Booth et al . , 2014 ) . Moreover , unlike WT , hkb mutant SGs invaginate from the center of the SG placode ( Myat and Andrew , 2002 ) . During ventral furrow formation , where apical constriction plays a major role in tissue invagination ( Guglielmi et al . , 2015 ) , myosin is absent from the cortex and accumulates in only the apicomedial region . Since AJ-associated cortical myosin has been linked to junctional remodeling and cell rearrangement in the intercalating ectodermal cells of the Drosophila embryo ( Bertet et al . , 2004; Blankenship et al . , 2006; Irvine and Wieschaus , 1994; Rauzi et al . , 2008; Zallen and Wieschaus , 2004 ) , one might expect that cell rearrangement does not take place during ventral mesoderm formation . Indeed , ventral furrow cells invaginate without changing their position relative to each other; cells constrict predominantly in the ventral/lateral direction , remaining longer along the A/P axis to form a long , narrow furrow along the axis ( Martin et al . , 2010; Sweeton et al . , 1991 ) . We show that apical constriction is not required for SG invagination per se , but is needed to acquire proper geometry of the invagination pit and the tube that forms from it ( Figures 5 and 6 ) . In the SG , where cells sequentially invaginate through a relatively small invagination pit to make a tube , combinatorial forces that drive different morphological changes are expected to contribute to tissue invagination ( Figure 8L ) . Correspondingly , in tracheal invagination , at least three independent processes have been implicated: EGF-dependent myosin accumulation ( Brodu and Casanova , 2006; Nishimura et al . , 2007 ) , an oriented final cell division ( Kondo and Hayashi , 2013 ) , and FGF-mediated migratory forces . Only when all three processes are disrupted , do tracheal cells fail to internalize ( Kondo and Hayashi , 2013 ) . However , since no cell divisions occur once the SG has been specified ( Chung et al . , 2014 ) , oriented cell divisions cannot contribute . Similarly , our finding that the non-autonomous functions of fog can lead to fully formed external SGs ( Figure 8 ) reduces the likelihood that signaling between the SG and neighboring tissues is critical . We show at least one additional force contributes to tissue invagination when apical constriction is blocked: the supracellular myosin cable that is under tension and sequentially closes in as SG cells internalize . This circumferential myosin cable forms normally in both fog mutant SGs and in Crb-overexpressing SGs , where invagination occurs relatively normally , but fails to form in fkh mutant SGs and in Dia-overexpressing SGs , where invagination either fails or is delayed ( Figure 7 ) . These findings support the idea that tension generated by the myosin cable contributes to and facilitates invagination ( Figure 7 ) . This closing process in the SG is reminiscent of the ‘purse-strings’ observed during wound healing in both flies and vertebrates ( Martin and Parkhurst , 2004 ) and also during Drosophila dorsal closure ( Franke et al . , 2005 ) . It is notable that multiple forces have been suggested to contribute to dorsal closure; dorsal closure fails only when both the forces of apical constriction of amnioserosa cells and of purse strings from surrounding epidermal cells are ablated ( Kiehart et al . , 2000 ) , although this model of a combinatorial force-component system has been challenged by recent studies showing that the epidermal actin cable tension may not contribute to dorsal closure ( Pasakarnis et al . , 2016 ) . The difference in invagination defects in Dia-CA-overexpressing SGs ( delayed ) and fkh H99 SGs ( completely failed ) , where both apical constriction and the circumferential myosin cable are absent , indicate that additional other forces contribute to SG invagination . We propose that cell rearrangement linked to junctional myosin may also be important . A strong junctional myosin pool is observed throughout SG invagination , with high levels at cell vertices ( Figure 2; Röper , 2012 ) . We have direct and indirect evidence for cell rearrangement in early WT SGs from live imaging and measurements of cell topology ( S . Chung , S . Kim and D . J . Andrew , unpublished data ) , suggesting that this process is linked to tissue invagination . Since myosin localizes in a broader cortical region in Dia-CA-overexpressing SGs ( Figure 5 ) , it is possible that cell rearrangement of Dia-CA-overexpressing cells is somewhat affected , explaining the delayed invagination and abnormal tissue morphology observed early ( Figure 6D and H ) . Nonetheless , cell rearrangement must still occur in the Dia-CA overexpressing glands since they form elongated tubes with an approximately WT arrangement of cells at later stages ( Figure 6J ) . The complete failure of invagination in fkh H99 SGs raises the possibility of failed cell rearrangement . Interestingly , changes in cell topology occur independently of fkh function . WT and fkh H99 mutant SGs show a similar distribution of n-sided cells during early stages of invagination ( S . Chung and D . J . Andrew , unpublished data ) . Consistent with the idea that cortical enrichment of myosin has a role in these topological changes , junctional myosin is intact in fkh H99 SGs ( Figure 2 ) . In any case , more direct measurements of cell rearrangement in both WT and fkh mutants will be necessary to learn how directed cell rearrangement contributes to SG internalization . Overall , our findings suggest that multiple forces are required for SG invagination and that the system is remarkably robust . Importantly , our finding that polarized and fully elongated tubes can form outside the embryo indicates that SG morphogenesis is largely a tissue-autonomous process .
fkh6 H99 recombinants ( RRID:DGGR_130500; RRID:BDSC_1576; Myat and Andrew , 2000a ) were used for the analysis of fkh mutant SGs . UAS-Crb ( RRID:DGGR_108289; Wodarz et al . , 1995 ) , UAS-HA-Dia-CA ( RRID:BDSC_27616; Somogyi and Rørth , 2004 ) and UAS-Myr-Wasp-GFP ( C-terminal GFP tag added to construct described in FBal0243602 ) driven by fkh-Gal4 ( FBtp0013253/FBti0027904; Henderson and Andrew , 2000 ) were used . fkh-Gal4 driving UAS-Tmem-GFP ( Ftp0011013 [aka UAS-GFP-CAAX]; Kakihara et al . , 2008 ) was used as a control . Two null alleles for fog ( fogS4 and fogRA67; RRID:DGGR_106665; RRID:BDSC_6218 ) were analyzed; both resulted in indistinguishable phenotypes . UAS-fog ( FBtp0021363; Dawes-Hoang et al . , 2005 ) driven by fkh-Gal4 in fogRA67 embryos at 25°C was used for SG-specific rescue experiment of fog . sqhAX3; sqh-GFP ( RRID:BDSC_57144; Royou et al . , 2004 ) and sqh-Rok-GFP ( RRID:BDSC_52289; Abreu-Blanco et al . , 2014 ) were used to analyze myosin and Rok accumulation , respectively . For most samples , dechorionated embryos were fixed in 1:1 heptane:formaldehyde for 40 min and devitellinized with 80% EtOH . For phalloidin staining , embryos were hand-devitellinized . Antibodies used include: mouse α-Crb ( RRID:AB_528181; DSHB; 1:10 ) , rabbit α-CrebA ( RRID:AB_10805295; Fox et al . , 2010; 1:3000 ) , chicken α-GFP ( RRID:AB_300798; Abcam; 1:1000 ) , rat α-E-Cad ( AB_528120; DSHB; 1:10 ) , rabbit α-Dia ( Afshar et al . , 2000; 1: 5000 ) , mouse α-HA ( RRID:AB_514506; Roche; 1:50 ) , guinea pig α-Sage ( RRID:AB_2632603; Fox et al . , 2013 ) ; 1:100 ) , rabbit α-Fkh ( RRID:AB_2632955; Fox et al . , 2013; 1:2000 ) , rabbit α-β-Gal ( AB_788167; Novus; 1:1000 ) . Alexa fluor 488 , 555 , 647-labelled secondary antibodies were used at 1:200 ( Molecular probes: RRID:AB_142924; RRID:AB_141822; RRID:AB_142754; RRID:AB_141693; RRID:AB_141373; RRID:AB_141778; RRID:AB_143165; RRID:AB_162543 ) . Alexa fluor-546-labeled phalloidin was used for F-actin labelling ( Molecular Probes: RRID_AB2632953; 1:250 ) . Fluorescent in situ hybridization with an RNA probe specific for fog and antibody staining with anti-Sage was performed as described ( Knirr et al . , 1999 ) . For the in situ analysis , a PCR-labelled product of the fog ORF was made from cDNA generated by reverse transcription of total RNA using the following primers: fog-5’: ATGTCTCCGCCCAATTGTCT and fog-3’: GATGACTGAAAAGCGGCGGC . All images were taken with a Zeiss LSM 780 confocal microscope . Dechorinated embryos were fixed in 25% glutaraldehyde in cacodylate buffer and hand-devitellinized . Embryos were post-fixed in 1% OsO4 and dehydrated through an EtOH series . They were dried using Hexamethyldislazane , coated with gold palladium , and examined and photographed in a LEO/Zeiss Field-emission SEM . E-Cad/CrebA stained SGs were imaged with a Zeiss LSM 780 confocal microscope . Two or three focal planes ( 0 . 37 µm apart ) of apical domain were used to generate a maximum intensity projection ( Zeiss Zen software; Germany ) . Cell segmentation was performed along the E-Cad signals and the apical area for each cell was calculated using the Bitplane Imaris program . During ‘Before invagination’ and ‘Clustered apical constriction’ , all cells were at the surface in the same focal plane . During ‘Beginning of invagination’ , a few cells in the posterior/dorsal region of the placode were found in a basal position relative to neighboring cells ( at 0 . 37–1 . 11 µm deeper focal planes ) . ‘Deep invagination’ included samples with cells that had internalized >2 µm . Frequency distribution of cells with a different apical area ( bin width = 4 ) was performed using a GraphPad Prism program . For median scaling normalization of fkh H99 SG cells ( Figure 1K ) , the apical area of each cell was multiplied by 1 . 2 . The x and y coordinates for each cell , which correspond to the position along the A/P and D/V axis of the tissue , respectively , were obtained using Imaris . The position of SG cells relative to the center of the invagination pit was calculated and plotted . Standard deviations for the positions along the A/P or D/V axis were compared to show dispersion of cells . Cells in the posterior/dorsal region of the placode were chosen for quantification . For all quantifications , maximum intensity projections that span the apical and the junctional region of SG cells were used ( Zeiss Zen software ) and measurements were performed with Fiji software . Regions were drawn manually along the inner or outer boundary of E-Cad signals of each cell to calculate the apical and junctional intensity . Integrated density for myosin signals was measured after background correction and mean values from three individual measurements were used . Cell segmentation for six WT SGs at deep invagination was performed as described above . Intensity mean was measured for the entire myosin signals for each segmented cell and re-scaled for 0 to 100 ( a . u . ) for plotting . Correlation ( Pearson ) and P values were calculated using the GraphPad Prism software . sqhAX3; sqh-GFP or sqh-GFP-Rok embryos were dechorinated and mounted ventral side up on a glass slide coated with heptane glue ( double-sided tape soaked in heptane ) . Spacer coverslips ( No . 1 . 5 ) were attached to prevent embryos from being squeezed . Halocarbon oil mixture 27:700 ( 1:1 ) was added on top of the embryos . A No . 1 coverslip was placed on top and sealed with nail polish . Images were taken every 5s with a 3i Marianas/Yokogawa Spinning Disk Confocal microscope to capture apicomedial myosin pulsing and every 20 s with a Zeiss LSM 780 confocal microscope to capture apicomedial Rok signals . sqhAX3; sqh-GFP embryos were collected and aged until stage 9 . Dechorinated embryos were immersed for five minutes in a 1:10 dilution of Citrasolv in water ( Citra-Solv , Danbury , CT ) to render embryos permeable ( Rand et al . , 2012 ) . Embryos were rinsed with water and PBS several times and were incubated in Y-27632 ( 500 µM in PBS ) for three hours . Embryos were fixed as described above and immuno-stained . 3D reconstruction of SGs was performed with the confocal stacks using the Imaris program . Cell boundaries were manually traced in ImageJ to quantify the length and area of the boundary corresponding either to the SG placode boundary or the boundary shifted one cell layer outside or inside the placode boundary . The ventral midline of the embryo was used as the ventral boundary of the traced circle in all cases . For fog mutants , only circularity of the SG boundary and inner boundary was measured , since the cells in the neighboring epithelia were often out of the focal plane due to the twisted phenotype of the embryos . Circularity was calculated based on the fact that for a perfect circle the circularity C = 1 = 4 π area/perimeter2 ( Lawrence et al . , 1999 ) . Circularity values were plotted in Prism and statistical significance in differences in the circularity of outer and inner boundaries compared to the circularity of the cable was analyzed using a two-tailed t test . | Many organs in the human body – like the kidneys , lungs , and salivary glands – are organized as a single layer of cells that surround a hollow tube . There are a number of ways that cells can achieve this particular arrangement . In one mechanism , a small group of cells bud out of a single cell layer to become the end of a new tube or a new branch of an existing tube . Since all the cells are still connected , the first cells bring their neighbouring cells along behind them , rearranging these cells to form the walls of a tube . In addition to changing position , the cells must change their shape to form a tube . One crucial change in cell shape is called apical constriction , and involves the side of the cell facing the inside of the tube becoming smaller than the other sides . This creates cells with a wedge-like shape that can fit together to form the curved wall of the tube , similar to shaped bricks in an archway . Apical constriction has been widely studied and is controlled by proteins that act like motors moving along protein-based filaments; however the roles of apical constriction in tube formation have not been fully explained . Using the developing salivary glands of the fruit fly Drosophila melanogaster , Chung et al . confirmed that the motor protein known as myosin II controls apical constriction during tissue invagination . Further examination showed that proteins ( called Fork Head and Fog ) activate and localize an enzyme ( Rho kinase ) to control the localized accumulation of myosin II and thereby control apical constriction . Chung et al . then showed that salivary glands could still form tubes if apical constriction was blocked , indicating that it is not an essential part of tissue invagination in this organ . However , blocking apical constriction led the tube to develop unusual shapes at intermediate stages . More work is now needed to better understand the links between apical constriction , cell rearrangement and tissue invagination . These processes are fundamental for organs to form correctly in many organisms and understanding their control could have wide-ranging impacts . A better understanding of these processes may provide insight into how the tubes can form while keeping all the cells adequately supplied with oxygen and nutrients , and into diseases that result if there are defects in the invagination process . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology",
"cell",
"biology"
] | 2017 | Uncoupling apical constriction from tissue invagination |
Dengue virus is a growing global health threat . Dengue and other flaviviruses commandeer the host cell’s RNA degradation machinery to generate the small flaviviral RNA ( sfRNA ) , a noncoding RNA that induces cytopathicity and pathogenesis . Host cell exonuclease Xrn1 likely loads on the 5′ end of viral genomic RNA and degrades processively through ∼10 kB of RNA , halting near the 3′ end of the viral RNA . The surviving RNA is the sfRNA . We interrogated the architecture of the complete Dengue 2 sfRNA , identifying five independently-folded RNA structures , two of which quantitatively confer Xrn1 resistance . We developed an assay for real-time monitoring of Xrn1 resistance that we used with mutagenesis and RNA folding experiments to show that Xrn1-resistant RNAs adopt a specific fold organized around a three-way junction . Disrupting the junction’s fold eliminates the buildup of disease-related sfRNAs in human cells infected with a flavivirus , directly linking RNA structure to sfRNA production .
Flaviviruses ( FVs ) are single-stranded , ( + ) -sense RNA viruses that include West Nile ( WNV ) , Yellow Fever ( YFV ) , Japanese Encephalitis ( JEV ) and other human disease-causing viruses ( Fields et al . , 2013 ) . Dengue ( DENV ) , the most pervasive FV , was the cause of >100 million symptomatic human infections during 2010 ( Bhatt et al . , 2013 ) . In humans , DENV infection can lead to severe hemorrhagic fever , dengue shock syndrome , and death ( Simmons and Farrar , 2012; The World Health Organization , 2014; United States Centers for Disease Control and Prevention , 2014 ) . Over 40% of the world’s population is at risk of contracting this disease , ( The World Health Organization , 2014; United States Centers for Disease Control and Prevention , 2014 ) and global trade and climate change continue to extend the habitat of the mosquito vector that spreads the virus ( Barclay , 2008; Morin et al . , 2013; Murray et al . , 2013; Brown et al . , 2014 ) . There is no broadly-effective therapy or vaccine against DENV or many other FVs , motivating continued research into the molecular basis of disease to identify new vulnerabilities in the viral lifecycle . FVs enter the cell through receptor-mediated endocytosis , culminating in the release of an infectious , ∼10 . 5 kB genomic RNA ( gRNA ) into the cytoplasm ( Kuhn et al . , 2002 ) . The capped but non-polyadenylated viral genomic RNA ( gRNA ) encodes a single open reading frame ( ORF ) , flanked by structured 5′ and 3′ untranslated regions ( UTRs ) . Translation of the FV ORF produces a single polypeptide that is processed by both viral and cellular proteases to yield ten viral proteins ( Fields et al . , 2013 ) . Both FV UTRs fulfill critical roles in the viral lifecycle , ( Iglesias and Gamarnik , 2011 ) including forming base pairing interactions during ( − ) strand synthesis of the viral RNA ( Filomatori et al . , 2006; Villordo and Gamarnik , 2009; Friebe and Harris , 2010; Villordo et al . , 2010; Gebhard et al . , 2011 ) . In addition to replicated copies of the gRNA , high levels of shorter non-coding viral RNAs are also produced during FV infection ( Naeve and Trent , 1978; Takeda et al . , 1978; Wengler and Gross , 1978; Urosevic et al . , 1997 ) . These subgenomic flaviviral RNAs ( sfRNAs ) contain most of the 3′UTR of the FV gRNA and play an important role in regulating the switch between translation and replication of the viral genome ( Lin et al . , 2004 ) . Production of sfRNAs is an evolutionarily-conserved trait in arthropod-borne FV’s and its accumulation is directly linked to human disease ( Pijlman et al . , 2008 ) . Importantly , through still poorly-understood mechanisms , sfRNA controls WNV’s ability to evoke cytopathicity in mammalian cell culture and the onset of pathogenesis in fetal mice ( Pijlman et al . , 2008; Liu et al . , 2014 ) . Virological studies of sfRNAs from WNV , DENV ( Liu et al . , 2010 ) and other flaviviruses showed that they disrupt aspects of the immune response , affecting RNAi mechanisms ( Schnettler et al . , 2012 ) , mRNA turnover pathways ( Moon et al . , 2012 ) , and the type-I interferon response ( Schuessler et al . , 2012; Chang et al . , 2013 ) . The importance of sfRNAs in FV-induced disease raised questions about the mechanism by which these RNAs are produced and maintained at levels approaching or exceeding those of viral gRNA ( Fan et al . , 2011 ) . Several studies showed that sfRNAs form by viral manipulation of the host cell’s RNA turnover machinery . Specifically , sfRNAs form as the result of incomplete degradation of viral gRNA by the cytoplasmic exoribonuclease Xrn1 ( Funk et al . , 2010; Silva et al . , 2010 ) . Xrn1 is a processive , 5′→3′ exonuclease responsible for degradation of roughly 30–40% of mRNA in actively dividing cells ( Jones et al . , 2012 ) . sfRNA production occurs when viral gRNA is loaded into Xrn1 for decay . Xrn1 degrades through nearly the entire viral gRNA before stalling near the beginning of the 3′UTR; what remains is an sfRNA ( Figure 1A ) . 10 . 7554/eLife . 01892 . 003Figure 1 . sfRNAs are formed by incomplete degradation of the flaviviral genomic RNA by the 5′→3′ exonuclease Xrn1 . ( A ) Xrn1 ( red ) likely loads onto decapped , monophosphorylated gRNA and degrades ∼10 kb of RNA before stopping within the viral 3′UTR . The remaining RNA is the sfRNA . ( B ) In the sfRNAs studied to date , stem-loop ( SL ) elements near the 5′ border of the 3′UTR/sfRNA appear to be signals for Xrn1 resistance , and are depicted here as cartoon secondary structures . In Dengue and many other FVs , two of these SLs are present in tandem , as shown . The most highly-conserved parts of the RNA are shaded red and putative PK interactions are indicated with dashed lines . Dumbbell elements ( not shown ) are located 3′ to these SLs . DOI: http://dx . doi . org/10 . 7554/eLife . 01892 . 003 Examination of Xrn1 illustrates how remarkable it is that FVs have evolved a way to resist the exonuclease at a specific and defined point within the gRNA . RNA degradation by Xrn1 takes place in the enzyme’s catalytic chamber buried within a conserved active site ( Chang et al . , 2011; Jinek et al . , 2011 ) . Progression of Xrn1 along an RNA being degraded is hypothesized to be powered by a combination of electrostatic interactions specific for 5′-monophosphorylated RNAs and by repetitious π-stacking of RNA nucleobases with aromatic residues in the enzyme′s active site ( Jinek et al . , 2011 ) . These features enable Xrn1 to degrade through complex RNA structures , including decapped mRNAs and ribosomal RNA ( rRNA ) . Thus , a popular practical application for Xrn1 is eliminating unwanted monophosphorylated RNA ( e . g . , rRNAs ) from samples bound for next-generation sequencing . How do sfRNAs escape degradation by Xrn1 ? Current evidence suggests that specific RNA sequences and structures located in the 3′UTR of different FVs are the signals that cause Xrn1 to stall . Presumably , these RNAs have unique features that make them impervious to Xrn1 . It has been proposed that both stem-loop ( SL ) and dumbbell ( DB ) type structures found in the FVs 3′UTRs can halt Xrn1 progression ( Figure 1B; Pijlman et al . , 2008; Funk et al . , 2010; Silva et al . , 2010 ) . Of these putatively resistant structures , the SL elements from WNV ( Funk et al . , 2010 ) and YFV ( Silva et al . , 2010 ) have been explored . Phylogenetic studies identified ∼19 conserved nucleotides within a secondary structure predicted to include a three-helix junction , a downstream hairpin , and an RNA pseudoknot ( PK ) ( Pijlman et al . , 2008 ) . The presence of the PK is supported by virological studies ( Silva et al . , 2010 ) , and this led to the proposal that these are ‘rigid’ structures . However , our current understanding of these RNA structures does not explain their ability to halt Xrn1 and in the case of DENV , the relationship of the structure of the 3′UTR to its Xrn1-resistant properties is unexplored . Here , we describe our interrogation of the structure and Xrn1-resistant properties of the complete 3′UTR of a serotype 2 Dengue Virus ( DENV2 ) . Chemical probing of the global structure of the 3′UTR revealed five independently-folded RNA structures within the longer RNA . We recapitulated Xrn1 resistance in vitro and developed a new fluorescence-based assay capable of quantitative , real-time monitoring of RNA degradation by Xrn1 and resistance to the enzyme . We used this assay to explore the Xrn1-resistant behavior of RNA structures throughout the DENV2 3′UTR , identifying two Xrn1-resistant RNA structures that confer quantitative protection to downstream segments of RNA . Using one of these structures we developed a series of mutants and precisely mapped where Xrn1 halts , revealing that a specifically-structured three-way junction is a functionally critical element of Xrn1 resistance . Similar mutations within the Xrn1-resistant structures of another FV ( the Kunjin strain of WNV ) , also impair the formation of sfRNAs in vitro and in infected human cells . Cumulatively , these results set the stage for detailed mechanistic and structural studies of these unique , disease-related viral RNAs .
We first characterized the global architecture of the DENV2 3′UTR and the secondary structures of five shorter RNA sequences within the UTR that are predicted to form discrete secondary structure domains ( Figure 2—figure supplement 1 ) . We probed the structure of these RNAs using dimethyl sulfate ( DMS ) and N-methylisatoic anhydride ( NMIA , SHAPE chemistry ) ( Tijerina et al . , 2007; Weeks , 2010; McGinnis et al . , 2012 ) , and used the resultant data to produce an experimentally-supported secondary structure of the complete 3′UTR ( Figure 2A , B ) . The chemical reactivity profiles obtained by probing individual regions of the 3′UTR in isolation match the profiles of those regions in the context of the intact 3′UTR ( Figure 2C ) . This correlation suggests that individual regions of the 3′UTR fold the same when in isolation as they do in the context of the intact 3′UTR , and thus can be studied as structurally discrete elements . The resultant secondary structure model is largely consistent with previous in silico predictions ( Hahn et al . , 1987; Proutski et al . , 1997; Olsthoorn and Bol , 2001 ) . Two stem-loop domains ( SL II and SL IV ) are followed by two ‘dumbbell’ motifs ( DB1 and DB2 ) and by an extended 3′ stem-loop structure ( 3′SL ) that is conserved among FVs . Each of these elements has distinct features:10 . 7554/eLife . 01892 . 004Figure 2 . Chemical probing and predicted secondary structure of the DENV2 3′UTR . ( A ) The Shapeknots predicted secondary structure of the DENV2 3′UTR with NMIA reactivity data overlaid as indicated ( blue ) . Secondary structure elements are labeled . ( B ) Same as panel ( A ) , but with DMS reactivity data overlaid ( green ) . ( C ) Chemical reactivity profiles of the DENV 3′UTR obtained when it is mapped in its entirety or a series of individually transcribed domains . The nucleotide position/number is on the x-axis , the y-axis is normalized reactivity . Locations of secondary structure elements are shown with dashed lines . An ‘ * ‘ indicates a region where the reverse transcriptase tended to stop in the full length 3′UTR . See Supporting Information for additional details . DOI: http://dx . doi . org/10 . 7554/eLife . 01892 . 00410 . 7554/eLife . 01892 . 005Figure 2—figure supplement 1 . Map of the RNAs used during the chemical probing experiments described in this work . ( A ) Schematic of the RNAs mapped in our chemical probing experiments . Numbering indicates the nucleotide positions at the borders of each individually transcribed RNA . Dashed lines indicate the inclusion of the 5′ and 3′ cassettes whose sequences are given in panel ( B ) . The specific sequence of any of these primers is available by request from the authors . DOI: http://dx . doi . org/10 . 7554/eLife . 01892 . 005 We next sought to determine the ability of each of the aforementioned discrete RNA structure from the DENV2 3′UTR to resist degradation by Xrn1 . These experiments were motivated in part by evidence that multiple Xrn1-resistant structures may be contained within a single FV 3′UTR ( Funk et al . , 2010; Silva et al . , 2010 ) . We therefore developed an in vitro assay capable of reporting Xrn1 resistance using RNAs transcribed with a 31 nucleotide ‘leader’ upstream of a putatively Xrn1-resistant structure . These RNAs were treated with a bacterial RNA pyrophosphate hydrolase from Bedelovibrio bacteriovorus ( BdRppH ) to convert the 5′-triphosphate generated during in vitro transcription to a monophosphate ( Figure 5A; Deana et al . , 2008; Messing et al . , 2009 ) , and render them effective substrates for Xrn1 . In the same reaction these RNAs were treated with Xrn1 from Kluveromyces lactis ( KlXrn1 ) ( Chang et al . , 2011 ) to degrade the RNA ( Figure 5B ) . Truncation of the input RNA to a specific shorter product was interpreted as diagnostic of Xrn1 resistance . 10 . 7554/eLife . 01892 . 009Figure 5 . Recapitulation of Xrn1 resistance in vitro . ( A ) Reactions carried out by RppH and Xrn1 . ( B ) Gel demonstrating the specificity of each component in the reconstituted reaction and the Xrn1-resistant behavior of ( +31 ) -DVxrRNA1 ( the input test RNA ) , which contains the isolated SL II element after a 31 nucleotide-long leader . The red arrow indicates the truncated product formed by Xrn1 resistance ( DVxrRNA1 , the resistant RNA ) . The 24-mer control RNA is labeled . DOI: http://dx . doi . org/10 . 7554/eLife . 01892 . 00910 . 7554/eLife . 01892 . 010Figure 5—figure supplement 1 . Purification of RNA processing enzymes . ( A ) SDS-PAGE gel showing the purity of RNA pyrophosphate hydrolase from Bdelovibrio bacteriovorous ( BdRppH , MW = 17 kDa ) . ( B ) SDS-PAGE gel showing the purity of Xrn1 from Kluveromyces lactis ( KlXrn1 , MW = 144 kDa ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01892 . 01010 . 7554/eLife . 01892 . 011Figure 5—figure supplement 2 . DVxrRNA1 demonstrates specific resistance to Xrn1 . When reacted with other exonucleases such as the 5′→3′ Phosphodiesterase I from Bovis tarsus ( BtPde1 ) or the 3→5′Phosphodiesterase II from Croatalus adamanteus venom ( CaPde2 ) , ( +31 ) -DVxrRNA1 RNA is readily degraded . DOI: http://dx . doi . org/10 . 7554/eLife . 01892 . 01110 . 7554/eLife . 01892 . 012Figure 5—figure supplement 3 . Examination of in trans protection of other Xrn1 substrates by DVxrRNA1 . ( A ) Outline of the experiment used to test for inhibition of Xrn1 by DVxrRNA1 constructs . Briefly , increasing amounts of DVxrRNA1 with a 31 nt ( +31 ) or no ( +0 ) added leader were included in reactions of a 24-mer monophophorylated substrate with Xrn1 , incubated for 30 min , and resolved by gel electrophoresis . Xrn1-resistant RNAs were added in stoichiometric excess of Xrn1 . Conditions are described in ‘Materials and methods’ . ( B ) dPAGE following the reaction described in ( A ) at 37°C and ( C ) at 0°C . At the lower temperature , some in trans protection may be observed . DOI: http://dx . doi . org/10 . 7554/eLife . 01892 . 012 Using this assay we first tested the SL II element of the DENV2 3′UTR for resistance to Xrn1 using a 5′ leader consisting of the 31 nucleotides that naturally precede SL II in the viral genome . Addition of BdRppH and KlXrn1 to a reaction containing this test RNA as well as a synthetically generated 5′-monophosphorylated 24 nucleotide-long internal control RNA ( 24-mer ) resulted in complete degradation of the control RNA and truncation of the test RNA to a discrete product , indicative of Xrn1 resistance ( Figure 5B ) . Because the enzymes used in this experiment were pure ( Figure 5—figure supplement 1 ) , this experiment establishes that Xrn1 resistance is conferred entirely by the SL II test RNA ( no auxiliary proteins are needed ) . We therefore conclude that RNA elements from within the DENV 3′UTR as short as ∼60 nucleotides can resist Xrn1 and that these RNAs operate effectively outside the context of the viral 3′UTR . Interestingly , we observed no enzymatic resistance using other commercially available exonucleases ( Figure 5—figure supplement 2 ) , suggesting this RNA is specifically resistant to Xrn1 . For clarity , here we refer to this ( and other ) discrete RNAs using the functionally-descriptive name of ‘Xrn1-resistant RNAs’ ( xrRNAs ) in order to distinguish them from the longer sfRNAs . To facilitate discussion of xrRNAs in this and future work we propose a simple naming scheme for xrRNAs that is outlined in the ( Figure 3—figure supplement 1 ) . Using this nomenclature , we here refer to the first Xrn1-resistant structure in the DENV 3′ UTR as ‘DVxrRNA1’ . To augment the aforementioned assay and enable monitoring of Xrn1 resistance in a quantitative and time-resolved manner , we engineered a fluorescence-based assay based on the malachite green ( MG ) aptamer ( Grate and Wilson , 1999; Baugh et al . , 2000 ) . The fluorescence of the triphenylmethylene MG dye is quenched when free in solution but dramatically increases when bound to the aptamer . This feature provides a fluorescent readout of the MG aptamer’s integrity ( Figure 6A; Babendure et al . , 2003 ) . We designed RNA substrates with the 40 nt-long MG aptamer placed downstream of each RNA structure to be tested for Xrn1 resistance ( following a 4 nucleotide-long poly ( U ) linker ) ( Figure 6B ) . In solution with MG , these RNAs produce an ∼2300-fold increase in MG fluorescence over free dye ( Babendure et al . , 2003 ) . If addition of RppH and Xrn1 results in degradation of the aptamer-tagged RNA ( no Xrn1 resistance ) , fluorescence disappears over time ( Figure 6C , bottom ) . If the RNA being tested provides Xrn1 resistance , the MG aptamer remains intact and fluorescence persists ( Figure 6C , top ) . A fuller description of the development and testing of this technique will be reported elsewhere ( Chapman et al . , in preparation ) . We used this fluorescence assay to test the Xrn1 resistance properties of the five discrete structures located within the DENV2 3′UTR ( Figure 6D ) . The 5′ leader of each RNA consisted of nucleotides 10 , 273–10 , 304 of the DENV 3′UTR , which is degraded when present before DVxrRNA1 ( Figure 5B ) and each had the MG aptamer downstream . For DVxrRNA1 , the test RNA was thus named ( +31 ) -DVxrRNA1-MG and other input RNAs were named following the same convention ( Figure 3—figure supplement 1 ) . As expected , upon the addition of both enzymes ( +31 ) -DVxrRNA1-MG maintains high fluorescence throughout the course of the experiment , consistent with Xrn1 resistance ( Figure 6D ) . Likewise , SL IV demonstrated resistance to Xrn1-mediated decay ( DVxrRNA2 ) . In contrast , reactions with the DB2 and the 3′ terminal stem loop [test RNAs: ( +31 ) -DVDB2-MG and ( +31 ) -DV3SL-MG] show a decrease in fluorescence that is consistent with no Xrn1 resistance . The construct encoding DB1 [ ( +31 ) -DB1-MG] shows more intermediate behavior . 10 . 7554/eLife . 01892 . 013Figure 6 . Design of a fluorescence assay for monitoring Xrn1 resistance and testing of individual DENV2 3′UTR structural elements . ( A ) Fluorescent properties and structure of the MG dye and MG aptamer . ( B ) Schematic of the RNA constructs used for monitoring the decay kinetics of individual elements of the DENV 3′UTR . Conserved nucleotides are colored red as in Figure 3B . Green indicates fluorescing MG dye . ( C ) Cartoon representing the expected outcome when using Xrn1-resistant ( top ) or non-resistant ( bottom ) RNAs . Green glow indicates fluorescence . ( D ) Fluorescence traces of different RNAs over the course of their reaction with Xrn1 . X-axis is time and y-axis is relative fluorescence intensity . Data are normalized to ( − ) Xrn1 controls and are averaged over three independent experiments . Error bars show one standard deviation from the mean . ( E ) dPAGE analysis of the products of the experiment of panel ( D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01892 . 013 To assess the products of putative Xrn1 resistance from the fluorescence assay , we analyzed the reactions from the completed experiment using denaturing polyacrylamide gel electrophoresis ( dPAGE ) ( Figure 6E ) . Consistent with the fluorescence traces , truncated Xrn1-resistant products accumulated in reactions with DVxrRNA1 and DVxrRNA2 ( SL II and SL IV ) -containing test RNAs . 3′ SL and DB2-contianing constructs show complete degradation of the input RNA ( Figure 6E ) , consistent with the fluorescence assay . In the case of the DB1-containing test RNA , the full-length RNA is partially depleted however there is no truncated product as would be expected if the DB1 sequence itself were Xrn1-resistant . The persistence of full-length RNA in this reaction suggests that Xrn1 may not be able to efficiently load onto this RNA and the lower intensity of the band suggests that when Xrn1 does load it degrades the RNA entirely . As >3 nucleotides of single-stranded RNA are required for Xrn1 to efficiently load on an RNA substrate ( Jinek et al . , 2011 ) , the behavior of the DB1-containing RNA may indicate an alternate secondary structure that precludes the 5′ end of this RNA from being recognized by Xrn1 . Cumulatively , our fluorescence and gel-based assays confirm two xrRNA structures in the DENV2 3′UTR that offer nearly quantitative protection to downstream RNA , of these we chose DVxrRNA1 as an archetype for deeper analyses . We first mapped the precise location where Xrn1 is blocked by DVxrRNA1 . To do this , we carried out a large-scale reaction using ( +31 ) -DVxrRNA1-MG and recovered the truncated product RNA by dPAGE ( DVxrRNA1-MGprod ) ( Figure 7A ) . We then used reverse transcription to determine the sequence of the 5′-end left by Xrn1 resistance ( Figure 7B ) . This experiment reveals that DVxrRNA1-MGprod RNA ends at a U corresponding to U10299 of the DENV2 3′UTR , five nucloetides upstream of the first phylogenetically-conserved nucleotide in the DVxrRNA1 structure . 10 . 7554/eLife . 01892 . 014Figure 7 . Characterization of the products left by Xrn1 resistance . ( A ) Gel of RNAs used to map the 5′ end of Xrn1-resistant product RNA . ( B ) Reverse transcription of RNAs from panel ( A ) . The cartoon inset shows a schematic of the reaction . Dideoxy sequencing lanes are labeled . The location of the stop site ( and hence the 5′ border of the product RNA ) is shown with a red arrow to the right , along with the sequence of the RNA surrounding this position . ( C ) Diagram of the experiment used to determine the phosphorylation state of the products left by Xrn1 resistance . Yellow balls indicate non-radioactive phosphates , a green star depicts 32P radioactive phosphates . ( D ) Gel containing the outcome of the experiment shown in panel ( C ) . Different species of RNA or DNA are labeled to the left and cartooned to the right . Red arrows: Xrn1-resistant product RNA . DOI: http://dx . doi . org/10 . 7554/eLife . 01892 . 014 We then interrogated the chemical nature of the 5′ end left by Xrn1 , which speaks to a possible mechanism for Xrn1 resistance . Substrate recognition by Xrn1 requires a 5′ monophosphate at the end of an RNA and each catalytic cycle regenerates this chemical species . Could Xrn1 resistance be based in part upon malfunction of this process ? If so , the products formed by xrRNAs might not bear a 5′ monophosphate and would be precluded from further degradation . We determined the phosphorylation state of the 5′ end of Xrn1-resistant products formed by DVxrRNA1 using the procedure outlined in Figure 7C and described in the Supplementary Description of ‘Materials and methods’ . Briefly , after ‘product’ and ‘control’ RNAs were isolated , half of each sample was dephosphoylated using calf intestinal alkaline phosphotase ( ‘CIP’ed’ ) . We then compared the ability of T4 polynucleotide kinase ( PNK ) to ligate a radioactive phosphate to the CIP’ed vs un-CIP’ed RNAs , expecting that the presence of a 5′-monophosphate would interfere with effective phosphorylation . In these experiments we included both a synthetically 5′ monophosphorylated 24-mer RNA and a 5′ hydroxylated 40-mer DNA as controls . As predicted , the dephosphorylated products of the DVxrRNA1 ‘control’ RNA are better substrates for phosphorylation than their un-CIP’ed counterparts ( Figure 7D , lanes A and B ) . Control reactions involving the synthetic 5′-monophosphorylated 24-mer produce similar results ( Figure 7D , lanes E and F ) . Similarly , the CIP’ed Xrn1-resistant product RNAs are better substrates for phosphorylation than their un-CIP’ed counterparts . These results indicate that the RNA product of Xrn1 resistance are 5′-monophosphorylated and are still viable substrates for Xrn1 ( Figure 7D , lanes C and D ) . Xrn1 resistance is thus not the result of chemical ‘mis-step’ of the exonuclease . The results presented above and those in the literature lead to the hypothesis that Xrn1 resistance depends on a specific RNA structure . To link RNA sequence , structure , and function and to identify the elements of the DVxrRNA1 that are important for conferring Xrn1 resistance , we designed eleven mutants targeting conserved xrRNA sequence and secondary structure elements and tested their Xrn1-resistant properties ( Figure 8 ) . As before , we transcribed these RNAs with 31 nucleotide-long 5′ leaders and compared their Xrn1-resistant behavior ( relative to a wild-type ( WT ) control ) using both dPAGE and our fluorescence assay . In these experiments DVxrRNA1 mutant RNAs displayed a range of Xrn1-resistant behaviors ( Figure 8A , B ) :10 . 7554/eLife . 01892 . 015Figure 8 . Mutational analysis of DVxrRNA1 . ( A ) Secondary structure of DVxrRNA1 with mutations labeled Conserved sequence elements in red . The stop site for Xrn1 is indicated . Inset is a bar graph containing the effects of mutations Xrn1 resistance when quantified using the fluorescence assay . The x-axis identifies each mutant , the y-axis the fraction of MG-tagged RNA degraded at 160 min . C is a non-resistant control RNA . Graph depicts the average of two independent experiments . ( B ) Gel analysis of the reaction of each mutant with Xrn1 and RppH as in Figure 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 01892 . 01510 . 7554/eLife . 01892 . 016Figure 8—figure supplement 1 . Native PAGE analysis of DVxrRNA1 mutants . Constructs and conditions are labeled within the figure . A secondary structure depicting the identity of each mutant is displayed in Figure 8 of the main text . DOI: http://dx . doi . org/10 . 7554/eLife . 01892 . 016 If Xrn1 resistance is due to the formation of a specific RNA structure , mutations that affect Xrn1 resistance should disrupt this structure . To test this , we subjected the mutants described above to non-denaturing ( native ) gel electrophoresis , which is sensitive to changes in the global structure of a folded RNA ( Figure 8—figure supplement 1; Woodson and Koculi , 2009 ) . In gels containing 5 mM MgCl2 , mutants of DVxrRNA1 display a spectrum of electrophoretic behaviors , indicating a variety of effects on RNA structure . One mutant that stood out was the non-Xrn1-resistant mutant 6 ( C10320→G ) ( Figure 9 ) . In EDTA-containing gels this mutant migrated at the same rate as WT DVxrRNA1 , however in gels containing MgCl2 mutant 6 demonstrates significantly retarded mobility , indicative of altered folding of the RNA . This suggests that a single point mutation made in the DVxrRNA1 three-helix junction is capable of disrupting a Mg2+-dependent tertiary fold that is linked to Xrn1 resistance . To further examine this behavior we used chemical probing experiments to assess the structure of the mutant 6 RNA . The normalized NMIA and DMS reactivity profiles of the WT and mutant 6 MG aptamer-tagged RNAs are similar but with several marked differences ( Figure 9A , D ) . These data reveal that the WT and mutant 6 RNA have similar secondary structures ( consistent with the EDTA-containing native gels , Figure 9C ) but indicate that the C10320→G mutation affects several structural elements and changes the global fold of the RNA ( consistent with the MgCl2-containing native gel Figure 9C ) . These findings support the conclusion that the DVxrRNA1 three-helix junction organizes a specific structure that is essential for Xrn1 resistance . 10 . 7554/eLife . 01892 . 017Figure 9 . Structural analyses of DVxrRNA1 mutant 6 ( mutant in the three-way junction ) . ( A ) Normalized NMIA and DMS reactivity profiles of mutant 6 compared to WT RNA , depicted as in Figures 3 and 4 . Colored bars ( green and blue ) represent the WT RNA and black bars represent mutant 6 . The location of the point mutation is shown with a purple shaded box . Reactivity changes are indicated by red boxes . ( B ) dPAGE of Xrn1 resistance by WT and mutant 6 . ( C ) Non-denaturing ( native ) PAGE analyses of WT and mutant 6 . ( D ) Secondary structure of the DVxrRNA1 with elements labeled . The purple shaded box shows the location of the mutation . Regions of the structure that show substantial changes in the chemical probing ( from panel ( A ) ) are indicated by red boxes . DOI: http://dx . doi . org/10 . 7554/eLife . 01892 . 017 Having established that xrRNA function depends on a structure organized by the P1-P2-P3 three-helix junction , we then assessed the generality of our findings in another FV and tested their relevance to sfRNA production in infected cells . Because sfRNA formation is readily explored in the context of the Kunjin strain of WNV ( WNVKUN ) , we used this virus system to assess the generality and relevance . We first tested two proposed xrRNA structures located within in the WNVKUN 3′UTR ( Figure 10A ) for Xrn1 resistance , again using a 31 nucleotide-long leader . Both of these RNAs , WNVKUNxrRNA1 ( nucleotides 10 , 499–10 , 574 ) and WNVKUNxrRNA2 ( nucleotides 10 , 659–10 , 728 ) , demonstrate resistance to Xrn1 in vitro ( Figure 10B ) . We also tested versions of these RNAs containing C→G point mutations made in positions analogous to the mutation ( mutant 6 ) that abolishes Xrn1 resistance in DVxrRNA1 ( C10519→G in WNVKUNxrRNA1 and C10680→G in WNVKUNxrRNA2 ) ( Figure 10B , C ) . We observe that the C→G point mutation in WNVKUNxrRNA1 partially compromises Xrn1 resistance , while the same point mutation made in WNVKUNxrRNA2 abolishes resistance . Together these results demonstrate the generality of the observations we made using DVxrRNA1 but also show some subtle variations in different xrRNA species . 10 . 7554/eLife . 01892 . 018Figure 10 . Identification and testing of WNVKUN xrRNA structures . ( A ) Sequence alignment of DVxrRNA1 , DVxrRNA2 , WNVKUNxrRNA1 and WNVKUNxrRNA2 . Nucleotide positions are indicated and correspond to Genbank accession numbers M20558 . 1and AY274504 . 1 for DENV2 and WNVKUN , respectively . Conserved nucleotides of these RNAs are highlighted in red . The point mutation we made in the three-was junction is indicated with a purple box . ( B ) dPAGE of the Xrn1 resistance assay run using WNVKUN xrRNAs and C10519→G and C10680→G point mutants . ( C ) Conserved secondary structure of WNVKUNxrRNA1 and WNVKUNxrRNA2 , with conserved nucleotides highlighted and the location of the mutation shaded purple . DOI: http://dx . doi . org/10 . 7554/eLife . 01892 . 018 The above results allowed us to examine the role played by each WNVKUN xrRNA structure in the formation of sfRNAs during infection . We infected human 293T cells with both wild-type and mutant WNVKUN and assessed sfRNA accumulation at 48 hours post infection ( h . p . i . ) . Northern blots of total RNA isolates show that infection by wild-type WNV produces three prominent sfRNA species ( Figure 11 ) . Based on their size , the largest of these RNAs , sfRNA1 and sfRNA2 , are the result of Xrn1 pausing at WNVKUNxrRNA1 and WNVKUNxrRNA2 , respectively . The third species , sfRNA3 corresponds with previous observations of Xrn1 stopping prior to DB1 of WNVKUN . Infections using a WNVKUN harboring a C10519→G point mutation in WNVKUNxrRNA1 results in a significant and reproducible decrease in sfRNA1 formation , corresponding with the incomplete loss of Xrn1 resistance we observe in vitro . Likewise , infection with WNVKUN containing a C10680→G point mutation in WNVKUNxrRNA2 results in the disappearance of sfRNA2 , and again mirrors our findings in vitro . Finally , infection by a WNVKUN containing both C10519→G and C10680→G point mutations produces low levels of sfRNA1 and essentially no sfRNA2 . The more substantial decrease in sfRNA1 formation in the double mutant as opposed to the C10519→G single mutant may suggest a degree of cooperation between the two xrRNA structures although establishing such an interaction will require additional experimentation . Cumulatively our results show that the structural and functional features we observe have applicability across xrRNAs from diverse FV’s and that the structural characteristics that confer Xrn1 resistance in vitro corresponds directly to the accumulation of disease-causing sfRNAs during viral infection . 10 . 7554/eLife . 01892 . 019Figure 11 . sfRNAs produced during infection by wild-type and mutated WNVKUN . Northern blot analysis of total RNA isolated from human 293T cells infected with WT and mutants WNVKUN , analyzed at 48 h . p . i . The location of molecular weight markers and the identity of sfRNA species are indicated to the left and right of the blot respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 01892 . 019
Non-coding RNAs play important roles in viral disease ( Steitz et al . , 2011 ) . During infections caused by FV’s , the formation of sfRNAs through incomplete degradation of the viral genome is directly linked to disease ( Pijlman et al . , 2008; Liu et al . , 2014 ) . In this study we identified discrete elements of the DENV2 3′UTR that confer resistance to Xrn1 ( xrRNAs ) and explored characteristics of the RNA structures responsible . We correlated these findings with similar RNAs in WNVKUN and demonstrated that the formation of specific , discrete folded structure is responsible for the production of sfRNAs . Here we discuss implications of these findings as they relate to the organization of the FV 3′UTR and the infection strategy used by these viruses . Based on our data we propose a model for how an RNA structure might confer Xrn1 resistance , and speculate about implications for other Xrn1-resistant RNAs . The DENV 3′UTR ( and correspondingly its sfRNA ) is comprised of individual structural elements that , based on our probing data , do not appear to interact with each other to form a higher-order structure . This differs from other viral 3′UTRs that we have shown can fold into a single higher-order structure ( Hammond et al . , 2010 ) . The presence of individually folded , discrete structures within the DENV 3′UTR correlates well with the observation that different elements are responsible for interacting with different proteins and perform different tasks during the FV replication cycle ( Polacek et al . , 2009; Manzano et al . , 2011; Ward et al . , 2011; Hussain et al . , 2012 ) . This ‘knots in a rope’ architecture of the DENV2 3′UTR and sfRNA may be important to enable and organize these multiple functions . We identified two resistant functional RNA structures ( DVxrRNA1 , DVxrRNA2 ) located in succession near the 5′ end of the DENV 3′UTR . These elements contain previously-proposed secondary structure stem-loop elements SL II and SL IV . We did not observe resistance from the DB elements in vitro; if they confer Xrn1 resistance during infection this may require interaction with a protein factor . The position of the two confirmed xrRNAs is consistent with a role in protecting downstream sequences and structures , including the DBs and 3′SL from degradation . Many other FV′s also contain two neighboring xrRNA structures at the 5′ end of their 3′UTR′s ( a notable exception is YF , which only contains one ) ( Pijlman et al . , 2008 ) . The evolutionary conservation of this pattern suggests that downstream RNA elements are critical for sfRNA function and viral success; however , the nearly quantitative resistance to Xrn1 demonstrated by these structures begs the question: why there are two xrRNAs in series when seemingly one should suffice ? Perhaps two xrRNAs provide redundancy to ensure production of the sfRNA or perhaps each xrRNA is tuned to operate most efficiently under different intracellular conditions encountered during the infection cycle ( Villordo and Gamarnik , 2013 ) . Our data suggest that the PK interaction is stable in only one of the two xrRNAs in DENV2 , hinting that subtle functional differences may exist between the two structures . Furthermore , our data from cells infected with mutant Kunjin Virus hint at cooperation or coupling between these elements . Cumulatively , our data suggest that a critical feature of Xrn1 resistance is the formation of a specifically structured three-helix junction . The presence of multiple conserved bases within the three-way junction , including several unpaired nucleotides , suggests that this core organizes the tertiary fold of the RNA . This idea is supported by that fact that this fold can be disrupted by a point mutation made in one of the unpaired nucleotides within the junction . The fold this junction organizes is important for Xrn1 resistance , and while it may be further stabilized by a PK , this tertiary interaction does not appear to be necessary for Xrn1 resistance in vitro ( at least in the context of DVxrRNA1 ) . Full understanding of the basis of Xrn1 resistance will require a high-resolution structure of an xrRNA , but our results allow some predictions . First , examination of the three-way junction features and comparisons with other RNA junctions suggest that it forms a ‘type-C’ three-helix junction in which P1 and P2 coaxially stack and P3 assumes an acute angle relative to P1 ( Lescoute and Westhof , 2006 ) . Based on this , we predict that folding of the junction ( in the absence of any tertiary interactions ) would bring L3 close to the base of P1 . Formation of the PK would further stabilize this helical arrangement . How does this conformation result in Xrn1 resistance ? We speculate that as the 5′ end of the RNA is drawn into the active site , the enzyme encounters a structure formed by the spatially adjacent P1 and P3 helices that prevents the helicase activity of the enzyme from unwinding the RNA . Jinek et al . ( 2011 ) have proposed that RNA entering the active site of Xrn1 is unwound when pulled through a narrow channel formed by the α1 helix and an adjacent loop . The xrRNA may have evolved an unusual and specific structure that prevents these features of the enzyme from effectively interacting with and unwinding the RNA . Such a structure need not be rigid or even exceptionally thermodynamically stable , but must be arranged in such a way that the enzyme cannot process it . We did not directly address the pathways by which viral genomic RNAs could be decapped and recognized by Xrn1 , or the pathways by which sfRNA alters infected cells once it has been formed , but our results may lend insight into these processes . One interesting idea is that sfRNA production causes dysregulation of Xrn1-dependent mRNA turnover in the infected cell . Indeed , studies of Dengue and Kunjin virus infections showed that sfRNA production stabilizes mRNAs by inhibiting Xrn1 ( Moon et al . , 2012 ) . One possible explanation of this effect could be that Xrn1 remains bound to xrRNAs , perhaps locking the enzyme in an inactive conformation or sequestering it from other substrates . There is evidence that an RNase L-inhibiting RNA of poliovirus operates as a competitive inhibitor through this type of mechanism ( Keel et al . , 2012 ) . However , in several of our experiments Xrn1 processes over 150 copies of the xrRNA substrate and over 300 copies of the 24-mer control RNA , indicating multiple turnover by the enzyme . Nonetheless , we did find conditions under which afforded some protection to an Xrn1 substrates in trans ( Figure 5—figure supplement 3 ) , but we have been unable to obtain evidence of formation of a stable complex between a resistant RNA and Xrn1 in a purified system ( data not shown ) . Overall , these observations argue against a simple binding model for sfRNA-induced changes in mRNA turnover and suggest that sfRNA may act as a reversible inhibitor of Xrn1 , perhaps changing the kinetics of the enzyme in infected cells . That Xrn1 resistance is conferred by a discrete , relatively short , and highly conserved RNA structure raises the notion that it could be targeted with therapeutics . The observation that mutation of a single conserved nucleotide in the three-way junction completely abrogates Xrn1 resistance and sfRNA formation in human cells suggests that only a few key intermolecular interactions need to be altered to prevent xrRNA folding and function . Perhaps mutations or deletions in individual xrRNA elements present in other FV genomes could be become part of a general route toward the development of attenuated vaccines . In conclusion , we mapped the architecture of the complete 3′UTR of DENV2 and identified Xrn1-resistant activity residing in discrete and portable structural elements . We identified similar Xrn1-resistant RNA structures in WNVKUN and demonstrated that these structures are directly linked to the formation of sfRNAs in vivo . Our studies regarding the characteristics of these uniquely , Xrn1-resistant RNAs suggest that a specific and unusual structure confers resistance to Xrn1 . These studies provide a framework for more detailed structure-based mechanistic investigations of these RNAs that are directly linked to disease .
In general , RNA was refolded prior to each experiment described in this manuscript . This was done using a thermocycler program that first ramped to 90°C for 2 min , held at 20°C for 5 min , and finally cooled and held at 4°C until used . The majority of experiments described here were carried out in 100 mM NaCl , 10 mM MgCl2 , 50 mM Tris , pH 7 . 9 , 1 mM DTT . For simplicity we will refer to these conditions as 1X EC3 throughout this supplement . Denaturing polyacrylamide gel electrophoresis was carried out using gels consisting of 8M urea , 1X TBE ( 80 mM Tris , 80 mM borate , 2 mM EDTA , pH 8 . 0 ) . In order to streamline experimentation and accommodate technical aspects of capillary electrophoresis , individual regions of the Dengue 3′UTR were transcribed between adjacent 5′ and 3′ ‘cassettes’ . These cassettes were uniformly installed before and after each of the five shorter RNA constructs as diagrammed Figure 2—figure supplement 1 . The 5′ cassette consisted of the first 31 nucleotides of the DENV2 3′UTR . This region of RNA is predicted to be predominantly unstructured in the context of these RNAs , agreeing well with the high level of modification we observe in our experiments . The 3′ cassette consisted of a 4 nucleotide-long uridine tract followed by a malachite green aptamer . In addition to providing a convenient place to anneal a primer for reverse transciption , this fluorophore binding aptamer was later used to observe the Xrn1 decay kinetics of these RNAs . In chemical probing experiments 2 μM RNA was folded in 1X EC3 and then reacted with NMIA or DMS . The RNAs mapped in this study were each subjected to a titration of these reagents in order to optimize the extent of modification and the corresponding efficiency of reverse transcription . When the full length 3′UTR was mapped using a primer corresponding to the 3′ end of this RNA we observed significant background pausing in RT just prior to the 3′ component of the DB1 pseudoknot ( data not shown ) . We therefore designed a second , internal primer to monitor chemical modification 5′ of this position in the full length 3′UTR/sfRNA ( Figure 2—figure supplement 1 ) . DMS modification was carried out in 10 μl reactions containing 2 μM RNA and 57 mM DMS in 1X EC3 . Reactions were incubated for 5 min at 0°C then quenched by the addition of 1 μl of neat β-mercaptoethanol . NMIA modification was carried out in 10 μl reactions containing 2 μM RNA and 2 . 5 mM NMIA ( Invitrogen ) in 1X EC3 . Reactions were incubated at 37°C for 1 hr . Following chemical modification , reactions were passed over homemade G-25 Sephadex ( GE Healthcare , Little Chalfont , United Kingdom ) spin columns and recovered . 1 . 25 pmol ( 0 . 1 μM , 0 . 063 equivalents ) of an appropriate 5′ 6-FAM end-labeled primer was then added . Reactions were then re-annealed in order to hybridize these primers to each RNA . Reverse transcription was then carried out using Superscript III reverse transcriptase in a reaction containing 0 . 1 μM primer and ∼1 . 5 μM RNA , 80 mM KCl , 3 . 23 mM MgCl2 , 6 mM DTT , 55 mM Tris , pH 8 . 3 , and 533 μM dNTPs . Following reverse transcription , these reactions were again desalted using G-25 spin columns and brought to 50% formamide ( HiDi , Applied Biosystems , Inc . , Foster City , CA ) . 1 μl of a ROX-labeled DNA ladder ( GeneScan-500 , ABI ) was then added to each ∼40 μl reaction . Reactions were then heated at 90°C for 10 min in order to denature cDNA products from the RNA templates and subsequently transferred to 96-well plates prior to capillary electrophoresis . Capillary electrophoresis was carried out using an ABI 3500 Gene Scanner equipped with a 50 cm capillary unit filled with POP-7 polymer ( ABI ) . Capillary electrophoresis data were exported as . fsa files and processed using the HITRACE-web online software ( Kim et al . , 2013 ) . This software was developed by and is maintained by the Das lab at Stanford University and Yoon lab at Seoul University in South Korea . The data files produced by HiTRACE were subsequently transferred to Matlab for further analysis , normalization , figure making and conversion into formats compatible with further processing ( e . g . , components of the RNAstructure suite ) ( Reuter and Mathews , 2010 ) . Because the Dengue 3′UTR is predicted to harbor multiple pseudoknot structures , we used the recently developed software program Shapeknots ( Hajdin et al . , 2013 ) to incorporate our chemical mapping data into secondary structure prediction algorithms capable of identifying such structures . Figures overlaying chemical probing data onto RNA secondary structures were made using the software program R2R ( Weinberg and Breaker , 2011 ) . To assess RNA decay kinetics through monitoring the fluorescence of the MG aptamer , 100 μl reactions of 4 μM RNA ( 400 pmol , ∼16 μg ) were refolded in 1X EC3 . Often several 100 μl reactions using the same RNA were carried out simultaneously and were pooled after being folded separately in an 8-tube PCR strip . While reactions were still pooled , 5 μl of >3 U/ μl RppH per reaction and 7 . 5–10 equivalents of malachite green ( 30 μM final concentration , 3 μmol per reaction ) were added . 105 μl aliquots of this mixture were then divided into separate wells of a black , flat bottomed 96-well plate ( Greiner Bio-One , Monroe , NC ) . Fluorescence was monitored using a Glomax Multi+ plate reader using a filter set that allowed excitation at 625 nm and monitoring of emission over 660–720 nm . Prior to the addition of Xrn1 , fluorescence was measured for 5 min in order to ensure reactions had reached equilibrium and to allow RppH to start producing 5′-monophosphorylated Xrn1 substrates . At 5 min , 5 μl of >3 U/μl Xrn1 was added and the reaction was then monitored for the next 160 min . Under these conditions RNA decay rates vary linearly with added Xrn1 ( data not shown ) . To produce the gel in Figure 6E , after the experiment had ended , 20 μl of each reaction was removed and subjected to 10% dPAGE . The gel was stained with methylene blue . In order to characterize the products left by Xrn1 resistance we conducted the experiment described for monitoring Xrn1 decay kinetics using ( +31 ) -DVxrRNA1-MG RNA on an eightfold scale . The products of these reactions , along with an equivalent number of ( − ) Xrn1 control reactions , were recovered after 160 min , pooled and subjected to 15% dPAGE . RNA products were visualized by staining with methylene blue , carefully excised and eluted from the gel overnight in water . Eluted RNA was concentrated using Amicon molecular weight cutoff filters and frozen at −20°C before additional analyses . Subsequent quantification revealed Xrn1-resistant RNA products were recovered with a 16% overall yield or as 0 . 52 μmol from a 3 . 2 μmol prep . Recovered DVxrRNA1-MGprod ( product ) and ( +31 ) -xrRNA1-MG ( control ) RNAs were annealed to a 5′ end-labeled primer complementary to nucleotides 10 , 328 through 10 , 342 in DVxrRNA1 . Reverse transcription was carried out as essentially as described for chemical mapping experiments . Reactions were quenched by the addition of 8 M urea , 10 mM EDTA loading dye and then heated to 90°C for 10 min in order to promote dissociation of cDNAs from the RNA . Reactions were then loaded onto 10% sequencing gels . Electrophoresis was carried out at 50 W for 4 hr . The resulting gels were removed , dried and exposed to Molecular Dynamics phosphor screens before being imaged on a Storm 720 scanner . Gels were visualized using ImageQuant software . The workflow for this experiment is outlined in Figure 7C . To determine the phosphorylation state of Xrn1-resistant RNA products and accompanying controls , fractions of these samples were with first treated with calf-intestinal alkaline phosphotase ( CIP ) ( line iii , Figure 7C ) . CIP reactions were carried out using ∼5 μg of each RNA in a reaction performed as prescribed by the manufacturer of the enzyme ( NEB ) . CIP’ed RNAs were purified by phenol chloroform extraction , washed once with 24:1 chloroform to isoamyl alcohol and then ethanol precipitated ( 70% ice cold EtOH , 300 mM NaOAC pH 5 . 2 , −20°C , 2 hr ) . CIP’ed RNAs were then resuspended and quantified . All RNAs: CIP’ed and un-CIP’ed , ‘product’ , ‘control’ and ’24-mer’ were then entered into phosphorylation reactions containing 50 pmol γ-32P-ATP and 8 pmol of each RNA ( line iv , Figure 7C ) . 1 pmol of an synthetic , 5′-hydroxylated 40mer DNA was included to serve as a internal control for PNK activity . RNAs were then phosphorylated using T4 PNK ( NEB ) in the buffer supplied by the manufacturer for 1 . 5 hr at 37°C . Individual samples were diluted and then passed over homemade G-25 Sephadex columns to remove unincorporated γ-32P-ATP . An equal volume of each reaction was the subjected to 15% dPAGE . The resulting gels were analyzed by phosphorimaging as described above . Prior to native gel electrophoresis 2 μg of each DVxrRNA1 mutant was brought to 18 μl in water ( ∼4 μM ) , refolded and then held at 4°C . Samples were brought to 1X EC3 by the addition of 2 μl of a 10X buffer and equilibrated on ice for 10 min . An equal volume of 50% ( wt/vol ) sucrose , 0 . 1% bromophenol blue , 0 . 1% xylene cyanol loading dye was added immediately prior to native gel electrophoresis . Electrophoresis was carried out at 4°C using 8% , 39:1 ( mono:bis ) polyacrylamide 1X TH gels ( 33 mM Tris , 66 mM HEPES , pH ∼8 ) with 5 mM MgCl2 or 2 mM EDTA added as indicated . In the experiments presented here , 300 pmol of ( +31 ) or ( +0 ) leadered xrRNAs were reacted with 150 pmol of Xrn1 in 1X EC3 for 30 min at the temperature indicated . At 30 min , 450 pmol of a synthetic , 5′-monophosphorylated 24-mer was added and the reaction continued for 30 min . At 1 hr these reactions were quenched by the addition of an 8 M urea , 10 mM EDTA loading buffer and subjected to 10% dPAGE . Gels were visualized by staining with methylene blue . Mutant Kunjin viruses predicted to have defective sfRNA formation during infection were generated by overlap extension PCR . The FLSDX ( pro ) HDVr Kunjin virus infectious clone ( provided by the Khromykh laboratory; described in Liu et al . , 2003 ) was used as a PCR template for the construction of each mutant virus . The high fidelity Phusion Hot Start II polymerase ( Thermo Scientific , Pittsburg , PA ) was used for PCR using the mutagenic primers listed below and assembled PCR products were inserted into the Age1 and Xho1 restriction sites . Plasmids were screened for the proper mutations by sequencing PCR amplicons using the following primers: FLSDX Fw: 5′- actttgttaattgtaaataaatattgttat; FLSDX Rv: 5′-gcgtgggacgttgattcgcctttgt . Plasmids were linearized with Xho1 and in vitro transcriptions performed using the MEGAscript Sp6 transcription kit ( Life Technologies , Foster City , CA ) followed by Turbo DNase treatment to remove template . RNAs were phenol-chloroform-isoamyl alcohol ( 25:24:1 ) extracted and ethanol precipitation with ammonium acetate . The baby hamster kidney cell line BHK-21 ( ATCC CCL-10; Mesocricetus auratus ) was used to generate Kunjin virus stocks and cells were maintained in MEM plus 10% fetal bovine serum ( Atlas Biologicals , Fort Collins , CA ) and 1% streptomycin and penicillin ( Fisher Scientific-Hyclone , Logan , UT ) at 37°C in the presence of 5% CO2 . BHK-21 cells were electroporated and virus was amplified by passaging once on BHK-21 cells to generate working stocks . Viral titers were assessed by plaque titrations , infections were performed in human 293T cells ( MOI of 10 ) , cells were washed twice after a 2 hr adsorption period , and total cellular RNA was collected at 48 h . p . i using TRIzol ( Invitrogen , Carlsbad , CA ) . Total RNA was treated with DNase I ( Fermentas , Vilnius , Lithuania ) to remove residual genomic DNA . Northern blotting was performed as described previously ( Moon et al . , 2012 ) using two micrograms of total RNA from each sample using a probe to the entire 3′ UTR . For Kunjin virus sfRNA detection , 2 μg of total RNA from infected cells at 48 h . p . i . ( or from mock infected cells ) was separated on a 5% denaturing polyacrylamide gel . RNA was then transferred onto a nylon membrane ( Hybond-XL; GE Healthcare ) for blotting and UV cross-linked before blocking for 30 min at 60°C in hybridization solution ( 50% formamide , 1 mg/ml bovine serum albumin , 750 mM sodium chloride , 75 mM sodium citrate , 0 . 1 mg/ml salmon sperm DNA , 1% sodium dodecyl sulfate , 1 mg/ml polyvinylpyrrolidone , 1 mg/ml ficoll ) . In vitro transcribed , internally radiolabeled RNA probes to the entire 3′ untranslated region of Kunjin virus ( GenBank: AY274504 . 1 nts 10 , 396-11 , 022 ) were generated using a template containing the Kunjin 3′ UTR inserted into the pGEM-4 vector ( X65303 . 1; GenBank ) in the EcoRI and HindIII restriction sites using these primers: 5′-GAATTCTAAATACTTTGTTAATTGTAAAT; 5’-AAGCTTAGATCCTGTGTTCTCGCACCACCA . Following an overnight incubation at 60°C to hybridize the probe to the membrane , blots were washed twice with wash solution ( 300 mM sodium chloride , 0 . 1% sodium dodecyl sulfate , 30 mM sodium citrate ) and twice with stringent wash solution ( 30 mM sodium chloride , 0 . 1% sodium dodecyl sulfate , 3 mM sodium citrate ) for 30 min each at 60°C . Hybridized RNAs were visualized by exposing the blot on storage phosphor screens and imaging on the Typhoon Trio Imager ( GE Healthcare ) . Mutation/primer setForwardReverseFlanking primersCATACCGGTCGGAAAAGTGATCGACCTTGGCATCTCGAGCAATTGTTGTTGTTAACTTGG→C10519TTCCCGGCACCGGAAGTTGAGCTCAACTTCCGGTGCCGGGAAG→C10680TGGCGTGGCACTCTGCGGAGCTCCGCAGAGTGCCACGCCA | More than 40% of people around the globe are at risk of being bitten by mosquitoes infected with the virus that causes Dengue fever . Every year , more than 100 million of these individuals are infected . Many develop severe headaches , pain , and fever , but some develop a life-threatening condition where tiny blood vessels in the body begin to leak . If not treated quickly , this more severe manifestation of the illness can lead to death . There are currently no specific therapies or vaccines against Dengue or many other closely related viruses such as West Nile and Japanese Encephalitis . These viruses use instructions encoded in a single strand of RNA to take over an infected cell and to reproduce . The viruses also exploit an enzyme that cells use to destroy RNA to instead produce short stretches of RNA called sfRNAs that , among other things , may help the virus to avoid the immune system of its host . Understanding exactly how Dengue and other viruses thwart this enzyme—which is called Xrn1—may help scientists develop treatments or vaccines for these diseases . Chapman et al . have now shown that Dengue virus RNA contains a number of RNA elements that prevent it being completely degraded by the Xrn1 enzyme . In particular , a junction formed by three RNA helixes is critical for stopping the enzyme in its tracks , leaving the disease-associated sfRNA behind . A single mutation in the Dengue RNA disrupts the structure of the three-helix junction and allows the enzyme to completely destroy the RNA . A similar mutation was also made in the West Nile virus RNA and when human cells were infected with the mutated West Nile virus , the short sfRNAs were not produced . Treatments or vaccines targeting this structure may therefore help reduce illness associated with Dengue and related viruses . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"microbiology",
"and",
"infectious",
"disease"
] | 2014 | RNA structures that resist degradation by Xrn1 produce a pathogenic Dengue virus RNA |
Although many high-risk mucosal and cutaneous human papillomaviruses ( HPVs ) theoretically have the potential to synthesize L1 isoforms differing in length , previous seroepidemiological studies only focused on the short L1 variants , co-assembling with L2 to infectious virions . Using the multimammate mouse Mastomys coucha as preclinical model , this is the first study demonstrating seroconversion against different L1 isoforms during the natural course of papillomavirus infection . Intriguingly , positivity with the cutaneous MnPV was accompanied by a strong seroresponse against a longer L1 isoform , but to our surprise , the raised antibodies were non-neutralizing . Only after a delay of around 4 months , protecting antibodies against the short L1 appeared , enabling the virus to successfully establish an infection . This argues for a novel humoral immune escape mechanism that may also have important implications on the interpretation of epidemiological data in terms of seropositivity and protection of PV infections in general .
Human Papillomaviruses ( HPVs ) are widely distributed in nature and more than 220 types were sequenced up to date ( PaVE: Papillomavirus Episteme ) . They cannot only be divided in mucosal and cutaneous types ( Bzhalava et al . , 2013 ) , but also whether the infection is acquired , for example via sexual intercourse ( as for high-risk genital HPVs ) ( Gravitt and Winer , 2017 ) or whether a commensal cohabitation ( as for cutaneous HPVs ) occurred shortly after birth ( Antonsson et al . , 2003; Weissenborn et al . , 2009 ) . Depending on environmental factors ( e . g . chronic UV exposure ) ( Rollison et al . , 2019; Uberoi et al . , 2016 ) , the individual immune status ( e . g . systemic immunosuppression ) ( Reusser et al . , 2015; Vinzón and Rösl , 2015 ) or genetic predispositions ( e . g . EVER1/2 mutations in Epidermodysplasia verruciformis patients ) ( de Jong et al . , 2018 ) , commensal cutaneous papillomaviruses can induce hyperproliferative lesions ( e . g . actinic keratosis ) which may progress to squamous cell carcinomas ( SCCs ) ( Hasche et al . , 2018 ) . The African multimammate rodent Mastomys coucha represents a unique model system to investigate the consequences of a natural PV infection in the context of skin carcinogenesis ( Hasche and Rösl , 2019 ) . The animals become infected with Mastomys natalensis papillomavirus ( MnPV ) soon after birth ( Schäfer et al . , 2011 ) and seroconversion against viral proteins can be detected shortly afterwards ( Schäfer et al . , 2010 ) . MnPV is a typical cutaneous PV that resembles human β-types by lacking an E5 open-reading frame ( ORF ) ( Tan et al . , 1994 ) . Characterization of the viral transcriptome in productive lesions revealed a complex splicing pattern with different promoters and transcriptional start sites ( Salvermoser et al . , 2016 ) , also described for some HPV types ( Sankovski et al . , 2014; Wang et al . , 2011 ) or for the mouse papillomavirus type 1 ( MmuPV1 ) ( Xue et al . , 2017 ) . Most of these transcripts are polycistronic , allowing ( at least hypothetically ) the translation of several different ORFs ( Salvermoser et al . , 2016 ) . Using Mastomys coucha as a preclinical model , we could show that immunization with MnPV virus-like-particles ( VLPs ) induces a long-lasting neutralizing antibody response that completely prevents the appearance of skin lesions both under immunocompetent and immunosuppressed conditions ( Vinzón et al . , 2014 ) . Furthermore , Mastomys coucha also represents a paradigm for SCC development in the context of MnPV infection and UV exposure , thereby reflecting many aspects found in humans where a ‘hit-and-run’ mechanism during carcinogenesis is supposed ( Hasche et al . , 2017; Hasche et al . , 2018 ) . Virions of PVs consist of 72 pentamers of the major ( L1 ) protein together with up to 72 molecules of the minor ( L2 ) capsid protein ( Buck et al . , 2013; Hagensee et al . , 1993; Wang and Roden , 2013 ) . The L1 protein has the capability to spontaneously form regular structures ( capsomers ) , triggered by a thermodynamically favored self-assembly process ( McManus et al . , 2016 ) . Due to their repetitive structures , PV particles are very immunogenic and induce the generation of neutralizing antibodies that block viral entry into the host cell via binding to conformational epitopes on the capsid ( Kwak et al . , 2011; Wang and Roden , 2013 ) . Considering the cross-talk between viral infections and the immune system , PVs have developed multiple strategies to escape from immune surveillance ( Bordignon et al . , 2017 ) . While there is plenty of information about how innate immunity as the first line of defense is circumvented ( Christensen , 2016; Smola et al . , 2017 ) , less is known about the humoral immune response in terms of generation of protecting antibodies during the natural course of a PV infection . In the present study , we show that MnPV , as a rodent equivalent for cutaneous PVs in humans , induces a strong seroconversion in its natural host early after infection . However , the raised antibodies are non-neutralizing and directed against a longer isoform of the L1 protein which is unable to assemble into viral particles . Only after a delay of around 4 months after infection , protecting antibodies appear . This argues for a novel PV immune escape mechanism , probably providing a selective advantage to establish an efficient infection . We characterized this mechanism in greater detail since it may also have important implications in understanding the humoral immune response during a normal infection cycle in general .
Based on two previous studies comparing the presence of initiation codons within the papillomavirus L1 ORF ( Joh et al . , 2014; Webb et al . , 2005 ) , their position was aligned according to the PV genera derivation ( Bzhalava et al . , 2015; Van Doorslaer et al . , 2013; Figure 1 ) . Notably , alternative ATGs can be found in various mucosal ‘high-risk’ HPV types such as 16 , 18 , 45 , 52 , 56 , 58 , but not in ‘low-risk’ types such as HPV6 , 11 , 40 , 42 , 43 , 44 , respectively ( Webb et al . , 2005 ) . Additional in-frame initiation codons can also be detected in cutaneous HPV types of several genera such as HPV1 , 2 , 8 , 38 , 41 , 57 and 77 , respectively , of which HPV8 and HPV38 are considered to be ‘high-risk’ cutaneous HPVs ( Rollison et al . , 2019; Tommasino , 2017 ) . Accordingly , due to the presence of potential alternative translation initiation sites , different L1 isoforms could be translated . As shown in Figure 1 , almost all outlined PV L1 proteins harbor a consensus Wx7YLPP motif within the N-terminal region ( Joh et al . , 2014 ) , independently from PV genus or cancer risk assessment , while the remaining N-terminal sequences are not very conserved . For the majority of PV types , the nearest methionine codon to this motif is located one to three amino acids upstream of the tryptophan ( W ) in the consensus motif . However , there are also exceptions from this rule since HPV31 , 35 and 51 , for instance , harbor one additional ATG followed by an interspersed in-frame TAA stop codon , thereby probably preventing the synthesis of additional L1 isoforms ( Webb et al . , 2005 ) . Intriguingly , the MnPV L1 ORF also contains three alternative ATGs , which are located at nucleotide positions ( nt ) 5704 , 5725 and 5797 , potentially leading to the expression of L1LONG , L1MIDDLE and L1SHORT proteins , respectively . To examine serological responses against the three putative MnPV L1 isoforms , 60 naturally infected animals were monitored during different stages of viral infection ( 682 sera in total ) encompassing an age between 8 and 76 weeks . Since most serological detection methods developed to date are based on the L1SHORT isoform , firstly we examined seroconversion against L1SHORT by glutathione S-transferase ( GST ) -capture ELISA ( Kricker et al . , 2020; Sehr et al . , 2002; Waterboer et al . , 2009 ) . Notably , only few animals ( 8/60 ) exhibited measurable seroresponses against L1SHORT , initially at an age of 28 weeks . The mean seroreactivity per time point exceeded the cut-off earliest at an age of 68 weeks ( Figure 2A ) . Conversely , in 27 . 5% of the animals , broad seroresponses against MnPV L1LONG were already detectable as early as 8 weeks of age which increased to 52 . 5% of the animals at 76 weeks ( Figure 2B ) . In the main comparison at the latest time point where most animals were still alive ( 68 weeks ) , a significant difference was observed ( p<0 . 001 , two-tailed McNemar’s test , see Figure 2—figure supplement 1 ) . Seroreactivity against L1MIDDLE shows a similar time course as L1LONG which is consistent with the correlation between both ( Figure 2C and Figure 2—figure supplement 2 . Seroresponses against the E2 protein , which is involved in viral DNA replication ( McBride , 2013 ) and considered as an early marker of infection ( Schäfer et al . , 2011 ) , developed shortly after birth and increased during the study ( Figure 2D ) . Conversely , seroconversion against the minor capsid protein L2 appeared only in a few of the animals and as late as seroconversion against L1SHORT ( Figure 2E ) . In order to exclude possible experimental bias merely working with a GST-fusion protein-based L1SHORT ELISA , additional ELISAs were performed using MnPV VLPs ( derived from L1SHORT ) produced via baculovirus expression system ( Christensen et al . , 1994; Rose et al . , 1993 ) . In accordance with the GST-L1SHORT ELISA , serum responses against VLPs were absent in early infection stages and did not exceed the cut-off before an age of 20 weeks ( Figure 2F ) . This suggests that anti-L1LONG antibodies during early infection fail to recognize epitopes of both GST-L1SHORT antigen and on the surface of intact VLPs . Accordingly , there is no correlation of anti-GST-L1LONG with anti-GST-L1SHORT or anti-VLP reactivity ( Figure 3A and B ) . Conversely , a significant correlation between GST-L1SHORT and VLP-ELISA ( Figure 3C ) strengthens the notion that the absence of a correlation between GST-L1LONG and L1SHORT ELISAs was indeed due to altered serological properties of L1 isoforms rather than due to different ELISA methodologies . Due to the different temporal order of seroconversion against L1LONG , L1SHORT and VLPs , we reasoned that MnPV escapes from adaptive immunity to establish an efficient infection and to maintain a persistent life cycle , which is indicated by the increased seroresponse against MnPV E2 ( Figure 2D ) . To get insight into this question , pseudovirion-based neutralization assays ( PBNA ) ( Pastrana et al . , 2004; Roden et al . , 1996; Vinzón et al . , 2014 ) were performed to monitor for the presence of protecting antibodies . As shown in Figure 4A , PBNA revealed a similar kinetics as previously demonstrated for the L1SHORT isoform , indicating that neutralizing antibodies in fact appeared delayed . This was further substantiated by correlating the serum titers measured by VLP-ELISA with data obtained by PBNA ( Figure 4B ) . Conversely , all sera directed against L1LONG ( but negative for L1SHORT ) lack neutralizing capability ( Figure 4C ) . In order to identify epitopes within L1 recognized by the sera , synthetic linear 15-mer peptides with 14 residue overlaps were spotted on microarrays . Incubation with a Mastomys serum mix obtained from five tumor-bearing animals , possessing high titers against L1LONG and L1SHORT identified three immunogenic epitopes within the region homologous between L1LONG and L1SHORT ( ITGHPLY , DYLGMSK and KRSLPASRN ) ( Figure 5A and B ) . Notably , two of them ( ITGHPLY , DYLGMSK ) coincide with the DE and FG loops ( Figure 5C; Bissett et al . , 2016 ) that form conformational epitopes on the surface of HPV virions ( Li et al . , 2017; Zhang et al . , 2016 ) and are known to be highly immunogenic . To further elucidate why anti-L1LONG antibodies lack neutralizing capacity , we dissected the seroresponse against L1 with respect to the 31 amino acids present only at the N-terminus of L1LONG ( Figure 1 ) . Analyzing 297 sera of 39 L1LONG-positive animals by GST-ELISA , no detectable or only weak positivity could be measured against this L1LONGaa1-31 ( Figure 6A ) . In contrast , when extending these 31 amino acids of L1LONG to 41 residues ( which includes nine residues of L1SHORT ) , the number of seropositive animals strikingly increased ( Figure 6B ) , whereas such a reactivity could not be observed in sera from MnPV-free animals vaccinated with VLPs ( made from L1SHORT ) obtained in a previous study ( Vinzón et al . , 2014; Figure 6C ) . Correlating reactivities between L1LONG and L1LONGaa1-41 ( Figure 6D ) shows that actually all sera which are positive for L1LONGaa1-41 are also positive for L1LONG , ( which is not the case for L1SHORT , see Figure 6—figure supplement 1 ) . It is therefore reasonable to assume that this antibody population is likely arising from exposure of the immune cells to the L1LONG antigen . These data , together with the finding that the first linear epitope recognized on the peptide array is located in L1SHORT ( Figure 5 ) , indicate that a conformational epitope is spanning L1LONG and L1SHORT , and represents both the most immunogenic epitope in L1LONG and a major immunogen in early stages of infection . Moreover , an algorithm that predicts antigenic sites on proteins ( Kolaskar and Tongaonkar , 1990 ) , calculates the abovementioned DE loop ( determinant no . 8 ) and partially the FG loop ( determinant no . 12 ) ( Figure 5—figure supplement 1 ) detected by the serum mix ( see Figure 5 ) and also predicts an N-terminal epitope between residues 11 and 34 ( determinant no . 1 ) . To further characterize the antigenic properties and to prove that a conformational epitope is formed at the N-terminus of L1LONG , ELISAs with denatured antigens ( VLPs or GST fusion proteins ) were performed . For this purpose , a panel of monoclonal antibodies raised against MnPV VLPs was used . These antibodies differ in terms of neutralization and sensitivity in VLP and GST-ELISAs ( Supplementary file 2 ) . Of these , only mAb 2E2 , 2D11 and 3H8 showed high reactivity against native VLPs ( Figure 7A , grey , yellow , and green lines ) , indicative for the recognition of conformational epitopes . Conversely , mAb 2D6 and 5E5 that possess low binding to intact VLPs were expected to represent antibodies recognizing linear epitopes ( Figure 7A , blue and purple lines ) . To test this assumption , we anticipated a reversed pattern upon denaturation where the latter antibodies can bind , while binding of mAb 2E2 , 2D11 and 3H8 should be abrogated . As shown in Figure 7B , this was indeed the case ( see also Supplementary file 2 ) . Sera from the collective of 60 naturally infected animals previously tested positive for L1LONG and L1SHORT were also tested in denatured-VLP ELISA ( 306 sera ) and denatured-GST-L1 ELISA ( 281 sera ) . Interestingly , denaturation of both VLPs and GST-L1 antigens abrogated the reactivity of all sera ( Figure 7C–E ) , suggesting that anti-L1LONG and anti-L1SHORT antibodies were indeed directed against conformational epitopes . To test the different isoforms for their capability to form virus-like structures , the ORFs of L1SHORT , L1MIDDLE and L1LONG were expressed in Sf9 insect cells by the use of recombinant baculoviruses . Different preparations were analyzed by CsCl density gradient centrifugation . For quality control , the gradients’ refractive indices were measured and corresponding fractions were analyzed via western blot , where all three L1 isoforms could be found , ranging between 55 and 70 kDa ( Figure 8—figure supplement 1 ) . To analyze the ability of the isoforms to form VLPs or similar structures , different fractions of the gradients were examined by EM . Considering L1SHORT , highly concentrated and spherically shaped particles with sizes of 60 nm and clearly visible capsomers could be found ( Figure 8A ) . Also in the lowest density fractions ( e . g . fraction 11 ) , particles with capsomer-like structures were detected . Conversely , inspecting the gradients of L1LONG and L1MIDDLE , similar assemblies were absent , although many particles of different sizes ( 20–50 nm ) were found ( Figure 8B and C ) . However , it is difficult to classify these substructures as regular capsomers , indicating that L1LONG and L1MIDDLE were apparently not able to form correctly assembled VLPs under the same experimental conditions . To examine whether the addition of L2 , regularly present in mature infectious virions , can facilitate the formation of L1LONG or L1MIDDLE composed particles , pseudovirions were produced . 293TT cells were transfected with expression plasmids encoding L2 and the different L1 isoforms in conjunction with a reporter plasmid . The assembled structures were purified via Optiprep gradients and infection assays with different fractions were performed . In contrast to L1SHORT ( Figure 8D ) , L1LONG and L1MIDDLE again did not yield virus-like structures even in the presence of L2 ( Figure 8E and F ) . Moreover , while infectious L1SHORT-based pseudovirions can be found in most of the gradient fractions , no signals of the reporter construct could be discerned for L1LONG and L1MIDDLE . Luciferase signals could be measured in non-fractionated 293TT cell lysates , indicating that the transfection was successful for all three L1 isotypes . To further investigate L1 isoform expression in their genuine host in vitro , Mastomys-derived fibroblasts ( Hasche et al . , 2016 ) were transfected either with plasmids exclusively encoding HA-tagged L1SHORT , L1MIDDLE or L1LONG or with the polycistronic plasmid vL1 encoding the genuine viral L1 ORFs as found in natural transcripts . While serum mix from tumor-bearing animals was able to detect all L1 isoforms in immunofluorescence stainings , mAb 2D11 ( which exclusively binds a conformational L1SHORT epitope , see Figure 7D and E , respectively ) could only detect L1SHORT but not L1MIDDLE and L1LONG , despite similar expression levels of all isoforms ( see HA-tag ) ( Figure 9A ) . This indicates that in Mastomys cells only the L1SHORT can form structures involved in the formation of virus particles . Western blot analysis of the corresponding cells showed similar expression of all HA-tagged L1 isoforms , which again could be detected by the serum mix ( Figure 9B ) . Notably , only in L1SHORT-transfected cells both mAb 2D11 and serum mix detected bands between 100 and 130 kDa and around 250 kDa , which disappeared in lysates treated with additional DTT , β-mercaptoethanol and extended heating prior to SDS-PAGE separation ( Figure 9C ) . Their sizes correlate to L1-dimers and trimers and since mAb 2D11 does not detect linear epitopes , this suggests that only the L1SHORT isoform is able to form capsomer-like structures , which are at least partially structured in non-reducing conditions due to stabilizing properties of inter-capsomeric disulfide bridges ( Buck et al . , 2005b; Sapp et al . , 1998 ) . Previous viral transcriptome analysis revealed the presence of three polycistronic transcripts ( referred to as Q , R and S ) that have the potential to encode both L1LONG and L1SHORT ( Salvermoser et al . , 2016 ) . Using the most abundant of these ( transcript Q ) for prediction of the start codons’ likelihood of being used for translation initiation ( Nishikawa et al . , 2000 ) , it turned out that the ORFs of E1^E4 ( reliability index , RI = 0 . 43 ) , L2 ( RI = 0 . 42 ) and L1LONG ( RI = 0 . 38 ) could be favored over L1SHORT ( RI = 0 . 17 ) ( Supplementary file 3 ) . Indeed , consistent with the immunofluorescence , when transfecting cells with the polycistronic construct vL1 , L1LONG as well as L1SHORT and its multimer band are detected , which confirms that both ORFs are functional and that L1LONG , although unable to form capsomers , can be expressed from such a polycistronic construct ( Figure 9 ) . Considering the PV life cycle , virions are released by shedding of terminally differentiated cells at the uppermost layer ( stratum corneum ) of the epidermis . Here , L1 and L2 are assembled to capsomers and virus particles ( Figure 10A ) . When staining MnPV-induced papillomas with serum that detects mature MnPV virions , virus particles can only be found in cornified structures above or within the epidermis and in islands of terminally differentiated keratinocytes ( Figure 10B ) . This is consistent with the expression of L2 , which also appears the earliest in nearly shed cells ( Figure 10C ) . Conversely , using serum from mice immunized with the N-terminal peptide exclusive for L1LONG , keratinocytes in the basal layer and the complete epithelium are positively stained ( Figure 10B ) . This suggests that its expression takes place already during early infection phases long before viral particles are formed .
The non-random distribution of several ATG initiation codons within the L1 open reading frame ( ORF ) of certain human and animal papillomaviruses ( Figure 1 ) potentially allows the translation of different L1 isoforms . Despite plenty of seroepidemiological studies on HPV , to our knowledge , seroresponses against different L1 isoforms have never been performed . Previous studies only focused on the shortest variant , known to efficiently form virus-like particles ( VLPs ) or , together with L2 , infectious virions ( Buck et al . , 2005b ) . The MnPV-infected rodent Mastomys coucha is a reliable preclinical model that mimics many aspects of the situation found in humans infected with cutaneous HPVs . Since the animals are immunocompetent and become naturally infected early in life ( Hasche and Rösl , 2019; Hasche et al . , 2018 ) , they represent a unique virus-host system to investigate the humoral response against MnPV L1 isoforms during the natural course of infection . Of note was the finding that seroconversion against the isoforms L1LONG and L1MIDDLE appeared strikingly earlier than for L1SHORT ( Figure 2 ) , raising the question of the selective advantage for the virus and in turn its permissive cycle . To address this issue , pseudovirion-based infection assays were performed . Here , only sera binding L1SHORT inhibited pseudovirion infection , while sera recognizing the L1LONG and L1MIDDLE isoforms were not neutralizing ( Figure 4 ) . Furthermore , in contrast to L1SHORT , the longer isoforms were also not able to form intact viral particles ( Figure 8 ) . Conversely , the time course of MnPV E2 seroreactivity ( Figure 2D ) in comparison to the appearance of neutralizing antibodies ( Figure 4 ) is indicative for viral replication and spread of infection ( Schäfer et al . , 2011; Xue et al . , 2010 ) . It is therefore reasonable to assume that the delay of neutralization capacity apparently allows the virus to accumulate . Pointing in this direction is also the kinetics of seroconversion against the L2 minor capsid protein ( Figure 2E ) , finally assembling with L1 to form new infectious progeny virions at the end of the permissive cycle . Moreover , the correlation of the different ELISAs showed an obvious consistency between VLP and GST-L1SHORT that is not maintained when VLP or GST-L1SHORT are compared with GST-L1LONG ( Figure 3 ) . As reported from mice immunized with recombinant HPV VLPs , neutralizing antibodies act via binding to linear or conformational epitopes ( Combita et al . , 2002; Senger et al . , 2009a; Zhang et al . , 2016 ) . Although most high-affinity neutralizing antibodies bind to conformation-dependent epitopes , antibodies recognizing linear epitopes can be raised against improperly maturated VLPs ( used for immunization ) or virions ( in the course of a natural infection ) ( Senger et al . , 2009a ) . Similar to HPV16 L1 ( Bissett et al . , 2016; Zhang et al . , 2016 ) , immunodominant epitopes could be identified within the DE and FG loops of MnPV L1 ( Figure 5 ) , known to form highly immunogenic conformational epitopes on the surface of virions ( Li et al . , 2017; Zhang et al . , 2016 ) . Interestingly , we identified a novel epitope in the N-terminal region of L1 , spanning at least some of the residues exclusive for L1LONG and extending to the first amino acids from L1SHORT . Seroconversion against this peptide ( referred to as L1LONGaa1-41 in Figure 6B ) followed roughly the same kinetics as L1LONG ( Figure 2B ) and most L1LONG-positive sera recognized these 41 residues ( Figure 6D ) . However , all sera only positive for L1LONG , but negative for L1SHORT did not neutralize PsVs ( Figure 4C ) and further lost their ability to bind to L1LONG when the antigen was denatured prior to the ELISA ( Figure 7D ) . Hence , the addition of N-terminal residues to L1SHORT , obviously leads to a distinct folding of L1LONG , thereby forming a new conformational epitope , which is not present in L1SHORT-derived VLPs and natural virions that do not induce such antibodies upon immunization or infection ( Figure 6C ) . Due to the fast kinetics of seroconversion against L1LONG , this conformational epitope is apparently the predominant immunogenic structure in L1LONG and most probably masks neutralization epitopes that would be accessible in VLPs ( Figures 2F and 4A ) . Indeed , in the case of HPV16 L1 , the length of the N-terminus is decisive for efficient and correct assembly of L1 into VLPs ( Chen et al . , 2000 ) . Consistently , neither MnPV L1LONG nor L1MIDDLE formed particles in the right size and shape in insect cells ( Figure 8 ) , a finding also reported for the mouse papillomavirus MmuPV L1 ( Joh et al . , 2014 ) . Moreover , while co-expression of MnPV L1SHORT and L2 results in well-shaped infective pseudovirions , L1LONG and L1MIDDLE strikingly inhibited pseudovirion formation ( Figure 8 ) , suggesting that the N-terminus is literally hindering the assembly of both isoforms to capsids . Previously , crystallization of HPV L1 for 3D structure determination was not successful with the ‘usual’ L1 ( being L1SHORT ) but required its N-terminal truncation ( Chen et al . , 2000 ) , probably because the crystallization conditions lead to a folding of the N-terminus that prevents particle formation . Therefore , even the most complete 3D model 3J6R for HPV16 L1 ( Cardone et al . , 2014 ) lacks eight N-terminal residues . It was speculated that the N-terminus of L1 fills a gap in interpentameric structures ( Figure 8—figure supplement 2 ) and since N-terminal deletion of eight residues has such a big impact on particle formation ( Chen et al . , 2000 ) it is likely that the additional 31 residues do not fit into this gap and furthermore completely sterically hinder L1 assembly . Additionally , the extended N-terminus not only inhibits assembly , but also causes a complete new folding of L1 , thereby exposing a novel immunogenic conformational epitope that induces non-protective antibodies . Nonetheless , using Mastomys-derived cells as genuine in vitro system to monitor expression of the three MnPV L1 variants ( Figure 9 ) , we found that L1SHORT but not the larger isoforms are recognized by mAb 2D11 , confirming the presence of conformational epitopes found in capsomers . Importantly , this experiment also showed that L1LONG and L1SHORT ORFs are functional in a polycistronic setting and can be translated in Mastomys cells . PV late gene expression is tightly regulated at transcriptional and post-transcriptional level ( Graham , 2017 ) . The MnPV transcription map obtained from skin lesions revealed L1 transcripts to be the most abundant in productive infections , which are encoded by polycistronic mRNAs ( coding for E1^E4 , L2 and L1 ) exclusively controlled by the late viral promoter ( Salvermoser et al . , 2016 ) . However , speculating about the mode of L1 isoform regulation , the following scenario could be envisioned: late PV transcripts are known to be strongly upregulated upon keratinocyte differentiation . However , due to insufficient suppressed late viral promoter activity , late transcripts can already be detected in undifferentiated cells ( Songock et al . , 2017 ) . Indeed , in contrast to mucosal types ( e . g . HPV11 , 16 and COPV ) , cutaneous papillomaviruses ( e . g . HPV1 , 2 and BPV1 ) initiate their late functions in the lower epithelial regions ( Peh et al . , 2002 ) . Consequently , this could lead to synthesis of the highly immunogenic L1SHORT already in early stages of infection , giving rise to neutralizing antibodies , thereby counteracting the permissive cycle and in turn the viral spread . However , the predicted presence of a stronger Kozak sequence in a upstream ORF from the conventional L1 ( L1SHORT ) ( Supplementary file 3 ) , generates an alternative longer isoform of L1 which is expressed already in basal and suprabasal layers of MnPV-induced lesions ( Figure 10 ) and similar to the pattern of L1 mRNA expression in MmuPV1-induced papillomas ( Xue et al . , 2017 ) . This L1LONG isoform is also immunogenic but does not raise neutralizing antibodies . Conversely and in agreement with other studies ( Biryukov and Meyers , 2015; Borgogna et al . , 2014 ) , capsid formation could only be found in the granular and uppermost layers ( Figure 10 ) , which are less accessible for immune effector cells . However , based on in silico prediction , the capsid-forming L1SHORT cannot be efficiently translated from polycistronic transcripts coding for E1^E4 , L2 and L1 ( Supplementary file 3 ) , reinforcing the existence of a transcript in which the splicing acceptor is located immediately upstream of the L1SHORT AUG ( Salvermoser et al . , 2016 ) . This splicing event is more tightly regulated than late promoter activity , based on a splicing silencer sequence suppressing the use of this site during mRNA processing ( Zhao et al . , 2004 ) . Thus , only the combination of splicing regulation and the presence of favored ORFs ensure L1SHORT synthesis in upper layers of the epithelium ( Zhao et al . , 2004 ) . This may allow a better escape from immune surveillance until the regular permissive cycle is completed . Finally , the question remains why only certain PVs produce an immunogenic L1 isoform that is not needed for its own life cycle . Considering a virus-host interaction as consequence of an evolutionary process , there is a sophisticated balance between host immune surveillance and immune escape by the respective virus to ensure virus progeny production ( French and Holmes , 2020 ) . The temporal and spatial change of such an equilibrium in favor to infection determines the efficiency of viral accumulation and maturation as well as the spread to infect another host ( Rothenburg and Brennan , 2020 ) . The occurrence of several L1 ORFs is not a peculiarity of MnPV , but also found in different HPV genera as well as in animal papillomaviruses such as MmuPV ( Joh et al . , 2014; Webb et al . , 2005 ) . It is intriguing that alternative ORFs are mostly present in mucosal ‘high-risk’ HPV types that cause clinical symptoms , but not in ‘low-risk’ types such as HPV6 and 11 ( Webb et al . , 2005 ) . Hence , there must be a selection pressure to maintain several L1 initiation codons in these PVs types , thereby allowing the synthesis of different isoforms to get an advantage for the virus . Naturally occurring mutations within neutralizing epitopes that reduce the antigenicity of HPV L1 and L2 proteins may contribute to humoral immune evasion ( Seitz et al . , 2013; Yang et al . , 2005 ) . Such L1 variants , isolated from premalignant HPV16 positive cervical tissue , showed impaired viral capsid assembly that can influence both B cell class switching and the production of non-neutralizing antibodies ( Yang et al . , 2005 ) . Indeed , assembly-defective HPV16 VLPs impair the activation of dendritic cells that play a decisive role in activating adaptive immunity ( Yang et al . , 2005 ) . Whether this also explains the late development of HPV16 L1 neutralizing antibodies ( eight and nine months after the first positive HPV DNA detection ) ( Gutierrez-Xicotencatl et al . , 2016 ) remains to be elucidated . In conclusion , it is tempting to speculate that our results show that early synthesis of alternative immunogenic L1 isoforms represents a novel mode of humoral immune escape mechanism , favoring persistent infections and viral spread due to a delay of immune recognition by the host .
The Mastomys coucha breeding colony naturally infected by MnPV is maintained under SFP conditions in individually ventilated cages ( Tecniplast GR900 ) at 22+/- 2°C and 55+/- 10% relative humidity in a light/dark cycle of 14/10 hr . Mastomys were fed with mouse breeding diet and allowed access to water ad libitum . For the follow-up experiment , animals were monitored for the duration of their lifetime until they had to be sacrificed due to tumor development or decrepitude . Blood was taken in intervals from 2–8 weeks by puncturing the submandibular vein of anesthetized animals ( 3% isoflurane ) , starting at the age of eight weeks . 293TT , HeLaT and MaFi132 cells were grown in DMEM supplemented with 10% fetal calf serum ( FCS ) , 1% Penicillin/Streptomycin and 1% L-glutamine . Media of HeLaT and 293TT cell were further supplemented with Hygromycin B ( 125 μg/ml ) to maintain additional SV40 large T-antigen expression . All cell lines were kept at 37°C , 5% CO2 and 95% humidity and regularly checked for Mycoplasma via PCR . Sf9 and TN-High Five insect cells were kept as described elsewhere ( Senger et al . , 2009b ) . As previously described ( Schäfer et al . , 2011; Schäfer et al . , 2010 ) , glutathione-casein was diluted in 50 mM carbonate buffer ( pH9 . 6 ) and 200 ng/well were coated overnight at 4°C to 96 well plates ( Nunc PolySorp ) . After blocking with 180 µl/well casein blocking buffer ( CBB , 0 . 2% casein in PBST: 0 . 05% Tween-20 in PBS ) for 1 hr at 37°C , the plate was incubated with the respective antigen ( bacterial lysate containing the GST-antigen-SV40-tag fusion protein ) for 1 hr at RT . Mastomys sera diluted 1:50 in CBB containing GST-SV40-tag were incubated for 1 hr at RT to remove unspecific reaction against bacterial proteins or the GST-SV40-tag fusion protein . Afterwards , ELISA plates were washed four times with PBST and incubated with pre-incubated sera for 1 hr at RT . After washing four times , 100 µl/well HPR-conjugated goat anti-mouse IgG ( H+L ) antibody ( Promega , 1:10 , 000 in CBB ) were applied for 1 hr at RT . Antibodies were quantified colorimetrically by incubating with 100 μl/well substrate buffer for 8 min ( 0 . 1 mg/ml tetramethylbenzidine and 0 . 006% H2O2 in 100 mM sodium acetate , pH6 . 0 ) . The enzymatic reaction was stopped with 50 μl/well 1 M sulfuric acid . The absorption was measured at 450 nm in a microplate reader ( Labsystems Multiskan , Thermo Fisher Scientific ) . To calculate the serum reactivity against the respective antigen , sera were tested in parallel against the GST-SV40-tag fusion protein and the reactivity was subtracted from the reactivity against the GST-antigen-SV40-tag . Each ELISA was performed in duplicates at least . The cut-offs were calculated individually for each antigen by measuring sera of virus-free animals . VLP-ELISAs were performed as described elsewhere ( Vinzón et al . , 2014 ) . Briefly , 100 ng/well purified high quality L1SHORT-VLPs were coated overnight in 50 mM carbonate buffer pH9 . 6 and blocked with CBB the next day . After incubation for 1 hr at RT with three-fold dilutions of sera in CBB , plates were washed four times with PBST and incubated with goat anti-mouse IgG-HRP ( 1:10 , 000 in CBB ) . After four washes , color development and measurement was performed as described for the GST-ELISA . Antibody titer represents the last reciprocal serum dilution above the blank . VLPs and GST fusion antigens were denatured at 95°C for 10 min in coating buffer ( 50 mM carbonate buffer pH9 . 6 ) prior to coating to ELISA plates overnight at 37°C . For the denatured VLP-ELISA , further steps were carried out according to the VLP-ELISA protocol described above . Denatured GST-ELISA antigens were then directly coated onto the ELISA plates overnight at 37°C and further steps ( blocking , washing , incubation with sera , color reaction ) were carried out according to the GST-ELISA protocol . The five monoclonal MnPV anti-L1SHORT antibodies mAb2E2 , mAb2D6 , mAb2D11 , mAb5E5 and mAb3H8 ( reactivities shown in Supplementary file 2 ) were generated via hybridoma technique from BALB/c mice vaccinated with MnPV L1SHORT-VLPs and were used together with a Mastomys serum mix ( sera from five tumor-bearing animals ) as controls for denaturation conditions . The peptide array ( PEPperPRINT GmbH , Germany ) was produced as previously described ( Stadler et al . , 2008 ) . The amino acid sequence of MnPV L1 was elongated with neutral GSGSGSG linkers at the C- and N-termini to avoid truncated peptides . Elongated antigen sequences were translated into 15 aa peptides with peptide-peptide overlaps of 14 aa . The resulting peptide microarray contained 530 different overlapping L1 peptides printed in duplicates and framed by additional HA ( YPYDVPDYAG , 86 spots ) control peptides . The peptide array was incubated for 10 min in PBST , followed by incubation in Rockland Blocking Buffer MB-070 ( RBB; Rockland Immunochemicals , USA ) for 1 hr . After short rinsing with PBST , the array was incubated for 16 hr at 4°C with Mastomys serum mix at a dilution of 1:300 in 10% RBB in PBST . The array was washed three times for 1 min with PBST and then incubated for 1 hr at RT with 0 . 2 µg/ml 10% RBB in PBST goat anti-mouse IgG ( Fc ) DyLight680 ( Rockland Immunochemicals , USA ) . Subsequently , the array was washed three times for 1 min with PBST and rinsed with 1 mM TRIS-HCL pH7 . 4 . As peptide controls , HA peptide spots were stained with monoclonal mouse-anti-HA IgG antibody ( 12CA5 , kindly provided by Dr . G . Moldenhauer , DKFZ ) conjugated with DyLight800 ( Lightning-Link , Innova Biosciences , UK ) , followed by washing as described above . The antibody was diluted to 1 µg/ml in 10% RBB in PBST and staining was performed for 1 hr at RT in the dark followed by washing as described above . After drying of the array , fluorescence images were acquired with an Odyssey Infrared Imager ( LICOR , USA ) at a resolution of 21 µm . Scanner sensitivity was set to 7 . 0 for the 700 and 800 nm channels respectively , the focal plane was set to +0 . 8 mm . Quantification of spot intensities , based on 16-bit gray scale tiff files and microarray image analysis , via PepSlide Analyzer ( SICASYS Software GmbH , Germany ) . A software algorithm breaks down fluorescence intensities of each spot into raw , foreground and background signal , and calculates averaged median foreground intensities and spot-to-spot deviations of spot duplicates . Averaged spot intensities of the assays with the sample were plotted against the antigen sequence from N- to C-terminus to visualize overall spot intensities and signal-to-noise ratios ( intensity plot ) . Pseudovirions were produced as previously described ( Buck and Thompson , 2007 ) . 293TT cells were co-transfected by calcium phosphate transfection with plasmids encoding humanized MnPV L1 isoforms ( L1LONG , L1MIDDLE and L1SHORT ) , L2 and a reporter plasmid encoding Gaussia luciferase . The 2nd and 3rd ATG of L1LONG and the 2nd ATG of L1MIDDLE were mutated to GCG ( alanine ) to exclusively guarantee L1LONG or L1MIDDLE expression . Transfected cells were incubated for 48 hr , harvested and resuspended in an equal volume of PBS and supplemented with 0 . 5% Brij 58 ( Sigma ) and 1% RNase A/T1 mix ( Thermo Fisher Scientific ) . Cells were lysed for 24 hr under rotation at 37°C to allow pseudovirion maturation prior to adjustment with 5 M NaCl to 0 . 85 M NaCl and treatment with 700 U Benzonase ( Merck ) for 1 hr at 37°C . For purification , the lysate was transferred on top of a three-step gradient of 27% , 33% and 39% Iodixanol ( Optiprep , Sigma ) diluted in 0 . 8 M NaCl/DPBS . and centrifuged at 37 , 000 rpm for 5 hr at 16°C in a swinging bucket rotor . Fractions of 500 µl each were collected in siliconized LoBind tubes ( Eppendorf ) and quantity and quality of pseudovirions in each fraction was assessed by electron microscopy ( EM ) and Gaussia luciferase reporter activity after infection of HeLaT cells . As previously described ( Buck et al . , 2005a ) , animal sera ( in duplicates , initial dilution 1:60 in medium ) were subjected to 1:3 serial dilutions in 96-well cell culture plates ( Greiner Bio-One GmbH ) . Then , the sera were mixed with 40 µl of diluted pseudovirions and incubated for 15 min at RT . Then , 50 µl of 2 . 5 × 105 HeLaT cells/ml were seeded to the pseudovirion-serum mixture and cultured for 48 hr at 37°C . The activity of secreted Gaussia luciferase was measured 15 min after adding coelenterazine substrate and Gaussia glow juice ( PJK Biotech , Germany ) according to the manufacturer’s instructions in a microplate luminometer reader ( Synergy 2 , BioTek ) . The neutralization titer represents the reciprocal of the highest dilution that reduces the signal by at least 50% . Two copies of MnPV wildtype L1LONG , L1MIDDLE and L1SHORT were inserted into the Multibac vector pFBDM using EcoRI/HindIII and XmaI/SphI . Comparable to the VLP production , to ensure that only L1LONG and L1MIDDLE are expressed in the Multibac expression system , the 2nd and 3rd ATG start codons of L1LONG and the 2nd ATG of L1MIDDLE were mutated to GCG ( alanine ) . The recombinant MultiBac bacmids were generated by electroporation ( 1 . 8 kV pulse ) of DH10MultiBacCre E . coli with the generated plasmids , followed by selection with antibiotics and blue/white screening ( Fitzgerald et al . , 2006 ) . Recombinant MultiBac bacmids were isolated by QIAGEN Plasmid Mini Kit followed by ethanol precipitation . Recombinant baculoviruses were generated as previously described ( Vinzón et al . , 2014 ) with some modifications . One µg of MultiBac bacmid containing L1LONG , L1MIDDLE or L1SHORT was diluted in 1 ml transfection buffer ( 25 mM Hepes , 125 mM CaCl2 , 140 mM NaCl , pH7 . 2 ) and added dropwise to Sf9 cells . After incubation at 27°C for 5 hr , cells were washed twice and then cultured for 6 days in supplemented TNM-FH medium ( Sigma ) . One ml supernatant was used for generation of a high-titer baculovirus stock by infecting 2 × 106 Sf9 cells in a T25 flask followed by virus amplification for 6 days . This step was repeated with the obtained supernatant two times with increasing cell numbers ( 3 ml supernatant for 1 × 107 cells in a T75 flask and 5 ml supernatant for 2 . 5 × 107 cells in a T175 flask ) . TN-High Five cells were cultivated to a density of 2 . 5 × 106/ml in 250 ml suspension culture , which were then pelleted and resuspended in 42 ml EX-CELL 405 serum-free medium ( Sigma ) and 8 ml high titer virus stock . The cells were shaken at a low speed for 1 hr at RT and then incubated within 250 ml final volume of medium for 3 days at 27°C . Cell pellets were harvested by centrifugation ( 3000 rpm for 10 min at 4°C in a Sorvall GS-3 rotor ) and washed in pre-chilled PBS for two times . Dry pellets were resuspended in 10 ml VLP extraction buffer ( 5 mM MgCl2 , 5 mM CaCl2 , 150 mM NaCl , 0 . 01% Triton X-100 and 20 mM Hepes pH7 . 4 ) containing 200 μl 100 mM PMSF , and then followed by three times sonication . A two-step gradient consisting of 7 ml of 40% sucrose on top of 7 ml CsCl solution was prepared . Clear cell lysate was obtained by centrifugation ( 10 , 000 rpm for 10 min at 4°C in a Sorvall F-28/50 rotor ) and carefully loaded onto the top of the CsCl layer . After centrifugation ( 27 , 000 rpm for 3 hr at 10°C in SW-31Ti rotor ) , the interphase between sucrose and CsCl together with the complete CsCl layer was transferred into a Quickseal tube . The fractions were collected in 1 ml aliquot after 16 hr centrifugation at 48 , 000 rpm at 20°C in a Beckman 70Ti rotor and analyzed by Coomassie blue dye and Western blot . Small aliquots from the fraction with highest and lowest protein yield were dialyzed against H2O on a membrane filter and analyzed by EM . VLP and PsV preparations or tissue were fixed with buffered aldehyde solution ( 2% formaldehyde , 2% glutaraldehyde , 1 mM MgCl2 , 2% sucrose in 100 mM calcium cacodylate , pH7 . 2 ) , followed by post-fixation in buffered 1% OsO4 , graded dehydration with ethanol and resin-embedding in epoxide ( 12 g glycid ether , 6 . 5 g NMA , 6 . 5 g DDSA , 400 μl DMP30 , all from Serva , Germany ) . Ultrathin sections at nominal thickness 60 nm and contrast-stained with lead-citrate and Uranylacetate were observed in a Zeiss EM 910 at 100 kV ( Carl Zeiss , Oberkochen , Germany ) and micrographs were taken with image-plates , scanned at 30 µm resolution ( Ditabis micron , Pforzheim , Germany ) . Variants of L1 ORFs were cloned into pPK-CMV-E3 expression plasmids . For exclusive and strong eukaryotic expression of the respective L1 isoform , viral codons were humanized , unwanted start codons mutated ( L1SHORT starts from 3rd ATG; L1MIDDLE starts from 2nd ATG , 3rd ATG mutated; L1LONG starts from 1st ATG , 2nd and 3rd ATG mutated ) and the ORFs were cloned downstream of an artificial Kozak sequence . Alternatively , the complete genuine viral L1 ORF encoding start codons and Kozak sequences of all isoforms was cloned and termed pPK-CMV_MnPV-vL1 . MaFi132 cells ( 400 , 000 cells/10 cm dish ) were transfected with 5 µg pPK-CMV_MnPV-L1SHORT or either 10 µg pPK-CMV_MnPV-L1MIDDLE , pPK-CMV_MnPV-L1LONG or pPK-CMV_vL1 24 hr after seeding using TurboFect ( Thermo Fisher Scientific ) according to the manufacturer’s protocol . Cells were collected 48 hr after transfection , washed in PBS and lysed for 30 min on ice in 1 . 25x Laemmli buffer ( 78 mM Tris pH6 . 8 , 2 . 5% SDS , 6 . 25% glycerol , 0 . 125% bromophenol blue , 2 . 5% β-mercaptoethanol ) . Lysates were then heated at 95°C for 5 min , chilled on ice and treated with 100 U/ml Benzonase ( Millipore ) for 5 min at RT . Protein concentrations were measured using a NanoDrop spectrophotometer . Forty μg lysate/lane were loaded to 8% SDS-PAGE . To guarantee complete denaturation of the samples , additional DTT and β-mercaptoethanol were added to a final concentration of 100 mM and 4% , respectively , prior to incubation for 1 hr at RT and heating at 95°C for 10 min . After blotting , proteins were detected with anti-HA ( 3F10 , 1:1000 , Roche ) , anti-vinculin ( 7F9 , 1:4000 , Santa Cruz ) , anti-MnPV-L1SHORT ( mAb 2D11 , 1:5 ) or Mastomys serum mix ( 1:1000 ) prior to detection with goat anti-mouse-HRP ( W4021 , 1:10 , 000 , Promega ) or goat anti-rat-HRP ( 1:10 , 000 , Jackson ImmunoResearch ) . For Coomassie staining , gels were incubated overnight in Coomassie stain and then destained in 20% methanol . MaFi132 cells were transfected with L1 isoforms as described above and seeded on glass cover slides after 24 hr . Additional 48 hr later the cells were washed with PBS and fixed for 10 min in 4% PFA . Cells were blocked in 10% goat serum/0 . 3% Triton X-100 in PBS for 1 hr and stained with anti-MnPV-L1SHORT ( 2D11 , 1:5 ) , Mastomys serum mix ( 1:1000 ) or anti-HA ( 3F10 , 1:1000 , Roche ) and the respective secondary goat anti-mouse or donkey anti-rat IgG ( conjugated to AlexaFluor488 , 1:1000 , Invitrogen ) . Nuclei were stained with DAPI . Cover slides were mounted with Faramount Aqueous Mounting Medium ( Dako ) and imaged with a Cell Observer ( Carl Zeiss ) . Staining of formalin-fixed , paraffin-embedded tumors was performed as previously described ( Hasche et al . , 2017 ) . Briefly , deparaffinized sections were heated in citrate buffer pH6 . 0 prior to blocking with 5% goat serum/5% FCS/1% BSA in PBS and incubation with primary antibodies ( serum of a VLP-vaccinated Mastomys ( Vinzón et al . , 2014 ) , serum of a mouse immunized with the N-terminal 31 aa of MnPV-L1LONG in the OVX313 platform ( Spagnoli et al . , 2017 ) or the respective pre-immune sera , cross-reactive anti-L2 ( K18L2 ) ( Rubio et al . , 2011 ) or serum of a guinea pig immunized with Gardasil9 ( unpublished ) overnight at 4°C . Detection of L1 isoforms was achieved with the Dako REAL Detection System , Peroxidase/AEC , Rabbit/Mouse . The color reaction with AEC/H2O2 substrate solution ( Sigma ) was stopped with distilled water followed by counterstaining with hemalum solution ( Carl Roth , Karlsruhe , Germany ) . Fluorescence stainings were detected with anti-mouse-IgG1-Alexa594 or anti-guinea-pig-Alexa488 ( Invitrogen ) and nuclei were stained with DAPI . Sections were mounted with Dako Faramount Aqueous Mounting Medium . Data analyses and graphic representations were performed with GraphPad Prism 6 . 0 Software and the respective statistical test indicated in the figure legends at 95% confidence interval and an alpha level of 5% to assess significance . For time course analyses , individual time points were compared to the eight-week starting time point . The rate of L1LONG- and L1SHORT- positive animals was calculated and compared with a two-tailed McNemar’s test at an alpha level of 5% to assess significance . | Cancer is not one disease but rather a collection of disorders . As such there are many reasons why someone may develop cancer during their lifetime , including the individual’s family history , lifestyle and habits . Infections with certain viruses can also lead to cancer and human papillomaviruses are viruses that establish long-term infections that may result in cancers including cervical and anal cancer , and the most common form of cancer worldwide , non-melanoma skin cancer . The human papillomavirus , or HPV for short , is made up of DNA surrounded by a protective shell , which contains many repeats of a protein called L1 . These L1 proteins stick to the surfaces of human cells , allowing the virus to get access inside , where it can replicate before spreading to new cells . The immune system responds strongly to HPV infections by releasing antibodies that latch onto L1 proteins . It was therefore not clear how HPV could establish the long-term infections and cause cancer when it was seeming being recognized by the immune system . Now , Fu et al . have used the Southern multimammate mouse , Mastomys coucha , as a model system for an HPV infection to uncover how papillomaviruses can avoid the immune response . This African rodent is naturally infected with a skin papillomavirus called MnPV which , like its counterpart in humans , can trigger the formation of skin warts and malignant skin tumors . Fu et al . took blood samples from animals that had been infected with the virus over a period of 76 weeks to monitor their immune response overtime . This revealed that , in the early stages of infection , the virus made longer-than-normal versions of the L1 protein . Further analysis showed that these proteins could not form the virus’s protective shell but could trigger the animals to produce antibodies against them . Fu et al . went on to show that the antibodies that recognized the longer variants of L1 protein where “non-neutralizing” , meaning that could not block the spread of the virus , which is a prerequisite for immunity . It was only after a delay of four months that the animals started making neutralizing antibodies that were directed against the shorter L1 proteins that actually makes up the virus’s protective coat . These findings suggest that virus initially uses the longer version of the L1 protein as a decoy to circumvent the attention of the immune system and provide itself with enough time to establish an infection . The findings also have implications for other studies that have sought to assess the success of an immune response during a papillomavirus infection . Specifically , the delayed production of the neutralizing antibodies means that their presence does not necessarily indicate that a patient is not already infected by a papillomavirus that in the future may cause cancer . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"microbiology",
"and",
"infectious",
"disease"
] | 2020 | Expression of different L1 isoforms of Mastomys natalensis papillomavirus as mechanism to circumvent adaptive immunity |
Out-of-plane tissue deformations are key morphogenetic events during plant and animal development that generate 3D shapes , such as flowers or limbs . However , the mechanisms by which spatiotemporal patterns of gene expression modify cellular behaviours to generate such deformations remain to be established . We use the Snapdragon flower as a model system to address this problem . Combining cellular analysis with tissue-level modelling , we show that an orthogonal pattern of growth orientations plays a key role in generating out-of-plane deformations . This growth pattern is most likely oriented by a polarity field , highlighted by PIN1 protein localisation , and is modulated by dorsoventral gene activity . The orthogonal growth pattern interacts with other patterns of differential growth to create tissue conflicts that shape the flower . Similar shape changes can be generated by contraction as well as growth , suggesting tissue conflict resolution provides a flexible morphogenetic mechanism for generating shape diversity in plants and animals .
Out-of-plane deformation of tissue sheets plays a fundamental part in morphogenesis . In animals , it underlies processes such as gastrulation , neurulation , convolution of the cerebral cortex , gut folding , development of imaginal discs and dorsal appendages . Several models have been proposed to account for such deformations , involving a variety of processes such as differential surface contractions , oriented tissue tensions and differential growth ( Chen and Brodland , 2008; Clausi and Brodland , 1993; Conte et al . , 2008; Hannezo et al . , 2014; Osterfield et al . , 2013; Savin et al . , 2011; Tallinen et al . , 2014; Wyczalkowski et al . , 2012 ) . In plants , cell rearrangements and contractions play little or no role in morphogenesis , which is largely driven by growth . Nevertheless , out-of-plane tissue sheet deformations lead to the formation of elaborate structures such as floral spurs , the orchid labellum and pitcher-shaped leaves . These observations raise the question of how out-of-plane tissue deformations are generated in such diverse systems and how they relate to underlying patterns of gene activity and cell behaviours . Plants are a good starting point for addressing this question as lack of cell movement simplifies analysis . A key feature of out-of-plane deformations is that they involve generation of curvature ( local rotations out of the plane ) . Two mechanisms might account for the generation of local rotations for a tissue sheet . The first is that local rotations arise through forces external to the sheet pulling or pushing on particular regions . For example , petals ( whorl 2 ) grow in between sepals ( whorl 1 ) and stamens ( whorl 3 ) , and these adjacent organs could apply forces to shape the petal . However , homeotic mutants that change the identity of stamens to petals do not have a major effect on the complexity of whorl 2 petal shape ( Bradley et al . , 1993 ) , making it unlikely that such a mechanism plays a major role in this case . The second mechanism is that regions within the tissue sheet are specified to grow at different rates and/or directions . Local rotations can arise because they reduce or resolve potential conflicts brought about by such differential growth ( Coen and Rebocho , 2016 ) , as without regions curving or rotating relative to each other , greater levels of stress would be generated . We refer to this second mechanism , in which heterogeneity of specified growth within the tissue leads to local rotations that reduce potential stresses , as tissue conflict resolution ( for a more mathematical definition of tissue conflict resolution see Materials and methods ) . To clarify the notion of tissue conflict resolution we distinguish between two types of growth: specified and resultant ( Kennaway et al . , 2011 ) . Specified growth is how a region of tissue would deform if it was free from the mechanical constraints of its neighbouring regions . Resultant growth is how a region deforms in the context of neighbouring mechanical constraints , and includes anisotropies and local rotations that emerge from such constraints . Specified growth therefore refers to the intrinsic or active properties of a region , which may be influenced by local gene expression , while resultant growth also includes the passive changes that arise through connectivity with other regions . It is usually not possible to infer specified growth patterns directly from observed deformations ( which reflects resultant growth ) . Modelling allows the consequences of particular hypotheses for specified growth to be evaluated and compared to the data on resultant growth , such as clones and shape deformations . To illustrate how patterns of specified growth may lead to out-of-plane deformations , consider a square sheet of tissue marked with circular spots ( virtual clones , Figure 1A ) . If specified growth is equal in all directions ( isotropic specified growth ) and a growth-promoting transcription factor , GTF ( red shading in Figure 1 ) , is expressed uniformly , the tissue simply gets larger ( Figure 1B , Video 1 ) . Alternatively , specified growth could also be anisotropic , in which case regions have the intrinsic property of growing preferentially in one orientation . A simple way to establish such orientations in a tissue is through a polarity field ( arrows Figure 1C ) . If specified growth is higher parallel to the local polarity , the tissue elongates ( Figure 1D , Video 2 ) . In both of these examples , all regions within the tissue grow in a similar way without constraining each other , so resultant growth is the same as specified growth . There is no tissue conflict and local rotations are not generated . 10 . 7554/eLife . 20156 . 003Figure 1 . Generation of 3D deformations through tissue conflict resolution . ( A–B ) Isotropic specified growth promoted by a uniformly expressed Growth-promoting Transcription Factor ( GTF , red shading ) . The initial square marked with circular clones ( A ) grows into a bigger square with enlarged isometric clones ( B ) . ( C–D ) A proximodistal polarity field ( white arrows ) with uniformly expressed GTF promoting growth parallel to the polarity ( specified anisotropic growth ) . The square ( C ) elongates to from a rectangle ( D ) . ( E–F ) Surface conflict . GTF promotes specified isotropic growth and is expressed at a higher level in the top surface ( red , adaxial , ad ) compared to the bottom ( pink , abaxial , ab ) ( E ) . The canvas deforms into a dome with downwards curled edges ( F ) . ( G–H ) Areal conflict . GTF promotes specified isotropic growth and is more highly expressed in the centre of the canvas ( G ) . The canvas deforms into a rounded dome with circular clones bigger at the apex ( H , side view in left panel and clipped view in right panel ) and slightly elliptical at the periphery of the dome ( blue square in H ) . ( I–J ) Directional conflict with a convergent polarity field ( white arrows ) and GTF promoting growth parallel to the polarity . The square deforms into an elongated dome with clones elongated parallel to the polarity field ( J , side view in left panel , clipped view in right panel ) . For each model the position of the clipping plane is indicated by black line in the side view . DOI: http://dx . doi . org/10 . 7554/eLife . 20156 . 00310 . 7554/eLife . 20156 . 004Figure 1—figure supplement 1 . Areal and directional conflicts with flat starting tissue . Tissue conflict resolutions as in Figure 1 but starting with a flat sheet with a small amount of random perturbation in height instead of an initial slight curvature . ( A–B ) Areal conflict as in Figure 1G . The tissue buckles to form a dome or wave depending on the simulation run ( A and B are outputs from two separate runs ) . ( C–D ) Directional conflict as in Figure 1I . The tissue buckles to form a dome upwards or downwards depending on the simulation run ( C and D are outputs from two separate runs ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20156 . 00410 . 7554/eLife . 20156 . 005Video 1 . Isotropic growth model as in Figure 1B . The size of the canvas is not rescale to better show the increase in canvas size . DOI: http://dx . doi . org/10 . 7554/eLife . 20156 . 00510 . 7554/eLife . 20156 . 006Video 2 . Anisotropic growth model as in Figure 1D . The size of the canvas is not rescale to better show the increase in canvas size . DOI: http://dx . doi . org/10 . 7554/eLife . 20156 . 006 Local rotations and curvature can result through spatial variation in specified growth , causing buckling or bending of the tissue . We may define three types of conflict leading to local rotations: surface , areal and directional . If GTF promotes isotropic growth and is expressed at higher level in the top compared to the bottom surface ( red vs pink shading in Figure 1E ) , the tissue folds as this reduces the potential conflict in growth between of the two surfaces ( surface conflict , Figure 1F , Video 3 ) . If GTF is expressed at a higher level in the centre of the tissue compared to the surround ( Figure 1G ) , the areal conflict is reduced by the tissue buckling and formation of a round dome ( Figure 1H , Video 4 ) . The direction ( up or down ) and pattern of buckling may be biased if the sheet has an initial slight curvature generated by surface conflict , or variable if it is initially flat with slight random perturbations in height ( Figure 1—figure supplement 1A–B ) . Even though specified growth is isotropic , anisotropies may result from areal conflict . For example , clones in regions with low specified growth become stretched circumferentially ( blue box in Figure 1H ) by nearby faster growing regions . These anisotropies are a passive result of residual stresses generated by differential growth , rather than being directly specified locally . Residual stresses arise because local rotations only partially resolve the areal conflict ( for more details , see description of tissue conflict resolution in Materials and methods ) . Examples of buckling arising through areal conflict have been described previously ( Conte et al . , 2008; Green , 1992; Nath et al . , 2003; Shi et al . , 2014 ) . 10 . 7554/eLife . 20156 . 007Video 3 . Surface conflict model as in Figure 1F . Size of canvas is rescaled to better visualise the deformation of the canvas . DOI: http://dx . doi . org/10 . 7554/eLife . 20156 . 00710 . 7554/eLife . 20156 . 008Video 4 . Areal conflict model as in Figure 1H . Size of canvas is rescaled to better visualise the deformation of the canvas . DOI: http://dx . doi . org/10 . 7554/eLife . 20156 . 008 A third type of conflict is generated when the orientations of specified growth vary . For example , if the polarity field converges at the centre of the tissue ( Figure 1I ) and specified growth rate is higher parallel to the polarity , the directional conflict is partially resolved by buckling and formation an elongated dome ( Figure 1J , Video 5 ) . Virtual clones are elongated along the radial axis of the dome ( Figure 1J ) . In this example , directional conflict involves a non-parallel polarity field , but it could also involve variation in specified growth rates in relation to polarity . As with areal conflict , the direction of buckling can be biased through initial curvature ( Figure 1—figure supplement 1C–D ) . 10 . 7554/eLife . 20156 . 009Video 5 . Directional conflict as in Figure 1J . Size of canvas is rescaled to better visualise the deformation of the canvas . DOI: http://dx . doi . org/10 . 7554/eLife . 20156 . 009 Unlike surface and areal conflicts , directional conflicts have been little explored . They have been proposed to be involved in generating the Snapdragon corolla shape , during a repatterning event ( Green et al . , 2010; Kennaway et al . , 2011 ) . However , the precise pattern these conflicts take and their operation at the cellular level has not been tested . Here we address this problem using a combination of cellular and tissue analysis , together with modelling at different levels of complexity . We show that an orthogonal pattern of directional conflict contributes to the out-of-plane deformation of the lower corolla . This pattern is likely established by a combination of orthogonal gene activity with a polarity pattern visualised by PIN1 localisation . The directional conflict interacts with surface and areal conflicts to generate the corolla shape and curvature . Dorsoventral genes modulate all three types of conflict accounting for a range of phenotypes in wild type and mutants . We propose that genetically controlled tissue conflict resolutions may provide a general mechanism for generating out-of-plane deformations in both plants and animals , despite different underlying cell behaviours .
The bilaterally symmetric Snapdragon ( Antirrhinum majus ) corolla has five petals united for part of their length to form a tube with five separate lobes , which diverge at their sinuses ( Figure 2A–B ) . The upper part of the corolla comprises two dorsal petals , while the lower part consists of a pair of lateral petals flanking a ventral petal ( Figure 2A ) . The corolla functions like a mouth , with the lower corolla articulated at a hinge to open or shut ( hinge , Figure 2B ) . Several genes controlling dorsoventral asymmetry and flower shape in Snapdragon have been characterised . CYCLOIDEA ( CYC ) , DICHOTOMA ( DICH ) and RADIALIS ( RAD ) encode dorsal-specific transcription factors that repress the ventral identity gene DIVARICATA ( DIV ) ( Almeida et al . , 1997; Corley et al . , 2005; Luo et al . , 1999 , 1996 ) . 10 . 7554/eLife . 20156 . 010Figure 2 . Wild-type and div Snapdragon corolla morphology . ( A–E ) Wild-type bilaterally symmetric Snapdragon flower , with dorsal ( D ) , lateral ( L ) and ventral ( V ) petals ( A ) . The flower has a closed mouth hinged at the dorsal to lateral sinuses ( hinge ) . The corolla is divided into a proximal region ( tube ) and distal region ( lobes ) ( B ) . Dissected and partially flattened lower petals , imaged from above ( C , adaxial surface , i . e . inside the flower ) and side ( D ) ( black and white images are used to allow labelling of petal regions ) . Along the proximodistal axis , the distal wedge limit is the boundary between the lip and the distal lobe ( white dashed line ) , and the proximal wedge limit is the boundary between the palate and the proximal tube ( red dashed line ) . Along the mediolateral axis , the lateral petal midveins ( orange lines ) are the lateral wedge limits . Ventral petal midvein ( yellow line ) , petal junctions ( blue dashed lines ) , foci ( asterisks ) and petal sinuses ( blue arrows ) are also labelled . A simplified clay model of the wedge shape ( E ) illustrates the slope ( palate , light pink ) , ridge ( rim , yellow dashed line ) and cliff ( lip , dark pink ) . Experimentally induced clonal sectors ( Green et al . , 2010 ) are superimposed on the 3D shape ( multi-coloured regions ) . The wedge spans the ventral petal and the ventral half of the lateral petal ( flanks ) while the dorsal half of the lateral petal forms a narrow region either side ( hinge , green shading in C and E ) . ( F–J ) div mutant flower , with normal dorsal petals ( D ) , modified lateral and ventral petals ( ‘L’ and ‘V’ ) ( F ) , and an open flower ( G ) . To compare the shape of the div mutant with wild type , the div mutant lower petals were dissected , flattened , imaged from above ( H , adaxial side ) and from the side ( I ) and labelled as for wild type . The div mutant has two domes at the foci ( asterisks in H–J ) . A simplified clay model ( J ) highlights the reduction of palate ( light pink shading ) and the flat lip ( dark pink shading ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20156 . 010 The most striking feature of the wild-type lower corolla is a wedge-shaped fold ( Figure 2C–D ) , represented in a simplified form in Figure 2E . The wedge comprises a platform ( palate ) that slopes up towards a ridge ( rim ) and a descending lip ( Figure 2C–E ) . The rim rises to form two small peaks at the junctions between lateral and ventral petals ( foci , Figure 2A and C–E ) . DIV is required for the development of the wedge-shaped fold . In div mutants , the wedge-shape of the lower corolla is lost and two small domes form at the foci ( Figure 2F–J ) . To determine the regional origins of the wedge of wild type , we created a fate map ( Figure 3 ) using morphological landmarks such as sinuses between petals ( blue arrows ) , main veins ( yellow and orange lines ) and trichomes ( purple ellipse , from 14 DAI = Days After petal Initiation ) ( Figure 3A–G ) . The first visible sign of wedge formation was a shallow furrow proximal to the petal sinuses at 12 DAI ( yellow bracket , Figure 3B ) . Prior to this , the lower corolla showed no out-of-plane deformation ( Figure 3A ) . During the next 5 days , the furrow became more pronounced and the regions that will form the palate and lip could be identified as two crescents by 14 DAI ( light and dark pink , Figure 3C ) . These crescents spanned the ventral petal together with half the lateral petal on each side , referred to as the flanks ( boundary marked by the lateral midvein , Figure 3A–G ) . The other halves of the lateral petals grew less in width and formed the hinge ( green shading , Figure 3C–G ) . Over the next seven days the crescent bulged further to form the wedge with a slope and steep lip ( Figures 3E–G and 2C–D ) . Thus , the wedge initiates from a narrow strip of tissue at the tube-lobe boundary that undergoes an out-of-plane deformation through a defined series of morphogenetic events . 10 . 7554/eLife . 20156 . 011Figure 3 . Morphogenesis and fate map of div and wild-type lower corollas . ( A–G ) Fate map of wedge emergence in the wild-type lower corolla . The 3D deformation of the lower petals was visualised by OPT ( Lee et al . , 2006 ) with a longitudinal midsection across the ventral petal ( highlighted with white shading in left panels of A-D ) and by photographing the partially flattened lower corolla ( right panels ) , labelled with various morphological landmarks as Figure 2 . The boundary between the palate and the proximal tube ( red dashed line ) can only be determined after palate trichome emergence at 14 DAI ( purple ellipses ) . At early stages of development , 11 DAI ( A ) and 12 DAI ( B ) , the lower petals develop a furrow at the rim ( yellow bracket in B ) although they still appear relatively flat . During the next five days ( C and D ) , the furrow gets more pronounced and can be clearly seen at 17 DAI ( yellow bracket in D ) . From 18 DAI , the out-of-plane deformation is visible from side ( left panel in E-G ) and top ( right panel in E-G ) views of the dissected corolla . The size of the wedge increases over the next four days ( compare E to G ) . Scale bars , 100 µm ( A–D ) and 1 mm ( E–G ) . ( H–J ) Fate map of dome emergence in the div lower corolla . The div lower corolla was dissected and partially flattened at 11 DAI ( H ) , 14 DAI ( I ) and 17 DAI ( J ) , and labelled with various morphological landmarks as for wild type , except for the boundary between the palate and the proximal tube , which is difficult to determine at these stages due to the lack of palate trichomes . Initially , petals are relatively flat ( H ) . At 14 DAI small adaxial bulges , with foci at their tips ( asterisks in I ) , can be seen at the intersection between the petal junctions and the rim . This is also the first timepoint when the lip region can be mapped ( dark pink ) . By 17 DAI , the domes extend half way into the lateral petals ( orange line in J ) . Scale bars , 100 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 20156 . 011 In contrast to wild type , the mature div flower has an open mouth and lacks an extended palate ( Figure 2F–G ) . The ventral corolla has two small domes where the tube-lobe boundary ( yellow dashed line ) intersects the ventral-lateral petal junctions ( blue dashed line ) ( Figure 2H–J ) . The centres of intersection correspond to the foci ( asterisks in Figure 2H–J ) . Initially , the div lower petals showed no out-of-plane deformation ( Figure 3H ) . The domes first became evident as small out-of-plane adaxial bulges at 14 DAI , and were clearly visible by 17 DAI ( Figure 3I–J ) . At these stages , we mapped the lip region to a crescent-shaped region distal to the domes ( Figure 3I–J ) . To determine how these out-of-plane deformations might relate to cellular behaviours , we analysed the pattern of cell files in div and wild type , at different stages of development by staining the petal tissue with calcofluor white . Calcofluor stains older walls less brightly than younger walls , and reveals a range of different cell patterns ( Figure 4 ) . For example , old walls ( highlighted in blue ) may surround a region of cells subdivided by randomly oriented recent walls ( highlighted in green or white , Figure 4A ) . Or old walls may surround a region subdivided by a ladder of perpendicular recent walls ( Figure 4B ) , or a region subdivided by recent perpendicular walls ( green ) , which are in turn further subdivided by more recent walls ( white ) perpendicular to these ( Figure 4C ) . 10 . 7554/eLife . 20156 . 012Figure 4 . Inferring growth orientations from cell division patterns . ( A–C ) Confocal images of petal tissue showing different shapes of cell files ( or clones ) . The cell walls were stained with calcofluor white ( left panels ) . The most recent cell walls stain the brightest and are the thinnest ( indicated by the white lines ) , intermediate cell walls stain less and are thicker ( green lines ) , and the oldest cell walls are the thickest and hardly stain ( blue lines ) . The three examples show different patterns observed on developing petals . Scale bars , 10 μm . ( D–F ) Simulated growth patterns that generate the shape of cell files ( or clones ) and pattern of cell divisions depicted in A–C . Original cell wall ( T0 ) blue , cell walls formed during T1 in green ( thinner than the T0 cell walls ) , and cell walls formed at T2 in white ( the thinnest cell walls ) . Specified growth isotropic ( D ) , oriented vertically ( E ) , or oriented vertically and then switching to horizontal ( F ) . Scale bars , 10 μm . ( G–I ) Inferring growth orientations from the patterns of cell division in the calcofluor data using lines perpendicular to cell division walls . For each panel , biological data is shown on left , simulation output on right . Lines were classified as roughly parallel ( red ) or perpendicular ( yellow ) to the proximodistal axis ( Px-Di for data ) or to the orientation of specified growth ( simulations ) . The simulations show output for isotropic growth ( G ) vertically oriented specified growth ( H ) or vertical followed by horizontally oriented specified growth ( I ) . Scale bars , 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 20156 . 01210 . 7554/eLife . 20156 . 013Figure 4—figure supplement 1 . Model output of cell division patterns generated from different specified growth patterns with various initial cell geometries . ( A-C ) Model output of cell files ( or clones ) and pattern of cell divisions for four different starting cell shapes under three different growth conditions . For each starting cell shape ( T0 ) , two images are captured at subsequent times , T1 and T2 , under specified isotropic growth ( A ) , vertically oriented growth ( B ) and vertically orientated and then switching to horizontal growth ( C ) . Original cell walls are shown in blue ( T0 ) , initially new cell walls are drawn in green ( T1 ) and then at later times drawn in white ( T2 ) . Note that in ( A ) new walls ( green and white ) are randomly orientated; in ( B ) new walls are mainly horizontal and in ( C ) green walls are mainly horizontal and white walls are mainly vertical . Thus initial cell wall geometry does not have a major effect on cell wall patterns . DOI: http://dx . doi . org/10 . 7554/eLife . 20156 . 013 To understand the origin of these patterns , we modelled cell divisions using the Growing Polarised Tissue ( GPT framework ) , in which the tissue is treated as a connected continuous material , termed the canvas ( Kennaway et al . , 2011 ) . This is the same framework used to model the deformations illustrated in Figure 1 . Within the GPT framework , specified anisotropic growth is oriented by a polarity field that propagates through the canvas . Regional gene activities modify local specified growth rates parallel ( Kpar ) or perpendicular ( Kper ) to the polarity field . Plant cells can be readily imposed on this framework because cell topology is preserved: once a vertex is created by division , its relation to other vertices is maintained through subsequent growth ( plant cells do not move relative to each other ) . We assume that when the area of a cell exceeds a threshold , it is divided along the shortest path through its centroid ( Errera , 1886; Besson and Dumais , 2011 ) . The new wall is then shortened slightly to give more realistic angles ( Nakielski , 2000 ) . Cell wall colour and thickness reflect wall age ( Figure 4D–F ) . If specified growth was isotropic and uniform , the resulting pattern of walls resembled that of Figure 4A , with the blue outline highlighting the clone derived from the original cell ( Figure 4D ) . Within the clone , cell walls were oriented randomly ( Figure 4D ) . If specified growth was uniformly anisotropic , with growth oriented proximodistally ( vertically in Figure 4 ) , a pattern similar to that seen in Figure 4B was generated , where division walls were largely perpendicular to the main orientation of growth ( Figure 4E ) . If specified growth was initially oriented proximodistally and then switches to mediolateral , a pattern similar to that in Figure 4C was generated ( Figure 4F ) . Similar patterns of division occurred irrespective of initial cell geometries , although the shape of the final clone varied ( Figure 4—figure supplement 1 ) . Thus , the calcofluor pattern gives an indication of the history of cell divisions , and orientations of growth . Lines perpendicular to the cell walls reflect growth oriented parallel ( red lines ) or perpendicular ( yellow lines ) to the proximodistal axis of the tissue ( Figure 4G–H ) . Calcofluor staining of developing div and wild-type petals showed that at 10 DAI , growth was largely oriented parallel to the proximodistal axis ( red lines in Figure 5A–B ) . In the next three days , cells in regions flanking the presumptive foci showed a switch to mediolateral growth ( yellow lines in Figure 5C–D , Figure 5—figure supplement 1A–B and F–G ) . This growth pattern continued so that by 15 DAI a clear orthogonal pattern of cell files was observed in the div mutant , with cell files being elongated medially at the tube-lobe boundary flanking the foci , proximodistally at the petal junctions and mostly isotropic at the intersection ( Figure 5E ) . In wild type , the orthogonal pattern of cell files at 15 DAI was less clear than in div as the rim flanking the ventral-lateral petal showed a mixture of cell wall orientations , indicating greater proximodistal growth ( compare Figure 5F to Figure 5E ) . There was also a region of enhanced mediolateral growth in the lateral ( lilac box , Figure 5F ) and ventral lip ( green box in Figure 5—figure supplement 1H ) . In addition , diagonally oriented cell files were observed near the sinus ( orange lines in blue box in Figure 5F ) . The observed pattern of growth at 15 DAI was maintained through later stages of div and wild-type development ( Figure 5—figure supplement 1C–D and I ) and correlated with the regions of the petal that formed out-of-plane deformations ( Figure 5—figure supplement 1E and I ) . Thus , there was a switch to mediolateral growth in regions flanking the foci prior to and during the out-of-plane deformation , which was modulated by the activity of DIV . The timing of this switch was in line with the repatterning of growth predicted previously ( Green et al . , 2010 ) . 10 . 7554/eLife . 20156 . 014Figure 5 . Pattern of cell file orientations at the ventral-lateral junctions during div and wild-type development . ( A–F ) Confocal images of div and wild-type ventral-lateral ( V–L ) junctions at 10 DAI ( A and B ) , 13 DAI ( C and D ) and 15 DAI ( E and F ) . The tissue was stained with calcofluor white to visualise the patterns of cell division and infer growth orientations as described in Figure 4G–H: proximodistal growth ( red lines ) , mediolateral growth ( yellow lines ) and diagonal growth ( orange lines ) . Each stage is shown in two magnifications: an overview of the junction region and a zoomed-in region ( coloured boxes in A-F ) . Only regions showing the clearest cell files and oriented patterns of division are shown in the overview images . The patterns of cell files at the div and wild-type V-L junction are mainly proximodistal at 10 DAI ( A and B ) and become increasingly mediolateral in the rim regions flanking the junction forming an orthogonal pattern of cell files by 13 DAI ( white dashed lines in C and D ) . At 15 DAI , the mediolateral region expands and together with the proximodistal files at the junction form a clear orthogonal pattern of growth orientations in div ( white dashed lines in E ) . In wild type , the orthogonal pattern is not as clear as the rim region flanking the foci and shows a mix of mediolateral and proximodistal growth ( blue box in F ) but extends to the lateral lip , where predominantly mediolateral growth is observed ( purple box in F ) in contrast to the mixed proximodistal and mediolateral growth in the div lateral lip ( purple box in E ) . Arrow at the lower right corner of each . panel indicates Proximal ( Px ) and Distal ( Di ) axis . Scale bar 100 μm , except for purple boxes of E and F where scale bar is 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 20156 . 01410 . 7554/eLife . 20156 . 015Figure 5—figure supplement 1 . Pattern of cell file orientations at the ventral-lateral junctions during div and wild-type development . ( A-D ) Confocal images of div lower petals , stained with calcofluor white to visualise the pattern of cell files , at different stages of development: 11 DAI ( A ) , 12 DAI ( B ) , 17 DAI ( C ) and 19 DAI ( D ) . For the early stages ( A and B ) , various maginfications are shown: 1 ) ‘ventral’ ( ‘V’ ) and ‘lateral’ ( ‘L’ ) petals ( left panel ) ; 2 ) enlargement ( pink box in left panel ) of the junction region labelled with the proximodistal files ( red lines ) and mediolateral files ( yellow lines ) ( middle panels ) , and 3 ) enlargement ( blue box in left panel ) of the ‘lateral’ rim labelled for proximodistal and mediolateral growth ( orthogonal to new walls , right panel ) . Cell files are predominantly proximodistal at early stages in development ( more red lines than yellow lines in A and B ) . The orthogonal pattern of cell files observed at day 15 ( Figure 4C ) is maintained later in development , at the time the emergence of the div domes ( C and D ) . Proximal ( Px ) and Distal ( Di ) axis indicated by arrow at the lower right corner of each panel . Scale bar , 100 μm . ( E ) Detail of a div ‘ventral-lateral’ junction at maturity ( day24 ) . Black dashed line indicatse the LIP boundary between the ‘ventral’ and ‘lateral’ petals . ( F-I ) Confocal images of wild-type lower petals stained with calcofluor white , at different stages of development: 11 DAI ( F ) , 12 DAI ( G ) , 15 DAI ( H ) and 16 DAI ( I ) . At each stage , magnifications are shown: 1 ) ventral ( V ) and lateral ( L ) petals ( left panel ) ; 2 ) enlargement of the junction region ( pink box in left panel ) labelled with the proximodistal files ( red lines ) , mediolateral files ( yellow lines ) and diagonal lines ( orange lines ) ( top right panel ) ; 3 ) enlargement of the lateral rim ( blue box in left panel ) labelled for proximodistal and mediolateral growth ( orthogonal to new walls ) ( bottom right panel ) ; 4 ) enlargement of the lateral lip , labelled as previous panel ( lilac box in left panel of I ) ; 5 ) enlargement of the ventral lip , labelled as above ( green box in left panel of H ) . Between 10 and 12 DAI ( F and G ) , there is an increase in mediolateral cell files in the rim region flanking ( yellow lines within white ellipses ) the junctions while in other regions growth is maintained mainly proximodistal ( red lines ) . By 15 DAI , the orthogonal pattern of cell files is not as clear as in div ( see Figure 4F ) . The mediolateral growth regions extend to the lateral ( purple box in Figure 4F ) and ventral lip ( green box , H ) . This pattern is maintained at later stages when the deformation of the rim region forming the ridge can be observed ( I ) . The positions of the lateral midvein and secondary vein are marked with an orange and green ‘line , respectively , in D . Scale bar , 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 20156 . 01510 . 7554/eLife . 20156 . 016Figure 5—figure supplement 2 . Pattern of cell file orientations at the lateral-dorsal junctions during div and wild-type development . ( A-E ) Confocal images of div lower petals , stained with calcofluor white to visualise the cell division orientations , at different stages of development: 10 DAI ( A ) , 11 DAI ( B ) , 12 DAI ( C ) , 13 DAI ( D ) and 15 DAI ( E ) . At each stage , two magnifications are shown: 1 ) left panel , an image of the ‘lateral’ ( ‘L’ ) and dorsal ( D ) petal junction labelled for proximodistal cell files ( red lines ) and mediolateral cell files ( yellow lines ) ; 2 ) right panel , enlarged region ( blue box in respective left panel ) of the ‘lateral’ rim labelled for proximodistal ( red lines ) and mediolateral ( yellow lines ) growth ( orthogonal to new walls ) . Similar to the ‘ventral-lateral’ junction , cell files are mainly proximodistal at early stages in development ( A ) and become increasingly mediolateral ( white ellipses in B , C and D ) in the rim region as the petals reach 15 DAI , when a clear orthogonal pattern of cell files can be observed ( E ) . Unlike the ‘ventral-lateral’ , the mediolateral files are organised around the sinus ( blue arrow ) , most likely due to the lack of growth at the lip region . Proximal ( Px ) and Distal ( Di ) axis indicated by arrow at the lower right corner of each panel . Scale . bar , 100 μm . ( F-J ) Confocal images of wild-type lower petals , stained with calcofluor , at different stages of development: day 10 ( F ) , day 11 ( G ) , day 12 ( H ) , day 13 I ) and day 15 ( J ) . At each stage , two magnifications are shown . The pattern of cell files is similar to the div ‘lateral’-dorsal junction , forming an orthogonal pattern of cell files around the junction . Scale bar , 100 μm and 10 μm ( for in zoomed-in images ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20156 . 016 To determine whether this growth repatterning was specific to the ventral-lateral junctions , we analysed cell files at the lateral-dorsal junction ( Figure 5—figure supplement 2 ) . An orthogonal pattern of cell files was again observed at a similar stage , but in contrast to the ventral-lateral junction , the distal arm of the orthogonal pattern did not grow . Thus , repatterning occurred at all petal junctions of the lower corolla but with varying growth dynamics . The orthogonal patterns of anisotropic growth observed in wild type and div suggests that an orthogonal directional conflict may be involved in generating these out-of-plane deformations . To explore this idea further , we simulated orthogonal conflicts using a square initial canvas , similar to that used in Figure 1 . We introduced several transcription factors with different expression domains into the initial canvas: a factor expressed in a vertical strip ( JUN , Figure 6A ) , a horizontal strip ( RIM , Figure 6B ) , the right half ( HALFSIDE , Figure 6C ) and upper half ( DISTALSIDE , Figure 6D ) . We first assumed a polarity field converging at the centre , similar to that employed in Figure 1I . To generate an orthogonal pattern of specified anisotropy , while keeping areal growth rate uniform across the canvas ( Figure 6E ) , we assumed both RIM and JUN promoted specified growth parallel to the polarity ( Kpar ) and inhibited specified growth perpendicular to the polarity ( Kper ) ( the ratio of Kpar to Kper is indicated by Kaniso in Figure 6 ) . Growth was specified to be isotropic where RIM and JUN overlap . This set of assumptions created an orthogonal pattern of specified anisotropic growth ( red and yellow regions Figure 6F ) next to quadrants of specified isotropic growth ( green regions , Figure 6F ) . Running a model based on these assumptions led to the formation of a dome ( Figure 6G , Video 6 ) , with clones more elongated within the orthogonal domains of high specified anisotropy than in adjacent quadrants ( red ellipse , Figure 6G ) . The anisotropy exhibited by the quadrants was the result of passive deformation rather than being specified directly . 10 . 7554/eLife . 20156 . 017Figure 6 . Generation of domes through orthogonal tissue conflicts . ( A–D ) Expression pattern of four growth regulatory factors . JUN is expressed as a vertical domain in the middle of the canvas ( A ) , RIM as a horizontal line in the middle of the canvas ( B ) , HALFSIDE in the right side of the canvas ( C ) and DISTALHALF in the upper half of the canvas ( D ) . In all models the Karea = ( Kpar + Kper ) was maintained uniform throughout the canvas ( E ) . ( F–O ) Orthogonal directional conflict models . In the first example Kpar is enhanced in an orthogonal domain established by JUN and RIM but not at the intersection of these factors ( F ) to generate a dome with clones more elongated in the arms of the orthogonal domains ( red ellipse ) than in the neighbouring quadrants ( G ) . Modifications to this orthogonal pattern generated variously shaped domes ( H–O ) . A T-shaped pattern ( H ) generated an asymmetric dome ( J ) . An L-shaped pattern ( J ) gave a less pronounced asymmetric dome ( K ) , while removing both side arms of high anisotropy ( L ) , gave a ridge with clones more elongated along it ( M ) . Removing all arms but one gave an asymmetric ridge ( N–O ) . ( P–U ) An orthogonal pattern of directional conflict in a parallel polarity field , generated by boosting Kpar by JUN while boosting Kper by RIM ( P ) . This specified growth pattern generated an elongated dome with clones elongated parallel to the polarity along the regions of high Kpar ( red ellipse in Q ) and perpendicular to the polarity along the regions of high Kper ( black ellipse in Q ) . Between these regions clones are more isodiametric ( Q ) . Boosting isotropic growth with JUN and RIM ( Karea in R ) resulted on the formation of four bulges but no dome ( S ) . An orthogonal pattern of directional conflict can also be generated with a channel of proximodistal polarity in the JUN domain , and mediolateral polarity in flanking regions ( T ) . An orthogonal domain of high Kpar generates a wide dome ( U ) . Kaniso = ln ( Kpar /Kper ) . The colour scale for Kaniso is –1 to +1 . DOI: http://dx . doi . org/10 . 7554/eLife . 20156 . 01710 . 7554/eLife . 20156 . 018Figure 6—figure supplement 1 . Combining tissue conflicts . ( A–D ) Combining orthogonal direction with areal conflict , by enhancing areal growth in the orthogonal domains while inhibiting growth at the intersection ( A ) , produced a rounded dome with protruding edges ( B ) . A 10% growth differential between surfaces ( surface conflict ) can be combined with an orthogonal conflict as in Figure 1I to generate an elongated dome with a wide base ( C ) or with an areal conflict as in Figure 1G to produce a rounded dome ( D ) . Kaniso = ln ( Kpar /Kper ) colour scale between the −1 and 1; Karea = Kpar + Kper . DOI: http://dx . doi . org/10 . 7554/eLife . 20156 . 01810 . 7554/eLife . 20156 . 019Video 6 . Orthogonal directional conflict model as in Figure 6G . Size of canvas is rescaled to better visualise the deformation of the canvas . DOI: http://dx . doi . org/10 . 7554/eLife . 20156 . 019 The contribution of the orthogonal directional conflict to shape could be further dissected by removing components . Removing one of the regions of high specified anisotropy by introducing dependence of RIM activity on HALFSIDE , gave an asymmetric dome ( Figure 6H–I , Video 7 ) . Removing two regions of anisotropy , by making JUN activity dependent on DISTALHALF , and RIM activity dependent on HALFSIDE ( Figure 6J ) , produced a less pronounced asymmetric dome ( Figure 6K , Video 8 ) . Removing both horizontal regions of high anisotropy by removing RIM activity , gave a ridge ( Figure 6L–M , Video 9 ) . Removing all regions but one , by making specified anisotropy dependent on the combination of JUN with DISTALHALF , resulted in a slightly arched and symmetric ridge ( Figure 6N–O , Video 10 ) . Thus , various types of out-of-plane deformation may be generated by varying the pattern and extent of directional conflict . 10 . 7554/eLife . 20156 . 020Video 7 . T-shaped orthogonal directional conflict model as in Figure 6I . Size of canvas is rescaled to better visualise the deformation of the canvas . DOI: http://dx . doi . org/10 . 7554/eLife . 20156 . 02010 . 7554/eLife . 20156 . 021Video 8 . L-shaped orthogonal directional conflict model as in Figure 6K . Size of canvas is rescaled to better visualise the deformation of the canvas . DOI: http://dx . doi . org/10 . 7554/eLife . 20156 . 02110 . 7554/eLife . 20156 . 022Video 9 . I-shaped directional conflict model as in Figure 6M . Size of canvas is rescaled to better visualise the deformation of the canvas . DOI: http://dx . doi . org/10 . 7554/eLife . 20156 . 02210 . 7554/eLife . 20156 . 023Video 10 . One arm directional conflict model as in Figure 6O . Size of canvas is rescaled to better visualise the deformation of the canvas . DOI: http://dx . doi . org/10 . 7554/eLife . 20156 . 023 To determine other ways of generating orthogonal directional conflicts , we considered a parallel rather than convergent polarity field . An orthogonal conflict could be generated by JUN promoting Kpar ( and inhibits Kper ) , while RIM promoted Kper ( and inhibits Kpar ) ( Figure 6P ) . A dome was again generated , with the polarity field being deformed by growth so that it passed over and around the dome ( Figure 6Q , Video 11 ) . Virtual clones were oriented parallel to the polarity in the JUN domain ( red ellipse , Figure 6Q ) and perpendicular to the polarity in the RIM domain ( black ellipse in Figure 6Q ) . 10 . 7554/eLife . 20156 . 024Video 11 . Orthogonal directional conflict model as in Figure 6Q . Size of canvas is rescaled to better visualise the deformation of the canvas . DOI: http://dx . doi . org/10 . 7554/eLife . 20156 . 024 In all the above cases , the axial information provided by the polarity field was critical for the resulting shape . Boosting isotropic growth in an orthogonal pattern ( areal conflict ) generated a series of bulges rather than a dome ( Figure 6R–S , Video 12 ) . 10 . 7554/eLife . 20156 . 025Video 12 . Orthogonal areal conflict model as in Figure 6S . Size of canvas is rescaled to better visualise the deformation of the canvas . DOI: http://dx . doi . org/10 . 7554/eLife . 20156 . 025 Orthogonal directional conflict resolution thus provides a mechanism for generating out-of-plane deformations , and can be specified in several ways through a combination of orthogonal regional identities and polarity fields ( convergent or parallel ) . Further dome shapes can be generated by combining the orthogonal directional conflict with areal and surface conflicts ( Figure 6—figure supplement 1A–C ) , or by combining surface and areal conflicts ( Figure 6—figure supplement 1D ) . In the above examples , we modelled deformations as arising through differential growth . In animals , contractions as well as growth may be involved in generating out-of-plane deformations . In early animal embryogenesis it is common for deformations to occur without overall change in size . To determine how principles of tissue conflict resolution could apply in these contexts , we modelled deformations that preserve overall size . For these purposes , contraction could be considered as equivalent to negative growth . For surface conflict , high specified isotropic growth of the upper surface was counterbalanced by an equal negative specified growth rate ( contraction ) for the lower surface , and gave a folded shape similar to that based on differential growth ( compare Figure 7A–B to Figure 1E–F ) . For areal conflict , negative specified isotropic growth at the periphery was counterbalanced by high specified growth in the centre , and gave a shape again similar to that based on increased growth alone ( compare Figure 7C–D to Figure 1G–H ) . For directional conflict , enhanced growth parallel to polarity ( positive Kpar ) was counterbalanced by negative specified growth perpendicular to the polarity ( negative Kper ) , and gave a similar result to overall positive growth ( compare Figure 7E–H to Figure 1I–J and Figure 6P–Q ) . Such a balance between growth in one orientation and contraction in the perpendicular orientation is equivalent to the process of convergent-extension ( Keller et al . , 2008 ) . Surface , areal and directional conflict resolutions , thus provide possible mechanisms for generating out-of-plane deformations with or without overall growth . 10 . 7554/eLife . 20156 . 026Figure 7 . Generation of domes by tissue conflicts through contraction and growth . Similar shape domes to Figure 1 can be generated using a combination of contraction and growth , with overall canvas size remaining the same . ( A–B ) Surface conflict . Uniform isotropic specified growth for one surface ( blue ) and isotropic specified contraction ( negative growth ) for the other surface ( lilac ) , causes an initial square ( A ) to deform into a dome with downward curled edges ( B ) . ( C–D ) Areal conflict . Isotropic specified growth in the centre counterbalanced by contraction at the edges causes the square canvas ( C ) to deform into a rounded dome with circular clones which are bigger at the apex ( D ) . ( E–F ) Directional conflict with a convergent polarity field . Uniformly high specified growth parallel to the polarity is counterbalanced by uniformly high specified contraction perpendicular to the polarity . The canvas deforms from an initial square ( E ) into a dome with elongated clones parallel to the polarity field ( F ) . ( G–H ) Orthogonal directional conflict with a parallel polarity field . High specified growth parallel to the polarity in the vertical domain ( red ) and perpendicular to the polarity in the horizontal domain ( blue ) are counterbalanced by contraction perpendicular to the polarity in the vertical domain and parallel in the horizontal domain ( G ) . A dome is generated with clones elongated parallel to the polarity in the red domain and perpendicular to the polarity in the blue domain ( H ) . Kaniso = ln ( Kpar /Kper ) . The colour scale for Kaniso is –1 to +1 . Karea = Kpar + Kper . DOI: http://dx . doi . org/10 . 7554/eLife . 20156 . 026 Given the cell file data , we hypothesise that orthogonal directional conflict is involved in generating the out-of-plane deformation in div and wild type . This hypothesis requires a polarity field for orienting the growth in each region . To test whether a polarity field is present and determine its pattern of orientations , we analysed PIN1 auxin transport protein distribution as this is a possible readout of cell polarity ( Steinmann et al . , 1999 ) . The Snapdragon genome has 11 PIN genes , three of which are PIN1 orthologues ( PIN1a , PIN1b and PIN1c ) . RNA in situ hybridisation revealed that the three PIN1 genes showed detectable expression in developing wild-type petals , with all PIN1 orthologues strongly expressed in the vascular tissue and epidermis in the palate region of the lower petals ( Figure 8—figure supplement 1 ) . To determine protein localisation , we raised peptide antibodies against each of the three proteins , as well as a generic antibody against a conserved domain in all three . The antibody raised to the PIN1a-based peptide gave the clearest signal . As this peptide shares some amino acids with the other PIN1s , we cannot be sure that the signal derives entirely from PIN1a and therefore refer to the signal obtained as PIN1 signal . Immunolocalisations on sectioned tissue showed that PIN1 signal is localised at the distal end of epidermal cells in early wild-type petal primordia , with abaxial and adaxial polarity converging at the tip of the petal primordia ( day 9 , Figure 8—figure supplement 2A–C ) . This is in agreement with PIN1 patterns observed in emerging primordia in other species ( O'Connor et al . , 2014; Reinhardt et al . , 2003 ) . To analyse PIN1 polarity in regions undergoing out-of-plane deformations , we developed a whole-mount immunolocalisation protocol for plant tissue sheets , and software to quantify the distribution of PIN1 signal for each cell ( PinPoint software ) . Prior to the repatterning of corolla growth in div and wild type , PIN1 is strongly expressed near the petal sinuses and margins , and is oriented proximodistally ( Figure 8A–D and Figure 8—figure supplement 2D–E ) . This pattern extends over more cells proximally when the orthogonal pattern of cell files begins to form ( 13 DAI , Figure 8E–H and Figure 8—figure supplement 2F–G ) , overlapping with the distal region showing proximodistal cell files ( Figure 8I–J ) . PIN1 is more extended in wild type than in div . Within the extended region , polarity is oriented proximodistally along the petal junction ( 3–4 central files within green dashed line in Figure 8G , Figure 8—figure supplement 2G ) . Either side of this , polarity points diagonally , towards the central files and nearby margins ( Figure 8G , Figure 8—figure supplement 2G ) . In both div and wild type , the region with diagonal PIN1 polarity corresponds to that showing diagonal cell files ( yellow boxes , Figure 8I–J ) . Strong PIN1 signal was not observed in the regions showing mediolateral cell files ( orange boxes in Figure 8I–J ) . By 15 DAI , PIN1 expression domain became restricted to the petal margins ( Figure 8—figure supplement 2H–I ) . The upregulation of PIN1 in wild type was not observed at the lateral-dorsal sinuses ( Figure 8—figure supplement 2J–K ) , where DIV was not strongly expressed at this stage ( Galego and Almeida , 2002 ) . Thus , PIN1 reveals an early proximodistal polarity pattern near the sinus which is modulated by DIV to become extended and deflected during repatterning . 10 . 7554/eLife . 20156 . 027Figure 8 . PIN1 polarity and correlation with cell file orientation . ( A–H ) Whole-mount immunolocalisation in div and wild-type ventral-lateral junctions using the antibody raised to the PIN1a peptide at 11 DAI ( A–D ) and 13 DAI ( E–H ) . Left panels show confocal images of PIN1 immunolocalisation ( red signal ) . Right panels depict the average PIN1 polarity ( yellow arrows ) calculated using the PinPoint software . At 11 DAI , PIN1 is expressed at the ventral-lateral petal junction ( blue dashed line ) just below the sinus ( blue arrow ) and points proximodistally towards the petal margin ( white outline ) in both div ( A and B ) and wild type ( C and D ) . At 13 DAI , PIN1 expression extends more proximally in div ( E and F ) and to an ever greater extent in wild type ( G and H ) . The PIN1 polarity in wild type has a central region of 2–3 proximodistal files ( within green dashed line in G ) and flanking regions with diagonal polarity deflected towards the central files ( G and H ) . Scale bar , 10 μm . Px , proximal , Di , distal . ( I and J ) Confocal images of div ( I ) and wild-type ( J ) ventral-lateral junctions at 13 DAI combining the immunohybridised PIN1a antibody signal ( red ) with the calcofluor white cell wall signal ( cyan ) at different magnifications ( yellow and orange boxes ) . When overlapping , the orientation of the cell files correlates with the pattern of cellular PIN1 polarity: proximodistal cell files have proximodistal PIN1 polarity ( e . g . files marked with white dashed lines in yellow boxes ) while diagonal cell files have diagonally oriented PIN1 polarity ( e . g . files marked with green dashed line in yellow boxes ) . The region of PIN1 expression does not overlap with the region where the mediolateral growth is observed ( white outlines in oranges boxes ) . Asterisks refer to subepidermal PIN1 signal in the vascular tissue . Scale bar ( E and F ) , 100 μm and ( G ) 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 20156 . 02710 . 7554/eLife . 20156 . 028Figure 8—figure supplement 1 . Expression patterns of AmPIN1 genes . ( A-L ) Expression patterns of AmPIN1 genes at 8 DAI ( A-C ) , 10 DAI ( D-F ) , 12 DAI ( G-I ) and 14 DAI ( J-L ) , determined by RNA in situ hybridisation of alternative sections hybridised with AmPIN1a ( left panels , A , D , G , J ) , AmPIN1b ( middle panels , B , E , H , K ) and AmPIN1c ( right panels , C , F , I , L ) probes . The AmPIN1 genes show similar expression patterns in the vascular tissue of petal primordia ( A-F ) , with some localised differences particularly at the tips of the petal primordia ( D-F and close-ups of red boxes in respective D-F images ) , in particular for AmPIN1c which shows stronger epidermal expression at the petal tips . At 12 DAI , the AmPIN1 genes continue to be expressed in the vascular tissue ( G-I ) but are additionally detected at the epidermis of the palate and lip regions , particularly strongly for AmPIN1c ( white arrows in close-ups of blue boxes in respective G-I images ) . At 13 DAI , the epidermal AmPIN1 expression starts to decrease ( J-L ) in comparison with the vascular tissue signal which continues to be similar to previous stages ( close-ups of green squares in respective J-L images ) . The upper and lower limits of the palate and lip domains are marked by pink dashed lines . The diagram in the upper right corner of every left panel represents the orientation of the section ( white line ) relative to the five petals ( red-dorsal , orange -lateral and yellow -ventral ) . V: ventral petal; D: dorsal petal; st: stamen; ca: carpel . Scale bars , 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 20156 . 02810 . 7554/eLife . 20156 . 029Figure 8—figure supplement 2 . PIN1 polarity patterns in div and wild type . ( A–C ) Scanning electron micrograph of a flower bud at 9 DAI , showing the dorsal ( D , red shading ) , lateral ( L , orange shading ) and ventral ( V , yellow shading ) petals ( A ) . Confocal images of PIN1 section-immuno ( red signal ) in calcofluor stained petal tissue ( cyan signal ) with a longitudinal section ( B , position of section indicated by blue line in A ) and a frontal section ( C , position of section indicated by magenta line in A ) . The PIN1 signal is strong in the epidermis , where polarity points proximodistally towards the tip of the petal ( white arrows in petal tip , yellow box inset in B ) and in the internal tissues where the vascular tissues starts to develop . This PIN1 pattern is confirmed in a frontal section , where the epidermal PIN1 converges towards the most distal cells at the tip of the petal while internally , the PIN1 polarity in the sub-epidermal cells points towards the developing vascular strand ( white arrows in green box inset of C ) . White line in B-C marks the petal edge . Px , proximal , Di , distal . Scale bar ( A , B and C ) , 100 μm . Scale bar ( D and E ) , 10 μm . ( D-I ) Confocal images of PIN1 whole-mount immuno in div ( D , F and H ) and wild-type ( E , G and I ) ventral-lateral sinus at 11 DAI ( D and E ) , 13 DAI ( F and G ) and 15 DAI ( H and I ) . Polarity of the PIN1 subcellular signal was estimated using the PinPoint software and is labelled with white arrow heads . Initially , PIN1 expression ( red ) is observed at the petal sinus and its cellular polarity is largely proximodistal at the sinus and proximomarginal along the marginal cells pointing towards the tips of the petals in both div and wild-type ( D and E ) . Although the expression expands in div , its polarity pattern is maintained proximodistal ( F ) . In wild type the extended PIN1 region is larger than in div ( compare G to F ) and is oriented proximodistally along the petal junction ( 3–4 central files ) ( files within green dashed line in H and zoomed-in image in blue box ) while either side of this junction , polarity points diagonally , towards the central files and nearby sinus ( G ) . At day 15 , PIN1 expression at the sinus becomes restricted to the petal margins , both in div and wild-type ( H and I ) . Scale bar , 10 μm . ( J and K ) PIN1 whole-mount immuno in wild-type lateral-dorsal sinus at 11 DAI ( J ) and 13 DAI ( K ) . PIN1 signal is restricted to the sinus region where it shows a proximodistal polarity in the margin cells ( next to white line ) and a proximomarginal polarity in the cells adjacent to the marginal cells ( J ) . The PIN1 expression at the lateral-dorsal junction is transient , as at 13 DAI no signal is observed around the sinus region ( blue arrow ) ( K ) . Vascular strands have strong PIN1 signal ( strands marked with asterisk ) . Scale bar , 10 μm . ( L ) Images of cells showing PIN1 in red ( leftmost column ) , calcofluor cell wall staining in blue ( second column ) , and composite image of PIN1 and calcofluor for use with the PinPoint software ( third column ) . Sample cell segmentations are shown in yellow and polarity directions calculated from PinPoint software are indicated with yellow arrows ( rightmost column ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20156 . 029 In light of the data above on fate mapping , cell files and PIN1 localisation , we modelled corolla morphogenesis , beginning with the simpler deformation observed in div . Rather than modelling the entire div corolla , we first modelled the region forming the out-of-plane deformation . Based on the fate map experiments , the div and wild-type out-of-plane deformations develop from a curved strip of tissue at the tube-lobe boundary , encompassing the presumptive palate and lip ( Figure 3 ) . We therefore used a strip of canvas as the initial configuration for our minimal model ( Figure 9A ) . The strip was convex on the abaxial and concave on the adaxial surface , as observed for this region of the corolla before out-of-plane deformation . Growth rates were modulated by regional identity factors distributed along three axes – proximodistal , mediolateral and dorsoventral – similar to those employed previously ( Green et al . , 2010 ) . The dorsoventral factor was a diffusible signal derived from the dorsal region , sRAD , and specified the hinge region of the lateral petal ( red shading in Figure 9A ) . The proximodistal factors included PALATE ( PLT ) , RIM and LIP ( Figure 9A ) , which corresponded to distinct regions and cell types in the flower ( Keck et al . , 2003 ) . RIM was a diffusible signal that activated the expression of BRIM to subdivide the regions of PLT and LIP expression ( Figure 9A ) . The mediolateral factors LAT and MED corresponded to the petal junctions and midvein regions , respectively ( Figure 9A ) . All of these factors interacted combinatorially to control the specified rates of growth parallel ( Kpar ) and perpendicular ( Kper ) to the polarity field ( details of interactions are given in Supplementary Material and methods ) . As PIN1 cellular localisation was only observed around the sinus and was mainly proximodistal , we made the simplest assumption that polarity is maintained as proximodistal throughout the canvas ( black arrows in Figure 9B ) . Alternatively , polarity outside the PIN1 upregulation domain may reorient to point towards the petal junction to create a region of mediolateral polarity flanking a channel of proximodistal polarity ( Figure 6T–U , Video 13 ) . 10 . 7554/eLife . 20156 . 030Figure 9 . Tissue-level model of div corolla development . ( A–F ) Modelling formation of div domes . ( A ) The initial shape represents a strip of tissue at the tube to lobe boundary , from which the domes develop ( Figure 3H–J ) . The strip is slightly curved to represent the initial petal shape , as shown in the side view of the canvas thickness ( Side ) . During the setup phase , several identity factors are established in domains along three different axes: ( 1 ) dorsoventral – sRAD which grades into the lateral petals from the dorsal petals , ( 2 ) proximodistal - LIP , PLT , RIM and BRIM and ( 3 ) mediolateral - LAT at petal junctions , and MED in the midvein region . ( B ) These factors interact combinatorially to control the specified rates of growth parallel ( Kpar ) and perpendicular ( Kper ) to the proximodistal polarity field ( black arrows ) . Growth is initially higher in Kpar than in Kper ( Kaniso ) . ( C ) Surface conflict introduced by differential growth at the rim . Kapar is specified Kpar on the abaxial surface of the canvas , whereas Kbpar is specified Kpar on the adaxial surface . The lower value of Kapar relative to Kbpar in rim region will lead to bending of the canvas at the rim . ( D ) Growth rate is higher in the LIPDISTAL region ( see darker grey in the distal region of the canvas ) , and inhibited at the hinge region by sRAD ( lighter shade of grey at the lateral edges of canvas . ( E ) The above pattern of growth produces a strip with short hinge region at 11 . 5 DAI just before the time of repatterning . At 12 DAI , an orthogonal pattern of directional conflict is introduced . Kpar is enhanced by LAT around the petal junctions ( white lines ) and inhibited by RIM , while Kper is enhanced by RIM and inhibited by MED and LAT ( the orthogonal pattern of anisotropy is marked by dashed white lines for the lateral-ventral junctions , and red dashed lines for the lateral-dorsal junction ) . ( F ) Growth following repatterning leads to a shape at 24 DAI similar to the div lower lip , with two domes at the foci region and an extended asymmetric lip ( left panel shows Kaniso , while right panel shows the LIP and PLT regional factors and virtual clones ) . Virtual clones are predominantly elongated along the proximodistal domain except in the RIM region where clones are mediolateral oriented . ( G–J ) Evaluating the contribution of growth conflicts to the shape of the div model . ( G ) Removal of surface conflict produced a similar shape as in F but the out-of-plane deformations protrude towards the abaxial side instead of the adaxial side ( left panel shows Kaniso , while right panel shows the LIP and PLT regional factors and virtual clones ) . ( H ) Removal of areal conflict , by normalising the areal growth across the canvas while maintaining the anisotropy , generated a curved canvas with long hinge region and abaxial protruding foci . ( I ) Removing all directional conflict by setting polariser to zero , so that all growth is isotropic , resulted in a curved narrow strip with slight bulges at the region of the high growth in the div ( dark grey areas around the petals junctions . ( J ) Removing only the orthogonal directional conflict , resulted in a canvas with a broad flat lip and a uniform bend at the RIM without domes . ( K ) Incorporating the above div patterns of areal , surface and directional conflict into the full Snapdragon model produced a div mutant corolla with an extended lip ( dark pink ) , short palate ( pink ) and two domes at the RIM ( yellow ) centred on the foci . Canvas shown at 14 DAI and maturity ( 24 DAI ) . Kaniso = ln ( Kpar /Kper ) . The colour scale for Kaniso is –1 to +1 . Karea = Kpar + Kper . DOI: http://dx . doi . org/10 . 7554/eLife . 20156 . 03010 . 7554/eLife . 20156 . 031Figure 9—figure supplement 1 . Virtual mutants in the Snapdragon model . ( A ) Ground-state Snapdragon corolla model , based on the published model but stripped of the genetic interaction that modulate the growth rates and orientations across the lip and palate regions of lower corolla . At 10 DAI ( left panel ) , the proximodistal polarity field is established by proxorg ( green ) and distorg ( blue ) and the polariser gradient is represented by the black arrows . The identity factors are defined throughout the canvas and control the rates of growth parallel and perpendicular to the polarity field . The different identity factors are established along three different axes , but here only the dorsoventral and the proximodistal axis are depicted . The dorsoventral axis is indicated by the darker shading of the dorsal petals . The proximodistal axis comprises of four regions: the distal lobe ( dist . lobe - light yellow ) , the lip ( dark pink ) , the rim ( yellow ) , the palate ( light pink ) and the tube ( white ) . The lower corolla at 14 DAI does not exhibit out-of-plane deformations and grows straight ( middle panel ) . The lower corolla shape is maintained until maturity ( 24 DAI panels , with top and front views marked with virtual clones in the left and middle panels , respectively , and a mid-longitudinal section in the right panel ) . ( B ) div mutant model without the orthogonal conflict ( with a front view and side marked with virtual clones in the left panel and middle panel , and a mid-longitudinal marked with polarity in the right panel ) . ( C ) Wild-type model without the orthogonal conflict ( with a front view and side marked with virtual clones in the left panel and middle panel , and a mid-longitudinal marked with polarity in the right panel ) . ( D ) Wild-type model at the repatterning stage without the deflection of the polarity at the sinus ( 14 DAI panel , compare black arrow pattern in black box of F with D ) . The shape at maturity is depicted in the 24 DAI panels ( with top and front views marked with virtual clones in the left and middle panels , respectively , and a mid-longitudinal section in the right panel ) . ( E-F ) Removing tissues conflicts . Wild-type shape when the surface conflict is removed ( E ) . Wild-type shape when the areal conflict is removed ( F ) . Red square in mid-longitudinal section indicates section plane in front view . DOI: http://dx . doi . org/10 . 7554/eLife . 20156 . 03110 . 7554/eLife . 20156 . 032Video 13 . Orthogonal directional conflict model as in Figure 6U . Size of canvas is rescaled to better visualise the deformation of the canvas . DOI: http://dx . doi . org/10 . 7554/eLife . 20156 . 032 For our minimal model of div , growth of the initial strip was specified to be higher parallel than perpendicular to the polarity ( ratio of Kpar to Kper is indicated by Kaniso in Figure 9B ) . Kpar was also inhibited on the lower ( abaxial ) surface in the rim to promote out-of-plane bending ( surface conflict , Figure 9C ) . Specified growth was further inhibited at the edges of the strip ( hinge ) through the activity of a dorsoventral gene ( sRAD ) ( areal conflict , Figure 9D ) . Kpar was also promoted in the distal lip ( areal and directional conflict , Figure 9D ) . These assumptions led to the generation of a bilaterally symmetric canvas with short lateral edges by 11 . 5 DAI ( Figure 9E ) . Orthogonal directional conflict was introduced at 12 DAI through combinatorial interactions that enhanced Kpar relative to Kper at the petal junctions , and enhanced Kper relative to Kpar along the rim ( Kaniso panel in Figure 9E showing orthogonal pattern marked by dashed white lines ) . At the hinge , growth in the distal part of the orthogonal domain was inhibited by sRAD ( red dashed lines in Figure 9E ) . In addition , Kper was inhibited by MED ( Figure 9E ) . Running this model generated two domes centred on the foci , and an extended lip , matching the shape observed in div ( Figures 9F and 2J , Video 14 ) . 10 . 7554/eLife . 20156 . 033Video 14 . div domes model as in Figure 9F . Size of canvas is rescaled to better visualise the deformation of the canvas . DOI: http://dx . doi . org/10 . 7554/eLife . 20156 . 033 This model involved three types of conflict: areal , directional and surface . To evaluate the contribution of each type of conflict to the final shape , we removed each individually . Removing surface conflict ( by equalising growth rates on the two surfaces ) led to the out-of-plane deformations protruding abaxially instead of adaxially ( Figure 9G and Video 15 ) . This was because the direction of bulging reflected the curvature of the initial strip of tissue , which was abaxially convex ( Figure 9A ) . Removing areal conflicts , by normalising the growth rates across the canvas , resulted in a crumpled strip with long hinge regions ( Figure 9H , Video 16 ) . Removing all directional conflict by eliminating polarity , while maintaining the patterns of areal and surface conflicts , generated a curved strip ( Figure 9I , Video 17 ) . Removing orthogonal directional conflict by removing the enhanced mediolateral growth at the rim generated a relatively flat shape with a bend at the rim ( Figure 9J , Video 18 ) . Thus , the out-of-plane deformation of div depends on all three types of conflict , with orthogonal directional conflict playing a key role in creating the domes . 10 . 7554/eLife . 20156 . 034Video 15 . div domes model without surface conflict as in Figure 9G . Size of canvas is rescaled to better visualise the deformation of the canvas . DOI: http://dx . doi . org/10 . 7554/eLife . 20156 . 03410 . 7554/eLife . 20156 . 035Video 16 . div domes model without areal conflict as in Figure 9H . Size of canvas is rescaled to better visualise the deformation of the canvas . DOI: http://dx . doi . org/10 . 7554/eLife . 20156 . 03510 . 7554/eLife . 20156 . 036Video 17 . div domes model without directional conflict as in Figure 9I . Size of canvas is rescaled to better visualise the deformation of the canvas . DOI: http://dx . doi . org/10 . 7554/eLife . 20156 . 03610 . 7554/eLife . 20156 . 037Video 18 . div domes model without orthogonal directional conflict as in Figure 9J . Size of canvas is rescaled to better visualise the deformation of the canvas . DOI: http://dx . doi . org/10 . 7554/eLife . 20156 . 037 The interactions from the above minimal model were incorporated into a model of the entire Snapdragon corolla to see if they could account for the overall final shape . We first modified the published full corolla model ( Green et al . , 2010 ) to create a ground state for the lower petals ( see Supplementary Material and methods ) . This produced a corolla with a wild-type looking upper corolla , but a lower corolla lacking out-of-plane deformations ( Figure 9—figure supplement 1A ) . Incorporating the interactions of the minimal div model above into this ground state produced a corolla with an open mouth , a reduced palate , two domes tipped by the foci and a flat extended lip , similar to the div phenotype ( Figure 9K 2F , Video 19 ) . As with the minimal model , removing orthogonal directional conflict generated a corolla with a bend at the rim for the lower corolla rather than two domes ( Figure 9—figure supplement 1B ) . 10 . 7554/eLife . 20156 . 038Video 19 . div full Snapdragon model as in Figure 9K . Size of canvas is rescaled to better visualise the deformation of the canvas . DOI: http://dx . doi . org/10 . 7554/eLife . 20156 . 038 To explore the role of DIV , we incorporated its activity within the minimal model . In accordance with previous experimental data , we assumed that DIV expression was graded , from low levels at the lateral-dorsal junctions to high levels in the ventral petals ( Figure 10A ) ( Galego and Almeida , 2002 ) . Unlike the div mutant , the wild-type corolla has an extended palate and lip ( forming the slope of the wedge , Figure 2E ) . We therefore assumed that DIV promotes Kpar in the palate and lip regions from an early stage ( Kaniso , Figure 10B; Karea , Figure 10D ) . DIV also enhanced surface conflict by increasing the extent of differential growth parallel to the polarity between the two surfaces near the rim , consistent with differential expression of cell division genes in this region ( Figure 10D; ( Gaudin et al . , 2000 ) ) . These assumptions generated a canvas similar to the div model at 11 . 5 DAI but with an extended palate ( compare 11 . 5 DAI , Figure 10E with Figure 9E ) . To simulate the observed deflection of polarity towards the wild-type sinus at 12 DAI , we assumed DIV activates expression of a new polariser sink at the sinus ( blue arrows , Figure 10E ) . To account for the observed patterns of orthogonal cell files in the wild type , we assumed that at 12 DAI DIV extends the region of high Kper flanking the foci into the lip region ( white dashed lines , Figure 10E ) , enhancing directional conflict . Kper was also inhibited at the distal and proximal boundaries as a proxy for the restraining effect of adjacent tissue . 10 . 7554/eLife . 20156 . 039Figure 10 . Tissue-level model of wild-type corolla development . ( A–J ) Modelling of the wild-type wedge . ( A ) The regional identities are similar to those in div ( Figure 9A ) with the addition of the graded DIV expression ( yellow shading ) and a region , SEC , between the midvein and petal junctions ( bottom panel ) . ( B ) Growth is initially higher in Kpar than in Kper ( Kaniso ) . ( C ) DIV inhibits Kapar at the ventral-lateral junctions ( whiter domain in Kapar ) , while the RIM region is kept short by DIV inhibiting Kpar slightly on both surfaces . ( D ) Growth is boosted in the region of the proximal lip and the palate ( see darker grey regions in Karea ) . ( E ) The above pattern of growth results in a taller strip , compared to div , at 11 . 5 DAI ( compare with Figure 9E ) . At 12 DAI ( repatterning stage ) , DIV promotes Kper in combination with SEC and LIP , extending the region of high mediolateral growth ( Kaniso , modified orthogonal pattern indicated with white dashed lines ) . DIV also activates expression of a minus organiser at the sinus ( blue arrows ) , deflecting the polarity field towards it . ( F ) The above specified pattern of growth leads to a shape and pattern of clone orientations similar to that of Figure 2E ( left panel shows Kaniso while right panel shows the LIP and PLT regional factors and virtual clones ) . ( G–J ) Evaluating the contribution of growth conflicts to the shape of the wild-type wedge models . ( G ) Removal of surface conflict produced a similar shape as in F but the out-of-plane deformations protrude towards the abaxial side instead of the adaxial side . ( H ) Removing areal conflict generated a small wedge with protruding foci and a long hinge region . ( I ) Removing all directional conflict , resulted in a narrow strip with big bulges at the lateral petal which would normally form the flanks of the wedge . ( J ) Removing only the orthogonal conflict resulted in a shape with a long palate , sharp bend at the rim ( left and middle panel ) and a narrow lip ( side view in right panel ) . ( K ) Incorporating the above growth patterns of areal , surface and directional conflict into the full Snapdragon model produced a closed mouth wild-type corolla , with an extended palate , a ridge at the rim and a steep lip . Canvas shown at 14 DAI and maturity ( 24 DAI ) . The region of deflection of polarity induced at the repatterning stage is shown enlarged at 14 DAI . Regions highlighted are lip ( dark pink ) , palate ( pink ) and rim ( yellow ) . Kaniso = ln ( Kpar /Kper ) . The colour scale for Kaniso is –1 to +1 . Karea = Kpar + Kper . DOI: http://dx . doi . org/10 . 7554/eLife . 20156 . 039 With these assumptions , a wedge shape was generated that resembled that observed in wild type ( Figure 10F , Video 20 ) . The pattern of growth could be compared with that of a wild-type flower by generating virtual clones , induced at 12 DAI . The pattern obtained was similar to that seen by mapping experimentally induced clones on the final wedge shape ( Figure 2E ) . In both cases , clones were predominantly proximodistal throughout the palate and distal lip regions , but mediolateral in the flanking lip regions ( compare Figure 10F to Figure 2E ) . 10 . 7554/eLife . 20156 . 040Video 20 . Wild-type wedge model as in Figure 10F . Size of canvas is rescaled to better visualise the deformation of the canvas . DOI: http://dx . doi . org/10 . 7554/eLife . 20156 . 040 As with the minimal model of div , we evaluated the contribution of each type of conflict to the final shape . Removing surface conflict led to the out-of-plane deformation protruding abaxially instead of adaxially ( Figure 10G , Video 21 ) . Removing areal conflicts resulted in a crumpled strip with long hinge regions and protruding domes with a narrow palate ( Figure 10H , Video 22 ) . Removing all directional conflict by eliminating polarity generated a curled canvas with two large domes in the lateral petal ( Figure 10I , Video 23 ) . Reducing orthogonal directional conflict by removing the enhanced mediolateral growth at the rim generated a sharp bend at the rim and a narrow lip region ( Figure 10J , Video 24 ) . Thus , as for div , morphogenesis of wild type depends on all three types of conflict , with directional conflict playing an essential role . 10 . 7554/eLife . 20156 . 041Video 21 . Wild-type wedge model without surface conflict as in Figure 10G . Size of canvas is rescaled to better visualise the deformation of the canvas . DOI: http://dx . doi . org/10 . 7554/eLife . 20156 . 04110 . 7554/eLife . 20156 . 042Video 22 . Wild-type wedge model without areal conflict as in Figure 10H . Size of canvas is rescaled to better visualise the deformation of the canvas . DOI: http://dx . doi . org/10 . 7554/eLife . 20156 . 04210 . 7554/eLife . 20156 . 043Video 23 . Wild-type wedge model without directional conflict as in Figure 10I . Size of canvas is rescaled to better visualise the deformation of the canvas . DOI: http://dx . doi . org/10 . 7554/eLife . 20156 . 04310 . 7554/eLife . 20156 . 044Video 24 . Wild-type wedge model without orthogonal directional conflict as in Figure 10J . Size of canvas is rescaled to better visualise the deformation of the canvas . DOI: http://dx . doi . org/10 . 7554/eLife . 20156 . 044 Integrating the activity of DIV into the full corolla model of div described above generated a corolla with a typical Snapdragon shape ( Figures 10K and 2A , Video 25 ) . Removing dorsoventral gene activities ( i . e . DIV , CYC , DICH ) from the model generated phenotypes resembling those of the respective mutants ( Figure 11A–D , Videos 26 and 27 ) . These phenotypes were distinct from virtual mutants obtained by removing each tissue conflict from the wild-type model ( Figure 9—figure supplement 1C and E–F ) . Removing the deflection of polarity at the sinus resulted in an overarching lower corolla ( Figure 9—figure supplement 1D ) , suggesting that the deflection of polarity could play a role in refining corolla shape . 10 . 7554/eLife . 20156 . 045Figure 11 . Tissue-level models of dorsoventral mutants . Comparison between corolla shapes of virtual mutants generated by removing gene activities from the tissue-level model ( A , C ) and corresponding phenotypes of real corollas ( B , D ) . Double mutant cyc dich ( A , B ) and triple mutant cyc dich div ( C , D ) are shown . For each panel , corollas are viewed from top ( left ) , side ( middle ) and in longitudinal mid-section ( right , position of section for the virtual corollas indicated by red line in middle panel ) . For the virtual mutants , regions highlighted are lip ( dark pink ) , palate ( pink ) and rim ( yellow ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20156 . 04510 . 7554/eLife . 20156 . 046Video 25 . Wild-type full Snapdragon model as in Figure 10K . Size of canvas is rescale to better visualise the deformation of the canvas . DOI: http://dx . doi . org/10 . 7554/eLife . 20156 . 04610 . 7554/eLife . 20156 . 047Video 26 . cyc dich double mutant model as in Figure 11A . Size of canvas is rescale to better visualise the deformation of the canvas . DOI: http://dx . doi . org/10 . 7554/eLife . 20156 . 04710 . 7554/eLife . 20156 . 048Video 27 . cyc dich div triple mutant model as in Figure 11C . Size of canvas is rescale to better visualise the deformation of the canvas . DOI: http://dx . doi . org/10 . 7554/eLife . 20156 . 04810 . 7554/eLife . 20156 . 049Figure 12 . Summaries of growth regulatory networks ( KRNs ) used in models See materials and methods for full explanation . ( A ) Phase I of div domes . ( B ) Phase II of div domes . ( C ) Phase I of div domes with no surface conflict . ( D ) Phase II of div domes with no surface conflict . ( E ) Phase II of div domes with no orthogonal conflict . ( F ) Phase I of wild-type wedge . ( G ) Phase II of wild-type wedge . ( H ) Phase II of wild-type wedge with no surface conflict . ( I ) Ground state of full corolla model . ( J ) div mutant full corolla model showing new interactions ( green ) and modified interactions ( blue ) . ( K ) Wild-type full corolla model showing only DIV-dependent interactions ( green and bold ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20156 . 049
Through analysis of cell files , polarity patterns and computational modelling , we show that an orthogonal directional conflict plays a key role in generating out-of-plane deformations underlying the 3D Snapdragon corolla shapes of wild-type and dorsoventral mutants . This directional conflict interacts with areal and surface conflicts to generate the observed shape . We propose these conflicts are generated through combinatorial patterns of gene activity along different axes of the flower , which modulate local rates and orientations of growth , and hence the pattern of conflicts . Dorsoventral genes , such as DIV , affect all three types of conflicts to various degrees , which resolve to generate flowers with altered shapes in different genotypes . Computational modelling allows the role of each conflict to be revealed by simulations in which each conflict is removed , while preserving the others , producing virtual mutants with altered shapes . These phenotypes are distinct from virtual mutants obtained by removing dorsoventral gene activity from the model ( Figures 9G–J and 10G–J and Figure 11A–D ) . The difference may reflect evolutionary history through which genes modulate conflicts in multiple ways to modify flower shape . The end result is that each gene , such as DIV , has an effect on multiple conflicts . Each conflict depends on combinatorial interaction of multiple genes , such as those specifying orthogonal domains needed for directional conflict . These expression domains are established along the mediolateral and proximodistal axes . Orthogonal gene expression domains are prevalent in development ( Wolpert et al . , 2015 ) , although the genes establishing the orthogonal domains in the Snapdragon corolla remains to be identified . In addition to orthogonal domains , directional conflict also requires a system for providing axial information within each region that allows specified growth to be oriented . Axial information cannot be specified directly by gene expression levels which provide only scalar information ( Lawrence et al . , 2007 ) . Axiality could be provided by a polarity field ( Abley et al . , 2013; Bosveld et al . , 2012; Goodrich and Strutt , 2011 ) or through mechanical cues such as stresses ( Aigouy et al . , 2010; Heisler et al . , 2010 ) ; ( Hervieux et al . , 2016 ) . We hypothesise that the axial information is primarily provided by a polarity field , with growth rates specified parallel or perpendicular to it . PIN1 localisation reveals a tissue cell polarity pattern that it is largely proximodistal at early stages . At the time of repatterning , PIN1 is upregulated at the petal junctions , where polarity remains largely proximodistal , rather than being inverted distal to the foci as had been previously hypothesised ( Green et al . , 2010 ) . Nevertheless , the growth orientations specified by these two polarity patterns would be similar , as specified anisotropic growth depends on axiality , which is the same whether the polarity is inverted or not . Thus , PIN1 localisation helps narrow down the possibilities for potential polarity fields guiding specified growth by providing data that cannot be inferred from growth analysis alone . Although remaining proximodistal along the petal junctions , PIN1 polarity is deflected during repatterning towards the sinus . Growth orientations follow this line of deflection , consistent with polarity orienting growth ( Figure 8I–J ) . The direction of polarity in the flanking regions which undergo mediolaterally oriented growth is unknown , as PIN1 signal is not observed in these regions . For simplicity we assume the polarity is maintained as proximodistal in these regions , and growth is enhanced perpendicular to the polarity . These assumptions can generate a Snapdragon corolla shape similar to that of a previous model ( Green et al . , 2010 ) . An alternative to polarity fields for orienting anisotropic specified growth is to employ the residual stresses generated by areal conflicts . This idea has been explored in the case of Arabidopsis sepal development ( Hervieux et al . , 2016 ) , where it was proposed that stresses generated by differential isotropic specified growth could feed back to reinforce tissue regions in the direction of the local stress . Regions of tissue would then no longer grow isotropically if mechanically isolated ( i . e . specified anistropic growth ) . However , simulations show that using local stresses to orient specified anisotropy does not allow a coordinated pattern of orientations to be specified ( Hervieux et al . , 2016 ) . This is because growth feeds back to modify the stresses and destabilises their orientations . To get around this problem , it was proposed that average stress orientation across the sepal orient specified anisotropic growth ( Hervieux et al . , 2016 ) . It is unclear , however , how such a sensing mechanism might operate and how it would allow an orthogonal pattern of orientations to be specified in the case of the Snapdragon flower . Whatever the mechanism for orienting specified growth , it needs be connected with cellular properties . A growing plant tissue can be considered as a deforming mesh of cell walls which yields continuously to cellular turgor pressure ( Lockhart , 1965; Moulia and Fournier , 2009 ) . This continuous process of mesh deformation is coordinated with introduction of new walls through cell division , keeping cell sizes within certain bounds and allowing mesh strength to be maintained as it grows . Specified growth depends on how genes control the degree of wall extensibility and cell turgor . Turgor pressure acts isotropically , while cell wall extensibility can be anisotropic because of the orientation and cross-linking of wall fibres such as cellulose . If a sheet of cellular tissue has uniform turgor and its walls have isotropic mechanical properties and yield to the pressure , the tissue gets uniformly larger . This situation corresponds to uniform isotropic specified growth ( Figure 1B ) . Conflicts arise through spatial variation in turgor and/or wall extensibility . For example , the areal conflict illustrated in Figure 1H could arise if walls in the central region of a square tissue yield more readily in the plane to turgor pressure . Similarly , the surface conflict in Figure 1F could reflect walls yielding more readily in the upper layer of cells of the tissue . Specified anisotropic growth depends on orientation and/or cross-linking of cellulose fibres in the cell wall , giving it anisotropic yielding properties . Tissue cell polarity could influence growth orientations by biasing alignments of microtubules ( Hashimoto , 2015 ) , or other processes influencing cell wall anisotropy ( Cosgrove , 2016a , 2016b ) . Directional conflicts would then arise through variation in tissue cell polarity orientations and/or how cells modify wall extensibility in response to these orientations ( e . g . wall stiffness parallel or perpendicular to the polarity , ( Coen and Rebocho , 2016 ) ) . We show that tissue conflict resolutions can generate similar deformations with or without overall growth , suggesting they provide a flexible morphogenetic mechanism applicable to many systems ( Figures 6 and 7 ) . For example , pollen wall folding has been proposed to arise through differential shrinkage following water loss , which would correspond to areal and directional conflicts in shrink ( negative growth ) rates ( Katifori et al . , 2010 ) . Gastrulation has been proposed to involve differential contraction of the apical and basal ends of cells , corresponding to a surface conflict ( Conte et al . , 2008 ) . The formation of the Drosophila wing involves deformation and out-folding of an epithelial sheet in the imaginal disc . Cell divisions within the dorsoventral boundary region of the disc tend to be oriented orthogonally to divisions in the nearby wing disc ( Baena-López et al . , 2005; Mao et al . , 2011 ) , mirrored by orthogonal expression domains ( e . g . Dpp , Wingless ) ( Neto-Silva et al . , 2009 ) . Formation of the wing may therefore involve a directional conflict . Animal epithelial deformations can also be generated in the absence of cell proliferation . The dorsal appendage in Drosophila is proposed to arise through differential tension oriented along orthogonal expression domains , corresponding to directional conflict ( Osterfield et al . , 2013 ) . Thus , similar types of tissue conflict resolution may underlie deformations for both plant and animal tissues , even though the underlying cell behaviours can be very different . According to the view presented here , tissue deformations arise through a dynamic interplay between gene activity and mechanical connectivity leading to local rotations through tissue conflict resolution . We show how genes can influence shape by changing the spatiotemporal pattern of tissue conflicts . This requires that genes modify local cell properties , such as wall extensibility in plants , actomyosin contractibility in animals , as well as tissue cell polarity ( Coen and Rebocho , 2016 ) . Outstanding questions are how such cellular properties are genetically controlled to produce the various types of tissue conflict , and how they are modulated during evolution to generate diverse shapes such as floral spurs and pitcher-shaped leaves .
Antirrhinum majus ( snapdragon ) wild type ( JI7 and JI2 ) and div-13 ( Almeida et al . , 1997 ) were grown in the greenhouse at the John Innes Centre . Using the published staging reference ( Vincent and Coen , 2004 ) for JI2 flower development , we adapted it to JI7 . Although these stocks are related , bud emergence and plastochron counts are slightly different in JI7 . Therefore , we developed a JI7 staging reference by dissecting and photographing several flower bud series and relating bud features such as carpel , stamen and petal morphology between stocks as well as measurements such as petal width and lobe length . Imaging of buds and flattened petals was done using a stereomicroscope ( Leica M205C ) . Bud age is referred to as Days After petal Initiation ( DAI ) of the flower meristem , and corresponds to the published reference system ( Vincent and Coen , 2004 ) . We used the JI7 reference system to stage the mutants . As the div-13 phenotypes have modified lower petals , we used the length and width of the dorsal petals to stage the mutant flowers at different stages of development . Snapdragon buds were dissected from inflorescences , sepals removed and the corolla opened and flattened on a drop of prostatic adhesive ( technovent ) placed onto a glass slide . At early stages the whole petal could be flattened onto the glue . At later stages , when wedge formation was deforming the lower petals , we tried to keep the 3D shape of the petals by only flattening the distal lobe and palate-tube regions . To follow the development of the wedge region from emergence until maturity in div-13 and wild-type , we used dissected and flattened material from staging experiments , as well as the OPT bud series , to mark different morphological landmarks that delimit the wedge , and made measurements on how this region grew . The measurements were done in imageJ . We used the same landmarks for div-13 and wild-type ( exceptions indicated below ) . We used the lateral petal midvein as the mediolateral limit of the wedge , which was visible at early stages just after petal emergence at 8 DAI . To mark the upper and lower limits of the wedge along the proximodistal axis , we used the lip-distal lobe boundary and the palate-distal tube boundary , respectively . The petal-distal lobe boundary ran along the sinus of the petals , so to mark this boundary we drew lines from the D-L petal sinus to the V-L sinus as well as the two V-L sinuses . This morphological landmark was easily marked from the time of petal emergence . The proximal wedge limit could only be marked after 14 DAI , when palate trichomes emerged . This limit could not be marked in div-13 as the lower petals did not develop palate trichomes ( due to absence of ventral identity ) . The rim position was visible from 14 DAI , as a furrow in flattened petals . Snapdragon flower buds were dissected from inflorescences , sepals removed and the buds prepared for OPT as previously described ( Lee et al . , 2006 ) . Scanning and reconstructions were performed using a prototype OPT Scanner at the John Innes Centre . The 3-D images were visualised and processed using VolViewer software ( http://cmpdartsvr3 . cmp . uea . ac . uk/wiki/BanghamLab/index . php/Main_Page ) . Snapdragon inflorescences were harvested and hand cut to 1 cm thick , before fixing in phosphate-buffered saline ( PBS ) pH 7 . 5 containing 4% paraformaldehyde ( RT-15710 , EMS ) , 0 . 1% Triton-X100 and 0 . 1% Tween-20 , overnight at 4°C . After dehydration ( 50% and 70% ethanol ) , samples were embedded in paraffin ( wax paramat-361148C , VWR ) using a Tissue-TEK VIP processor ( Sakura ) . 8 µm sections were cut , mounted on polysine slides ( VWR ) and dried overnight at 42°C . Slides were dewaxed in histoclear ( 2x for 10 min ) ( HS-200 , National Diagnostics ) and hydrated through a decreasing ethanol series ( made in 1x saline solution: 0 . 85% ( w/v ) NaCl ) . Slides were washed in PBS solution before digesting the tissue with Pronase ( 0 . 125 mg/ml , P6911 Sigma-Aldrich ) for 12 min . A 0 . 2% ( v/v ) Glycine-PBS solution was used to block the Pronase activity ( 3 min , RT ) , followed by a PBS wash and another fixation step with 4% paraformaldehyde in PBS for 10 min . After a 2 × 10 min PBS washes , the tissue charges were neutralised using a 0 . 5% ( v/v ) acetic anhydride solution made in 0 . 1M triethanolamine pH 8 . 0 , to minimise unspecific binding and background . Slides were washed in PBS and dehydrated though an increasing ethanol series ( made in 1x saline solution ) and left to dry for at least 15 min . To generate specific digoxigenin-labelled riboprobes probes , the ORF of AmPIN1a ( F , 5’ATGATAACTTTATCTGATTTTTACCTG-3’ and R , 5’-TCATAGTCCCAGCAAAATATAGTAG-3’ ) , AmPIN1b ( F , 5’-AAAAATGATATCTTTATCTGATTTTTACC-3’ and R , 5’TCAAGTCCCAACAAAATATAGTAGATG-3’ ) and AmPIN1c ( F , 5’ATGATCACTTTAACAGACTTATATCACG-3’ and R , 5’-CTAGAGTCCAAGCAAGATGTAGTAG-3’ ) were amplified using TAQ DNA polymerase PCR ( 201205 , Qiagen ) and cloned into pCR4-TOPO TA vector ( K4575-01 , Life Technologies ) following manufacturer’s instructions . The cloned fragments were amplified using forward specific primers ( AmPIN1a , 5’-GAGAAAGTGAAAGTGATGCCTCC-3’ , AmPIN1b , 5’-CAATGAAGATGGTTACTTGGAG-3’ and AmPIN1c , 5’CCAAACCAACTGCAATGCCACC −3’ ) and the M13F primer ( pCR4-TOPO kit ) , column purified ( 28104 , Qiaquick PCR purification ) followed by a phenol/chloroform extraction . Antisense probes were obtained by RNA transcription using the T7 promoter ( 10881767001 , Roche ) and DIG-UTP ( 11209256910 , Roche ) , according to manufacture instructions . The AmPIN1a , AmPIN1b and AmPIN1c probes were hydrolysed at 60°C using 200 mM carbonate buffer pH 10 . 2 solution , for 80 , 8 and 60 min , respectively . The hydrolysed probes were boiled in 50% formamide ( 4311320 , Life Technologies ) solution for 2 min and left on ice to cool , before adding the hybridisation buffer ( per 5 ml: 625 µl 10x Salts ( 3M NaCl , 0 . 1M Tris-HCl pH6 . 8 , 0 . 1M NaPO4 , 50 mM EDTA ) , 2500 µl Deionized formamide ( 4311320 , Life Technology Ltd ) , 62 . 5 µl tRNA ( R4251 , Sigma-Aldrich ) , 125 µl Denhardt’s ( 30915 , Sigma-Aldrich ) , 437 . 5 µl RNase-free water , 1250 µl 50% Dextran Sulphate ( D8906 , Sigma-Aldrich ) ) . Two microliters of hydrolysed probe plus 100 µl of hybridisation buffer were used per slide , and covered with Hybrislip hybridisation cover ( HS6024-CS , Grace biolabs ) . Slides were hybridised in a humid box overnight at 50°C . Next day , slides were washed 3 times in 0 . 2x SSC solution at 55°C for 25 min . Single stranded RNA was digested with RNaseA ( 0 . 02 mg/ml , R4875 Sigma-Aldrich ) in NTE buffer ( 10 mM Tris-HCl pH 7 . 5 , 1 mM EDTA and 500 mM NaCl ) at 37°C for 30 min . Slides were washed in 100 mM Tris . NaCl pH7 . 5 solution and then blocked using a freshly made 0 . 5% ( w/v ) blocking reagent ( 11096176001 , Roche ) in the same buffer , for 1 hr at room-temperature ( RT ) . After a 30 min wash in 1% ( w/v ) BSA ( A7906 , Sigma-Aldrich ) ( made in 100 mM Tris . NaCl pH7 . 5 with 0 . 3% Tritonx100 ) , slides were hybridised with a 1:3000 dilution of Anti-digoxigenin-AP antibody ( 11093274910 , Roche ) for 90 min . Slides were washed in 100 mM Tris . NaCl pH7 . 5 with 0 . 3% ( v/v ) tritonx100 , 4 times for 25 min . Triton-x100 was removed from slides by washing them in 100 mM Tris . NaCl pH7 . 5 for 5 min . Before detection , the slides were washed in a 100 mM Tris . NaCl pH9 . 5 to equilibrate the pH . Detection was performed overnight using a NBT/BCIP ( Promega ) in 100 mM Tris . NaCl pH9 . 5 plus 50 mM MgCL2 . The in situ experiments were performed three times . Protocol adapted from Conti and Bradley ( Conti and Bradley , 2007 ) . For section immunolocalisation , snapdragon inflorescences were harvested and hand cut to 2 . 5 mm thickness . Tissue was fixed in a 3 . 7% ( v/v ) formaldehyde ( F8775 , Sigma-Aldrich ) -ethanol-acetic acid solution ( FAA ) with 1% ( v/v ) DMSO and 0 . 5% ( v/v ) Triton-x100 , overnight at 4°C . After dehydration ( 50% and 70% ethanol ) , samples were embedded in paraffin ( wax paramat-361148C , VWR ) using a Tissue-TEK VIP processor ( Sakura ) . 8 µm sections were cut , mounted on polysine slides ( VWR ) and dried overnight at 42°C . Slides were dewaxed in histoclear ( 2x for 10 min ) ( HS-200 , National Diagnostics ) and hydrated through a decreasing ethanol series . To improve antigen retrieval , slides were boiled in 10 mM citrate ( pH 6 ) for 15 min and then left to cool for 30 min at room temperature . Slides were rinsed in deionized water and blocked in BB ( 5% ( w/v ) non-fat dry milk in PBS pH 7 . 5 ) at RT for 3 hr . Immunohybridisation of AmPIN1 was carried out overnight at 4°C using 1:500 primary anti-AmPIN1 antibody in BB . After washing three times ( 10 min ) with PBS containing 0 . 3% ( v/v ) Triton-x100 , the slides were hybridised with Alexa594-conjugated secondary antibody ( Jackson ImmunoResearch , 1:500 ) for 3 hr . After incubation , slides were washed as after primary antibody incubation . Finally , slides were counterstained with 0 . 1% ( w/v ) calcofluor ( F3543 , Sigma-Aldrich ) and mounted and imaged in 1% ( w/v ) DABCO ( D27802 , Sigma-Aldrich ) /PBS/50% glycerol solution . For whole-mount immunolocalisation , buds were dissected and fixed flat onto a glass slide by glue as described in the dissection and flattening of lower petals section . The tissue was then fixed in a 3 . 7% ( v/v ) formaldehyde ( F8775 , Sigma-Aldrich ) -ethanol-acetic acid solution ( FAA ) with 1% ( v/v ) DMSO and 0 . 5% Triton-x100 , overnight at 4°C , after a short vacuum treatment . After two washes with 50% ethanol , the slides were washed with deionized water and boiled in 10 mM citrate ( pH 6 ) for 15 min and then left to cool for 30 min at room temperature . Slides were rinsed in deionized water and blocked in BB ( 3% ( w/v ) BSA in PBS pH 7 . 5 ) at RT for 3 hr . Primary and secondary antibody hybridisation , washes and calcofluor staining were performed as described above . Tissue was mounted and imaged in 1% ( w/v ) DABCO/PBS solution . Number of replicas for calcofluor staining and whole mount PIN1 immuno of div mutant lower petals at various stages of development = 6 . Number of replicas for calcofluor staining and whole mount PIN1 immuno of wild-type mutant lower petals at various stages of development = 11 . Replicas gave similar results for similar developmental stages . The AmPIN1a , AmPIN1b , AmPIN1c and AmPIN1 ( conserved peptide between all three PIN1 proteins ) antibodies were produced in rabbit by Cambridge Research Discovery ( CRB ) using the following predicted antigenic peptide sequences present in the cytosolic loop domain . Peptide sequence for AmPIN1a -SRGPTPRPSNFEEE , AmPIN1b – HRGNNEDGYLERDEL , AmPIN1c – FSPASTKKKGENGKD and AmPIN1 -IYSMQSSRNPTPRGS . The clearest signal was obtained with the AmPIN1a antibody . The AmPIN1 antibody gave similar results but with more background . Scanning electron microscopy was carried out as described in Vincent and Coen ( Vincent and Coen , 2004 ) . In situ hybridised material was imaged using a Leica DM6000 and Nomarski settings , x10 dry lens , and a DFC420 digital camera was used to photograph in situ sections . All measurements made from in situ images were calculated using ImageJ software ( http://imagej . nih . gov/ij/ ) . Fluorescent immunolocalisation results were imaged using a Leica SP5 II confocal microscope and x25 water dipping lens . Images were acquired in a format of 1024 × 1024 , with a line average of 4 , scan speed of 600 Hz and pinhole of 1 airy unit . Excitation for Alexa 594 on laser 20% was with a 561 laser line , with a detection band of 570–630 nm . Calcofluor was imaged with a 405 diode laser and detection at 450–500 nm . Z stack images were at 1 µm/section . Image reconstruction was performed using the ImageJ software . PIN1 polarity was quantified using the PinPoint software , developed in MATLAB to analyse gene expression in cell boundaries . A composite stack of Anti-PIN1-Alexa594 and calcofluor staining was loaded into the program . For each cell segmentation , we determined which cell the PIN signal belonged to by manually scrolling through the z-stack . In ambiguous cases this determination was aided by PIN signal following the curvature of cell walls ( Figure 8—figure supplement 2 ) . A representative segmentation plane was chosen near to the cell’s midplane and the cell boundary was manually segmented . The segmentation was used to create an image mask of the cell . A second image mask was created for each cell by morphologically eroding the cell boundary with a disk-shaped structuring element of radius five pixels ( ≈1 . 5 µm ) . A difference image was created by subtracting the eroded mask from the original cell mask . For each pixel in the difference image , a vector was calculated using the PIN1 pixel intensity as magnitude and angle measured from the cell centroid to the pixel , relative to the x-axis as orientation . The sum of these vectors was used to represent the polarity of each cell . Average vector fields were calculated by summing vectors for each cell within a 30 × 30 pixel window . Mathematically , growth of a region is described by a growth tensor which can have both a strain ( symmetric ) and rotational ( skew-symmetric ) component ( Hejnowicz and Romberger , 1984 ) . A specified growth tensor has a strain component but no rotational component , as we assume that each tissue region exerts no intrinsic rotational force . However , a resultant growth tensor may have both strain and rotational components . If all regions are specified to grow in a similar manner and no external forces act on the tissue , specified and resultant growth are the same . This is a conflict-free situation and no local rotations are generated ( the resultant tensor has no rotational component ) . By contrast , if specified growth varies across a tissue it may create situations of potential conflict which are resolved or reduced through local rotation ( i . e . the resultant growth tensor has a rotational component ) . Local rotations provide degrees of freedom not part of the specified growth which can allow potential stresses to be reduced . We refer to this mechanism for generating local tissue rotations as tissue conflict resolution . In some cases , the potential conflicts in growth may be fully resolved through rotations , in which case the only difference between specified and resultant growth tensors is in the rotational component and there is no residual strain . For example , a conformal map is a planar deformation in which local rotations are generated through a gradient in isotropic specified growth without producing residual strain ( Alim et al . , 2016; Mitchison , 2016 ) . If the tissue did not rotate locally to accommodate the gradient in specified growth , residual stresses would be generated . Thus , local rotations fully resolve the potential tissue conflict . In most cases , however , local rotations only partially resolve the conflicts and the specified and resultant growth tensors differ in the strain component ( residual strain ) as well as the rotational component . Some residual strain is inevitable in cases of tissue buckling ( curvature through areal or directional conflicts ) , as the two surfaces of the tissue grow to a different extent , even though their specified strains are identical . See Supplementary Material and methods for a full description of tissue-level modelling . All models used can be downloaded from: http://cmpdartsvr1 . cmp . uea . ac . uk/downloads/software/OpenSourceDownload_Elife_Rebocho_2016/GPT_TissueConflicts . zip Models are based on the Growing Polarised Tissue framework ( GPT-framework ) ( Green et al . , 2010; Kennaway et al . , 2011 ) . Tissue , is considered as a continuous sheet , termed a canvas , with two surfaces ( A and B ) . Identity and signalling factors can be specified throughout the canvas and interact through a gene regulatory network ( GRN ) . A growth regulatory network ( KRN ) controls the specified growth parallel ( Kpar ) and perpendicular ( Kper ) to the local polarity , established by taking the gradient of a diffusible factor POLARISER ( POL ) . The production and degradation of POL depends on a polarity regulatory network ( PRN ) . To visualise specified anisotropy , we calculate Kaniso = ln ( Kpar /Kper ) and display its local value using a colour scale . If growth is isotropic ( Kpar = Kper ) , Kaniso = 0 . Specified areal growth rate is given by Karea = Kpar + Kper . To illustrate how various types of tissue conflict resolution lead to curvature , we use an initial square with very slight curvature and marked with circular clones . | Plant and animal organs come in many different shapes , from pitcher-shaped leaves and butterfly wings , to orchid flowers and the convoluted shape of the brain . Unlike pottery or sculpture , no external hand guides the formation of these biological structures; they arise on their own , through sheets of cells developing into particular three-dimensional shapes . But how does this process of self-making operate ? We know that patterns of gene activity are important , because mutations that disrupt these patterns change the shape of the organ . But it is not clear how these patterns lead to sheets of cells curving and bending themselves into their characteristic three-dimensional shapes . Plants are particularly useful tools for studying how three-dimensional organs form because , unlike animals , their cells do not slide relative to each other , which makes the analysis simpler . Rebocho et al . used a combination of computational modelling and cell analysis to study how the intricately shaped flowers of plants known as Snapdragons form . The experiments show that genes control the shape of Snapdragon flowers by causing groups of cells to grow at different rates and in different directions . This pattern of growth creates internal conflicts that are resolved by sheets of cells curving in particular ways , accounting for the three-dimensional shape . Rebocho et al . propose that the principles of tissue conflict resolution described in this work may also underlie the development and evolution of many other plant and animal organ shapes . A future challenge is to identify the cellular mechanisms that link patterns of gene activity to the generation and resolution of conflicting cell behaviours . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Material",
"and",
"methods"
] | [
"plant",
"biology",
"developmental",
"biology"
] | 2017 | Generation of shape complexity through tissue conflict resolution |
The Ras-superfamily GTPases are central controllers of cell proliferation and morphology . Ras signaling is mediated by a system of interacting molecules: upstream enzymes ( GEF/GAP ) regulate Ras’s ability to recruit multiple competing downstream effectors . We developed a multiplexed , multi-turnover assay for measuring the dynamic signaling behavior of in vitro reconstituted H-Ras signaling systems . By including both upstream regulators and downstream effectors , we can systematically map how different network configurations shape the dynamic system response . The concentration and identity of both upstream and downstream signaling components strongly impacted the timing , duration , shape , and amplitude of effector outputs . The distorted output of oncogenic alleles of Ras was highly dependent on the balance of positive ( GAP ) and negative ( GEF ) regulators in the system . We found that different effectors interpreted the same inputs with distinct output dynamics , enabling a Ras system to encode multiple unique temporal outputs in response to a single input . We also found that different Ras-to-GEF positive feedback mechanisms could reshape output dynamics in distinct ways , such as signal amplification or overshoot minimization . Mapping of the space of output behaviors accessible to Ras provides a design manual for programming Ras circuits , and reveals how these systems are readily adapted to produce an array of dynamic signaling behaviors . Nonetheless , this versatility comes with a trade-off of fragility , as there exist numerous paths to altered signaling behaviors that could cause disease .
Many dynamic processes in the cell such as proliferation , differentiation , or morphological change are regulated by signaling through members of the Ras superfamily of small GTPases ( Chang et al . , 2003; Sjölander et al . , 1991; Hofer et al . , 1994; Vojtek and Der , 1998; Bourne et al . , 1990 ) . Mutations in these important molecules are often associated with cancer or other diseases ( Bos , 1989; Schubbert et al . , 2007 ) . These small GTPases act as macromolecular 'switches' at cell membranes , cycling between an ON state when bound to GTP and an OFF state when bound to GDP ( Bourne et al . , 1990 ) ( Figure 1A ) . This notion of an ON and OFF state of the GTPase is manifest in differences in the conformation of the protein such that , in most cases , only the GTP-bound state is able to interact with downstream effector molecules and assemble signaling complexes ( Krengel et al . , 1990; Milburn et al . , 1990; Nassar et al . , 1995; Herrmann , 2003 ) . 10 . 7554/eLife . 12435 . 003Figure 1 . Multiple activities that are frequently perturbed in disease dynamically regulate Ras activity to control the assembly of downstream effectors during signal processing . ( A ) Depiction of the proximal architecture of Ras signaling systems . Ras is activated by guanine exchange factors ( GEFs ) that exchange GDP for GTP and is inactivated by GTPase-activating proteins ( GAPs ) that accelerate the hydrolysis of GTP . Activated Ras interacts with downstream effectors such as Raf or PI3 Kinase to assemble signaling complexes and elicit signaling outputs . ( B ) Abstraction of the proximal biochemical machinery underlying Ras processing of inputs into outputs , raising the question as to how the network configuration shapes signaling to multiple effector outputs . DOI: http://dx . doi . org/10 . 7554/eLife . 12435 . 003 As enzymes , these GTPases are formally capable of binding GTP , hydrolyzing it to GDP+Pi , and releasing product to complete the catalytic cycle on their own , but , in practice , the GTPase is incredibly slow at each stage of this cycle except for the initial binding of nucleotide ( Neal et al . , 1988; Gibbs et al . , 1984; McGrath et al . , 1984 ) . As such , molecules that can accelerate these slow steps in the catalytic cycle function as essential regulators of GTPase activity during signaling events: guanine exchange factors ( GEFs ) , which promote product release by emptying the nucleotide pocket of the GTPase and allowing subsequent reloading of the GTPase with nucleotide ( OFF->ON transition ) ; and GTPase-activating proteins ( GAPs ) which accelerate the hydrolysis of GTP to GDP+Pi ( ON->OFF transition ) ( Figure 1A ) ( Boguski and McCormick , 1993; Trahey and McCormick , 1987; Bos et al . , 2007; McCormick et al . , 1991 ) . How Ras processes information , then , is not determined by Ras alone , but rather is also highly dependent on a system of molecules comprising the upstream GEFs and GAPs that regulate its activity , and the downstream effector molecules that are engaged and regulated by the activated GTPase ( Figure 1B ) . Our biochemical view of Ras and Ras-associated proteins , however , is largely focused on individual molecules , rather than the system of molecules . Considerable work in vitro over several decades has provided structural and biochemical insights into how individual GEFs , GAPs and effectors function , as well as how their activity can be controlled through mechanisms like autoinhibition and allostery ( Lenzen et al . , 1998; Sondermann et al . , 2004; Margarit et al . , 2003; Iwig et al . , 2013; Scheffzek et al . , 1997; Boriack-Sjodin et al . , 1998; Feng et al . , 2004; Bollag and McCormick , 1991 ) . These studies have provided many of the most fundamental insights into the mechanisms of Ras activation and inactivation as well as clarified our understanding of the nature of oncogenic mutations . To study the regulators , however , these reconstitutions are almost always performed under conditions in which Ras cannot productively cycle: GEF assays monitor a single exchange of fluorescent nucleotide for non-fluorescent nucleotide; GAP assays monitor a single turnover of GTP without the possibility of the nucleotide reloading ( Eberth and Ahmadian , 2001 ) . Likewise , studies of effector interactions with activated GTPase are typically done under non-cycling conditions using non-hydrolyzable nucleotide analogs to measure equilibrium binding constants ( Geyer , 1996; Herrmann et al . , 1996; Sydor et al . , 1998 ) . However , we know that signaling dynamics are critical for many cellular responses , and yet these features have not been analyzed in most in vitro studies of Ras signaling . By comparison , cell-based investigations of signaling are inherently multi-component and multi-turnover with respect to Ras , but provide far less control over the internal system parameters . In vivo assays usually contain fewer observables that are often far removed from the proximal signaling events . A variety of ingenious experiments have aimed to directly probe Ras activation during signaling in cells using fluorescent reporter molecules or super-resolution microscopy ( Rubio et al . , 2010; Murakoshi et al . , 2004; Nan , 2015 ) . However , the complications of using these reporter tools in living cells have made it difficult to systematically probe the behavior of the Ras signaling module . For example , it is difficult to know whether the response of one FRET reporter represents that of the diverse Ras effector species in the cell . To date , there has been little systems reconstitution of Ras signaling – methodically exploring in vitro how the multiple activities that regulate Ras work together to dynamically cycle Ras and control the assembly of competing effectors on activated Ras during signal processing . As such , we know little about how the concentration and identity of the components within a Ras system define its signaling properties . Understanding how these systems-level parameters shape behavior is critical , given that different cell types can harbor different configurations of network components ( both in identity and expression levels ) and also because distinct receptors may differentially activate key network components . In addition , the fact that many diseases are associated with perturbations to Ras and its associated regulators suggests that a systems level reconstitution of Ras signaling systems could be highly informative as to types of outputs and behaviors that these systems are capable of , as well as how these systems respond to perturbations such as mutation . Here , to address this problem of systems reconstitution , we develop a new microscopy-based bead reconstitution assay of dynamic signal processing by human H-Ras , a canonical member of the Ras subfamily of GTPases . Our system includes both upstream regulators of Ras activity as well as multiple downstream effectors that bind to and perceive Ras•GTP signals . This allows us to follow multiple cycles of Ras turnover in real-time using the bead recruitment of fluorescently tagged effectors on Ras as the measured output – precisely analogous to the way cells measure and couple Ras•GTP levels to signaling outputs . In addition to giving an output that reflects a critical biological function ( effector recruitment ) , this system allows us to precisely control the components and their concentration . Using this system , we have explored how Ras signaling changes in response to network changes . We explore how oncogenic substitutions in Ras impact output behavior . We have scanned how the concentration of each type of network component sculpts effector outputs in response to a simple step input of GEF activity , how systems that contain multiple competing effector molecules behave , and how different mechanisms for implementing positive feedback reshape the landscape of output behaviors . More generally , the methods we develop herein provide a framework for studying the dynamics of other assembly driven signaling systems or more complex systems that incorporate multiple interconnected signaling nodes . This kind of analysis provides an overall design manual for Ras circuits , including those that could have originated through evolution or disease perturbation .
To gain insight into the dynamics of how Ras transmits signals to downstream effectors under different network configurations or perturbations , we sought a dynamic in vitro reconstitution of Ras signal processing that would allow us to track effector outputs across multiple Ras turnovers . We reasoned that a microsphere surface charged with Ras could serve as a platform for the assembly and disassembly of fluorescent effector molecules from solution in response to inputs , much like the native Ras system ( bound to the plasma membrane ) functions in cells ( Figure 2A ) . Signaling networks of defined composition could then be prepared from recombinant proteins and robust measurements of the dynamic output behavior could be determined by tracking the amount of effector on the surface over time for many individual beads and averaging their responses . Although such a system would not fully capture all biophysical features of cellular Ras signaling , such as GTPase diffusion in a fluid plasma membrane , partitioning between membrane microdomains , or GTPase exchange from the membrane , it serves as an excellent starting point to understand how these systems behave with fixed Ras molecules in a highly controlled setting ( Tian et al . , 2007; Silvius et al . , 2006 ) . Moreover , although the signaling activity of most Ras effectors is more complex than binding alone ( see , for example: Jelinek et al . , 1996; Stokoe et al . , 1994 ) , the regulated interaction of effectors with GTPase is the foundation on which any other complex signaling mechanisms will unfold , and thus represents a universal and fundamental feature of all Ras signaling systems that demands our understanding . 10 . 7554/eLife . 12435 . 004Figure 2 . A network-level multi-turnover reconstitution of dynamic signal transmission from Ras to downstream effectors . ( A ) Bead-based approach used to study how Ras systems assemble effector complexes in response to inputs . By incubating Ni-NTA microspheres that have been loaded with Ras in solutions containing GEFs , GAPs , and fluorescent effectors , system outputs can be observed by monitoring the accumulation of effector on the bead-bound Ras . ( B ) Example of GEF-catalyzed GTP-dependent translocation of fluorescent effector to Ras-loaded bead . The amount of fluorescent effector bound to an individual bead before or after ( 10 min ) addition of 2 μM GEF +/- 5 mM GDP or GTP is shown . ( C ) Schematic depicting multiplexed assay workflow in which the output dynamics for many different system configurations can be measured by microscopy . ( D ) Dose-dependent signaling response of effector translocation in response to increasing amounts of indicating RasGRF GEF activity . ( E ) Dose-dependent turn-off of output in the presence of saturating effector and increasing amounts of indicated NF1 GAP activity . ( F ) Combined turn on and turn off behavior of effector response when the system was activated with 2 μM RasGRF GEF and after 30 min NF1-GAP was added . GAPs , GTPase-activating proteins; GEFs , guanine exchange factors . DOI: http://dx . doi . org/10 . 7554/eLife . 12435 . 00410 . 7554/eLife . 12435 . 005Figure 2—figure supplement 1 . RasGRF GEF and NF1 GAP dose-dependent effects on effector output behaviors . ( A ) Dose-dependent effect of increasing RasGRF GEF concentratiosn on initial rates of the c-Raf RBD effector to Ras-loaded beads . ( B ) Dose-dependent effect of increasing RasGRF GEF concentrations on steady state levels of the c-Raf RBD effector on Ras-loaded beads . ( C ) Dose-dependent effect of increasing NF1 GAP concentrations on disappearance of c-Raf RBD effecro from Ras•GTP-loaded beads . GAP , GTPase-activating protein; GEF , guanine exchange factor; RBD , Ras-binding domain . DOI: http://dx . doi . org/10 . 7554/eLife . 12435 . 005 We first asked whether we could observe GTP-dependent translocation of an effector molecule to a Ras-coated bead catalyzed by a guanine nucleotide exchange factor ( GEF ) . For our initial studies , we chose to use the catalytic domain from the RasGRF GEF , which is constitutively active and , unlike other GEFs , contains no allosteric feedback sites ( Freedman , 2006 ) . Ni-NTA microspheres were charged with a 16x-histidine tagged H-Ras•GDP ( OFF state ) that could not dissociate from the bead and incubated in the presence of 50 nM ( ~KD ) of a model effector: a fluorescently tagged Ras-binding domain ( RBD ) from the C-Raf kinase ( Block et al . , 1996 ) . Under these basal conditions , the amount of fluorescence on the bead was comparable to the background levels of fluorescence from the effector in solution . We then added as input 2 μM of the catalytic domain of the RasGRF GEF and 5 mM of either GDP or GTP and monitored the output of effector fluorescence on the bead ( Figure 2B ) . This amount of nucleotide in solution is in vast excess of the small amount of bead-bound Ras present in the reactions , providing essentially an infinite supply of nucleotide for these reactions on the timescale we examine ( detailed in 'Materials and methods' ) . Upon GEF and nucleotide addition , there was noticeable accumulation of fluorescent effector on the bead surface of the GTP containing reactions within seconds , and considerable fluorescent signal was observed by 10 min . In contrast , no fluorescent effector accumulated on the surface of reactions containing GDP , indicating that GEF-catalyzed translocation of the effector was dependent on Ras becoming GTP loaded . Having seen GTP-dependent GEF-catalyzed translocation of an effector to a Ras-charged bead surface , we were now in position to prepare signaling networks of arbitrary configuration and assay their output dynamics in multiplex using our microscopy-based assay ( Figure 2C ) . We first asked whether we could observe quantitative differences in the system’s output behavior when different amounts of GEF activity were used as inputs . When identically loaded beads were stimulated with increasing amounts of RasGRF , we observed both faster rates of effector translocation and higher steady state amplitudes of effector output ( Figure 2D , Figure 2—figure supplement 1A–B ) . Initial rates of effector translocation appeared to show a hyperbolic response to increasing GEF , while steady states effector levels showed a linear response . Together , these data demonstrate that our reconstituted Ras signal processing system produced outputs that responded quantitatively to the amount of GEF present in the system . In addition to being able to turn on , a dynamic reconstitution of Ras signal processing must be reversible and be able to turn off . To test the reversibility and turn-off of our system , we prepared beads loaded with H-Ras•GTP ( ON state ) , incubated them with a saturating excess amount of C-Raf RBD effector ( 2 . 5 μM total , 50 nM fluorescently labeled ) , and monitored the loss of effector signal from the bead over time ( Figure 2E ) . In the absence of any GAP , effector fluorescence decayed with a rate constant of 0 . 01 min-1 , which is similar to the expected rate of intrinsic hydrolysis by H-Ras under our assay conditions ( Neal et al . , 1988; Gibbs et al . , 1984 ) . This result is consistent with previous observations that intrinsic hydrolysis by the GTPase continues to occur even in the presence of saturating amounts of effector ( Herrmann et al . , 1995 ) . When increasing amounts of the catalytic domain from purified Neurofibromin-1 GAP ( NF1-GAP [Scheffzek et al . , 1998; Xu et al . , 1990] ) were included in the reactions , effector signal disappeared from the bead at an increased rate in a dose-dependent manner with a hyperbolic dependence on the GAP concentration ( Figure 2E , Figure 2—figure supplement 1C ) . Thus , as with the turn-on of the system , these data indicate that our reconstituted Ras signal processing system displays turn-off that responds quantitatively to the amount of GAP present in the system . Having found that our system can produce effector outputs that are turned on by GEF and turned off by GAP , we wanted to verify that the system was truly multi-turnover and that the output dynamics would respond to both these activities working in concert . To this end , we incubated Ras•GDP-loaded beads with 50 nM fluorescent effector , initiated signaling with 2 μM RasGRF GEF and 5 mM GTP , and then added 1 μM NF1-GAP at 25 min ( Figure 2F ) . As before , addition of GEF stimulated assembly of effector on the bead as the levels of Ras•GTP increased . When NF1-GAP was added to the reactions , the system responded with a rapid decrease in effector levels before stabilizing at a non-zero plateau corresponding to the non-equilibrium steady state maintained by the balance of effector , GEF and GAP activities present in the reaction . Taken together , these data imply that our on bead reconstitution of H-Ras signal processing can semi-quantitatively track dynamic effector outputs across multiple cycles of Ras activation and deactivation during signaling . This system now puts us in position to explore how different mutational states , network configurations , protein identities , or feedback mechanisms affect signal processing by Ras GTPase systems . Mutations of Ras ( especially at the G12 , G13 , or Q61 positions ) are frequently associated with cancer or other diseases ( Barbacid , 1987 ) . These alleles are primarily thought to impact Ras signaling through three mechanisms: 1 ) decreasing the intrinsic hydrolysis rate of the GTPase , 2 ) blocking GAP-mediated hydrolysis of the GTPase , and 3 ) potentially altering the interaction and preference of the GTPase for downstream effectors ( Figure 3A ) ( Trahey and McCormick , 1987; Barbacid , 1987; Rajalingam et al . , 2007; Smith and Ikura , 2014 ) . The same mutant allele of Ras can elicit different phenotypes in different cell types and tissues . Thus , we wanted to use our in vitro systems reconstitution assay to determine which system configurations are most sensitive to these oncogenic perturbations . 10 . 7554/eLife . 12435 . 006Figure 3 . The extent of signal processing distortion by oncogenic alleles of Ras depends on the balance of positive and negative regulatory activities in the network . ( A ) Depiction of wild-type ( WT ) Ras and oncogenic G12V Ras , illustrating the modes by which mutation is thought to impact the network behavior: changing intrinsic hydrolysis rate , blocking GAP-mediated hydrolysis , and modulating effector interactions . ( B ) Absolute and normalized effector responses to a 2 μM RasGRF GEF step input in the absence of any GAP activity . ( C ) Absolute and normalized responses of the same step input as in ( B ) , but with 1 μM NF1 GAP activity present in the network . ( D ) Experimentally determined phase diagram derived from 16 output responses showing the magnitude of signal distortion caused by G12V substitution ( defined as fold-change in integrated signal of G12V relative to WT ) in different GEF and GAP network configurations . GAP , GTPase-activating protein; GEF , guanine exchange factor . DOI: http://dx . doi . org/10 . 7554/eLife . 12435 . 00610 . 7554/eLife . 12435 . 007Figure 3—figure supplement 1 . The extent of signal processing distortion by oncogenic G12C and Q61L alleles of Ras depends on the balance of positive and negative regulatory activities in the network . ( A ) Depiction of wild-type Ras and oncogenic G12C/Q61L Ras illustrating the modes by which mutation is thought to impact the network behavior: changing in intrinsic hydrolysis rate , blocking GAP-mediated hydrolysis , and modulating effector interactions . ( B ) Absolute effector responses for G12C and WT Ras in response to a 2 μM RasGRF GEF step input in the absence of any GAP activity . ( C ) As in ( B ) but with 1 μM NF1 GAP activity present in the network . ( D ) Absolute effector responses for Q61L and WT Ras in response to a 2 μM RasGRF GEF step input in the absence of any GAP activity . ( E ) As in ( D ) but with 1 μM NF1 GAP activity present in the network . GAP , GTPase-activating protein; GEF , guanine exchange factor; WT , wild type . DOI: http://dx . doi . org/10 . 7554/eLife . 12435 . 00710 . 7554/eLife . 12435 . 008Figure 3—figure supplement 2 . Kinetic modeling and simulations suggest competition and intermediate GTPase states contribute to transient system behavior . Kintek simulations for a variety of models . Each simulation contains initial conditions of 50 nM effector , 10 nM GDP bound Ras , a GEF activity of ~1 υM , and an 'infinite' supply of nucleotide ( 100000 nM ) . ( A ) Output of Kintek simulation based on the simplest two-state model described in the main text methods . [GAP] is titrated by varying the hydrolysis rate constant over 5 orders of magnitude . ( B ) Output of Kintek simulation based on extending the simplest two-state model to include competition between GAP and effectors as described in the main text methods . The output from six different GAP concentrations are shown for two different GAP parameter choices . Overshoot is observed for the non-physiologic GAP parameter set ( koff = 0 . 0001 s-1 and kcat = 0 . 0001 s-1 ) . However , this is not observed when physiologic GAP parameters are used ( koff = 0 . 01 s-1 , kcat = 1s-1 ) . ( C ) Output from a three-state GTPase model that includes competition between GAP and effector , as described in the main text methods . The output from six different GAP concentrations are shown for a physiologic choice of GAP parameters ( koff = 0 . 01 s-1 , kcat = 1 s-1 ) . Transient overshoot behavior mirroring the experimentally obtained data in Figure 3 are obtained in this simulation . GAP , GTPase-activating protein; GEF , guanine exchange factor . DOI: http://dx . doi . org/10 . 7554/eLife . 12435 . 008 Using our dynamic , multi-turnover reconstitution of Ras signal processing , we examined how signaling networks bearing the G12V allele of the Ras GTPase distorted effector outputs relative to the wild-type Ras GTPase . By labeling wild type and G12V Ras GTPases with different fluorophores , we could distinguish beads loaded with each variant in a common solution of network components to see differences in effector outputs from each system side-by-side . For these and future experiments , we display the output response data in two ways: ( 1 ) we show the absolute response , which conveys information about both the amplitude and the shape of the output response , and ( 2 ) we show the responses after normalizing to the maximum value of the response , which conveys information only about the shape or dynamic profile of the output . The latter is particularly useful for seeing how the shape of two responses differs when the amplitudes are substantially different . With this approach , we first examined the output of 50 nM C-Raf RBD effector from G12V or wild-type Ras networks without GAP activity in response to a step input of 2 μM of the GEF RasGRF ( Figure 3B , Video 1 ) . Under this network configuration , wild type and G12V Ras systems produced very similar outputs with almost no difference in the total integrated effector output and only small differences in the overall dynamics of their responses . This suggests that neither the intrinsic hydrolysis nor changes in C-Raf RBD effector interactions of the G12V substitution is particularly perturbative to the output of the signaling system under this GAP-free network configuration . 10 . 7554/eLife . 12435 . 009Video 1 . Response of wild type and G12V Ras systems in GAP-free network context . The effector output ( red ) from a representative bead loaded with wild-type Ras ( blue ) or G12V Ras ( green ) is shown . 2 μM RasGRF was provided as an activating input . Time-steps are separated by 15 min . Associated with data in main-text Figure 3B . DOI: http://dx . doi . org/10 . 7554/eLife . 12435 . 009 We then looked at the system responses of wild type and G12V Ras systems to the exact same step-input ( 2 μM RasGRF GEF ) but in networks that now included 1 μM basal NF1-GAP ( Figure 3C , Video 2 ) . Unlike in the GEF-only networks , both the dynamics and amplitude of the effector output were substantially distorted by the Ras-G12V allele . In this network configuration , wild-type Ras produced a transient response that peaked within an hour and declined to a steady state less than 20% its maximum value . In contrast , outputs from G12V were sustained and increased in magnitude for over 6 hr before settling at a steady state more than 40 times higher than wild-type Ras . Thus , the G12V mutation is significantly perturbative in a high-GAP network context . 10 . 7554/eLife . 12435 . 010Video 2 . Response of wild type and G12V Ras systems in high-GAP network context . The effector output ( red ) from a representative bead loaded with wild-type Ras ( blue ) or G12V Ras ( green ) is shown . 2 μM RasGRF was provided as an activating input and the system contained 1 μM NF1-GAP . Time-steps are separated by 15 min . Associated with data in main-text Figure 3C . DOI: http://dx . doi . org/10 . 7554/eLife . 12435 . 010 Together these data and model imply that the balance of positive and negative regulatory activities in a signaling network impacts the severity by which Ras-G12V distorts signals . Similar results were also observed for G12C and Q61L alleles ( Figure 3—figure supplement 1 ) . To determine which particular configurations are most distorted by the G12V allele , we measured the effector output response across four different input strengths ( [GEF] activity ) and four different NF1-GAP levels . We then calculated a distortion score as the fold-change integrated output from Ras-G12V relative to wild-type Ras and interpolated the responses from these 16 configurations to produce a phase diagram of signal distortion by G12V under different network conditions ( Figure 3D ) . This revealed that G12V alleles were most perturbative with low-GEF inputs and a high-GAP network context , conditions in which the GAP activity would , for wild-type Ras , completely dominate over the small amount of activating GEF input . These observations are consistent with models of oncogenic Ras signaling in which low-level inputs or noise from the environment that would normally be filtered out by basal GAP-activity are misinterpreted by the cell as bona fide activating signals . Our comparison of wild type and G12V Ras signaling systems and our associated model illustrated the importance of the network composition in shaping signal processing outputs . Each individual network component is , in essence , a separate 'dial' of the Ras signaling system that can be turned by adjusting the concentration of that component ( Figure 4A ) . Because expression levels of signaling components vary across different cell types and are often different in oncogenic states , we wondered how the level of each network component impacted signal processing wild-type Ras signaling networks . To this end , we fixed a particular input ( 2 μM RasGRF GEF ) and , starting from a particular initial system configuration ( ~2500 Ras molecules × μm2 , 50 nM C-Raf effector , no GAP activity ) with an associated output response , asked how titration of individual system components modulated the effector output . 10 . 7554/eLife . 12435 . 011Figure 4 . The concentration and identity of each Ras network component can modulate the timing , duration , shape , or amplitude of effector outputs . ( A ) Depiction of the experimental setup: a fixed step-input is applied to a panel of Ras signaling systems in which the concentration of a single network component is varied to determine how each network component individually modulates system output . ( B ) Absolute and normalized effector responses to step-input in the presence of increasing amounts of the NF1 gap . ( C ) Absolute and normalized effector responses to step-input in the presence of increasing amounts of the p120 GAP . ( D ) Absolute and normalized responses to step-input in the presence of different densities of Ras on the bead surface . ( E ) Absolute and normalized responses to step-input in the presence of increasing amounts of the C-Raf RBD effector . GAP , GTPase-activating protein; RBD , Ras-binding domain . DOI: http://dx . doi . org/10 . 7554/eLife . 12435 . 011 So far , our characterization of Ras signal processing has used the C-Raf RBD as the sole downstream effector , but in living cells these networks typically contain multiple effectors targeting different output responses that are in competition with one another for access to activated Ras , with each of these effectors possessing its own affinity for Ras•GTP and expression level in the cell ( Herrmann , 2003; Smith and Ikura , 2014 ) . Indeed , our kinetic modeling implied that competition was another important source of dynamic complexity in these systems ( Figure 3—figure supplement 2 ) . Because our reconstituted signal processing is microscopy based , we can track the behavior of multiple distinct competing effectors processing signals on the same bead simultaneously by labeling each effector with a different color fluorophore ( Figure 6A ) . To this end , we purified and labeled RBDs from the A-Raf and B-Raf kinases , which have lower ( koff = 5 . 52x10-4 s-1 , kon = 7 . 20x103 M-1s-1 ) and higher ( koff = 1 . 48x10-4 s-1 , kon = 1 . 32x104 M-1s-1 ) affinities for Ras•GTP than C-Raf ( koff = 2 . 15x10-4 s-1 , kon = 1 . 02x104 M-1s-1 ) , respectively ( Fischer et al . , 2007 ) , to examine the signal processing behavior of two-effector systems in either GAP-free or high NF1-GAP networks . 10 . 7554/eLife . 12435 . 016Figure 6 . Unique interpretation of Ras•GTP signals by different effectors in multi-effector networks encodes multiple distinct temporal outputs in the system response . ( A ) Depiction of the experimental design: a fixed step-input is applied to a particular network configurations in which more than one effector molecule is , resulting in multiple simultaneous system outputs that are measured . ( B ) Absolute and normalized responses to step-input of C-Raf RBD and B-Raf RBD in the absence of any GAP activity . ( C ) as in ( B ) but with 1 μM NF1-GAP present in the signaling network . ( D ) Absolute and normalized responses to step-input of C-Raf RBD and A-Raf RBD with 1 μM NF1-GAP present in the signaling network . ( E ) Absolute and normalized responses to step-input of C-Raf RBD and the C-RafN64A mutant RBD with 1 μM NF1-GAP present in the signaling network . RBD , Ras-binding domainDOI: http://dx . doi . org/10 . 7554/eLife . 12435 . 01610 . 7554/eLife . 12435 . 017Figure 6—figure supplement 1 . Additional examples of how the unique interpretation of Ras•GTP signals by different effectors in multi-effector networks encodes multiple distinct temporal outputs in the system response . ( A ) Depiction of the experimental design: a fixed step-input is applied to a particular network configurations in which more than one effector molecule is , resulting in multiple simultaneous system outputs that are measured . ( B ) Absolute and normalized responses to step-input of C-Raf RBD and A-Raf RBD in the absence of any GAP activity . ( C ) Absolute and normalized responses to step-input of C-Raf RBD and C-RafN64A RBD in the absence of any GAP activity . RBD , Ras-binding domainDOI: http://dx . doi . org/10 . 7554/eLife . 12435 . 01710 . 7554/eLife . 12435 . 018Figure 6—figure supplement 2 . Kinetic modeling and simulations show that competition between effectors allows multiple temporal responses to be encoded in the system output . ( A ) Output of Kintek simulation using a three-state GTPase model with competition between GAP and effectors as described in the main-text 'Materials and methods' , in which two effectors ( one c-Raf like ( koff = 0 . 001 s-1 ) , one B-Raf like ( koff = 0 . 00025 s-1 ) are present in the system at 50 nM . Other initial conditions were 50 nM effector , 1 μM GEF , 1 μM GAP , and 'infinite' nucleotide ( 100000 nM ) . This simulation recovers the observation that B-Raf can respond in a sustained way while C-Raf can respond in a transient way . ( B ) Output of Kintek simulation using a three-state GTPase model with competition between GAP and effectors as described in the main-text Materials and methods , in which two effectors have very similar concentrations and parameters ( as indicated on the figure ) . Other initial conditions were 50 nM effector , 1 μM GEF , 1 μM GAP , and 'infinite' nucleotide ( 100000 nM ) . This simulation recovers the observation that small parameter differences between effector can alter the timing and duration of transient signaling outputs . ( C ) Output of Kintek simulation using a three-state GTPase model with competition between GAP and effectors as described in the main-text 'Materials and methods' , in which three effectors with different parameters and concentrations ( as indicated in the figure ) are present in the system . Other initial conditions were 50 nM effector , 1 μM GEF , 1 μM GAP , and 'infinite' nucleotide ( 100000 nM ) . This simulation shows that a complex sequence of effector outputs can be produced ( 3 THEN 2 THEN 1 ) in response to a step input simply by titration of levels and altering effector parameters . GAP , GTPase-activating protein; GEF , guanine exchange factor . DOI: http://dx . doi . org/10 . 7554/eLife . 12435 . 018 We first considered networks containing equivalent , physiological amounts of C-Raf and B-Raf effectors , which both have high affinity for Ras•GTP ( Smith and Ikura , 2014 ) . In response to a 2 μM RasGRF GEF step input in a GAP-free network , C-Raf and B-Raf processed these signals with different amplitudes and completely different dynamics ( Figure 6B ) . Initially , C-Raf and B-Raf outputs assembled at comparable rates , but within 1 hr C-Raf output peaked and began to decrease while B-Raf continued to increase in output monotonically over the entire time course . When these step-responses were re-examined in a high NF1-GAP network context , we continued to see different effector responses between C-Raf and B-Raf: C-Raf output peaked within 30 min before sharply declining to a steady state value 25% of its maximum . In contrast , B-Raf output peaked later at 1 hr , and declined to a 75% its peak maximum , a much higher steady state compared to C-Raf ( Figure 6C ) . Thus , in this case , one effector ( B-Raf ) produces a transient response while a different effector ( C-Raf ) produces a more sustained output . Why do different effector molecules interpret upstream Ras signals with different output dynamics as we observed ? Two factors contribute to this phenomenon . First , as we previously saw with one-effector systems , the time it takes for a binding process to equilibrate impacts the extent to which it can track the transient dynamics of its target . Distinct effector molecules have different binding affinities that are determined by different on and off rates from the target , and thus will interpret Ras dynamics differently . Second , these effector molecules are in competition with one another for access to the supply of time-varying activated Ras . Because we are observing the nonequilibrium binding dynamics of effectors to Ras , an effector may be competitive in the short-term ( kinetically determined phase ) but less so in the long-term ( equilibrium-determined phase ) ( Motulsky and Mahan , 1984 ) . Appropriately , the inclusion of competing effectors with physiological parameters in our model produced dynamics recapitulating such different effector responses ( Figure 6—figure supplement 2 ) . We then performed a similar analysis for two-effector systems containing equivalent , physiological amounts of C-Raf and A-Raf RBDs , where A-Raf has a weaker affinity for Ras•GTP and higher off-rate from the GTPase than C-Raf ( Smith and Ikura , 2014 ) . In GAP-free networks , the step-responses for A-Raf and C-Raf differed most significantly in terms of their amplitudes , with A-Raf output 20 times lower than C-Raf output ( Figure 6—figure supplement 1 ) . The normalized traces of each effector output show only minor differences in the dynamics of these outputs . In high NF1-GAP networks , we could only detect an output response from the C-Raf effector; the ability of A-Raf to assemble on the bead was either filtered out by the low affinity combined with high levels of GAP activity , or was simply too weak to detect above the background A-Raf in solution ( Figure 6D ) . Clearly , effector molecules with distinct identities result in differential interpretation of inputs , but how easy is it for a given effector molecule to acquire a new dynamic behavior ? Given that A-Raf , B-Raf , and C-Raf have different affinities for the GTPase , we wondered whether mutations in C-Raf that changed its affinity would be sufficient to drive new dynamic behaviors . To this end , we characterized the step-response of two-effector network containing equivalent amounts ( 50 nM ) of the C-Raf RBD and the C-RafN64A point mutant that has decreased affinity for Ras•GTP ( Block et al . , 1996 ) . Different dynamic output behaviors for each effector were observed for both GAP-free ( Figure 6—figure supplement 1 ) and high NF1-gap ( Figure 6G ) networks . In both contexts , the lower affinity C-Raf N64A mutant peaked earlier and had lower overall amplitude than the wild-type C-Raf RBD , and like before , high GAP-networks accentuated these temporal differences . Consistent with this , we found that parameter changes as small as 2-fold could produce subtle shifts in the timing of effector reponses of our kinetic model ( Figure 6—figure supplement 2 ) . This shows that new dynamic behaviors are readily realized by mutation of an effector molecule . Taken together , this analysis shows that distinct effector molecules can perceive the same input to a Ras signaling system with different dynamics and amplitudes depending on their affinities and biochemical properties . Consequently , a single-step input can be in principle be used encode multiple classes of temporally distinct outputs that peak and decline out of phase with one another , allowing for a sequence of different activities to be organized during signal processing . For example , we were able to produce a three-wave activation response of three distinct effectors in our kinetic model by simply modifying concentrations and off-rates ( Figure 6—figure supplement 2 ) . Furthermore , the context of other regulators ( e . g . extent of GAP activity in the network ) can influence how these different dynamic responses unfold , magnifying temporal distinctions in some cases while restricting the ability of certain effectors to assemble productively at all in other cases . Finally , because even simple point mutations to an effector can dramatically alter its output dynamics , new dynamic patterns are not difficult to produce and can be easily accessed during evolution . The signaling networks we have examined thus far are solely the product of constitutive enzymatic activities and effector assembly processes unfolding in the simplest possible Ras GTPase signaling circuit . Our analysis found that in high-GAP systems , the 'ground state' output for a step-response will transiently overshoot the final steady state . In some instances , this behavior could be useful for the cell , for example to create an adaptive response or to produce distinct temporal phases in multiple downstream effector outputs; in other instances this overshoot behavior could prove undesirable , for example if the overshoot provoked a proliferative response to non-proliferative level of input . Many cellular circuits modulate intrinsic behaviors of a signaling system by including additional layers of regulation and feedback control that could alter the signaling properties of the system . To gain insight into how such regulation might alter the ground state signaling behavior of Ras GTPase systems , we examined the effect of introducing two different modes of GTPase→GEF positive feedback ( defined as active Ras promoting more activation of Ras ) on system signaling behavior .
In this work , we developed a multi-turnover reconstitution of Ras signaling to explore the space of dynamic output behaviors that could be produced by Ras GTPase systems and to characterize how each network component contributes to these behaviors . Using these assays explored how different perturbations such as oncogenic mutation , component levels , inclusion of additional effector molecules , or introducing positive feedback altered the landscape of available outputs . Our experiments imply that , much in the same way that a single genome can encode multiple cell types that are regulated through differential gene expression , a single signaling system like Ras can encode multiple dynamic signal processing behaviors by regulating the concentration and identity of network components . This regulation can be direct by acting at the level of gene expression . For example , a simple survey of published p120GAP , Ras , and c-Raf mRNA expression levels across a variety of tissue types reveals a staggering amount of diversity in what types of network configurations are present in different cell types and tissues ( Figure 9A , Table 1 ) . The true diversity in these configurations is likely even greater given the plethora of additional GEFs , GAPs , Ras variants and effectors that cells can deploy . Regulating the concentration of these activities can also be achieved indirectly by the differential recruitment of these molecules by the receptors that initiate Ras signaling , which changes their effective concentration at the plasma membrane . Indeed , many of the catalytic domains that we looked at in this study show regulated interaction dynamics with the plasma membrane in response to extracellular signals ( Gureasko et al . , 2010 ) . Thus , different cells can position their signaling systems at different points in the space of available Ras network configurations and modulate these configurations in response to extracellular cues to provide versatile top-level control of the amplitude and duration of proximal signal processing events ( Figure 9B ) . 10 . 7554/eLife . 12435 . 025Table 1 . List of plasmids used this study . A description of each construct used in this study , the bacterial antibiotic resistance associated with that plasmid , and a pSC reference index to facilitate any plasmid requests . DOI: http://dx . doi . org/10 . 7554/eLife . 12435 . 025 DescriptionBacteria MarkerpSC353pMal-H . s . SOS1cat-StrepIIamppSC354pMal-H . s . p120GAP ( RASA ) -StrepIIamppSC369pMalStrep-RasGRF ( MusGRF1cat ) amppSC427pSNAP-Mal-cRaf-RBD-StrepIIamppSC451pSNAP_Mal_H-Ras_2xHis ( 6xHis-linker-10xHis ) amppSC465pMalStrep-RasGRF-30xGAGS-RBDamppSC485pMalStrep-NF1 Ras GAPamppSC486pSNAP-Mal-H-rasG12v-2xHisamppSC488pSNAP-Mal-RafRBD ( N64A ) -StrepIIamppSC490pSNAP-Mal-H-RasG12C-2xHisamppSC492pSNAP-Mal-H-RasQ61L-2xHisamppSC501pSNAP-Mal-ARafRBD-StrepIIamppSC502pSNAP-Mal-BRafRBD-StrepIIamp This versatility is not without trade-offs , however . In particular , we observed many different paths in network-space from one signaling processing behavior to another with much higher or sustained amplitude ( Figure 9C ) . These paths include classic oncogenic substitutions like G12V in Ras , but can also be realized by increased GEF activity , decreased GAP activity , or inclusion of high-affinity effectors that increase , extend and sustain signaling responses . While some of these perturbations have not been definitively recognized as drivers of cancer , many are associated with other RASopathies in humans , like Noonan syndrome or type 1 Neurofibromatosis ( Schubbert et al . , 2007; Bollag et al . , 1996 ) . Thus , the same flexibility that allows Ras systems to realize many different signaling behaviors creates many opportunities for misregulation in response to perturbation . How any particular perturbation distorted signaling output was highly dependent on network configuration . This was most obvious from comparing the effect of Ras G12V perturbation on GAP-free and high-GAP networks ( see Figure 3 ) but is also readily apparent from inspection of the structure of our experimental maps between network configuration and signaling outputs ( see Figure 5—figure supplement 2 ) . Even networks configurations that produced highly similar output behaviors could nonetheless respond divergently to perturbations . These are network configurations along a contour in the phase diagrams and represent neutral paths along which a cell or species can move without immediate consequence to the system . For example , a low signaling output that is maintained by a weak GEF activity alone might also be produced by a higher GEF activity balanced by a high GAP activity . However , these two configurations would respond very differently to substitution with oncogenic alleles of Ras like G12V . This observation demonstrates the limited predictive power of static steady-state measurements of cellular states and highlights the need to obtain dynamic data about the pre-steady state and impulse-response behavior of cellular systems using fine-grained time courses or new methods such as optogenetic pathway activation ( Toettcher et al . , 2013 ) . The dependence of a perturbation on network configuration can also afford cells new opportunities that may be positive rather than deleterious . As an example , our analysis of the feedback gain produced by the RasGRF-RBD fusion revealed that some configurations had almost no impact on the system output , while others were highly impacted and produced little to no signal without the presence of feedback ( see Figure 7B–D ) . Thus , access to certain regions of this space is not feasible without first acquiring permissive modifications to the feedback architecture of the system . Acquisition of this permissive architecture though , can occur in regions in which there is minimal consequence to the system output . Once this feedback mechanism is present , the space of behaviors available to the system changes and previously non-functional regions of the space can now be accessed . By interrogating the space of available behaviors to a signaling system in an unbiased way as we have in the present work , we learn not only what the behavior of any particular system configuration is , but also how systems respond to change and what paths exist to travel to new configurations with new behaviors . For Ras , this space appears rich with dynamic possibilities and sufficient neutral network structure to provide evolution with ample fodder to facilitate the use of Ras for the wide array of diverse signaling roles at it plays across different cell types and species , but at the risk of harmful perturbation by diseased alleles or expression states . One striking observation from this work was the importance of effector molecules in determining how a dynamic Ras•GTP signal is interpreted . This is , in fact , a critical aspect of how these particular signaling systems work as activated Ras itself has no enzymatic activity toward other molecules , but instead serves only as a platform for the recruitment of many possible competing effector molecules within the cell . Moreover , activated Ras cannot engage more than one effector simultaneously and thus competition between effectors as well as upstream regulators like GAPs contributes to the system’s output dynamics . For simple one-effector systems that we studied with the C-Raf RBD effector , the concentration of effector shaped not only the amplitude of the output , but also the dynamics of that output as well ( see Figure 4D ) . This is because higher effector concentrations equilibrate against their target faster than lower effector concentrations , resulting in different abilities to capture transient features of the target dynamics . Effector concentration is thus more than a passive 'volume' knob that reports on Ras•GTP levels and instead is an active system component that interprets Ras activity to sculpt effector-specific output dynamics owing to its own assembly and disassembly kinetics . The importance of this property of effectors was even more apparent in two-effector Ras signaling systems , in which we found that equivalent amounts of different competing effectors interpreted the same system inputs with markedly different outputs that differed both in amplitudes as well as in duration and dynamics ( see Figure 6 ) . These systems showed a variety of interesting multi-effector programs , such as one effector exhibiting a transient response while another was sustained , one effector responding while another did not at all , or two transient responses that peaked and declined with different timing and duration . Moreover , we found that different dynamic behaviors could readily be produced by introducing point mutations in an effector that altered affinity for activated Ras . These observations stress the important roles that both kinetic and thermodynamic aspects of effector assembly and competition play in shaping how an individual effector interprets dynamic Ras•GTP levels in the context of the rest of the network . An interesting consequence of these different effector behaviors and dynamics is that it can naturally result in the temporal partitioning of distinct activities during a signaling response . This can allow some effector outputs to be restricted to early phases of signaling , only to decline and be displaced by other more dominant effectors at later stages . These observations extend recent observations of hierarchies of binding by different effectors to Ras under equilibrium conditions with non-hydrolyzable analogs ( Smith and Ikura , 2014 ) . Thus , the differential perception of Ras•GTP signals by distinct effectors may not be a flaw in the method by which cells make measurements , but a useful feature by which cells can use a single upstream signaling molecule like Ras to dictate a complex temporal program of multiple downstream outputs . Binding to activated Ras is only the first step in signal propagation for many effectors . For example , binding of Raf kinases we used in this study to Ras primarily serves to deliver the kinase to the plasma membrane where its interaction with lipids ( Ghosh and Bell , 1997 ) , other Raf kinases ( Freeman et al . , 2013 ) , scaffolds ( Brennan et al . , 2011; Ritt et al . , 2006 ) and other macromolecules alters its kinase activity and thus how it sends out signals downstream ( REF ) . In fact , Raf can even activate downstream signaling in the absence of Ras by artificial membrane recruitment ( Stokoe et al . , 1994; Leevers et al . , 1994 ) . However , this only further emphasizes the importance of effector binding dynamics in the context of cellular signal processing , as binding to Ras is a physiological prerequisite for these other mechanisms to take place . Moreover , different Ras effectors such as PI3 Kinase or Ral-GDS will have their own molecule-specific layers of regulation that take place upon interaction with Ras at the plasma membrane . These processes will be influenced by the underlying effector-binding dynamics in different ways depending on the kinetics of these downstream steps . Our work demonstrates that cells have simple systems for modulating and controlling these fundamental binding dynamics and further indicates that known control mechanisms should be analyzed in this complex dynamic context . Another layer of dynamic and regulatory complexity is also likely to arise as more classes of Ras GTPases are included in the signaling networks . Indeed , our present study has only investigated effector interaction with H-Ras , but there exist many additional Ras isoforms such as K-Ras and N-Ras , K4A-Ras , and K4B-Ras that may engage these effectors in different ways to produce different dynamics . Additionally , there exist related GTPases such as Rap which can serve as platform for Ras effectors but that do not necessarily promote signal propagation ( Wynne et al . , 2012; Cook et al . , 1993 ) . Understanding how the underlying distribution of GTPase isoforms dictates signal processing behavior is another critical component of cellular signaling and we hope to extend our in vitro system to explore these fundamental questions in the future . The network level biochemical approach to interrogating signaling systems we employed in this study occupies a relatively underexplored area in our understanding of cellular decision making systems , but is similar to approaches used to understand dynamic mechanical systems in cells like microtubules . Indeed , because additional complexities can emerge when multiple energetically driven processes are coupled together to promote the dynamic assembly and disassembly of competing effectors , exploring how these systems behave in vitro under different configurations sheds new light on the phenomenology of how biochemical signaling devices function and respond to perturbation . The ability to prepare non-equilibrium steady states in which Ras is actively cycling between ON and OFF states may also prove useful in developing new strategies to ameliorate erroneous signaling associated with diseased states . For example , the fact that we could reconstitute radically different signaling behaviors for wild type and G12V Ras under high GAP conditions is consistent with the notion that wild-type Ras ordinarily cycles quickly but G12V does not . This difference in the lifetime of a Ras•GTP molecule compared to RasG12V•GTP could potentially serve as an additional dynamic selectivity handle for small molecules in which we want to only target the oncogenic form of Ras . The assays we developed are well suited to compare how small molecules differentially impact signaling through these different forms of Ras side by side under active energy-consuming conditions . More generally , the simplicity of the approach we present here paves the way for further studies on other types of non-equilibrium signaling systems that center around the assembly of molecules from the cytoplasm on a surface such as the other members of the Ras superfamily like Rho , Rac , Cdc42 , and Rab GTPases , as well as other completely different multi-turnover signaling systems like receptor tyrosine kinases or lipid kinases . Some aspects of the H-Ras system may be shared with these systems , while other aspects may be different owing to idiosyncratic features of a particular system of molecules . These systems could also be extended in other ways as well to explore how other biophysical constraints impact these signaling processes . Our reconstitutions could , for example , be extended to lipid-coated beads to explore how membrane fluidity or lipid identity shape effector outputs . Multi-currency networks that include multiple Ras isoforms or contain more than one type of GTPase could also be examined to look at higher-order networks and cascades . Only by building these systems , turning them on and watching them run can we begin to understand how they actually perform and operate in different signaling regimes .
A combination of standard and custom ImageJ macros were used to prepare the primary image data for further analysis . First , drift in the stage was corrected using a macro based around the MultiStackReg plugin . Two or three color multi-tiff timecourses were split into separate channels . Matrix transformations to register timecourses were obtained using the constant GTPase fluorescence on every bead from the blue channel . These matrices were then used to register the timecourses of the red or green channels . The three channels were then recombined to produce the properly registered multi-tiff used for analysis of the beads . The ImageJ macro code was tweaked depending on the particular experiment , but a representative example of the code is shown below: matrix path= "C:\Users\scoyl_000\Google Drive\MICROSCOPY\150403\matrix\"; inputpath = "C:\Users\scoyl_000\Google Drive\MICROSCOPY\150403\raw\"; outputpath = "C:\Users\scoyl_000\Google Drive\MICROSCOPY\150403\processed\"; function scott_register ( input , output , filename ) { open ( input+filename ) ; run ( "Split Channels" ) ; run ( "MultiStackReg" , "stack_1=[C3-" + filename + "] action_1=Align file_1=["+matrixpath+"matrix . txt] stack_2=None action_2=Ignore file_2=[] transformation=[Rigid Body] save" ) ; run ( "MultiStackReg" , "stack_1=[C1-" + filename + "] action_1=[Load Transformation File] file_1=["+matrixpath+"matrix . txt] stack_2=None action_2=Ignore file_2=[] transformation=[Rigid Body]" ) ; run ( "MultiStackReg" , "stack_1=[C2-" + filename + "] action_1=[Load Transformation File] file_1=["+matrixpath+"matrix . txt] stack_2=None action_2=Ignore file_2=[] transformation=[Rigid Body]" ) ; run ( "Merge Channels . . . " , "c1=C1-" + filename + " c2=C2-" + filename + " c3=C3-" + filename + " create" ) ; } list = getFileList ( inputpath ) ; for ( i = 0; i < list . length; i++ ) scott_register ( inputpath , outputpath , list[i] ) ; We then corrected for uneven sample illumination or background artifacts using a rolling ball background substraction of 50 pixels . These processed images were subsequently used to analyze the signaling behavior of each bead . To analyze the images , beads were either identified and stored as regions-of-interest ( ROIs ) automatically from the GTPase fluorescent signal on beads using a custom macro that makes use of ImageJ’s 'find particles' function , or in the case of experiments with beads that deliberately contain differing levels of Ras density , beads were identified and stored as ROIs by hand . The particular parameters of the automated bead finding varied depending on the particular experiment , but a representative ImageJ macro is shown below: inputpath = "C:\Users\scoyl_000\Google Drive\MICROSCOPY\150403\processed\"; outputpath = "C:\Users\scoyl_000\Google Drive\MICROSCOPY\150403\beads\"; function scott_findbeads ( input , output , filename ) { open ( input+filename ) ; run ( "Split Channels" ) ; selectWindow ( "C3-" + filename ) ; run ( "Smooth" , "stack" ) ; //set threshold and find particles Stack . setPosition ( 1 , 18 , 1 ) ; setSlice ( 20 ) ; setThreshold ( 3744 , 23285 ) ; run ( "Analyze Particles . . . " , "size=400-15000 circularity=0 . 60-1 . 00 display exclude clear include add slice" ) ; //close any and all open windows close ( ) ; close ( ) ; close ( ) ; //reopen original file open ( input+filename ) ; //transfer ROIs to overlay and save run ( "From ROI Manager" ) ; saveAs ( "Tiff" , output+filename ) ; //clear manager roiManager ( "Delete" ) ; close ( ) ; } list = getFileList ( inputpath ) ; for ( i = 0; i < list . length; i++ ) scott_findbeads ( inputpath , outputpath , list[i] ) ; Once beads were identified and stored as ROIs , a variety of measurements were made for each bead using a custom macro . We measured the average total Ras fluorescence in the ROI , and the average fluorescent effector signal in the ROI at every timepoint in the experiment . We also measured the area and perimeter of the bead . As with other macros , the exact details of the code varied depending on the number of effectors we were simultaneously examining or other aspects of the setup , but a representative ImageJ macro is shown below: inputpath = "C:\Users\scoyl_000\Google Drive\MICROSCOPY\150318\one_shot_processed\"; outputpath = "C:\Users\scoyl_000\Google Drive\MICROSCOPY\150318\one_shot_data\"; setBatchMode ( true ) ; list = getFileList ( inputpath ) ; for ( i = 0; i < list . length; i++ ) crunchimage ( inputpath , outputpath , list[i] ) ; setBatchMode ( false ) ; //The Functions that are used in the macro are below function crunchimage ( input , output , filename ) { //open the image open ( input+filename ) ; //clear log file and reulsts table run ( "Clear Results" ) ; print ( "\Clear" ) ; //import ROIs from overlay run ( "To ROI Manager" ) ; //count ROis count=roiManager ( "count" ) ; //record Ras AMPS and RBD timecourse for each ROI for ( i=0;i<count;i++ ) { // first record the AMP for the Ras field; recordAMP ( i ) ; // next calculate the time series for the data recordTimecourse ( i ) ; // print newline marker print ( " ! " ) ; } selectWindow ( "Log" ) ; saveAs ( "Text" , output+filename ) ; close ( ) ; } function recordAMP ( index ) { //function that reports area , mean intensity , and perimieter // of an ROI used to get everything EXCEPT the timecourse data run ( "Clear Results" ) ; roiManager ( "Select" , index ) ; // select channel 2 and select the midpoint of hte stack Stack . setPosition ( 2 , 18 , 18 ) ; // make measurements run ( "Measure" ) ;BeadArea=getResult ( "Area" , 0 ) ;BeadMean=getResult ( "Mean" , 0 ) ;BeadPerim=getResult ( "Perim . " , 0 ) ;print ( BeadArea+" , "+BeadMean+" , "+BeadPerim+" , " ) ; } function recordTimecourse ( index ) { run ( "Clear Results" ) ; roiManager ( "Select" , index ) ; sliceCount=nSlices ( ) /2; // print ( sliceCount ) ; for ( k=0;k<sliceCount;k++ ) { Stack . setPosition ( 1 , 18 , k+1 ) ; run ( "Measure" ) ; timepointK=getResult ( "Mean" , k ) ; print ( timepointK+" , " ) ; } } The output of this macro is a file that contains a list of every single bead trace in the multi-tiff image . Each trace begins with the area measurement of the bead , the mean total Ras intensity of the bead , and the perimeter measurement of the bead , followed by the effector measurement of the bead at every timepoint . Each measurement ends with a ' , ' and is on a newline . At the end of each bead trace , a stop marker ' ! ' is printed . These data are then transformed by GREP and shell script into a CSV where each line contains all the relevant information about each single bead’s signaling trace . These data can then be loaded into either Matlab or Excel for further analysis . Once in Excel , data were typically further analyzed as follows: ( i ) intensity measurements were normalized to perimeter instead of area , ( ii ) the time-series data for a given bead was normalized such that the time-zero effector measurement was zero , and ( iii ) single bead traces were binned based on total GTPase levels to obtain statistics on the signaling behavior . Beads were assigned to the nearest of the 6 Ras density beads: 150 molecules / μm2 , 300 molecules / μm2 , 600 molecules / μm2 , 1200 molecules / μm2 , 2500 molecules / μm2 , 10 , 000 molecules / μm2 . The individual bead traces within a given bin were then averaged together to produce an average response for the associated density bin and network configuration . Each trace was typically the average of 15–80 beads from the combination of two independent experiments . The standard error of the mean for a given trace was typically <15% . Kintek Student Explorer ( Johnson et al . , 2009 ) was used to simulate the dynamic behavior of a variety of models for GTPase activation . Time was modeled in seconds and concentrations in nanomoles . Rate constants for association and dissociation of molecules from the GTPase were based on published Biacore measurements ( Fischer et al . , 2007 ) , and catalytic rate constants for GEF and GAP activities were based published solution measurements ( Bollag and McCormick , 1991; Freedman , 2006 ) . GEF was not modeled explicitly but rather directly incorporated in the rate constant for nucleotide release . Each simulation was allowed to run for 42000 s ( ~700 min ) . Three models were initially explored . The first model was a two-state GTPase model that did not account for competition between GAP and effector . This was modeled in Kintek using the following equations and parameters: k-k+G + T = GT10GT = GD0 . 0001*[GAP]0GD = G + D0 . 050GT + EFF = GT_EFF0 . 00010 . 001GT_EFF = GD + EFF0 . 00010 The second model was a two-state GTPase model that explicitly modeled competition between GAP and effector . This was modeled in Kintek using the following equations and ( physiological ) parameters: k+k-G + T = GT10GT = GD0 . 00010GD = G + D0 . 005050GT + EFF = GT_EFF0 . 00010 . 001GT_EFF = GD + EFF0 . 00010GT + GAP = GT_GAP0 . 00010 . 01GT_GAP = GD + GAP10 The third model was a three-state GTPase model that included an additional post-hydrolysis GTPase state ( GI ) which was refractory to GEF activation . This state converts to the GDP form on a slow timescale . This was modeled in Kintek as: k+k-G + T = GT10GT = GD0 . 00010GI = GD0 . 00010GD = G + D0 . 0050GT + EFF = GT_EFF0 . 00010 . 001GT_EFF = GD + EFF0 . 00010GT + GAP = GT_GAP0 . 00010 . 01GT_GAP = GD + GAP10 The third model was the best at explaining the transient behaviors of the system that we observed as well as the differences between WT and G12V Ras , and thus was used as the basis of all subsequent modeling and simulations . For any given simulation in the text , the initial conditions and any changes to associated rate-constants are indicated in the figure legend . Relative log-transformed expression levels for p120GAP , C-Raf , and H-Ras across a variety of cell-types and tissue-types were obtained from data contained within the Genevestigator software package ( see data in Table 1 ) . The three-dimensional phenotypes associated with each cell or tissue type was plotted as a 3D scatterplot using Matlab . | Cells sense and respond to the world around them using signaling “circuits” made of proteins and other molecules , and when an important cell circuit breaks , diseases like cancer may arise . Much like with electrical circuits , a given set of molecular components can be used to build different signaling circuits that behave in different ways . However , unlike for electrical circuits we generally do not have design manuals that allow us to work out how a signaling circuit behaves based on the components it includes . Doing this would involve identifying all the molecular parts of a circuit , using them to build every possible circuit , and carefully measuring the associated behavior . A group of proteins called the Ras-superfamily GTPases are important controllers of cell behavior . To investigate the behavior of Ras GTPase signaling circuits , Coyle and Lim built up different circuits from their components and “watched” their behavior with a microscope . Analyzing these behaviors provided the information needed to produce a ‘design manual’ for programming Ras circuits . Coyle and Lim found that the makeup of a Ras signaling circuit strongly affects the timing , duration , shape and size of its output . This means that different cells can use the same core components in different ways to build circuits customized to their specific needs . Nonetheless , this versatility comes with a trade-off: the circuits are fragile , and can break in many different ways to cause disease . In the future Coyle and Lim aim to build other types of important cellular signaling circuits from their component parts . Only by building these systems , turning them on and watching them run can we begin to understand how they actually perform and what they are capable of . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"computational",
"and",
"systems",
"biology"
] | 2016 | Mapping the functional versatility and fragility of Ras GTPase signaling circuits through in vitro network reconstitution |
Neuropeptides play a key role in the regulation of behaviors and physiological responses including alertness , social recognition , and hunger , yet , their mechanism of action is poorly understood . Here , we focus on the endocrine control ecdysis behavior , which is used by arthropods to shed their cuticle at the end of every molt . Ecdysis is triggered by ETH ( Ecdysis triggering hormone ) , and we show that the response of peptidergic neurons that produce CCAP ( crustacean cardioactive peptide ) , which are key targets of ETH and control the onset of ecdysis behavior , depends fundamentally on the actions of neuropeptides produced by other direct targets of ETH and released in a broad paracrine manner within the CNS; by autocrine influences from the CCAP neurons themselves; and by inhibitory actions mediated by GABA . Our findings provide insights into how this critical insect behavior is controlled and general principles for understanding how neuropeptides organize neuronal activity and behaviors .
Understanding how ensembles of neurons produce behaviors is an important aim of neuroscience . The mapping of the neural circuits that underlie a behavior is considered a necessary first step toward this goal , and efforts to determine the ‘connectome’ of different parts of the nervous system have been present since the beginnings of modern neuroscience . They start with the classical inferred synaptic relationships in Cajal’s anatomical analyses ( Ramón y Cajal , 1899 ) , through the detailed information on the wiring of some invertebrate circuits ( e . g . , Carew et al . , 1981; Comer and Robertson , 2001; King , 1976a , 1976b ) , culminating with the complete map of the Caenorhabditis elegans central nervous system ( CNS ) ( White et al . , 1986 ) , and the wiring diagrams of the Drosophila optic lobes ( Takemura et al . , 2013 ) and the mammalian retina ( Helmstaedter et al . , 2013 ) . Yet , research into the functioning of neuronal networks has revealed that a wiring diagram is usually not enough to understand what a neuronal network can do , although it does inform on its possible outcomes . One of the elements that adds tremendous multiplicity to the universe of possible outputs of a neural circuit is the action of neuromodulators , including biogenic amines and neuropeptides . In conjunction with classical transmitters , they can gate the input to a circuit or reconfigure its pattern of activity , thereby causing the same circuit to produce qualitatively different outputs ( Bargmann , 2012; Bargmann and Marder , 2013; Brezina , 2010; Leinwand and Chalasani , 2013; Marder , 2012; Nusbaum and Blitz , 2012 ) . The influence of neuropeptides can be profound and underlies the expression of entire behavioral states , such as hunger and satiation ( Atasoy et al . , 2012; Chambers et al . , 2013; Gao and Horvath , 2007 ) , pair bonding and stress ( Lieberwirth and Wang , 2014; Neumann and Landgraf , 2012 ) , and arousal and attention ( Li et al . , 2016 ) , and can involve many brain regions in addition to sensory and physiological inputs . How these actions are effected is poorly understood because neuropeptides can be broadly released within the CNS and exert combinatorial and non-linear effects ( Brezina , 2010 ) . In addition , we know little about how these gatekeepers are themselves regulated , yet such knowledge is also critical for understanding how the expression of a behavioral state is controlled . Here , we report on the genetic dissection , using Drosophila , of the response of a network of peptidergic neurons that controls the stereotyped and sequential insect behavior of ecdysis , and which represents a tractable system in which these questions can be addressed . Ecdysis is the vital behavior that is used by arthropods to shed the remains of the old cuticle at the end of every molt ( Reynolds , 1980 ) . It includes several behavioral subroutines and physiological events that are expressed sequentially to loosen and then shed the old cuticle , then expand and harden the new one . In insects , it is triggered by the sudden release of the neuropeptide , Ecdysis-Triggering Hormone ( ETH ) ( Ewer and Reynolds , 2002; Zitnan and Adams , 2012 ) , which activates sequentially a number of peptidergic neurons , each expressing the A isoform of the ETH receptor ( ETHR ) ( Diao et al . , 2016; Kim et al . , 2006 , 2015 ) . A current model proposes that each class of peptidergic ETH targets then activates or modulates specific phases of ecdysis behavior ( Kim et al . , 2006 ) . However , the mechanism responsible for producing the sequential activation of these targets in response to a common ETH stimulus is currently unknown . Here , we combined the use of genetically encoded calcium- and voltage-sensitive probes , targeted RNAi expression , mutants null for particular neuropeptides downstream of ETH , and pharmacology , to understand how this response is produced . For this we focused on the subset of neurons that produce CCAP ( Crustacean cardioactive peptide ) , which play a key role in the control of ecdysis ( Kim et al . , 2006 , 2015; Lahr et al . , 2012; Park et al . , 2003 ) . We show that the response of CCAP neurons to ETH and the ensuing ecdysis behaviors depend on direct actions mediated by the ETH trigger as well as on the actions of neuropeptides downstream of ETH together with inhibition mediated by GABA . Importantly , we found that removal of a downstream neuropeptide can eliminate the rhythmic pattern of neuronal activity induced by ETH , revealing that such neuropeptides are critical for the expression of the fundamental features of this neural response . Our findings have implications for understanding how this vital insect behavior is controlled . The principles that emerge are also relevant for understanding how peptidergic networks control behavioral states .
At the end of larval life , Drosophila enters the prepupal stage , then ecdyses to a pupa to initiate the transformation to the adult that occurs during metamorphosis . Pupal ecdysis consists of a sequence of behavioral subroutines , which starts with the preparatory behavior of pre-ecdysis , during which the hardened larval cuticle of the puparium is loosened from the underlying pupal cuticle through slow anteriorly directed movements of the body . This phase is followed by ecdysis proper , during which alternating left-right contractions lead into a phase of anteriorly directed peristalses that eventually cause head eversion , during which the brain is pushed anterior to the mouth . During the final phase of the behavioral sequence ( post ecdysis ) , alternating left-right contractions and then posteriorly directed movements produce a body with the external shape of an adult fly ( Kim et al . , 2006; Lahr et al . , 2012; Park et al . , 2003 ) . Each of these phases has a stereotyped duration and pattern of activity , which can be recorded in intact ( Kim et al . , 2015 , Lahr et al . , 2012; Park et al . , 2003 ) and puparium-free preparations ( Kim et al . , 2006; cf . , Figure 2 , below ) . In addition , fictive ecdysis can be visualized in ex vivo CNS preparations challenged with ETH that express the calcium sensor , GCaMP , in motoneurons ( Figure 1A , B ) . Consistent with the behaviors observed at ecdysis , the pattern of motor activity expressed in vitro in response to ETH consists of an initial phase that primarily recruits activity in the posterior region of the ventral nervous system ( VNS ) ( 'P' region , Figure 1Ab , Ba; corresponding to pre-ecdysis ) followed by a barrage of activity throughout the left and right sides of the VNS neuropils ( 'L' , Figure 1Ac; Bb; and 'R' regions , Figure 1Ad; Bd ) . Importantly , expansion of this latter section of the record reveals that the 'L' and 'R' regions are active in an alternating pattern ( Figure 1Be ) , consistent with the prominent left-right repetitive contractions seen at ecdysis in the intact animal . 10 . 7554/eLife . 19686 . 003Figure 1 . Fictive ecdysial behavior in normal and ETHR knockdown animals . ( A ) Snapshots of the pattern of GCaMP signal recorded from motoneurons of control animals at 0 min ( a ) , 10 min ( b ) , and at around 20 min ( c , d ) after in vitro challenge with 600 nM ETH . ( B ) Corresponding recording of GCaMP signal , color-coded according to regions indicated in ( Ab and Ac ) : ( a ) 'P' ( posterior , black trace; cf . , Ab ) ; ( b ) : 'L' ( left , blue trace; cf . , Ac ) ; ( c ) : 'C' ( center , green trace; cf . , Ac ) ; ( d ) 'R' ( right , red trace; cf . , Ac ) . ( e ) Expanded segment of recording ( recordings for 'P' , 'L' , 'C' and 'R' regions superimposed ) during 18–20 min period ( indicated by small bar beneath time axis of ( d ) ) . Note the alternating activity in 'L' and 'R' regions . ( C ) ( a ) Motoneuron activity patterns ( recordings for 'P' , 'L' , 'C' and 'R' regions superimposed ) from animals that express ETHR RNAi in CCAP neurons; ( b ) Expanded segment of recording shown in ( a ) ( recordings for 'P' , 'L' , 'C' and 'R' regions superimposed ) during 42–48 min period ( indicated by small bar beneath time axis in ( a ) ) . Note that 'L' and 'R' activity no longer alternate ( compare with Be ) . Zero min indicates time of ETH challenge in all records . Genotypes: Controls ( A , B ) : CCAP>GCaMP ( Ccap-GAL4 + UAS-GCaMP ) ; ETHR knockdown in CCAP neurons ( C ) : CCAP+MN>GCaMP+ETHR RNAi ( MN: C164 motoneuron GAL4; see Materials and methods ) . Note that this genotype also knocks down ETHR expression in motoneurons ( MNs ) . Nevertheless , knockdown of ETHR only in MNs had only a slight effect on ecdysis behavior ( cf . Figure 2 ) , suggesting that most defects observed here were due to knockdown of ETHR in CCAP neurons . In all experiments using RNAi , its effectiveness was boosted by including a UAS-dcr2 transgene . DOI: http://dx . doi . org/10 . 7554/eLife . 19686 . 003 At the endocrine level , ecdysis is initiated by the sudden and near-complete release of ETH from peripheral epitracheal cells , which is fueled by an endocrine positive feedback with centrally produced Eclosion Hormone ( EH ) ( Ewer et al . , 1997; Kingan et al . , 1997 ) . In the CNS , direct targets of ETH include neurons that express the neuropeptides , FMRFamide ( FMRF ) , Kinin , EH , and Crustacean Cardioactive Peptide ( CCAP ) , either alone or in combination with bursicon ( made up of two subunits , BURS and PBURS ) and/or Myoinhibitory Peptides ( MIPs ) ( Diao et al . , 2016; Kim et al . , 2006 ) . By targeting a calcium-sensitive GFP ( GCaMP ) to different peptidergic ensembles , Kim et al . ( 2006 ) showed that each of these sets of peptidergic neurons is activated at a particular time and for a specific duration following a challenge by ETH in vitro . By correlating these times with the expression of the different behavioral phases of ecdysis , each set was assigned a role in the control of particular ecdysial subroutine , which was also broadly consistent with functional and genetic evidence ( Diao et al . , 2016; Kim et al . , 2006 , 2015; Krüger et al . , 2015; Lahr et al . , 2012; Park et al . , 2003 ) . Nevertheless , the mechanisms by which the direct targets of ETH would be activated at different times after an ETH challenge remained unanswered . Here , we show that the pattern of activity of ETH targets depends on direct ETH actions , as well on actions effected by targets downstream of ETH . In order to understand how the temporal pattern of activity of peptidergic ETH targets is produced , we first identified relevant neuronal sets by determining the effects on ecdysis behavior of expressing ETHR RNAi in each set of peptidergic ETH targets . Expression of ETHR RNAi in FMRFa and EH neurons caused no measurable effect on ecdysis behavior in intact ( Figure 2A ) or puparium-free preparations ( Figure 2B ) ( 'FMRF>ETHR RNAi' and 'EH>ETHR RNAi' , respectively ) , other than an increase in the frequency of pre-ecdysial contractions . The corresponding pattern of neural activity induced by ETH was not visibly altered by expression of ETHR RNAi in these neurons ( not shown ) . In the case of EH neurons , this lack of effect could be due to their high sensitivity to ETH ( Kim et al . , 2006 ) , which could make them insensitive to the levels of reduction in ETHR expression that can be achieved by RNAi ( see below ) . Recent work has shown that the role of kinin neurons in the control of ecdysis behavior ( Kim et al . , 2015 ) may be mostly indirect and a consequence of its role in diuresis ( Diao et al . , 2016 ) . 10 . 7554/eLife . 19686 . 004Figure 2 . Impact on ecdysial behaviors of manipulating ETH effectiveness on downstream targets . ( A ) Ecdysial behaviors in intact puparium . Duration of pre-ecdysis ( left; open bars ) and ecdysis ( right; filled bars ) in controls ( CS and CCAP>GCaMP ) , and in animals expressing ETHR RNAi in different subsets of ETH targets . ( B ) Corresponding ecdysial behaviors of puparium-free preparations . Data are mean ± SEM . In ( A ) and ( B ) duration of ecdysial phases is indicated as not significantly different ( ‘ns' ) or significantly different ( '*': p<0 . 5; '**’: p<0 . 01 ) than those of CCAP>GCaMP control ( one-way ANOVA , Dunnett's post-hoc to control ) . Comparison of frequency of contractions per minute for pre-ecdysis and ecdysis ( labeled Fpre and Fecd , respectively ) in puparium-free preparations is indicated as not significantly different ( ‘ns' ) or significantly smaller ( '<': p<0 . 05; '<<’ p<001 ) or greater ( '>': p<0 . 05; '>>’ p<001 ) than that of CCAP>GCaMP controls ( one-way ANOVA , Dunnett's post-hoc to control ) . Actual p values can be found in Supplementary file 1 . Genotypes: all animals expressed GCaMP under control of Ccap-GAL4 ( Ccap-GAL4 + UAS-GCaMP ) . ETHR RNAi: UAS-ETHR RNAi; Df ( ETHR ) /+: hemizygosity for ETHR; MN>: motoneuron GAL4 ( C164; see Materials and methods ) . In all experiments using RNAi , its effectiveness was boosted by including a UAS-dcr2 transgene . DOI: http://dx . doi . org/10 . 7554/eLife . 19686 . 004 By contrast , expression of ETHR RNAi in CCAP neurons had a considerable effect on ecdysis behavior ( Figure 2; Kim et al . , 2015 ) . In particular , animals monitored free of the puparium expressed a significantly longer and weaker pre-ecdysis than did controls , and none ( 0/14 ) ecdysed ( Figure 2B ) . The defects in pre-ecdysis were rendered more severe in animals hemizygous for ETHR ( Figure 2 ) , revealing that the phenotype obtained by knockdown of ETHR using ETHR RNAi is functionally equivalent to that of an ETHR hypomorph . Fictive ecdysis behaviors , monitored by expressing GCaMP in motor neurons , showed that the underlying motor program was also severely disrupted in these animals , consistent with the expressed behavior . Indeed , unlike the sequences of calcium responses observed in the motoneurons of control animals ( Figure 1A , B ) , those of animals expressing ETHR RNAi in CCAP neurons responded to ETH with a smaller and sustained pattern of activity in the posterior region ( Figure 1C ) , consistent with the longer and weaker pre-ecdysis behavior observed ( Figure 2 ) , and a severe reduction and disorganization in the activity of L and R regions of the CNS , consistent with the observed lack of ecdysis behavior ( Figure 2 ) . These results confirm previous findings ( Kim et al . , 2006 , 2015; Lahr et al . , 2012; Park et al . , 2003 ) and reveal that CCAP neurons play a critical role in the control of Drosophila ecdysis behavior . Knockdown of ETHR in CCAP neurons also caused significant changes in the calcium response induced by ETH in vitro . In control animals , the two pairs of serial CCAP homologs showed a characteristic response to 600 nM ETH: whereas one set of neurons ( called here α neurons , Figure 3C–E ) responded ca . 20 min after ETH challenge with a barrage of GCaMP spikes , the other ( called here β neurons , Figure 3C , D ) responded after a similar delay with a sustained increase in fluorescence . Based on the electrophysiological activity recorded in Manduca ( Gammie and Truman , 1997 ) , we assume that the α neurons correspond to the projection ‘Cell 27’ neurons , whereas the β neurons correspond to interneurons ‘IN704’ . Expression of ETHR RNAi in CCAP neurons caused a significant delay in the onset of the response ( Figure 3F , H; Kim et al . , 2015 ) , and the period of activity was shorter and included a smaller number of peaks ( Figure 3F , I , J ) ( In these records , the responses of α and β neurons have been combined as they could not be unambiguously distinguished ) . This weakening of the response was seen in CCAP neurons in thoracic segment 3 ( TN3 ) as well as in serial homologs in abdominal segments 1–4 ( AN1-4 ) but was not detected in AN8-9 , which may not express ETHR ( Diao et al . , 2016; Kim et al . , 2006 ) . Further reduction of ETHR function , accomplished using ETHR hemizygosity , did not simply reduce the response further , revealing that indirect as well as direct ETH actions may be involved in determining the response of CCAP neurons to ETH . Indeed , in such animals , the latency to respond was generally more similar to that of the control ( Figure 3G , H ) , whereas its duration and number of spikes was more similar to those observed when ETHR RNAi was expressed in a wild-type background ( Figure 3I , J ) . Significantly , however , as illustrated in Figure 3F , G and quantitated in Figure 3—figure supplement 1 , both manipulations severely reduced the amplitude of the response . Thus , for animals expressing ETHR RNAi it was reduced four-fold for CCAP neurons AN1-4 and AN8-9 , and eight-fold in CCAP neurons TN3 , AN1-4 , and AN8-9 for animals expressing ETHR RNAi in a ETHR hemizygous background . This reduced response may be the basis for the much weaker behavior expressed by these genotypes ( Figures 1C and 2 ) . We do not know the cause of the differential effects on CCAP neurons TN3 vs . AN1-4 for these and other manipulations carried out in this study; we assume that they are due to differential ETHR expression and/or synaptic inputs . 10 . 7554/eLife . 19686 . 005Figure 3 . Impact on CCAP neuron activation of ETHR knockdown in CCAP neurons . ( A ) Schematic of Drosophila nervous system indicating the location of EH , FMRFa , and CCAP neurons . ( B–C ) Snapshots of the pattern of GCaMP signal in CCAP neurons in segments AN1-4 , recorded from wild-type animals at 0 min ( B ) and 20 min ( C ) after in vitro challenge with 600 nM ETH . ( D ) Calcium dynamics of AN1 α ( top , red trace ) and β ( lower , blue trace ) neurons ( cf . , 3C ) after ETH challenge . ( E–G ) Pattern of GCaMP activity recorded from CCAP neurons AN1-4 following in vitro challenge with 600 nM ETH in wildtype CNSs ( E ) , in CNSs expressing ETHR RNAi in CCAP neurons ( F ) , and in CNSs of ETHR hemizygous animals expressing ETHR RNAi in CCAP neurons ( G ) . ( H–J ) Quantitation of time of onset ( H ) , duration ( I ) , and number of spikes ( J ) for the different genotypes tested . TN3: thoracic ganglion 3; AN: abdominal ganglion . Zero min indicates time of ETH challenge . N = 10 preparations for all genotypes . Data are mean ± SEM . Different letters indicate statistically significant groups ( p<0 . 05 ) ; one-way ANOVA , Tukey's post-hoc multiple comparison analyses . Actual p values can be found in Supplementary file 1 . Genotypes: all animals expressed GCaMP under control of Ccap-GAL4 ( Ccap-GAL4 + UAS-GCaMP ) ; ETHR RNAi: UAS-ETHR RNAi; Df ( ETHR ) /+: hemizygosity for ETHR . In all experiments using RNAi , its effectiveness was boosted by including a UAS-dcr2 transgene . DOI: http://dx . doi . org/10 . 7554/eLife . 19686 . 00510 . 7554/eLife . 19686 . 006Figure 3—figure supplement 1 . Amplitude of GCaMP response induced by ETH in CCAP neurons . Quantitation of mean amplitude of GCaMP response to ETH of CCAP neurons in TN3 , AN1-4 , and AN8-9 for different genotypes and conditions ( Note: some of the results shown in this figure relate to data shown in Figures 4 , 5 , and 7 ) . ( A ) Results obtained for genotypes that used Ccap-GAL4 to drive GCaMP expression . ( B ) Results obtained for genotypes involving pburs mutants , for which Ccap-LexA was used to drive GCaMP expression ( see Figure 6 for further details ) . N = 6–13 for all conditions . Data are mean ± SEM . One-way ANOVA , followed by Dunnett's post-hoc analyses to control ( A ) or unpaired two-tailed t test to control ( B ) ; ‘*’ p<0 . 05 . Actual p values for all analyses can be found in Supplementary file 1 . All animals expressed Ccap-GAL4+GCaMP ( A ) or Ccap-LexA+LexAop-GCaMP ( B ) ; Df ( ETHR ) /+ indicates hemizygosity for ETHR; Eh[-]: Eh[-]/Df ( 3 ) Eh; pburs[-]: pburs[-]/Df ( 2 ) pburs . In all experiments using RNAi , its effectiveness was boosted by including a UAS-dcr2 transgene . DOI: http://dx . doi . org/10 . 7554/eLife . 19686 . 006 As an alternative approach to studying the response of CCAP neurons to ETH by changing their sensitivity to ETH , we investigated the effects of challenging the CNS with lower doses of ETH . As was observed when ETHR signaling was reduced using RNAi , such manipulations again revealed the presence of non-linear effects . Thus , for instance the lengthening of the latency observed using 300 nM was in general reversed with the lower concentrations ( 150 nM and 60 nM; Figure 4E ) . Interestingly , none of these concentrations significantly affected the duration of the response ( Figure 4F ) . Nevertheless , the responsiveness of neurons was severely affected , with only around 75% , 50% and 25% of neurons responding when challenged with 300 nM , 150 nM and 60 nM ETH1 , respectively ( vs . 100% for 600 nM ETH1 ) . Furthermore , the amplitude of the response was significantly decreased for all the lower concentrations used ( ten- to forty-fold , depending on segmental location and ETH concentration; Figure 4B–D-Figure; see also Figure 3—figure supplement 1 ) . 10 . 7554/eLife . 19686 . 007Figure 4 . Calcium responses induced by different concentrations of ETH in CCAP neurons . ( A–D ) Pattern of GCaMP activity recorded from CCAP neurons in AN1-4 from control CNSs challenged in vitro with 600 nM ( A ) , 300 nM ( B ) , 150 nM ( C ) , and 60 nM ( D ) ETH . Zero min indicates time of ETH challenge . ( E–G ) Quantitation of results , shown as described in Figure 3H–J . N = 10 for all preparations . Data are mean ± SEM . Different letters indicate statistically significant groups ( p<0 . 05 ) ; one-way ANOVA , Tukey's post-hoc multiple comparison analyses . Actual p values can be found in Supplementary file 1 . The responsiveness and amplitude of the responses were also affected; see text and Figure 3—figure supplement 1 . Individual measurements have been superimposed on the summary histogram when <8 neurons showed a measurable response . Genotypes: all animals expressed GCaMP under control of Ccap-GAL4 ( Ccap-GAL4 + UAS-GCaMP ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19686 . 007 These results show that the response of CCAP neurons to ETH affects the expression of ecdysis behaviors and that this response depends in a nonlinear fashion on the levels of both ligand and receptor . This non-linearity could simply reflect the sigmoidal dose-response curve of GPCRs or may also be caused by inputs to CCAP neurons that are also ( direct or indirect ) targets of ETH . Eclosion hormone ( EH ) is a key neuropeptide in the control of ecdysis ( Ewer and Reynolds , 2002; Gammie and Truman , 1999; Krüger et al . , 2015 ) . It is involved in potentiating and accelerating the release of ETH that causes the sudden onset of ecdysis behaviors ( Ewer et al . , 1997; Kingan et al . , 1997 ) ; in addition , it plays a critical role within the CNS for the expression of ecdysis behaviors . In Drosophila larvae , for instance , mutants lacking EH do not express pre-ecdysis , transitioning directly into a prolonged ecdysis-like phase , which is , nevertheless , generally ineffective in causing the shedding of the old cuticle ( Krüger et al . , 2015 ) . Eh mutant animals also expressed specific behavioral defects at pupal ecdysis , which , interestingly , differed from those expressed in the larva . Thus , intact Eh hemizygous mutants expressed much longer pupal pre-ecdysis ( Figure 5A ) and ecdysis behavior was also longer , with most animals ( 4/10 ) failing to express the behavior . Similar defects were observed in puparium-free preparations ( Figure 5B ) , although in this case none ( 0/10 ) expressed the ecdysial phase of the behavior . These defects were all qualitatively rescued using a transgene containing the Eh gene , indicating that they were mostly due to the lack of EH ( Figure 5A , B , for intact and puparium-free preparations , respectively ) . 10 . 7554/eLife . 19686 . 008Figure 5 . Absence of EH abolishes ecdysis and alters the response of CCAP neurons to ETH . ( A , B ) Duration of pre-ecdysis ( left; open bars ) and ecdysis ( right; filled bars ) behavior of animals hemizygous for Eh in intact ( A ) and puparium-free preparations ( B ) , shown as described in Figure 2 . Behavioral defects in ( A , B ) were qualitatively rescued by transgene containing Eh gene , indicating that they are caused by lack of EH neuropeptide . ( C , D ) Pattern of GCaMP activity in CCAP neurons AN1-4 induced by ETH in CNS from control animals ( C ) and from animals hemizygous mutant for Eh ( D ) . Zero min indicates time of ETH challenge . ( E ) Summary of results obtained for latency ( left ) and duration ( right ) of response . ( F ) Recording from motoneurons from CNS of controls ( left; cf . Figure 1B ) and animals hemizygous for Eh ( right ) ; traces color-coded as described in Figure 1 . Zero min indicates time of ETH challenge . N = 6–10 for all genotypes . Data in ( A , B , and E ) are mean ± SEM . One-way ANOVA , Dunnett's post-hoc to control; ( *: p<0 . 05; **: p<0 . 01 ) ; actual p values can be found in Supplementary file 1 . The amplitude of the responses was also affected; see text and Figure 3—figure supplement 1 . In ( E ) , individual measurements have been superimposed on the summary histogram when < 8 neurons showed a measurable response . Genotypes: in ( A–D ) all animals expressed GCaMP under control of Ccap-GAL4; in ( F ) , they expressed GCaMP under control of MN-GAL4 ( C164-GAL4 ) . Eh[-]: Eh[-]/Df ( 3 ) Eh; Eh[-]+P{Eh}: Eh[-]/Df ( 3 ) Eh; P{Eh}; see Materials and methods for exact genotypes . DOI: http://dx . doi . org/10 . 7554/eLife . 19686 . 008 Drosophila EH neurons are direct targets of ETH ( Diao et al . , 2016; Kim et al . , 2006 ) and respond to ETH with a shorter latency than do CCAP neurons ( Kim et al . , 2006 ) ; in addition , the response of larval CCAP neurons to ETH is severely weakened in the absence of EH ( Krüger et al . , 2015 ) . These results suggest that EH neurons are an ETH target that is upstream of CCAP neurons . In order to explore the role of EH in determining the features of the response of CCAP neurons to ETH at pupal ecdysis , we examined the GCaMP signal of CCAP neurons to ETH in animals hemizygous mutant for Eh . As shown in Figure 5D , the temporal organization of the response was dramatically disorganized in these animals , with a significant loss in the regular pattern of GCaMP spikes that is typical of the control ( Figure 5C ) . This effect was partially reflected in the quantitation of the response , where the lack of EH caused significant increases in the latency to respond ( Figure 5E ) and a two-fold decrease in the amplitude of the response in CCAP neurons AN1-4 and AN8 , 9 ( Figure 5D; see also Figure 3—figure supplement 1 ) , although the severely disrupted response ( cf . , Figure 5D ) precludes a quantitative comparison . These results show that , as occurs in the larva , the response of CCAP neurons to ETH depends critically on the action of EH . Animals hemizygous mutant for Eh also expressed a severely weakened and disrupted pattern of motoneuron activity in response to ETH ( Figure 5F ) . This aberrant response is consistent with the failure of ecdysis behavior observed in this genotype ( Figure 5A , B ) , although in the intact animal the absence of EH would likely also prevent ETH release ( Krüger et al . , 2015 ) , which may also contribute to the failure of ecdysis , Bursicon , the so-called tanning hormone , has traditionally been implicated in post-ecdysial events following adult emergence ( Honegger et al . , 2008 ) . However , genetic analyses have shown that pburs , which encodes one of the subunits of the heterodimeric bursicon hormone , also plays a role in the control of ecdysial behaviors ( Lahr et al . , 2012 ) . Bursicon is of particular interest in the context of this work because it is expressed by a subset of the CCAP neurons themselves and could therefore play a ( direct or indirect ) autocrine role in determining the response of these neurons to ETH . Consistent with the findings of Lahr et al . ( 2012 ) , the lack of PBURS ( and thus of the heteromeric bursicon ) in intact puparia caused a lengthening of ecdysis , with only 5/8 animals executing the behavior ( Figure 6A ) . Puparium-free preparations lacking PBURS , by contrast , expressed a significantly shorter ecdysis motor program ( Figure 6B ) ; a similar defect was obtained expressing pburs RNAi in CCAP neurons ( Figure 6B ) . Importantly , the defects expressed by hemizygous mutant animals in both types of preparations were rescued by a transgene containing the pburs gene ( Figure 6A , B ) indicating that they are due to the lack of pburs function . The apparently opposite phenotypes expressed by these two types of preparations are likely due to the different criteria that are used to define the ecdysial phases in intact vs . puparium-free animals . At the level of the response of CCAP neurons , the lack of PBURS caused , on average , a two-fold decrease in the amplitude of the response to ETH of CCAP neurons AN1-4 ( Figure 7B; see also Figure 3—figure supplement 1 ) . ( These and other experiments involving mutations in pburs had to be done using the LexA/LexAop binary expression system due to interference from GAL4; see legend to Figure 6 for further details . ) It also accelerated the time of onset and shortened the response for some of the CCAP neurons ( Figure 7C , D ) . No effects on the timing or amplitude of the response were detected in CCAP neurons AN8-9 , suggesting that they may not express the bursicon receptor . In summary , the absence of pburs function caused quantitative changes to the response of CCAP neurons to ETH , revealing that a neurohormone made by the CCAP neurons themselves participates in determining their response to the ETH trigger . 10 . 7554/eLife . 19686 . 009Figure 6 . Ecdysial behaviors of pburs mutant animals . ( A , B ) Duration of pre-ecdysis ( left; open bars ) and ecdysis ( right; filled bars ) behavior of animals hemizygous for pburs in intact ( A ) and in puparium-free preparations ( B ) , shown as described in Figure 2 . Data are mean ± SEM; one-way ANOVA , Dunnett's post-hoc to control . Actual p values can be found in Supplementary file 1 . Lack of pburs function did not significantly alter the duration of the pre-edysis phase ( unless CCAP neurons also expressed GAL4 , see below ) , whereas it lengthened ( intact preparations ) or shortened ( puparium free preparations ) the duration of ecdysis; a similar result was obtained when pburs function was knocked down using RNAi . In ( A ) , note that expression of GAL4 in CCAP neurons accentuated the defects expressed by pburs hemizygotes; these defects could not be rescued by pburs-containing transgene and were therefore not caused only by the lack of pburs function . For this reason , GCaMP responses to ETH of pburs mutants were assessed using Ccap-LexA driver ( cf . , Figure 7 ) . Genotypes: Control ( GAL4 ) : CCAP-GAL4+UAS-GCaMP; Control ( LexA ) : CCAP-LexA+LexAop-GCaMP . All genotypes including ' ( GAL4 ) ' or ( 'LexA' ) contained CCAP-GAL4+UAS-GCaMP or CCAP-LexA+LexAop-GCaMP , respectively . pburs[-]: pburs[-]/Df ( 2 ) pburs; pburs[-]+P{pburs}: pburs[-]/Df ( 2 ) pburs; P{pburs}; see Materials and methods for exact genotypes . In all experiments using RNAi , its effectiveness was boosted by including a UAS-dcr2 transgene . DOI: http://dx . doi . org/10 . 7554/eLife . 19686 . 00910 . 7554/eLife . 19686 . 010Figure 7 . Absence of PBURS affects response of CCAP neurons to ETH . ( A , B ) Pattern of GCaMP activity in CCAP neurons AN1-4 induced by ETH in CNS from control animals ( A ) and in CNS of animals hemizygous mutant for pburs ( B ) . Zero min indicates time of ETH challenge . ( C , D ) Summary of results obtained for latency ( C ) and duration ( D ) of response , summarized as described in Figure 5 . N = 7–10 for all genotypes and preparations . Data in ( C , D ) are mean ± SEM . Significant differences ( p<0 . 05 ) compared to control are indicated by '*'; t test ( unpaired , two-tailed ) . Actual p values can be found in Supplementary file 1 . The amplitude of the responses was also affected; see text and Figure 3—figure supplement 1 . Genotypes: Control: CCAP-LexA+LexAop-GCaMP . pburs[-]: pburs[-]/Df ( 2 ) pburs + CCAP-LexA+LexAop-GCaMP; see Materials and methods for exact genotypes . DOI: http://dx . doi . org/10 . 7554/eLife . 19686 . 010 A striking feature of the responses triggered by ETH on its peptidergic targets is that , despite all being direct targets of this triggering hormone , the various subsets of peptidergic neurons are not activated simultaneously following an in vitro ETH challenge . Thus , although EH ( Kim et al . , 2006 ) and Kinin ( Diao et al . , 2016; Kim et al . , 2015 ) neurons respond within ca . 10 min of an ETH challenge , CCAP neurons do not respond until around ca . 20 min after ETH stimulation , with different subsets showing different and characteristic latencies ( Figure 8C , F; Kim et al . , 2006 ) . The delay in the execution of the ecdysial phase of ecdysis behavior , which correlates with the time of activation of CCAP neurons , has been hypothesized to occur through inhibitory influences originating from the brain and/or subesophageal ganglion ( Baker et al . , 1999; Ewer and Reynolds , 2002 ) ; Ewer and Truman , 1997; Fuse and Truman , 2002; Zitnan and Adams , 2000 ) . Nevertheless , the exact origin and nature of this inhibition is currently unknown . In order to investigate the role of inhibition in the response of CCAP neurons to ETH , we first examined their time course of activation in the presence of the GABAA receptor ( GABA-RA ) blocker , picrotoxin ( 100 µM; Rohrbough and Broadie , 2002 ) . As shown in Figure 8D , F , pharmacological inhibition of GABA-RA caused a significant reduction in the latency to respond; this effect was most dramatic in neurons from segments AN1-4 , where the time to onset was reduced by almost 50% ( from around 20 to 10 min ) . Other changes were also apparent in these preparations . Most noteworthy was the more sustained nature of the response ( Figure 8D ) , which suggests that the pronounced spikes that are normally seen ( e . g . , Figure 8C ) are partly accomplished through inhibitory processes mediated by GABA . Similar results were obtained using the GABA-RA inhibitor , gabazine ( 100 µM; Hosie and Sattelle , 1996; data not shown ) . 10 . 7554/eLife . 19686 . 011Figure 8 . GABA inhibition controls latency to ecdysis and of CCAP response to ETH . ( A , B ) Duration of pre-ecdysis ( left; open bars ) and ecdysis ( right; filled bars ) behavior in intact ( A ) and in puparium-free ( B ) preparations expressing GABA-RA RNAi in CCAP neurons; results are summarized as shown in Figure 2 . Animals also expressed GCaMP under control of Ccap-GAL4 . ( C–E ) Pattern of GCaMP activity in CCAP neurons AN1-4 induced by ETH in CNSs from control animals ( C ) , in CNSs from control animals recorded in the presence of GABA-RA antagonist picrotoxin ( 100 µM ) ( E ) , and in CNSs expressing GABA-RA RNAi in CCAP neurons ( E ) . ( F–H ) Quantitation of results , shown as described in Figure 3H–J . Zero min indicates time of ETH challenge . N = 9–13 for all genotypes and preparations . Data in ( A , B , F–H ) are mean ± SEM . For panels A and B , data for experimental groups were compared to those of control ( one-way ANOVA , Dunnett's post-hoc to control ) , and summarized as described in Figure 2 . For panels F–H , different letters indicate statistically significant groups ( p<0 . 05 ) ; one-way ANOVA , Tukey's post-hoc multiple comparison analyses . Actual p values for all analyses can be found in Supplementary file 1 . All animals also expressed GCaMP under control of Ccap-GAL4 ( Ccap-GAL4 + UAS-GCaMP ) . In all experiments using RNAi , its effectiveness was boosted by including a UAS-dcr2 transgene . DOI: http://dx . doi . org/10 . 7554/eLife . 19686 . 011 In order to determine the contribution of inhibitory processes acting directly on CCAP neurons , we expressed GABA-RA RNAi in these ETH targets . As illustrated in Figure 8E and quantitated in Figure 8F , this manipulation caused a significant shortening of the response latency , which , furthermore , was similar to that observed following picrotoxin inhibition ( Figure 8D , F ) , but without significantly affecting the other features of the response ( duration and number of spikes , Figure 8G and H , respectively ) . Importantly , the corresponding animals expressed a significantly shorter pre-ecdysis , thereby reducing the latency to ecdysis ( Figure 8A , B ) . These results show that around 50% of the latency of the response of CCAP neurons to ETH is caused by GABA inhibition acting directly on these neurons . They also again show the importance of CCAP neurons in the control of ecdysis since reducing the latency of the onset of the GCaMP response was accompanied by a comparable shortening in the time of onset of ecdysis . In order to more directly visualize the inhibitory processes acting on CCAP neurons , we carried out optical voltage recordings of these neurons using the genetically encoded voltage sensor , ArcLight ( Cao et al . , 2013 ) . As shown in Figure 9B , two different responses could be recorded in preparations where an α + β pair of CCAP neurons was in focus ( neurons from neuromeres A8 and A9 could not be visualized so are omitted from these analyses ) : although both neurons showed an initial plateau ( top red and blue records in Figure 9B ) followed by a depolarization at around 20 min ( black arrowheads in Figure 9B ) , one of them ( top red record in Figure 9B ) then showed a series of spikes , whereas the other ( top blue record in Figure 9B ) showed a more sustained response . We assume that the neuron expressing the spikes corresponds to the α neuron ( Figure 3D ) , whereas the other corresponds to the β member of the pair . Most importantly , the main effect of challenging the preparations with ETH in the presence of picrotoxin was a shortening of the latency to depolarize , without significantly affecting the total duration of the entire response ( Figure 9C ) . Finally , a quantitative analysis of the latency and duration of the voltage vs . GCaMP response ( Figure 9D ) reveals that both responses show an overall very similar timecourse . These results suggest that GABA inhibition delays the depolarization of CCAP neurons caused by ETH . This inhibition would then be lifted at around 20 min , causing the depolarization and firing of CCAP neurons , which results in the calcium spikes recorded using the GCaMP sensor ( Figure 8C ) and the activation of the ecdysis motor program . 10 . 7554/eLife . 19686 . 012Figure 9 . GABA delays depolarization of CCAP neurons induced by ETH . ( A ) ArcLight fluorescence in CCAP neurons α and β from ganglion AN1-2 . ( B ) ArcLight signal recorded from α ( red trace ) and β ( blue trace ) neurons following ETH stimulation in control CNS ( +ETH; top traces ) and in the presence of picrotoxin ( +ETH+Px , lower traces ) . Zero min indicates time of ETH challenge . Inverted triangles indicate start and end of depolarization; 'Duration' corresponds to the time between these two events . ( C ) Quantitation of results showing time of onset of depolarization ( top ) and duration of depolarization ( bottom ) in CCAP neurons from TN3 and AN1-4 ( neurons in AN8 and AN9 could not be visualized ) . ( D ) Comparison between timecourse of GCaMP and ArcLight signal . Data in ( C , D ) are mean ± SEM . N = 10–13 for all genotypes . In ( C ) , statistically significant differences ( p<0 . 05 ) relative to control are indicated with '*'; t test ( unpaired , two-tailed ) . In ( D ) , statistically significant differences ( p<0 . 05 ) are indicated by different letters; one-way ANOVA , Tukey's post-hoc multiple comparison analyses . Actual p values for all analyses can be found in Supplementary file 1 . Individual measurements have been superimposed on the summary histogram when <8 neurons showed a measurable response . Genotype of animals expressing ArcLight sensor: CCAP>Arclight; of those expressing GCaMP: CCAP>GCaMP . DOI: http://dx . doi . org/10 . 7554/eLife . 19686 . 012
Ecdysis behavior consists of behavioral routines and physiological events that are expressed in a specific sequence . At the neural level , the sequential nature of ecdysis is based on the sequential activation of different ETH targets . Here , we have identified some of the elements involved in determining the time course of activation of CCAP neurons . The CCAP neurons express ETHR , and by targeting ETHR RNAi to these neurons we showed that the timecourse and intensity of their response to ETH is sensitive to the dosage of ETHR . The associated lengthening of pre-ecdysis and failures in ecdysis underscore the well-established role for these neurons as key regulators of ecdysis ( Kim et al . , 2006 , 2015; Lahr et al . , 2012; Park et al . , 2003 ) . In addition to ETH , however , we show that several elements downstream of this triggering hormone including EH , PBURS , and GABA play a key role in determining the response of CCAP neurons ( whether GABA neurons mediating this inhibition are direct or indirect targets of ETH is currently unknown ) . Thus , we found that the response of CCAP neurons is qualitatively changed in the absence of EH , causing a decrease in the amplitude and dramatically altering the temporal features of the response , with a concomitant failure to ecdyse . Interestingly , we show that PBURS , which is produced by a subset of CCAP neurons , is also involved in patterning the response of these neurons to ETH , revealing a ( direct or indirect ) autocrine regulation of CCAP activation . Finally , we show that GABA inhibition mediates at least part of the delay between exposure to ETH and CCAP neuron activation and may also sculpt the shape of the resulting pattern of activation . Our findings are summarized in the model shown in Figure 10; the model details the contributions to the activation of CCAP neurons from ETH , peptides produced by targets of ETH including the CCAP neurons themselves , as well as inhibitory effects mediated by GABA , and define the times when these influences participate during the expression of the different phases of ecdysis . In addition to these actions , there are several features of the response to ETH whose origins await elucidation . For instance , decreasing ETH effectiveness using RNAi or different concentrations of ETH revealed the existence of non-linear processes whose basis is currently unknown . They could simply reflect the non-linear activation of GPCRs and/or may occur because the various targets that are activated or inhibited by ETH may respond differentially and non-linearly to different concentrations of this neuropeptide , as occurs in other cases of neuromodulation ( Brezina , 2010; Marder et al . , 2014 ) . Similarly , there remains a 10-min delay in the onset of the response of CCAP neurons that is not mediated by GABA inhibition . Dissecting the contributions of different modulators and their targets to the response of this neuronal ensemble will undoubtedly be aided by the availability of recently developed genetic tools ( Diao et al . , 2015; Luan et al . , 2006 ) that can be used for the precise manipulation of receptor spatial expression . Finally , we still do not know how the activation of the various peptidergic targets of ETH causes the production of the ecdysial motor programs ( cf . , Figure 1A , B ) . Preliminary evidence indicates that CCAP may play an important role in directly activating these motor programs because the timing of GCaMP activity in motoneurons follows closely the pattern of GCaMP activity of CCAP neurons ( Mena and Ewer , unpublished ) . 10 . 7554/eLife . 19686 . 013Figure 10 . Model for endocrine control of ecdysial behaviors . ( A ) ETH released from peripheral endocrine cells acts directly on EH and CCAP neurons , and directly or indirectly on GABA neurons . Release of ETH is further potentiated by ETH-induced EH release ( reciprocal arrows ) . Direct GABA inhibition of CCAP neurons prevents onset of response . ( B ) EH and/or ETH turn on the preparatory phase of ecdysis; waning of GABA inhibition , and EH and autocrine PBURS action ( presumably mediated by BURS+PBURS bursicon heterodimer ) on CCAP neurons activates CCAP neurons . ( C ) Activation of CCAP neurons causes expression of ecdysis proper and silencing of pre-ecdysis phase . Except for reciprocal relationship between EH and ETH , and ETH actions on CCAP neurons , none of the actions indicated are known to be direct . Arrows indicate stimulation; cross bars indicate inhibition . DOI: http://dx . doi . org/10 . 7554/eLife . 19686 . 013 In many motor systems , repetitive motor outputs are produced by central pattern generators ( CPG’s ) whose exact pattern of activity is then modulated by biogenic amines and neuropeptides ( Brezina , 2010; Marder et al . , 2014; Nusbaum and Blitz , 2012 ) . In the case of the ecdysis motor program , intrinsic modulatory actions ( Katz , 1995 ) appear to play a defining role in the expression of the neural response elicited by ETH and of its accompanying motor output . Thus , for example , the lack of EH destroys the repetitive pattern of firing of CCAP neurons induced by ETH ( Figure 5D ) as well as that of the resulting motor output ( Figure 5A , B and F ) . What might the advantage be of not hardwiring the neural bases of such a vital behavior as ecdysis ? One clue may lie in the fact that , despite the diversity in ecdysial behaviors expressed by different arthropods during different molts , the neuropeptides that drive ecdysis as well as their receptors are extremely well conserved , and clear homologs can be identified even in chelicerates ( Christie et al . , 2011; Grbić et al . , 2011; Veenstra et al . , 2012 ) , which diverged from the insects ca . 600 million years ago ( Regier et al . , 2005 , 2010 ) . One way for such a conserved signaling pathway to produce different behavioral outputs would be to change the pattern of receptor expression . Consistent with this hypothesis , the exact function of particular ecdysial neuropeptides can differ in different insect groups and stages ( White and Ewer , 2014 ) . For instance , whereas Drosophila lacking CCAP express normal ecdysis behaviors ( Lahr et al . , 2012 ) , RNAi inhibition of CCAP causes ecdysial failures in Tribolium ( Arakane et al . , 2008 ) . Likewise , bursicon is involved in postecdysial maturation following the emergence of the adult fly ( Honegger et al . , 2008 ) , yet is involved in the control of the earlier ecdysial phase at pupation ( Lahr et al . , 2012 ) . Finally , it is especially interesting to identify an inhibitory input to CCAP neurons mediated by GABA , as it could provide a route through which sensory input could modulate the time of onset of the ecdysial phase . Such an input appears to be absent in Drosophila larval and pupal ecdysis but plays an important role in orthopteran ecdysis ( Carlson and Bentley , 1977; Ewer and Reynolds , 2002 ) as well as in dipteran ( Baker et al . , 1999 ) and lepidopteran adult emergence ( Ewer and Truman , 1997 ) . It will be very interesting to determine the extent to which differences in neuropeptide receptor expression underlie the diversity of ecdysial behaviors expressed by insects with different body plans . In other systems , the different responses elicited by well-conserved neuropeptides may also be accomplished through changes in receptor expression . Such appears to be the case for oxytocin in voles , where differences in pair bonding in monogamous vs . gregarious species are at least in part due to differences in receptor distribution ( Young et al . , 1999 ) . Neuropeptides can be released at synapses , along axons , and even from dendrites , from where they can influence distant targets with a timecourse that can span from seconds to hours ( Leng and Ludwig , 2008; van den Pol , 2012 ) . Despite this mode of delivery , with low spatial and temporal specificity , their actions are typically very specific and can lead to the expression of tightly regulated responses . In the case of ecdysis behavior , for example , EH and CCAP are both broadly released into the ventral CNS , yet the timing of pre-ecdysis and ecdysis is normally extremely precise ( Figure 2 ) . Similarly , in the case of the mammalian circadian system , robust and precise circadian rhythmicity of pacemaker activity in the suprachiasmatic nucleus ( SCN ) and of the behavioral output depends on Vasoactive Intestinal Peptide ( VIP ) , which is released in a broad paracrine manner in the SCN ( Hastings et al . , 2014 ) . A similar situation may occur in the Drosophila circadian system , where the neuropeptide , Pigment Dispersing Factor ( PDF ) , is critical for the coupling of the different circadian oscillators ( Lin et al . , 2004; Shafer and Yao , 2014; Yao and Shafer , 2014; Yoshii et al . , 2009 ) . Overall , these examples show that neuropeptides play a critical role in determining the precise output of a neural network and of the behavior it controls . In the case of ETH , the precision of the timing and phasing of the different ecdysial phases depends on its combinatorial actions with other neuropeptides including EH and CCAP , and which are likely to be highly non-linear ( Brezina , 2010; Marder et al . , 2014 ) . A similar situation applies for other neuropeptide controlled behaviors such as hunger , where neurons in the hypothalamic arcuate nucleus that express Agouti-related protein ( Agrp ) regulate feeding by integrating inputs mediated by circulating hormones ( ghrelin and letpin ) , transmitters ( glutamate , and GABA ) and other peptidergic neurons ( POMC neurons ) ( Atasoy et al . , 2012; Gao and Horvath , 2007; Sohn et al . , 2013 ) . These inputs provide a readout of the animal’s physiological state and are then translated into a non-linear switch-like behavior , consistent with the switch-like features of hunger states ( Yang et al . , 2011 ) . For most systems , however , we lack information on the detailed actions exerted by each neuropeptide within a network . The inner working of the peptidergic network that controls ecdysis provides a tractable system for understanding how neuropeptides trigger and modulate complex patterns of neuronal activity and behaviors .
Flies stocks were maintained at room temperature ( 22–25°C ) on standard agar/cornmeal/yeast media . Unless indicated otherwise , all strains were obtained from the Drosophila Bloomington stock center ( Bloomington , Indiana , USA ) . The following GAL4 drivers were used; expression and reference are indicated in parenthesis: Ccap-GAL4 ( CCAP neurons; Park et al . , 2003 ) ; Eh-GAL4 ( EH neurons; McNabb et al . , 1997 ) ; FMRFa-GAL4 ( subset of FMRFa neurons; Suster et al . , 2003 ) ; C164-GAL4 ( motoneurons; Torroja et al . , 1999 ) . RNAi lines were obtained from Vienna Drosophila RNAi Center ( VDRC; Vienna , Austria ) or the National Institute of Genetics ( NIG; Shizuoka , Japan ) : UAS-ETHR RNAi ( VDRC #42717 ) ; UAS-pburs RNAi ( NIG 15284 R-1 ) . In all experiments using RNAi , its effectiveness was boosted by including a UAS-dcr2 transgene . The null allele of pburs is described in Lahr et al . ( 2012 ) ; the null allele of Eh and an Eh rescue transgene are described in Krüger et al . ( 2015 ) . UAS-GCaMP 3 . 2 was kindly provided by J . Simpson ( HHMI , Janelia farms , USA ) . Homozygous null animals were always heterozygous for a null allele in combination with a genetic deletion that included the relevant gene ( Df ( 2 ) Exel6036 for pburs; Lahr et al . ( 2012 ) and Df ( 3 ) Eh[-] for Eh; Krüger et al . , 2015 ) . Stocks with homozygous lethal mutations were maintained heterozygous with actin-GFP expressing balancer chromosomes . All UAS-transgene bearing flies were crossed with wild-type animals to create heterozygous controls . Imaging of Ca2+ dynamics was carried out essentially as described in Kim et al . ( 2006 ) . Briefly , animals containing a bubble in the mid-region of the puparium ( late stage p4 ( i ) ; Bainbridge and Bownes , 1981 ) were selected . They were then further staged to be within 4 hr before onset of pupal ecdysis . The animals were then dissected under cold PBS ( 137 mM NaCl , 2 . 7 mM KCl , 10 mM Na2HPO4 and 2 mM KH2PO4 , pH 7 , 3 ) , placed on the surface of 200 µl , 1 . 5% low melting temperature agarose solution ( Sigma type VII-A; Sigma-Aldrich Chemical Co . , MO ) , which was then left to harden for 30 min in a humidified chamber kept at 10–15°C . Preparations were then covered with Schneider´s Insect Medium ( Sigma-Aldrich Chemical Co . , MO ) and imaged under an Olympus DSU Spinning Disc microscope ( Olympus Corporation , Shinjuku-ku , Tokyo , Japan ) using a 20 X W NA 0 . 50 or 40 X W NA 0 . 80 immersion lens . Fluorescent images were acquired using an ORCA IR2 Hamamatsu camera ( Hamamatsu Photonics , Higashi-ku , Hamamatsu City , Japan ) using the Cell^R Olympus Imaging Software ( Olympus Soft Imaging Solutions , Munich , Germany ) . Preparations were first imaged for 5 min ( exposures taken every 5 s ) and those showing spontaneous activity ( ca . 5% of the preparations ) were discarded . They were then stimulated with synthetic ETH1 ( Bachem Co . , USA; referred to here simply as ETH ) and GFP fluorescence captured every 5 s for 30 , 60 , or 90 min for the Ca2+ sensor , GCaMP , and every 2 s for 60 min for the voltage sensor , ArcLight . When used , antagonists for GABA receptor A , Picrotoxin ( Sigma-Aldrich Chemical Co . , MO ) or Gabazine ( SR95531; Tocris Bioscience , Bristol , UK ) , were added 10 min prior to ETH1 challenge . Recordings were analyzed using Cell^R Olympus Imaging Software ( Version 2 . 6 ) and fluorescence intensity calculated as ∆F/F . The data were further processed with Excel ( Microsoft , WA ) ; statistical analyses were carried out using Prism 6 . 0 ( Graphpad Software Inc , CA ) . In general , only 50–75% of CCAP neurons were in focus in a given preparation and could therefore be quantitated . However , since for most genotypes and manipulations 90–100% of neurons showed a measurable response , each preparation yielded at least one independent measurement per segment . Exceptions to this were neurons in segment T3 , where only one pair of neurons showed a measurable signal; and for some genotypes and manipulations , where many neurons failed to respond ( e . g . preparations challenged with 60 nM ETH1; cf . Figure 4 ) . In cases where <8 measurements were obtained , individual data points have been superimposed on the relevant summary histograms ( Figures 4 , 5 , and 9 ) . | Most behaviors occur only under specific circumstances: we eat when we are hungry , for example . But how does the nervous system decide when to start or stop a particular behavior ? Molecules called neuropeptides are thought to play a key role in these decisions . Neuropeptides are produced by organs throughout the body and also by the nervous system itself . When neuropeptides act on neurons responsible for a particular behavior – such as feeding – they can inform those neurons about conditions elsewhere in the body and the brain . This enables the nervous system to decide whether to start or stop the behavior . Yet , how the signals from the different neuropeptides are integrated is poorly understood . As immature insects grow , they regularly molt then shed their outer skeleton – or cuticle – in a process called ecdysis . This requires a series of behaviors to occur in a particular order . The old cuticle is first loosened and shed , and then the new cuticle expands and hardens . A number of neuropeptides control ecdysis: for example , a key neuropeptide called ecdysis-triggering hormone ( ETH ) triggers the process . However , it was not clear how each of the other neuropeptides that are released at this time contributes to the behaviors involved in ecdysis . By studying ecdysis in developing fruit flies , Mena et al . now show that the various ecdysial neuropeptides work together to produce the precise behaviors that are observed . For instance , the effect that ETH has on the nervous system depends on whether another neuropeptide called eclosion hormone is also present . ETH can therefore cause different behavioral outcomes depending on the actions of the other neuropeptides . Further work is needed in order to work out exactly how the nervous system integrates information from different neuropeptides . Do certain neurons respond to specific neuropeptide combinations ? It also remains to be seen how different insects are able to use the same neuropeptides to control ecdysis despite their different body shapes . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology",
"neuroscience"
] | 2016 | Stereotyped responses of Drosophila peptidergic neuronal ensemble depend on downstream neuromodulators |
Unesterified cholesterol accumulation in the late endosomal/lysosomal ( LE/LY ) compartment is the cellular hallmark of Niemann-Pick C ( NPC ) disease , caused by defects in the genes encoding NPC1 or NPC2 . We previously reported the dramatic stimulation of NPC2 cholesterol transport rates to and from model membranes by the LE/LY phospholipid lysobisphosphatidic acid ( LBPA ) . It had been previously shown that enrichment of NPC1-deficient cells with LBPA results in cholesterol clearance . Here we demonstrate that LBPA enrichment in human NPC2-deficient cells , either directly or via its biosynthetic precursor phosphtidylglycerol ( PG ) , is entirely ineffective , indicating an obligate functional interaction between NPC2 and LBPA in cholesterol trafficking . We further demonstrate that NPC2 interacts directly with LBPA and identify the NPC2 hydrophobic knob domain as the site of interaction . Together these studies reveal a heretofore unknown step of intracellular cholesterol trafficking which is critically dependent upon the interaction of LBPA with functional NPC2 protein .
Cholesterol is a small , hydrophobic molecule that is a vital building block of cell membranes and a precursor for steroid hormones , bile salts , vitamin D , and oxysterol ligands for transcription factors . Intracellular transport of cholesterol is a highly regulated but , as yet , incompletely understood process . Perturbations can lead to detrimental outcomes such as in the lysosomal storage disorder Niemann Pick Type C ( NPC ) disease , where LDL-derived cholesterol becomes trapped within the late endosomal/lysosomal ( LE/LY ) system . The sterol particularly enriches LE/LY inner membranes which develop during endosome maturation as a means of compartmentalizing its contents ( Gruenberg , 2001; Gruenberg , 2003; Matsuo et al . , 2004 ) . The accumulation of cholesterol in NPC disease is associated with amassing of other lipids in the LE/LY , disruption of post-lysosomal cholesterol metabolism , and ultimately clinical manifestations including organomegaly and neurological deterioration . In 95% of NPC cases , mutations in the large LE/LY transmembrane protein , NPC1 , prevent proper export of cholesterol from the LE/LY to other cellular compartments . The remaining 5% of cases are caused by mutations in the small , 132-amino acid , soluble LE/LY protein , NPC2 ( Peake and Vance , 2010; Vanier , 2010 ) . Similarities in the cellular and clinical phenotypes resulting from either NPC1 or NPC2 deficiency led to the suggestion that these two proteins function cooperatively in normal LE/LY cholesterol trafficking ( Kwon et al . , 2009; Sleat et al . , 2004 ) ; a proposed model shows cholesterol directly transferred from NPC2 , in the LE/LY lumen , to NPC1 , located in the limiting membrane of the compartment ( Estiu et al . , 2013; Wang et al . , 2010 ) . Recent tertiary structural analyses identifying a potential NPC2 interacting domain on the NPC1 protein support this mode of cholesterol egress from the LE/LY compartment ( Zhao et al . , 2016 ) . It is further proposed that cholesterol transfer from NPC2 to the luminally localized N-terminal domain of NPC1 allows sterol passage through the glycocalyx found at the luminal surface of the LE/LY limiting membrane , via concerted effort from membrane glycoproteins ( Li et al . , 2016 ) . NPC2 binds cholesterol with a 1:1 stoichiometry ( Xu et al . , 2007 ) . Ko et al . ( 2003 ) showed that cholesterol binding was necessary but not sufficient for its cholesterol efflux function , and we later demonstrated that cholesterol transport was the second critical component of NPC2 functionality ( Cheruku et al . , 2006; Xu et al . , 2008 ) . We further showed that cholesterol transfer by NPC2 occurs via protein-membrane interactions ( Cheruku et al . , 2006; Xu et al . , 2008 ) . Using model membranes , we demonstrated a remarkable , order of magnitude stimulation in wild type NPC2 cholesterol transfer rates by the incorporation of lysobisphosphatidic acid ( LBPA ) , also known as bis-monoacylglycerol phosphate ( Cheruku et al . , 2006; Xu et al . , 2008 ) . LBPA is a structural isomer of phosphatidylglyerol with an atypical phospholipid stereoconfiguration . It is localized primarily to inner LE/LY membranes and is thought to be involved not only in the formation of these internal membranes and their architecture , but also in the sorting and efflux of LE/LY components , including cholesterol ( Gruenberg , 2003; Hullin-Matsuda et al . , 2007; Kobayashi et al . , 1998; Kobayashi et al . , 1999; Möbius et al . , 2003 ) . Incubation of BHK cells and macrophages with an anti-LBPA monoclonal antibody results in cholesterol accumulation in the LE/LY compartment , resembling the NPC phenotype ( Delton-Vandenbroucke et al . , 2007; Kobayashi et al . , 1999 ) . Interestingly , Chevallier et al . ( 2008 ) showed that by enriching NPC1-deficient cells with exogenously added LBPA , the cholesterol accumulation was reversed ; the underlying molecular mechanism for LBPA stimulated LE/LY cholesterol egress remains unknown . Based on the dramatic effects of LBPA on NPC2-mediated cholesterol transfer , we hypothesized that the mechanism of LBPA action involves its specific functional interaction with NPC2 , such that LBPA enrichment of cells deficient in NPC2 would not reverse cholesterol accumulation , in contrast to what had been found in NPC1-deficient cells . In the present studies we demonstrate that , indeed , LBPA enrichment does not lead to the clearance of cholesterol in NPC2-deficient cells despite the presence of functional NPC1 . We show for the first time that the mechanism involves an obligate direct interaction of NPC2 with LBPA , identify the LBPA-sensitive domain on the NPC2 surface , and establish the essential functional nature of NPC2-LBPA interactions in cholesterol egress from the LE/LY compartment . The results identify a novel , heretofore unknown step in LE/LY cholesterol egress which is dependent upon LBPA interaction with the hydrophobic knob of NPC2 protein . This in turn suggests that LBPA enrichment may be used to effect cholesterol egress in cells with defective NPC1 but functional WT NPC2 , and NPC2 with disease-causing mutations outside the hydrophobic knob domain .
Our previous kinetics analyses strongly suggested that the mechanism of cholesterol transfer between NPC2 and membranes was via protein-membrane interaction ( Cheruku et al . , 2006; McCauliff et al . , 2015; Xu et al . , 2008 ) . NPC2 does not contain any apparent transmembrane domains , nor are there experimentally documented membrane interactive domains to date . For de novo predictions we therefore employed the Orientation of Proteins in Membranes ( OPM ) Database , a curated online resource that predicts the spatial positions of known protein structures relative to the hydrophobic core of a lipid bilayer ( Lomize et al . , 2012 ) . Using the crystal structure of bovine NPC2 ( PDB ID: 1NEP ) , a loop domain consisting of hydrophobic residues as highlighted in Figure 1 was predicted to be highly membrane interactive , with a ΔG of −4 . 6 kcal/mol . This domain corresponds to 56-HGIVMGIPV-64 and consists primarily of the hydrophobic residues I58 , V59 , M60 , I62 , P63 , and V64 , plus the non-polar residues G61 and G57 . Structurally , this domain forms a hydrophobic knob which presents prominently on the surface of the NPC2 protein . OPM predicts that this knob domain inserts into the hydrophobic space of the membrane model , positioning the sterol binding pocket of NPC2 near the membrane surface ( Figure 1A ) and , thus , in proximity to membrane sterols . These computational observations suggest that this knob domain may play a key role in the mechanism by which NPC2 is able to transport cholesterol between inner membranes of the LE/LY compartment . Of note , and as shown in Figure 1B , the primary sequence of the hydrophobic knob domain is conserved in mammalian NPC2 proteins but not in the yeast NPC2 homologue . Hydrophobicity scales for NPC2 as well as a Kyte-Doolittle plot of the hydrophobicity scores along the primary protein sequence indicate that the hydrophobicity of the knob domain is conserved amongst mammalian NPC2 proteins , in contrast to the low hydrophobicity in the yeast NPC2 protein ( Figure 1C ) . NPC2 is a soluble protein but we demonstrated using tryptophan quenching that it is membrane-interactive and that cholesterol transfer between NPC2 and membranes occurs via transient protein-membrane interactions ( Xu et al . , 2008 ) . We previously found that incorporation of 25 mol% LBPA in egg phosphatidylcholine ( EPC ) membranes resulted in cholesterol transfer rates from vesicles to NPC2 that were markedly accelerated relative to 100% EPC membranes ( Cheruku et al . , 2006; Xu et al . , 2008 ) LBPA accounts for approximately 15 mol% of total LE/LY phospholipids , with the likelihood of higher lateral concentrations in the highly heterogeneous inner LE/LY membranes ( Kobayashi et al . , 2002 ) . Thus , we examined the rates of cholesterol transfer from NPC2 to membranes as a function of increasing levels of LBPA . The results in Figure 2 indicate an exponential relationship between the LBPA content of the vesicles and the NPC2 cholesterol transfer rate; increasing the mol% of LBPA in SUV from 0% to 30% effectively increases the NPC2 cholesterol transfer rates by approximately 100 fold . The transfer of cholesterol between NPC2 and vesicles was also examined using 3H-cholesterol , where an NPC2-3H-cholesterol complex was added to vesicles followed by centrifugal separation . The reverse reaction was also examined , where membranes with 3H-cholesterol were added to apo NPC2 followed by centrifugation to separate protein and membranes . The results showed that the labeled cholesterol transfers from NPC2 to membranes and from membranes to NPC2 , and in agreement with previous experiments using tryptophan quenching to determine sterol distribution , the relative partition of cholesterol between NPC2 and phospholipid vesicles was >30:1 , mol NPC2:mol PL ( Xu et al . , 2008 ) . Since cholesterol transfer rates to and from NPC2 are very rapid , on the order of seconds or faster , this physical separation method serves only to provide an equilibrium distribution of the sterol between protein and membranes , but nevertheless indicates that the changes in NPC2 tryptophan fluorescence are reflecting sterol transfer between protein and membranes . Although there is order of magnitude more rapid sterol transfer from NPC2 to LBPA-containing vesicles relative to EPC vesicles , no appreciable differences in the equilibrium distribution of cholesterol between NPC2 and the different membranes were found . Using zwitterionic PC membranes we recently reported that point mutations in multiple regions on the NPC2 surface , including in the hydrophobic knob domain , led to diminished rates of cholesterol transfer between NPC2 and membranes ( McCauliff et al . , 2015 ) ; the impact of LBPA on cholesterol transfer by these NPC2 mutants was not investigated . Here we examined the rates of cholesterol transfer from these and additional point mutations in NPC2 protein to membranes containing 25 mol% LBPA . Mutations were confirmed by DNA sequencing and all mutant proteins were found to bind cholesterol similar to WT NPC2 ( Friedland et al . , 2003; Ko et al . , 2003 ) , with submicromolar affinity ( McCauliff et al . , 2015 ) . The results in Figure 3A show that when acceptor membranes included 25 mol% of LBPA , cholesterol transfer rates for NPC2 proteins with mutations in regions other than the hydrophobic knob were similar to those of WT . Surprisingly , though mutations at H31 , Q29 , D113 , and E108 exhibited sterol transfer rates to zwitterionic EPC membranes that were ≤15% of WT NPC2 , the inclusion of LBPA in acceptor membranes resulted in rates of cholesterol transfer that were ≥85% of WT rates . By contrast the I62 and V64 mutations , both in the hydrophobic knob and which also resulted in markedly defective cholesterol transfer to EPC vesicles , were unaffected by the inclusion of LBPA in the acceptor membranes , with cholesterol transfer by these mutants remaining barely detectable . The G61A mutation , also in the hydrophobic knob , resulted in cholesterol transfer deficiencies similar to the I62 and V64 mutants , though changes are less extreme; sterol transfer to EPC vesicles was reduced by 70% , and it remained highly defective in the presence of LBPA , with rates of cholesterol transfer of only 16% relative to WT NPC2 . Mutations in hydrophobic knob residues H56 , G57 , and I58 had little effect on cholesterol transfer rates to EPC vesicles , however unlike the WT NPC2 , these mutants were insensitive to the presence of LBPA in acceptor membranes . The NPC2 residues where mutations cause large decreases in cholesterol transfer rates to EPC are shown in red in Figure 3B; multiple surface regions are highlighted , in agreement with our aforementioned studies ( McCauliff et al . , 2015 ) . In striking contrast , Figure 3C shows that the mutations which remained insensitive to membrane LBPA ( shown in red ) were localized exclusively in the hydrophobic knob domain; all other surface mutations , including many that were markedly defective in sterol transfer to EPC membranes , were sensitive to LBPA inclusion and displayed normalized rates of sterol transfer ( shown in green ) . These results strongly indicate that the hydrophobic knob domain is the sole LBPA-sensitive region on the protein surface . We and others have shown that WT NPC2 promotes membrane-membrane interactions ( Abdul-Hammed et al . , 2010; Berzina et al . , 2018; McCauliff et al . , 2011; McCauliff et al . , 2015 ) . We further showed that NPC2 point mutants with deficient cholesterol transfer abilities are also unable to cause EPC membrane aggregation ( McCauliff et al . , 2015 ) . In the present studies , we investigated whether the presence of LBPA in the vesicles affected vesicle aggregation by WT and mutant NPC2 proteins . The results in Table 1 show that inclusion of 25 mol% LBPA in LUVs resulted in a 16-fold increase in the rate of membrane-membrane interaction by WT NPC2 , relative to 100% EPC LUVs . Incorporation of LBPA into membranes normalizes the membrane aggregation rates for NPC2 proteins with mutations outside the hydrophobic knob , for example H31 , D113 , and Q29 . By contrast , the hydrophobic knob domain mutations were relatively insensitive to membrane LBPA . The results for this membrane-membrane interaction assay map virtually identically onto the NPC2 structure as did those for the cholesterol transport rates , as seen in Figure 3C . To determine whether the relationship between LBPA and NPC2 in cholesterol trafficking involves direct interactions , protein-lipid binding assays were conducted . For studies of WT NPC2 protein , custom LBPA Snoopers ( Avanti Polar Lipids ) containing various LBPA isomers were incubated at pH 7 . 4 with WT NPC2 and relative binding was assessed via densitometric analysis of an antibody-probed strip , as described in Methods . The results shown in Figure 4A demonstrate that WT NPC2 binds to LBPA , showing greater interaction with isomers containing oleoyl ( C18:1 ) as opposed to myristoyl ( C14:0 ) fatty acyl chains , and overall the greatest degree of binding to the S , S 18:1 LBPA . NPC2 binding to S , S 18:1 LBPA was also greater than binding to egg PC . Binding of WT NPC2 to other typical membrane phospholipid species , in comparison to di-oleoyl LBPA , was also examined at pH 7 . 5 using membranes spotted with di-oleoyl phospholipids . Figure 4B shows that NPC2 interacts more strongly with LBPA than with PC , PA , PG , and PS species . As two other anionic phospholipids assayed , PG and PA , show even less interaction with NPC2 than does zwitterionic PC , the mechanism by which NPC2 binds to LBPA is likely not solely dependent on electrostatic interactions . In protein-lipid overlay assays the phospholipids are not necessarily present in a physiological orientation , therefore we further examined NPC2–lipid interaction using Homogenous Time Resolved Fluorescence ( HTRF ) , in which the phospholipids are present as lamellar structures . HTRF , which has been recently demonstrated to be an effective and sensitive assay for lipid-protein interaction ( Fleury et al . , 2015 ) , is a fluorescence resonance energy transfer ( FRET ) based technology that utilizes specific fluorophores that emit long-lived fluorescence signals when involved in a FRET process . Notably the europium cryptate energy donor has been shown to be impervious to photobleaching and to exhibit significant stability in a homogenous assay environment ( Degorce et al . , 2009 ) . This allows for time-resolved measurements of , in this case , protein-lipid interactions , with the ability to easily subtract background signals . The NPC2-interaction with LBPA at pH 5 . 0 was substantially greater than with other negatively charged lipids or with zwitterionic PC ( Figure 4D ) , in general agreement with the lipid-blot results . Taken together the results support direct interactions between NPC2 and membrane phospholipids , the greatest interaction being with LBPA . To determine whether there is a specific LBPA interactive domain on the surface of NPC2 , we further employed the LBPA lipid blots to examine interactions with NPC2 point mutants . The results show that mutations in most of the hydrophobic knob residues markedly reduce interactions , relative to WT NPC2 . For example , studies analyzing binding to various isomers of LBPA show that the hydrophobic knob mutants I62N and V64A bound the S , S and S , R di-oleoyl isomers at only ~20% of WT levels , and the C18:1 R , R isomer at approximately 40% of WT . In contrast , the Q29A and D113A mutants , in regions outside the hydrophobic knob , bound nearly all LBPA isomers similar to WT; only Q29A was observed to bind the C18:1 R , R isomer at approximately 50% relative to WT binding ( Figure 5A ) . The I62D and V64A mutants exhibited greater interaction with the 18:1 Semi LBPA species , which has three oleoyl acyl chains ( Figure 5A ) , binding approximately 70% relative WT NPC2 . Mutations in other hydrophobic knob residues also reduced the interaction of NPC2 with the di-oleyol S , S LBPA on blots using various phospholipids . Indeed , similar to what was observed with the LBPA isomer blots , the H56 , G57 , I58 , and G61 mutants showed only 30% to 40% degree of interaction relative WT NPC2 . In marked contrast , surface mutations outside the hydrophobic knob had virtually no impact on NPC2 interaction with LBPA , with Q2A , H31A , D113 , and E108 mutants exhibiting WT levels of interaction with LBPA ( Figure 5B ) . Overall , mutations within the hydrophobic knob domain of NPC2 resulted in diminished binding of the protein to LBPA while mutations outside the knob region presented proteins with LBPA interactions similar to WT ( Figure 5B ) . HTRF analysis of the mutant NPC2 proteins was generally consistent with the lipid blot results , also indicating reduced binding of the hydrophobic knob mutants to LBPA ( Figure 5C and D ) . In agreement with several previous reports , incubation of NPC2 deficient fibroblasts with WT NPC2 protein resulted in a dramatic decrease in filipin staining , reaching levels similar to healthy fibroblasts ( Figure 6 ) ( Ko et al . , 2003; Liou et al . , 2006; McCauliff et al . , 2011; McCauliff et al . , 2015 ) . The H56A mutant , with rates of sterol transfer and membrane aggregation similar to WT protein , also reduced filipin stain area , similar to WT NPC2 . In contrast , the G57D and I58A hydrophobic knob mutants , with markedly attenuated cholesterol transfer and membrane aggregation rates , were unable to reverse cholesterol accumulation in NPC2 cells; filipin staining remained at a level comparable to that of the unsupplemented cells . G61A , also in the knob domain , was able to lessen cholesterol accumulation in NPC2 cells to a moderate extent , and its defect in cholesterol transfer was also more modest than that of other hydrophobic knob mutants ( Figure 6 ) . These results are in keeping with our previously reported results for two other hydrophobic knob mutants , I62D and V64A ( McCauliff et al . , 2015 ) which are also deficient in cholesterol transfer and membrane aggregation ability . The consistency between results of the sterol transfer assays and membrane-membrane interaction assays , with the cholesterol clearance in patient cells , strongly supports the physiological relevance of the structure-function studies , and points to a particularly important role for the hydrophobic knob of NPC2 in effecting normal sterol trafficking . To increase cell LBPA levels , NPC patient fibroblasts were incubated with PG , known to be its precursor ( Bouvier et al . , 2009; Poorthuis and Hostetler , 1978; Thornburg et al . , 1991 ) . The results in Figure 7 show that incubation of the cells with 100% PG SUVs led to substantial increases in cellular content of LBPA in all fibroblast types; in WT cells the increase was nearly 6-fold . In agreement with previous reports LBPA levels in NPC patient cells were found to be increased relative to WT cells prior to enrichment ( Chevallier et al . , 2008; Davidson et al . , 2009; Sleat et al . , 2004; Vanier , 1983 ) ; PG incubation resulted in 2–3 fold increases in NPC1- and NPC2-deficient cells , respectively , relative to unsupplemented cells ( Figure 7A ) . Levels of other cellular phospholipids appeared unchanged ( Figure 7B ) . Direct addition of LBPA-containing SUVs also led to approximately 2 to 3-fold increases in cellular LBPA levels ( data not shown ) , in agreement with ( Chevallier et al . , 2008 ) . Supplementation with PC SUVs as a control had no effect on the phospholipid composition of any of the cell types ( data not shown ) . Following PG supplementation , cholesterol content remained unchanged in the WT fibroblasts ( Figure 8 ) . NPC1 deficient fibroblasts exhibited a dramatic reduction in cholesterol accumulation following PG supplementation , approaching levels observed for WT cells and similar to the direct addition of LBPA ( Chevallier et al . , 2008 ) . In marked contrast to NPC1 deficient cells , the cholesterol accumulation in NPC2 deficient fibroblast remained elevated following PG supplementation , despite increased LBPA content ( Figure 8 ) . While the PG/LBPA-dependent cholesterol egress observed in NPC1 deficient cells appears to indicate an NPC1-independent mechanism for cholesterol egress , the NPC1 cell line used in these studies is a compound heterozygote ( C709T and T3182C ) which , despite its well established deficiency in cholesterol egress , is still expressed in the limiting LE/LY membrane and thus could potentially be involved in the egress of cholesterol from cells supplemented with PG . To address this possibility , we generated NPC1 null HeLa cells via CRISPR-Cas9 and found that PG supplementation led to significant reductions in the cholesterol accumulation phenotype ( Figure 8C ) . We additionally performed PG supplementation in WT cells treated with the U18666A compound , known to induce the NPC disease phenotype at the cellular level by targeting NPC1 ( Lu et al . , 2015 ) . In these cells too , PG supplementation resulted in cholesterol clearance ( Figure 8D ) . Overall , these results suggest that LBPA can promote cholesterol egress from the endolysosomal compartment despite the absence of NPC1 , but not in the absence of NPC2 . The point mutagenesis analysis of NPC2 shows a direct relationship between the cholesterol transfer rate of a particular NPC2 mutant and its ability to rescue the cholesterol accumulation of NPC2-deficient cells ( McCauliff et al . , 2015; Figures 3 and 6 ) . In the present studies we discovered an unanticipated impact of LBPA incorporation into membranes , in which some NPC2 mutants that were highly defective in sterol transfer to phosphatidylcholine membranes , were essentially normalized when LBPA was present . These ‘LBPA-sensitive’ NPC2 mutations were , almost exclusively , present in surface domains outside of the hydrophobic knob . NPC2 proteins with mutations in the hydrophobic knob , that were highly defective in cholesterol transfer to phosphatidylcholine membranes , remained insensitive to LBPA in the membranes ( Figure 3 ) . Based on this structure-based difference in LBPA sensitivity , we hypothesized that increasing the LBPA levels in NPC2 deficient cells would enhance the activity of mutants outside the hydrobphobic knob that responded to LBPA in the kinetic assays , whereas cellular LBPA enrichment would not enhance the action of the hydrophobic knob mutants that were insensitive to LBPA . To test these predictions , NPC2 deficient fibroblasts were incubated with purified wild type or mutant NPC2 proteins , with or without PG supplementation , and cholesterol accumulation was assessed by filipin staining as described . The results in Figure 9 demonstrate that NPC2 proteins with mutations outside the hydrophobic knob , such as Q29A , D113A , and D72A , which were unable to reduce cholesterol accumulation in unsupplemented NPC2 deficient cells , were indeed able to ‘rescue’ cells that were enriched with LBPA via PG supplementation . By contrast , the hydrophobic knob mutants I62D and G61A , which were insensitive to LBPA in cholesterol transfer assays , were unable to clear cholesterol from LBPA-enriched NPC2 deficient cells . Thus , the results show that a combination of increased cellular levels of LBPA and LBPA-sensitive NPC2 protein was able to rescue the NPC2 deficient cells , beyond the ability of the mutant protein alone ( Figure 9 ) . As before , WT cells supplemented with PG alone showed no change in cholesterol accumulation , and cells supplemented with purified WT NPC2 , with or without PG , showed significantly reduced filipin staining . The results in patient cells again mirror the results of sterol kinetics experiments , suggesting that return to normal sterol trafficking can be achieved for NPC2 mutations outside the hydrophobic knob that are otherwise dysfunctional , when the LBPA content of the cells is increased .
LBPA was proposed to be involved in intracellular cholesterol trafficking based on the sterol accumulation which accompanies incubation of cells with an anti-LBPA antibody ( Kobayashi et al . , 1999 ) , however the mechanism of LBPA action has remained unknown . In this study we have , for the first time , shown that LBPA function in cholesterol trafficking is obligately dependent upon its interaction with the NPC2 protein within the endo-lysosomal system . At the molecular level , we show that this novel step in intracellular cholesterol trafficking involves direct interaction of LBPA with the hydrophobic knob domain on NPC2 . This surface domain is located near the cholesterol binding pocket , thus its insertion into the bilayer would position the protein to efficiently exchange cholesterol with the membrane . Recent molecular dynamic simulations show that LBPA , but not other phospholipids , may position NPC2 in an orientation that could promote protein-membrane sterol exchange ( Enkavi et al . , 2017 ) , potentially inducing a conformational change in the binding pocket of NPC2 such that the off-rate of cholesterol is markedly increased , allowing for rapid transfer of cholesterol between protein and LBPA-enriched membranes . A decade ago , the Gruenberg laboratory reported that viral-mediated supplementation of NPC1-deficient cells with exogenous LBPA reversed cholesterol accumulation in the diseased cells ( Chevallier et al . , 2008 ) . Here we show that LBPA enrichment is , conversely , completely ineffective in cells expressing NPC1 but lacking NPC2 , underscoring the required functional interaction of LBPA with the NPC2 protein . In our efforts to establish the molecular basis of this interaction , we found that several NPC2 mutants that could not reverse cholesterol accumulation in unsupplemented NPC2-deficient cells , became effective if the cells were first enriched with LBPA . The NPC2 mutations which were ‘rescued’ by LBPA enrichment are all localized outside of the hydrophobic knob domain . By contrast , mutations within the hydrophobic knob were insensitive to LBPA enrichment apart from H56A , which may be on the limiting edge of this region . Thus , LBPA is able to normalize deficient transfer by mutant NPC2 as long as the mutation is outside of the hydrophobic knob . Interestingly , direct interaction between NPC2 and NPC1 , residing in the limiting LE/LY membrane , has also been identified as involving the hydrophobic knob region on NPC2 ( Li et al . , 2016; Wang et al . , 2010 ) . This highlights further the critical nature of this domain in intracellular cholesterol trafficking , such that the NPC2 hydrophobic knob is involved in both LBPA-dependent binding of cholesterol by NPC2 at inner-LE/LY membranes , and the delivery of cholesterol by NPC2 to the NPC1 N-terminal domain . Primary amino acid sequence comparisons reveal a high degree of conservation within the NPC2 hydrophobic knob domain for all species with the exception of yeast; interestingly , yeast lacks the LBPA phospholipid , indirectly supporting a functional interaction between the mammalian NPC2 hydrophobic knob and LBPA . Thus , we propose that the hydrophobic knob on NPC2 interacts with LBPA in the inner LE/LY membranes , leading to cholesterol extraction , with NPC2 then subsequently involved in the ‘handoff’ of the sterol to NPC1 in the limiting membrane of these organelles ( Deffieu and Pfeffer , 2011; Infante et al . , 2008; Li et al . , 2016; Wang et al . , 2010 ) . In healthy cells , the enrichment of inner LE/LY membranes with LBPA and concurrent decline in cholesterol content may be reflective of the efficient role LBPA plays in normal LE/LY cholesterol efflux . If direct interactions between LBPA and NPC2 are integral to this mechanism of efflux , a rate determining step may be their frequency of interaction and , thus , be partially dependent upon the concentration of both phospholipid and protein in the LE/LY compartment . While the concentration of LE/LY LBPA has been shown to increase in parallel with cholesterol and other lipids in NPC disease ( Davidson et al . , 2009; Sleat et al . , 2004; Vanier , 1983 ) , as also found here , it is possible that LBPA levels nevertheless remain too low , as previously suggested ( Chevallier et al . , 2008 ) , to provide support for NPC2-mediated transport of the elevated cholesterol load . Our results using PG supplementation to augment LBPA levels further indicate that LBPA may become limiting in NPC1 disease . Importantly , we show here that LBPA enrichment cannot reverse cholesterol accumulation caused by NPC2 deficiency unless the mutation is outside the hydrophobic knob domain in which case PG supplementation is effective , as it is in NPC1-deficient cells . Interestingly , Chen et al . observed a pronounced association of the NPC2 protein with inner LE/LY membranes in cells of Npc1-/- mice ( Balb/c Npc1nih ) , relative to WT ( Chen et al . , 2005 ) , suggesting that increased recruitment of NPC2 to inner LE/LY membranes in NPC1 disease , where LBPA concentration is high , may be a compensatory mechanism to alleviate sterol accumulation . Beyond its obligate interaction with NPC2 within the LE/LY , it is not yet clear how increased LBPA leads to the efflux of LE/LY cholesterol in the absence of NPC1 protein , although in recent studies we have found that increased concentrations of cellular LBPA increase macroautophagy in NPC1 deficient cells ( data not shown ) . It is well known that impaired autophagy is present in NPC disease ( Liao et al . , 2007; Lieberman et al . , 2012; Seranova et al . , 2017 ) , and stabilization of this pathway has been shown to reduce cholesterol accumulation in NPC1 disease ( Dai et al . , 2017; Ordonez et al . , 2012; Sarkar et al . , 2013 ) . LBPA also displays a variety of other unique functions within the endo/lysosomal system that could potentially promote cholesterol efflux . Studies have demonstrated , for instance , that its transient interaction with the ESCRT-associated protein , ALIX , is required for the proper formation of multivesicular structures within the LE/LY ( Matsuo et al . , 2004 ) , where cholesterol is localized ( Fivaz , 2002 ) ; it is also found , along with cholesterol , on exosomes derived from fusion of inner lamellae with the limiting LE/LY membrane ( Matsuo et al . , 2004 ) . Furthermore , it has recently been proposed that LBPA may influence cholesterol homeostasis beyond the confines of late endosomes and lysosomes , having been shown to be necessary for lipid droplet formation via Wnt signaling within the endoplasmic reticulum , where cholesteryl esters are synthesized ( Scott et al . , 2015 ) . Within the endo/lysosomal system , however , our results demonstrating that cellular enrichment of LBPA cannot reverse cholesterol accumulation in cells lacking functional NPC2 protein indicate that the cholesterol efflux mechanism utilized by LBPA is dependent upon its interaction with NPC2 . LBPA is reported to comprise only 1% of total cellular phospholipids , but about 15 mol% of total phospholipids in the LE/LY ( Chevallier et al . , 2000; Kobayashi et al . , 1998; Kobayashi et al . , 2002 ) . There is no agreement regarding which isomeric form of LBPA is predominant in cells , as the fatty acids can be linked at the sn-2 or sn-3 positions of each glycerol in the S or R conformation , though the sn-2 , sn-2’ positions are currently favored ( Chevallier et al . , 2000; Kobayashi et al . , 1998; Kobayashi et al . , 2002; Mason et al . , 1972; Matsuo et al . , 2004 ) . The 2 , 2’-LBPA was shown to be quite effective at mobilizing cholesterol in NPC1 disease cells while the 3 , 3’- and semi-LBPA isoforms were unable to promote sterol efflux . No difference in efficacy between S , S , S , R , and R , R LBPA isomers were noted , however , suggesting that the conformation of LBPA has no bearing on its ability to reverse cholesterol accumulation in NPC1 deficient cells ( Chevallier et al . , 2008 ) . In agreement with this observation , we previously showed that the S , S , S , R , and R , R configurations of LBPA had little to no effect on in vitro cholesterol transfer rates by NPC2 protein ( Xu et al . , 2008 ) . Similarly , in the present studies we observed little variation in NPC2 binding to these different LBPA stereoisomers . Regardless , we chose to use the presumed precursor and structural isomer of LBPA , phosphatidylglycerol , to increase the cellular LBPA content in these studies , obviating any potential concern about the LBPA isoform as the cells presumably generate the physiologically accurate form . PG is thought to convert to LBPA along the endo/lysosomal system ( Hullin-Matsuda et al . , 2007; Poorthuis and Hostetler , 1978 ) , and studies have demonstrated that exogenously administered PG can be converted to LBPA in vivo ( Somerharju and Renkonen , 1980 ) , although the anabolic pathway of this conversion remains unknown . The present demonstration that PG supplementation specifically increases the LBPA content of all cells tested supports the hypothesis that PG is a precursor to LBPA . Cellular conversion of PG to LBPA has also been demonstrated in mammalian alveolar macrophages ( Waite et al . , 1987 ) , lymphoblasts ( Hullin-Matsuda et al . , 2007 ) and RAW macrophages ( Bouvier et al . , 2009 ) . In addition to variations in stereochemistry , the cellular fatty acyl chain components of LBPA have been found to vary ( Bouvier et al . , 2009 ) , with oleic acid and docosahexaenoic acid ( DHA ) reported to be selectively incorporated ( Besson et al . , 2006; Luquain et al . , 2001 ) . Here we found acyl chain-dependent differences in NPC2-LBPA interactions , with reduced binding to the 14-carbon saturated dimyristoyl-LBPA species relative to the 18-carbon monounsaturated dioleoyl-LBPA species . In prior work we showed that cholesterol transfer from NPC2 to membranes containing dioleoyl-LBPA was >2 fold faster than transfer to vesicles with dimyristoyl-LBPA ( Xu et al . , 2008 ) . Taken together , the results suggest that the acyl composition but not the steroconfiguration of LBPA may be important in normal LE/LY cholesterol efflux . Ongoing studies focused on identifying the specific acyl-chain species that are most effective at stimulating LE/LY cholesterol clearance in an NPC phenotype , for instance , should inform both our understanding of intracellular cholesterol transport and the development of therapies for lysosomal storage disorders . Based on the present results , we propose that the NPC2 hydrophobic knob domain inserts into LBPA enriched inner LE/LY membranes , interacting directly with the phospholipid . Given the ability of NPC2 to promote membrane-membrane interactions ( Abdul-Hammed et al . , 2010; McCauliff et al . , 2011; McCauliff et al . , 2015 ) , we speculate that this interaction between LBPA and NPC2 is involved in the formation of membrane contact sites which could potentially exist between closely apposed inner LE/LY membranes and facilitate rapid transfer of sterol . Membrane contact sites have been shown to be important in intermembrane lipid transfer ( Helle et al . , 2013; Holthuis and Levine , 2005; Prinz , 2014 ) , although none have yet been specifically described within the multivesicular LE/LY . Our results show that NPC2 promotes membrane-membrane interaction , and further indicate the ability of NPC2 to bind to LBPA; this interaction may represent one membrane contact point on inner LE/LY membranes . Our previous kinetic studies suggested that NPC2 interacts with other membrane phospholipids as well , and preliminary molecular dynamics simulations indicate that unlike LBPA , where NPC2 interacts at the hydrophobic knob , these interactions occur at NPC2 surface sites other than the knob domain ( data not shown ) ; such interactions could represent a second contact point within the inner LE/LY membranes . Interestingly , it has now been demonstrated that NPC2 also binds directly to NPC1 which resides in the limiting LE/LY membrane , and possibly also with LAMP proteins in these same membranes ( Li et al . , 2016 ) . These could also be potential tether points for the NPC2 and would effectively bring inner LE/LY cholesterol-laden membranes in closer proximity to the limiting LE/LY membrane , which cholesterol must ultimately cross to exit the compartment . The key finding of the present studies is that the primary mechanism by which LBPA stimulates LE/LY cholesterol efflux is critically dependent upon its interaction with functional NPC2 . LBPA first appears on the internal vesicles and membranes of cholesterol rich multivesicular bodies characteristic of late endosomes ( Gruenberg , 2003; Möbius et al . , 2003 ) and increases in concentration through the pathway to cholesterol depleted lysosomes ( Möbius et al . , 2003 ) . This localization overlaps with NPC2 , targeted to late endosomes/lysosomes via the mannose-6-phosphate receptor ( Naureckiene et al . , 2000 ) . While NPC1 is also targeted to late endosomes ( Garver et al . , 2000; Higgins et al . , 1999; Neufeld et al . , 1999 ) , Blanchette-Mackie and colleagues observed that most NPC1 localizes to a specific set of LAMP2 positive , mannose 6-phosphate receptor negative vesicles that are distinct from cholesterol enriched LAMP2 positive lysosomes , where LBPA and NPC2 reside , suggesting that transient interactions between NPC1 positive organelles and cholesterol rich lysosomes occur to effect normal cholesterol egress via NPC2-NPC1 interaction ( Neufeld et al . , 1999 ) . While several groups have reported NPC1-independent egress of cholesterol from the LE/LY ( Boadu et al . , 2012; Goldman and Krise , 2010; Kennedy et al . , 2012 ) , it is not clear whether these function in the normal situation , or whether they are manifested secondary to NPC1 dysfunction . Under normal conditions it is likely that the critical functional interaction between NPC2 and LBPA , demonstrated in the present studies , is followed by an interaction between NPC2 and NPC1 at the limiting lysosomal membrane , allowing for cholesterol to egress from the endolysosomal system . However , our studies have also implied that in the presence of intact NPC2 , bypassing dysfunctional NPC1 may be achieved via PG-mediated or direct LBPA enrichment . LBPA enrichment may also be effective in NPC2 cases where the mutation is in a residue outside of the hydrophobic knob . For example , a human mutation in D72 has been reported to be disease causing ( Biesecker et al . , 2009 ) , and we found here that PG supplementation/LBPA enrichment allowed NPC2 deficient cells to be effectively cleared by the D72A protein; supplementation with D72A-NPC2 was entirely ineffective prior to LBPA enrichment . In a recent high-throughput screen for drugs that raise LBPA levels , ( Moreau et al . , 2019 ) showed that a compound which increased LBPA levels cleared cholesterol from an NPC2 patient cell line; the mutated residue in this instance , C93 , lies outside the hydrophobic knob . Thus , the in vitro evidence strongly suggests that LBPA enrichment can affectively ameliorate cellular cholesterol accumulation in the majority of NPC disease cases , including potentially all NPC1 mutations as well as NPC2 mutations outside the hydrophobic knob . Currently there is no cure for NPC disease . Pharmacological options are limited , and palliative care remains the standard for treatment of the disease , focusing on increasing the length and quality of life for affected patients . Based on the existing in vitro evidence , we propose that cellular LBPA enrichment is worth exploring as a possible therapy . Aerosolized phospholipids such as dipalmitoyl phosphatidylcholine are widely used to enhance pulmonary drug delivery ( Duret et al . , 2014 ) , and can be adequately nebulized without losing compositional integrity ( Schreier et al . , 1994 ) . PG itself , administered intranasally , has been used as a therapy by the Voelker group to effectively inhibit respiratory syncytial virus infection ( Numata et al . , 2010; Numata et al . , 2013 ) and influenza A virus ( Numata et al . , 2012 ) . Moreover , lipid based colloidal carriers are able to cross the blood brain barrier ( BBB ) when administered intranasally ( Ganesan et al . , 2018; Mittal et al . , 2014; Patel and Patel , 2017; Tapeinos et al . , 2017 ) and have recently been shown to be efficient drug delivery vehicles in the treatment of intrinsic brain tumors ( van Woensel et al . , 2013 ) and neurodegenerative diseases including Alzheimer’s ( Agrawal et al . , 2018; Tapeinos et al . , 2017 ) and Parkinson’s ( Tapeinos et al . , 2017; Yang et al . , 2016 ) . Given the historical difficulties in treating the neurological effects of NPC disease , the development of a minimally invasive , effective treatment with intrinsic ability to cross the BBB is of interest .
The Orientation of Proteins in Membranes ( OPM ) database ( http://opm . phar . umich . edu/ ) was used to predict spatial orientation of NPC2 protein with respect to the hydrophobic core of lipid bilayers . Protein structure of NPC2 ( PDB: 1NEP ) was searched against the OPM database and the resulting coordinate file and orientation predictions were obtained . Protein sequences for human NPC2 ( NCBI Accession: NP_006423 . 1 ) , rat NPC2 ( NCBI Accession: NP_775141 . 2 ) mouse NPC2 ( NCBI Accession: NP_075898 . 1 ) , bovine NPC2 ( NCBI Accession: NP_776343 . 1 ) , cat NPC2 ( NCBI Accession: XP_003987882 . 1 ) , chimpanzee NPC2 ( NCBI Accession: NP_001009075 . 1 ) and the yeast NPC2 ( NCBI Accession: KZV12184 . 1 ) were aligned with CLUSTAL Omega ( Sievers et al . , 2011 ) . Protein conservation was scored using a PAM250 scoring matrix , which is extrapolated from comparisons of closely related proteins , similar to the current application ( Pearson , 2013 ) . Domain specific conservation of the hydrophobic knob between each NPC2 sequence was analyzed by taking the sum of the conservation scores of each residue from 56 to 64 , relative to the human NPC2 sequence . Protein hydrophobicity was scored using the Kyte and Doolittle Amino acid Hydropathicity scale ( Kyte and Doolittle , 1982 ) . Domain specific hydrophobicity of the hydrophobic knob of each NPC2 sequence was analyzed by taking the sum of the Kyte and Doolittle Amino acid Hydropathicity score of each residue from 56 to 64 . Whole protein hydrophobicity of aligned sequences were analyzed using the ProtScale Tool on the ExPASy server ( Gasteiger et al . , 2005 ) , based on the Kyte and Doolittle Amino acid Hydropathicity scale ( Kyte and Doolittle , 1982 ) . Chinese hamster ovary ( CHO ) cells transfected with a human NPC2 expression vector ( NPC2-800#7 ) ( Liou et al . , 2006 ) , kindly provided by Peter Lobel , were maintained in F12-K media ( Invitrogen ) supplemented with 10% FBS and 1 mg/mL gentamycin . Human WT ( GM03652 ) , NPC1 ( GM03123 ) , and NPC2 ( GM18455 ) fibroblasts ( Coriell Institute , Camden , NJ ) and HeLa ( ATTC CCL-2 ) ( ATCC , Manassas , VA ) cells were maintained in DMEM media ( Invitrogen ) supplemented with 15% FBS and 1% penicillin-streptomycin . All cells were at passage 18 or below . Authentication of all fibroblasts and HeLa CCL-2 cells was obtained via STR analysis . All cultures were confirmed mycoplasma free at receipt and were cultured aseptically using only mycoplasma free reagents . NPC1 CRISPR-Cas9 KO construct was purchased from Santa Cruz Biotechnology ( Dallas , TX ) and transfected into HeLa CCL-2 cells using the Lipofectamine 3000 reagent ( Invitrogen ) . Cells were selected with puromycin at 2 μg/ml for 4 days . Single cell clones were isolated and subsequently screened for loss of NPC1 expression using Western Blot analysis with an anti-NPC1 antibody ( Abcam , Cambridge , MA ) . Point mutations were created with the Stratagene QuikChange Site Directed Mutagenesis Kit ( Agilent , Santa Clara , CA ) , using myc 6xHis-tagged murine NPC2 plasmid , according to the manufacturer’s directions and as described previously ( Ko et al . , 2003; McCauliff et al . , 2015 ) . Plasmid isolation was performed using the PureYield Plasmid Miniprep system ( Promega , Madison , WI ) . Wild type and mutant myc 6xHis-tagged NPC2 proteins were purified from transfected NPC2-800#7 CHO cells , which secrete large amounts of the NPC2 protein into CD CHO media ( Invitrogen ) , using a 10 kDa cutoff flow filtration membrane ( Millipore , Bedford , MA ) to initially concentrate the media , as previously described ( McCauliff et al . , 2015 ) . The presence of purified NPC2 was confirmed by SDS-PAGE ( Cheruku et al . , 2006; Ko et al . , 2003; McCauliff et al . , 2015 ) ; proteins used were >90% pure by silver staining . Buffer exchange was performed using Sartorius Vivaspin Turbo four filters with a 10 kDa cutoff membrane followed by dilution in sodium citrate buffer ( in 20 mM sodium citrate , 150 mM NaCl , pH 5 . 0 ) . Equilibrium binding constants for cholesterol binding by WT and mutant NPC2 proteins were determined by quenching of tryptophan emission , as previously described ( Cheruku et al . , 2006; Ko et al . , 2003; McCauliff et al . , 2015; Xu et al . , 2008 ) . Purified proteins were delipidated via acetone precipitation ( Liou et al . , 2006 ) and resuspended in sodium citrate buffer . The delipidated NPC2 proteins were incubated with increasing concentrations of cholesterol ( >99% ) ( Sigma Aldrich ) , in DMSO , for 20 min at 25°C and tryptophan emission spectrum were acquired on an SLM fluorimeter ( Horiba Jobin Yvon , Edison , NJ ) . Final DMSO concentration was >1% ( v/v ) . AUC were determined for all spectrum and binding constants were determined by hyperbolic fit of the data using Sigma Plot software ( San Jose , CA ) . Small unilamellar vesicles ( SUV ) were prepared by sonication and ultracentrifugation as previously described ( Storch and Kleinfeld , 1986 ) . Large unilamellar vesicles ( LUV ) were prepared via freeze-thaw cycling and extrusion through a 100 nm membrane , as previously described ( McCauliff et al . , 2015; Xu et al . , 2008 ) . The final phospholipid concentration of all vesicles was determined by quantification of inorganic phosphate ( Gomori , 1942 ) . Vesicles were maintained above the phase transition temperatures of all constituent lipids . Standard vesicles were composed of 100 mol% egg phosphatidyl choline ( EPC ) ( Avanti Polar Lipids , Alabaster , AL ) . Where noted , LBPA ( Avanti ) replaced 5–25 mol% of EPC in SUV and/or LUV preparations , as indicated . All vesicles used for in vitro transfer assays were prepared in sodium citrate buffer , pH 5 . 0 . For incubation with cells , vesicles were composed of 100 mol% PG ( Avanti ) , 25 mol% LBPA or 100 mol% PC ( Avanti ) and prepared in sterile phosphate-buffered saline , pH 7 . 4 . As detailed previously ( McCauliff et al . , 2011; McCauliff et al . , 2015; Xu et al . , 2008 ) , transfer of cholesterol from WT or mutant NPC2 protein to membranes was monitored by the dequenching of tryptophan fluorescence over time using a stopped-flow mixing chamber interfaced with a Spectrofluormeter SX20 ( Applied Photophysics , Leatherhead , UK ) . Cholesterol transfer rates from 1 µM WT NPC2 to 125 µM EPC membranes containing increasing mol percentages of LBPA ( 0% , 10% , 20% and 30% ) were determined at 25°C . Additionally , to determine the effects of LBPA on the sterol transport properties of mutant NPC2 proteins , transfer of cholesterol from 1 µM WT or mutant NPC2 to 125 µM SUVs composed of either 100% EPC or 25 mol% LBPA/EPC was monitored at 25°C . Instrument settings to ensure the absence of photobleaching were established before each experiment . Data were analyzed with the Pro-Data SX software provided with the Applied Photophysics stopped-flow spectrofluorometer , and the cholesterol transfer rates were obtained by single exponential fitting of the curves , as previously described ( McCauliff et al . , 2015; Xu et al . , 2008 ) . Large unilamellar vesicles ( LUVs ) were prepared by extrusion as described previously ( Wootan and Storch , 1994 ) . Vesicles were composed of 100% EPC or 75% EPC/25% LBPA ( mol/mol ) , with a trace amount of 14C-cholesterol ester as a nonexchangeable marker of the LUVs . 3H-cholesterol was incubated with NPC2 for >15 min , and the complex mixed with LUVs . The LUV were pelleted after 30 s or 5 min by ultracentrifugation at 100 , 000 x g for 45 min . For the reverse reaction , the LUVs contained 3H-cholesterol and were mixed with apo NPC2 . Following correction for unpelleted LUVs and for NPC2 in the pellet , determined by tryptophan emission , the relative distribution of 3H-cholesterol between NPC2 and phospholipid membranes was determined ( Storch and Bass , 1990; Xu et al . , 2008 ) . Effects of NPC2 on vesicle-vesicle interactions were assessed in two ways , both using light scattering approaches . 200 µM LUVs were mixed with 1 µM WT or mutant NPC2 proteins in a 96-well plate reader , and absorbance at 350 nm monitored every 10 s over a period of 30 min ( Petruševska et al . , 2013 ) . Increases in A350nm ( light scattering ) are indicative of vesicle-vesicle interaction , the rate of which was determined by a three-parameter hyperbolic fit of the data using Sigma Plot software . Additionally , 750 µM LUVs were mixed with 1 µM WT or mutant NPC2 proteins in a spectrophotometer ( Hitachi U-2900 , Pleasanton , CA ) and A350nm was measured over a period of 60 s ( Schulz et al . , 2009 ) ; rates of vesicle-vesicle interaction were obtained by single exponential fitting of the curves . To assess WT NPC2 binding to various LBPA isomers , LBPA Snoopers ( Avanti ) , containing 1 μg spots of pure LBPA isomers , were blocked with tris-buffered saline ( TBS ) ( 0 . 8% NaCl , 20 mM Tris-HCl pH 7 . 4 ) + 3% BSA ( fatty-acid free ) , followed by a one hour incubation at room temperature with 5 μg of WT NPC2 in TBS pH 7 . 4 + 3% BSA , at a final concentration of 0 . 5 μg/ml protein . The protein solution was removed and the Snoopers were washed with TBS . NPC2 bound to LBPA isomers was detected by incubating the Snoopers with rabbit polyclonal anti-c-myc-tag antibody ( GenScript , Piscataway , NJ ) at a concentration of 0 . 5 μg/ml in TBS + 3% BSA for one hour at room temperature . Following removal of the primary antibody , the strips were washed with TBS and incubated with anti-rabbit IgG HRP-conjugated antibodies ( GE Healthcare , Pittsburgh , PA ) at a 1:20 , 000 dilution in TBS + 3% BSA . After a one-hour incubation with the secondary antibody , the Snoopers were washed with TBS + 0 . 05% Tween and developed with ECL reagents ( GE Healthcare ) . For further analysis of NPC2-lipid interaction , Hybond-C membranes ( GE-Healthcare ) were spotted with either 500 pmol of 18:1 LBPA/BMP ( S , R ) ( Avanti ) , 18:1 PA ( Avanti ) , 18:1 PG ( Avanti ) , 18:1 PS ( Avanti ) , and Egg PC , or with increasing concentrations of LBPA ( 125 , 250 , 375 and 500 pmol ) to analyze binding of WT NPC2 to various phospholipid species , or binding of NPC2 mutants to LBPA , respectively . Following the protocol of Dowler et al . ( 2002 ) , each phospholipid was spotted in duplicate and allowed to dry for one hour . Membranes were blocked for 1 hr in blocking buffer containing TBS ( 50 mM Tris/HCl , pH 7 . 5 , 150 mM NaCl ) and 5% ( w/v ) non-fat dry milk . Membranes were then incubated overnight at 4°C with either WT or mutant NPC2 protein diluted to a final concentration of 1 ug/ml in TBS and 3% ( w/v ) non-fat dry milk . The membranes were then washed at room temperature in TBST ( 0 . 1% Tween 20 ) six times for 5 min each , followed by incubation with mouse monoclonal anti-myc antibody ( Millipore ) at a 1:2000 dilution in TBS and 3% ( w/v ) milk . After washing with TBST , the membrane was then incubated with anti-mouse IgG IRDye-800CW conjugated antibody ( LI-COR , Lincoln , NE ) at a 1:10 , 000 dilution in TBS , 0 . 1% SDS , and 3% ( w/v ) milk . The membranes were finally washed in TBST 12 times for 5 min each at room temperature before acquiring images on the LI-COR Odyssey . Assays were performed as described in Fleury et al . ( 2015 ) , with minor modifications . Reaction mixtures for the interaction assays were prepared in white 384-well polystyrene non-binding surface NBS microplates ( Corning , Corning , NY ) with a final volume of 20 μL per well . Each reaction mix contained 6 μL of buffer A ( 20 mM sodium citrate , 150 mM NaCl , 1 mM EDTA pH 5 . 0 ) , 2 μL of the recombinant WT or mutant His-tagged NPC2 at a final concentration of 75 nM , 2 μL of biotinylated lipid solution in buffer A at a final concentration range of 1 μM - 58 . 5 nM , 5 μL of streptavidin-d2 conjugate ( Cisbio Bioassays , Bedford , MA ) and 5 μL of monoclonal anti-6His-europium cryptate antibody ( Cisbio ) in detection buffer ( 20 mM of HEPES pH 8 . 5 , 200 mM of potassium fluoride , 1% bovine serum albumin ) . The biotinylated lipids ( PS , PA , PG , PC and LPBA; Avanti ) were dried under nitrogen and the resultant film was initially reconstituted in EtOH and secondarily in binding buffer at a ratio of 1:10 EtOH:buffer . The final concentration of ethanol in the reaction was 1% ( v/v ) . Following an 18 hr incubation of the reaction mixture at room temperature , the fluorescence was measured with an Envision plate reader ( Perkin Elmer; λex = 320 nm , λem = 615 and 665 nm; 100 μs delay time ) . The HTRF ratio value was represented as Ch1/Ch2*10 , 000 where Ch1 is the energy transfer signal at 665 nm , and Ch2 is the europium cryptate antibody signal at 615 nm . The negative control wells contained donor and acceptor fluorochromes without NPC2 or biotinylated lipid . The negative control ( background ) readout was subtracted from all the sample readings . As detailed previously , a single dose of purified WT or mutant NPC2 protein was added to the media of NPC2 mutant fibroblasts cultured on 8-well tissue culture slides ( Nalgene ) , and allowed to incubate for 3 days . The final concentration of added protein was 0 . 4 nM . In keeping with prior literature ( Ko et al . , 2003; Liou et al . , 2006; McCauliff et al . , 2015; Wang et al . , 2010 ) , since identical amounts of NPC2 protein were added to equivalent samples of cultured cells , we reasonably assume equivalent uptake of the various NPC2 proteins; notably all bind cholesterol similarly , implying that they fold normally and thus are of similar shape and size , precluding any potential difference in uptake via fluid phase endocytosis . Following incubations the cells were fixed and stained with 0 . 05 mg/mL filipin III ( Fisher ) and subsequently imaged on a Nikon Eclipse E800 epifluorescence microscope using a DAPI filter set . Filipin stain was quantified as a ratio of fluorescence intensity per unit cell area in treated and untreated conditions with the accompanying NIS-Elements software ( Nikon Inc ) . Results are corrected for background fluorescence and are representative of an average of 80 to 100 cells per condition . WT , NPC1 , and NPC2 mutant fibroblasts were cultured to confluence in 100 mm petri dishes and passaged by trypsinization at a 1:3 ratio . After 24–48 hr , media was removed and replaced by media supplemented with either 30 , 100 , or 250 µM PG SUVs ( Bouvier et al . , 2009; Luquain-Costaz et al . , 2013 ) , or with 100 µM LBPA or PC SUVs . After 24 hr , cells were collected and 4–6 dishes of the same treatment were pooled . Protein levels were analyzed using the Bradford method ( Bradford , 1976 ) . Total cell lipids were extracted from 2 mL of 1 mg/mL protein via the method of Bligh and Dyer ( 1959 ) , resuspended in 200 µL 2:1 chloroform:methanol , and were run on HPTLC plates ( EMD Chemicals , Inc ) in a solvent of 65:35:5 chloroform:methanol ( v/v ) :30% ammonium hydroxide ( v/v ) ( Akgoc et al . , 2015 ) . Lipid spots were quantified by densiometric analysis ( ImageJ ) from standard curves of authentic standards . WT , NPC1– , and NPC2 mutant fibroblasts were plated onto 8-well tissue culture slides ( BD falcon ) at a density of approximately 20 , 000 cells per well and incubated at 37°C , 5% CO2 for 24 hr . Culture media was then removed and the cells were incubated with media supplemented with 100 µM PG SUVs for 24 hr . Cells were subsequently fixed and stained with 0 . 05 mg/mL filipin III . In some experiments , NPC2 deficient fibroblasts were secondarily incubated for 24 hr with WT or mutant NPC2 proteins at a final concentration of 0 . 4 nM , 24 hr after the 24 hr supplementation with PG . Cells were imaged on a Nikon Eclipse E800 epifluorescence microscope using a DAPI filter set to detect filipin . Filipin and antibody stain intensity was quantified with the accompanying NIS-Elements software ( Nikon Inc ) ; cholesterol accumulation was calculated as the ratio of filipin stain intensity to cell area ( Delton-Vandenbroucke et al . , 2007; McCauliff et al . , 2011; McCauliff et al . , 2015 ) . HeLa NPC1 knockout cells were seeded in 6 cm dishes at a density of 2 × 105 per dish , in DMEM ( Sigma ) supplemented with 10% FBS , 1% Penicillin-Streptomycin . Forty-eight hours after seeding , cells were incubated with 100 µM PG SUVs or vehicle ( PBS ) for 24 to 48 hr . One dish of cells per treatment condition was analyzed for cholesterol accumulation via mutant 125I-perfringolysin O ( PFO* ) staining . PFO* plasmid was kindly provided by Arun Radhakrishnan ( University of Texas Southwestern , Dallas ) ( Das et al . , 2013 ) and PFO* purification and staining was performed as described by Li et al . ( 2017 ) . Stained cells in PBS were analyzed using a BD Acuri TM C6 flow cytometer . WT fibroblasts cells were seeded in 6 cm dishes at a density of 1 . 2 × 105 cells per dish , in EMEM ( Sigma ) containing Earle’s Salts and Nonessential Amino Acids supplemented with 15% FBS , 1% Penicillin-Streptomycin , 2 mM L-glutamine and 1 mM sodium pyruvate . Forty-eight hours after seeding , cells were incubated with 1 µM U18666A in DMSO ( 0 . 1% v/v ) for 24 hr in order to induce an NPC1-like phenotype . Cells were then incubated with 100 µM PG SUVs for 16 , 34 and 40 hr without changing media . Following incubation with PG SUVs , two dishes of cells per sample were trypsinized , pelleted and used for PFO* labeling for flow cytometry as described above . Statistical analysis was performed using OriginPro 2016 ( OriginLab Corporation ) and SigmaPlot 12 . 0 . Means were compared using Student’s t-test for independent samples or one-way ANOVA where indicated with p<0 . 05 considered as significantly different . | Cholesterol is a type of fat that is essential for many processes in the body , such as repairing damaged cells and producing certain hormones . Normally , cholesterol enters cells from the bloodstream and is then moved to the parts of the cell that need it via a process known as ‘trafficking’ . When cholesterol trafficking goes wrong , abnormally large amounts of cholesterol and other fats accumulate within the cell . Over time , these fatty deposits become toxic to cells and eventually damage the affected tissues . Niemann-Pick type C disease ( NPC ) is a severe genetic disorder affecting cholesterol trafficking . It is characterized by cholesterol build-up in multiple tissues , including the brain , which ultimately causes degeneration and death of nerve cells . Two proteins , NPC1 and NPC2 , are involved in NPC disease . Both proteins normally help move cholesterol out of important trafficking compartments ( known as the endosomal and lysosomal compartments ) to other areas of the cell where it is needed . Patients with the disease can have mutations in either the gene for NPC1 or the gene for NPC2 . This means that cells from NPC1 patients do not make enough functional NPC1 protein ( but contain working NPC2 ) , and vice versa . Previous studies had shown that giving cells with NPC1 mutations large amounts of the small molecule lysobisphosphatidic acid ( LBPA for short ) could compensate for the loss of NPC1 , and stop the toxic build-up of cholesterol . McCauliff , Langan , Li et al . therefore wanted to explore exactly how LBPA was doing this . They had shown that LBPA dramatically increased the ability of purified NPC2 protein to transport cholesterol , and wondered if the effect of LBPA in the cells without NPC1 depended on NPC2 . They predicted that boosting LBPA levels would not work in cells lacking NPC2 . Biochemical experiments using purified protein showed that LBPA and NPC2 did indeed interact directly with each other . Systematically changing different building blocks of NPC2 revealed that a single region of the protein is sensitive to LBPA , and when this region was altered , LBPA could no longer interact with NPC2 . Since LBPA is naturally produced by cells , they then stimulated cells grown in the laboratory to generate more LBPA using its precursor phosphatidylglycerol . They used cells from patients with mutations in either NPC1 or NPC2 and demonstrated that LBPA’s ability to reverse the accumulation of cholesterol was dependent on its interaction with NPC2 . Thus , increasing LBPA levels in cells from patients with NPC1 mutations was beneficial , but had no effect on cells from patients with NPC2 mutations . These results shed new light not only on how cells transport cholesterol , but also on potential methods to combat disorders of cellular cholesterol trafficking . In the future , LBPA could be developed as a genetically tailored , patient-specific therapy for diseases like NPC . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"cell",
"biology"
] | 2019 | Intracellular cholesterol trafficking is dependent upon NPC2 interaction with lysobisphosphatidic acid |
Disorders of consciousness are a heterogeneous mixture of different diseases or injuries . Although some indicators and models have been proposed for prognostication , any single method when used alone carries a high risk of false prediction . This study aimed to develop a multidomain prognostic model that combines resting state functional MRI with three clinical characteristics to predict one year-outcomes at the single-subject level . The model discriminated between patients who would later recover consciousness and those who would not with an accuracy of around 88% on three datasets from two medical centers . It was also able to identify the prognostic importance of different predictors , including brain functions and clinical characteristics . To our knowledge , this is the first reported implementation of a multidomain prognostic model that is based on resting state functional MRI and clinical characteristics in chronic disorders of consciousness , which we suggest is accurate , robust , and interpretable .
Severe brain injury can lead to disorders of consciousness ( DOC ) . Some patients recover consciousness after an acute brain insult , whereas others tragically fall into chronic DOC . The latter cannot communicate functionally or behave purposefully . Most patients remain bedridden , and require laborious care . The medical community is often confronted with an inability to meet the expectations of the chronic DOC patients' families . The social , economic , and ethical consequences are also tremendous ( Bernat , 2006 ) . In parallel , although more validations are required , recent pilot studies have proposed new therapeutic interventions , which challenge the existing practice of early treatment discontinuation for a chronic DOC patient ( Schiff et al . , 2007; Corazzol et al . , 2017; Yu et al . , 2017 ) . However , before using these novel therapeutic interventions , clinicians first need to determine whether the patient is a suitable candidate . The availability of an accurate and robust prognostication is therefore a fundamental concern in the clinical management of chronic DOC patients , as medical treatment , rehabilitation therapy and even ethical decisions depend on this information . To date , the prognostication for a DOC patient is based on physician observation of the patient's behavior over period that is sufficient to allow determination of whether there is any evidence of awareness . On the one hand , a patient's motor impairment , sensory deficit , cognitive damage , fluctuation of vigilance and medical complications could give rise to misjudgments; on the other hand , for the assessor , a lack of knowledge regarding DOC , poor training and non-use of adequate behavioral scales are additional elements that may contribute to a high possibility of mistakes . Consequently , careful and repeated behavioral assessments are considered to be particularly important for a precise diagnostic and prognostic judgment ( Wannez et al . , 2017 ) . Nonetheless , behavioral assessments are inevitably subjective and vulnerable to a variety of personal interferences ( Giacino et al . , 2009 ) . Physicians and scientists have therefore been seeking accurate and objective markers for diagnosis and prognosis ( Demertzi et al . , 2017; Noirhomme et al . , 2017 ) . Several pioneering studies have suggested that the etiology , incidence age and duration of DOC are important indicators for prognosis ( Multi-Society Task Force on PVS , 1994 ) . Specifically , patients who have non-traumatic brain injury are expected to have a worse functional recovery than traumatic brain injury patients , and young patients were considered more likely to have a favorable outcome than older ones . During the recent decades , some pilot prognostic models have also been explored that are based on features of neurological examination ( Zandbergen et al . , 1998; Booth et al . , 2004; Dolce et al . , 2008 ) , abnormalities detected with electroencephalogram ( EEG ) and evoked potentials ( Steppacher et al . , 2013; Kang et al . , 2014; Hofmeijer and van Putten , 2016; Chennu et al . , 2017 ) , anatomical and functional changes identified with brain computed tomography ( CT ) , positron emission tomography ( PET ) and magnetic resonance imaging ( MRI ) ( Maas et al . , 2007; Sidaros et al . , 2008; Neuro Imaging for Coma Emergence and Recovery Consortium et al . , 2012; Luyt et al . , 2012; Stender et al . , 2014; Wu et al . , 2015 ) , and physiological and biochemical disturbances at both the brain and body levels ( Kaneko et al . , 2009; Rundgren et al . , 2009 ) . Despite many efforts , however , identifying efficient biomarkers for the early prediction of outcome is still challenging and requires additional research . One of the reasons for this is that the DOC could have many different causes and could be associated with several neuropathological processes and different severities , such that any method when used alone carries the risk of false prediction ( Bernat , 2016; Rossetti et al . , 2016 ) . Recently , resting state functional MRI ( fMRI ) has been widely used to investigate the brain functions of DOC patients . Research suggests that these patients demonstrate multiple changes in brain functional networks , including the default mode ( Vanhaudenhuyse et al . , 2010; Silva et al . , 2015 ) , executive control ( Demertzi et al . , 2014; Wu et al . , 2015 ) , salience ( Qin et al . , 2015; Fischer et al . , 2016 ) , and sensorimotor ( Yao et al . , 2015 ) , auditory ( Demertzi et al . , 2015 ) , visual ( Demertzi et al . , 2014a ) and subcortical networks ( He et al . , 2015 ) . The within-network and between-network functional connectivity appeared to be useful indicators of functional brain damage and the likelihood of consciousness recovery ( Silva et al . , 2015; Di Perri et al . , 2016 ) . Taken together , these studies suggest that the brain networks and functional connectivity detected with resting state fMRI could be valuable biomarkers that can be used to trace the level of consciousness and predict the possibility of recovery . With advances in medicine , prognostication of a DOC patient has moved toward a multidomain paradigm that combines clinical examination with the application of novel technologies ( Gosseries et al . , 2014 ) . Multidomain assessment has the potential to improve prediction accuracy . More importantly , it can provide reassurances about the importance of each predictor for prognostication by offering concordant evidence ( Stevens and Sutter , 2013; Rossetti et al . , 2016 ) . More than 20 years ago , the Multi-Society Task Force on persistent vegetative state ( PVS ) suggested that the etiology , incidence age and duration of DOC could help to predict the outcome ( Multi-Society Task Force on PVS , 1994 ) , and numerous studies have subsequently validated the clinical utility of these features ( Jennett , 2005; Bruno et al . , 2012; Estraneo et al . , 2013; Celesia , 2016 ) . Therefore , it is possible that a multidomain model that combines these clinical characteristics and resting state fMRI data could improve prognostic predictions at an individual level and could lead to the early identification of patients who could recover consciousness . The present work had two major objectives . The first aim was to develop an approach to predict the prognosis of an individual DOC patient using clinical characteristics and resting state fMRI . The second aim , building on the first , was to further explore the different prognostic effects of these clinical and brain imaging features .
This study involved three datasets . The datasets referred to as ‘Beijing 750’ and ‘Beijing HDxt’ were both collected in the PLA Army General Hospital in Beijing , and the same medical group diagnosed and managed the patients . However , the MRI scanners and imaging acquiring protocols were different for these two datasets: the ‘Beijing HDxt’ cohort was scanned with a GE signa HDxt 3 . 0T scanner between May 2012 and December 2013 , whereas the ‘Beijing 750’ cohort was scanned with a GE Discovery MR750 3 . 0T scanner between January 2014 and May 2016 . The dataset referred to as ‘Guangzhou HDxt’ was collected from the Guangzhou General Hospital of Guangzhou Military Command in Guangzhou , and the MRI data were obtained with a GE signa HDxt 3 . 0T scanner between April 2011 and December 2014 . The inclusion criterion was that the patients should be at least 1 month after the acute brain insult so that they met the DOC diagnosis . Patients were excluded when there was an unstable level of consciousness ( continuous improvement or decline within the two weeks before the T0 time point ) , uncertain clinical diagnosis ( ambiguity or disagreement between examiners ) , contraindication for MRI or large focal brain damage ( >30% of total brain volume ) . A total of 160 DOC patients were initially enrolled in this study . Eleven patients were excluded due to large local brain lesions or movement artifacts during MRI scanning . Nine patients died during the period of the follow-up , 16 patients were lost to follow-up , and for 12 patients no definite outcome information was collected at the 12-month endpoint of the follow-up . Thus , according to the inclusion and exclusion criteria and the follow-up results , the ‘Beijing 750’ dataset included 46 vegetative state/unresponsive wakefulness syndrome ( VS/UWS ) patients and 17 minimally conscious state ( MCS ) patients . The ‘Beijing HDxt’ dataset contained 20 VS/UWS patients and 5 MCS patients , and the ‘Guangzhou HDxt’ dataset contained 16 VS/UWS patients and 8 MCS patients . The demographic and clinical characteristics of the patients are summarized in Table 1 , with additional details provided in Appendix 1—table 1 , 2 and 3 . The ‘Beijing 750’ dataset also included 30 healthy participants , and the ‘Beijing HDxt’ dataset included 10 healthy participants . All of the healthy participants were free of psychiatric or neurological history . These healthy participants are referred to as ‘normal controls’ . See Appendix 1—table 4 and 5 for details . As the ‘Beijing 750’ dataset involved more patients than the other two datasets , it was used as the training dataset for model development and internal validation , whereas the ‘Beijing HDxt’ and ‘Guangzhou HDxt’ datasets were only used for external validation . The study was approved by the Ethics Committee of the PLA Army General Hospital ( protocol No: 2011–097 ) and by the Ethics Committee of the Guangzhou General Hospital of Guangzhou Military Command ( protocol No: jz20091287 ) . Informed consent to participate in the study was obtained from the legal surrogates of the patients and from the normal controls . All of the participants in the three datasets were scanned with resting state fMRI and T1-weighted 3D high-resolution imaging . During the MRI scanning , the participants did not take any sedative or anesthetic drugs . The resting state fMRI scan was obtained using a T2*-weighted gradient echo sequence , and a high-resolution T1-weighted anatomical scan was obtained to check whether the patients had large brain distortion or focal brain damage . For the training dataset , ‘Beijing 750’ , the resting state fMRI acquisition parameters included TR/TE = 2000/30 ms , flip angle = 90° , axial 39 slices , thickness = 4 mm , no gap , FOV = 240 × 240 mm , matrix = 64 × 64 , and 210 volumes ( i . e . 7 min ) . For the testing dataset , ‘Beijing HDxt’ , the resting state fMRI acquisition parameters were as follows: axial 33 slices , TR/TE = 2000/30 ms , flip angle = 90° , thickness = 4 mm , no gap , FOV = 220 × 220 mm , matrix = 64 × 64 , and 240 volumes ( i . e . 8 min ) . For the testing dataset , ‘Guangzhou HDxt’ , the resting state fMRI acquisition parameters included axial 35 slices , TR/TE = 2000/30 ms , flip angle = 90° , thickness = 4 mm , no gap , FOV = 240 × 240 mm , matrix = 64 × 64 , and 240 volumes ( i . e . 8 min ) . The data analysis pipeline is illustrated in Figure 2 . Preprocessing and connectivity calculation were performed in the same way for the training dataset and the two testing datasets . All resting-state fMRI scans were preprocessed using SPM8 ( SPM , RRID:SCR_007037 ) and in-house Matlab codes . Specifically , the first five volumes of each subject were discarded . The remaining resting-state fMRI volumes were corrected for slice timing differences and realigned to the first volume to correct for inter-scan movements . The functional images were then spatially smoothed with a Gaussian kernel of 6 × 6 × 6 mm full-width at half maximum . Linear regression was used to remove the influence of head motion , whole brain signals and linear trends . The variables regressed out included 12 motion parameters ( roll , pitch , yaw , translation in three dimensions and their first derivatives ) , the average series of the signals within the brain , and the regressors for linear trends . Motion artifact is increasingly recognized as an important potential confound in resting state fMRI studies . Any particular motion may produce a wide variety of signal changes in the fMRI data , and may thus introduce complicated shifts in functional connectivity analysis . This problem was particularly serious for the DOC patients , as they were unlikely to follow the experimental instructions and control their head motion . To balance the demands of noise reduction and data preservation , we censored volumes that preceded or followed any movement ( framewise displacement , FD ) greater than 1 . 5 mm . The FD is a summarization of the absolute values of the derivatives of the translational and rotational realignment estimates ( after converting the rotational estimates to displacement at 50 mm radius ) ( Power et al . , 2015 ) . The head motion measures were achieved in the preprocessing step of realignment using SPM . To obtain reliable Pearson's correlations for functional connectivity , the patients with less than 50 volumes worth of remaining data were excluded . More information about the analysis and validation of controls for motion-related artifacts are provided in Appendix 4 . Finally , to reduce low-frequency drift and high-frequency noise , band-pass filtering ( 0 . 01–0 . 08 Hz ) was only performed on volumes that survived motion censoring . As noted in the introduction , multiple functional brain networks are disrupted in DOC patients . Among these impaired networks , six ( the default mode , executive control , salience , sensorimotor , auditory , and visual networks ) show system-level damages and significant correlations with behavioral assessments ( Demertzi et al . , 2014 , 2015 ) . We therefore defined a total of 22 regions of interest ( ROIs ) to probe these six brain networks . The definitions of the 22 ROIs were based on the results of a series of previous brain functional studies ( Seeley et al . , 2007; Raichle , 2011; Demertzi et al . , 2015 ) , and their names and Montreal Neurological Institute ( MNI ) coordinates are listed in Appendix 2 . The connection templates of the six brain networks were first investigated within the normal control group . In addition to the above-mentioned preprocessing stages , the resting state fMRI scans of the normal controls in the training dataset were transformed into MNI standard space . For each of the six networks , time series from the voxels contained in the various ROIs were extracted and averaged together . The averaged time series were then used to estimate whole-brain correlation r maps that were subsequently converted into normally distributed Fisher’s z-transformed correlation maps . Group functional connectivity maps for each of the six networks were then created with a one-sample t test ( see Appendix 3 for details ) . Notably , the T map included both positive and negative values . We used the six T maps as the brain connection templates of the corresponding brain networks in the healthy population , which would assist to define one type of imaging features , that is the connection feature of the ROI . More information about the connection features of the ROIs are provided in the following section . The conventional fMRI preprocess normalizes individual fMRI images into a standard space defined by a specific template image . Our goal was to extend this conventional approach to generate a functional connectivity image for each patient in his/her own imaging space . During the preprocessing of each patient’s fMRI scans , the 22 ROIs and six brain connection templates were therefore spatially warped to individual fMRI space and resampled to the voxel size of the individual fMRI image . We also developed tools to check the registration for each subject visually , some examples of which are provided in Appendix 5 and Supplementary file 1 . We designed two types of imaging features from the resting state fMRI , one being the functional connectivity between each pair of 22 ROIs , and the other being the spatial resemblance between the functional connection patterns of each ROI and the brain connection templates across the whole brain . The functional connectivity was based on the Pearson’s correlation coefficients , while the spatial resemblance was conceptually similar to the template-matching procedure ( Greicius et al . , 2004; Seeley et al . , 2007; Vanhaudenhuyse et al . , 2010 ) . The basis of template matching is that the greater the spatial consistency that exists between the template of a brain network and a specific connectivity map ( for example , a component in an independent component analysis ) , the stronger the possibility that the connectivity map belongs to that brain network . Here , for each ROI of an individual DOC patient , we first computed the Pearson’s correlation coefficients between the time-course of the ROI and that of each voxel within the brain so as to obtain a functional connectivity map , and subsequently converted the functional connectivity map to a normally distributed Fisher’s z transformed correlation map . Next , we calculated the Pearson’s correlation coefficients between the Fisher’s z transformed correlation map and the corresponding brain connection template wrapped to individual fMRI space across each voxel within the brain . A greater correlation coefficient between the two maps suggests that there is more spatial resemblance between the functional connectivity map of the ROI and the normal brain connection template . Our assumption was that the more spatial consistency that existed between the connectivity map of the ROI in a DOC patient and the brain connection template , the more intact the corresponding brain function of the ROI in this individual . In this way , we defined the connection feature of the ROI with the spatial resemblance . Overall , for each participant in this study , there were 231 ( 22 × 21/2 ) functional connectivity features and 22 brain area connection features . Feature selection techniques have been widely adopted in brain analysis studies , in order to produce a small number of features for efficient classification or regression , and to reduce overfitting and increase the generalization performance of the model ( Fan et al . , 2007; Dosenbach et al . , 2010; Drysdale et al . , 2017 ) . Feature ranking and feature subset selection are two typical feature selection methods ( Guyon and Elisseeff , 2003 ) . Feature subset selection methods are generally time consuming , and even inapplicable when the number of features is extremely large , whereas ranking-based feature selection methods are subject to local optima . Therefore , these two feature selection methods are usually used jointly . Here , we first used a correlation-based feature selection technique to select an initial set of features , and then adopted a feature subset selection method for further selection . As a univariate method , correlation-based feature selection is simple to run and understand , and measures the linear correlation between each feature and the response variable . Here , the image features ( i . e . functional connectivity features and brain area connection features ) that significantly correlated to the CRS-R scores at the T1 time point across the DOC patients in the training dataset were retained for further analysis . Competitive adaptive reweighted sampling coupled with partial least squares regression ( CARS-PLSR , http://libpls . net/ ) was then used for further feature subset selection ( Li et al . , 2009 , 2014 ) . Briefly , CARS-PLSR is a sampling-based feature selection method that selects the key informative variables by optimizing the model's performance . As it provides the influence of each variable without considering the influence of the remainder of the variables , CARS-PLSR is efficient and fast in carrying out feature selection ( Mehmood et al . , 2012 ) , and has therefore been used to explore possible biomarkers in medicine ( Tan et al . , 2010 ) and for wavelength selection in chemistry ( Fan et al . , 2011 ) . Using CARS-PLSR , we selected a subset of key informative imaging features . Notably , both the correlation-based and CARS-PLSR feature selection methods filtered the features from the original feature set without any transformations . This made the prognostic regression model easier to interpret , as the imaging predictors were associated with either brain regions or functional connectivity . PLSR is able to handle multicollinearity among the predictors well ( Wold et al . , 2001; Krishnan et al . , 2011 ) . It was therefore used to generate the prognostic regression model in the training dataset ‘Beijing 750’ . Given that clinical characteristics—including the etiology , incidence age and duration of DOC—have been verified as useful prognostic indicators , we designated the selected imaging features and the three clinical characteristics at the T0 time point as independent co-variates and the CRS-R score at the T1 time point as the dependent variable . Among the three clinical characteristics , the incidence age and duration of DOC were quantitative variables , whereas the etiology was a qualitative variable . In accordance with a previous study ( Estraneo et al . , 2010 ) , we categorized the etiology into three types: traumatic brain injury , stroke and anoxic brain injury . Thus , two dummy variables for etiology were designed and included in the model . Prior to model training , all involved predictors were centered and normalized ( i . e . transformed into Z-scores ) . The prognostic regression model therefore took the imaging and clinical features as input and returned a predicted score as output . In the training dataset ‘Beijing 750’ , we used cross-validation to decide that the number of latent variables for PLSR was three . To evaluate the regression model , the coefficient of determination R2 between the predicted scores and the CRS-R scores at the T1 time point was calculated , and the Bland-Altman plot was used to measure the agreement between them . Next , receiver operating characteristic ( ROC ) curves were plotted for the predicted scores . The optimal cut-off value for classifying an individual patient as having recovered consciousness or not was appointed to the point with the maximal sum of true positive and false negative rates on the ROC curve . Individual patients were classified as exhibiting recovery of consciousness if their predicted scores were higher than or equal to the cut-off value , otherwise as consciousness non-recovery . The classification accuracy was calculated by comparing the predicted label and the actual GOS score , that is 'consciousness recovery' ( GOS ≥ 3 ) versus ‘consciousness non-recovery’ ( GOS ≤ 2 ) . As model interpretation is an important task in most applications of PLSR , there has been considerable progress in the search for optimal interpretation methods ( Kvalheim and Karstang , 1989; Kvalheim et al . , 2014 ) . In this study , using the Significant Multivariate Correlation ( sMC ) method ( Tran et al . , 2014 ) , we assessed predictor importance in the prognostic regression model . The key points in sMC are to estimate the correct sources of variability resulting from PLSR ( i . e . regression variance and residual variance ) for each predictor , and use them to determine statistically a variable's importance with respect to the regression model . The F-test values ( termed the sMC F-values ) were used to evaluate the predictors' importance in the prognostic regression model . The prognostic regression model was internally validated using bootstrap sampling ( Steyerberg , 2008 ) . Specifically , bootstrap samples were drawn with replacement from the training dataset ‘Beijing 750’ such that each bootstrap sampling set had a number of observations equal to that of the training dataset . Using a bootstrap sampling set , correlation-based feature selection and CARS-PLSR were first used to select the feature subset , after which the PLSR was used to generate a prognostic model . We then applied the model to the bootstrap sampling set and the original training dataset , and calculated the coefficient of determination R2 of each of the two datasets . The difference between the two coefficients of determination was defined as the optimism . This process was repeated 1000 times to obtain a stable estimate of the optimism . Finally , we subtracted the optimism estimate from the coefficient of determination R2 of the ‘Beijing 750’ training dataset to obtain the optimism-corrected performance estimate . In addition , out-of-bag ( OOB ) estimation was used as an estimate of model classification performance in the training dataset ( James et al . , 2013 ) . Specifically , for the original training dataset x , we left out one sample at a time and denoted the resulting sets by x ( –1 ) , . . . , x ( n ) . From each leave-one-out set x ( –i ) , 1000 bootstrap learning sets of size n–1 were drawn . On every bootstrap learning set generated from x ( –i ) , we carried out feature selection , built a PLSR regression and classification model , and applied the model to the test observation xi . A majority vote was then made to give a class prediction for observation xi . Finally , we calculated the accuracy for the whole training dataset x . External validation is essential to support the general applicability of a prediction model . We ensured external validity by testing the model in two testing datasets , neither of which included samples that were considered during the development of the model . First , using the prognostic regression model , we calculated one predicted score for each patient in the two testing datasets . As the ‘Beijing HDxt’ dataset assessed the patients' CRS-R scores at the T1 time point , we calculated the coefficient of determination R2 between the predicted scores and the patients' CRS-R scores at this time point . The Bland-Altman plot was also determined . Finally , the patients in the two testing datasets were assessed as achieving consciousness recovery or not on the basis of the cut-off threshold obtained using the training dataset . The performance of the classification , including the accuracy , sensitivity and specificity , was determined . Using the modeling and validation method described above , we examined the predictability and generalizability in the two testing datasets on the basis of the clinical features alone or the imaging features alone . In addition , to compare the two types of single-domain models and the combination model , we used bootstrap resampling to obtain the distribution of the prediction accuracies in the two testing datasets based on each of the three types of models . We first resampled with replacement from the training dataset , and built a regression and classification model based on the clinical features alone , the neuroimaging features alone , or the combination of the two-domain features . We then tested the classification accuracy in the two testing datasets using the three types of models . In this way , we obtained the distribution of the prediction accuracies using each of the three types of models . Next , we used repeated measures ANOVA to determine whether or not the performances of the three types of models were the same; we also used Ψ , the root-mean-square standardized effect , to report the effect sizes of the mean differences between them . We compared the prediction performances between the proposed modeling method and linear SVM . The code for SVM was downloaded from LIBSVM ( LIBSVM , RRID:SCR_010243 ) . The 253 imaging features and the four clinical features were integrated into one feature vector . No feature selection was adopted in the linear SVM-based classification . The patients with GOS ≥3 were labeled as 1 , with the others being designated as −1 ( i . e . GOS ≤2 ) . Similarly , the OOB estimation process was used to estimate the performance of linear SVM in the training dataset ‘Beijing 750’ . Next , using all the samples in the training dataset ‘Beijing 750’ , we trained a linear SVM-based classifier and then tested the predictive accuracy in the two testing datasets .
The prognostic regression model is presented in Figure 4 . On the basis of the regression formula , we noted some interesting findings . First , there were both positive and negative weights . In particular , the weights were all positive for the three brain area connection features , whereas the weight for the functional connectivity feature between the aMPFC in the default mode network and the DMPFC in the executive control network was negative . Interestingly , this connection had the maximum sMC F-value as shown in Figure 4B . In addition , the age and the anoxic etiology had negative weights , and the age predictor had the largest sMC F-value among the four clinical features . Figure 5A presents the predicted score for each patient in the training dataset . As shown in Figure 5B , there was good agreement between the CRS-R scores at the T1 time point and the predicted scores . The apparent coefficient of determination R2 was equal to 0 . 65 ( permutation test , p=0 . 001 ) , and the Bland-Altman plot verified the consistency between the predicted and achieved scores ( one sample T test , p = 1 . 0 ) . The prognostic regression model was internally validated using bootstrapping . The optimism-corrected coefficient of determination R2 was equal to 0 . 28 . Figure 5C illustrates the number and proportion of DOC patients in different bands of predicted scores . We found that the proportion of the patients with a ‘consciousness recovery’ outcome in the patient cohorts rose in conjunction with an increase in the predicted score . The higher the predicted score , the higher the proportion of patients who exhibited a favorable outcome . Figure 5D shows the area under the ROC curve ( AUC = 0 . 96 , 95% CI = 0 . 89–0 . 99 ) . On the basis of the ROC curve for the training dataset , a threshold 13 . 9 was selected as the cut-off point to classify the recovery of individual patients . In other words , if the predicted score for a patient was equal to or larger than 13 . 9 , the classification model designated the label ‘consciousness recovery’ for this patient , otherwise ‘consciousness non-recovery’ . The classification accuracy was assessed by comparing the predicted and actual outcomes , that is 'consciousness recovery' ( GOS ≥ 3 ) versus ' consciousness non-recovery’ ( GOS ≤ 2 ) . Using this method , the classification accuracy in the training dataset was up to 92% . Specifically , the sensitivity was 85% , the specificity was 94% , the positive predictive value ( PPV ) was 79% , the negative predictive value ( NPV ) was 96% , and the F1 score was 0 . 81 . The OOB was able to provide the mean prediction error on each training sample and to estimate the generalizability of our method in the training dataset . Using the OOB estimation , we found that the prediction accuracy in the training dataset ‘Beijing 750’ was 89% , and the sensitivity , specificity , PPV and NPV were 69% , 94% , 100% , and 87% , respectively . The performance of the prediction model on the two testing datasets is illustrated in Figure 6 . As we assessed the CRS-R scores at the T1 time point for the patients in the ‘Beijing HDxt’ dataset , we calculated the coefficient of determination R2 between these scores and the predicted scores . The R2 was equal to 0 . 35 ( permutation test , p=0 . 005 ) , with the Bland-Altman plot verifying the consistency between the predicted and actual scores ( one sample T test , p=0 . 89 ) . Using the predicted score 13 . 9 as the threshold , we then tested the classification accuracy on the two testing datasets . We found that , for the ‘Beijing HDxt’ dataset , the prediction accuracy was up to 88% ( sensitivity: 83% , specificity: 92% , PPV: 92% , NPV:86% , F1 score: 0 . 87; permutation test , p<0 . 001 ) , while for the ‘Guangzhou HDxt’ dataset it was also up to 88% ( sensitivity: 100% , specificity: 83% , PPV: 67% , NPV: 100% , F1 score: 0 . 80; permutation test , p<0 . 001 ) . Notably , our model demonstrated good sensitivity and specificity for both the ‘subacute’ patients ( i . e . duration of unconsciousness ≤3 months ) and those in the chronic phase ( i . e . duration of unconsciousness >3 months ) , as shown in Figure 7 . More interestingly , for the testing dataset ‘Beijing HDxt’ , eight DOC patients who were initially diagnosed as VS/UWS subsequently recovered consciousness . Using the proposed model , we could successfully identify seven of these and there was only one false-positive case . That is , for the VS/UWS patients , the model achieved 90 . 0% accuracy ( sensitivity: 87 . 5% , specificity: 91 . 7% , PPV: 87 . 5% , NPV: 91 . 7% , F1 score: 0 . 875 ) . To test robustness , we evaluated whether the present prognostic regression model generalized to the healthy subjects scanned in the ‘Beijing 750’ training dataset ( n = 30 ) and to the ‘Beijing HDxt’ testing dataset ( n = 10 ) . We found that both the healthy subjects and the ‘consciousness recovery’ patients had significantly higher predicted imaging subscores than the ‘consciousness non-recovery’ patients ( two-sample T test , p<0 . 05 ) . Additional information is provided in Appendix 8 . When only the clinical features were used to build the predictive model , the prediction accuracy for the ‘Beijing HDxt’ dataset was 68% ( sensitivity: 58% , specificity: 77% , PPV: 70% , NPV: 67% , F1 score: 0 . 64 ) , whereas for the ‘Guangzhou HDxt’ dataset , it was 83% ( sensitivity: 100% , specificity: 78% , PPV: 60% , NPV: 100% , F1 score: 0 . 75 ) . When only the imaging features were involved in the model , the prediction accuracy for the ‘Beijing HDxt’ dataset was 80% ( sensitivity: 67% , specificity: 92% , PPV: 89% , NPV: 75% , F1 score: 0 . 76 ) , whereas for the ‘Guangzhou HDxt’ dataset , it was 79% ( sensitivity: 100% , specificity: 72% , PPV: 55% , NPV: 100% , F1 score: 0 . 71 ) . Using bootstrapping , we obtained the distribution of the prediction accuracies in the two testing datasets with each of the three types of models . In the 'Beijing HDxt' testing dataset , the means±standard deviations of the distribution of the prediction accuracies were 0 . 815±0 . 050 , 0 . 811±0 . 044 , and 0 . 666±0 . 037 for the combination model , the model using imaging features alone , and the model using clinical features alone , respectively . We found that there were significant differences between the means of the classification accuracies using the three types of models ( repeated measures ANOVA , p<0 . 001 ) . Subsequently , we conducted pairwise comparisons . We found that there was significant difference between the combination model and the model s using the imaging feature alone ( paired sample t-test , p=0 . 001 ) or using the clinical feature alone ( paired sample t-test , p<0 . 001 ) . We also found that there was significant difference between the model using the imaging feature alone and the model using the clinical feature alone ( paired sample t-test , p<0 . 001 ) . Using effect-size analysis , we found that there was a mean difference of Ψ=0 . 004 ( 95% CI = [0 . 002 , 0 . 007] ) between the combination method and the method using only imaging features , and Ψ=0 . 149 ( 95% CI = [0 . 147 , 0 . 152] ) between the combination method and the method using only clinical features . We also observed a mean difference of Ψ = 0 . 145 ( 95% CI = [0 . 142 , 0 . 147] ) between the methods using only imaging features and only clinical features . In the 'Guangzhou HDxt' testing dataset , the mean±standard deviation of the distribution of the prediction accuracies was 0 . 863±0 . 051 , 0 . 783±0 . 044 , and 0 . 829±0 . 086 for the combination model , the model using imaging features alone , and the model using clinical features alone , respectively . Similarly , we found that there were significant differences between the mean of the classification accuracies using the three types of models ( repeated measures ANOVA , p<0 . 001 ) , and there was significant difference between the combination model and the models using imaging features alone ( paired sample t-test , p<0 . 001 ) or using clinical features alone ( paired sample t-test , p<0 . 001 ) . Using effect-size analysis , we found a mean difference of Ψ = 0 . 080 ( 95% CI = [0 . 076 , 0 . 084] ) between the combination model and the model using the imaging features alone , and Ψ = 0 . 034 ( 95% CI = [0 . 028 , 0 . 040] ) between the combination model and the model using only clinical features . We also observed a mean difference of Ψ = –0 . 046 ( 95% CI = [–0 . 053 , –0 . 040] ) between the model using imaging features alone and that using only clinical features . Therefore , in both testing datasets , the combination of imaging and clinical features demonstrated higher accuracy than the use of the single domain features alone . In addition , use of the imaging features alone had higher predictive power in comparison to use of the clinical features alone in the ‘Beijing HDxt’ dataset , whereas the opposite condition was observed in the ‘Guangzhou HDxt’ dataset , suggesting that the two testing datasets might be heterogeneous . More information about the single-domain models are provided in Supplementary file 2 . Using the OOB estimation , we found that the accuracy of the linear SVM-based classification method in the training dataset ‘Beijing 750’ was 83% ( sensitivity: 31% , specificity: 96% , PPV: 100% , NPV: 81% ) , which was lower than the accuracy of our proposed modeling method ( i . e . accuracy: 89% , sensitivity: 69% , specificity: 94% , PPV: 100% , NPV: 87% ) . On the other hand , the linear SVM-based classification method achieved an accuracy of 76% ( sensitivity: 58% , specificity: 92% , PPV: 88% , NPV: 71% ) and 88% ( sensitivity: 100% , specificity: 83% , PPV: 67% , NPV: 100% ) in the ‘Beijing HDxt’ testing dataset and the ‘Guangzhou HDxt’ testing dataset , respectively . That is , the accuracy in the ‘Beijing HDxt’ testing dataset was lower than that in our method , whereas the accuracy in the ‘Guangzhou HDxt’ testing dataset was similar to that of our approach . Therefore , taking together the performance comparisons in both the training dataset and the two testing datasets , we believe that our method based on feature selection and PLSR should have higher prediction accuracy and better generalizability in comparison to linear SVM .
In this paper , we describe the development of a prognostic model that combines resting state fMRI with three clinical characteristics to predict one-year outcomes at the single-subject level . The model discriminated between patients who would later recover consciousness and those who would not with an accuracy of around 88% in three datasets from two medical centers . It was also able to identify the prognostic importance of different predictors , including brain functions and clinical characteristics . To our knowledge , this is the first reported implementation of a multidomain prognostic model that is based on resting state fMRI and clinical characteristics in chronic DOC . We therefore suggest that this novel prognostic model is accurate , robust , and interpretable . For research only , we share the prognostic model and its Matlab code at a public download resource ( https://github . com/realmsong504/pDOC; copy archived at https://github . com/elifesciences-publications/pDOC ) . Brain functions are mediated by the interactions between neurons within different neural circuits and brain regions . Functional imaging can detect the local activity of different brain regions and explore the interactions between them , and has demonstrated potential for informing diagnosis and prognosis in DOC . On the one hand , many studies have focused on one modality of brain functional imaging , such as PET ( Phillips et al . , 2011 ) , SPECT ( Nayak and Mahapatra , 2011 ) , task fMRI ( Owen et al . , 2006; Coyle et al . , 2015 ) , or resting state fMRI ( Demertzi et al . , 2015; Qin et al . , 2015; Wu et al . , 2015; Roquet et al . , 2016 ) . On the other hand , some cross-modality studies have compared the diagnostic precision between imaging modalities , for example , comparing PET imaging with task fMRI ( Stender et al . , 2014 ) or comparing PET , EEG and resting state fMRI ( Golkowski et al . , 2017 ) . In our study , by combining brain functional networks detected from resting state fMRI with three clinical characteristics , we built a computational model that allowed us to make predictions regarding the prognosis of DOC patients at an individual level . We compared the models separately using only the imaging features or only the clinical characteristics and found that the combination of these predictors achieved greater accuracy . This validated the need for the use of accumulative evidence stemming from multiple assessments , each of which has different sensitivity and specificity in detecting the capacity for recovery of consciousness ( Demertzi et al . , 2017 ) . Validations in additional and unseen datasets were also undertaken to evaluate the feasibility of the predictive model . Our results showed about 88% average accuracy across the two testing datasets , and good sensitivity and specificity in both the ‘subacute’ patients ( i . e . 1 months ≤ duration of unconsciousness ≤ 3 months ) and those in the prolonged phase ( i . e . duration of unconsciousness >3 months ) , which suggested good robustness and the generalizability of our model . Further , the sensitivity of 83% and 100% obtained across the two testing datasets demonstrated a low false-negative rate , which would avoid predicting non-recovery in a patient who can actually recover . Our method successfully identified 16 out of the total 18 patients who later recovered consciousness in the two testing datasets . In parallel , the specificity across the two testing datasets was up to 92% and 83% , respectively . Taken together , these results suggest that our method can precisely identify those patients with a high-potential for recovery consciousness and concurrently reduce false positives in predicting low-potential patients . In addition , although it has been widely considered that the prospect of a clinically meaningful recovery is unrealistic for prolonged DOC patients ( Wijdicks and Cranford , 2005 ) , our model correctly predicted 9 out of 10 DOC patients with longer than or equal to three months duration of DOC who subsequently recovered consciousness , including three patients with DOC longer or equal to 6 months duration , suggesting that it can also be applied to prolonged DOC patients . According to the surviving consciousness level , DOC patients can be classified into distinct diagnostic entities , including VS/UWS and MCS . As MCS is often viewed as a state with a potentially more favorable outcome ( Luauté et al . , 2010 ) , a misdiagnosis of VS/UWS could heavily bias the judgment of the prognosis , the medical treatment options and even the associated ethical decisions . Some influential studies have found that a few VS/UWS patients exhibit near-normal high-level cortical activation in response to certain stimuli or commands ( Owen et al . , 2006; Monti et al . , 2010 ) , and that late recovery of responsiveness and consciousness is not exceptional in patients with VS/UWS ( Estraneo et al . , 2010 ) . Instead of predicting diagnosis , this study used one-year outcome as a target for regression and classification . Our method , which is based on the combined use of clinical and neuroimaging data , successfully identified seven out of the eight VS/UWS patients in the testing dataset who were initially diagnosed as VS/UWS but subsequently achieved a promising outcome . By analyzing the sMC F-value for each predictor in the regression model , we investigated the prognostic effects of these predictors . In particular , the sMC F-value of the incidence age was greater than that of the other clinical characteristics , suggesting that incidence age might be the most important predictor among the clinical characteristics . Notably , the sMC F-value for the imaging features as a whole seemed to be greater than that for the clinical features , as shown in Figure 4B . We therefore speculate that the patient's residual consciousness capacity , indicated by brain networks and functional connectivity detected from resting state fMRI , might have a stronger prognostic effect than their clinical characteristics . Some previous studies have shown that the resting state functional connectivity within the default mode network is decreased in severely brain-damaged patients , in proportion to their degree of consciousness impairment , from locked-in syndrome to minimally conscious , vegetative and coma patients ( Vanhaudenhuyse et al . , 2010 ) . Moreover , the reduced functional connectivity within the default mode network , specifically between the MPFC and the PCC , may predict the outcome for DOC patients ( Silva et al . , 2015 ) . Our model also validates that the aMPFC and PCC in the default mode network play important roles in predicting prognosis . Above all , we found that the functional connectivity between the aMPFC and the DMPFC had the maximum sMC F-value in the prognostic regression model . The aMPFC is one of the core brain areas in the default mode network , whereas the DMPFC is located in the executive control network . Previous studies have demonstrated that this functional connectivity is negative connectivity in normal healthy populations , with the anti-correlation being proposed as one reflection of the intrinsic functional organization of the human brain ( Fox et al . , 2005 ) . The default mode network directly contributes to internally generated stimulus-independent thoughts , self-monitoring , and baseline monitoring of the external world , while the executive control network supports active exploration of the external world . Correct communication and coordination between these two intrinsic anti-correlated networks is thought to be very important for optimal information integration and cognitive functioning ( Buckner et al . , 2008 ) . A recent study reported that negative functional connectivities between the default mode network and the task-positive network were only observed in patients who recovered consciousness and in healthy controls , whereas positive values were obtained in patients with impaired consciousness ( Di Perri et al . , 2016 ) . Further , our regression model suggests that the anti-correlations between these two diametrically opposed networks ( i . e . the default mode network and the executive control network ) should be the most crucial imaging feature for predicting the outcomes of the DOC patients . We therefore conclude that our prognostic model has good interpretability , and that it not only verifies the findings of previous studies but also provides a window to assess the relative significance of various predictors for the prognosis of DOC patients . This study involved two testing datasets , which were found to be quite different , as shown in Table 1 . First , the distributions of the etiology of the patients were remarkably different in the two datasets . The numbers of patients suffering from trauma/stroke/anoxia were 12/6/7 and 8/0/16 in the ‘Beijing HDxt’ and ‘Guangzhou HDxt’ datasets , respectively . The outcomes were also different . In the ‘Beijing HDxt’ dataset , 12 out of the total 25 patients recovered consciousness , compared with six out of the total 24 patients in the ‘Guangzhou HDxt’ dataset . Given that the characteristics of the two medical centers and their roles in the local health care system are different , we speculate that this could be one of the main reasons why the enrolled patient populations were heterogeneous . As described in the Introduction , DOC can have many causes and can be associated with several neuropathological processes and different severities , leading to the suggestion that information from different domains should be integrated to improve diagnosis and prognostication ( Bernat , 2016 ) . Our study demonstrates that the combination of imaging and clinical features can achieve a better performance than the use of single domain features . However , some caution is warranted . First , although this study involved 112 DOC patients , the number of patients who would later recover consciousness was relatively low ( i . e . 31/112 ) . So , a much larger cohort will be needed for further validation . Second , the PPVs for the two testing datasets were remarkably different , with that for the ‘Guangzhou HDxt’ dataset being relatively low ( 67% versus 91% ) . Although our method predicted that nine patients in this dataset would recover consciousness , only six of them did ( i . e . GOS ≥3 ) , with the other three remaining unconscious at the end of the follow-up ( i . e . GOS ≤2 ) . Given that many additional factors are associated with the outcome of DOC patients , including medical complications , nutrition and so on , future work will need to integrate more information in order to provide more precise predictions . Third , the signal-to-noise ratio of resting state fMRI can influence functional connectivity analysis , so calibrations will be necessary when applying the predictive model across different sites , including standardizing the MRI acquisition protocols ( e . g . scanner hardware , imaging protocols and acquisition sequences ) and the patients' management strategies ( see Appendix 10 for more information ) . Finally , a DOC patient’s prognosis can be considered according to at least three dimensions: survival/mortality , recovery of consciousness , and functional recovery . This study focused on predicting recovery of consciousness , and the patients who died during the follow-up were excluded . In the future , we will consider more outcome assessments , including survival/mortality and functional recovery . In summary , our prognostic model , which combines resting state fMRI with clinical characteristics , is proposed to predict the one-year outcome of DOC patients at an individual level . The average accuracy of classifying a patient as ‘consciousness recovery’ or not was around 88% in the training dataset and two unseen testing datasets . Our model also has good interpretability , thereby providing a window to reassure physicians and scientists about the significance of different predictors , including brain networks , functional connectivities and clinical characteristics . Together , these advantages could offer an objective prognosis for DOC patients that will optimize their management and deepen our understanding of brain function during unconsciousness . | Severe brain injury can lead to disorders of consciousness ( DOC ) , such as a coma . Some patients regain consciousness after injury , while others do not . Those who do not recover are unable to communicate or move in purposeful ways , and need long-term care . It can be difficult for physicians to predict which patients will mend . This is mainly based on their observations of the patient’s behavior over time . But such perceptions are subjective and vulnerable to errors . More accurate and objective methods are needed . Several studies suggest that the cause of the injury , the age of the person at the time of injury , and how long the person has had a DOC may predict recovery . Recent studies have shown that using a brain-imaging tool called resting state functional magnetic resonance imaging ( fMRI ) to measure communication between different parts of the brain may help to calculate the likelihood of recovery . Now , Song , Yang et al . show that combining resting state fMRI with three pieces of clinical information may help to better predict who will improve . Song et al . created a computer model that forecasts recovery from DOC based on fMRI results , the cause of the person’s injury , their age at the time of injury , and how long they have had impaired consciousness . The model could tell which patients would regain consciousness 88% of the time for 112 patients from two medical centers . It also identified several patients who got better despite initial predictions from doctors that they would not . The experiments show that combining multiple types of information can better predict which patients with DOC will convalesce . Larger studies are needed to confirm that the computer model is reliable . If they do , the model may one day help physicians and families to better plan and manage patients’ care . | [
"Abstract",
"Introduction",
"Materials",
"and",
"methods",
"Results",
"Discussion"
] | [
"medicine",
"neuroscience"
] | 2018 | Prognostication of chronic disorders of consciousness using brain functional networks and clinical characteristics |
Viruses like influenza are infamous for their ability to adapt to new hosts . Retrospective studies of natural zoonoses and passaging in the lab have identified a modest number of host-adaptive mutations . However , it is unclear if these mutations represent all ways that influenza can adapt to a new host . Here we take a prospective approach to this question by completely mapping amino-acid mutations to the avian influenza virus polymerase protein PB2 that enhance growth in human cells . We identify numerous previously uncharacterized human-adaptive mutations . These mutations cluster on PB2’s surface , highlighting potential interfaces with host factors . Some previously uncharacterized adaptive mutations occur in avian-to-human transmission of H7N9 influenza , showing their importance for natural virus evolution . But other adaptive mutations do not occur in nature because they are inaccessible via single-nucleotide mutations . Overall , our work shows how selection at key molecular surfaces combines with evolutionary accessibility to shape viral host adaptation .
Viruses are exquisitely adapted to interact with host-cell machinery to facilitate their replication . Despite significant differences in this machinery across host species , some viruses like influenza can evolve to infect divergent hosts ( Parrish et al . , 2008; Webster et al . , 1992 ) . Such zoonotic transmissions can have severe public health consequences: transmission of influenza virus from birds or pigs to humans has resulted in four pandemics over the last century ( Taubenberger and Kash , 2010 ) . These pandemics require the virus to adapt to the new host ( Long et al . , 2019 ) . Delineating how viruses adapt to new hosts will aid in our ability to understand what determines if a chance zoonotic infection evolves into a human pandemic . One critical determinant of influenza host range is the viral polymerase ( Long et al . , 2019 ) , which transcribes and replicates the viral genome ( Eisfeld et al . , 2015; te Velthuis and Fodor , 2016 ) . Avian influenza polymerases perform poorly in mammalian cells ( Cauldwell et al . , 2014; Mänz et al . , 2012; Massin et al . , 2001; Naffakh et al . , 2000 ) . This host range restriction likely arises from the need for the viral polymerase to interact with host proteins such as importin-α ( Resa-Infante and Gabriel , 2013 ) and ANP32A ( Long et al . , 2016 ) , which differ between avian and mammalian hosts . However , it remains unclear exactly how the molecular interfaces between the polymerase and these host proteins are altered during adaptation to humans ( Long et al . , 2019 ) . Studies of natural zoonoses and experimental passaging of viruses in the lab have identified a number of mutations that adapt avian influenza polymerases to mammalian hosts ( Bussey et al . , 2010; Cauldwell et al . , 2014; Cauldwell et al . , 2013; Chen et al . , 2006; Finkelstein et al . , 2007; Gabriel et al . , 2005; Hu et al . , 2017; Hu et al . , 2014; Kim et al . , 2010; Mänz et al . , 2016; Mehle and Doudna , 2009; Miotto et al . , 2008; Mok et al . , 2011; Naffakh et al . , 2000; Reperant et al . , 2012; Taft et al . , 2015; Tamuri et al . , 2009; Yamada et al . , 2010; Zhou et al . , 2011 ) . The best known of these mutations is E627K in the PB2 subunit of the polymerase ( Subbarao et al . , 1993 ) . This mutation alone significantly improves avian influenza polymerase activity in mammalian cells ( Long et al . , 2013; Massin et al . , 2001; Mehle and Doudna , 2008; Naffakh et al . , 2000 ) , and was considered a key step in adaptation to humans ( Taubenberger et al . , 2005 ) . But surprisingly , the recent 2009 H1N1 pandemic lineage lacks the E627K mutation . Instead , it has acquired mutations to PB2 at sites 590 and 591 that similarly confer improved polymerase activity ( Mehle and Doudna , 2009; Yamada et al . , 2010 ) . This fact underscores the possibility that natural evolution has explored only a small fraction of the possible host-adaptation mutations . Examining only the currently available instances of adaptation in nature or the lab may therefore overlook additional mechanisms of adaptation and evolutionary paths to future zoonoses . Here , we map all single amino-acid mutations to an avian influenza PB2 protein that enhance growth in human cells versus avian cells . We do so by leveraging deep mutational scanning ( Boucher et al . , 2014; Fowler and Fields , 2014 ) , which previously has only been used to measure the functional effects of mutations to several influenza proteins in mammalian cells ( Ashenberg et al . , 2017; Bloom , 2014; Doud and Bloom , 2016; Du et al . , 2018; Jiang et al . , 2016; Lee et al . , 2018; Wu et al . , 2014a; Wu et al . , 2014b ) . We show that comparative deep mutational scanning in human versus avian cells identifies numerous human-adaptive mutations that have never before been described . These mutations cluster on the surface of the PB2 protein , highlighting potential interfaces with host factors . Some of these mutations are enriched in avian-human transmission of H7N9 influenza , demonstrating the utility of our experiments for anticipating PB2’s adaptation in nature . The human-adaptive mutations that have not been observed in nature are often inaccessible by single-nucleotide mutations . Overall , our complete map of human-adaptive mutations sheds light on how species-specific selection and evolutionary accessibility shape influenza virus’s evolution to new hosts .
To identify host-adaptation mutations in PB2 , we used deep mutational scanning to measure the effects of all amino-acid mutations to this protein in both human and avian cells . We performed these experiments using the PB2 from an avian influenza strain , A/Green-winged Teal/Ohio/175/1986 ( also previously referred to as S009 ) ( Jagger et al . , 2010; Mehle and Doudna , 2009 ) . The PB2 from this strain is representative of avian influenza PB2s , most of which are highly similar ( average pairwise amino-acid identity of 98 . 7% ) ( Figure 1—figure supplement 1 ) . We mutagenized all codons in PB2 to create three replicate mutant plasmid libraries with an average of 1 . 4 codon substitutions per clone ( Figure 1A , Figure 1—figure supplement 2A–F ) . Since there are 759 residues in PB2 , there are 759 × 19 = 14 , 421 amino acid mutations , virtually all of which are represented in our libraries ( Figure 1—figure supplement 2G ) . We generated a mutant virus library from each of the triplicate plasmid mutant libraries using a helper-virus approach , which reduces bottlenecks during generation of complex viral libraries ( Doud and Bloom , 2016 ) ( Figure 1A ) . For biosafety reasons , we rescued reassortant virus using polymerase ( PB2 , PB1 , PA ) and nucleoprotein ( NP ) genes from the avian influenza strain and the remaining viral genes ( HA , NA , M , NS ) from the lab-adapted A/WSN/1933 ( H1N1 ) mammalian influenza strain . We wanted to minimize selection for host-adaptive mutations during the initial library generation . Therefore , we generated the libraries in human HEK293T cells with a co-transfected protein-expression plasmid encoding the human-adapted PB2-E627K protein variant , so that all cells had a PB2 protein that could complement poorly functioning library variants . To select for functional PB2 variants in human versus avian cells , we passaged each replicate mutant virus library at low MOI in the A549 human lung epithelial carcinoma line and CCL-141 duck embryonic fibroblasts ( Figure 1A , Figure 1—figure supplement 2A ) . To quantify the functional selection on each mutation during viral growth , we deep sequenced the initial plasmid mutant libraries and the passaged mutant viruses to measure the frequency of mutations before and after selection ( Figure 1B , Figure 1—figure supplement 3 ) . All experiments were also performed in parallel on virus carrying wild-type PB2 as a control to quantify the rate of errors arising during sequencing , library preparation , and viral replication ( Figure 1—figure supplement 2H ) . To assess the efficacy of selection without the complication of errors arising from sequencing and passaging , we examined the post-selection frequency of stop and nonsynonymous mutations accessible by >1 nucleotide substitution . Stop and nonsynonymous mutations fell to 2–7% and 26–35% of their initial frequencies respectively ( Figure 1—figure supplement 2H ) . In contrast , synonymous mutations remained at 68–87% of their initial frequency . Therefore , the experiments effectively selected for functional PB2 mutants . We quantified selection at the amino-acid level in terms of the ‘preference’ of each site in the protein for each amino acid ( Figure 1B ) ( Bloom , 2015 ) . The preference for an amino acid is proportional to its enrichment during functional selection . We assessed the reproducibility of our experiments across biological replicates by examining the correlations of preferences for all 14 , 421 amino acid ( Figure 1—figure supplement 2I ) . Biological replicates passaged in each cell type were well correlated ( Pearson’s R in human cells was 0 . 74 to 0 . 79; Pearson’s R in avian cells was 0 . 76 to 0 . 79 ) , and were generally better correlated within cell types than between cell types ( Pearson’s R between cell types was 0 . 67 to 0 . 78 ) . For downstream analyses , we rescaled our preferences to match the stringency of selection in nature ( see Materials and methods , Supplementary file 4 , Figure 2—source data 1 ) . Our experiments reflect known functional constraints on PB2 ( Figure 2A , Figure 2—figure supplement 1 ) . As expected , the start codon shows a strong preference for methionine in both human and avian cells . PB2’s cap-binding function is mediated by a hydrophobic cluster of five phenyalanines ( F404 , F323 , F325 , F330 , F363 ) , H357 , E361 , and K376 ( Guilligay et al . , 2008 ) . Phenylalanines are strongly preferred in the hydrophobic cluster in both host cell types , with the exception of site 323 , which also tolerates aliphatic hydrophobic residues in human cells ( Figure 2A ) . E361 is also strongly preferred in both cell types , as is K376 in the duck cells . A number of other amino acids are tolerated at site 376 in human cells , and at site 357 in both cell types . At site 357 , aromatic residues tyrosine , tryptophan , and phenylalanine are preferred in addition to histidine , consistent with previous observations that the H357W substitution enhances binding to the m7GTP base ( Guilligay et al . , 2008 ) . Finally , the two motifs comprising the C-terminal bipartitite nuclear import signal , 736-KRKR-739 and 752-KRIR-755 ( Tarendeau et al . , 2007 ) , are strongly and similarly preferred in both host cell types . Thus , our experimentally measured preferences largely agree with what is known about PB2 structure and function , and further suggest that functional constraints at these critical sites are similar in both human and avian cells . To more broadly investigate whether functional constraints are similar between both cell types across the entire PB2 protein , we computed the entropy of the amino acid preferences at each site . A larger site entropy indicates a higher tolerance for mutations at that site . Site entropies are well correlated between cell types ( Figure 2B , R = 0 . 78 ) , indicating that sites are usually under similar functional constraint in both cell types . These protein-wide measures of mutational tolerance are also consistent with natural sequence variation . Natural variation is generally low in natural avian sequences , probably because influenza A virus is highly adapted to avian hosts so there is little pressure for additional adaptation . Natural variation is generally higher in natural human sequences , likely because of increased genetic diversity generated as a result of directional selection to adapt to the human host ( dos Reis et al . , 2011 ) , and diversifying selection to escape immune selection on PB2-derived T-cell epitopes ( Assarsson et al . , 2008 ) . Sites that are highly variable among publicly available natural influenza sequences tend to also be ones that we experimentally measured to be mutationally tolerant ( Figure 2C , D ) . To identify mutations that are adaptive in human versus avian cells , we quantified the host-specific effect of each mutation using two different metrics . The first metric , differential selection , quantifies how much a mutation is selected in one condition versus another ( Doud et al . , 2017 ) . Differential selection is computed by taking the logarithm of the relative enrichment of the mutation relative to the wild-type residue in human versus avian cells ( Figure 1C , Figure 3—figure supplement 1A , Figure 2—source data 1 ) . Differential selection greater than zero indicates that a mutation is relatively more favorable in human than avian cells . To test if differential selection accurately identifies host-specific mutations , we asked if a set of 25 previously experimentally verified human- or mammalian-adaptive mutations ( Figure 3—source data 1 ) have differential selection values greater than zero . Indeed , most of these previously characterized mutations had positive differential selection values , as expected for human-adaptive mutations ( Figure 3A ) . In contrast , all other mutations have a distribution of differential selection values centered around zero . However , there are many previously uncharacterized mutations that have differential selection values similar to or greater than those of known human-adaptive mutations ( Figure 3A ) . Of course , differential selection only quantifies the extent to which a mutation is more beneficial in human than avian cells . But importantly , for a mutation to be truly adaptive , it must also be more beneficial than the wild-type amino acid in human cells . To quantify each mutation’s effect relative to wild type in each cell type , we computed the logarithm of the ratio of preferences of the mutant versus wild-type amino acid ( Figure 3B , Figure 2—source data 1 ) . Mutation effect values greater than zero indicate that a mutation is more preferred than the wild-type residue . We identified top experimentally adaptive mutations using both differential selection and mutation effect metrics ( Figure 3B , Figure 3—figure supplement 1B ) . We focused on the 34 mutations most adaptive in our human cell selection ( differential selection >1 . 5 and mutational effect in human cells > 1 ) . Among these 34 mutations , only one ( D701N ) has already been described as human adaptive . The E627K mutation is favored in human cells in our experiments , though it is not in this set of top 34 mutations . However , two other mutations at this site ( E627C and E627S ) are among the top 34 mutations ( Figure 3C ) . S627 is naturally encoded by bat influenza and , in the context of bat influenza polymerase , supports high polymerase activity in human cells ( Tong et al . , 2013; Tong et al . , 2012 ) . In fact , it appears that many mutations at site 627 are human adaptive , with the exception of E627D . These observations are consistent with prior findings that a wide range of amino acid residues can be accommodated at site 627 , and in fact improve polymerase activity in human HEK293T cells over the consensus avian wild-type glutamic acid ( Chin et al . , 2014 ) . We additionally identified 42 mutations as adaptive in avian cells ( differential selection <-2 and mutational effect in avian cells > 1 ) , and seven mutations that are more favorable than the wild-type amino acid in both cell types ( mutational effect >1 in both human and avian cells ) . From these top adaptive mutations identified in our deep mutational scanning , we chose 26 for experimental validation . Specifically , we chose 18 , four , and three mutations adaptive in human , avian , or both cell types , respectively ( Figure 3C , D ) . We prioritized mutations that had consistent measurements across biological replicates . When there were multiple strongly adaptive mutations at a site , we chose just one mutation at that site to test mutations across more sites . Finally , we also validated additional mutations of particular interest , such as the only mutation at site 627 ( E627D ) that appeared to be favored in avian over human cells ( negative differential selection ) , and mutations observed in avian-to-human transmission of H7N9 influenza ( see below for more details ) . The main function of the influenza polymerase is to transcribe and replicate the viral genome . We quantified the effect of mutations on polymerase activity using a minigenome assay which measures transcription of an engineered viral RNA encoding GFP by reconstituted influenza polymerase . To test whether the results of our deep mutational scanning are generalizable to human cells beyond the A549 cell line used in the scanning , we performed the minigenome polymerase activity assay in HEK293T as well as A549 cells . Almost all the putative human-adaptive mutations identified in the deep mutational scanning improved transcriptional activity in human cells relative to the wild type or a synonymous mutant ( Figure 4A , B , Figure 4—source data 1 ) . Mutations that did not improve transcriptional activity retained at least wild-type activity . The effect of mutations in both human cell lines were remarkably consistent , suggesting that our deep mutational scanning yielded results that generalize across human cells . In contrast , all but one of the putative avian-adaptive mutations decreased transcriptional activity in human cells compared to wild type , as expected ( Figure 4A , B ) . The one mutation that did not decrease transcriptional activity , I382D , had comparable activity to wild type . Finally , mutations that are putatively adaptive in both human and avian cells had modestly improved or comparable transcriptional activity in human cells compared to wild type . We also tested the effect of some of the mutations on viral growth to capture any effects of mutations beyond transcriptional activity . We selected some mutations that increased transcriptional activity , and others that had no effect of transcriptional activity . Our rationale in these choices was to determine if mutations that were identified in our screen but did not increase transcriptional activity still resulted in improved viral growth in human over avian cells . For each competition , we infected human ( A549 ) and avian ( CCL-141 ) cells with mutant and wild-type viruses mixed at a 1:1 ratio of transcriptionally active particles as determined by flow cytometry for HA expression in infected cells . We then measured the frequencies of mutant to wild-type virus by deep sequencing . To measure kinetics of viral genome replication from a single round of infection , we infected cells at MOI of 0 . 1 , collected samples at 10 hr post infection , and sequenced vRNA from cellular extract . To measure multi-cycle replication kinetics , we infected cells at MOI of 0 . 01 , collected samples at 48 hr post infection , and sequenced vRNA from the supernatant . At the end of the infection , we calculated the ratio of mutant to wild-type virus in human cells divided by the same ratio in avian cells . Because this quantity is a ratio of ratios , it corrects for any possible deviations from a 1:1 ratio of infectious particles in the initial inoculum . Therefore , ratios greater than one indicates that the mutation confers a relative benefit for viral growth in avian versus human cells . Almost all putative human-adaptive mutations identified in the deep mutational scanning improved growth in human over avian cells , as reflected by an increase in the ratio of mutant to wild type in human cells versus avian cells over the time course of the competition ( Figure 4C , Figure 4—source data 2 ) . Of the putative human-adaptive mutations that did not improve polymerase activity , one mutation , R355G , improved growth in human over avian cells at both 10 and 48 hr post-infection . An additional mutation , R355K , slightly improved growth in human cells by 48 hr post-infection . These two mutations may therefore confer a human-specific growth benefit due to some mechanism other than polymerase activity . As expected , both putative avian-adaptive mutations resulted in poorer growth in human versus avian cells . Finally , the N82K mutation that is putatively adaptive in both human and avian cells resulted in comparable growth in both human and avian cells , as expected . Thus , our deep mutational scanning identified numerous previously undescribed PB2 mutations that improve polymerase activity or viral growth in human cells . In addition , it also identified an intriguing small set of mutations that enhance viral growth but not polymerase activity in human cells . The additional human-adaptive mutations we identified may improve PB2’s ability to interact with important human cell factors . To identify potential interfaces for such interactions , we mapped the sites of top human-adaptive mutations identified in the deep mutational scanning onto the structure of PB2 . Many of the sites cluster in regions of PB2 that may play a role in host adaptation ( Figure 5 , Figure 5—figure supplement 1 ) . Most adaptive mutations are surface exposed in at least one of the two conformations of the polymerase we examined ( Figure 5—figure supplement 1A , relative solvent accessibility >0 . 2 ) . In the transcription pre-initiation form of the polymerase ( PDB: 4WSB ) ( Reich et al . , 2014 ) ( Figure 5A , B ) , the experimentally identified sites 532 and 292 are located on the surface near sites of known human-adaptive sites 286 , 534 , and 535 ( Cauldwell et al . , 2013; Mänz et al . , 2016 ) ( Figure 5B: i ) . Experimentally identified sites 69 and 82 are located on the surface close to sites 64 and 81 ( Figure 5B: ii ) , which are located near the template exit channel and were recently shown to modulate generation of mini viral RNAs that act as innate-immune agonists ( Te Velthuis et al . , 2018 ) . Experimentally identified site 698 is located near known sites 701 and 702 ( Gabriel et al . , 2005 ) ( Figure 5B: iii , Figure 5D: i ) . Experimentally identified mutations are also located in regions not yet identified to be important for host-adaptation: we find a cluster of experimentally identified sites on the surface of the PB2-N terminal domain ( 169 , 176 , 183 , 190 , Figure 5B: iv ) , partially occluded by the flexible PA endonuclease domain in the transcription pre-initiation structure . Experimentally identified sites 163 and 182 are not surface exposed , but are buried right underneath the cluster of four sites . Almost all sites that are not highly surface exposed in the transcription pre-initiation conformation of the polymerase become exposed upon conformational rearrangement of the polymerase from the transcription pre-initiation form to the apo form ( PDB: 5D98 ) ( Hengrung et al . , 2015 ) . The experimentally identified sites 521 and 522 which face the product exit channel , site 355 which faces the core of the polymerase , and site 669 which faces PB1 in the transcription pre-initiation structure ( Figure 5B: v – vii ) , are more fully surface-exposed in the apo structure ( Figure 5C , D: ii – iv ) . However , some sites remain inaccessible: site 156 faces the internal core of the polymerase in both transcription pre-initiation and apo forms ( Figure 5B: viii , Figure 5D: v ) . Therefore , with only a few exceptions , the human-adaptive mutations cluster in patches on the surface of PB2 . Taking a comprehensive approach has therefore allowed us to map surfaces of PB2 that might mediate host-interactions . Next , we asked if host-adaptation mutations occur at known interaction interfaces with host proteins . One known interacting host protein is importin-α , which mediates nuclear import of PB2 and has been proposed to have a role in viral transcription and replication ( Resa-Infante et al . , 2008; Tarendeau et al . , 2007 ) . PB2 of avian viruses uses importin-α3 in human cells , whereas PB2 of mammalian-adapted viruses uses importin-α7 ( Gabriel et al . , 2011; Gabriel et al . , 2008 ) . We mapped total positive differential selection on each site of PB2 ( see legend for Figure 5E ) , and asked how this selection on PB2 relates to its interaction with importin-α . Sites on PB2 that interact with the major and minor NLS-binding surfaces of importin-α ( Pumroy et al . , 2015; Tarendeau et al . , 2007 ) generally have low differential selection , indicating that host-adaptation mutations do not occur at these sites ( Figure 5E; PDB: 4UAD ) ( Pumroy et al . , 2015 ) . This is expected , since all importin-α isoforms share an invariant NLS-binding surface . However , adjacent PB2 sites have higher differential selection ( Figure 5E: i-iii ) . Some of these PB2 sites are in close proximity to regions of importin-α that differ between the α−7 and α−3 isoforms ( Figure 5D: i , iii ) , suggesting that adaptation at these PB2 sites affects importin-α usage . PB2 also interacts with the C-terminal domain of RNA polymerase II , and this interaction is proposed to stabilize the polymerase in the transcription-competent conformation ( PDB: 6F5O ) ( Serna Martin et al . , 2018 ) . Similar to what we observe with importin-α , PB2 sites thought to directly interact with RNA polymerase II tend to have low differential selection , whereas adjacent PB2 sites have higher differential selection ( Figure 5—figure supplement 1D ) . Thus , it appears that host adaptation may involve mutations at sites adjacent to the core residues that directly interact with host proteins . A challenge in the surveillance of non-human influenza and assessment of pandemic risk is determining which of the many mutations that occur during viral evolution are human-adaptive ( Lipsitch et al . , 2016; Russell et al . , 2014 ) . We investigated whether our experimental measurements can identify host-adaptation mutations that occur during the actual transmission of avian influenza to humans . Avian H7N9 influenza viruses have recently caused a large number of sporadic human infections ( Su et al . , 2017 ) . We examined mutations occurring during the evolution of H7N9 viruses that have jumped from avian to human hosts to determine whether they were enriched for changes predicted by our deep mutational scanning to be human adaptive . First , we constructed a phylogeny of H7N9 PB2 sequences , inferred ancestral sequences for all internal nodes , and assigned mutations to specific branches of the phylogenetic tree ( Figure 6A , Figure 6—figure supplement 1–5 ) . We then classified each mutation on the phylogenetic tree as ‘avian’ or ‘human’ based on whether it occurred on a branch connecting two avian isolates , or on a branch leading to a human isolate respectively ( Figure 6—source data 1 ) . As human infections by H7N9 are evolutionary dead-ends , mutations occurring during human infections should appear immediately proximal to human isolates in the phylogeny , while mutations occurring during bird infections will be ancestral . We next asked whether mutations occurring in human hosts had higher differential selection values in our deep mutational scanning than mutations in avian hosts . Indeed , human mutations more often had high differential selection ( a value >0 . 5 ) than avian mutations ( Fisher’s exact test , p=2 . 69e-7 ) ( Figure 6B ) . The H7N9 human mutations with differential selection >0 . 5 include the well-studied human-adaptive mutations 627K and 701N . Indeed , these two mutations make up the majority of H7N9 human mutations with high differential selection . But we also identified a number of other mutations with high differential selection that occurred at least four independent times in jumps of H7N9 influenza into humans: 627V , 534F , 355K , and 521I ( Figure 6A , Figure 6—figure supplement 1–5 , 6B , 3D ) , only one of which has been previously characterized ( E627V , Taft et al . , 2015 ) . A second mutation at site 355 ( 355G ) with high differential selection also occurs during jumps of H7N9 influenza into humans . Thus , our deep mutational scanning identifies both previously characterized and novel mutations that occur in natural avian-to-human transmission of influenza . Our deep mutational scanning identifies many human-adaptive mutations . Why do we not observe all of them in nature ? One possible explanation is that some of these mutations are inaccessible by single nucleotide substitution from existing sequences , and are therefore less likely to arise during natural evolution ( Fragata et al . , 2018 ) . We examined if accessibility by single nucleotide substitution imposes constraint on which human-adaptive mutations arise in nature . To do so , we calculated the mean nucleotide substitutions required to access known human-adaptive mutations from all avian influenza PB2 sequences collected in the past three years . This mean number of nucleotide substitutions can range from less than one ( if the mutation is already present in some avian PB2 sequences ) to three ( if the mutation requires three nucleotide changes from all avian PB2 sequences ) . The majority of previously characterized human-adaptive mutations are accessible by single nucleotide substitutions from avian PB2 sequences ( Figure 7 ) , suggesting that these mutations have already been characterized because they readily occur in the context of current avian influenza viruses . In contrast , most of the top human-adaptive mutations identified in our deep mutational scanning require multiple nucleotide substitutions from current avian PB2 sequences ( Figure 7 ) . Therefore , many of the novel human-adaptive mutations uncovered by our experiment have probably not been previously identified because they are evolutionarily inaccessible from current avian influenza sequences . The importance of evolutionary accessibility is especially obvious if we examine the adaptive mutations that actually occur in natural influenza virus evolution . Of the five top human adaptive mutations accessible by single nucleotide substitution ( Figure 7 ) , three have occurred repeatedly during recent transmissions of H7N9 influenza to humans ( 355G , 521I , 701N; see Figure 6A ) . Thus , it may be possible to combine deep mutational scanning measurements of phenotypes with analyses of evolutionary accessibility to anticipate which mutations are most likely to arise in the context of any given starting sequence ( Fragata et al . , 2018 ) .
We have measured how all single amino-acid mutations to an avian influenza PB2 affect viral growth in both human and avian cells . Our results separate the constraints on PB2 into those that are maintained across cells from diverse species versus those that are specific to human or avian cells . The vast majority of sites are under extremely similar constraint in human and avian cells , including at residues already known to be critical for PB2 function . Layered upon this conserved constraint are mutations with host-specific effects . Our approach therefore represents a powerful strategy for mapping viral determinants of cross-species transmission . The viral determinants of influenza host specificity in the PB2 protein have been intensely studied for decades . Earlier studies addressing this question focused on mutations that have fixed during viral adaptation in nature or in the lab ( Bussey et al . , 2010; Cauldwell et al . , 2013; Chen et al . , 2006; Finkelstein et al . , 2007; Gabriel et al . , 2005; Hu et al . , 2017; Hu et al . , 2014; Kim et al . , 2010; Mänz et al . , 2016; Mehle and Doudna , 2009; Miotto et al . , 2008; Mok et al . , 2011; Naffakh et al . , 2000; Tamuri et al . , 2009; Yamada et al . , 2010; Zhou et al . , 2011 ) . Recent studies have sampled more mutations by random mutagenesis at key sites such as positions 627 , 701 , and 702 of PB2 ( Chin et al . , 2014; Chin et al . , 2017 ) , or error prone PCR across the genome ( Taft et al . , 2015 ) . We can now place these previously characterized mutations in the context of a complete map , thus revealing how selection at molecular interfaces and evolutionary accessibility shapes influenza’s adaptation to humans . Examination of our maps of differential selection at known molecular interfaces , such as with importin-α ( Pumroy et al . , 2015; Tarendeau et al . , 2007 ) and RNA polymerase II ( Serna Martin et al . , 2018 ) suggest an interesting hypothesis: host-adaptive mutations are located adjacent to but not at core residues that directly interact at host proteins , suggesting that host adaptation may involve mutations at sites at the periphery of core interactions . Of course , it remains unclear precisely how any of these new human-adaptive mutations act , though initial validations show that most improve polymerase activity in human cells . However , an intriguing set of mutations such as at site 355 improve viral growth but not polymerase activity in human cells . We speculate that the effects of these mutations on viral growth are mediated by other effects of PB2 , such as in modulating the innate-immune response ( Graef et al . , 2010; Te Velthuis et al . , 2018 ) . Even though we have focused our analyses mostly on host-adaptive mutations , our measurements are also useful to studies of PB2 function independent of host adaptation . Our complete maps of amino acid preferences in each cell type provide context for the handful of residues well-known to be critical for PB2 functions . For example , although our data recapitulate most of the known functional constraints on sites previously defined to be critical for mRNA cap-binding ( Guilligay et al . , 2008 ) , they also reveal that some of these sites can tolerate alternate residues . Thus , our complete survey of the mutational constraints at each site complements other functional and structural knowledge . We also show that our comprehensive experimental measurements can identify human-adaptive mutations that occur during avian-to-human transmission of H7N9 influenza . These measurements therefore help address a fundamental challenge in assessing the risk of potential pandemic influenza virus strains: determining which of the many mutations observed during viral surveillance affect whether a virus will be successful in human hosts ( Lipsitch et al . , 2016; Russell et al . , 2014 ) . Our high-throughput approach therefore enables phenotypic measurements to keep pace with the challenge of interpreting the many viral mutations that are observed during genotypic surveillance of virus evolution ( Grubaugh et al . , 2019 ) . However , although some amino-acid mutations we experimentally identified as human-adaptive occur frequently in nature , others have never been observed . In addition , the strengths of differential selection measured in our experiments do not necessarily correspond to observed frequencies of mutations in nature . These apparent inconsistencies highlight the important role of evolutionary accessibility in shaping influenza’s host adaptation . First , most human-adaptive mutations observed in natural influenza evolution are accessible by single nucleotide substitution from current avian genotypes , demonstrating how the architecture of the genetic code impacts viral adaptation ( Fragata et al . , 2018 ) . Second , even among single-nucleotide substitutions , transition mutations are about ten-fold more frequent than transversions during influenza replication ( Bloom , 2014 ) . This might explain why E627K ( which requires a G→A transition ) is much more frequently observed than E627V ( which requires a A→T transversion ) , despite the latter being more strongly selected in human cells in our experiments . Of course , many additional factors not captured by our experiments can impact which adaptive mutations fix in nature . Integrating our complete maps of the effects of amino-acid mutations with other data that sheds light on evolutionary opportunity , such as nucleotide accessibility , mutation rates , transmission bottlenecks , and environmental and epidemiological factors ( Geoghegan et al . , 2016; Geoghegan and Holmes , 2017; Moncla et al . , 2016; Olival et al . , 2017; Peck and Lauring , 2018; Varble et al . , 2014 ) will help us understand how viruses cross species barriers in nature .
HEK293T , MDCK-SIAT1 , and A549 ( ATCC CCL-185 ) cells were maintained in D10 media ( DMEM supplemented with 10% fetal bovine serum , 2 mM L-glutamine , 100 U/ml penicillin , and 100 μg/ml of streptomycin ) . CCL-141 cells ( ATCC CCL-141 ) were maintained in E10 media ( identical to D10 except that EMEM is used in place of DMEM ) . Cells were grown in WSN growth media ( WGM: Opti-MEM supplemented with 0 . 5% FBS , 0 . 3% BSA , 100 µg/ml CaCl2 , 100 U/ml penicillin , and 100 μg/ml of streptomycin ) for viral infections . For an avian cell line , we chose to use a duck rather than a chicken cell line because ducks are natural hosts of influenza that ( unlike chickens ) possess RIG-I , a key innate-immune sensor of influenza ( Barber et al . , 2010 ) . For expansion of helper virus , we generated MDCK-SIAT1 cells expressing S009 PB2-E627K under control of a doxycycline-inducible promoter ( MDCK-SIAT1-tet-S009-PB2-E627K ) using a Sleeping Beauty transposon system ( Kowarz et al . , 2015 ) . Briefly , MDCK-SIAT1 cells were transfected with pSBtet_RP_S009_PB2_E627K and pSB100X transposase vector using Lipofectamine 3000 ( ThermoFisher Scientific , L3000015 ) , and then subject to selection with 1 µg/ml puromycin . At three days post-transfection , we sorted for individual transfected cells expressing mCherry . All subsequent experiments were performed with a clonal expansion of a single transfected cell . A549 cells were authenticated using the ATCC STR profiling service . CCL-141 cells were obtained from ATCC and used without extensive passaging . All cell lines tested negative for mycoplasma at the time they were expanded for either generating helper virus or passaging mutant plasmid libraries . Sequences for plasmids generated in this study are provided in Supplementary file 1 . Avian influenza polymerase plasmids: Original plasmids for PB2 , PB1 , PA , and NP genes from avian influenza strain A/Green-winged Teal/Ohio/175/1986 ( S009 ) were gifts of Jeffrey Taubenberger ( Jagger et al . , 2010 ) . For generating the mutant plasmid library , we cloned the S009 PB2 coding sequence into a pHW2000 vector ( Hoffmann et al . , 2000 ) from which we removed the CMV promoter ( final plasmid pHW_noCMV_S009_PB2 ) . The mutant library insert was cloned into the recipient vector pHW_noCMVnoTerm_BsmBI , which lacks the Pol I terminator ( terminator sequence is part of the insert ) . The reason that we generated a pHW plasmid without a CMV promoter is that we were unable to maintain a stable bacterial clone of the S009 PB2 coding sequence on the pHH21 plasmid backbone – we observed frequent deletions in the coding sequence during plasmid propagation , suggesting that the insert on the pHH21 plasmid backbone is toxic to the bacterial host . For generating helper virus and virus for viral competitions , we cloned the S009 PB2 , PB1 , PA , and NP coding sequences into pHW2000 ( pHW_S009_PB2 , pHW_S009_PB1 , pHW_S009_PA , pHW_S009_NP ) . In all cases we used non-coding viral-RNA termini from the respective A/WSN/1933 ( H1N1 ) gene segment . For protein expression and the minigenome assay , we cloned the PB2 , PB1 , PA , and NP coding sequences from S009 into a protein-expression plasmid with a CMV promoter ( HDM_S009_PB2 , HDM_S009_PB1 , HDM_S009_PA , HDM_S009_NP ) . All mutants of PB2 were made by site-directed mutagenesis on the appropriate plasmid backbone . Helper virus plasmids: To generate a PB2 vRNA lacking a functional PB2 protein , we cloned GFP flanked by PB2 sequence into the pHH21 vector ( Neumann et al . , 1999 ) ( pHH_PB2_S009_flank_99_eGFP_100 ) . The flanking non-coding viral-RNA termini are from WSN PB2 , and the coding sequences are from S009 PB2 . The length of flanking sequences , 99 and 100 bases on the 5’ and 3’ end of the PB2 coding sequence respectively , are based on prior experiments analyzing how much terminal sequence is needed for effective genome packaging ( Liang et al . , 2005 ) . We mutated out start codons 5’ to the GFP start site in the mRNA sense . Our helper virus rescue also required the reverse genetics plasmids encoding HA , NA , M , and NS from WSN ( pHW184_HA , pHW186_NA , pHW187_M , pHW188_NS ) ( Hoffmann et al . , 2000 ) . To generate a cell line with doxycycline-inducible expression of S009 PB2 for expansion of PB2-deficient helper virus , we cloned the S009 PB2-E627K coding sequence into the pSBtet vector ( pSBtet_RP_S009_PB2_E627K ) ( Kowarz et al . , 2015 ) . Minigenome assay: In addition to protein expression plasmids described above , we used a pHH-PB1-flank-eGFP reporter ( Bloom et al . , 2010 ) , and pcDNA3 . 1-mCherry as a transfection control . All primer sequences used in this study are provided in Supplementary file 2 . Note that this Excel file has several worksheets giving primers for different aspects of the experiments . We generated all possible codon mutations of the entire PB2 coding sequence using the PCR-based strategy described in Bloom ( 2014 ) with the modifications described in Dingens et al . ( 2017 ) . Briefly , we designed mutagenic primers tiling across the entire coding region ( https://github . com/jbloomlab/CodonTilingPrimers; Bloom and Dingens , 2019; copy archived at https://github . com/elifesciences-publications/CodonTilingPrimers ) . We performed 10 cycles of fragment PCR using the mutagenic primers and end primers flanking the vRNA , followed by 20 cycles of joining PCR using only end primers ( Supplementary file 2: Mutagenesis worksheet ) . We generated three independent libraries starting from mutagenesis of independent bacterial clones . The PB2 variants were cloned into the BsmBI-digested vector pHW_noCMVnoTerm_BsmBI using NEBuilder HiFi DNA Assembly Master Mix ( NEB , E2621S ) , and electroporated into ElectroMAX DH10B competent cells ( Invitrogen , 18290015 ) . We obtained 18–22 million transformants for each replicate library , from which we extracted plasmid by maxiprep . We randomly selected 48 clones for Sanger sequencing to evaluate the library mutation rate ( https://github . com/jbloomlab/SangerMutantLibraryAnalysis; Bloom et al . , 2019; copy archived at https://github . com/elifesciences-publications/SangerMutantLibraryAnalysis ) . ( Figure 1—figure supplement 2A–F ) . We generated mutant virus libraries using the helper-virus approach in Doud and Bloom ( 2016 ) , with modifications . We rescued reassortant virus using polymerase and nucleoprotein genes ( PB2 , PB1 , PA , NP ) from S009 , and remaining genes ( HA , NA , M , NS ) from A/WSN/1933 ( H1N1 ) ( WSN ) . Helper virus: We plated a co-culture of 4 × 105 HEK293T and 0 . 5 × 105 MDCK-SIAT1-tet-S009-PB2-E627K in D10 media per well of a 6-well plate . At 18 hr after seeding cells , we added 1 µg/ml doxycycline to induce PB2 expression . One hour after adding doxycycline , we transfected each well with 250 ng each of pHH_PB2_S009_flank_99_eGFP_100 , HDM_S009_PB2-E627K , pHW_S009_PB1 , pHW_S009_PA , pHW_S009_NP , pHW184_HA , pHW186_NA , pHW187_M , and pHW188_NS using BioT ( Bioland Scientific , B01-01 ) . At four hours post-transfection , we replaced D10 media with WGM supplemented with 1 µg/ml doxycycline . We collected viral supernatant 52 hr post-transfection . To expand the helper virus , we seeded MDCK-SIAT1-tet-S009-PB2-E627K cells in D10 media 4 hr prior to infection at 4 × 106 cells per 15 cm dish . We then infected each dish with 40 µl of fresh viral supernatant , using WGM supplemented with 1 µg/ml doxycycline to induce PB2 expression . At 48 hr post-infection , we collected viral supernatant containing expanded helper virus and clarified the supernatant by centrifugation at 400x g for 4 min . We measured the infectious particle ( IP ) /µl titer of the helper virus by infecting HEK293T cells with a known volume of viral supernatant , and quantifying the number of GFP+ cells by flow cytometry 18 hr post-infection . Mutant virus library rescue: For each mutant plasmid library , we seeded 36 wells of a 6-well dish with 1 × 106 HEK293T cells , and transfected each well 17 hr later with 375 ng each of HDM_S009_PB2-E627K , HDM_S009_PB1 , HDM_S009_PA , HDM_S009_NP , and 500 ng of PB2 mutant plasmid library using BioT . For the wild-type control , we seeded six wells and used pHW_noCMV_S009_PB2 in place of the mutant plasmid library . At 6 hr post-transfection , we infected cells with helper virus at MOI of 1 IP/cell in WGM . At 2 hr post-infection , we replaced the inoculum with fresh WGM . At 20 hr post-infection , we harvested viral supernatants and clarified the supernatant by centrifugation at 400x g for 4 min . Supernatants were titered by TCID50 on MDCK-SIAT1 cells . The titers for the three library replicates and wild-type control were 262 , 68 , 100 , and 1467 TCID50/µl respectively . Passaging of mutant virus libraries: We aimed to passage 1 × 106 TCID50 of each mutant virus library in A549 and CCL-141 cells at MOI of 0 . 01 TCID50/cell as determined in MDCK-SIAT1 cells . Therefore , we aimed to have 1 × 108 total cells each at the time of infection . The day prior to each infection , we seeded between 8 × 107 to 1 × 108 A549 cells in D10 in 4 × 5 layered cell culture flasks ( Corning , 353144 ) , and 8 × 107 to 1 × 108 CCL-141 cells in E10 in 8 × 5 layered cell culture flasks . To estimate the total number of each cell type at the time of each infection , we plated an equivalent density of cells in a T225 flask . Just prior to infection , we counted the number of cells in the T225 flask , and extrapolated the total number of cells to be infected . We calculated , for an MOI of 0 . 01 , the number of TCID50s to be passaged for each of the three library replicates and wild-type control in A549 cells to be 9 . 39 × 105 , 1 . 02 × 106 , 1 . 20 × 106 , and 9 . 30 × 105 respectively , and in CCL-141 cells to be 8 . 46 × 105 , 8 . 92 × 105 , 9 . 54 × 105 , and 8 . 58 × 105 respectively . We infected cells by removing D10 or E10 media , rinsing each flask with PBS , and then adding the calculated amount of virus diluted in WGM . At 3 hr post-infection , we replaced the inoculum with fresh WGM . The low MOI passage is expected to purge any non-replicative virus , including those containing the GFP segment . We confirmed this by our observation that there is limited spread of GFP expression over the course of infection . We harvested viral supernatant 48 hr post-infection , and clarified the supernatant by centrifugation at 400x g for 4 min . We ultracentrifuged clarified viral supernatant at 27 , 000 rpm in a Beckman Coulter SW28 rotor , for 2 hr at 4°C . We resuspended the virus the residual media , then extracted RNA from 280 µl of concentrated virus using the Qiagen RNeasy Mini Kit ( Qiagen , 74104 ) . We titered the concentrated viral supernatant by TCID50 on MDCK-SIAT1 cells to estimate the total TCID50 from which we extracted RNA ( ranged from 2 . 80 × 105 to 8 . 8 × 106 TCID50 ) ; we expect the TCID50 titer to be a lower-bound and underestimate of the total viral variants present in the supernatant , since it measures only infectious virus , whereas we would extract RNA from both infectious and non-infectious virions . We used a barcoded-subamplicon deep sequencing strategy that reduces the sequencing error rate ( Doud and Bloom , 2016 ) , https://jbloomlab . github . io/dms_tools2/bcsubamp . html ) . Briefly , we reverse transcribed the full PB2 vRNA with Accuscript Reverse Transcriptase ( Agilent , 200820 ) ( primer S009-PB2-full-1F , Supplementary file 2: Barcoded subamplicon sequencing worksheet ) , and then PCR amplified the full PB2 vRNA ( primers S009-PB2-full-1F and S009-PB2-full-8R , Supplementary file 2 ) using KOD Host Start Master Mix ( EMD Millipore , 71842 ) , making sure to have amplified from an estimated 1 × 107 cDNA molecules . During this amplification , we observed that the band corresponding to full-length PB2 was generally more intense than the smaller bands likely corresponding to the PB2-GFP gene segment , as well as PB2 deletions , suggesting that full-length PB2 was the most prevalent ( Figure 1—figure supplement 3 ) . We then PCR amplified the PB2 gene in eight subamplicons using primers containing a random barcode to uniquely identify each template cDNA molecule ( Supplementary file 2 ) . Approximately 7 . 5 × 105 uniquely barcoded molecules from each subamplicon library were then amplified by primers that add Illumina sequencing adaptors ( Supplementary file 2 ) . Finally , these libraries were deep sequenced on an Illumina HiSeq 2500 using 2 × 250 bp paired-end reads to a target 3 . 3x coverage per barcode . Deep mutational scanning sequence data was analyzed using dms_tools2 ( https://jbloomlab . github . io/dms_tools2 , version 2 . 3 . 0 ) . The GitHub repository https://github . com/jbloomlab/PB2-DMS ( Soh and Bloom , 2019; copy archived at https://github . com/elifesciences-publications/PB2-DMS ) contains Jupyter notebooks that perform all steps of the analyses and provide detailed step-by-step explanations and plots . The README file explains the organization of the notebooks and other files . HTML renderings of the notebooks are provided in Supplementary file 3 . Processed results on preferences , differential selection , and mutation effect are provided in Figure 2—source data 1 . Rescaling of preferences: We rescaled our preferences to match the stringency of selection in nature . To do so , we first asked how well the preferences measured by the deep mutational scanning in either human or avian cells describes evolution of PB2 in both human and avian hosts . We used the preferences to generate an experimentally informed codon substitution model ( ExpCM ) ( Bloom , 2017; Hilton et al . , 2017 ) , and asked if the ExpCMs described PB2’s natural evolution better than a standard phylogenetic substitution model . The ExpCMs using amino-acid preferences vastly outperformed standard phylogenetic substitution models , suggesting that our experiments do capture some of the natural evolutionary constraint on PB2 ( Supplementary file 1 ) . The ExpCM stringency parameter had a value of 2 . 5 , indicating that natural selection favors the same amino acids as our experiments , but with greater stringency ( Hilton et al . , 2017 ) . We thus rescaled our preferences to match the stringency of selection in nature using the ExpCM stringency parameter , and use these rescaled preferences for all subsequent analyses ( Figure 2—figure supplement 1 , Supplementary file 1 , Figure 2—source data 1 ) . Minigenome assays were performed in biological triplicate ( starting from independent bacterial clones of each PB2 mutant ) in both A549 and HEK293T cells . We seeded 2 . 5 × 104 A549 or HEK293T cells per well of a 96-well plate . Cells were transfected the next day with 10 ng each of HDM_S009_PB2 ( for the respective mutant ) , HDM_S009_PB1 , HDM_S009_PA , HDM_S009_NP , 30 ng of pHH-PB1-flank-eGFP reporter , and 30 ng of pcDNA-mCherry as transfection control , using Lipofectamine 3000 ( A549 ) or BioT ( HEK293T ) . At 22 hr post-transfection , cells were trypsinized and analyzed by flow cytometry . We report minigenome activity as the percent of mCherry-positive cells that are GFP-positive . Mutant virus: We generated mutant virus by reverse genetics using pHW_S009_PB2 ( for the respective mutant ) , pHW_S009_PB1 , pHW_S009_PA , pHW_S009_NP , pHW184_HA , pHW186_NA , pHW187_M , and pHW188_NS . We seeded 5 × 105 HEK293T cells per well of a 6-well plate , and transfected cells the next day with 250 ng each of the 8 pHW plasmids using BioT . At 2 hr post-transfection , we replaced media with WGM . At 40 hr post-infection , we collected viral supernatant . Viruses were titered by measuring the number of transcriptionally active particles as determined by flow cytometry for HA expression in infected A549 and CCL-141 cells . Briefly , we infected A549 and CCL-141 cells with a known volume of viral supernatant , and quantified the number of HA+ cells 9 hr post-infection . Cells were stained for HA using the H17-L19 antibody ( Gerhard et al . , 1981 ) , which reacts with the WSN HA ( Doud et al . , 2017 ) . Competitions: Viral competition assays were performed in biological duplicate ( starting from independent bacterial clones of each PB2 mutant ) . For each competition , A549 and CCL-141 cells were infected with a mixture of wild-type and PB2-mutant virus at a 1:1 ratio of transcriptionally active particles as measured for that cell type . For samples collected at 10 hr post infection , we infected a minimum of 3 . 78 × 105 cells at MOI of 0 . 1 . For samples collected at 48 hr post infection , we infected a minimum of 7 . 62 × 105 cells at MOI of 0 . 01 . At 2 hr post-infection , we replaced either D10 or E10 media with fresh WGM . At 10 hr post-infection , cells infected at MOI of 0 . 1 were lysed in buffer RLT and cellular RNA was extracted using the Qiagen RNeasy Mini Kit . At 48 hr post-infection , we collected viral supernatant from cells infected at MOI of 0 . 01 , and extracted RNA using the QIAamp Viral RNA Mini Kit ( Qiagen , 52904 ) . Sequencing to determine mutant frequency: We reverse transcribed the full length PB2 vRNA from the extracted RNA with SuperScript III ( ThermoFisher Scientific , 18080051 ) ( primer S009-PB2-full-1F , Supplementary file 2: Viral competition ) . For each PB2 mutant , we PCR amplified from the cDNA the region of PB2 centered around that mutated codon site ( Supplementary file 2 ) . This product was then subject to a second PCR using primers that add Illumina sequencing adaptors . Finally , these libraries were deep sequenced on an Illumina MiSeq using 50 bp single-end reads . Computational analyses to quantify mutant versus wild-type frequency are provided in Supplementary file 3 , and at https://github . com/jbloomlab/PB2-DMS . The phylogenetic tree was generated using Nextstrain’s augur pipeline ( Hadfield et al . , 2018 ) , and ancestral state reconstruction and adjustment of branch lengths according to sequence isolation date were performed with TreeTime ( Sagulenko et al . , 2018 ) . Ancestral state reconstruction was only performed for nucleotide states , and was not used to infer ancestral host states . Instead , we inferred host-state transitions associated with each branch of the tree in a way that leveraged the prior knowledge that most H7N9 viruses circulate in avian hosts , and that most human infections arise from direct avian-to-human transmissions ( Su et al . , 2017 ) . Specifically , for each node in the tree , starting from the root , we gathered all tips descending from that node . If that clade included only human sequences , and its parent node also included only human sequences , then the current clade falls within a monophyletic human clade , and the branch leading to it was labeled human-to-human . If the current clade includes only human sequences but the parent node includes non-human sequences , then the branch leading to the clade was labeled avian-to-human . If the current clade includes both human and non-human sequences , then the branch leading to the clade was labeled avian-to-avian . Mutations were classified as human if they occurred on human-to-human and avian-to-human branches , and are classified as avian if they occurred on avian-to-avian branches . Note that since H7N9 human influenza typically results from avian-to-human transmissions , the branches we label as human-to-human in reality likely arise from avian-to-human transmissions , but which we cannot accurately reconstruct due to insufficient sampling of avian sequences . For this reason , human-to-human and avian-to-human branches were grouped together . Further details are provided in Jupyter notebook at https://github . com/jbloomlab/PB2-DMS , or in Supplementary file 3 . We calculated the accessibility , or mean nucleotide substitutions required to access an amino-acid mutation , from avian influenza PB2 sequences collected from 2015 through 2018 . The accessibility of codon c to amino-acid a by single-nucleotide mutations is defined as the minimum number of nucleotide mutations needed to generate any codon for that amino-acid . For a collection of sequences , we calculate the accessibility as the weighted average of the accessibilities of all codons observed at that site in the collection of sequences . Accessibility was calculated using code as documented here ( https://jbloomlab . github . io/dms_tools2/dms_tools2 . utils . html ? highlight=accessibility#dms_tools2 . utils . codonEvolAccessibility ) , in Jupyter notebook at https://github . com/jbloomlab/PB2-DMS , and in Supplementary file 3 . Quantification and statistical analysis was performed in Python and a complete description is available in main text , methods , associated figure legends , and computational Jupyter notebooks . Deep sequencing data have been deposited in the NCBI Sequence Read Archive under BioProject accession number PRJNA511556 . The GitHub repository https://github . com/jbloomlab/PB2-DMS contains Jupyter notebooks that perform all steps of computational analyses and provide detailed step-by-step explanations and plots . The README file explains the organization of the notebooks and other files . HTML renderings of the notebooks are provided in Supplementary file 3 . | Viruses copy themselves by hijacking the cells of an infected host , but this comes with some limitations . Cells from different species have different molecular machinery and so viruses often have to specialize to a narrow group of species . This specialization consists largely of fine-tuning the way that viral proteins interact with host proteins . For instance , in bird flu viruses , a protein known as PB2 does not interact well with the machinery in human cells . Because PB2 proteins form part of the viral polymerase ( the structure that copies the viral genome ) , this prevents bird flu viruses from replicating efficiently in humans . Sometimes however , changes in the PB2 protein allow bird flu viruses to better replicate in humans , potentially leading to deadly flu pandemics . To understand exactly how this happens , researchers have previously used two approaches: examining the changes that have happened in past flu viruses , and monitoring the evolution of bird flu viruses grown in human cells in the lab . However , these approaches can only look at a small number of the many possible genetic changes to the virus . This makes it hard to anticipate the new ways that flu might adapt to human cells in the future . To overcome this problem , Soh et al . systematically created all of the single changes to the bird flu PB2 , altering every element of the protein sequence one-by-one . They then tested which of the changes to PB2 helped the virus grow better in human cells . The modifications that made the viruses thrive were on the surface of the protein , suggesting that they might improve interaction with the cell machinery of the host . Some changes have been found in bird flu viruses that have recently jumped into humans in nature , although fortunately none of these viruses have yet spread widely to cause a pandemic . Many factors affect the evolution of viruses , and their ability to infect new species . Understanding which changes in proteins help these microbes adapt to new hosts is an important element that scientists could consider to assess future risks of pandemics . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"epidemiology",
"and",
"global",
"health",
"microbiology",
"and",
"infectious",
"disease"
] | 2019 | Comprehensive mapping of adaptation of the avian influenza polymerase protein PB2 to humans |
Dedifferentiation of acini to duct-like cells occurs during the physiologic damage response in the pancreas , but this process can be co-opted by oncogenic Kras to drive carcinogenesis . Myeloid cells infiltrate the pancreas during the onset of pancreatic cancer , and promote carcinogenesis . Here , we show that the function of infiltrating myeloid cells is regulated by oncogenic Kras expressed in epithelial cells . In the presence of oncogenic Kras , myeloid cells promote acinar dedifferentiation and carcinogenesis . Upon inactivation of oncogenic Kras , myeloid cells promote re-differentiation of acinar cells , remodeling of the fibrotic stroma and tissue repair . Intriguingly , both aspects of myeloid cell activity depend , at least in part , on activation of EGFR/MAPK signaling , with different subsets of ligands and receptors in different target cells promoting carcinogenesis or repair , respectively . Thus , the cross-talk between epithelial cells and infiltrating myeloid cells determines the balance between tissue repair and carcinogenesis in the pancreas .
The highly specialized epithelial cells in the adult pancreas ( Slack , 1995 ) derive from common progenitors during embryogenesis ( Cano et al . , 2007; Gittes , 2009; Means and Leach , 2001 ) . The cells forming the exocrine pancreas , namely acinar , ductal and centroacinar cells , are believed to constitute the likely origin of pancreatic ductal adenocarcinoma , the most common type of pancreatic cancer and one of the deadliest human malignancies ( for review , see ( Puri and Hebrok , 2010; Rooman and Real , 2012; Stanger and Dor , 2006; Zorn and Wells , 2007 ) . Kras mutations are common in human pancreatic cancer ( Bailey et al . , 2016; Hezel et al . , 2006; Jones et al . , 2008 ) , and are present with high frequency in the precursor lesions to pancreatic cancer known as Pancreatic Intraepithelial Neoplasia ( PanIN ) ( Kanda et al . , 2012; Klimstra and Longnecker , 1994 ) . Genetically engineered mice that express oncogenic Kras in the pancreas develop PanINs that progress to pancreatic cancer ( Aguirre et al . , 2003; Hingorani et al . , 2003; Hingorani et al . , 2005 ) , a process that is accelerated by the introduction of other common mutations , such as mutation or loss of tumor suppressors p53 and Ink4a/ARF ( Aguirre et al . , 2003; Hingorani et al . , 2005 ) . In mice , both ductal and acinar cells can serve as the cell of origin for pancreatic cancer ( De La O et al . , 2008; Guerra et al . , 2007; Habbe et al . , 2008; von Figura et al . , 2014 ) . However , acini are more susceptible to transformation , through a process of dedifferentiation known as acinar-ductal metaplasia ( ADM ) ( Kopp et al . , 2012 ) . ADM is a reversible physiological process that protects the pancreas upon tissue injury , such as acute pancreatitis ( Halbrook et al . , 2017; Houbracken et al . , 2011; Strobel et al . , 2007 ) in part by reducing the production of digestive enzymes . In presence of oncogenic Kras , pancreatitis-induced ADM becomes irreversible and progresses to PanIN lesions ( Carrière et al . , 2009; Guerra et al . , 2007 ) . We have previously described the iKras* mouse model , which allows inducible and reversible expression of oncogenic Kras upon administration of doxycycline ( DOX ) ( Collins et al . , 2012a ) . By inactivating oncogenic Kras at different stages of carcinogenesis , we showed that sustained Kras activity is necessary to maintain ADM as well as PanIN lesions , and inactivation of oncogenic Kras leads to redifferentiation of acinar cells ( Collins et al . , 2012a ) . Kras-driven de-differentiation of the acinar cell compartment is mediated , at least in part , by activation of the Kras effector pathway MAPK/ERK ( Collins et al . , 2014 ) . Conversely , the acinar cell specific transcription factors PTF1A and BHLHA15 , as well as the pancreatic transcription factor PDX1 , protect cellular identity and thus counteract transformation ( Krah et al . , 2015; Roy et al . , 2016; Shi et al . , 2013 ) . Epithelial cells within the pancreas exist in the context of a complex microenvironment that rapidly reacts to tissue damage ( for review see[Ying et al . , 2016] ) . Myeloid cells are an abundant component of the immune infiltrate during the onset of pancreatic carcinogenesis ( Clark et al . , 2007; Stromnes et al . , 2014 ) . Macrophages and other myeloid subsets are required for PanIN formation ( Liou et al . , 2015; Zhang et al . , 2017 ) and might be sufficient to induce ADM ( Liou et al . , 2015; Liou et al . , 2013 ) . Macrophages are similarly important for pancreas regeneration following acute damage , such as experimental loss of acinar cells ( Criscimanna et al . , 2014 ) . While mechanisms regulating acinar cell identity and the regulation of the microenvironment have been addressed separately , we have little understanding of how the cross-talk between different cell types affects these aspects of pancreatic biology . Here , we have set out to study the interactions between pancreatic epithelial cells and infiltrating myeloid cells and determine the effect of oncogenic Kras expression in modulating this interaction .
To investigate the cross-talk between epithelial cells and myeloid cells , we generated iKras*;CD11b-DTR mice ( Figure 1A ) . CD11b-DTR mice express the simian Diphtheria Toxin Receptor gene in myeloid cells thus allowing depletion of these cells at will by administration of Diphtheria Toxin ( DT ) ( Duffield et al . , 2005 ) . To validate myeloid cell depletion in the pancreas , we treated mice with a single dose of Diphtheria Toxin , and the induced acute pancreatitis , a process accompanied by myeloid cell infiltration ( Figure 1—figure supplement 1A ) . Compared to control , DT injection resulted in a 40–45% decrease of pancreas infiltrating CD11b+ cells; we observed similar depletion of macrophages and Myeloid-derived suppressor cells ( MDSCs ) , but little change in the dendritic cell population ( Figure 1—figure supplement 1B ) . We then depleted myeloid cells in oncogenic Kras-expressing pancreata , following formation of low-grade PanINs . In brief , doxycycline was added to the drinking water to induce oncogenic Kras* expression in adult mice . Acute pancreatitis was induced 72 hr later by caerulein administration for two consecutive days to promote PanIN formation as previously described ( Collins et al . , 2012a ) . A subset of the mice was sacrificed 3 weeks later , while the remaining animals were administered DT and harvested either 3 days or 1 week later ( Figure 1B , n = 5–7 mice/cohort ) . Histopathological analysis 3 weeks post caerulein revealed low-grade PanINs and ADM surrounded by fibrotic stroma throughout the pancreas parenchyma both in iKras* and in iKras*-CD11b mice ( Figure 1C ) . DT treatment had no effect on lesion progression in iKras* mice , compared to untreated control . Pancreata from iKras*-CD11b mice harvested 3 days following DT treatment were histologically indistinguishable from control . In contrast , 1 week following myeloid depletion , we observed occasional acini , increased ADM and fewer mucinous lesions and PanINs than in corresponding iKras* tissues ( Figure 1C , quantification in Figure 1D ) . Furthermore , upon myeloid cell depletion , we observed a reduction in MAPK activation in epithelial cells ( as determined by p-ERK1/2 immunostaining ) notwithstanding the continuous presence of oncogenic Kras ( Figure 1E ) . This reduction in MAPK signaling correlated with an increase of acinar differentiation in the tissue , as determined by staining for Basic helix-loop-helix family member a15 ( BHLHA15 , also known as MIST1 ) ( Figure 1F ) and for Amylase , a digestive enzyme ( Figure 1G ) . We also observed co-expression of acinar markers ( BHLHA15 and Amylase ) with the ductal marker CK19 , possibly indicating ongoing re-differentiation of acinar cells ( Figure 1F and G ) . To distinguish between re-differentiation and outgrowth of cells that had escaped recombination , we stained the tissue for EGFP . The Rosa26 locus in iKras* mice expresses rtTa-IRES-EGFP following Cre recombination ( Collins et al . , 2012a ) , thus EGFP expression serves as lineage tracing for cells that have undergone recombination and activated oncogenic Kras in a rtTa-dependent manner . Our results showed that both PanIN/ADM lesions and recovered acinar cells expressed EGFP , thus validating the redifferentiation of acini from low-grade lesions ( Figure 1—figure supplement 1C ) . We also observed a reduction in intracellular mucin , as measured by Periodic Acid–Schiff ( PAS ) staining ( Figure 2A ) . We did not observe changes in apoptosis ( Cleaved Caspase three staining , Figure 2B ) . Immunostaining for the macrophage marker F4/80 confirmed depletion of this cell population in the pancreas ( Figure 2C ) . In parallel with changes in the epithelial compartments , myeloid cell depletion led to changes in the stroma . Although tissue fibrosis was still evident by histology ( Figure 1C ) , we observed reduced expression of Smooth Muscle Actin ( SMA ) , a fibroblast activation marker ( Figure 1—figure supplement 1D and quantification in Figure 1—figure supplement 1E ) . Consistently , the expression of genes for the production of extracellular matrix components , such as Fibronectin 1 ( Fn1 ) , Collagen type I alpha one chain ( Col1a1 ) and Collagen type III alpha one chain ( Col3a1 ) , was reduced ( Figure 2D ) . Sonic hedgehog ( Shh ) , secreted by pancreatic neoplastic cells to activate surrounding fibroblasts ( Bailey et al . , 2008; Yauch et al . , 2008 ) , was similarly reduced upon myeloid cell depletion . Even in presence of oncogenic Kras , ligand-mediated activation of EGFR is required to maintain elevated Kras/MAPK activity ( Ardito et al . , 2012 ) . Given the reduction in MAPK signaling levels , we investigated the expression of EGFR ligands by qRT-PCR . Intriguingly , the expression of the EGFR ligand Heparin-Binding epidermal-growth-factor ( EGF ) -like growth factor ( Hbegf ) −previously shown to promote pancreatic carcinogenesis ( Ardito et al . , 2012; Ray et al . , 2014 ) − decreased upon myeloid cell depletion , suggesting that myeloid cells might be a source of this factor or regulate its expression in other compartments ( Figure 2D ) . We observed a similar pattern for Epiregulin ( Ereg ) , which decreased upon myeloid cell depletion . In contrast , there was no change in three other EGFR ligand genes , Amphiregulin ( Areg ) , Transforming growth factor α ( Tgfα ) and Egf . Immunostaining for the active , phosphorylated form of EGFR ( p-EGFR ) , showed expression in in epithelial cells in the control as well as up to three days following myeloid cell depletion , but virtually complete loss of expression by one week ( Figure 2E ) . Our data support the notion that myeloid cells – either directly or through interaction with other cell types – are required for activation of EGFR/MAPK signaling in epithelial cells , thus promoting carcinogenesis while preventing acinar re-differentiation and tissue repair . In advanced malignancy , myeloid cells promote tumorigenesis by inhibiting CD8+ T cell mediated immune responses ( Mitchem et al . , 2013; Stromnes et al . , 2014; Zhang et al . , 2017; Zhu et al . , 2014 ) , and myeloid cell depletion causes CD8+ T cell mediated epithelial cell death . To determine whether a similar immune suppressive mechanism was at play in early lesions , we depleted CD8+ T cells along with myeloid cells in mice bearing low-grade lesions ( Figure 2—figure supplement 1A ) . CD8+ T cell depletion alone had no effect on PanIN progression in iKras* mice . Conversely , limited acinar cell recovery was induced by myeloid cell depletion ( as described above ) and similarly observed when both CD8+ T cell and myeloid cells were depleted in iKras*;CD11b-DTR mice ( Figure 2—figure supplement 1B ) . Thus , during the early stages of carcinogenesis , suppression of T cell-mediated immune responses does not appear to be the main function of infiltrating m cells . To further investigate the cross-talk between oncogenic Kras expressing epithelial cells and infiltrating myeloid cells , we inactivated oncogenic Kras in PanIN bearing iKras* mice ( Figure 3A , n = 4–7 mice/cohort ) ( Collins et al . , 2012a ) , and harvested pancreata after 3 days , one week or two weeks . We detected abundant macrophages ( CD11b+CD64+F4/80+ ) in pancreata expressing oncogenic Kras as well as 3 days following Kras inactivation , as determined by immunostaining and flow cytometry ( Figure 3B–C ) . The total number of macrophages was significantly lower 1 week following Kras inactivation . We then used a combination of surface markers to measure different subsets of macrophages . In the presence of oncogenic Kras , most infiltrating macrophages were CD11b+CD64+F4/80+CD11c+CD206- , consistent with surface characteristics of tumor associated macrophages ( TAMs ) ( Franklin et al . , 2014;Noy and Pollard , 2014; Pollard , 2004 ) . TAM infiltration decreased following Kras* inactivation , while CD11b+CD64+F4/80+CD206+CD11c- macrophages transiently increased ( Figure 3D ) . We sorted total myeloid cells ( DAPI-EGFP-CD45+CD11b+ ) from iKras* pancreata prior to or after Kras inactivation . In myeloid cells extracted from oncogenic Kras expressing pancreata , we detected elevated expression of Arginase 1 ( Arg1 ) and Chitinase 3-like 3 ( Chil3 ) –also known as Ym1–mediators of the immune response and commonly expressed in TAMs ( Geiger et al . , 2016; Munder et al . , 1998; Raes et al . , 2002 ) ; both markers were downregulated in myeloid cells sorted following Kras inactivation ( Figure 3E ) . Thus , macrophage polarization is regulated by the Kras status of epithelial cells . To determine whether direct interactions between epithelial cells and myeloid cells mediated expression of Arg1 , we used an in vitro indirect co-culture system . iKras* primary cells ( Collins et al . , 2012 ) were cultured in presence or absence of DOX to modulate the expression of oncogenic Kras . Conditioned medium from these cells was then used to culture the mouse macrophage cell line RAW264 . 7 . Analysis of the RNA derived from the macrophages by qRT-PCR revealed that cancer cell conditioned media induced Arg1 expression in macrophages in an oncogenic Kras dependent manner ( Figure 3F ) . Further characterization of myeloid cells extracted from oncogenic Kras expressing pancreata revealed high levels of Hbegf , Tgfβ and Tumor necrosis factor-α ( Tnfα ) ( Figure 3G ) . The expression of these ligands was reduced in myeloid cells isolated following Kras inactivation , while other secreted molecules , such as EGF and Tgfα , did not change ( Figure 3E and G ) . In parallel with these changes in myeloid cells , we observed a change in the receptor subsets expressed in sorted EGFP+ epithelial cells . While Egfr expression was high when oncogenic Kras was expressed , and decreased upon its inactivation , Erbb4 was expressed at a lower level when Kras was active , but increased upon Kras inactivation ( Figure 3—figure supplement 1 ) . Thus , oncogenic Kras expression regulates the specific EGFR receptor expressed in the epithelium , as well as regulating polarization and expression of EGFR ligands in infiltrating myeloid cells through a non-cell autonomous mechanism . During the early stages of carcinogenesis , oncogenic Kras , through activation of MAPK signaling , promotes dedifferentiation of acinar cells to ADM ( Collins et al . , 2014; Halbrook et al . , 2017; Houbracken et al . , 2011; Strobel et al . , 2007 ) . Conversely , acinar re-differentiation occurs upon inactivation of oncogenic Kras ( Collins et al . , 2012a ) . While intrinsic factors are known to regulate acinar redifferentiation ( Krah et al . , 2015; Roy et al . , 2016; Shi et al . , 2013 ) , the role of the microenvironment is less clear . We investigated the functional role of myeloid cells upon Kras inactivation , during the re-differentiation of acinar cells . We activated oncogenic Kras in adult iKras* or iKras*;CD11b-DTR mice and induced acute pancreatitis , to induce widespread PanIN formation . Three weeks later , we inactivated oncogenic Kras by withdrawing DOX and simultaneously treated the mice with DT , and harvested the pancreata 3 days or 1 week later ( see scheme in Figure 4A ) . In DT-treated iKras* mice , inactivation of oncogenic Kras during the early neoplastic stages leads to re-differentiation of acinar cells and remodeling of the extracellular matrix and fibro-inflammatory stroma ( Figure 4B , Top Row ) . In contrast , depletion of myeloid cells severely impaired this process . A week after Kras inactivation , DT-treated iKras*;CD11b-DTR pancreata remained fibrotic with very few acinar units identified by histology ( Figure 4B , Bottom Row , Trichrome staining in Figure 4C and pathological quantification in Figure 4D ) . Furthermore , PAS staining , indicating mucinous ducts and low-grade PanINs , were still present in a subset of the ductal structures ( Figure 4—figure supplement 1A ) . Immunostaining for EGFP , a lineage tracing marker for Cre recombination , showed that the epithelial cells in both iKras* and iKras*;CD11b-DTR mice after Kras* inactivation were derived from PanIN cells that had previously expressed the Kras* transgene ( Figure 4—figure supplement 1B ) . In both iKras* and iKras*;CD11b-DTR mice the ductal marker CK19 was prevalent with Kras* ON . Upon Kras inactivation , iKras* mice presented with a transient phase of co-expression of CK19 and amylase , prior to re-establishment of normal pancreas architecture ( Figure 5A , Top Row ) . In contrast , in iKras*;CD11b-DTR mice , ductal structures with co-expression of CK19 and amylase persisted a week after Kras inactivation ( Figure 5A , Bottom Row ) . The newly recovered acinar cells in iKras* mice were highly proliferative ( Figure 5—figure supplement 1A ) . In contrast , proliferation was low in iKras*;CD11b-DTR mice a week after Kras* inactivation . Conversely , apoptotic cells were rare in iKras* pancreata , but abundant in the epithelial compartment of iKras*;CD11b-DTR pancreata as shown by E-cadherin and cleaved caspase three co-immunostaining ( Figure 5B ) . To determine whether myeloid cell depletion during tissue repair resulted in CD8+ T cell mediated immune responses against epithelial cells , we depleted CD8+ T cells along with myeloid cells upon Kras* inactivation ( Figure 5—figure supplement 2A ) . In iKras* mice , CD8+ T cell depletion had no effects on either stroma remodeling or epithelial cell survival , and tissue remodeling occurred as expected . Interestingly , CD8+ T cell depletion didn’t improve cell survival in iKras*;CD11b-DTR mice ( Figure 5—figure supplement 2B ) . Thus , acinar cell plasticity and survival was regulated by infiltrating myeloid cells , independently from their ability to regulate anti-tumor immune responses . Inactivation of oncogenic Kras in pre-neoplastic iKras* pancreata results in a reduction of epithelial p-ERK expression ( Collins et al . , 2012a ) . Surprisingly , p-ERK downregulation in the epithelium coincided with a transient activation of p-ERK in the stroma ( Figure 5C ) . To positively identify the stromal components expressing p-ERK , we performed a panel of co-immunostaining . SMA , a marker of activated fibroblasts , was rapidly reduced upon oncogenic Kras inactivation ( Collins et al . , 2012a ) ; yet we observed expression of p-ERK in a subset of SMA+ cells ( Figure 5C , yellow arrows ) . We observed extensive co-localization of p-ERK with the fibroblast markers Vimentin and Platelet-derived growth factor receptor β ( PDGFRβ ) ( Figure 5—figure supplement 1B ) . In contrast , F4/80+ macrophages only rarely expressed measurable but low levels of p-ERK ( Figure 5—figure supplement 1B ) . Interestingly , myeloid cell ablation prevented p-ERK up-regulation in the stroma upon Kras inactivation ( Figure 5C ) . These data are consistent with the hypothesis that myeloid cells provide essential factors that activate EGFR/MAPK signaling in stromal fibroblasts . We then examined the expression of EGFR ligands and downstream matrix metalloproteinases ( MMPs ) in our models using qRT-PCR . Egf and Tgfα levels were significantly up-regulated in iKras* pancreata 3 days following Kras* inactivation , whereas the levels of other EGFR ligands Hbegf , Areg and Ereg were high in neoplastic pancreata ( Kras* ON ) and dramatically down-regulated when Kras* was inactivated . Depletion of myeloid cells prevented the increase in Egf and Tgfα upon Kras inactivation ( Figure 6A ) . Among the MMPs we examined , Mmp1 was upregulated in iKras* pancreata 3 days following Kras* inactivation . We observed a similar trend for Mmp2 and Mmp9 while the expression of Mmp12 and Mmp14 did not change . However , their expression was slightly ( but not significantly ) decreased upon myeloid cell depletion ( Figure 6A ) . To identify the source of EGFR ligands and MMPs during Kras inactivation induced tissue repair , we flow sorted myeloid cells ( DAPI-EGFP-CD45+CD11b+ ) , fibroblasts ( DAPI-EGFP-CD45-CD3-CD11b-CD31- ) and epithelial cells ( EGFP+CD45- ) for RNA extraction . qRT-PCR analysis showed that Egf and Tgfα were present in both myeloid cells and fibroblasts , at similar levels independently from the oncogenic Kras status ( Figure 3G and Figure 6B ) . Hbegf , as mentioned earlier , was expressed in a Kras-dependent manner in myeloid cells ( Figure 3G ) , but not expressed in fibroblasts ( data not shown ) . EGFR was expressed in fibroblasts independently of epithelial Kras status ( Figure 6B ) . The EGFR/MAPK pathway regulates expression of ECM degrading enzymes in various types of cells including fibroblasts ( Kajanne et al . , 2007 ) . Accordingly , MMPs expression was detected in both myeloid cells and fibroblasts . In particular , Mmp2 expression in fibroblasts derived from iKras* was up-regulated when Kras* was OFF for 3 days and significantly higher compared to that in fibroblasts derived from iKras*;CD11b-DTR . Further , Mmp9 expression in fibroblasts decreased upon myeloid cell depletion ( Figure 6B ) . Western-blot analysis of the pancreata showed a reduction in overall Mmp2 protein levels , and specifically the active form of the protein , upon myeloid cell depletion ( Figure 6C ) . Interestingly , Western-blot analysis also revealed a decrease in Collagen I levels following Kras* inactivation in iKras* mice but not in iKras*;CD11b-DTR mice , indicating impaired remodeling in the latter ( Figure 6C ) . In addition to myeloid cells , epithelial cells might constitute a source of EGFR ligands . By q-PCR analysis , we detected expression of Egf , Tgfα and Hbegf mRNA in sorted epithelial cells; the expression of Egf was decreased upon myeloid cell depletion while expression of the other ligands was unchanged ( Figure 6—figure supplement 1 ) . Based on our data , myeloid cells might contribute to EGFR ligand levels both by expressing them directly , and by inducing their expression in other cells types ( epithelial , and possibly others ) . To determine whether EGFR/MEK activation in fibroblasts was required for tissue remodeling , we inhibited EGFR or MEK – a key component of MAPK signaling- with Erlotinib and Tramatinib , respectively . First , we let iKras* mice develop low-grade PanINs , as described above . Then , upon inactivation of oncogenic Kras , we treated the animals with the inhibitors or vehicle controls ( Figures 7A and 8A ) . EGFR inhibition ( EGFRi ) blocked MAPK activation in the stroma as measured by reduced p-ERK levels ( Figure 7B ) . Similar to myeloid cell depletion , EGFRi treatment resulted in delayed tissue repair . Abundant stroma was still present at 1 week post Kras* inactivation ( Figure 7B , HE and Trichrome staining ) . Further , Mmp2 and Mmp9 expression was inhibited in EGFRi treated pancreata 3–7 days post Kras* inactivation ( Figure 7C ) . However , acinar re-differentiation was not affected by EGFRi treatment , as shown by co-immunofluorescent staining for CK19 and amylase ( Figure 7B ) . We made similar observations upon MEK inhibition ( MEKi ) upon Kras* inactivation , with reduced ECM degradation and MMPs expression ( Figure 8B and C ) , but unimpaired acinar re-differentiation . Therefore , EGFR-MAPK signaling is required for ECM degradation and remodeling . Conversely , in the epithelium , repression of EGFR/MAPK promoted re-differentiation , consistent with previous studies ( Ardito et al . , 2012; Collins et al . , 2014 ) .
The pancreas is formed by a limited number of progenitor cells and , in the adult , it has a limited ability to regenerate following injury ( Dor et al . , 2004 ) , although it can grow in response to increase need for its exocrine or endocrine function ( Holtz et al . , 2014; Karnik et al . , 2007 ) . The pancreas is highly plastic; in particular , acinar cells can de-differentiate into duct-like cells during a process known as acinar-ductal metaplasia ( ADM ) . While ADM is important during tissue damage such as acute pancreatitis -where it might protect acinar cells from further damage and set the stage for repair- it also leads to a cell type that is susceptible to transformation by oncogenic Kras ( for review see [Morris et al . , 2010; Roy and Hebrok , 2015] ) . Thus , the mechanisms regulating pancreas plasticity are relevant to both damage/repair in this organ and carcinogenesis . ADM is characterized by loss of acinar differentiation and acquisition of a duct-like phenotype which is accompanied by expression of pancreas progenitor markers ( Puri and Hebrok , 2010; Roy and Hebrok , 2015; Stanger and Hebrok , 2013; Storz , 2017 ) . Transcription factors driving acinar differentiation are down-regulated during ADM . BHLHA15 expression is lost during ADM and re-established when ADM re-differentiates to acini . Importantly , BHLHA15 plays a functional role in this process , and while BHLHA15 loss facilitates ADM ( and , consequently , carcinogenesis ) , forced expression of BHLHA15 is protective against both ADM and carcinogenesis ( Shi et al . , 2013 ) . The transcription factor Ptf1a is expressed throughout the pancreatic bud early in development , but it is restricted to acinar cells in the adult organ ( Kawaguchi et al . , 2002 ) . Ptf1a loss facilitates ADM and carcinogenesis ( Krah et al . , 2015 ) . Further , PDX1 , a key determinant of pancreas development expressed at low levels in adult acini , is similarly important to maintain acinar cell identity ( Roy et al . , 2016 ) . Thus , signals intrinsic to epithelial cells regulate the differentiation status of acinar cells . We have , and others , have previously shown that oncogenic Kras induces ADM through activation of MAPK signaling ( Ardito et al . , 2012; Collins et al . , 2014; Collisson et al . , 2012 ) and consequent repression of acinar-specific transcription factors . Conversely , inhibition of MAPK signaling using MEK inhibitors allows re-expression of acinar-specific transcription factors and re-differentiation of acinar cells ( Collins et al . , 2014 ) . Thus , a complex network of intrinsic signals regulates acinar differentiation in the adult pancreas . In our initial characterization of the iKras* mouse model , we investigated the role of oncogenic Kras during very early stages of carcinogenesis . While oncogenic Kras promotes transdifferentiation of acinar cells to acinar-ductal metaplasia , inactivation of oncogenic Kras in ADM or even low-grade PanIN lesions leads to regression of these lesions and re-differentiation of the epithelial compartment to acinar cells ( Collins et al . , 2012a ) . Inactivation of oncogenic Kras also results in profound remodeling of the surrounding fibroinflammatory reaction . Here , we set out to understand the interaction between oncogenic Kras expressing epithelial cells and the surrounding microenvironment . We show that reciprocal interactions between oncogenic Kras expressing epithelial cells and the surrounding microenvironment control pancreatic plasticity ( see working model in Figure 9 ) . First , we determined that Kras expressing epithelial cells alter myeloid cell polarization in the pancreas , inducing expression of Arginase1 , Chil3 and Hbegf . These markers have been previously described in tumor associated macrophages ( TAMs , for review see [Mantovani et al . , 2017] ) . Second , Inactivation of oncogenic Kras led to the loss of Arg1 , Chil3 and Hbegf from myeloid cells . Conversely , a subset of macrophages positive for the surface markers CD206 and CD11c , transiently accumulated in the pancreas , coinciding with the remodeling process . Interestingly , the surface marker expression of this population is consistent with M2 macrophages previously shown to be important in regeneration of pancreatic acini and islets following experimental ablation ( Criscimanna et al . , 2014 ) , and similarly involved in tissue repair in other organs ( for review , see [Wynn and Vannella , 2016] ) . In this study , we show that myeloid cells play an instructive role regulating epithelial cell identity and plasticity . In the presence of oncogenic Kras , myeloid cells are required to maintain dedifferentiation of ADM/low-grade PanIN lesions . Depletion of myeloid cells induces BHLHA15 expression and occasionally expression of the digestive enzyme amylase in low-grade PanINs , notwithstanding expression of oncogenic Kras , thus presumably preventing further progression to malignancy . We show that myeloid cells are required for the expression of EGF ligands , and activation of MAPK signaling in pancreatic epithelial cells . This finding fits with the notion that oncogenic Kras is insufficient to induce a high enough level of MAPK activation to induce transformation ( Daniluk et al . , 2012 ) , thus EGFR ligands are required for carcinogenesis ( Ardito et al . , 2012 ) . In tumor-bearing mice , myeloid cells inhibit CD8+ T cell mediated anti-tumor immune responses , and this function explains their requirement in cancer growth ( Stromnes et al . , 2014; Zhang et al . , 2017 ) . However , both during the progression of early PanIN lesions and during their regression upon Kras inactivation , myeloid cell-requirement was independent from the presence of CD8+ T cells , indicating that they play a function distinct from immune suppression . Upon Kras inactivation , myeloid cells including re-polarized M2 macrophages are required for tissue remodeling . First , we show that myeloid cells are required for epithelial cell re-differentiation and survival . Thus , myeloid cell depletion results in epithelial cell death and persistence of clusters of cells co-expressing acinar and ductal markers . Second , we show that myeloid cells are required for remodeling of the extracellular matrix . To gain mechanistic insight , we investigated the cross-talk between epithelial cells and myeloid cells . We have previously shown that myeloid cells are required to sustain activation of EGFR/MAPK signaling in epithelial cells ( Zhang et al . , 2017 ) . In turn , MAPK signaling is necessary for PanIN formation and progression ( Ardito et al . , 2012; Collins et al . , 2014; Collisson et al . , 2012 ) . Surprisingly , remodeling of the extracellular matrix was also regulated by EGFR/MAPK signaling . Inactivation of oncogenic Kras in the pancreas led to changes of expression of specific EGFR ligands in the pancreas . In presence of active Kras , Hbegf , Areg and Ereg were the main ligands . Upon Kras* inactivation , their expression decreased while that of Egf and Tgfα increased . In parallel , we observed changes in expression of EGFR family receptors . Remarkably , inactivation of oncogenic Kras resulted in transient activation of MAPK signaling in stromal fibroblasts , simultaneous to loss of activation in the epithelium . Consistent with the notion that MAPK signaling in the stroma is important for remodeling , treatment with Erlotinib ( EGFR inhibitor ) or Tramatinib ( MEK inhibitor ) resulted in the persistence of pancreatic fibrosis . The observation that activation of EGFR/MAPK signaling in different cellular compartments might , in turn , favor carcinogenesis or remodeling has potential clinical implications , suggesting that specific inhibition of distinct EGFR ligands or receptors might be preferable to overall inhibition . While our data support the notion that myeloid cells are a source of EGFR ligands , they also support the idea that myeloid cells induce EGFR ligands in other cellular compartments , including epithelial cells; future studies will need to address the role of specific EGFR ligands and their specific cell sources . In summary , in this study we show that the cross-talk between epithelial cells and myeloid cells regulates pancreatic plasticity and fibrosis . Further , we show that this cross-talk is important for pancreatic tissue repair following injury , but can be co-opted , in presence of oncogenic Kras , to promote carcinogenesis . Manipulating this cross-talk to promote repair while inhibiting carcinogenesis should therefore be prioritized in future studies .
iKras*;CD11b-DTR mice were generated by crossing iKras* mice ( ptf1a-Cre;R26-rtTa-IRES-EGFP;TetO-KrasG12D ) ( Collins et al . , 2012a ) with CD11b-DTR mice ( B6 . FVB-Tg ( ITGAM-DTR/EGFP ) 34Lan/J , Jackson Laboratory ) ( Duffield et al . , 2005 ) . Double mutant littermates of iKras* were used as controls . Male and female mice were included equally . All animal studies were conducted in compliance with the guidelines of Institutional Committees on Use and Care of Animals at the University of Michigan . Acute pancreatitis was induced in 4–6 week-old mice by caerulein injection and KrasG12D expression was induced by doxycycline as previously described ( Collins et al . , 2012a ) . Three weeks post pancreatitis induction doxycycline was withdrawn from the drinking water for tissue repair study . Mice were also treated with EGFR inhibitor Erlotinib ( 50 mg/kg , oral gavage , daily ) ( Selleckchem ) , or MEK inhibitor Tramatinib ( GSK1120212 ) ( 1 mg /kg , i . p . daily ) ( Selleckchem ) or vehicle . For myeloid cell depletion , CD11b-DTR and iKras*;CD11b-DTR mice were treated with diphtheria toxin ( DT ) ( 25 ng/g i . p . ) ( Enzo Life Science ) and repeated every 4 days . For CD8+ T cell depletion , anti-CD8 mAb ( BioXcell #BE0061; clone 2 . 43; 200 µg/mouse ) was injected i . p . twice per week . All cells were cultured in IMDM supplemented with 10% FBS and 1% penicillin/streptomycin ( Gibco ) . Primary mouse pancreatic cancer cell line iKras* derived from iKras*p53* ( ptf1a-Cre; TetO-KrasG12D; Rosa26rtTa/+; p53R172H/+ ) tumor ( Collins et al . , 2012 ) was used to generate conditioned medium ( CM ) in presence or absence of Doxycycline at 1 µg/ml ( Sigma ) for 2–3 days . These cells were used at low passage , genotyped for the Kras , Cre and p53 transgenes , and tested negative for mycoplasma . Mouse macrophage cell line RAW264 . 7 ( ATCC Cat# TIB-71 , RRID:CVCL_0493 ) were similarly used at low passage and mycoplasma negative . CM was filtered through 0 . 2 µm filter before use . 1–2 × 105 cells of RAW264 . 7 were plated in 6-well plates overnight and then cultured with CM ( iKras* CM diluted 1:1 in fresh IMDM with 10% FBS ) for 24 hr before harvest for RNA isolation . Hematoxylin and eosin ( H&E ) , Periodic Acid Schiff ( PAS ) , Gomori’s Trichrome , immunohistochemical and immunofluorescent staining were performed on formalin-fixed , paraffin embedded mouse pancreatic tissues as described before ( Zhang et al . , 2013a ) . Antibodies used are listed in Supplementary file 1 . For immunofluorescence , Alexa Fluor ( Invitrogen ) secondary antibodies were used . Cell nuclei were counterstained with Prolong Gold with DAPI ( Invitrogen ) . Images were taken with Olympus BX-51 microscope , Olympus DP71 digital camera , and DP Controller software . The immunofluorescent images were acquired using the Olympus IX-71 confocal microscope and FluoView FV500/IX software . For histopathological analysis , five non-overlapping H&E images ( 20x objective ) per slide were examined by a pathologist ( W . Y . ) as described before ( Zhang et al . , 2013a ) . Image-Pro Plus 4 . 5 was used to measure the percentage of positive area of immunofluorescent staining per high power field image . Three samples per group , and 4–6 images per sample were analyzed . Western blotting was conducted as previously described ( Collins et al . , 2012a ) , and Collagen I was detected under non-reduced and non-denatured condition . Antibody information is included in Supplementary file 1 . Single-cell suspensions of fresh spleen or pancreas were prepared as previously described ( Zhang et al . , 2013b ) and stained with fluorescently conjugated antibodies listed in Supplementary file 1 . Flow cytometric analysis was performed on a Cyan ADP analyzer ( Beckman Coulter ) and data were analyzed with Summit 4 . 3 software . Cell sorting was performed using a MoFlo Astrio ( Beckman Coulter ) . Myeloid cells ( DAPI-EGFP-CD45+CD11b+ ) , epithelial cells ( DAPI-EGFP+CD45- ) and fibroblasts ( DAPI-EGFP-CD45-CD11b-CD31-CD3- ) were sorted and lysed in RLT buffer ( Qiagen ) . Total RNA was prepared using RNeasy ( Qiagen ) and reverse-transcripted using High Capacity cDNA Reverse Transcription kit ( Applied Biosystems ) . Samples for quantitative PCR were prepared with 1X SYBR Green PCR Master Mix ( Applied Biosystems ) and various primers ( primer sequences are listed in Supplementary file 2 ) . All primers were optimized for amplification under reaction conditions as follows: 95°C 10mins , followed by 40 cycles of 95°C 15 secs and 60°C 1 min . Melt curve analysis was performed for all samples after the completion of the amplification protocol . Cyclophilin A was used as the housekeeping gene expression control . Graphpad Prism six software was used for all statistical analysis . All data were presented as means ± standard error ( SEM ) . Intergroup comparisons were performed using Two-tailed unpaired t-test , and p<0 . 05 was considered statistically significant . | The pancreas contains many types of highly specialized cells . For example , the acinar cells produce enzymes that help to digest food , and the ductal cells build the ducts to transport these enzymes to the gut . When the pancreas gets injured , the acinar cells start to transform into duct-like cells . The cells can revert to normal acinar cells once the tissue has repaired itself . However , when a protein named Kras becomes faulty , the transformed acinar cells can no longer revert to normal ones . Scientists believe that is one of the first signs of pancreatic cancer , as mutated Kras proteins are very common in this disease . Injury and cancer both attract immune cells to the pancreas , including a type called myeloid cells . However , until now it was not known how myeloid cells and acinar cells interact . In 2012 , scientist showed that when the faulty Kras protein is removed , the tissue of a damaged pancreas can repair itself again . To investigate this further , Zhang et al . – including some of the researchers involved in the 2012 work – created genetically modified mice in which the faulty Kras protein could be experimentally activated or deactivated . The results showed that when Kras was activated , the myeloid cells helped the transformed acinar cells to develop into cancer cells . When Kras was inactivated , myeloid cells helped to repair the damaged tissue . Moreover , myeloid cells used similar molecular signals to either activate the tissue repair or to stimulate the cells to turn into cancer cells . At the moment , pancreatic cancer cannot be cured . A better understanding of how this disease develops may help scientists develop new treatments . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cancer",
"biology"
] | 2017 | Epithelial-Myeloid cell crosstalk regulates acinar cell plasticity and pancreatic remodeling in mice |
Appropriate physiological signaling by primary cilia depends on the specific targeting of particular receptors to the ciliary membrane , but how this occurs remains poorly understood . In this study , we show that D1-type dopaminergic receptors are delivered to cilia from the extra-ciliary plasma membrane by a mechanism requiring the receptor cytoplasmic tail , the intraflagellar transport complex-B ( IFT-B ) , and ciliary kinesin KIF17 . This targeting mechanism critically depends on Rab23 , a small guanine nucleotide binding protein that has important effects on physiological signaling from cilia but was not known previously to be essential for ciliary delivery of any cargo . Depleting Rab23 prevents dopamine receptors from accessing the ciliary membrane . Conversely , fusion of Rab23 to a non-ciliary receptor is sufficient to drive robust , nucleotide-dependent mis-localization to the ciliary membrane . Dopamine receptors thus reveal a previously unrecognized mechanism of ciliary receptor targeting and functional role of Rab23 in promoting this process .
Primary cilia are microtubule-based protrusions of the plasma membrane that support a wide range of specialized receptor-mediated signaling functions . Physiological signaling from cilia critically depends on the selectivity of receptor targeting to the ciliary membrane , and disturbances in this targeting are thought to underlie a variety of pathological conditions ( Hsiao et al . , 2012 ) . The remarkable specificity of ciliary membrane targeting is clear among G protein-coupled receptors ( GPCRs ) . Some members of this large receptor family robustly accumulate in the ciliary membrane while others , including closely related homologues , are found throughout the extra-ciliary plasma membrane but are effectively excluded from cilia ( Schulz et al . , 2000; Marley and von Zastrow , 2010 ) . Understanding how particular GPCRs localize to primary cilia with such exquisite selectivity is a fundamental problem with broad physiological significance ( Emmer et al . , 2010 ) . The ciliary membrane compartment is separated from the surrounding extra-ciliary plasma membrane by a transition zone complex that impedes lateral exchange of membrane proteins ( Gilula and Satir , 1972; Hu et al . , 2010; Chih et al . , 2011; Williams et al . , 2011 ) . This can explain how GPCRs are retained in cilia once delivered , but not how they are delivered in the first place . Two basic routes of ciliary membrane delivery have been described: first , receptors can originate from an intracellular source , through fusion of post-Golgi transport vesicles with the ciliary membrane in or near the transition zone . A number of membrane proteins are targeted to cilia by this route , and molecular machineries supporting it have been identified ( Deretic and Papermaster , 1991; Geng et al . , 2006; Mazelova et al . , 2009 ) . Second , receptors can originate from the extra-ciliary plasma membrane . This route , first described in a seminal study of flagellar agglutinins in Chalmydomonas ( Hunnicutt et al . , 1990 ) , contributes to ciliary targeting of the atypical seven-transmembrane protein Smoothened ( Smo; Milenkovic et al . , 2009 ) in mammalian cells . Is the lateral delivery route relevant to ciliary localization of conventional GPCRs ? Molecular mechanisms that underlie specific ciliary delivery pathways also remain incompletely understood . A number of proteins are already known to play a role , including the BBSome ( Nachury et al . , 2007; Berbari et al . , 2008b; Jin et al . , 2010 ) , Tulp3 ( Mukhopadhyay et al . , 2010 , 2013 ) , Arf4 ( Deretic et al . , 2005 ) , ASAP1 ( Wang et al . , 2012 ) , and intraflagellar transport ( IFT ) -B and IFT-A ( Mukhopadhyay et al . , 2010; Keady et al . , 2011 , 2012; Crouse et al . , 2014; Kuzhandaivel et al . , 2014 ) . Are there additional machineries not yet identified that function in targeting specific GPCRs to cilia ? We addressed these questions through study of the D1-type dopamine receptor ( D1R ) , a conventional GPCR that robustly localizes to cilia in diverse cell types ( Marley and von Zastrow , 2010; Domire et al . , 2011 ) . Here , we show that D1Rs are delivered to the cilium from the extra-ciliary plasma membrane . Further , we show that the D1R cytoplasmic tail is both necessary and sufficient to direct receptor targeting to the ciliary membrane , and this requires a distinct set of cellular proteins including the anterograde IFT-B complex and ciliary kinesin , KIF17 . Moreover , we identify an essential role of the small GTP-binding protein , Rab23 , in the ciliary targeting mechanism . Rab23 is not only necessary for D1R access to cilia , it is also sufficient to drive strong ciliary localization of a non-ciliary GPCR . D1Rs thus reveal a discrete route and mechanism of ciliary GPCR targeting in which Rab23 plays an unprecedented and essential role .
The D1R is a cilia-localized GPCR whose mechanism of targeting to the cilium is poorly understood ( Marley and von Zastrow , 2010; Domire et al . , 2011; Zhang et al . , 2013 ) . We investigated this question using recombinant receptors expressed in inner medullary collecting duct ( IMCD3 ) cells . Using an N-terminal Flag tag on the D1R to label the overall surface pool , D1Rs were visualized throughout the plasma membrane and highly enriched in cilia marked by acetylated tubulin ( AcTub ) ( Figure 1A ) , like the cilia-localized somatostatin-3 receptor ( SSTR3 ) ( Figure 1B; Händel et al . , 1999; Schulz et al . , 2000; Berbari et al . , 2008a ) . In contrast , the delta opioid peptide receptor ( DOP-R or DOR ) localized throughout the extra-ciliary plasma membrane but was not detectable on cilia ( Figure 1C ) . 10 . 7554/eLife . 06996 . 003Figure 1 . D1Rs specifically localize to primary cilia . ( A–C ) Representative epifluorescence microscopy images of Flag-D1R ( panel A ) , Flag-SSTR3 ( panel B ) , and Flag-DOR ( panel C ) localization on the surface of inner medullary collecting duct ( IMCD3 ) cells . Insets show a cropped region of the plasma membrane containing the cilium , with Flag immunoreactivity marking receptor ( top ) and acetylated tubulin ( AcTub ) immunoreactivity marking the cilium ( middle ) . Merged view is at bottom with Flag in green and AcTub in red . Flag-D1R and Flag-SSTR3 localize robustly to cilia , while Flag-DOR is detectable in the extra-ciliary plasma membrane but not on cilia . ( D ) Quantification of ciliary localization by determining the fraction of receptor ( Flag ) -positive cilia , judged by the presence of Flag immunoreactivity visible by epifluorescence microscopy , and expressed as a percentage of total cilia counted in the transfected cell population . ( E ) Scheme for quantification of ciliary localization by determining enrichment of receptor ( Flag ) signal in an ROI containing the cilium , when compared to an adjacent region of the extra-ciliary plasma membrane . Representative ROIs are shown for a Flag-D1R-transfected cell . ( F ) Fold-enrichment calculated as a ratio of background-subtracted Flag signal present in the ciliary ROI divided by background-subtracted Flag signal present in the adjacent extra-ciliary plasma membrane ROI ( cilia/PM ) . Error bars represent SEM from n = 3 independent experiments , with 10–15 cilia analyzed for each receptor in each experiment . ( *** ) p < 0 . 001 . Scale bars , 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 06996 . 003 We first quantified ciliary localization by counting the number of receptor-expressing cells with visible receptor immunoreactivity on the cilium . This normative metric verified ubiquitous D1R localization to cilia , similar to SSTR3 , and high specificity of ciliary localization relative to DOR ( Figure 1D ) . Second , because cilia scored as receptor-positive varied in degree of apparent receptor concentration , we determined average fold-enrichment of receptors on the cilium relative to the extra-ciliary plasma membrane ( Figure 1E ) . This graded metric further verified robust ciliary localization of the D1R and SSTR3 ( but not DOR ) and indicated that the D1R is enriched on cilia even more strongly than the SSTR3 ( Figure 1F ) . In principle , D1Rs could be concentrated on cilia relative to the extra-ciliary plasma membrane by immobilization or by a diffusion barrier at the base of the cilium ( Huang et al . , 2007; Hu et al . , 2010; Francis et al . , 2011 ) . To distinguish these possibilities , we fused the D1R to photoactivatable green fluorescent protein ( Flag-D1-PAGFP ) and investigated mobility by live cell imaging coupled to local photoactivation ( Figure 2A ) . We verified appropriate ciliary localization of engineered receptors by anti-Flag Alexa555 surface labeling ( Figure 2B , top row ) . Locally photoactivated D1Rs were distributed non-uniformly on cilia immediately after the 405-nm illumination pulse and then equilibrated throughout the cilium within seconds ( Figure 2B , C; whole-cell images shown in Figure 2—figure supplement 1 ) . However , there was no visible spread of labeled receptors outside of the cilium on this time scale . Consistent with this , total PA-GFP fluorescence intensity integrated over the ciliary length remained unchanged throughout an 80-s interval after local photoactivation ( Figure 2D ) . Further , a robust fluorescence signal representing photoactivated D1Rs was still visible when the same cilia were re-imaged minutes thereafter ( Figure 2E; whole-cell images shown in Figure 2—figure supplement 2 ) . Together , these observations indicate that ciliary D1Rs are laterally mobile in the ciliary membrane compartment but restricted from freely diffusing into the extra-ciliary plasma membrane . 10 . 7554/eLife . 06996 . 004Figure 2 . D1Rs are mobile in the ciliary membrane and delivered from the extra-ciliary plasma membrane . ( A ) Schematic for local labeling of D1-type dopamine receptor ( D1R ) in a cilium using PA-GFP . IMCD3 cells expressing Flag-D1-PAGFP were labeled with anti-Flag antibody conjugated to Alexa555 to visualize the overall surface receptor pool . A point-focused 405-nm laser spot was used to locally photoactivate receptors on the mid-portion of the cilium . Non-fluorescent PA-GFP is depicted in gray , fluorescent state in green . ( B ) Live cell confocal images of a representative cilium showing the frame immediately before the photoactivation pulse ( left column ) , and frames acquired 1 s ( middle column ) and 10 s ( right column ) after local photoactivation . The Flag-Alexa555 signal labeling the entire surface receptor pool ( top row ) was present throughout the cilium at all time points . PA-GFP fluorescence representing the photoactivated pool was non-uniformly distributed at 1 s and uniformly distributed along the cilium within 10 s . ( C ) Line scan analysis of PA-GFP fluorescence along the cilium from the example in panel B . ( D ) Integrated PA-GFP fluorescence signal in the cilium as a function of time after the 405-nm laser pulse . The PA-GFP fluorescence at time = 0 was set at 100% . Points represent the mean fraction of PA-GFP fluorescence present in the cilium over an 80-s imaging interval . Error bars represent SD from analysis of n = 6 cilia . There was no detectable loss of ciliary PA-GFP signal quantified over an 80-s interval . ( E ) Confocal images of a representative cilium acquired immediately after ( 0 min ) and 10 min after photoactivation , showing that the locally photoactivated receptor pool was largely retained in the cilium even after this longer interval . ( F ) Assessing new D1R delivery to the cilium by saturation photoactivation and the sequential ‘image-photoactivate-image’ scheme described in the ‘Materials and methods’ . Bars represent mean fractional increase in ciliary PA-GFP fluorescence elicited by the subsequent test pulse . Error bars represent SD for n = 7 cilia . ( G ) Schematic for modifying the saturation photoactivation method to assess source of newly delivered D1Rs , based on the ratio of integrated PA-GFP/Alexa555 fluorescence ( PA/555 ) measured in the cilium as a function of time . The initial condition is depicted on the left ( ‘0 min’ ) with the fluorescence ratio ( PA/555 ) arbitrarily set to 1 . If new receptors enter the cilium from an internal membrane pool during the 30-min incubation period ( depicted in center , ‘30 min’ ) , they contribute neither Alexa555 nor PA-GFP signal , so the fluorescence ratio is unchanged from the initial condition ( PA/555 = 1 ) . After the subsequent 405-nm test pulse ( depicted at right , ‘30 min + PA’ ) , the PA-GFP signal increases without any change in Alexa555 signal , elevating the fluorescence ratio above the initial condition ( PA/555 > 1 ) . If new receptors enter the cilium from the extra-ciliary plasma membrane pool , they contribute Alexa555 but not PA-GFP signal during the 30-min incubation , reducing the fluorescence ratio from the initial condition ( PA/555 < 1 ) . The subsequent 405-nm pulse restores the fluorescence ratio to the initial value ( PA/555 = 1 ) . ( H ) Experimental results from the strategy depicted in panel G . Bars represent the mean ratio of integrated PA-GFP/Alexa555 fluorescence measured in the cilium . Error bars represent SD from n = 6 cilia . ( *** ) p < 0 . 001 . Scale bars , 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 06996 . 00410 . 7554/eLife . 06996 . 005Figure 2—figure supplement 1 . Whole-cell images corresponding to the images shown in Figure 2B . Flag immunoreactivity is shown in red and PA-GFP in green . Dashed blue line indicates outline of an individual cell . Scale bar , 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 06996 . 00510 . 7554/eLife . 06996 . 006Figure 2—figure supplement 2 . Whole-cell images corresponding to the images shown in Figure 2E . Channels are shown individually in gray scale . Dashed blue line indicates outline of an individual cell . Scale bar , 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 06996 . 00610 . 7554/eLife . 06996 . 007Figure 2—figure supplement 3 . New D1R delivery to cilia increases over time . The sequential ‘image-photoactivate-image’ scheme was applied as described in Figure 2F except that the time interval between the initial 405-nm pulse series and the subsequent assessment of PA-GFP fluorescence increment was varied from 2 min to 50 min . Each square represents an individual determination . The line indicates a least squares best fit . DOI: http://dx . doi . org/10 . 7554/eLife . 06996 . 00710 . 7554/eLife . 06996 . 008Figure 2—figure supplement 4 . Control experiment for the ciliary delivery assay described in Figure 2F . The scheme used in Figure 2F was applied to fixed cells . Error bars represent SD for n = 6 cilia . There was no significant PA-GFP fluorescence increment at either time point . DOI: http://dx . doi . org/10 . 7554/eLife . 06996 . 00810 . 7554/eLife . 06996 . 009Figure 2—figure supplement 5 . Images of cilia from ciliary delivery assay . Representative images of a cilium from the ‘image-photoactivate-image’ scheme described in Figure 2F , G showing the PA-GFP signal and the Flag signal separately in gray scale . Scale bar , 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 06996 . 00910 . 7554/eLife . 06996 . 010Figure 2—figure supplement 6 . Bleaching control for the ciliary delivery assay . Cilia were photoactivated and consecutive PA-GFP and Alexa555 images were acquired . The integrated fluorescence intensity in the cilium was normalized to that measured in the initial image after photoactivation . Error bars represent SD from analysis of n = 3 cilia . There was no detectable loss of ciliary PA-GFP signal or ciliary Alexa555 ( Flag ) signal after 28 consecutive images . DOI: http://dx . doi . org/10 . 7554/eLife . 06996 . 01010 . 7554/eLife . 06996 . 011Figure 2—figure supplement 7 . Bleed-through control for the ciliary delivery assay . Representative images of cilia expressing Flag-D1-GFP in the absence and presence of M1-555 , which recognizes Flag , showing negligible bleed-through of the GFP signal into the 555 channel . The merged image displays the GFP channel in green and 555 ( Flag-receptor ) in red . Insets show a cropped region of the plasma membrane containing the cilium . Dashed blue line indicates outline of an individual cell . Scale bar , 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 06996 . 011 Achieving net ciliary concentration of receptors that are laterally mobile requires the ability of new receptors to enter the cilium . We sought to measure this delivery by taking advantage of the irreversible nature of PA-GFP photoactivation ( Lippincott-Schwartz et al . , 2003 ) . Multiple 405-nm light pulses were delivered in rapid succession to photoactivate the majority of D1Rs present in the cilium . We then assessed the effect of administering a subsequent 405-nm pulse at a later time . We reasoned that , because D1Rs delivered from outside the cilium would not have been previously photoactivated , their arrival in the cilium would result in an increment in the ciliary PA-GFP signal elicited by a subsequent 405-nm pulse . This was indeed the case: when a subsequent 405-nm pulse was administered shortly ( ∼30 s ) after the initial photoactivation series , little ( ∼15% ) increase of ciliary PA-GFP fluorescence was observed , consistent with a small residual fraction of D1Rs escaping photoactivation in the initial pulse series ( Figure 2F , left bar ) . However , when the subsequent 405-nm pulse was administered 30 min after the initial series , the increment of ciliary PA-GFP fluorescence was markedly increased ( Figure 2F , right bar ) . Additionally , the degree of the PA-GFP fluorescence increment increased with time ( Figure 2—figure supplement 3 ) , and we verified that the PA-GFP fluorescence increment was negligible when measured in fixed cells ( Figure 2—figure supplement 4 ) . These results directly verify active delivery of D1Rs to the ciliary membrane compartment and provide a rough estimate of the rate of this delivery , on the order of ∼1% of the total ciliary D1R pool per min . D1Rs delivered to the cilium could originate from an internal vesicular pool or from the extra-ciliary plasma membrane ( Deretic and Papermaster , 1991; Milenkovic et al . , 2009; Wang et al . , 2009 ) . To distinguish these possibilities , we further elaborated the sequential photoactivation technique by taking advantage of dual labeling of surface D1Rs using anti-Flag conjugated to Alexa555 , whose fluorescence was not affected by the 405-nm pulses ( Figure 2G ) . If D1R delivery originates from an internal membrane pool , we expected ( 1 ) unchanged PA-GFP/Alexa555 ratio over the 30-min incubation after initial photoactivation because receptors delivered during this interval would not contribute fluorescence in either channel and ( 2 ) increased PA-GFP/Alexa555 ratio above the initial value after the subsequent photoactivation pulse because newly delivered receptors would contribute PA-GFP but not Alexa555 signal . On the other hand , if ciliary D1R delivery originates from a plasma membrane pool , we expected ( 1 ) decreased PA-GFP/Alexa555 ratio during the 30-min incubation after initial photoactivation because newly delivered D1Rs would be labeled with Alexa555 but lack PA-GFP signal and ( 2 ) return to the initial value after the subsequent 405-nm pulse because newly delivered D1Rs would then contribute both Alexa555 and PA-GFP signal . We observed precisely the latter result: PA-GFP/Alexa555 decreased by approximately 50% during the 30-min incubation after initial photoactivation and returned to a value close to the initial ratio after the subsequent photoactivation pulse ( Figure 2H; example images from separate fluorescence channels shown in Figure 2—figure supplement 5 ) . These observations indicate that D1Rs are delivered to the ciliary membrane compartment primarily from the extra-ciliary plasma membrane pool . Supporting the validity of this fluorescence ratio determination , bleaching of both fluorophores was negligible after sequential image acquisitions exceeding the number required for this experiment ( Figure 2—figure supplement 6 ) , and there was negligible bleed-through from the 488 channel into the 555 channel ( Figure 2—figure supplement 7 ) . To begin to explore the biochemical mechanism of D1R ciliary targeting , we searched for structural determinants within the receptor that are required for ciliary localization . Previous studies of other cilia-localized receptors have identified targeting determinants located either in a cytoplasmic loop ( Berbari et al . , 2008a ) or the cytoplasmic tail ( C-tail; Deretic et al . , 1998; Corbit et al . , 2005; Geng et al . , 2006; Jenkins et al . , 2006 ) . Progressive truncation of the D1R C-tail ( Figure 3A ) strongly reduced ciliary localization of receptors . Truncating the distal end had no effect ( e . g . , D1-415T; Figure 3B ) , but removing a larger portion strongly reduced ciliary receptor localization ( D1-382T; Figure 3C ) . A 15-residue sequence within this region markedly reduced D1R ciliary localization when selectively deleted ( D1∆381–395; Figure 3D; whole-cell images for Figure 3B–D are shown in Figure 3—figure supplement 1 ) . This effect was verified by both metrics of ciliary receptor targeting ( Figure 3E , F ) , and the deletion did not disrupt overall surface expression of receptors ( Figure 3—figure supplement 2 ) . 10 . 7554/eLife . 06996 . 012Figure 3 . The D1R cytoplasmic tail is necessary and sufficient for ciliary receptor targeting . ( A ) Schematic representation of D1R C-tail mutations used in the present analysis . ( B ) Representative images of cells expressing Flag-tagged wild-type D1R ( D1R ) or a receptor construct truncated at residue 415 ( D1-415T ) , showing robust ciliary localization of both . The merged image at bottom displays Flag-receptor in green and AcTub in red . ( C ) Representative image of a receptor construct truncated at residue 382 ( D1-382T ) , showing near complete loss of receptor localization to cilia marked by AcTub . ( D ) Representative images of a D1R construct with internal deletion of residues 381–395 ( D1Δ381-395 ) , showing the range of phenotypes observed , from a complete loss of receptor localization in cilia to a pronounced reduction of ciliary receptor localization . ( E ) Quantification of the fraction of receptor ( Flag ) -positive cilia for Flag-tagged wild-type ( D1R ) or mutant ( D1Δ381-395 ) receptor . The analysis is described in Figure 1D . ( F ) Quantification of the average fold-enrichment of wild-type ( D1R ) or mutant ( D1Δ381-395 ) receptors on cilia . The analysis is described in Figure 1E , F . ( G ) Schematic representation of chimeric mutant receptors containing portions of the D1R cytoplasmic tail ( in red ) fused to the delta opioid peptide receptor ( DOR ) cytoplasmic tail ( in green ) . ( H ) Representative images of cilia in cells expressing Flag-DOR , Flag-DOR-D1 ( 338–446 ) , or Flag-DOR-D1 ( 368–446 ) showing that the D1R C-tail is sufficient to drive ciliary targeting of chimeric receptors . ( I ) Fraction of receptor ( Flag ) -positive cilia . ( J ) Average fold-enrichment of receptor ( Flag ) signal on cilia . Error bars represent SEM from n = 3 experiments with 10–20 cilia analyzed per experiment . ( *** ) p < 0 . 001 . Scale bars , 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 06996 . 01210 . 7554/eLife . 06996 . 013Figure 3—figure supplement 1 . Whole-cell images corresponding to images shown in Figure 3B–D . The merged image displays Flag-receptor immunoreactivity in green and AcTub in red . Dashed blue line indicates outline of an individual cell . Scale bars , 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 06996 . 01310 . 7554/eLife . 06996 . 014Figure 3—figure supplement 2 . Overall surface expression of D1R C-tail mutant . Surface-accessible Flag immunoreactivity was quantified for the cilia-defective mutant D1R ( D1∆381–395 ) relative to wild-type D1R by fluorescence flow cytometry , as described in the ‘Materials and methods’ . The ciliary targeting defective mutant receptor was expressed in the overall plasma membrane at a level indistinguishable from the cilia-localized wild-type D1R , supporting the hypothesis that the C-tail region mutated is required specifically for targeting surface receptors to cilia , but not for targeting receptors to the extra-ciliary surface . DOI: http://dx . doi . org/10 . 7554/eLife . 06996 . 01410 . 7554/eLife . 06996 . 015Figure 3—figure supplement 3 . Whole-cell images corresponding to images shown in Figure 3H . The merged image displays Flag-receptor immunoreactivity in green and AcTub in red . Dashed blue line indicates outline of an individual cell . Scale bar , 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 06996 . 01510 . 7554/eLife . 06996 . 016Figure 3—figure supplement 4 . Overall surface expression of DOR-derived constructs . Surface-accessible Flag immunoreactivity for the indicated DOR-derived chimeric mutant constructs expressed relative to wild-type DOR . The chimeric mutant receptors were expressed at similar levels , further supporting the specific function of the D1R-derived sequence in targeting receptors to the ciliary plasma membrane compartment . ( * ) p < 0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 06996 . 01610 . 7554/eLife . 06996 . 017Figure 3—figure supplement 5 . The 15 residue sequence required for full ciliary targeting of D1R is not sufficient to confer ciliary localization on DOR . Representative image of a cilium in cells expressing the chimeric mutant receptor Flag-DOR-D1 ( 379–400 ) , which includes the D1R residues 381–395 , showing that this sequence is not sufficient to confer ciliary targeting of receptors . The merged image displays Flag-receptor immunoreactivity in green and AcTub in red . Insets show a cropped region of the plasma membrane containing the cilium . Dashed blue line indicates outline of an individual cell . Scale bar , 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 06996 . 017 We next asked if the D1R C-tail is sufficient to confer ciliary localization on a non-ciliary GPCR . Fusion of the entire D1R C-tail to DOR ( Figure 3G ) conferred robust ciliary localization . However , fusion of a shorter fragment of the D1R C-tail was not sufficient to confer ciliary localization ( Figure 3H–J; whole-cell images for Figure 3H shown in Figure 3—figure supplement 3 ) , even though it contained the necessary 381–395 sequence , and chimeras exhibited similar levels of overall surface expression ( Figure 3—figure supplement 4 ) . Additionally , we found that fusion of a smaller region of the D1R C-tail that also contains residues 381–395 ( Flag-DOR-D1 ( 379–400 ) ) was not sufficient to confer ciliary localization on DOR . 23 out of 23 cells expressing the chimeric receptor had no visible enrichment of receptor in the cilium ( Figure 3—figure supplement 5 ) . Together , these results indicate that the structural information required for D1R ciliary targeting is contained in the receptor's C-tail and requires residues 381–395 for full activity . However , the targeting determinant is clearly not restricted to this 15-residue sequence , and it likely represents a more extended structure including residues in the proximal C-tail . We next pursued a candidate-based RNA interference screen to search for trans-acting proteins required for ciliary D1R targeting ( Table 1 ) . Duplexes were transfected individually into IMCD3 cells stably expressing the Flag-D1R , using a clone with particularly strong ciliary receptor accumulation , and ciliary D1R localization was scored visually . No effect was found for siRNAs targeting several proteins implicated in ciliary localization other GPCRs , including TULP3 and BBSome components important for ciliary localization of SSTR3 , MCHR1 , and Gpr161 ( Table 1; Berbari et al . , 2008b; Jin et al . , 2010; Mukhopadhyay et al . , 2010 , 2013 ) . Instead , two components of the IFT-B complex , IFT57 and IFT172 , were identified ( Figure 4A; whole-cell images shown in Figure 4—figure supplement 1 ) and knockdown was verified by qRT-PCR ( Figure 4—figure supplement 2 ) . IFT-B knockdown was complicated by reduced ciliogenesis ( Figure 4B ) , but in the ciliated cells remaining in the transfected cell population , reduced D1R targeting was clearly evident . This was verified by both quantitative metrics of ciliary D1R localization , and this was specific to ciliary targeting because knockdown produced little or no effect on the overall surface expression of receptors ( Figure 4C , D; Figure 4—figure supplement 3 ) . As IFT57 tolerates an N-terminal epitope tag while IFT172 does not ( Follit et al . , 2009 ) , we tested rescue of the IFT57 knockdown effect using an N-terminally HA-tagged IFT57 construct engineered with silent mutations in the sequence targeted by IFT57 siRNA ( HA-IFT57-NTM ) . Verifying a specific requirement for IFT57 in ciliary D1R targeting , HA-IFT57-NTM restored normal ciliary D1R localization in 30 out of 30 cells analyzed ( Figure 4A , rescue; whole-cell images verifying HA-IFT57-NTM expression in Figure 4—figure supplement 1 ) . 10 . 7554/eLife . 06996 . 018Table 1 . siRNA knockdown screenDOI: http://dx . doi . org/10 . 7554/eLife . 06996 . 018GeneImplication+/− effect on D1 in ciliaTULP3SSTR3 , MCHR1 cilia localization−Bbs4 , Bbs2SSTR3 , MCHR1 cilia localization−Arf4Rhodopsin localization to rod outer segment−Asap1Rhodopsin localization to rod outer segment−Kif7 , Kif27Hedgehog signaling−Vps35Receptor trafficking−Rab15 , Rab14 , Rab8 , Rab11Cilia associated Rabs−Rab4 , Rab35Trafficking of receptors−Rab23Hedgehog signaling+Arl6Cilia associated small GTPase−IFT57Intraflagellar transport+IFT172Intraflagellar transport+Clathrin heavy chainReceptor endocytosis−Pacs1Olfactory CNG channel cilia localization−Kif5cApical trafficking of cargo−Septin2Cilia diffusion barrier−10 . 7554/eLife . 06996 . 019Figure 4 . IFT-B complex proteins are necessary for D1R localization to cilia . ( A ) Representative images of cilia on cells stably transfected with Flag-D1R and transiently transfected with a non-silencing duplex ( control ) or siRNA targeting IFT57 ( IFT57-1 and IFT57-4 ) or IFT172 ( IFT172-3 and IFT172-4 ) . In the merged image , Flag-D1R immunoreactivity is shown in green and AcTub in red . Duplexes targeting IFT57 and IFT172 caused a visually obvious reduction in ciliary D1R localization . The right column of images shows the rescue condition in which cells transfected with siRNA against IFT57 were additionally transfected with IFT57 that is not targetable by the IFT57 siRNA ( HA-IFT57-NTM ) . Scale bar , 5 μm . ( B ) Effect of siRNAs on the fraction of cells in the population possessing a visible cilium marked by AcTub . Error bars represent SEM from 150 cells counted in n = 3 independent experiments . ( C ) Fraction of D1R ( Flag ) -positive cilia . Error bars represent SEM for n = 3 experiments with 50 cells counted in each experiment . ( D ) Average fold-enrichment of D1R ( Flag ) signal on cilia . Error bars represent SEM for n = 3 experiments with 15–40 cilia analyzed per experiment . ( E ) Association of IFT57 with the D1R but not DOR demonstrated by co-immunoprecipitation . Cells were transfected with the constructs indicated above each lane . Cell extracts were blotted for HA and Flag; HA-IFT57 resolved as a sharp band at its expected apparent molecular mass and Flag-tagged receptors resolved as heterogeneous species consistent with complex glycosylation as shown previously . Specific co-immunoprecipitation is indicated by HA-IFT57 detected in the Flag-D1R pull-down but not in Flag-DOR pull-down . Molecular mass markers ( in kDa ) are shown on right side of blots . The results in panel E are representative of n = 3 independent experiments . ( F ) Increased association of IFT57 with D1Δ381-395 relative to D1R demonstrated by co-IP . Cells were transfected with the constructs indicated above each lane . Cell extracts were blotted for HA and Flag . More HA-IFT57 was detected in the Flag-D1Δ381-395 pull-down than the Flag-D1R pull-down . Molecular mass markers ( in kDa ) are shown on right side of blots . The results in panel E are representative of n = 3 independent experiments . ( G ) Immunoblots from multiple experiments were scanned in the linear range , as described in ‘Materials and methods’ , to estimate the amount of IFT57 co-IPed with the indicated receptors . Expressed as a fold increase over control where control is D1R . Error bars represent SD from n = 3 experiments . ( * ) p < 0 . 05; ( ** ) p < 0 . 01; ( *** ) p < 0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 06996 . 01910 . 7554/eLife . 06996 . 020Figure 4—figure supplement 1 . Whole-cell images corresponding to the images shown in Figure 4A . The merged image displays Flag-D1R immunoreactivity in green and AcTub in red . Expression of HA-IFT57-NTM is verified by HA immunoreactivity . Dashed blue line indicates outline of an individual cell . Scale bars , 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 06996 . 02010 . 7554/eLife . 06996 . 021Figure 4—figure supplement 2 . Verification of IFT-B knockdown . Cells stably expressing Flag-D1R were transfected with either control siRNA or siRNA targeting intraflagellar transport ( IFT ) -B components , IFT57 and IFT172 , and RNA levels were measured via qRT-PCR . Error bars represent SEM for n = 3 experiments . ( ** ) p < 0 . 01; ( *** ) p < 0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 06996 . 02110 . 7554/eLife . 06996 . 022Figure 4—figure supplement 3 . IFT-B knockdown has little effect on overall surface receptor expression . Surface-accessible Flag immunoreactivity representing Flag-D1Rs present in the plasma membrane was quantified by flow cytometry . Results are normalized to control siRNA . DOI: http://dx . doi . org/10 . 7554/eLife . 06996 . 022 Co-immunoprecipitation analysis detected physical association of HA-IFT57 with Flag-D1R , and this was specific because HA-IFT57 did not detectably co-IP with Flag-DOR even when the latter was expressed at higher levels than Flag-D1R ( Figure 4E ) . To ask if IFT57 interaction with D1R is directly congruent with ciliary targeting activity , we carried out co-immunoprecipitation analysis comparing HA-IFT57 pull-down with wild-type Flag-D1R and FlagD1Δ381-395 . Deletion of resides 381–395 of the D1R C-tail , a mutation that profoundly impairs ciliary targeting , did not reduce IFT57 co-IP . To the contrary , the co-IP signal was significantly enhanced ( Figure 4F , G ) . This suggests that the D1R C-tail , while both necessary and sufficient for ciliary receptor targeting , likely functions in a more complex manner than can be explained by a single interaction surface with IFT-B . IFT-B is known to associate with KIF3A ( kinesin-II , a heterotrimeric kinesin-2 ) and KIF17 ( a homodimeric kinesin-2 ) , plus end-directed kinesins that mediate anterograde cargo movement toward or within cilia ( reviewed in Pedersen and Rosenbaum , 2008 ) . KIF3A is required for delivery of axonemal components and overall ciliogenesis , while KIF17 is not essential for ciliogenesis and is proposed to have more specialized cargo transport functions ( Jenkins et al . , 2006; Zhao et al . , 2012 ) . We verified the presence of KIF17-IFT57 complexes in IMCD3 cells , as reported previously by others ( Insinna et al . , 2008; Howard et al . , 2013 ) , by co-immunoprecipitation ( Figure 5A ) . Thus , we hypothesized that KIF17 may function as a specific motor supporting D1R delivery to cilia . To test this , we introduced a point mutation in the KIF17 motor domain at a conserved residue in the switch II region ( KIF17-G234A ) that is essential for kinesin motor activity ( Figure 5—figure supplement 1; Rice et al . , 1999 ) . D1Rs were visible on cilia in essentially all cells expressing the motor-defective HA-KIF17-G234A , but the degree of ciliary receptor enrichment was greatly reduced ( Figure 5B ) . In contrast , expression of wild-type KIF17 did not visibly affect D1R ciliary localization ( Figure 5B; whole-cell images verifying HA-KIF17-G234A and HA-KIF17 expression shown in Figure 5—figure supplement 2 ) . Verifying this , KIF17-G234A , but not KIF17 , selectively reduced the fold-enrichment metric while having little effect on the fraction of receptor-positive cilia ( Figure 5C , D ) . As an independent assessment of the role of KIF17 on D1R ciliary localization , we examined the effect of expressing a different dominant negative KIF17 construct ( HA-KIF17-DN ) containing only the cargo-binding domain ( Jenkins et al . , 2006 ) . Expression of HA-KIF17-DN also significantly reduced D1R ciliary enrichment ( Figure 5—figure supplement 3 ) . Further , this effect was specific to D1Rs because KIF17-G234A did not reduce ciliary enrichment of SSTR3 ( Figure 5E , F; whole-cell images verifying HA-KIF17-G234A expression in Figure 5—figure supplement 4 ) . Expression of HA-KIF17-G234A did not significantly affect overall surface expression of either D1R or SSTR3 ( Figure 5—figure supplement 5 ) . 10 . 7554/eLife . 06996 . 023Figure 5 . KIF17 motor activity is required for full D1R enrichment in cilia . ( A ) Association of IFT57 with KIF17 indicated by co-immunoprecipitation . Cells were transfected with expression constructs indicated at top of the panel , and extracts were blotted for HA to detect IFT57 and Flag to detect KIF17 . HA-IFT57 resolved as expected and described in Figure 4E . KIF17 resolved as two species with the top band corresponding to the expected molecular mass of the full-length protein . Specific co-immunoprecipitation is indicated by HA-IFT57 detected in the Flag pull-down from cells expressing Flag-KIF17 pull-down ( arrow ) but not from cells in which Flag-KIF17 was not expressed . Molecular mass markers ( in kDa ) shown on right . ( B ) Representative images of cilia on cells co-transfected with Flag-D1R and control empty vector ( +pcDNA ) , a plasmid encoding HA-tagged KIF17 ( +KIF17 ) , or a plasmid encoding an HA-tagged KIF17 construct harboring a point mutation in a conserved residue that disrupts kinesin motor activity ( +KIF17-G234A ) . Robust ciliary localization of Flag-D1R was observed in cells expressing control plasmid or the wild-type KIF17 construct , but markedly reduced ciliary enrichment of D1Rs was observed in cells expressing motor-defective mutant KIF17 . ( C ) Quantification of the effect of disrupting KIF17 motor activity on the fraction of D1R ( Flag ) -positive cilia . ( D ) Quantification of the effect of disrupting KIF17 motor activity on average fold-enrichment of D1R ( Flag ) on cilia . Disrupting KIF17 motor activity strongly reduced the degree of D1R enrichment on the ciliary membrane without blocking D1R access to cilia . ( E ) Representative images of cilia on cells co-transfected with Flag-SSTR3 and with control empty vector ( +pcDNA ) or a plasmid encoding motor domain-mutant KIF17 ( +KIF17-G234A ) . Disrupting KIF17 motor activity did not detectably affect Flag-SSTR3 localization to cilia . ( F ) Quantification of the effect of disrupting KIF17 motor activity on average fold-enrichment of somatostatin-3 receptor ( SSTR3 ) ( Flag ) on cilia . Disrupting KIF17 motor activity did not detectably affect ciliary enrichment of SSTR3 . Error bars represent SEM from n = 3 independent experiments with 10–20 cilia analyzed in each experiment . ( *** ) p < 0 . 001 . Scale bars , 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 06996 . 02310 . 7554/eLife . 06996 . 024Figure 5—figure supplement 1 . Switch II mutation in KIF17 . Amino acid sequence alignment of the switch II regions of Kinesin-1 and KIF17 showing identical sequences . Switch II residues are shown in magenta with a box around the G residue mutated in the motor domain mutant . DOI: http://dx . doi . org/10 . 7554/eLife . 06996 . 02410 . 7554/eLife . 06996 . 025Figure 5—figure supplement 2 . Whole-cell images corresponding to the images shown in Figure 5B . HA-KIF17-G234A and HA-KIF17 expression is verified by HA immunoreactivity detected throughout the cytoplasm and accumulated in the nucleus . The HA signal detected in cilia is likely contributed in large part by cross reactivity of the secondary anti-rat antibody with the mouse primary antibody recognizing AcTub . Dashed blue line indicates outline of an individual cell . Scale bars , 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 06996 . 02510 . 7554/eLife . 06996 . 026Figure 5—figure supplement 3 . Independent verification of KIF17 requirement for D1R ciliary enrichment . Cells were co-transfected with Flag-D1R and an empty vector ( +pcDNA ) or a plasmid encoding an HA-tagged version of a previously reported dominant negative KIF17 ( +KIF17-DN ) . Quantification of the effect of KIF17-DN on average fold-enrichment of D1R ( Flag ) on cilia . Expression of KIF17-DN strongly reduced the degree of D1R ciliary enrichment . Error bars represent SEM from n = 4 independent experiments with 10–20 cilia analyzed per experiment . ( *** ) p < 0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 06996 . 02610 . 7554/eLife . 06996 . 027Figure 5—figure supplement 4 . Whole-cell images corresponding to the images shown in Figure 5E . HA-KIF17-G234A expression is verified by HA immunoreactivity . Dashed blue line indicates outline of an individual cell . Scale bars , 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 06996 . 02710 . 7554/eLife . 06996 . 028Figure 5—figure supplement 5 . The KIF17 motor domain mutation has little effect on overall surface expression of receptors . Effects of KIF17 transfection on surface-accessible Flag immunoreactivity from Flag-tagged D1R ( D1R ) or Flag-tagged SSTR3 ( SSTR3 ) were quantified by fluorescence flow cytometry and normalized to the mock-transfected ( pcDNA ) condition . Bars represent mean normalized surface expression . ( * ) p < 0 . 05; ( ** ) p < 0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 06996 . 028 Our siRNA screen also identified Rab23 as a candidate whose knockdown caused a pronounced reduction of D1R localization to cilia ( Figure 6A; whole-cell images in Figure 6—figure supplement 1 ) without affecting ciliogenesis ( Figure 6B ) . Knockdown was verified by qRT-PCR ( Figure 6—figure supplement 2 ) . Rab23 knockdown strongly reduced both quantitative metrics of ciliary D1R targeting , without affecting overall surface expression of receptors ( Figure 6C , D; Figure 6—figure supplement 3 ) . Rab23 knockdown also blocked ciliary targeting activity of the D1R C-tail assessed through fusion to the normally cilia-excluded DOR ( Figure 6E , F ) . This was unexpected because Rab23 was not known previously to be required for ciliary targeting of any signaling receptor or other membrane cargo . 10 . 7554/eLife . 06996 . 029Figure 6 . Rab23 is necessary for D1R localization to cilia . ( A ) Representative images of cilia on cells stably transfected with Flag-D1R and transiently transfected with a non-silencing duplex ( control ) or siRNA targeting Rab23 ( Rab23-4 and Rab23-2 ) . Rab23 knockdown strongly reduced Flag-D1R localization to cilia . ( B ) Quantification of the siRNA effect on the fraction of cells in the population possessing a visible cilium marked by AcTub . Error bars represent SEM from 150 cells counted in n = 3 experiments . ( C ) Quantification of the Rab23 knockdown effect on the fraction of D1R ( Flag ) -positive cilia . Error bars represent SEM from n = 3 experiments with 50 cells counted per experiment . ( D ) Quantification of the Rab23 knockdown effect on average fold-enrichment of D1R ( Flag ) signal on cilia . Error bars represent SEM from n = 3 independent experiments with 20–30 cilia analyzed per experiment . ( E ) Schematic representation of wild-type D1R and the cilia-targeted DOR-D1 ( 338–446 ) chimeric mutant receptor ( duplicated from Figure 3G ) . ( F ) Quantification of the Rab23 knockdown effect on average fold-enrichment of Flag-D1R ( D1R ) and the Flag-tagged chimeric mutant receptor ( DOR-D1 ( 338–446 ) ) on cilia of transiently transfected cells . Error bars represent SEM from n = 3 independent experiments with 10–20 cilia analyzed per experiment . ( G ) Representative images of cilia on cells stably transfected with SSTR3-GFP and transiently transfected with a non-silencing duplex ( control ) or siRNA targeting Rab23 ( Rab23-4 and Rab23-2 ) . Rab23 knockdown strongly reduced SSTR3-GFP localization to cilia . ( H ) Quantification of the Rab23 knockdown effect on the fraction of SSTR3-GFP ( GFP ) positive cilia . Error bars represent SEM from n = 3 experiments with 50 cells counted per experiment . ( I ) Quantification of the Rab23 knockdown effect on average fold-enrichment of SSTR3 ( GFP ) signal on cilia . Error bars represent SEM from n = 3 experiments with 10–20 cilia analyzed in each experiment . ( * ) p < 0 . 05; ( ** ) p < 0 . 01; ( *** ) p < 0 . 001 . Scale bars , 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 06996 . 02910 . 7554/eLife . 06996 . 030Figure 6—figure supplement 1 . Whole-cell images for corresponding images shown in Figure 6A . The merged images display Flag-D1R immunoreactivity in green and AcTub in red . Dashed blue line indicates outline of an individual cell . Scale bar , 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 06996 . 03010 . 7554/eLife . 06996 . 031Figure 6—figure supplement 2 . Verification of Rab23 knockdown . Results of qRT-PCR analysis . Error bars represent SEM for n = 3 experiments . ( ** ) p < 0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 06996 . 03110 . 7554/eLife . 06996 . 032Figure 6—figure supplement 3 . Rab23 knockdown has little effect on overall surface receptor expression . Results of flow cytometric analysis . Results are normalized to control siRNA . Bars represent mean normalized surface expression . DOI: http://dx . doi . org/10 . 7554/eLife . 06996 . 03210 . 7554/eLife . 06996 . 033Figure 6—figure supplement 4 . Whole-cell images for corresponding images shown in Figure 6G . The merged images display SSTR3-GFP in green and AcTub in red . Dashed blue line indicates outline of an individual cell . Scale bar , 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 06996 . 033 To ask if Rab23 affects additional receptor cargoes , we carried out the same experiment investigating ciliary localization of SSTR3 . Duplexes were transfected individually into IMCD3 cells stably expressing SSTR3-GFP . Rab23 knockdown significantly reduced ciliary targeting of SSTR3 , as assessed by epifluorescence microscopy and both quantitative metrics ( Figure 6G–I; whole-cell images shown in Figure 6—figure supplement 4 ) . Rab23 was not only necessary for ciliary localization of wild-type D1Rs and the ciliary targeting activity of the D1R C-tail , but it was also sufficient to rescue ciliary targeting of targeting-defective D1Rs when fused to the C-tail . Fusing wild-type Rab23 to the C-tail of the ciliary localization-defective D1Δ381-395 mutant receptor ( D1Δ381-395-Rab23 , Figure 7A ) rescued robust ciliary targeting ( Figure 7B; whole-cell images shown in Figure 7—figure supplement 1 ) . This effect was dependent on the nucleotide state of Rab23 because a GTP binding-defective mutant allele ( D1Δ381-395-Rab23-S23N ) failed to produce detectable ciliary receptor localization , while fusion of an activated Rab23 allele ( D1Δ381-395-Rab23-Q68L ) drove ciliary localization even more robustly than the D1R C-tail itself ( Figure 7B–D; overall surface expression is shown in Figure 7—figure supplement 2 ) . Rab23 was detectable but not highly concentrated in cilia ( Figure 7E ) . Therefore , we do not think Rab23 fusion confers ciliary localization on receptors by simple tethering . 10 . 7554/eLife . 06996 . 034Figure 7 . Rab23 is sufficient to drive ciliary localization of a non-ciliary GPCR . ( A ) Schematic representation of the D1R-derived constructs examined . D1R-derived sequence is depicted in red and Rab23 sequence in blue . Flag-tagged wild-type D1R was compared to the ciliary targeting-impaired mutant D1R ( D1Δ381-395 ) , and to the ciliary targeting-impaired mutant D1R fused to wild-type Rab23 ( D1Δ381-395-Rab23 ) , inactive mutant Rab23 ( D1Δ381-395-Rab23-S23N ) or activated mutant Rab23 ( D1Δ381-395-Rab23-Q68L ) . ( B ) Representative images of cilia on cells transiently expressing Flag-tagged versions of the indicated receptor constructs . Fusion of either wild-type or activated Rab23 to the cilia targeting-defective D1R visibly enhanced ciliary localization of receptors . ( C ) Quantification of the fraction of receptor ( Flag ) -positive cilia . ( D ) Average fold-enrichment of receptor ( Flag ) on cilia . ( E ) Representative live-cell images of GFP-tagged Rab23 or Rab23-Q68L localization relative to cilia marked by Arl13b-mRuby after transient co-transfection . ( F ) Schematic representation of the DOR-Rab fusions . DOR-derived sequence is depicted in green , Rab23 in blue , Rab11 in violet , and Rab8 in orange . ( G ) Representative images of cilia on cells transiently expressing Flag-tagged versions of the indicated receptor constructs . Wild-type DOR was not detected on cilia . Fusion of activated ( Q68L ) Rab23 produced strong ciliary localization , while fusion of activated ( Q70L ) Rab11 or activated ( Q67L ) Rab8 failed to do so . ( H ) Fraction of receptor ( Flag ) -positive cilia . ( I ) Average fold-enrichment of receptor ( Flag ) signal on cilia . Error bars represent SEM from n = 3 independent experiments with 10–20 cilia analyzed in each experiment . ( ** ) p < 0 . 01; ( *** ) p < 0 . 001 . Scale bars , 5 μm . GPCR , G protein-coupled receptor . DOI: http://dx . doi . org/10 . 7554/eLife . 06996 . 03410 . 7554/eLife . 06996 . 035Figure 7—figure supplement 1 . Whole-cell images corresponding to images shown in Figure 7B . The merged images display Flag-receptor immunoreactivity in green and AcTub in red . Dashed blue line indicates outline of an individual cell . Scale bars , 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 06996 . 03510 . 7554/eLife . 06996 . 036Figure 7—figure supplement 2 . Overall surface expression of D1R-Rab23 fusion constructs . Results of flow cytometric analysis . Bars represent mean normalized to Flag-D1R surface expression . ( ** ) p < 0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 06996 . 03610 . 7554/eLife . 06996 . 037Figure 7—figure supplement 3 . Whole-cell images corresponding to images shown in Figure 7G . The merged images display Flag-receptor immunoreactivity in green and AcTub in red . Dashed blue line indicates outline of an individual cell . Scale bars , 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 06996 . 03710 . 7554/eLife . 06996 . 038Figure 7—figure supplement 4 . Overall surface expression of DOR-Rab23 fusion . Results of flow cytometric analysis . Bars represent mean normalized to Flag-DOR surface expression . ( ** ) p < 0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 06996 . 03810 . 7554/eLife . 06996 . 039Figure 7—figure supplement 5 . Rab23 is sufficient to drive ciliary localization of several non-ciliary GPCRs . Both mu-opioid receptor ( MOR ) and B2AR were fused to activated mutant Rab23 ( MOR-Rab23-Q68L , B2AR-Rab23-Q68L ) . Representative images of cells transiently expressing Flag-tagged versions of the indicated constructs are shown . Wild-type MOR and B2AR were not detected on cilia , while fusion of MOR and B2AR to activated ( Q68L ) Rab23 produced strong chimeric receptor ciliary localization . The merged images display Flag-receptor immunoreactivity in green and AcTub in red . Insets show a cropped region of the plasma membrane containing the cilium . Dashed blue line indicates outline of an individual cell . Scale bars , 5 μm . GPCR , G protein-coupled receptor . DOI: http://dx . doi . org/10 . 7554/eLife . 06996 . 039 To ask if Rab23 fusion is fully sufficient to promote ciliary targeting of receptors , we carried out similar experiments fusing Rab23-Q68L to DOR , which is normally undetectable on cilia . Remarkably , activated Rab23 drove robust ciliary localization of DOR ( DOR-Rab23-Q68L , Figure 7F–I; whole-cell images are shown in Figure 7—figure supplement 3 ) . While fusion of Rab23-Q68L to the mutant D1R increased overall receptor surface expression ( Figure 7—figure supplement 2 ) , the opposite was observed for fusion to DOR ( Figure 7—figure supplement 4 ) . This indicates that the ciliary targeting activity of Rab23 is separable from its effects on overall surface receptor expression . Moreover , this ciliary targeting activity was specific for Rab23 because fusion of an activated allele of the closely related Rab paralogue , Rab11 ( DOR-Rab11-Q70L ) , or of Rab8 ( DOR-Rab8-Q67L ) that is known to localize to cilia , failed to drive detectable ciliary targeting ( Figure 7F , G ) . The ability of activated Rab23 to confer ciliary localization on a non-ciliary GPCR was not limited to DOR . Fusion of activated Rab23 to the C-tails of the mu-opioid receptor ( MOR ) and beta-2-adrenergic receptor ( B2AR ) , which are normally excluded from cilia , conferred robust ciliary localization on both receptors ( Figure 7—figure supplement 5 ) . Together , these results support the hypothesis that Rab23 plays a key role in determining the specificity of ciliary receptor targeting . We next sought to investigate if the identified protein components required for D1R ciliary targeting function in an integrated pathway . As noted above , disrupting KIF17 motor activity strongly reduced ciliary enrichment of the wild-type D1R . In contrast , ciliary enrichment driven by direct fusion of the activated Rab23 to the D1R ( D1Δ381-395-Rab23-Q68L ) was unaffected by this manipulation ( Figure 8A–C; whole-cell images verifying HA-KIF17-G234A expression are shown in Figure 8—figure supplement 1 ) . Additionally , full ciliary enrichment of D1Δ381-395-Rab23-Q68L remained in the presence of IFT172 knockdown ( Figure 8D ) , suggesting that fusion to activated Rab23 can also override the IFT-B requirement . We also noted that Rab23 knockdown did not prevent or reduce D1R association with IFT-B , as estimated by co-immunoprecipitation of IFT57 ( Figure 8E ) . To the contrary , Rab23 knockdown tended to increase the IFT57-D1R co-IP signal ( Figure 8F ) . In contrast to its clear effect on the ciliary concentration of D1Rs , disrupting KIF17 motor activity did not prevent Rab23 localization to cilia ( Figure 8—figure supplement 2 ) . Together , these results suggest that IFT-B/KIF17 and Rab23 are all required for efficient targeting of D1Rs to cilia and may indeed function in an integrated pathway . 10 . 7554/eLife . 06996 . 040Figure 8 . Evidence IFT-B , KIF17 , and Rab23 function in an integrated ciliary delivery pathway . ( A ) Representative images showing the effect of motor domain-mutant KIF17 ( +KIF17-G234A ) on Flag-tagged wild-type D1R localization to the cilium ( from Figure 5B ) . ( B ) Representative images from an identical experiment examining localization of the Flag-tagged D1R fusion to activated Rab23 ( D1Δ381-395-Rab23-Q68L ) . Scale bars , 5 μm . ( C ) Average fold-enrichment of D1R and D1Δ381-395-Rab23-Q68L ( Flag ) signal on cilia . Wild-type D1R localization to cilia was strongly reduced by motor-defective KIF17 , but direct Rab23 fusion effectively bypassed this requirement . ( D ) Effect of IFT172 knockdown on average fold-enrichment of D1R and D1Δ381-395-Rab23-Q68L ( Flag ) signal on cilia . Wild-type D1R localization to cilia was strongly reduced by IFT172 knockdown , but direct Rab23 fusion effectively bypassed this requirement . Error bars represent SEM from n = 3 independent experiments with 10–20 cilia analyzed in each experiment . ( *** ) p < 0 . 001 . ( E ) Co-immunoprecipitation analysis showing that Rab23 is not necessary for D1R association with IFT57 . The analysis and presentation of data are described in Figure 4E . ( F ) Immunoblots from multiple experiments were quantified in the linear range to estimate the amount of IFT57 co-IPed . The result is expressed as a fold-increase over the control siRNA condition . Error bars represent SD from n = 3 experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 06996 . 04010 . 7554/eLife . 06996 . 041Figure 8—figure supplement 1 . Whole-cell images corresponding to the images shown in Figure 8B . HA-KIF17-G234A expression is verified by HA immunoreactivity . Dashed blue line indicates outline of an individual cell . Scale bars , 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 06996 . 04110 . 7554/eLife . 06996 . 042Figure 8—figure supplement 2 . Disruption of KIF17 motor activity does not affect Rab23 ciliary localization . Representative live-cell images of cells co-transfected with Flag-Rab23-Q68L , Arl13b-YFP as a cilia marker , and either empty vector ( +pcDNA ) or motor domain mutant KIF17 ( +KIF17-G234A ) . In merged image , Flag-Rab23-Q68L ( Flag ) immunoreactivity is shown in red , Arl13b-YFP in green , and HA-KIF17-G234A in blue . Rab23-Q68L was observed in 8/18 cilia in cells expressing pcDNA . Similarly , Rab23-Q68L was clearly visible in 9/15 cilia in cells expressing KIF17-G234A . Scale bar , 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 06996 . 042
The physiological signaling functions of primary cilia critically depend on the specificity with which particular receptors are targeted to the ciliary membrane ( Corbit et al . , 2005; Garcia-Gonzalo and Reiter , 2012 ) . We identified a discrete mechanism of ciliary receptor targeting through study of the D1R . Past work has focused primarily on receptor delivery from a post-Golgi membrane source . In contrast , we found that D1Rs are delivered to cilia from the extra-ciliary plasma membrane . We also found that ciliary D1R delivery is directed by the receptor C-tail . However , we were unable to find any sequence in the D1R C-tail conforming to a previously defined ciliary targeting motif ( Deretic et al . , 1998; Geng et al . , 2006; Jenkins et al . , 2006; Berbari et al . , 2008a ) . Also , our mutational studies suggest that the structural determinant required for ciliary D1R targeting is a relatively extended structure . We identified a distinct set of trans-acting proteins important for ciliary targeting of D1Rs . We were unable to detect a requirement for TULP3 or BBSome components , although these are essential for ciliary localization of SSTR3 and MCHR1 ( Berbari et al . , 2008b; Jin et al . , 2010; Mukhopadhyay et al . , 2010 ) . We also did not detect a requirement for Arf4 , ASAP1 , or Rab11 , which are essential for rhodopsin delivery to the rod outer segment ( Deretic et al . , 2005; Wang et al . , 2012 ) . Thus , D1Rs add to a growing appreciation that there exist receptor-specific differences in mechanisms of ciliary membrane targeting . Nevertheless , some similarities are evident . The present results are consistent with IFT-B functioning in anterograde cargo delivery to the cilium ( Silverman and Leroux , 2009; Garcia-Gonzalo and Reiter , 2012 ) . Several IFT-B components ( IFT20 , IFT57 , and IFT140 ) have been implicated in rhodopsin delivery to the rod outer segment ( Krock and Perkins , 2008; Keady et al . , 2011; Crouse et al . , 2014 ) , and IFT172 was recently implicated in Smo localization to primary cilia ( Kuzhandaivel et al . , 2014 ) . The present finding that IFT57 and IFT172 are required for ciliary D1R targeting provides further evidence that IFT-B contributes to ciliary delivery of select membrane cargoes . Our findings are also consistent with KIF17 functioning as an ancillary motor supporting ciliary cargo delivery . KIF17 is necessary for ciliary localization of an olfactory cyclic nucleotide-gated ion channel ( Jenkins et al . , 2006 ) , but to our knowledge , KIF17 has not been shown previously to function in localizing a GPCR to cilia . Our results identify a specific role of KIF17 , and of KIF17 motor activity , in promoting ciliary concentration of the D1R but not the SSTR3 . Further supporting the proposed role of KIF17 as an anterograde transport motor for D1Rs , we found that D1Rs specifically co-IP IFT57 from intact cells ( Figure 4E ) and verified IFT57 association with KIF17 ( Figure 5A; Insinna et al . , 2008; Howard et al . , 2013 ) . An interesting related observation is that deletion of residues 381–395 in the D1R tail , a manipulation that inhibits D1R ciliary targeting , did not disrupt the D1R-IFT57 interaction , but rather , increased it . This supports the idea that ciliary targeting of D1Rs is likely a complex process . A remarkable finding in the present study is that Rab23 is an essential component of the D1R ciliary targeting mechanism . To our knowledge , this represents the first evidence that Rab23 is required for ciliary localization of any signaling receptor or membrane cargo . Previous studies of other cilia-localized membrane proteins , such as polycystin-2 and Smo , have not observed such a requirement ( Eggenschwiler et al . , 2006; Boehlke et al . , 2010; Hoffmeister et al . , 2011 ) . Thus , we think it likely that Rab23's function in ciliary membrane targeting is specific to a subset of cilia-localized cargoes . We note that ciliary localization of SSTR3 is also sensitive to Rab23 knockdown , even though , in contrast to D1R , its ciliary targeting is insensitive to manipulation of KIF17 motor activity . This provides further support for the existence of receptor-specific differences in the ciliary targeting mechanism . Altogether , the present findings support the conclusion that D1Rs are targeted to the cilium from the extra-ciliary plasma membrane through a complex mechanism involving IFT-B , KIF17 , and Rab23 . Our results support the hypothesis that these components function together in an integrated ciliary delivery pathway and suggest that they have distinguishable functional effects in the delivery pathway . First , disrupting KIF17 motor activity strongly reduced ciliary D1R enrichment without affecting the fraction of receptor-positive cilia . This is potentially consistent with KIF17 motor activity promoting cililary D1R concentration , but not being essential for ciliary D1R access . Second , direct fusion of activated Rab23 to the D1R C-tail can promote ciliary targeting in a manner that is apparently insensitive to KIF17 motor activity . Third , Rab23 knockdown did not disrupt the D1R/IFT-B interaction as estimated by co-immunoprecipitation . These results also support the idea that the lateral delivery mechanism is complex , but additional studies are required to further delineate the precise functions of each of these identified components . The present results raise a number of interesting questions for future study . First , given that receptor accumulation in the ciliary membrane is dependent on Rab23 nucleotide state , an important next question is how this nucleotide state is controlled . A second question , which may be related to the first , is how selective cargo engagement with the ciliary delivery mechanism is determined . The present data strongly suggest a functional interaction between Rab23 and the D1R ciliary targeting determinant , but we have been unable , so far , to establish direct physical interaction between the D1R and Rab23 . It is conceivable that IFT-B links D1Rs and Rab23 , or that unidentified additional protein ( s ) explain the functional interaction observed . A third question is to ascertain precisely how Rab23 directs receptor delivery to the ciliary membrane . Based on precedent of other Rab protein functions , we speculate that there is a specific effector of Rab23 that operates at or near the ciliary diffusion barrier . A fourth question is what broader physiological significance the discrete , Rab23-dependent ciliary targeting mechanism has . We note that a mutation in Rab23 produces excessive Hedgehog signaling in vivo ( Eggenschwiler et al . , 2001 ) . One possibility is that this reflects a disruption of ciliary signaling normally mediated by a Rab23-dependent receptor . Thus , further investigation of the receptor-specific ciliary targeting mechanism identified here may provide fundamental insight into the role of primary cilia as physiological signaling devices and toward understanding pathologies associated with ciliary defects .
IMCD3 cells ( ATCC ) were grown in DMEM/Ham's F-12 Medium supplemented with 10% fetal bovine serum ( UCSF Cell Culture Facility , San Francisco , California , USA ) . Flag-D1R , Flag-DOR , Flag-MOR , and Flag-B2AR constructs were described previously ( Vickery and von Zastrow , 1999; Tanowitz and Von Zastrow , 2003; Yu et al . , 2010 ) . KIF17 and KIF17-DN cDNA ( Jenkins et al . , 2006 ) was a gift from Kristen Verhey ( University of Michigan , Ann Arbor ) . 2XHA-KIF17 was created using PCR and ligation into pIRES . HA-KIF17-DN was created using PCR and ligation into pIRES . 2XHA-KIF17-G234A was generated using site-directed mutagenesis ( Phusion Site-Directed Mutagenesis Kit , Thermo Scientific , Waltham , MA ) . SSTR3-GFP IMCD3 stable cells were a gift from Maxence Nachury ( Stanford University ) . IFT57 cDNA was a gift from Wallace Marshall ( UCSF , San Francisco ) . HA-IFT57 was created using PCR and ligation into pIRES . HA-IFT57-NTM was generated using site-directed mutagenesis ( Phusion Site-Directed Mutagenesis Kit , Thermo Scientific ) of HA-IFT57 to change the site targeted by IFT57-4 siRNA using primers 5′-GTCACCCCAGAGTCTGCGATAGGGTTCTACTAAACACGTGGGCTTCC-3′ and 5′-AGCTGCATGCATGTCCCTGGTCATGTTGC-3′ . Flag-tagged D1-415T , D1-382T , and D1Δ381-395 were created by site-directed mutagenesis ( Phusion Site-Directed Mutagenesis Kit , Thermo Scientific ) . Receptor chimeras , DOR-D1 ( 338–446 ) , DOR-D1 ( 368–446 ) , and DOR-D1 ( 379–400 ) were generated using PCR and homology-directed ligation ( In-Fusion HD Cloning kit , Clontech ) . D1 oligos were fused to DOR residue 340 . Flag-D1-PAGFP was generated using PCR and ligation into p-PAGFP-N1 . Flag-SSTR3 was created using PCR and ligation into pIRES . Rab8a and Arl13b-YFP cDNAs were gifts from Jeremy Reiter ( UCSF , San Francisco ) . Rab23 , Rab23-S23N , Rab23-Q68L , and Rab11 cDNA were gifts from Keith Mostov ( UCSF , San Francisco ) . Flag-Rab23-Q68L was created using PCR and ligation into pIRES . Receptor chimeras , Flag-D1Δ381-395-Rab23 , Flag-D1Δ381-395-Rab23-S23N , Flag-D1Δ381-395-Rab23-Q68L , Flag-DOR-Rab23-Q68L , and Flag-DOR-Rab11-Q70L , Flag-DOR-Rab8-Q67L , Flag-MOR-Rab23-Q68L and Flag-B2AR-Rab23-Q68L , were generated using PCR and homology-directed ligation into pIRES ( In-Fusion HD Cloning kit , Clontech ) . Rab23 constructs were fused to the C-terminal end of D1Δ381-395 with a 2 residue linker . Rab23-Q68L , Rab11-Q70L , and Rab8-Q67L were fused to the C-terminal end of DOR with an 8-residue linker . Transfection of constructs was performed using Lipofectamine 2000 and RNAi-max ( Invitrogen ) for cDNA or siRNA , respectively , in accordance with manufacturer's instructions . Stably transfected cell clones expressing Flag-D1R were generated by selecting for neomycin resistance with 500 μg/ml G418 ( Geneticin , Invitrogen ) . Target sequences for knockdown mIFT57 ( 1: 5′-CAGCAATTGGCTTCTATTAAA-3′ , 2: 5′-TACAATGAATATAGTATTTAA-3′ ) , mIFT172 ( 1: 5′-AAGGAGCATTTACAAGAACAA-3′ , 2: 5′-CCCACAGAATTTCAACATCTA-3′ ) , mRab23 ( 1: 5′-AAGATTGGTGTCTTTAATGCA-3′ , 2: 5′-TAGCCACTAAATGCATGGTAA-3′ ) , and control ( 1027281 , Qiagen ) . Duplex RNA ( 30 pm , Qiagen ) was transfected into 40% confluent cells in a 6-well dish 72 hr before experimentation . Antibodies used were rabbit anti-Flag ( Sigma ) , mouse anti-Flag M1 ( Sigma ) , rat anti-HA ( Roche Applied Science ) , and mouse anti-AcTub ( Sigma ) . Cells expressing indicated constructs were grown to confluency in 10-cm dishes . 48 hr after transfection , cells were lysed in 0 . 2% Triton X-100 , 200 mM NaCl , 50 mM Tris pH 7 . 4 , and 1 mM CaCl2 supplemented with a standard protease inhibitor mixture ( Roche Applied Science ) and cleared by centrifugation ( 12 , 000×g for 10 min ) . Samples were pre-cleared by incubation with mouse IgG agarose ( Sigma ) at 4°C for 30 min . Samples were incubated with anti-Flag M2 affinity gel IgG ( Sigma ) at 4°C for 1 hr , washed with lysis buffer five times , and incubated with SDS sample buffer ( Invitrogen ) supplemented with dithiothreitol to elute proteins . Western immunoblot analysis was performed using rat anti-HA-peroxidase ( Roche ) or rabbit anti-Flag ( Sigma ) . Immunoprecipitation signals were quantified by scanning densitometry of films exposed in the linear range . Linearity was verified by generating a standard curve using a dilution series of the indicated sample . Cells were transfected with the indicated construct ( s ) and then plated on glass coverslips 24 hr later . Cells were grown to confluency to induce ciliation over 24 hr and then fixed . Surface Flag-tagged receptors were labeled by addition of rabbit anti-Flag antibody ( 1:500; Sigma ) to the media for 20 min at 37°C . Cells were then washed with phosphate-buffered saline 2× and fixed by incubation in 4% formaldehyde ( Fisher Scientific ) diluted in PBS for 15 min at room temperature . Cells were permeabilized and blocked in 0 . 1% Triton X-100 and 3% milk diluted in PBS . Primary labeling of AcTub and HA was performed with mouse anti-AcTub ( 1:1000; Sigma ) or rat anti-HA ( 1:1000; Roche Applied Science ) , respectively , for 1 hr . Secondary labeling was performed using donkey anti-rabbit Alexa Fluor 488 ( 1:1000; Invitrogen ) , goat anti-mouse Alexa Fluor 594 ( 1:1000; Invitrogen ) , and goat anti-rat Alexa Fluor 647 ( 1:1000; Invitrogen ) . Specimens were mounted using ProLong Gold antifade reagent ( Life Technologies ) . Fixed cells were imaged by epifluorescence microscopy using a Nikon inverted microscope , 60× NA 1 . 4 objective ( Nikon ) , mercury arc lamp illumination , and standard dichroic filter sets ( Chroma ) . Cells were imaged at 37°C in Dulbecco's Modified Eagle Medium , D-MEM Glucose ( DME H-21 ) , w/o Phenol Red supplemented with 30 mM Hepes . Surface Flag-D1-PAGFP receptors were labeled by addition of mouse anti-Flag M1 antibody ( 1:500; Sigma ) conjugated to Alexa Fluor 555 . Cells were imaged on a spinning disk confocal microscope ( Nikon TE-2000 with Yokogawa confocal scanner unit CSU22 ) using a 100× NA 1 . 45 objective . To photoactivate Flag-D1-PAGFP specifically in the cilium , 405-nm laser illumination was directed through a second light path via a single-mode fiber and focused in the image plane . Photoactivation was achieved by delivering brief ( 100 ms ) pulses of 405-nm illumination . For epifluorescence microscopy of fixed cells , images were acquired using a cooled CCD camera ( Princeton Instruments MicroMax ) with illumination and exposure times adjusted to remain in the linear range of the camera . For spinning disc confocal microscopy of live cells , images were collected at 37°C using an electron multiplying CCD camera ( Andor iXon 897 ) operated in the linear range . Images were processed at full bit depth for all analysis and rendered for display by converting to 8 bit format using ImageJ software ( http://imagej . nih . gov/ij/ ) and a linear lookup table . To measure the receptor fluorescence in the cilium , a region of interest ( ROI ) was manually created by outlining the cilium in the image . To correct for background fluorescence , the ROI was moved to a region outside the cell to measure representative fluorescence . This value was subtracted from the ciliary fluorescence . To measure the D1-PAGFP diffusion in cilium , the total PAGFP fluorescence was measured and normalized to the Flag-555 label to account for accumulation of receptor or focal plane ( orientation ) of the cilium . To estimate lateral mobility of receptors in the cilium , the 405-nm laser spot was positioned so that it illuminated the center of the cilium but not the ends . A single photoactivation pulse was delivered with continuous confocal imaging at 0 . 5 Hz to monitor changes in the distribution of photoactivated Flag-D1-PAGFP in the cilium over time . To estimate new receptor delivery to the cilium , three 405-nm pulses were delivered over 10 s , determined empirically to photoactivate the majority of Flag-D1-PAGFP present in the cilium . We then acquired a subsequent GFP image , delivered another ( single ) 405-nm pulse , and acquired the GFP image again . This sequential ‘image-photoactivate-image’ sequence was applied either immediately ( approximately 30 s ) after the initial 405-nm pulse series or 30 min after . New delivery was estimated by the increment of integrated PA-GFP fluorescence intensity measured after the subsequent 405-nm pulse minus before , normalized to the integrated fluorescence intensity measured after . The anti-Flag Alex555 channel ( unaffected by 405-nm pulses ) was used to optimize focus on the cilium and to verify that the specimen did not move significantly between the ‘before’ and ‘after’ images . To assess the source of Flag-D1-PAGFP delivery to the cilium , we used a similar strategy as described above but quantified in the photoactivation series both PA-GFP and Alexa555 channels and determined their ratio . New receptor delivery from internal relative to plasma membrane sources was distinguished by changes in the PA-GFP/Alexa555 ratio , based on selective labeling with Alexa555 of only the plasma membrane pool , as discussed in text . For line scan analysis , a straight line was drawn on the cilium , and the PlotProfile tool in ImageJ was used to determine the fluorescence intensity along the line . For integrated fluorescence determinations , the rectangular ROI tool was used . Results are displayed as mean of results from each experiment involving imaging of multiple specimens and cilia ( specified in the figure legends ) . Error bars represent standard error of the mean based on at least n = 3 independent experiments done on different days unless noted otherwise . The statistical significance between conditions was analyzed using Student's unpaired t-test , calculated using Prism 6 . 0 software ( GraphPad Software , Inc ) and applying the Hochberg correction for multiple comparisons . The threshold for significance was p < 0 . 05 with exact p value ranges indicated in the figure legends . Surface-accessible Flag immunoreactivity was quantified by fluorescence flow cytometry as described previously ( Tsao and Von Zastrow , 2000 ) . Briefly , cells were dissociated from culture dishes , labeled in suspension 4°C with anti-Flag M1 conjugated to Alexa 647 , and analyzed using a FACS-Calibur instrument ( Becton Dickenson ) . In each experiment , mean fluorescence intensity was determined from 10 , 000 cells and averaged over triplicate determinations for each construct . For all conditions shown , experiments were performed in triplicate , on separate days and from separate transfections . Error bars represent SEM across the experimental days . For determination of mutant receptor surface expression , transiently transfected cells were analyzed 48 hr after transfection . For evaluation of the effects of siRNA knockdown , stably transfected cells were analyzed 72 hr after transfection with the indicated siRNA duplex . | Slender structures called primary cilia protrude from the outer membrane of nearly every human cell . Each cell generally has one cilium , which helps the cell to sense its environment and respond to the signaling molecules sent to the cell to influence its behavior . Proteins called receptors , which are embedded in the surface of the cilia , bind to these molecules and help to transmit information about these signals into the cell . The receptor proteins are not made in the cilia . So far , two broad ways of transporting the receptors to the cilia have been established: some receptors are carried through the interior of the cell inside small membrane-enclosed compartments , and some are pulled from another part of the cell's membrane . However , the exact steps and molecules involved in transporting different receptor types are not fully known . One type of receptor that is commonly found in the cilia of many different cells interacts with a signaling molecule called dopamine . Leaf and Von Zastrow tagged the dopamine receptor with a fluorescent protein that allowed its movement through a cell to be followed under a microscope . This revealed that the dopamine receptor slides into cilia from another part of the cell's membrane . To do so , the dopamine receptor binds to a pair of proteins known to help move receptors around the cell . A third protein called Rab23 then acts as a gatekeeper that allows the dopamine receptor to pass into the cilia . Reducing the amount of Rab23 in a cell prevented dopamine receptors from entering the cilia . Furthermore , Leaf and Von Zastrow found that other receptor proteins incorrectly moved into the cilia if they were bound to Rab23 . Rab23 was already known to play a role in controlling signaling in cilia , and mutations in the Rab23 gene cause disease in humans , but Rab23 was not known to be necessary for delivering receptors to cilia . Future studies to further reveal how Rab23 directs receptors to cilia , and to determine which other ciliary proteins might use this transportation process , will lead to a better understanding of the normal structure and activity of primary cilia . This could lead to new strategies for treating the human diseases that involve defective cilia . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology"
] | 2015 | Dopamine receptors reveal an essential role of IFT-B, KIF17, and Rab23 in delivering specific receptors to primary cilia |
Agouti-related-peptide ( AgRP ) neurons—interoceptive neurons in the arcuate nucleus of the hypothalamus ( ARC ) —are both necessary and sufficient for driving feeding behavior . To better understand the functional roles of AgRP neurons , we performed optetrode electrophysiological recordings from AgRP neurons in awake , behaving AgRP-IRES-Cre mice . In free-feeding mice , we observed a fivefold increase in AgRP neuron firing with mounting caloric deficit in afternoon vs morning recordings . In food-restricted mice , as food became available , AgRP neuron firing dropped , yet remained elevated as compared to firing in sated mice . The rapid drop in spiking activity of AgRP neurons at meal onset may reflect a termination of the drive to find food , while residual , persistent spiking may reflect a sustained drive to consume food . Moreover , nearby neurons inhibited by AgRP neuron photostimulation , likely including satiety-promoting pro-opiomelanocortin ( POMC ) neurons , demonstrated opposite changes in spiking . Finally , firing of ARC neurons was also rapidly modulated within seconds of individual licks for liquid food . These findings suggest novel roles for antagonistic AgRP and POMC neurons in the regulation of feeding behaviors across multiple timescales .
The homeostatic drive to feed is at least partially driven by agouti-related-peptide ( AgRP ) neurons in the arcuate nucleus of the hypothalamus ( ARC ) . These neurons have privileged access to slow hormonal signals of energy balance , such as ghrelin and leptin ( Willesen et al . , 1999 , Morton and Schwartz , 2001 , Zigman and Elmquist , 2003 , Varela and Horvath , 2012 , Wang et al . , 2014 ) , and also receive long-range glutamatergic , GABAergic , and peptidergic synaptic inputs from multiple central brain nuclei , including the paraventricular and dorsomedial hypothalamus ( Krashes et al . , 2014 ) . Both opto- and pharmaco-genetic activation of AgRP neurons drive intense feeding in ad libitum-fed mice ( Aponte et al . , 2011 , Krashes et al . , 2011 ) , while loss-of-function experiments in food-restricted mice lead to a reduction in food consumption ( Gropp et al . , 2005 , Luquet et al . , 2005 , Krashes et al . , 2011 ) . These studies suggest that AgRP neurons represent a critical node in the neural pathway ( or pathways ) linking interoceptive sensing of energy deficit with the decision to seek and consume food . A requirement for pinpointing the precise role of AgRP neurons in driving various aspects of this complex feeding process involves the direct evaluation of their endogenous spiking activity . Previous attempts to directly record spiking activity of AgRP neurons have been restricted to in vitro approaches , due to the technical challenges of extracellular electrophysiological recordings in the ARC in living animals , and the fact that AgRP neurons are intermingled with pro-opiomelanocortin ( POMC ) neurons with opposing effects on food intake ( Varela and Horvath , 2012 , Zhan et al . , 2013 ) . Consistent with the hypothesized enhancement of AgRP neuron firing in times of caloric deficit , in vitro recordings of AgRP neurons in brain slices from mice during their dark cycle or during a period of fasting revealed enhanced action potential firing and spontaneous subthreshold currents as compared to recordings during the light cycle ( Yang et al . , 2011 , Liu et al . , 2012 , Krashes et al . , 2013 ) . Interestingly , opposite effects were observed in satiety-promoting POMC neurons ( Yaswen et al . , 1999 , Aponte et al . , 2011 , Zhan et al . , 2013 ) , which are known to be inhibited by GABA release from AgRP neurons ( Cowley et al . , 2001 , Atasoy et al . , 2012 ) . Studies performed in vitro further suggest that AgRP and POMC neurons exert antagonistic influences , not only on each other's activity ( Cone et al . , 2001 , Yang et al . , 2011 , Atasoy et al . , 2012 ) , but also on the activity of common long-range target nuclei ( Bagnol et al . , 1999 , Cowley et al . , 1999 , Atasoy et al . , 2012 , Atasoy et al . , 2014 ) . However , these in vitro experiments were performed under conditions in which most endogenous circulating factors are absent , and most sources of slow and fast afferent neuronal input are severed . Indeed , in the presence of realistic levels of synaptic inhibition , recordings from de-afferented AgRP neurons show minimal action potential firing in vitro ( Yang et al . , 2011 ) . High-temporal resolution in vivo recordings of spiking activity in identified single neurons in the intact ARC would be necessary to confirm these findings regarding sensitivity to slow changes in energy deficit and could potentially reveal novel roles for AgRP and other ARC neurons in guiding food-seeking and feeding behaviors at shorter timescales . Here , we used an optetrode approach to investigate the in vivo spiking activity of AgRP neurons and a group of nearby neurons inhibited by AgRP neuron photostimulation ( ARCinh neurons ) . Because the only ARC neurons currently known to be inhibited by AgRP neurons are POMC neurons ( Cowley et al . , 2001 , Atasoy et al . , 2012 ) , a large fraction of ARCinh neurons are likely to be POMC neurons . We found that AgRP neuron firing rates increased across hours over the course of the light period . Surprisingly , we also found that AgRP neurons exhibited a sudden and sustained decrease in spiking over the course of minutes , in response to feeding as well as to cues that predicted the availability of food . This decreased level of spiking , which persisted throughout the meal , nevertheless exceeded spiking rates during recordings from ad libitum fed , sated mice at the onset of the light cycle . This abrupt change in spiking cannot simply reflect homeostatic changes but instead suggests that the drop in AgRP neuron spiking reflects a reduction in the drive to seek food at the initiation of food consumption . Finally , we observed that the activity of ARC neurons could be modulated on the timescale of seconds by feeding-related behaviors , including individual licks for liquid food . In general , neurons inhibited by AgRP neuron photostimulation show opposite effects to AgRP neurons . Together , these results suggest that , in addition to sensing slow systemic changes in energy balance , AgRP and POMC neurons may also integrate this information with complex environmental cues regarding food availability and feeding context , in order to dynamically adjust feeding behaviors at timescales from hours to seconds .
We recorded spiking activity from AgRP neurons and other nearby ARC neurons , by selectively expressing cre-dependent channelrhodopsin ( AAV9-FLEX-hSYN-ChR2-mCherry ) in the ARC of Agrp-IRES-Cre mice , and by subsequently performing extracellular optetrode recordings in awake , behaving mice in which putative AgRP neurons were identified via a significant increase in firing during optogenetic photostimulation ( see below ) . Because AgRP neurons are densely packed in the ARC ( Figure 1A ) , we used tetrodes with high impedance that allowed for isolation of large spike waveforms ( Figure 1B ) from neurons proximal to the tetrodes ( 4–8 bundles of 4 wires , <70 µm total diameter per bundle ) . Several neurons could be recorded and discriminated on each tetrode and clustered via differences in spike waveform amplitudes across electrodes within a tetrode ( Figure 1B , C ) . We verified that recordings were located within the ARC; Figure 1A . 10 . 7554/eLife . 07122 . 003Figure 1 . Stable optetrode recordings from arcuate hypothalamic neurons . ( A ) An optetrode was implanted into the arcuate nucleus of the hypothalamus to identify genetically-defined , ChR2-mCherry-expressing agouti-related-peptide ( AgRP ) neurons ( see below and ‘Materials and methods’ ) . Left: coronal section , 1 . 5 mm posterior to Bregma ( inset ) and example histological section , showing AgRP neurons in the ARC ( mCherry expression , red ) , and localization of optetrode recording site ( as determined by visualization of optetrode track ) . White inverted ‘T’ shape denotes location of optetrode track ( vertical line ) and approximate width of optetrode ( horizontal line ) , which estimates the medial-lateral range of potential locations of recorded single-units . Right: schematic showing optetrode locations across 12 mice for which optetrode tracks were recovered . ( B ) Example voltage traces from recordings of spontaneous spiking from one tetrode . Note differences in scale bar across electrode channels , reflecting difference in waveform amplitude across channels . ( C ) Cluster-plots showing discriminability of spikes from different cells using tetrodes . Each dot represents the peak amplitude of a single-spike waveform , measured on three different electrodes within the four-wire tetrode bundle . In this example , each spike waveform was designated as belonging to one of three separable single-units ( colored dots ) , or to multi-unit activity ( gray dots ) . Colors for different single-units match the ticks above the spike traces in B . ( D ) Example of a single-unit defined as a putative AgRP neuron , with peri-photostimulation ( blue lines ) spike raster plot ( top ) , average peri-stimulus time histogram ( PSTH ) across trials ( middle ) , and mean normalized PSTH ( average of individual neuron PSTHs after normalization by pre-pulse-train firing rate ) across all 19 AgRP neurons recorded from 9 ad libitum-fed mice ( bottom ) . Shaded areas denote SEM . ( E ) Raster and PSTH plots ( top , middle ) for an example single-unit defined as significantly and strongly ( >20% ) inhibited by AgRP neuron photostimulation ( ARCinh ) , and mean normalized PSTH ( bottom ) across all ARCinh units in ad libitum-fed mice ( n = 14 ) . ( F ) Firing rate timecourses , in 2-s bins ( gray ) and 10-s bins ( colored ) , for the two example cells in D and E . In ad libitum-fed mice in the absence of food cues or food , AgRP neurons and ARCinh neurons exhibited stable minute-to-minute firing rates across recordings ranging from 30 to 90 min . DOI: http://dx . doi . org/10 . 7554/eLife . 07122 . 00310 . 7554/eLife . 07122 . 004Figure 1—figure supplement 1 . Light-evoked responses in different populations of arcuate neurons . ( A and B ) Activity aligned to the onset of a train of photostimulation ( blue ticks: 20-ms square laser pulses at 20 Hz for 1 s ) . Examples of PSTHs from 6 putative AgRP neurons ( A ) and 6 arcuate neurons that were strongly ( >20% ) and significantly suppressed by AgRP neuron photostimulation ( ARCinh; B ) . ( C ) Phasic entrainment to individual laser light pulses in AgRP neurons . Examples of cycle histograms ( which include firing rates during all pulses from all trials , see ‘Materials and methods’ ) , from four putative AgRP neurons that were phase-locked to light stimulation ( light blue ) . Asterisks denote phases relative to laser pulse onset ( time 0 ) with firing rates that were significantly higher than phase-shuffled control levels ( p < 0 . 0001; see ‘Materials and methods’ ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07122 . 004 We separated ARC neurons into classes by matching spikes with near-identical waveforms obtained during ongoing activity and periods of photostimulation and identifying neurons that were either driven ( putative AgRP neurons , henceforth termed ‘AgRP neurons’ ) or suppressed ( ‘ARCinh neurons’ ) by laser photostimulation ( Lima et al . , 2009 , Cohen et al . , 2012; see below , Figure 1A–E and ‘Materials and methods’ ) . Across 75 daily sessions in 15 mice , we recorded spiking activity in 100 ARC neurons , of which 41 were optogenetically identified as AgRP neurons on the basis of sustained firing increases during photostimulation ( see Figure 1D , Figure 1—figure supplement 1A; see ‘Materials and methods’ for additional details of classification ) . We also recorded activity of 26 nearby ARC neurons that were significantly and strongly suppressed ( by at least 20% ) by photostimulation ( Figure 1E and Figure 1—figure supplement 1B; see also ‘Materials and methods’ ) . Because the only ARC neurons currently known to be inhibited by AgRP neurons are POMC neurons ( Cowley et al . , 2001 , Atasoy et al . , 2012 ) , a large fraction of ARCinh neurons are likely to be POMC neurons . We also recorded from an additional 33 nearby neurons that were unaffected by photostimulation ( ‘ARCother’ ) . All recordings were performed in mice habituated to head restraint ( see ‘Materials and methods’ ) , as this enabled recordings with greater stability from a larger number of electrodes . In a first experiment in ad libitum-fed mice , we measured the firing of ARC neurons during daily 1-hr recording sessions at different phases of the light period , as the stomach is emptying ( Kentish et al . , 2013 ) , levels of ghrelin , a hormone known to increase AgRP neuron activity , are rising ( Tschop et al . , 2000 , Cummings et al . , 2001 , Wang et al . , 2002 , Bodosi et al . , 2004 ) , and minimal feeding is occurring as compared to the subsequent dark period ( Lu et al . , 2002 ) . Stable firing across tens of minutes ( Figure 1F ) allowed reliable estimation of mean firing rate . As predicted by diurnal variations in in vitro AgRP neuron activity ( Yang et al . , 2011 , Krashes et al . , 2013 ) and in ARC expression of Agrp mRNA ( Lu et al . , 2002 ) , AgRP neurons demonstrated a significant , approximately fivefold increase in firing in afternoon vs morning recordings ( p = 0 . 001; n = 10 vs 9 neurons , respectively; Figure 2A ) . In contrast to AgRP neurons , we observed a trend towards decreased firing in afternoon vs morning recordings across all non-AgRP neurons ( p = 0 . 09 , n = 32 neurons ) , with ARCinh neurons showing a similar trend ( p = 0 . 13 , n = 14; Figure 2A ) . 10 . 7554/eLife . 07122 . 005Figure 2 . Arcuate neurons demonstrate changes in firing rate across the light period . ( A ) AgRP neurons ( green dots ) fired significantly more in the afternoon ( when caloric deficiency is increased and the dark period is approaching ) than in the morning ( AM: 1 . 4 ± 0 . 3 Hz , n = 10; PM: 7 . 6 ± 1 . 7 Hz , n = 9; t-test , p = 0 . 001 ) , while all other ARC neurons showed the opposite trend ( AM: 12 . 0 ± 4 . 0 Hz , n = 15; PM: 5 . 3 ± 1 . 1 Hz , n = 17; t-test , p = 0 . 08 ) . ARCinh neurons ( purple dots ) showed a similar trend ( AM: 18 . 5 ± 7 . 6 Hz , n = 7; PM: 5 . 9 ± 2 . 4 Hz , n = 7; t-test , p = 0 . 14 ) . Note the presence of ARCinh neurons with very high mean spiking rates above 30 Hz . ( B ) Same plots as in A , but displaying the rate of short inter-spike interval events ( ISI; spikes occurring <50 ms apart ) in morning vs afternoon recordings . AgRP neurons showed a 14-fold increase in short ISI events ( AM: 0 . 03 ± 0 . 11 Hz; PM: 2 . 7 ± 1 . 1 Hz; t-test p = 0 . 02 ) , while non-AgRP ARC neurons showed a trend toward a decrease in short ISI events in the afternoon ( AM: 8 . 7 ± 3 . 8 Hz; PM: 2 . 2 ± 0 . 7 Hz; t-test , p = 0 . 08 ) ; neurons that were inhibited by photostimulation showed a similar trend ( AM: 15 . 3 ± 7 . 4 Hz; PM: 2 . 9 ± 1 . 7 Hz; t-test , p = 0 . 13 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07122 . 00510 . 7554/eLife . 07122 . 006Figure 2—figure supplement 1 . Characterization of ISI statistics in arcuate neurons . Plot of all ISI distributions ( each row depicts one neuron ) in the ad libitum feeding experiment ( A and B ) and in the instrumental conditioning task ( C ) . Distributions were grouped by class ( green: AgRP; purple: ARCinh; gray: ARCother ) . All distributions were normalized by their peak values . Distributions were relatively broad compared to in vitro recordings , as characterized by coefficients of variation ( defined as the ( mean ) / ( standard deviation ) of the log10 ( ISI ) distribution ) greater than 1 ( coefficient of variation across all neurons recorded in both experiments: AgRP: 1 . 21 ± 0 . 07; ARCinh: 1 . 64 ± 0 . 21; ARCiother: 1 . 51 ± 0 . 13; all values are mean ± SEM; no significant differences across classes , KS-test , p > 0 . 05 ) . Firing rate distributions were tested for bimodality using Hartigan's Dip test; in total , 6/100 ARC neurons ( red asterisks ) had significantly bimodal distributions of ISIs ( p < 0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07122 . 006 AgRP neuron photostimulation of food-seeking and food intake has been shown to be dependent on stimulation frequency , with greater potency at 20 Hz than at 5 Hz or 10 Hz ( Aponte et al . , 2011 ) . Further , the release of peptides is also likely to depend on spike frequency ( Summerlee and Lincoln , 1981 , Aponte et al . , 2011 , Arrigoni and Saper , 2014 , Schone et al . , 2014 ) . Thus , we sought to gain insight into instantaneous spiking frequency in our sample of ARC neurons by considering the distribution of inter-spike intervals ( ISIs ) . Most neurons ( 94/100 ) demonstrated tonic firing with unimodal ISI distributions ( Figure 2—figure supplement 1; Hartigan's Dip Test , p < 0 . 05 ) , suggesting the absence of pronounced burst-like behavior ( van den Top et al . , 2004 ) . In contrast to in vitro recordings , where spiking often demonstrates machine-like regularity , we observed that ISI distributions in vivo were quite broad ( ratio of standard deviation to mean , or coefficient of variation , exceeding a value of 1; see Figure 2—figure supplement 1 for details ) for all three classes of neurons , likely reflecting strong moment-to-moment fluctuations in synaptic input in vivo , as well as differences in the cellular milieu in vivo vs . ex vivo . It is notable that , in the context of recordings from AgRP neurons in free-feeding mice , we rarely observed short ISIs ( <50 ms ) in morning recordings , but observed a roughly 14-fold increase in occurrence of such events in afternoon recordings ( Figure 2B; p = 0 . 02 ) . In contrast , the non-AgRP neurons ( ARCinh and ARCother ) show the opposite trend , with short ISIs occurring more often in the morning than in the afternoon ( p = 0 . 13 ) . Taken together , these in vivo increases in AgRP firing and decreases in ARCinh firing from morning to afternoon recordings are consistent with the homeostatic roles proposed for AgRP and POMC neurons in feeding behavior ( Krashes et al . , 2011 ) . In a second experiment , we recorded activity of ARC neurons in food-restricted mice trained to lick a lickspout to receive high-calorie liquid food ( Ensure; see ‘Materials and methods’ ) . As described in Figure 3A , during each daily session , we recorded spiking activity ( i ) prior to presence of any food-predicting cues ( ‘baseline period’ ) and ( ii ) prior to feeding but following the presentation and positioning of the lickspout in front of the snout , and ( iii ) during a period of at least 45 min in which the mouse could lick to receive food rewards . Across sessions , the onset of food availability had a variable delay following lickspout placement in order to disambiguate ARC responses to initiation of feeding ( Figure 3 ) from any pre-feeding responses to food-associated cues ( Figure 4 ) . The main findings are illustrated in two example ARC neurons ( Figure 3B , C ) . We observed a dramatic decrease in firing rate in the AgRP neuron ( Figure 3B ) when comparing the 5-min baseline period before lickspout placement ( orange dashed line ) to the 45-min period following onset of food consumption ( red dashed line ) . By contrast , we observed a large increase in firing in the example ARCinh neuron ( Figure 3C ) . 10 . 7554/eLife . 07122 . 007Figure 3 . Arcuate neurons are modulated on the timescale of minutes by feeding . Following instrumental conditioning for liquid food rewards ( Ensure ) in food-restricted mice , we recorded arcuate neuron changes during feeding . ( A ) Experimental paradigm . First , baseline spiking was recorded for at least 5 min . A lickspout was then positioned close to the snout . After a variable duration ( 0 . 5–15 min ) , food was made available , at which point licking resulted in a delivery of 10 μl of liquid food . Typically , the mouse continued to eat for at least 45 min , beginning with almost continuous licking and gradually transitioning to sparser feeding bouts ( see below ) . ( B ) An example AgRP neuron demonstrating a fast and sustained decrease in firing within minutes of presentation of a lickspout ( orange vertical dashed line; see also Figure 4A ) and access to food ( maroon vertical dashed line ) . Dots above x-axis signify 10-s bins in which licking occurred . Gray trace: firing rate in 2-s bins; colored trace: 10-s bins . Significant decreases in firing were observed from pre-lickspout baseline to the periods following access to food ( p < 0 . 001 for early- , mid- , and late-feeding periods ) . ( C ) Similar to B , but for an example ARCinh neuron that demonstrated significant increases in firing post-feeding onset ( p < 0 . 001 , for early- , mid- , and late-feeding periods; see also Figure 4A ) . ( D ) Timecourses of increases ( red ) , decreases ( blue ) , or no reliable change ( white ) in firing from pre-lickspout baseline ( gray vertical dashed line ) for each cell recorded during this task ( n = 49 ) . For visualization purposes , this plot employs a normalized index called the area under the Receiver Operating Characteristic Curve ( auROC; see ‘Materials and methods’ ) . Short vertical black lines denote the onset of food availability . Example neurons in B and C are denoted by ‘B’ and ‘C’ , respectively . ( E ) Proportion of cells recorded that responded with a significant ( two-sample KS-test , p < 0 . 025 ) increase ( red ) , decrease ( blue ) , or with no change in firing at 0–5 min ( left ) , 5–15 min ( middle ) , and 15–45 min ( right ) post-feeding onset . Data include 22 AgRP neurons , 12 ARCinh neurons , and 15 ARCother neurons from 5 mice . ( F ) Comparison of auROC values , across AgRP , ARCinh , and ARCother ( green , purple , and gray , respectively ) neurons , during early- , mid- , and late-feeding periods ( left , middle , and right panels , respectively ) . Left: bar plot showing averaged auROC ( a value of 0 . 5 reflects no change in distributions of firing rate ) . For early- , mid- , and late-feeding periods , mean auROC for ARCinh ( early: 0 . 68 , mid: 0 . 75 , late: 0 . 68 ) is significantly larger than those of AgRP ( early: 0 . 35 , mid: 0 . 33 , late: 0 . 38; Analysis of variance , p = 0 . 046 , 0 . 00004 , 0 . 014 , respectively ) . Error bars denote SEM . Right: cumulative distribution of auROC values across neurons in each class , for all feeding periods . The abscissa value at an ordinate of 50% indicates the median auROC for each class . DOI: http://dx . doi . org/10 . 7554/eLife . 07122 . 00710 . 7554/eLife . 07122 . 008Figure 3—figure supplement 1 . Different feeding effects in separate populations of arcuate neurons . ( A and B ) Comparisons of auROC firing modulation index ( A; modulation index: ( 5–15 min post ensure—baseline ) / ( 5–15 min post ensure + baseline ) ) and firing rate ( B ) mid feeding ( 5–15 min post-feeding onset ) as compared to baseline , for AgRP neurons ( green ) , ARCinh neurons ( purple ) , and ARCother neurons ( gray ) . More darkly colored lines represent significantly modulated neurons ( KS-test; see Figure 2D ) . Population changes in absolute firing rate ( mean: black horizontal bars; SEM: gray ) appeared generally consistent with changes at the level of single neurons ( Figure 3B , C ) but were significant only for the ARCinh class ( AgRP neurons: p = 0 . 16; ARCinh neurons , p = 0 . 01; ARCother neurons; p = 0 . 8; paired sample t-test ) , likely due to the large variability in baseline firing rates ( see main text ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07122 . 00810 . 7554/eLife . 07122 . 009Figure 3—figure supplement 2 . Neurons recorded across multiple days show similar firing changes to food-associated cues and feeding . In two mice , we recorded the same AgRP neuron in two subsequent sessions during the feeding task . ( A ) Similarity of spike waveforms in the four recording channels across day 1 ( top ) and day 2 ( middle ) . Similarity of spike waveforms ( bottom ) from prior to ( black ) and during ( blue ) AgRP photostimulation . Inset , one individual spike waveform from each epoch . ( B ) Ratio of waveform amplitudes across electrode channels was also similar . ( C ) Spiking activity during day 1 and 2 of recording from the same AgRP neuron demonstrated strikingly similar responses across days to both lickspout placement ( orange dashed line ) and to the onset of feeding ( maroon dashed line ) . Clear feeding-related drops in firing occurred that were distinguishable from the earlier drops following lickspout placement . ( D–F ) Example data from a second AgRP neuron . Same notations as in A–C . Note partial recovery following initial post-lickspout drop in firing , but not following post-feeding drop in firing . DOI: http://dx . doi . org/10 . 7554/eLife . 07122 . 00910 . 7554/eLife . 07122 . 010Figure 4 . Many ARC neurons are modulated within minutes following food cue presentation , but prior to feeding . ( A ) Spiking activity of the same cells shown in Figure 3B , C , zoomed-in to illustrate the drop in spiking in response to presentation of a food cue ( lickspout placement near the snout; orange dashed line ) but prior to onset of food delivery ( maroon dashed line ) . Top: AgRP neuron showing a significant food cue-induced decrease ( two-sample KS-test , p < 0 . 001 ) ; Bottom: ARCinh neuron showing a significant food cue-induced increase ( p < 0 . 001 ) . ( B ) Proportion of cells recorded that responded with a significant ( two-sample KS-test , p < 0 . 025 ) increase in firing ( red ) , decrease in firing ( blue ) , or with no change in firing ( gray ) following lickspout placement but prior to feeding . ( C ) Comparison between averaged auROCs of AgRP , ARCinh , and ARCother neurons ( green , purple , and gray , respectively ) for the period between lickspout placement and feeding ( cf . Figure 3F ) . Left: bar plot of mean auROC ( 0 . 5 indicates no change in a neuron's distribution of spike rates ) across classes . Mean auROC for ARCinh ( 0 . 67 ) is significantly larger than that of AgRP ( 0 . 42 ) ( Analysis of variance , p = 0 . 0046 ) . Error bars denote SEM . Right: cumulative distribution of auROC values for all ARC classes . DOI: http://dx . doi . org/10 . 7554/eLife . 07122 . 010 Abrupt and sustained decreases in AgRP neuron firing , and increases in ARCinh neuron firing , were also observed at the population level . We visualized firing changes at the population level as follows: because neurons in our sample exhibited a range of baseline firing rates and firing rate variability ( Figure 3—figure supplement 1 ) , we computed an index of reliable changes in firing rate between the baseline period ( 5 min ) prior to lickspout placement , and a 1-min sliding window of time following this baseline period . This index , called the area under the Receiver Operating Characteristic ( auROC; see ‘Materials and methods’ ) , has a value of 0 ( blue ) in periods where a neuron's firing has decreased robustly ( i . e . , the distribution of firing rates at the tested period is entirely below the firing rate distribution of the baseline period ) , a value of 1 ( red ) in periods where firing has increased robustly , and a value of 0 . 5 for periods with no change in firing ( i . e . , where the distribution of firing rates in the test period is indistinguishable from baseline ) . Using this index , we observed that AgRP neurons typically showed a rapid and sustained decrease in firing ( Figure 3D ) when comparing pre-lickspout baseline firing ( prior to gray vertical bar ) with firing rates in a sliding window following onset of feeding ( black vertical bars ) . By contrast , ARCinh neurons , a class that likely includes POMC neurons , typically showed a rapid and sustained increase in firing ( Figure 3D ) . We quantified these feeding-related changes from baseline firing ( pre-lickspout placement ) in three windows of time: ( i ) between 0–5 min post-feeding onset , a time that likely precedes most nutrient absorption and counter-regulatory visceral and hormonal changes ( e . g . , de Araujo et al . , 2008 ) , ( ii ) between 5–15 min post-feeding onset , a time when mice were still licking for food at a maximal rate but when systemic changes may begin to occur ( de Araujo et al . , 2008 ) , and ( iii ) between 15–45 min post-feeding onset , a time when licking for food decreased and became more sporadic ( see dots above x-axis in Figures 3B , C ) . During each analysis window , we tested for significant differences in mean firing between baseline and post-feeding firing rates ( by K-S tests; using distributions of firing rates from each 5 s bin in each period ) , and we quantified population estimates of the auROC index described in Figure 3D ( which reflects discriminability of firing rate distributions between pre-lickspout and post-feeding periods ) . Surprisingly , during early feeding ( 0–5 min post-feeding onset ) , mean firing rates were significantly different than baseline in 82% ( 40/49 ) of all ARC neurons ( Figure 3E ) . In particular , 64% ( 14/22 ) of AgRP neurons had significantly decreased mean firing during this period , while only 23% ( 5/22 ) of AgRP neurons had increased mean firing ( Figure 3E ) . The average auROC index across all AgRP neurons was significantly less than 0 . 5 ( i . e . , a discriminable decrease in distribution of firing rates; Figure 3F , top left panel; p = 0 . 016 ) . Indeed , auROC indices for most AgRP neurons were <0 . 5 during this period , as illustrated in the cumulative distributions plot ( green distribution in Figure 3F , bottom left panel ) . In contrast to these decreases in firing , ARCinh neurons demonstrated opposite changes during this early pre-feeding period , as 66% ( 8/12 ) of ARCinh neurons showed a significant increase in firing , while only 17% ( 2/12 ) showed a significant decrease ( see Figure 3E; mean auROC significantly above 0 . 5 , p = 0 . 01 , see Figure 3F ) . These drops in firing rate in AgRP neurons , and increases in ARCinh neurons , generally persisted across later feeding periods ( Figure 3E , F; see also Figure 3—figure supplement 1 ) . In general , feeding-induced changes from pre-lickspout baseline period remained consistent but began to abate by 15–45 min post-feeding onset , a time of decreased licking and consumption ( Figure 3E; see also Figure 3F , bottom panels; 53% of all 49 ARC neurons showed significant changes in mean firing at 15–45 min after feeding onset , vs 82% and 86% of ARC neurons during earlier post-feeding periods ) . Notably , AgRP neurons with higher initial firing rates prior to lickspout placement were more likely to show a larger subsequent drop in firing after feeding ( e . g . , at 5–15 min post-feeding onset: r = −0 . 54; p = 0 . 0091; Figure 3—figure supplement 1B ) . It is possible that certain AgRP neurons may have already decreased their activity prior to the start of recording , due to other contextual food-associated cues appearing at each session's onset . As such , our findings likely provide a conservative estimate of the number of AgRP neurons with cue- and food-related decreases in firing ( see ‘Discussion’ ) . In general , these findings demonstrate clear and opposite changes in spiking activity in the majority of AgRP and ARCinh neurons during feeding behavior , consistent with recent studies measuring changes in calcium activity in ARC neurons during feeding ( Betley et al . , 2015 , Chen et al . , 2015 ) . In addition to these changes in firing , our data provided additional information regarding absolute activity levels following the onset of feeding . Interestingly , while feeding reduced firing rates , it did not abolish firing in AgRP neurons , and firing rates remained elevated relative to certain ad libitum conditions . Specifically , we found that the spiking rate of AgRP neurons recorded in food-restricted mice following refeeding ( 15–45 min after onset of feeding , whether during epochs that involve or do not involve licking , 5 . 9 ± 1 . 3 and 5 . 8 ± 1 . 3 spikes/s , respectively; N = 22 neurons ) was significantly higher than firing rates measured in free-feeding mice at the onset of the light cycle following night feeding ( 1 . 4 ± 0 . 3 spikes/s; N = 10; p = 0 . 03 in both cases ) . These data suggest that the early , feeding-related drop in AgRP neuron activity represents a partial drop in firing that does not reach the lower levels of firing exhibited by these neurons in the absence of caloric deficiency . We next assessed the possibility of even earlier changes in firing , immediately after presentation of the food-predicting cue ( lickspout placement near the snout ) but prior to onset of feeding . As shown in Figure 4A , B ( zoomed-in plots of same example AgRP and ARCinh neurons as in Figure 3B , C ) , a closer inspection of this period ( between the orange and maroon dashed lines ) suggests that ARC firing rates may begin to change from baseline before feeding has even started , likely due to the presentation of the food-predicting cue ( lickspout placement ) . Indeed , almost half of all ARC neurons recorded ( 45% , 22/49; 2 sample KS-test , p < 0 . 025 ) showed significant changes in mean firing in the 5-min period following lickspout placement , but prior to feeding . As with post-feeding changes , 32% ( 7/22 ) of AgRP neurons showed a significant ‘anticipatory’ drop in firing , while only 5% ( 1/22 ) showed a significant increase . By contrast , 50% ( 6/12 ) of ARCinh neurons showed significant anticipatory increases in firing , while none showed a significant drop . Similar effects were evident in the auROC change index , plotted in Figure 3D ( in the period following the black vertical dashed line , but prior to the black vertical bar in each row ) and quantified in Figure 4C . Thus , a large subset of AgRP neurons decrease spiking activity in the minutes following presentation of a food-predicting cue , while ARCinh neurons showed the opposite effect . These findings suggest that AgRP neurons decrease their firing upon identification of an upcoming source of food , either to decrease the drive to continue the search for food ( see ‘Discussion’ ) , or in anticipation of future meal-induced restoration of caloric deficit . While we ensured that neurons recorded across daily sessions were distinct ( by electrode adjustment and analysis of spike waveforms ) , in two cases , we did record from what appeared to be the same AgRP neuron in back-to-back sessions ( including the example neuron in Figures 3B , 4A; based on similarity of waveforms and amplitudes across channels of the same tetrode across days , see Figure 3—figure supplement 2A , B , D , E ) . These examples illustrate that our findings of rapid decreases in AgRP neuron firing , both prior to and following the onset of feeding , were highly consistent for the same neuron when assessed across days . Interestingly , the second example neuron ( Figure 3—figure supplement 2C , F ) was recorded during two sessions involving long imposed durations between lickspout placement and commencement of feeding ( orange and maroon dashed vertical lines ) . On both days , this AgRP neuron showed a transient drop following lickspout placement that partially recovered over minutes , followed by a more sustained drop following onset of feeding ( Betley et al . , 2015 , Chen et al . , 2015 ) . Thus , such stable recordings of the same ARC neurons across days should be possible using our approach ( see e . g . , Kentros et al . , 2004 , Siegle and Wilson , 2014 , Thorn and Graybiel , 2014 ) and can provide valuable insights into slow and fast motivational changes . Optogenetic stimulation of AgRP neurons drives voracious feeding ( Aponte et al . , 2011 ) . Previous studies , however , were performed in freely moving mice with access to solid food . We confirmed that AgRP photostimulation in a headfixed mouse trained to lick for liquid Ensure also drives feeding . First , at the end of feeding experiments ( described above ) , when mice were sated ( defined by voluntary abstinence from licking for liquid food ) , we repeated the photostimulation procedure to confirm identification of recorded cells . Here , we investigated whether this AgRP photostimulation would induce additional licking for liquid food ( n = 11 sated sessions from 4 mice , including all sessions in which the mouse did not lick for Ensure in the 3 min preceding laser stimulation onset ) . Mice showed a marked increase in licking behavior after photostimulation onset compared to the period preceding photostimulation ( Figure 5A; p = 0 . 03 ) . In a separate experiment , we habituated another cohort of mice to head-fixation and trained them ( for 1–2 days , under mild food restriction ) to lick for Ensure , followed by at least 5 days with ad libitum access to food . Under these free-feeding conditions , which more faithfully approximate published data on AgRP-driven feeding behavior ( Aponte et al . , 2011 , Krashes et al . , 2011 ) , mice consistently increased licking behavior and food consumption in response to AgRP photostimulation ( Figure 5B , C; pre-photostim . : 12 ± 2 μl/min; during photostim . : 67 ± 5 μl/min; post-photostim . : 11 ± 2 μl/min , mean ± SEM across 5 mice; pre-photostim vs during photostim: p = 0 . 004 , during photostim vs post-photostim: p = 0 . 015; paired t-test ) . 10 . 7554/eLife . 07122 . 011Figure 5 . Optogenetic activation of AgRP neurons promotes licking behavior and food consumption in head-restrained mice . ( A ) Mice that have been fed Ensure to satiety ( see Figures 3 , 4 ) subsequently increased licking for Ensure in response to optogenetic photostimulation of AgRP neurons . ( B ) Ad libitum-fed mice also increased licking for Ensure in response to optogenetic photostimulation of AgRP neurons , even when head-restrained ( Top: individual single-session examples from 5 separate mice; Bottom: mean ± SEM of lick rate from 5 mice ) . ( C ) Ad libitum-fed mice increased food consumption in response to optogenetic stimulation of AgRP neurons , even when head-restrained ( asterisks indicate paired t-tests , p<0 . 02 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07122 . 011 These feeding-related changes at the timescale of minutes led us to ask whether ARC neurons can be modulated at the even faster timescale of seconds , by individual licks and/or by bouts of licking preceded by several seconds without licking ( see Figure 6; Chen et al . , 2015 ) . If such modulation existed , it could imply that AgRP neurons and other ARC neurons might contribute to momentary fluctuations in the drive to seek and/or consume food , even at the level of licking microstructure ( Davis , 1989 ) . 10 . 7554/eLife . 07122 . 012Figure 6 . Arcuate neurons are modulated on short timescales by licking activity . We evaluated whether ARC neuron firing could be modulated by bouts of licking for food ( defined by >8 s without licking , followed by a burst of >3 licks ) and/or by the occurrence of an individual lick ( timescale of <1 s ) . ( A ) Example firing rate traces ( gray: 2-s bins; green/purple: 10-s bins ) and licking ( orange ) from an AgRP neuron ( left ) and an ARCinh neuron ( right ) that both showed positive correlations between firing rate and licking bouts . ( B ) Top: raster plots of spiking ( black ticks ) and licking ( orange ticks ) for four example neurons , aligned to the onset of a licking bout . Bottom: average firing ( mean ± SEM ) relative to bout onset for AgRP neurons ( green ) and ARCinh neurons ( purple ) . Dark orange traces are average lick rates . Note that firing of some cells appears linearly related to frequency of individual licks ( e . g . , third panel from left ) , while firing in other cells appeared more strongly modulated by bout onset ( first and fourth panles from left ) . Several AgRP neurons ( e . g . , second panel from left ) showed a reliable decrease in firing at bout onset . ( C ) We estimated the degree of modulation of firing across time relative to an individual lick ( gold traces ) , and relative to an individual bout onset ( orange ) , using multiple linear regression . Asterisks indicate times of significant modulation , relative a lick or lick bout ( F-test , p < 0 . 002 , corrected for multiple comparisons; see ‘Materials and methods’ ) . ( D ) Population distribution of times relative to a single lick ( gold ) or lick bout ( orange ) at which significant modulation of firing ( asterisks in C ) occurred across 44 cells . Note that many neurons began changing their firing before the onset of a lick or lick bout ( gray vertical lines ) , and that modulation mostly occurred within ±1–2 s of onset of a lick or lick bout , demonstrating modulation of ARC neuron firing at a surprisingly fast timescale . ( E ) Proportions of neurons in each class that were significantly modulated by individual licks ( gold ) , lick bouts ( orange ) , both ( red ) , or neither ( gray ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07122 . 01210 . 7554/eLife . 07122 . 013Figure 6—figure supplement 1 . Modulation of firing by licking across classes of arcuate neurons . ( A ) Distribution of times relative to lick onset ( left ) or lick-bout onset ( right ) where ARC neuron firing was significantly modulated , as in Figure 3D but plotted separately for AgRP neurons ( green ) , ARCinh neurons ( purple ) , and ARCother neurons ( gray ) . ( B ) Linear relationship ( kernel estimate from multiple linear regression , see ‘Materials and methods’ ) of ARC firing rate modulation ( in 0 . 5-s bins ) surrounding lick onset ( left ) or lick-bout onset ( right ) , as in Figure 3C but plotted for all neurons exhibiting significant modulation for at least one time point ( F-statistic for regression , p < 0 . 002 , corrected for multiple comparisons across time points ) . Colors indicate cell class , as in A . The left panel indicates , for example , that a putative POMC neuron ( purple arrow ) increased its firing , on average , by approximately 2 . 4 spikes in the 0 . 5-s bin at which an individual lick occurs , while an AgRP neuron ( green arrow ) decreased its firing , on average , by 1 . 3 spikes in the 0 . 5 s prior to the occurrence of each individual lick . ( C ) To provide an estimate of the sign and relative magnitude of the modulation of ARC firing by lick events , we estimated the time point ( within ±2 s of the lick or bout onset ) with the larger significant modulation coefficient , for all traces in B . We then normalized this coefficient to the mean firing rate of the cell ( offset estimate from regression analysis ) to provide an estimate of relative modulation of firing by licking for individual cells . The data show that licking and onsets of lick bouts could decrease or , more often , increase ARC neuron firing , with lick bouts increasing firing by up to ∼300% ( three fold ) , and individual licks by up to ∼50% . DOI: http://dx . doi . org/10 . 7554/eLife . 07122 . 01310 . 7554/eLife . 07122 . 014Figure 6—figure supplement 2 . Feeding-related effects are observed independent of licking behavior . ( A ) As in Figure 3D , timecourses of increases ( red ) , decreases ( blue ) , or no reliable change ( white ) in firing from pre-lickspout baseline ( gray vertical dashed line ) are plotted for each cell recorded during this task . For visualization purposes , this plot employs a normalized firing index called the auROC; see ‘Materials and methods’ . Short black lines denote the onset of food availability . In the left panel , all time bins within 5–10 s of any lick are grayed out , to better visualize feeding-related responses not contaminated by licking effects . In the right panel , all firing changes data surrounding any lick events are plotted . Note that sustained decreases ( blue ) in AgRP neuron firing , and increases ( red ) in ARCinh firing , appear consistent irrespective of licking behavior . ( B ) Comparisons of absolute ( top ) and normalized ( bottom ) firing rates , in the 5-min pre-lickspout placement , compared to 15–45 min post-feeding onset , calculated using only periods without licking activity . More darkly colored lines and circles represent significantly modulated neurons ( KS-test ) . Population changes in firing ( mean: black horizontal bars; SEM: gray ) were not significant for any class ( AgRP neurons: p = 0 . 15; ARCinh neurons , p = 0 . 11; ARCother neurons; p = 0 . 37 ) , likely due to within-class variability in baseline firing . Pie charts ( right ) demonstrate the number of neurons with increased ( red ) , decreased ( blue ) , or no significant change ( gray ) in activity pre-lickspout compared to 15–45 min post-feeding onset including only periods without licking . ( C ) Same as B , but only including periods containing licking activity . In this case , mean firing was significantly increased in ARCinh neurons ( p = 0 . 04 ) , while mean changes in other cell classes were not significant ( AgRP: p = 0 . 31; ARCother: p = 0 . 64 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07122 . 014 Modulation of firing by bouts of licking ( defined as periods of intense licking for food , preceded by at least 8 s without licking ) is clearly evident in the example neuron in Figure 6A . We focused analyses on the period from 15 to 45 min post-feeding onset , when initiation of licks and lick bouts was more sporadic ( see Figure 3 ) . Firing activity aligned to each lick bout for this and three other neurons ( see spike raster plots and average peri-bout time histograms in Figure 6B ) revealed that some neurons appear particularly modulated at lick bout onset , while others appear to change their firing proportionally to the number of individual licks at any time ( third example from left ) . We estimated whether each neuron was significantly modulated by licks and/or by lick bouts , using multiple linear regression between timecourses of firing rate , licking , and lick bouts ( using 0 . 5-s bins; see Figure 6C ) . For the four example neurons , yellow traces in Figure 6C ( left panels ) show changes in firing rate relative to the moment that a lick occurred ( asterisks denote significant firing modulation , F-test , p < 0 . 002 , corrected for multiple comparisons ) . Similarly , orange traces in Figure 6C ( right panels ) show changes in firing relative to onset of a bout of licking for liquid food . At the population level , almost half of ARC neurons were significantly modulated by individual licks and/or lick bouts ( Figure 6D , E ) , with most significant lick- or bout-induced changes in firing ( asterisks in Figure 6C ) occurring approximately 1 s before and/or after lick/bout onset ( Figure 6D and Figure 6—figure supplement 1A , B ) . The modulation of firing at times substantially preceding the onset of a lick or lick bout suggests that these changes do not only reflect efference motor copies but could partially help drive initiation of upcoming licks . These analyses show that firing rates could be influenced by licking events . However , this rapid source of modulation did not influence the main conclusions in Figure 3 regarding feeding-related decreases in firing in most AgRP neurons , or increases in firing in most ARCinh neurons . These findings remained intact when considering only epochs with or without lick events ( Figure 6—figure supplement 2 ) . Feeding-related changes in firing near the end of the meal persisted in the absence of any licking or consumption , suggesting that these changes do reflect actual persistent changes in motivational drive . Of the 5 AgRP neurons that were significantly modulated by lick bouts , 2/5 showed reliable decreases in firing within seconds of the onset of a licking bout ( Figure 6B , C , second neuron from left ) , similar to the decrease in firing observed over the course of several minutes during the initial onset of feeding behavior ( Figure 3 ) . While the rest of our small sample of significant bout-modulated AgRP and ARCinh neurons tended to increase their firing at the time of bout onset ( AgRP: 3/5 increased , ARCinh: 2/2 increased; Figure 6—figure supplement 1C ) , future studies will be required to confirm whether populations of AgRP vs ARCinh neurons , on average , also show opposite , lick-related changes in firing at the 1-s timescale . One potential means to gain additional insight into the question of whether AgRP and ARCinh neurons show opposite changes in firing at the 1-s timescale , and whether AgRP neurons show synchronized activity at this timescale , is to evaluate simultaneously recorded pairs of neurons . Overall , we found significant but modest pairwise correlations in firing rate , at the 1-s timescale , in 40/44 pairs of simultaneously recorded ARC neurons ( see Figure 7A , B and legend ) . While 3/3 paired recordings from a simultaneously recorded AgRP neuron and ARCinh neuron showed significant correlations , correlation coefficients were modest ( Figure 7A , <0 . 2 ) . Similarly , while 7/10 pairs of AgRP neurons showed significant correlations , these were also quite modest ( mostly <0 . 2 ) . Thus , the correlations in endogenous activity across ARC neurons differ significantly from the hypersynchronous correlations at this timescale that likely occur across ARC neurons during periodic , 1-s duration photostimulation often employed in vivo ( Figure 1—figure supplement 1 ) . 10 . 7554/eLife . 07122 . 015Figure 7 . Endogenous correlations between simultaneously recorded pairs of arcuate neurons . ( A ) Correlation coefficients ( at zero time-lag ) across pairs of simultaneously recorded neurons within a class or across classes of ARC neurons . While 7/10 pairs of AgRP neurons showed significant correlation coefficients ( p < 0 . 05 ) , the correlations were modest ( all <0 . 3 ) . Similar results were observed for other pairs . To ensure that correlation coefficients did not simply reflect slow concurrent changes in firing across neurons , we first removed slow trends in firing from each cell's spike-rate timecourse ( slower than ∼100 s , by high-pass filtering firing rate timecourses above 0 . 01 Hz ) . ( B ) To examine the timescale of correlation between pairs of neurons , we calculated correlations between pairs at lags up to ±20 s . Most pairs with significant correlations ( black lines ) peaked near zero time-lag , with correlations falling off by 5 s of lag . These data suggest that pairs of ARC neurons within and across classes can show modest but significant correlations or anti-correlations at the timescale of ∼1 s . DOI: http://dx . doi . org/10 . 7554/eLife . 07122 . 015
In this study , we have developed a means for stable extracellular recordings from individual neurons in the arcuate nucleus of the hypothalamus in awake mice . Using this approach , we demonstrated dynamic changes in spiking activity in classes of ARC neurons across multiple timescales ( Figure 8 ) . 10 . 7554/eLife . 07122 . 016Figure 8 . ARC neurons are modulated on multiple timescales . We observed slow , likely homeostatic , changes in ARC neuron activity , consistent with the established role for ARC AgRP and pro-opiomelanocortin ( POMC ) neurons in regulating energy balance . However , we also demonstrated fast changes , on the order of minutes and even seconds , in response to feeding , food-associated cues , and licking behavior . Differential modulation of these opposing populations of neurons on these timescales may enable both homeostatic and more rapid adjustment of downstream circuits that underlie complex feeding-related behaviors , including food-seeking and food consumption . PVH: paraventricular hypothalamus; DMH: dorsomedial hypothalamus; LH: lateral hypothalamus . DOI: http://dx . doi . org/10 . 7554/eLife . 07122 . 016 First , consistent with in vitro studies , we found an increase in spiking of putative AgRP neurons from morning to afternoon recordings in ad libitum-fed mice ( Figure 2 ) , despite the fact that food intake in mice is minimal during this period ( Lu et al . , 2002 ) . Because the observed increase in endogenous afternoon firing in AgRP neurons does not drive feeding , additional circuits may exist downstream of AgRP neurons ( Figure 8 ) that prevent significant daytime feeding until dark period onset , at which point this ‘drive state’ is released and food-seeking and feeding rapidly occur ( Lu et al . , 2002 ) . The slow changes in firing that we observed across the light cycle as caloric deficit increases are consistent with prior studies employing indirect measures of AgRP activity ( Lu et al . , 2002 , Ellis et al . , 2008 ) , as well as with in vitro recordings from AgRP neurons across times of day ( Yang et al . , 2011 , Krashes et al . , 2013 ) . Similarly , the sustained decrease from baseline spiking in AgRP neurons at 15–45 min following onset of feeding ( Figure 3 ) was consistent with previous studies reporting a decrease in AgRP cFos activity 2 hr after scheduled refeeding ( Tan et al . , 2014 ) or after post-fast refeeding ( Becskei et al . , 2009 ) . Second , in the context of the instrumental feeding task in food-restricted mice , our direct measurement of absolute spiking rates revealed that while spiking in AgRP neurons is reduced following refeeding , even in epochs late in the meal when mice are not licking for liquid food , firing persists and remains elevated ( by approximately four fold ) as compared to low levels of spiking observed in AgRP neurons recorded from ad libitum-fed mice in the early stages of the light cycle ( ∼5 . 9 ± 1 . 3 spikes/s vs ∼1 . 4 ± 0 . 3 spikes/s ) . These data further support the notion that additional circuits downstream of AgRP neurons may prevent additional feeding , even in contexts where AgRP neurons continue to fire at intermediate spike rates . Moreover , these data suggest that the rapid decrease in AgRP firing , associated with food cues and meal-initiation ( Figures 3 , 4; Betley et al . , 2015 , Chen et al . , 2015 ) , may be separable from the firing that persisted at intermediate spiking rates during the meal , which may be associated with homeostatic signals of residual negative energy balance . We found that AgRP neurons responded with rapid and persistent decreases in activity to both a food-predicting cue ( lickspout placement near the snout ) and to feeding onset , and that such decreases were sustained at later periods ( e . g . , 15–45 min post-feeding onset ) . In contrast to AgRP neurons , ARCinh neurons—including putative POMC neurons—responded with fast and persistent increases in firing to both food-predicting cues and food reward . Furthermore , acute modulation of ARC firing rates even occurred down to a timescale of approximately 1 s . Such seconds-long modulations could occur prior to , as well as during , individual licks or bouts of licking for food , which themselves elicited a transient decrease in firing in several AgRP neurons . These fast changes in firing that we observed on the timescale of minutes to seconds in classes of ARC neurons were surprising , as these neurons were presumed , until recently , to be mainly driven by slow homeostatic signals ( Lu et al . , 2002 , Ellis et al . , 2008 , Kim et al . , 2014 ) . However , as reviewed by Berridge 2004 , many eating behaviors may ‘co-opt or pre-opt the cue-depletion detectors that trigger hunger in emergency cases of real deficit’ . In the case of AgRP neurons , initial evidence of non-homeostatic influences was suggested based on observations of a decrease in AgRP neuron c-Fos expression 2 hr after consumption of a calorie-free meal ( Becskei et al . , 2009 ) , though limited temporal resolution and an uncertain relationship between c-Fos and in vivo firing limit the interpretation of these data . More recently , in vivo fiber photometry was used to monitor bulk changes in population calcium activity , pooled from many ARC AgRP neurons ( Chen et al . , 2015 ) , while another recent study used epifluorescence measurements of calcium activity in individual AgRP neurons in awake mice ( Betley et al . , 2015 ) . Consistent with our spiking data from individual neurons , these studies observed fast reductions in AgRP neuron calcium activity ( within <1 s ) during presentation of food-predicting cues and of food ( Betley et al . , 2015 , Chen et al . , 2015 ) . Taken together , our direct measurements of spiking activity in AgRP neurons are generally consistent with recent findings observed using measurements of calcium activity in the ARC ( see below for further comparison of these different technical approaches ) . The full complement of feeding-related behaviors includes sensing caloric deficiency , initiation of food-seeking , consummatory activity , and , ultimately , cessation of feeding . While manipulations of several different brain areas can produce initiation or cessation of food consumption , it remains unknown whether neurons exist whose endogenous in vivo spiking activity parallels the slowly increasing and rapidly terminating drive state associated with food seeking . The motivational drive to find food grows during periods of increasing energy deficit ( Saper et al . , 2002 ) . During appropriate contexts , this drive should bias actions towards food-seeking , as demonstrated in rodents with increased exploration at the time of scheduled feeding ( Moran and Tamashiro , 2007 , Mistlberger , 2009 , Tan et al . , 2014 ) . This behavior persists until a source of food is secured , at which point seeking must stop so that feeding can begin ( Craig , 1917 , Berridge , 2004 ) . Our findings of slow diurnal increases and fast feeding-related decreases in AgRP neuron firing support the hypothesis that AgRP neurons may also encode a food-seeking drive that can be transiently shut off when food becomes available ( Tan et al . , 2014 ) , possibly to avoid further ‘appetitive phase’ food-seeking in lieu of consummatory behavior ( Craig , 1917 , Berridge , 2004 ) . Similar to our recordings in AgRP neurons , previous studies involving a neural circuit that regulates drinking reported rapid reductions in firing of vasopressin-secreting supraoptic neurons during drinking of water ( Arnauld and du Pont , 1982 , Stricker and Hoffmann , 2007 ) . These findings suggest that rapid reductions in firing of neurons promoting specific motivational drives during consumption may be a general mechanism . It is also possible that the fast modulation in spiking activity represents the anticipation of future meal-induced restoration of caloric deficit . To further distinguish between contributions of AgRP neurons to appetitive , consummatory , and/or anticipatory feeding behaviors , it will be instructive , in future studies , to ( i ) employ larger delays between delivery of food rewards ( Cohen et al . , 2012; currently inter-reward interval was ≥ 2 . 5 s ) , ( ii ) use lever pressing rather than instrumental licking in order to distinguish requests for food from food consumption ( Histed et al . , 2012 ) , and ( iii ) use virtual reality methods in head-fixed mice ( Dombeck et al . , 2007 ) to study simulated foraging activity with precise behavioral control and monitoring . Our demonstration that AgRP neuron activity recorded in awake , head-restrained mice shows similar changes as observed in freely moving mice ( Betley et al . , 2015 , Chen et al . , 2015 ) , and that activation of AgRP neurons in head-fixed mice has similar effects on consummatory behaviors as previous studies in freely behaving mice ( Aponte et al . , 2011 , Krashes et al . , 2011 ) , sets the stage for using head-fixation as a means to probe this feeding circuit during well-controlled behaviors with precise behavioral monitoring , large numbers of trials , and easier experimental access to stable , dense recordings using electrophysiology or two-photon calcium imaging with fewer weight restrictions . In contrast to AgRP neurons , ARCinh neurons , a group that likely includes POMC neurons that exert opposite effects to AgRP neurons on common downstream targets , demonstrated concomitant increases in firing ( Figure 8 ) that may act synergistically with decreases in AgRP firing to more effectively drive these common targets . The recent observations of a rapid increase in population POMC calcium activity in response to food cues , to sustained feeding , as well as during brief individual bouts of licking of liquid food ( Chen et al . , 2015 ) are consistent with our spiking data in ARCinh neurons . Optogenetic photostimulation of POMC neurons results in reduced food consumption , although this effect is only evident at longer timescales ( Aponte et al . , 2011 , Zhan et al . , 2013 ) . Our data suggest that at least some POMC neurons may act on faster timescales to reduce the drive to seek food . It remains an open question whether these opposing changes in firing are primarily driven by direct inhibition by AgRP neurons ( Cowley et al . , 2001 , Atasoy et al . , 2012 ) , or by additional sources of fast input to these neurons ( see above ) . Future studies can begin to elucidate which of the local and long-range sources of fast synaptic input to AgRP and POMC neurons ( Figure 8; Krashes et al . , 2014 ) are responsible for various fast and slower changes in firing that we have observed , using optetrode recordings from these inputs as well as by combining ARC recordings with methods for selective silencing of each source of input . The current study benefited from single-spike resolution to monitor spiking of AgRP neurons in awake mice , down to a timescale of milliseconds . This allowed sub-second measurements of correlations between ARC neuron firing and lick microstructure , as well as sub-second correlations in firing between ARC neurons . Optetrode methods present some challenges , particularly in the ARC: the optetrode ( ∼300 μm in diameter , with 4–8 70 μm diameter tetrodes extending into the ARC ) must penetrate approximately 5 . 5 mm into the brain to gain access to the relatively small ARC nucleus , likely resulting in lower yield . Similar to previous optetrode studies ( e . g . , Cohen et al . , 2012 ) , some actual AgRP neurons may be classified as ARCother neurons in cases of insufficient photostimulation . While a previous study found that AgRP neurons almost exclusively decrease their calcium activity upon food presentation ( Betley et al . , 2015 ) , we only observed significant decreases in spiking in approximately two-thirds of AgRP neurons and noted increases in firing in 14–23% of AgRP neurons . These differences may reflect real heterogeneity across AgRP neurons , as previous studies have demonstrated anatomically and functionally distinct populations of AgRP neurons in the ARC ( Betley et al . , 2013 ) . Alternatively , experimental differences ( e . g . different states of baseline meal expectation across paradigms ) and/or a possible misclassification of a small number of non-AgRP cells as AgRP neurons could explain the heterogeneity in our data ( for additional discussion , see ‘Materials and methods’ ) . As discussed above , recordings of GCaMP6 activity in populations of ARC neurons ( Chen et al . , 2015 ) and single cells ( Betley et al . , 2015 ) provide powerful complementary approaches to the method described here . Recordings using calcium indicators allow chronic monitoring of calcium activity in AgRP neurons , albeit with some caveats including temporal resolution , estimation of relative contributions of spike-evoked vs intracellular release of calcium , and the potential for neuropil contamination during single-cell recordings . By contrast , our current approach provides robust sensitivity to single action potentials , information regarding inter-spike interval structure of tonic vs bursting ARC neurons , and the capacity to record from both AgRP neurons and other ARC neurons simultaneously . In addition , our recordings of absolute spiking levels ( in contrast to relative changes in calcium activity ) revealed differences in baseline firing across times of day and across satiety states following refeeding as compared to free-feeding conditions . Together , these mutually informative studies demonstrate consistent changes in the activity of classes of ARC neurons across timescales and feeding behaviors , leading to refinements in the hypotheses regarding the potential roles of these neurons in feeding , food-seeking , valence coding and reinforcement learning , and suppression of competing drive states ( Betley et al . , 2015 , Chen et al . , 2015 , Garfield et al . , 2015 , Palmiter , 2015 , Seeley and Berridge , 2015 ) across multiple timescales and behavioral paradigms . In summary , our recordings of spiking activity from identified classes of ARC neurons in awake mice provide direct support for previous hypotheses , based on in vitro recordings and in vivo manipulations , of slow , opposite changes in AgRP and POMC neuron spiking across hours , during slow changes in energy balance . In addition , our data suggest that AgRP and POMC neurons can be modulated on timescales inconsistent with a purely homeostatic role in feeding . In particular , the rapid drop in spiking activity of AgRP neurons at meal onset may reflect a termination of the drive to seek sources of food , while residual , persistent spiking in these neurons may reflect a sustained drive to consume food .
All animal care and experimental procedures were approved by the Beth Israel Deaconess Medical Center Institutional Animal Care and Use Committee . We used 12 adult male mice , heterozygous for Cre recombinase under the control of the Agrp gene ( Agrp-Ires-cre; Tong et al . , 2008 ) . Animals were housed at 22°C–24°C on a 12:12 light/dark cycle ( light cycle: 6:00 am to 6:00 pm ) with standard mouse chow and water provided ad libitum , unless specified otherwise . To selectively express channelrhodopsin-2 ( ChR2 ) in AgRP neurons , we injected Agrp-Ires-cre mice with 200 nl of adeno-associated virus , serotype 9 , carrying an inverted ChR2-mCherry flanked by double loxP sites ( UPenn Vector Core , Philadelphia , PA ) into the arcuate nucleus of the hypothalamus ( ARC; coordinates relative to Bregma: anterior-posterior , −1 . 50 mm; dorsal-ventral , −5 . 80 mm; lateral , 0 . 25 mm ) . 3 weeks after viral injection , mice were prepared for awake , head-fixed electrophysiology recordings by surgical implantation of a head post and an optetrode microdrive ( See ‘Optetrode electrophysiology’ section , below; see also Cohen et al . , 2012 ) as follows: first , mice were anesthetized using isoflurane in 100% O2 ( induction , 3%–5%; maintenance , 1%–2% ) and placed into a stereotaxic apparatus ( Kopf , Model 940 Small Animal Stereotaxic Instrument with Digital Display Console ) on a heating pad ( CWE ) . Ophthalmic ointment ( Vetropolycin ) was applied to the eyes . Using procedures identical to those described previously ( Goldey et al . , 2014 ) , a two-pronged head post was affixed to the skull using C&B Metabond ( Parkell; cat . no . 242-3200 ) , and a 0 . 5-mm diameter burr hole was drilled over the mouse ARC . The optetrode was then implanted with distal electrode tips ending well above the arcuate nucleus ( 4 . 8 mm ventral to Bregma ) , and the implant was secured in place using a light-cured glue ( FLOW-IT ALC part #N11VH , Pentron Clinical ) around the craniotomy , followed by metabond and dental cement ( Grip cement kit , powder and solvent; Dentsply; cat . no . 675570 ) . Analgesia ( 0 . 5 mg/kg meloxicam , s . c ) was administered post-operatively and on the following day . Following recovery from surgery , mice were habituated to tolerate 1–2 hr of head restraint ( typically requiring 3–4 days ) . Note that physiology studies in mice and primates commonly employ head-restraint , which enables more precise control and monitoring of feeding behavior and other behavioral and neurophysiological parameters ( Paz et al . , 2003 , Niell and Stryker , 2010 ) . It was previously shown that stress responses largely normalize after 3–4 days of habituation to restraint stress ( Ma and Lightman , 1998 ) . Furthermore , the fast changes in spiking activity we report cannot be explained by stress , since spontaneous activity over the course of tens of minutes ( during experiments in Figure 2 in mice fed ad libitum , see below ) was relatively stable under near-identical conditions . A cohort of 9 mice , used to investigate diurnal rhythms of ARC neurons ( Figure 2 ) , had free access to food and maintained normal , ad-libitum feeding weight . In a typical session , recordings began with spontaneous activity for ∼5 min followed by a laser stimulation protocol ( ∼5–10 min ) . We then recorded spontaneous activity as long as recordings were stable . To minimize stress , head-fixation of habituated mice was restricted to <3 hr . Therefore , we restricted our daily recordings to either morning or afternoon , and advanced the electrodes between recordings to ensure that different neurons were recorded in each daily session . A cohort of 5 mice , used to investigate feeding effects on ARC neurons ( Figures 3–6 ) , was maintained between 85 and 90% of ad-libitum weight ( median across mice: 87 . 5%; averaged weight fluctuation within a mouse across sessions: 3 . 2 ± 1 . 8% ) . Mice were trained to consume a high-calorie liquid meal replacement ( Ensure ) from a lickspout , while head-fixed on a spherical treadmill . Licking was detected via disruption of an infrared beam positioned in front of the lickspout . Upon detection of a lick , a 10 µl drop of Ensure was released using a solenoid and MonkeyLogic software ( Asaad and Eskandar , 2011 ) in Matlab . After ∼2–4 days of training , food-restricted mice would readily consume large quantities of Ensure . In a typical session , recordings began with spontaneous activity for ∼5 min followed by a laser photostimulation protocol ( ∼5–10 min ) . After a variable period of time ( 5–10 min ) , a lickspout was positioned in front of the mouse's snout . After an additional variable period of time ( 3–15 min ) , Ensure was made available , and the mouse subsequently engaged in instrumental licking for food reward ( delivered in 10-µl increments ) . We then recorded activity as long as the mouse continued to eat ( typically 1 hr , with consumption of 3–5 ml of Ensure , equivalent to 4 . 5–7 . 5 Cal ) . The hour-long recording typically includes ∼30 min of constant feeding , during which mice lick almost continuously and delivery of Ensure was contingent on detection of a lick . Of note , a minimal duration between Ensure drops was defined to be 2 . 5 s , even in periods of near-continuous licking behavior . This period of continuous feeding was then typically followed by more sparse bouts of feeding ( Figure 6A , orange highlights ) . If the mouse stopped drinking for a substantial period of time , the software delivered a drop of Ensure to encourage additional licking . However , these Ensure drops constituted <15% of all drops delivered . When feeding further diminished , a second round of laser stimulation was performed to help with cell identification ( Figure 5A ) . Mice were additionally given chow in their home cage ( 1 . 5–2 g , given between the hours of 6–8 PM ) to help maintain a weight of approximately 87% of free-feeding weight . We confirmed that the chow was fully consumed by 8 AM the following morning , such that the subsequent recordings took place after at least 8 hr without access to food . In practice , since 2 g of chow is typically consumed within 4–5 hr , it is more likely that these recordings took place following at least 16 hr without feeding . Note that we deliberately randomized the time until lickspout placement and the time from lickspout placement until the onset of Ensure availability , as well as occasionally interleaving feeding sessions with sessions of head-fixation without feeding , in order to avoid having the onset of lickspout placement and food delivery be perfectly predictable by prior cues ( e . g . , initial head restraint prior to recording ) , though it is likely that such cues nevertheless induced ‘predictive’ changes in firing . A cohort of 5 mice was used to investigate whether optogenetic activation of AgRP neurons induced food consumption under head-fixed conditions ( Figure 5 ) . Mice were trained to consume Ensure from a lickspout , while head-fixed on a spherical treadmill . To this end , mice were first habituated for 1–2 days to head-fixation on the trackball . They were then mildly food restricted until their body weight decreased to 95% of their free-feeding weight . After another 1–2 days of habituating to feeding while head-fixed , mice were returned to their free-feeding weight . After at least 5 additional days , mice were tested . Each session began with spontaneous licking and consumption of up to 0 . 5 ml Ensure . After 30 min of baseline activity ( during which mice rarely licked; Figure 5B ) , we photostimulated AgRP neurons for 30 min ( 20-ms pulses at 20 Hz; 1 s on / 3 s off ) , followed by an additional 30 min without photostimulation . During the session , licking and the delivery of Ensure were continuously recorded . We recorded extracellularly from multiple neurons simultaneously using a custom-built , 200-μm diameter optic fiber-coupled microdrive ( an ‘optetrode’ ) with between four and eight manually constructed tetrodes ( comprising of 4 twisted strands of electrode wire ) attached to the sides of the fiber ( Cohen et al . , 2012 ) . In the days following optetrode insertion ( to depth of 4 . 8 mm ventral to Bregma ) , and prior to the beginning of recordings , tetrodes were gradually lowered by 0 . 5 mm ( 0 . 15 mm/day ) . All tetrodes were glued to the fiber with epoxy , such that the ends of the tetrodes were 400–600 μm beyond the end of the fiber . Each tetrode was then gold plated ( Gold NC Solution , Neuralynx Inc ) , to reach a final impedance between 500 and 800 KΩ . Neural signals and time stamps for behavior were recorded using a Digital Lynx SX recording system ( Neuralynx ) . Broadband signals from each wire , filtered between 0 . 1 and 9000 Hz , were recorded continuously at 32 kHz . To extract the timing of spikes , signals were band-pass filtered between 400 and 6000 Hz . Spikes were detected whenever a signal crossed a selected threshold value . For each electrode , the threshold was defined as four times an estimated noise level . The standard deviation of the background noise was estimated as the median of the absolute value of the band-pass filtered recording , divided by 0 . 6745 ( Donoho and Johnstone , 1994 , Quiroga et al . , 2004 ) . Spike waveforms for each electrode within a tetrode ( a 1-ms epoch around each time stamp ) were extracted using a broadband signal ( 300–9000 Hz ) sampled at 32 kHz . This ensured that minimal information about spike waveform was lost to additional filtering . Waveform spikes were then sorted offline using Neuralynx spike sorting software ( SpikeSort ) as follows: first , each spike waveform consisted of a 1-ms window surrounding its peak amplitude . Second , for each spike , we defined two features , amplitude of the peak and amplitude of the valley , for each of the four electrodes within a tetrode ( a total of 8 features ) . Clusters were then defined according to these feature distributions , manually selecting the dimensions that best separated different clusters . We used several criteria to include a neuron in our data set . First , we inspected the ISI distribution . A histogram of the ISI distribution for the spikes within each cluster is expected to show a refractory period , that is , a dearth of spikes that occur within milliseconds of each other ( Hill et al . , 2011 ) . Therefore , only clusters in which none of the ISIs were less than 1 ms and less than 5% of the ISIs were smaller than 5 ms were considered for further examination as candidates for single-units , thus ensuring minimal contamination . Second , the clustered waveforms were also inspected by eye to exclude those with aphysiological shapes . The waveform shape and amplitude were examined across the duration of the recording to ensure stability and reject the possibility of contamination by multiple neurons or potential loss of a neuron at an intermediate time within the recording . Finally , we performed cross-correlation between each spike waveform and the averaged waveform , and specified that the averaged correlation coefficient must exceed 0 . 95 . To ensure stable recordings , we confirmed that the correlation coefficients between spikes in the first and last 5 min of recordings were not significantly different than those between the same number of randomly selected spikes across the recording . Recording sites were also verified histologically with electrolytic lesions at the termination of the experiment , when possible , using 15–20 s of 100 μA direct current , or by visualizing the optical fiber track ( Figure 1A ) . We adapted recent methods for optogenetic identification ( Lima et al . , 2009 , Cohen et al . , 2012 , Kravitz et al . , 2013 ) of well-isolated single-units , to classify these units into three categories , as follows . First , to classify AgRP neurons , we delivered blue-light photostimulation pulses at 20 Hz , a stimulation frequency shown to elicit feeding and as well as sustained spiking in ChR2-expressing AgRP neurons in vitro ( Aponte et al . , 2011 ) . Specifically , we delivered 1-s-long trains of 20-ms light pulses at 20 Hz ( wavelength: 473 nm; intensity: 5–20 mW/mm2 ) , with 3 s between pulse trains ( typically 50–100 trains were used at a given laser intensity ) . The laser beam was passed through a Pockels cell to ensure accurate control of laser pulse shape ( <0 . 2 ms timing accuracy ) and amplitude ( calibrated with a power meter and a photodiode ) . To ensure that spontaneous and light-evoked waveforms originated from the same cell , we validated that the correlation coefficients of the cross-correlations between spontaneous and light-evoked waveforms were above 0 . 95 and were not significantly different than the correlation between pairs of spontaneous spike waveforms ( see also Figure 3—figure supplement 2 ) . To determine whether a neuron showed a significant light-evoked response , we used a paired sample t-test comparing firing rates in the 2 s prior to a 20-Hz pulse train with the first half or the second half of the pulse train ( p < 0 . 025 , corrected for number of tests ) . This method was chosen because some clearly driven neurons showed more pronounced excitation after a delay of several 100's of milliseconds ( see Figure 1—figure supplement 1; peri-stimulus time histograms show binned firing rates relative to laser train onset; estimated with 100-ms bins ) . For neurons that fired significantly below pre-train baseline ( inhibited by AgRP neuron photostimulation ) according to the above t-test , we added an additional criterion that the cells be suppressed by greater than 20% relative to baseline , which removed a subset of weakly but significantly inhibited cells ( 7% of all recorded cells ) . This class of cells was labeled ‘ARCinh’ . Cells not significantly modulated according to the t-test were assigned to the ‘ARCother’ category . Finally , we noticed that in a small subset of recordings ( <10% ) , the initial trial of a 1-s laser pulse train in the series of 1-s trains could lead to a sharp increase or decrease in firing that did not return to pre-photostimulation baseline until 0–2 min after the end of the final laser photostimulation trial . We reasoned that these effects were clearly laser-evoked ( two-sample Kolmogorov–Smirnov ( KS ) test , p < 0 . 05 ) , and thus , we also used this information in our classification . This additional criterion only changed the cell classification in 7% ( 2/33 ) of AgRP neurons and 12% ( 3/25 ) of ARCinh neurons and did not affect the main conclusions of the study . While histology showed reasonably high penetrance of ChR2 expression in AgRP neurons , the ARCother category may include a small subset of AgRP neurons lacking sufficient or any ChR2 expression , or that an insufficient intensity of light reached the tetrode on which the unit was recorded . Note that ARCother and ARCinh neurons were only included from recordings during or subsequent to identification of a putative AgRP neuron , to ensure that no neurons from regions dorsal to the ARC were included . While many AgRP neurons showed classical entrainment to the pulse train at 20 Hz , some clearly laser-driven AgRP neurons did not show strong entrainment . Neurons with low excitability and low-spontaneous firing rates in vivo may be unsuitable for identification protocols demanding entrainment at high frequencies ( Kravitz et al . , 2013 ) . This may be the case with AgRP neurons , whose excitability we found to be low in the morning ( in ad libitum-fed mice , Figure 2 ) and following feeding ( in food-restricted mice , Figure 3 ) . Further , the intrinsic membrane properties of AgRP neurons have themselves been shown to be state dependent ( Baver et al . , 2014 ) , thus providing an additional source of in vivo variability that may affect entrainment , but that is not present during most in vitro recordings ( e . g . , Aponte et al . , 2011 ) . Moreover , AgRP neurons might also not show entrainment due to other cell-intrinsic mechanisms that would only be influential in certain in vivo contexts , in the presence of strong , summating synaptic input ( Jo et al . , 2005 ) . To quantify fast entrainment of spiking activity to each laser pulse , 50-ms cycle histograms were calculated in 5-ms bins ( Figure 1—figure supplement 1C ) . To determine significant laser entrainment , a shuffling procedure was applied to spikes during individual laser cycles ( 20 ms of laser stimulation followed by 30 ms of no laser stimulation ) . We created a distribution of shuffled cycle histograms by shuffling the spikes within a given cycle 5000 times while maintaining the same total number of spikes per cycle . We compared the firing rate in our cycle histogram of individual 5-ms bins with the distribution of shuffled cycle histograms to determine if any bins were significantly modulated by photostimulation ( p < 0 . 0001 , corrected for number of neurons and number of bins ) . To identify whether ARC neurons were tonic firing , or had burst-like behavior involving occasional , short ISIs between longer ISIs , we used the Hartigan's dip test of unimodality on the distribution of the logarithm ( to base 10 ) of each ISI ( Figure 2—figure supplement 1; p < 0 . 05 ) . To quantify the changes in firing rate ( estimated in 5-s bins ) during the feeding paradigm , auROC timecourses ( Cohen et al . , 2012 ) were calculated for each cell ( Figure 3D ) . This analysis compares the distribution of firing rates during a baseline period ( up to 5 min prior to lickspout placement ) with the distribution of firing rates post baseline , using a sliding window of 1 min . This analysis quantifies how discriminable these two distributions are . For example , if the two distributions of firing rates are completely non-overlapping , the auROC reflects an estimate of 1 ( clear increase in firing; all post-baseline firing rate values are larger than all baseline firing rate values; red in Figure 3D ) or 0 ( clear decrease in firing; all post-baseline firing rate values are smaller than all baseline firing rate values; blue in Figure 3D ) , while an auROC estimate of 0 . 5 indicates that the distribution of baseline and post-baseline firing rates is indistinguishable ( white in Figure 3D ) . This analysis effectively normalizes the responses across a population , allows concurrent visualization of neurons with very different firing rates , accounts for local firing rate variability , and speaks to the reliability of the difference between baseline and post-baseline time windows . To quantify the percentage of AgRP , ARCinh , and ARCother neurons modulated by placement of the lickspout or feeding , we compared the distribution of firing rates ( estimated in 5-s bins ) before lickspout placement ( up to 5 min prior to lickspout placement ) with the distribution of firing rates post-lickspout placement ( ‘food cue predictive responses’; up to 5 min post lickspout , but only including time bins prior to food availability; Figure 4 ) or post-Ensure delivery ( early-feeding responses: 0–5 min following onset of Ensure availability; mid-feeding response: 5–15 min following onset of Ensure availability; late-feeding response: 15–45 min following onset of Ensure availability; Figure 3 ) via a two-sample KS-test ( Figure 3E; p < 0 . 025 ) . For analyses of the relationship between firing rate and licks or lick bouts , we use multiple linear regression analysis . Simply stated , this analysis assumes a linear relationship between a neuron's firing and the occurrence of individual licks or lick bouts and estimates this relationship . The main advantage of this approach over the generation of lick-triggered ( or bout-triggered ) average firing rate histograms is that it assesses the impact of each lick or lick bout , irrespective of the occurrence of other prior or future licks or bouts . The analysis determines the linear ‘kernel’ ( one for the relationship between firing and licking , another for the relationship between firing and lick bouts ) —a set of coefficients that adjust the firing rate up or down at each moment in time relative to the occurrence of each individual lick or lick bout . If the relationship is indeed linear , then one should be able to perfectly predict the moment-by-moment firing of the neuron as the sum of the following three terms: ( i ) a fixed constant firing rate ( in spikes/s ) , ( ii ) the convolution of the ‘lick’ kernel ( units: spikes/s/lick; Figure 6C , left panels ) with each individual lick , and ( iii ) the convolution of the ‘lick bout’ kernel ( units: spikes/s/lick; Figure 6C , left panels ) with each individual lick bout . Note that these kernels include coefficients at times both prior to and following onset of the lick or lick bout , to estimate changes in firing that precede and follow the licking event , respectively . The F-statistic assesses whether a given kernel coefficient at a given time relative to lick/bout onset explains a significant amount of variance . P-values were corrected for multiple comparisons ( p < 0 . 05/19 = 0 . 0026 , corrected for time bins , reflecting the 19 time bins at 2-Hz sampling rate , from −4 . 5 s to 4 . 5 s relative to lick or lick bout onset ) . All statistical tests and analyses were performed using Matlab . At the conclusion of recordings , which lasted between 10 and 60 days , we performed histological verification of the recording site . In a subset of mice ( 5/12 ) , an electrolytic lesion was made 400 µm above the final recording location by passing a mild current between two electrodes ( 25 mA for 30 s ) . Mice were given an overdose of tribromoethanol , perfused with 10% formalin , and brains were cut in 50-µm coronal sections . Sections were stained with 4′ , 6-diamidino-2-phenylindole ( DAPI ) to visualize nuclei . Recording sites , identified by the presence of the fiber tract and/or electrolytic lesion , were all verified to be among ChR2-mCherry-expressing AgRP neurons . | Appetite is controlled in part by the opposing actions of the ‘hunger hormone’ ( called ghrelin ) and the ‘satiety hormone’ ( called leptin ) . Ghrelin is released by the stomach when empty and stimulates appetite , whereas leptin is released by fat stores and induces feelings of fullness . Both hormones travel via the bloodstream and are detected by a region of the brain called the hypothalamus . Ghrelin and leptin act specifically on a group of cells in the hypothalamus that contains at least two major cell types: AgRP neurons and POMC neurons . Electrode recordings from slices of mouse brain show that AgRP neurons fire more rapidly at night—when mice normally feed—than during the day , whereas POMC neurons do the opposite . This suggests that the activity of AgRP neurons drives food-seeking behavior , whereas POMC firing inhibits it . However , the absence of circulating hormones such as leptin and ghrelin in brain slices makes it difficult to draw firm conclusions about the role of these cells in controlling appetite . Mandelblat-Cerf , Ramesh , Burgess et al . have addressed this issue by performing the first recordings of spiking activity in individual AgRP neurons and other cells that are likely to be POMC neurons in awake mice . Consistent with the results of slice experiments , the firing rate of AgRP neurons increased steadily over the course of the day , suggesting that their activity signals an increasing need for food . Furthermore , as soon as food became available , the firing rate of the AgRP neurons suddenly dropped—even though the animals' energy reserves would still have been low . These results are consistent with the findings of two recent studies reported earlier this year that used different methods to indirectly measure neuronal activity in awake mice . Notably , even after the drop in activity , the firing rates of AgRP neurons remained above those recorded in fully sated mice—which possibly reflects the fact that the animals' energy reserves were still low . The putative POMC neurons generally showed opposite effects to the AgRP neurons . The results of these electrode recordings in awake mice thus suggest that AgRP and POMC neurons together maintain a drive to seek out food sources as energy reserves fall , and to refrain from doing so when energy reserves are plentiful . Moreover , the seemingly paradoxical drop in AgRP firing and increase in POMC firing upon receiving food may act as a signal to temporarily stop searching for food , so that feeding itself can begin . Alternatively , since the release of satiety hormones after eating a meal is slow , these rapid changes in firing may provide more immediate feedback to the neuronal circuits that regulate the drives to seek and consume food . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2015 | Arcuate hypothalamic AgRP and putative POMC neurons show opposite changes in spiking across multiple timescales |
Trimethylamine-oxide ( TMAO ) is present in seafood which is considered to be beneficial for health . Deep-water animals accumulate TMAO to protect proteins , such as lactate dehydrogenase ( LDH ) , against hydrostatic pressure stress ( HPS ) . We hypothesized that TMAO exerts beneficial effects on the circulatory system and protects cardiac LDH exposed to HPS produced by the contracting heart . Male , Sprague-Dawley and Spontaneously-Hypertensive-Heart-Failure ( SHHF ) rats were treated orally with either water ( control ) or TMAO . In vitro , LDH with or without TMAO was exposed to HPS and was evaluated using fluorescence correlation spectroscopy . TMAO-treated rats showed higher diuresis and natriuresis , lower arterial pressure and plasma NT-proBNP . Survival in SHHF-control was 66% vs 100% in SHHF-TMAO . In vitro , exposure of LDH to HPS with or without TMAO did not affect protein structure . In conclusion , TMAO reduced mortality in SHHF , which was associated with diuretic , natriuretic and hypotensive effects . HPS and TMAO did not affect LDH protein structure .
Some clinical studies have shown that increased levels of trimethylamine-oxide ( TMAO ) in the plasma are associated with an increased risk of adverse cardiovascular events ( Tang et al . , 2015; Trøseid et al . , 2015; Tang et al . , 2013 ) . However , other studies have not confirmed this relationship ( Meyer et al . , 2016; Yin et al . , 2015; Stubbs et al . , 2019 ) . Furthermore , basic research data regarding the effect of TMAO on the circulatory system are contradictory ( Huc et al . , 2018; Aldana-Hernández et al . , 2019; Collins et al . , 2016; Organ et al . , 2016; Savi et al . , 2018 ) . In the plasma , TMAO originates from the liver oxidation of trimethylamine ( TMA ) , a product of gut bacteria metabolism of l-carnitine and choline ( Zeisel and Warrier , 2017; Ufnal et al . , 2015 ) . However , another direct source of TMAO in humans is TMAO-rich seafood ( Cheung et al . , 2017; Yancey and Siebenaller , 2015 ) . Therefore , populations whose diet are rich in seafood , such as the Japanese , have higher urine TMAO concentrations than those that do not , for example , Americans ( Dumas et al . , 2006; Holmes et al . , 2008 ) . Interestingly , prevalence and mortality rates of heart failure ( HF ) are lower in Japan compared to the US or Europe , despite the fact that Japan has the highest proportion of elderly people in the world ( Nagai et al . , 2018; Ogawa et al . , 2007 ) . Marine animals living in deep water , and thus exposed to hydrostatic pressure stress ( HPS ) , accumulate TMAO ( Yancey and Siebenaller , 2015; Ma et al . , 2014; Yancey and Siebenaller , 1999; Sarma and Paul , 2013; Yin et al . , 2017 ) . Data from biophysical experiments suggest a protective effect of TMAO on cell proteins exposed to HPS . For example , TMAO has been found to stabilize teleost and mammalian lactate dehydrogenase ( LDH ) , a complex tetramer protein that plays an essential role in cellular metabolism ( Yancey and Siebenaller , 1999 ) . Notably , in a heart exposed to catecholamine-induced stress or in a hypertensive heart , diastolic-systolic changes in pressure may exceed 220 mmHg . These fluctuations in pressure can happen in humans up to 200 times per minute . These numbers are even higher in small animals . These events may produce an environment in which the hydrostatic pressure changes approximately 100 000 times in 24 hr . However , the effect of HPS produced by the contracting heart on cardiac proteins is obscure . Recently , we hypothesized that TMAO may benefit the circulatory system by protecting cardiac LDH exposed to HPS produced by diastole-systole-driven changes in the hydrostatic pressure of the contracting heart ( Ufnal and Nowiński , 2019 ) . Here , we investigated whether a continuous , 12-months-long oral administration of TMAO in healthy Sprague-Dawley rats ( SD ) , in SD exposed to catecholamine stress , and in animal model of heart failure ( HF ) with reduced ejection fraction ( SHHF ) exerts beneficial effects on the circulatory system . Furthermore , we evaluated whether TMAO protects the protein structure of cardiac LDH exposed to HPS . To examine the effects of HPS in the context that mimics the environment produced by the contracting heart , we developed a novel experimental system using microfluidics chambers with piezoelectric valves and pressure controllers .
In general SD rats showed no pathological findings ( Table 1 , Figures 2 and 3 , and Figure 3—figure supplement 1 ) . In general , the SHHF animals showed characteristics of hypertrophic cardiomyopathy with compromised systolic function including substantially increased heart mass and plasma NT-proBNP , decreased stroke volume and ejection fraction as well as lung edema ( Table 2 , Figure 2 ) . Histological evaluation of the heart revealed dilated cardiomyopathy that is a moderate increase in the diameter of cardiomyocytes , enlargement of the nucleus and a reduction of cytoplasmic acidity ( Figure 6 ) . In the lungs , a passive hyperemia with thickening of the interalveolar septa , a weak focal parenchymal edema and a moderate stromal connective tissue hyperplasia were present . There were no significant pathological changes in the kidneys ( Figure 6 ) . In general , the SD animals subjected to catecholamine stress ( isoprenaline ) showed some characteristics of Takotsubo-like cardiomyopathy , including a mild degree of apical akinesis/dyskinesis , edema of cardiomyocytes , increased NT-proBNP level and mild lung edema ( Figures 2 and 8 , Table 3 ) . Numerous , scattered foci of banded mononuclear cell infiltration were present in the myocardium . Severe hyperemia of myocardial capillaries and arterioles and small organized foci of myocardial extravasation were present . Nevertheless , the majority of the myocardium remained normal in structure . The lungs presented transudate in the alveoli . In the kidneys , there was a weak congestion in the medulla and renal bodies . A small number of tubules filled with an acidic substance was present . Stimulation of stromal fibrocytes without production of connective tissue fibers was observed ( Figure 8 ) . We evaluated changes in renal excretion induced by TMAO , urea and saline intravenous administration in acute experiments . Results are summarized in Figure 9 . Only TMAO induced diuresis . The pattern of diuresis and total solutes excretion induced by TMAO were similar . Increases of V and UosmV induced by TMAO were associated with transient decrease of Uosm , whereas UNaV and UKV were not affected . This indicates that TMAO did not affected the tubular transport of sodium and potassium but induced osmotic diuresis . The bolus infusions of TMAO or saline produced a transient increase in ABP with no changes in HR , which was followed by a decrease in ABP below the baseline ( by 6 ± 3 mmHg ) . There was no significant correlation between changes in ABP and diuresis . Labeled LDH was stable in PBS solution , showing no tendency for spontaneous aggregation . The value of diffusion coefficients measured by FCS at 25°C was 49 . 2 ± 3 . 3 μm2/s . This corresponds to a hydrodynamic radius of around 5 . 0 nm , which is in line with previously reported values ( Zipper and Durchschlag , 1998 ) . The tertiary and quaternary structures of LDH , with and without TMAO , were not influenced by a 24 hr treatment with HPS ( pressure oscillations mimicking those of a rat heart ) , ( Figure 10A ) . Tests performed using a different pressure oscillation system , where pressures up to 1000 mmHg were applied , did not detect observable changes in the protein structure ( see Figure 10—figure supplement 1 ) . The incubation of LDH at atmospheric pressure and elevated temperatures ( 50–80°C ) changed the diffusion coefficient of LDH , indicating the dissociation of LDH tetramers , as well as protein denaturation and aggregation . The addition of 1M TMAO produced a moderate stabilizing effect on LDH , shifting the threshold of observed protein morphology change towards higher temperatures ( Figure 10B ) . Specifically , it seems that the gradual dissociation of tetramers to monomers occurred above 55°C , which was followed by the denaturation of the tertiary protein structure at higher temperatures . At a concentration of LDH >3 nM , the aggregation of monomers prevailed . At a lower concentration , the aggregation progressed more slowly , if at all , while any aggregates that did form , were too sparse to influence the measurement results . Further work , including probing a broader matrix of LDH concentrations and temperatures , is needed to confirm these initial findings . Nevertheless , in all the experiments , presence of TMAO shifted the threshold of change in LDH morphology towards higher temperatures and diminished the magnitude of the change . This suggests a stabilizing effect of TMAO on the native structure of the protein .
A limitation of our study is that biochemical and hemodynamic measurements were performed only at the end of the treatment . This is because we aimed to avoid stress-related circulatory complications in SHHF rats , which are very prone to lethal cardiovascular events . TMAO , a molecule present in seafood and a derivate of gut bacteria metabolism , exerts beneficial effects in HF rats . These benefits might be derived from the diuretic , natriuretic and hypotensive properties of TMAO . The hydrostatic pressure stress generated by the contracting heart does not affect LDH protein structure . Further studies designed to evaluate TMAO-dependent diuretic and natriuretic effects are needed , as TMAO may serve as a naturally occurring diuretic agent in diseases associated with fluid retention e . g . heart failure .
The study was performed according to Directive 2010/63 EU on the protection of animals used for scientific purposes and approved by the Local Bioethical Committee in Warsaw ( permission:100/2016 and 098/2019 ) . 4–5 week-old , male , lean Spontaneously Hypertensive Heart Failure ( SHHF/MccGmiCrl-Leprcp/Crl ) rats were purchased from Charles River Laboratories ( USA ) . Age-matched Sprague-Dawley ( SD ) rats were obtained from the Central Laboratory for Experimental Animals , Medical University of Warsaw , Poland . Six-week-old SHHF ( n = 18 ) and SD ( n = 40 ) were randomly assigned to either control groups ( rats drinking tap water ) or the TMAO groups ( rats drinking TMAO solution in tap water , TMAO - abcr GmbH - Karlsruhe , Germany , 333 mg/l ) . While no specific randomization method was used , rats from one cage were assigned to different groups . The dose of TMAO was selected in order to increase the plasma TMAO concentration by 3–5 times ( to mimic possible physiological concentrations ) and to avoid suprapharmacological effects of TMAO , based on our previous study ( Huc et al . , 2018 ) . Rats were housed in groups of 2–3 animals , in polypropylene cages with environmental enrichment , 12 hr light/12 hr dark cycle , temperature 22–23oC , humidity 45–55% , fed standard laboratory diet ( 0 . 19% Na , Labofeed B standard , Kcynia , Poland ) and water ad libitum . SHHF-TMAO ( n = 9 ) , SHFF-control ( n = 9 ) , SD-TMAO ( n = 10 ) , SD-control ( n = 10 ) were not subjected to any interventions except standard animal care until the age of 58 weeks . At the age of 56 weeks the ISO-control ( n = 10 ) and ISO-TMAO ( n = 10 ) series were given ( s . c . ) isoprenaline at a dose of 100 mg/kg b . w . ( isoprenaline hydrochloride , Sigma-Aldrich , Saint Louis , MO , USA ) to produce catecholamine stress as previously described ( Sachdeva et al . , 2014 ) . The experimental protocol is depicted in Figure 1 . 58-week-old rats were maintained in metabolic cages for 2 days to evaluate the 24 hr water and food balance and to collect urine for analysis . The next day , the rats underwent an echocardiogram using a Samsung HM70: an ultrasound system equipped with a linear probe 5–13 . MHz . After the echo examination the rats were anaesthetized with urethane ( 1 . 5 g/kg b . w . i . p . , Sigma-Aldrich ) and the left femoral artery was cannulated with a polyurethane catheter for arterial blood pressure ( ABP ) recordings . The recordings were started 40 min after the induction of anesthesia and 15 min after inserting the arterial catheter . After 10 min of ABP recordings , a Millar Mikro-Tip SPR-320 ( 2F ) pressure catheter was inserted via the right common carotid artery and simultaneous left ventricle pressure ( LVP ) and ABP recordings were performed . The catheter was connected to a Millar Transducer PCU-2000 Dual Channel Pressure Control Unit ( Millar , USA ) and Biopac MP 150 ( Biopac Systems , USA ) . After hemodynamic recordings , blood from the right ventricle of the heart was taken and rats were euthanized by decapitation . The heart , the lungs and the kidneys were harvested for histological and molecular analysis . 56-week-old rats were housed in metabolic cages for 2 days to evaluate the 24 hr water and food balance and to collect urine for analysis . Echocardiography was performed as described above . The next day , rats were given isoprenaline ( 100 mg/kg , s . c . ) . 24 hr after the administration of ISO , the echocardiogram was repeated . Eight days after the ISO-treatment , the 24 hr food and water intake was evaluated and an echocardiogram was performed . Afterwards , the rats were anaesthetized with urethane ( 1 . 5 g/kg b . w . i . p . , Sigma-Aldrich , Poland ) and the hemodynamic measurements were taken , including ABP and LVP recordings as described above for SHHF and SD rats . Plasma and urine concentrations of TMAO were measured using Waters Acquity Ultra Performance Liquid Chromatograph coupled with a Waters TQ-S triple-quadrupole mass spectrometer . The mass spectrometer was operated in the multiple-reaction monitoring ( MRM ) - positive electrospray ionization ( ESI ) mode , as previously described ( Jaworska et al . , 2017 ) . Serum and urine sodium , potassium , creatinine and urea were analyzed using a Cobas 6000 analyzer ( Roche Diagnostics , Indianapolis , USA ) . The following ELISA kits were used for the evaluation: NT-proBNP ( FineTest , cat . no . ER0309 ) , aldosterone ( Cayman Chemicals , cat . no . 501090 ) , vasopressin ( Biorbyt , cat . no . orb410987 ) , angiotensin II ( FineTest , cat . no . ER1637 ) , TNFα and IL-10 ( cat . no . RTA00 and R1000 , respectively , R and D System ) . All procedures were carried out according to the standard protocol supplied with the ELISA Kit . The absorbance intensity was measured at 450 nm with a Multiskan Microplate Reader ( Thermo Fisher Scientific ) . All experiments were performed in duplicate ( technical replicates ) . Tissues sections were fixed in 10% buffered formalin , dehydrated using graded ethanol and xylene baths and embedded in paraffin wax . Sections of 3–4 μm were stained with hematoxylin and eosin ( HE ) and van Gieson stain ( for connective tissue fibers ) . General histopathological examination was evaluated at a magnification of 10x , 40x and 100x ( objective lens ) and 10x ( eyepiece ) and photographic documentation was taken . Morphometric measurements were performed at magnification of 40x ( objective lens ) . Heart and kidney samples were collected from rats under urethane anesthesia and frozen at −80°C . Next , the samples were homogenized on BeadBug microtube homogenizer ( Benchmark Scientific , Inc ) . Total RNA was isolated from the samples according to the TRI Reagent protocol . cDNA was transcribed from RNA samples according to the iScript Reverse Transcription Supermix #1708841 protocol ( Bio-Rad ) . The qPCR mixes were prepared according to the Bio-Rad SsoAdvanced Universal SYBR Green Supermix protocol #1725271 . Amplifications were performed on a Bio-Rad CFX Connect Real-Time System under standardized conditions using commercial assays . Data were analyzed using CFX Manager 3 . 0 software . The genes investigated in this study were: angiotensinogen ( Atg , qRnoCED0051666 ) , angiotensin II receptor type 1a ( At1a , qRnoCID0052626 ) , angiotensin II receptor type 1b ( At1b , qRnoCED0005729 ) , angiotensin II receptor type 2 ( At2 , qRnoCED0007551 ) , transforming growth factor-beta ( Tgf-b , qRnoCED0007638 ) , renin ( Rn , qRnoCID0008721 ) , metalloproteinase inhibitor 2 ( Timp2 , qRnoCID0001559 ) . Beta-actin was used as housekeeping gene ( Actbl2 , qRnoCED0018219 ) . We evaluated the effect of TMAO on bovine , cardiac LDH ( Merck , Poland ) exposed to pressure oscillations and increased temperature . The pressure oscillations were generated in a custom-built system . In order to mimic the conditions in the heart the pressure changed from 0 to 180–250 mmHg ( or to higher values ) at oscillation rate of 280 min-1 . In general , the setup consisted of two main parts: i ) a custom-built oscillatory pressure controller with solenoid micro valves to control the inner pressure and ‘pulse’ frequency and ii ) a sample chamber ( Figure 11 ) . We designed and constructed three different samples chambers ( Figures 12 , 13 and 14 ) , which permitted the samples to be exposed to pressure oscillations in different ways . Blinding was provided for each treatment ( TMAO vs control ) . Unblinding was performed after statistical analysis . Any encountered outliers were included in the analysis . Diastolic arterial blood pressure ( DBP ) , systolic arterial blood pressure ( SBP ) and heart rate ( HR ) were calculated from the arterial blood pressure tracing . Left ventricular end-diastolic pressure ( LVEDP ) , maximal slope of systolic pressure increment ( +dP/dt ) and diastolic pressure decrement ( -dP/dt ) were calculated from the left ventricle blood pressure tracing using AcqKnowledge Biopac software ( Biopac Systems , Goleta , USA ) . The Shapiro-Wilk test was used to test normality of the distribution . Differences between the TMAO and control groups were evaluated by an Independent-Samples t-test or by Mann-Whitney U test for data that were not normally distributed . In the acute experiments , the differences in the mean values between groups were first analyzed by the classic one-way ANOVA followed by a modified Student’s t-test for independent variables , using Bonferroni’s correction for multiple comparisons . The log-rank test was used to test the survival differences between TMAO and control animals . A value of two-sided p<0 . 05 was considered significant . Analyses were conducted using Statistica , version 13 . 3 ( Tibco , Palo Alto , CA , USA ) . Replicates were not used unless otherwise stated . The basic definitions of technical and biological replicates are as follows . Technical replicates: a test performed on the same sample multiple times . Biological replicates: a test performed on biologically distinct samples representing an identical time point or treatment dose . Sample size was calculated at the start of the study , based on plasma levels of the investigated markers and hemodynamic parameters in rats , which were reported in our previous studies ( Savi et al . , 2018 ) . We have chosen between-group difference in plasma NT-proBNP , ejection fraction , stroke volume and ABP as primary end-points with the following parameters , respectively: the difference between the tested ( groups ) 30% , 15% , 30% and 13%; the average for the entire population of 30 pg/mL , 80% , 0 . 35 mL , 100 mmHg; a common standard deviation of 7 pg/mL , 9% , 0 . 08 mL , 10 mmHg; for an alpha error of 0 . 05 , test power 0 . 8 . Other biochemical parameters were used as secondary end points . | Heart failure is a common cause of death in industrialized countries with aging populations . Japan , however , has lower rates of heart failure and fewer deaths linked to this disease than the United States or Europe , despite having the highest proportion of elderly people in the world . Dietary differences between these regions may explain the lower rate of heart failure in Japan . The Japanese diet is rich in seafood , which contains nutrients that promote heart health , such as omega-3 fatty acids . Seafood also contains other compounds , including trimethylamine oxide ( TMAO ) . Fish that live in deep waters undergo high pressures , which can damage their proteins , but TMAO seems to protect the proteins from harm . In humans , eating seafood increases TMAO levels in the blood and urine , but it is unclear what effects this has on heart health . Increased levels of TMAO in the blood are associated with cardiovascular diseases , but scientists are not sure whether TMAO itself harms the heart . A toxic byproduct of gut bacteria called TMA is converted in TMAO in the body , so it is possible that TMA rather than TMAO is to blame . To assess the effects of dietary TMAO on heart failure , Gawrys-Kopczynska et al . fed the compound to healthy rats and rats with heart failure for one year . TMAO had no effects on the healthy rats . Of the rats with heart failure that were fed TMAO , all of them survived the year , while one third of rats with heart failure that were not fed TMAO died . TMAO-treated rats with heart failure had lower blood pressure and urinated more than untreated rats with the condition . The experiments suggest that dietary TMAO may mimic the effects of heart failure treatments , which remove excess water and salt and lower pressure on the heart . More studies are needed to confirm whether TMAO has this same effect on humans . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"medicine"
] | 2020 | TMAO, a seafood-derived molecule, produces diuresis and reduces mortality in heart failure rats |
The emergence of multicellularity in Animalia is associated with increase in ROS and expansion of tRNA-isodecoders . tRNA expansion leads to misselection resulting in a critical error of L-Ala mischarged onto tRNAThr , which is proofread by Animalia-specific-tRNA Deacylase ( ATD ) in vitro . Here we show that in addition to ATD , threonyl-tRNA synthetase ( ThrRS ) can clear the error in cellular scenario . This two-tier functional redundancy for translation quality control breaks down during oxidative stress , wherein ThrRS is rendered inactive . Therefore , ATD knockout cells display pronounced sensitivity through increased mistranslation of threonine codons leading to cell death . Strikingly , we identify the emergence of ATD along with the error inducing tRNA species starting from Choanoflagellates thus uncovering an important genomic innovation required for multicellularity that occurred in unicellular ancestors of animals . The study further provides a plausible regulatory mechanism wherein the cellular fate of tRNAs can be switched from protein biosynthesis to non-canonical functions .
The ever-growing genome information has shed light on the expansion of the number of tRNA genes , which increases with the complexity of multicellular organisms ( Goodenbour and Pan , 2006 ) . Such an expansion of tRNA genes led to the appearance of tRNA isodecoders that is tRNAs with identical anticodon but with sequence variation ( s ) elsewhere . This sequence diversification has aided the functional expansion of the tRNA molecules from mere adapters in protein translation to versatile molecules involved in many non-canonical functions . Recent studies have shown the indispensable physiological roles of tRNA and tRNA-derived fragments in diverse functions , such as intergenerational inheritance , RNAi , gene regulation , apoptosis inhibition , ribosome biogenesis , stress granules and also in pathologies such as carcinoma ( Chen et al . , 2016; Gebetsberger et al . , 2017; Goodarzi et al . , 2015; Ivanov et al . , 2011; Saikia et al . , 2014; Schorn et al . , 2017; Sharma et al . , 2016 ) . The tRNA is abundantly saturated with identity elements which are essential for processing , structure , and also performing the canonical functions in translation . However , the emergence of tRNA isodecoders has led to the degeneracy of idiosyncratic features of a few tRNAs , which are essential for the canonical translation function ( Saint-Léger et al . , 2016 ) . Moreover , the advent of multicellularity has led to the production and use of reactive oxygen species ( ROS ) as signaling molecules , unlike bacteria and yeast that only respond to oxidative stress . It has been suggested that ROS adds the required diversity to the gamut of signaling events involved in cell differentiation and proliferation essential for multicellular systems such HIF pathway and signaling cascade by NRF-2 , NFКB , Oct4 , p53 , etc ( Bigarella et al . , 2014; Bloomfield and Pears , 2003; Covarrubias et al . , 2008; Lalucque and Silar , 2003; Peuget et al . , 2014; Shadel and Horvath , 2015; Sies , 2017 ) . tRNA synthetases are essential enzymes responsible for maintaining fidelity during protein biosynthesis by selecting both the correct amino acid and tRNA ( Ibba and Soll , 2000; Ramakrishnan , 2002 ) . Alanyl-tRNA synthetase ( AlaRS ) is unique among this class of enzymes in using the universally conserved G3•U70 wobble base pair in the acceptor stem as an identity element for recognition and charging of tRNAAla ( Hou and Schimmel , 1988 ) . Due to the phenomenon of a subtle recognition slippage , the archaeal and eukaryotic AlaRSs have a relaxed specificity for tRNA and can charge alanine on G4•U69 containing ( non-cognate ) tRNAs also ( Sun et al . , 2016 ) . Recently , we have shown that G4•U69 is predominant in kingdom Animalia , and specifically in tRNAs coding for Thr ( 18% in humans and sometimes as high as 43% , in Takifugu rubripes ) , thereby generating L-Ala-tRNAThr ( G4•U69 ) ( Kuncha et al . , 2018a ) . Such a tRNA misselection error can result in Thr-to-Ala mistranslation . However , the absence of Thr-to-Ala mutations at the proteome level indicated the presence of a dedicated proofreading factor ( Sun et al . , 2016 ) . We further showed that L-Ala-tRNAThr ( G4•U69 ) is robustly edited by Animalia-specific tRNA deacylase ( ATD ) in vitro ( Kuncha et al . , 2018a ) . ATD is a paralog of a ‘Chiral Proofreading’ enzyme D-aminoacyl-tRNA deacylase ( DTD ) , which is one of the key chiral checkpoints during protein synthesis and present across Bacteria and Eukarya ( Kuncha et al . , 2019 ) . Earlier studies using ligand-bound crystal structures , in combination with biochemical assays , have shown that DTD operates via L-chiral rejection mechanism and as a result also avoids mistranslation of L-Ala to achiral glycine ( Ahmad et al . , 2013; Kuncha et al . , 2018b; Pawar et al . , 2017 ) . The L-chiral rejection by DTD is attributed to an active site invariant cross-subunit Gly-cisPro motif that acts as a chiral-selectivity filter ( Routh et al . , 2016 ) . The Gly-Pro ( GP ) motif conformation has switched from Gly-cisPro in DTD to Gly-transPro in ATD . In Animalia , the appearance of tRNAThr ( G4•U69 ) is invariably correlated with the presence of ATD ( Kuncha et al . , 2018a ) . Thus , ATD is the first and the only known proofreader of tRNA misselection errors . However , the physiological significance and its role in translation quality control are not known . Given the emerging evidence of stress-induced regulation of mistranslation ( Netzer et al . , 2009; Steiner et al . , 2019 ) , the role of ATD in higher organisms is of immense interest . Here , we show that not only ATD but threonyl-tRNA synthetase ( ThrRS ) is also involved in correcting the tRNA-induced misselection error by AlaRS . As oxidative stress inhibits the proofreading activity of ThrRS , the protective activity of ATD is indispensable for avoiding mistranslation of Thr codons that is deleterious . Strikingly , the origin of ATD and the G4•U69 containing tRNAs are rooted in Choanoflagellates suggesting its importance for the origin of metazoans at the convergence of multicellularity-induced oxidative stress and tRNA isodecoder expansion .
Based on the earlier biochemical observations , ATD is expected to avoid Thr-to-Ala mistranslation , which would affect protein homeostasis and also cell survival . Hence to understand the physiological role and relevance of ATD , we generated a CRISPR-Cas9-based ATD knockout of HEK293T cells ( Figure 1A , Figure 1—figure supplement 1 ) . Surprisingly , the ATD null cells did not show any noticeable phenotypic effect which was puzzling , given that ATD is the only known editor of L-Ala-tRNAThr ( G4•U69 ) . To further analyze the cells at the molecular levels , Hsp70 was chosen as the marker for proteome stress . Hsp70 is a chaperone which is upregulated in response to protein misfolding and hence used as a marker to study proteostasis stress , and is also implicated in aging and neurodegenerative diseases ( Gupta et al . , 2011; Hartl , 1996; Mosser et al . , 2000; Rampelt et al . , 2012; Sala et al . , 2017 ) . The ATD null cells also showed no significant difference in molecular proteome stress marker Hsp70 ( Figure 1B ) . Lack of proteome errors is further confirmed by the identification of reporter protein peptides ( explained later ) using mass spectrometry experiments wherein no missubstitutions of alanine for threonine were observed . The above unexplainable results prompted us to check whether ATD is indeed expressed in vivo . The expression of ATD was initially checked in different cell lines ( CHO , Neuro2A , SKOV3 , HEK293T , HeLa , NIH/3T3 , and mouse embryonic stem cells E14Tg2a ) , followed by different tissue samples of the mouse . We could see that ATD is ubiquitously expressed in all the cell lines and tissues of mouse tested thus underscoring its physiological importance ( Figure 1C , D; Figure 1—figure supplement 2 ) . The expression data obtained is in line with the available databases such as Human protein Atlas ( https://www . proteinatlas . org/ ) , Zfin database ( https://zfin . org/ ) , Genecards ( https://www . genecards . org/ ) and Bgee ( https://bgee . org/ ) . Despite the ubiquitous presence of ATD , the absence of any noticeable phenotypic and molecular changes in the ATD knockout cell lines prompted us to search for the presence of any other proofreading factor responsible for clearing L-Ala-tRNAThr in addition to ATD ( Figure 1E ) . The probable proofreading players in the cell that can deacylate L-Ala-tRNAThr ( G4•U69 ) and maintain the fidelity of translation in the absence of ATD include both cis ( AlaRS , threonyl-tRNA synthetase ( ThrRS ) ) and trans editors ( AlaXp’s and ProX ) present in Animalia . Since L-Ala is a cognate amino acid for AlaRS and AlaXp’s , the deacylation of L-Ala-tRNAThr by these editing domains is very unlikely Beebe et al . , 2008; Sokabe et al . , 2005 ) . ProX is specific to tRNAPro and its homolog ( Ybak ) is known to achieve substrate specificity by forming a binary complex with ProRS , which imparts tRNA specificity for editing ( Ahel et al . , 2003; An and Musier-Forsyth , 2004; An and Musier-Forsyth , 2005; Chen et al . , 2019 ) . Thus , the only known and possible player with a high affinity for tRNAThr is ThrRS . ThrRS has an N-terminal editing domain , which is known to deacylate L-Ser erroneously charged on tRNAThr ( Dock-Bregeon et al . , 2004 ) . To test whether ThrRS indeed possess deacylation activity towards L-Ala-tRNAThr , we incubated M . musculus ThrRS ( MmThrRS ) with the substrate . Intriguingly , MmThrRS was able to deacylate L-Ala-tRNAThr at 10 nM concentration but not L-Thr-tRNAThr ( Figure 2A , B , Table 1 ) . This is the first identified instance where one aminoacyl-tRNA synthetase , ThrRS in this case , corrects the error caused by another aminoacyl-tRNA synthetase ( AlaRS ) , which misselects the tRNA ( Figure 2C ) . These results identified a two-tier functional redundancy in the cell for clearing L-Ala mistakenly attached to tRNAThr suggesting that ThrRS can correct the tRNA misselection error and hence compensate for the absence of ATD in the knockout cells that do not show any mistranslation of Thr-codons . To validate the universality of the cross-synthetase error correction mechanism , we checked the activity of D . rerio ( Dr ) and H . sapiens ( Hs ) ThrRS on L-Ala-tRNAThr ( Figure 3A ) . In line with the MmThrRS activity , DrThrRS and HsThrRS also deacylated L-Ala-tRNAThr robustly , indicating that the activity is conserved across higher eukaryotes . Therefore , in Animalia there exists a functional redundancy in clearing L-Ala-tRNAThr in the form of ATD and ThrRS ( Figure 2C ) . It was tempting to check whether the L-Ala activity of ThrRS is conserved across different domains of life or has it been acquired over the course of evolution only in Animalia ? It is worth noting here that it is only in Animalia that the error inducing tRNA species is present and hence the problem of L-Ala mischarging on tRNAThr . To test this , we checked the deacylation activity of ThrRS from Bacteria . We were surprised to see that bacterial ThrRS from E . coli ( EcThrRS ) was active on L-Ala-tRNAThr , even though neither Bacteria ( bacterial AlaRS is discriminatory and charges only tRNAs containing G3•U70 [Sun et al . , 2016] ) nor lower eukaryotes ( which lack tRNAs containing G4•U69 ) possess the problem of tRNA misselection ( Figure 3—figure supplement 1A–C ) . Therefore , it is perplexing as to why the L-Ala editing activity of ThrRS is universally conserved . In any case , these biochemical results clearly suggest that the editing site of ThrRS can accommodate and deacylate L-Ala mischarged on tRNAThr , despite lacking the critical interactions emanating from the side chain hydroxyl group of serine ( Figure 3D , E ) . Thus , in Animalia , ThrRS maintains the fidelity of decoding Thr codons by performing the dual function of clearing both amino acid ( L-Ser-tRNAThr ) and tRNA misselection ( L-Ala-tRNAThr ) errors ( Figure 3A–C ) . Recruitment of a new factor , ATD , even in the presence of ThrRS , a housekeeping gene , suggests the importance of avoiding Thr-to-Ala mistranslation . It raises an important question as to whether this functional redundancy is important for maintaining cellular protein homeostasis during any kind of physiological or stress conditions . Recent thiome studies using HeLa cell lysates show that ThrRS is oxidized at physiological conditions ( Leonard et al . , 2009 ) and the site of oxidation was found to be an active site invariant cysteine ( C182 ) of ThrRS editing domain , which plays a crucial role in catalysis ( Figure 3D , E ) . Oxidation of C182 abolishes the L-Ser-tRNAThr editing activity of E . coli ThrRS has been noted earlier ( Ling and Söll , 2010 ) . We could establish that oxidizing conditions ( using H2O2 ) do not affect the aminoacylation activity but results in a complete loss of L-Ala-tRNAThr and L-Ser-tRNAThr editing activity of MmThrRS ( Figure 3F; Figure 3—figure supplement 2 ) . Interestingly , the synthetase activity of MmThrRS is not affected even in the presence of 10-fold excess ( 50 mM ) of H2O2 . To establish the universality of oxidation-induced inactivation of ThrRS editing activity , we tested EcThrRS , DrThrRS , MmThrRS , and HsThrRS ( Figure 3F–I; Figure 3—figure supplement 1D–F ) . Irrespective of the organism ( from bacteria to mammals ) , in the presence of oxidizing conditions the editing domain of ThrRS is inactive on L-Ala-tRNAThr , however , this did not affect the aminoacylation activity of the enzyme . These deacylation experiments clearly show that the editing site cysteine is prone to oxidation and this feature is conserved across Bacteria and Eukarya . Since ThrRS is inactive in the presence of ROS , the obvious question is whether ATD is active in the presence of oxidizing conditions ? To this end , we checked for MmATD’s activity on L-Ala-tRNAThr in the presence and absence of H2O2 and found that it was totally unaffected ( Figure 3F ) . The inertness of ATD towards oxidative stress was further validated by performing L-Ala-tRNAThr deacylation assays using ATDs from different organisms such as Hydra vulgaris ( HvATD ) from the invertebrate Animalia , Danio rerio ( DrATD ) from class Pisces , Gallus gallus ( GgATD ) which belongs to Aves and Homo sapiens ( HsATD ) which is a mammal . Irrespective of the origin/class of organisms , all the ATDs were robust in deacylating L-Ala-tRNAThr even in the presence of oxidizing conditions ( H2O2 ) ( Figure 3J; Figure 3—figure supplement 3A , B ) . Hence , unlike ThrRS , ATD is insensitive to a significant amount of H2O2 . This is in accordance with the absence of any cysteine residue in the active site of ATD , unlike that of ThrRS editing site ( Figure 3—figure supplement 4A , B ) . We then asked if ATDs activity is not affected by H2O2 , would this factor avoid Thr-to-Ala mistranslation during oxidative stress in vivo ? To see the cellular effect and the essentiality of ATD , we initially treated the HEK293T wild type and ATD knockout cells with different concentrations of H2O2 ( 50 , 75 and 100 µM ) for 24 hr , followed by cell viability assay ( MTT ) , which revealed that ATD knockout cells show pronounced cell death ( Figure 4A; Figure 4—figure supplement 1 ) . We could further show that this is a direct effect of ATD by rescuing the cells from H2O2 induced toxicity by expressing FLAG-tagged ATD from a plasmid copy ( Figure 4B , C ) . To further confirm that the rescue is due to the enzymatic activity of ATD , we used an enzymatically inactive G115F mutant of ATD . This mutant was generated based on our earlier studies on DTD’s ligand-bound structures wherein an analogous mutation A112F ( PfDTD , PDB id: 4NBI ) sterically excludes the binding of adenine ( A76 ) of the incoming substrate ( aa-tRNA ) and hence renders it inactive ( Ahmad et al . , 2013 ) . As expected , we could show that the G115F mutant of ATD also is similarly inactive for deacylation activity even with 1000-fold excess concentrations ( Figure 4D; Figure 4—figure supplement 2 ) . The G115F mutant of ATD does not rescue ATD knockout cells from the oxidative stress-induced toxicity and therefore unequivocally establishes that it is ATDs enzymatic activity which is essential for the rescue during oxidative stress ( Figure 4B , C ) . To further validate the in vivo results of HEK293T cells , we generated ATD knockout in mouse embryonic stem ( mES ) cells using the CRISPR-Cas9 paired guide strategy ( Figure 4—figure supplement 3A , B ) . mES cells are known to have higher internal ROS and can be cultured only in the presence of antioxidants such as β-mercaptoethanol , which is usually added to the culture media ( Czechanski et al . , 2014; Han et al . , 2008 ) . We altered the culture conditions by simply decreasing the concentration of the reducing agent , β-mercaptoethanol ( βME ) . mES cells colonies were initially stained using the Leishman stain method and counted under the microscope . Interestingly , in the usual concentration of β-mercaptoethanol ( 100 µM ) the wild type and ATD knockout mES cells were growing normally without any significant change in proliferation and survival . However , at 0 . 1X ( 10 µM ) concentration of βME , the number of colonies in the knockout cells ( from two independently derived clones ) was significantly lower compared to that of wild type ( ( Figure 4E–H ) . While the wild type and knockout have suffered at 0 . 05X concentration βME and no colonies were observed in the plate ( Figure 4—figure supplement 3C ) . These results in combination with HEK293T data unequivocally establish that ATD is essential during physiological conditions of high oxidative stress . It is worth noting here that oxidative stress in ATD-/- cells mimic the scenario of a double knockout condition in which ATD is absent and the editing domain of ThrRS is inactivated . Hence , the hypersensitivity of ATD knockout towards oxidative stress is due to the absence of both the deacylators of L-Ala-tRNAThr and therefore is expected to result in the mistranslation of Thr codons . To check for mistranslation , initially , we set out to look for markers of protein misfolding such as Hsp70 . The level of Hsp70 in ATD knockout cells was significantly upregulated ( >2 fold ) in response to oxidative stress , while the wild type cells showed a marginal upregulation ( Figure 4I ) . The mistranslation of Thr codons would also affect the overall cellular proteome homeostasis . To visualize the disturbance in the cellular proteostasis , we used EGFP tagged FlucDM as the sensor of proteome stress ( Gupta et al . , 2011 ) . FlucDM-EGFP aggregates in response to proteome stress and can be visualized either by microscopy or by quantifying its ratio of insoluble to the soluble fraction in the cell . H2O2-treated ATD knockout cells have more speckles of GFP ( due to aggregation of FlucDM ) and show a dose-dependent increase in oxidative stress , while the number of aggregates in wild type increases only marginally ( Figure 5A; Figure 5—figure supplement 1A ) . These observations are further validated by quantifying the ratio of insoluble-to-soluble FlucDM-EGFP using western blots , which showed a marked two-fold increase compared to wild type ( Figure 5B; Figure 5—figure supplement 1B ) , which is in line with the already observed 1 . 5 to 2-fold increase in FLucDM insoluble-to-soluble fraction ratio during different stress ( heat and proteasome inhibitor ) conditions ( Gupta et al . , 2011; Rawat et al . , 2019 ) . As further evidence to signify the role of ATD in oxidative stress , we observe the upregulation of ATD expression in H2O2-treated wild-type cells ( Figure 5C ) . Therefore , ATD is an important factor needed to maintain cellular proteostasis during oxidative stress . The earlier observed proteome stress is very likely due to Thr codons' mistranslation . To pinpoint the exact cause of proteome stress , we overexpressed a reporter protein ( GFP using pEGFP vector ) with and without oxidative stress , followed by GFP pull-down and subjected to peptide identification using mass spectrometry ( MS/MS ) . The MS/MS data was examined for amino acid substitutions using a GFP-mutant database . We could read increased mistranslation of Thr-to-Ser in a few peptides of samples purified from H2O2-treated wild type cells ( Figure 5—figure supplement 2 ) . This quickly rules out the possibility of ATD proofreading L-Ser-tRNAThr , which is in accordance with the low activity of ATD and its likely inability to resample from elongation factor thermo unstable ( EF-Tu ) ( Figure 5G , H ) . However , no toxicity is observed in these cells since the mutation of Thr-to-Ser is a milder substitution and therefore tolerable . In the case of H2O2-treated ATD knockout cells , the level of Thr-to-Ser mistranslation is similar to that of wild type and further validates our biochemical data that ATD cannot proofread L-Ser-tRNAThr in vivo . Unlike Thr-to-Ser substitutions , the levels of Thr-to-Ala mistranslation was significantly higher in cells devoid of ATD and treated with H2O2 ( Figure 5D–F ) . As expected , except for Ser and Ala , Thr codons were not mistranslated to other amino acids such as smaller Gly , and bigger Leu , Tyr , and Phe . Since Thr-to-Ala is a drastic change of size , shape , and polarity , it destabilizes the structural integrity of the proteome producing the observed toxicity to the cell . These findings unambiguously provide direct evidence that ATD and ThrRS can proofread L-Ala-tRNAThr in vivo and the former plays a critical role during oxidative stress . ATD is present in kingdom Animalia and so is tRNAThr possessing G4•U69 . Since ATD is a paralog and evolved from a pre-existing DTD , we were curious to find out the first event of emergence or an intermediate of this transition . By extensive bioinformatics search , we have identified that ATD is present in Salpingoeca rosetta -a Choanoflagellate ( Figure 6A , B ) , but not in other unicellular eukaryotes ( like yeast , Plasmodium , Ichthyosporeans , and Filastereans ) . Choanoflagellates are the colonial unicellular cousins of multicellular animals , and recent studies have shown that these organisms possess genes that are unique to kingdom Animalia and essential for multicellularity i . e . , cell cycle regulation ( p53 ) , adhesion molecules ( cadherin’s ) , hypoxia signaling pathway ( prolyl hydroxylase ) , immune system , sexual reproduction , meiotic division , etc . ( Table 2; Brunet et al . , 2019; Fairclough et al . , 2010; King et al . , 2003; Sogabe et al . , 2019; Woznica et al . , 2017 ) . A thorough phylogenetic analysis of all the available ATD sequences has shown that the emergence of ATD is rooted in Choanoflagellates , which possesses ATD in addition to having a canonical DTD ( Figure 6A; Figure 6—figure supplement 1 ) . Intriguingly , Choanoflagellate ATD has mixed characteristics of both ATD and DTD . The two signature sequence motifs characteristic of all bacterial and eukaryotic DTD are SQFTL and NXGPXT , while PQATL and TNGPY/FTH typify ATD . Choanoflagellate ATD is unusual in having one motif each from ATD and DTD i . e . , PQATL and NDGPFT respectively , thus capturing an intermediate in the transition and emergence of a paralog of DTD ( Figure 6A ) . Using the available transcriptome and genomics data of different Choanoflagellates , we could identify ATD in a few more Choanoflagellates ( S . macrocollata , S . dolichothecata , S . helianthica , Codosiga hollandica , Mylnosiga fluctuans , and Choanoeca flexa ) ( Brunet et al . , 2019; Richter et al . , 2018 ) and all represent the transition state ( Figure 6—figure supplement 2 ) . Based on the aforementioned mixed features , Choanoflagellate represents a unique case of an intermediate in the metamorphosis of DTD to ATD from the amino acid sequence perspective . We further wanted to check whether this ancestor of Animalia ATD would perform a similar proofreading function . Indeed , biochemical assays showed that S . rosetta ATD ( SrATD ) was active on L-Ala-tRNAThr thereby proving it to be a functional ATD and not DTD , which is a ‘Chiral Proofreader’ that does not act on L-aminoacyl-tRNA substrates ( Ahmad et al . , 2013; Figure 6C ) . Interestingly , SrATD was not only active on L-Ala mischarged on non-cognate tRNA but also could discriminate the correctly charged substrates ( L-Ala-tRNAAla ) ( Figure 6D ) by 10-fold . The discrimination potential between cognate and non-cognate will be further enhanced in the presence of Elongation factor which binds L-Ala-tRNAAla strongly compared to that of L-Ala-tRNAThr ( Kuncha et al . , 2018a; LaRiviere et al . , 2001 ) . These bioinformatic analysis in combination with biochemical assays provide the basis for the emergence of ATD in the ancestors/cousins of Animalia ( Figure 6B ) . tRNA isodecoder containing G4•U69 also emerged in the unicellular ancestors of animals tRNA misselection by AlaRS is due to the appearance of tRNA isodecoders containing G4•U69 which is in turn linked to the expansion of tRNA genes . The number of unique tRNA genes is highly correlated to the complexity of the organisms ( number of cell types ) ( Figure 6E ) . The presence of G4•U69 in tRNAThr can be traced from Cnidaria ( Hydra vulgaris ) to the recently evolved mammals ( Homo sapiens ) as is the case for ATD ( Figure 6—figure supplement 3 ) . Interestingly , Choanoflagellates ( S . rosetta ) represent a stage of transition with 89 unique tRNAs , of which G4•U69 is seen in tRNAGln and tRNAHis , but not in tRNAThr ( Figure 6—figure supplement 4 ) . Since in eukaryotes , tRNAHis undergoes a post-transcriptional modification of adding guanine at the −1 position , which acts as a major determinant for histidyl-tRNA synthetase and a negative determinant for rest of the synthetases , and therefore would make it inert towards AlaRS ( Giegé et al . , 1998; Jackman and Phizicky , 2006; Tian et al . , 2015 ) . However , the presence of G4•U69 in tRNAGln possibly necessitates that Choanoflagellates have a proofreading factor that can edit L-Ala erroneously charged on tRNAGln ( G4•U69 ) . We could show that tRNAGln ( G4•U69 ) could be charged by AlaRS albeit at much lower levels of 4% to 7% than observed for tRNAThr ( G4•U69 ) at 30% to 40% ( Figure 6—figure supplement 5 ) . Due to the low levels of aminoacylation of Choanoflagellate tRNAGln ( G4•U69 ) , we could not perform the deacylation assays using L-Ala-tRNAGln . In any case , even such low levels of tRNA misselection would be precarious and Choanoflagellate ATD , which is in transition , is expected to act on these substrates and this aspect requires further probing . Therefore , the presence of ATD strictly alongside tRNA isodecoders containing G4•U69 even in Choanoflagellates , but not in fungi or other protists , underscores that ATD’s emergence is strongly linked with the appearance of tRNA isodecoders containing G4•U69 ( Figure 6F ) .
ROS is an inevitable cellular metabolite of increased metabolism in animals , produced by mitohormesis , NOX enzymes and also β-oxidation of lipids in peroxisomes ( Balaban et al . , 2005; Lambeth , 2004; Mohanty and McBride , 2013 ) . Unlike bacteria and lower eukaryotes , in Animalia ROS is an important signaling molecule and implicated in many cellular pathways ( activating NFκB , NRF2 , p53 , Oct4 , etc . ) and also for the origin of multicellularity ( Bigarella et al . , 2014; Bloomfield and Pears , 2003; Covarrubias et al . , 2008; Lalucque and Silar , 2003; Peuget et al . , 2014; Shadel and Horvath , 2015; Sies , 2017 ) . Cells avoid or nullify the toxic effects of ROS on protein quality control either by improving the fidelity of aminoacyl-tRNA synthetase or by increasing the overall Met content in the cellular proteome ( Goodenbour and Pan , 2006; Ling and Söll , 2010 ) . In the case of proofreading tRNA misselection , the functional redundancy of editing L-Ala-tRNAThr by ThrRS and ATD is broken in the presence of oxidative stress . As mentioned earlier , thiome studies have shown that Cys182 ( residue number corresponds to E . coli ThrRS ) , which is essential for editing activity , gets oxidized at physiological conditions ( cellular ROS ) . Since oxidative stress is an important component of multicellular systems , the innovation of ATD in Choanoflagellates seems to be essential to avoid the deleterious effects of tRNA misselection on the cellular proteome . Choanoflagellates mark the origin of many Animalia-specific genes which are essential for multicellularity ( Brunet and King , 2017; Brunet et al . , 2019; Fairclough et al . , 2010; Richter et al . , 2018; Sogabe et al . , 2019; Young et al . , 2011 ) . Therefore , the emergence of this metazoan specific proofreader , ATD , in Choanoflagellates is strictly linked to the appearance of tRNA isodecoders containing G4•U69 and underscoring a strong coevolution of these two , thus implicating their role in the emergence of multicellularity ( Figure 6E , F and Figure 7 ) . Recent studies have demonstrated the versatile roles of tRNA and tRNA-derived fragments in multiple cellular functions such as regulation , sperm fertility , intergeneration inheritance , and integrated stress response ( Doowon et al . , 2018; Fricker et al . , 2019; Gebetsberger et al . , 2017; Nätt et al . , 2019; Schwenzer et al . , 2019 ) . The appearance of tRNAThr isodecoders with G4•U69 in Animalia , and its conservation at the cost of tRNA misselection suggests a strong physiological role . The presence of 3’ tRNA-derived fragments ( tRF id: 3020a and 3020b ) of the tRNAThr ( G4•U69 ) in the tRF database ( http://genome . bioch . virginia . edu/trfdb/ ) ( Kumar et al . , 2015 ) indicates the possible role of these isodecoders in non-canonical functions . One of the unknown puzzles in RNA biology is how these tRNAs are effectively routed for non-canonical functions . Post-transcriptional modifications of tRNA are implicated in regulating the tRNA folding and also fragments generation ( Durdevic and Schaefer , 2013; Lyons et al . , 2018; Pan , 2018 ) . However , tRNAs once aminoacylated are channeled to the ribosome by elongation factor . Therefore , the presence of deacylators like ATD and ThrRS can strip the non-cognate amino acid from these tRNAs and thus provides one of the plausible ways of diverting a part of the tRNA pool away from translation ( Figure 7 ) . It appears that the emergence of ATD as a translational quality control factor in Choanoflagellates has been utilized to selectively regulate the free tRNA pool for other non-canonical functions in Animalia , which remains to be explored . Similar to genetic variations allowing to adapt/provide an advantage in stress conditions , the generation of variations at the proteome levels are known to be advantageous ( Kelly et al . , 2019; Mohler and Ibba , 2017 ) . In lower systems such as Mycoplasma , mistranslation has been shown to be advantageous , wherein the presence of editing defective aaRSs ( LeuRS , ThrRS , PheRS ) allows the organism to gain phenotypic plasticity by generating proteome diversity ( Li et al . , 2011; Mohler and Ibba , 2017; Pezo et al . , 2004 ) . The current work involves the use of acute oxidative stress ( H2O2 ) and further prompts to look at the behavior of ATD knockout cells in the presence of chronic oxidative stress such as activators of NADPH oxidase . ROS-induced inactivation of the editing activity of ThrRS can be used as a conserved mechanism for generating subtle variations in the cellular proteome . At the same time , the spatiotemporal regulation of ATD in combination with oxidative stress can generate a huge repertoire of proteome diversity that can be beneficial under certain cellular scenarios . ATD's expression levels are very likely modulated by the level of oxidative stress as seen in the case of reproductive tissues ( Figure 1C , D ) , which also suggests a mode of regulation via a feedback mechanism . It is worth noting here that the appearance of multicellular animals ~ 750 million years ago is also marked by a significant increase in the global atmospheric oxygen levels which moved from <10% to nearly 20% ( He et al . , 2019; Kump , 2008 ) . Overall , the confluence of tRNA expansion ( Figure 7 ) and oxidative stress , both external and internal , necessitates the emergence of ATD to maintain cellular proteostasis in Animalia .
All the components for biochemical assays were generated as mentioned in Kuncha et al . , 2018a . The sequence coding for threonyl-tRNA synthetase ( Homo sapiens , Danio rerio , and Escherichia coli ) was amplified from cDNA for human and zebrafish and from genome for E . coli and cloned into pET28b with an N-terminal 6X His-tag . All the proteins were expressed in E . coli BL21-CodonPlus ( DE3 ) -RIL strain , except E . coli ThrRS , which was expressed in E . coli BL21 . All the proteins were purified using the affinity-based column ( Ni-NTA ) , followed by size exclusion chromatography . Purified proteins were stored at −30°C in 150 mM NaCl , 200 mM Tris ( pH 7 . 5 ) and 50% glycerol . Genes coding M . musculus tRNAThr ( G4•U69 ) , tRNAAla is in vitro transcribed using MEGAshortscript T7 Transcription Kit ( Thermo Fisher Scientific , USA ) , followed by 3’ end labeling using CCA-adding enzyme ( Ledoux and Uhlenbeck , 2008 ) . Alanylation of tRNAThr ( G4•U69 ) was done using M . musculus alanyl-tRNA synthetase . EF-Tu activation experiments were performed using pyruvate kinase and phosphoenolpyruvate as explained in Routh et al . , 2016 . Deacylations were performed using a range of enzyme concentrations , while the concentration which gave a gradual curve was used for curve fitting in GraphPad Prism software , and each data point in the graph represents mean of at least three reading , while the error bars represent the standard deviation from the mean . HEK293T cells were acquired from ATCC and confirmed using STR profiling . HEK293T cells were cultured and maintained in DMEM containing 10% fetal bovine serum , 50 IU ml−1 penicillin , and 50 μg ml−1 streptomycin . Cultures were grown in a humidified incubator at 37°C and 5% CO2 . Transfections were done using Lipofectamine 3000 reagent ( Invitrogen ) according to the manufacturer’s manual . A routine check for mycoplasma contamination was checked using DAPI staining . For microscopy the cells were grown on coverslips , fixed with 10% formaldehyde , permeabilized with Triton X100 , and stained with DAPI ( Sigma ) . The cells were imaged in Zeiss Axioimager Z . 1 Microscope or Leica TCS SP8 . Mouse ES cells ( E14Tg2a ) were grown as described in Jana et al . , 2019 . Briefly , cells were cultured on tissue culture-treated plates/dishes coated with 0 . 1% gelatine , in GMEM media supplemented with L- glutamine , 100 µM β-mercaptoethanol , 1 mM non-essential amino acids ( Gibco ) , 100 units/ml human LIF supplemented with 10% fetal bovine serum ( Gibco ) . Cultures were grown in a humidified incubator at 37°C and 5% CO2 . mES cells were transfected using P3 Primary cell 4D-Nucleofector Kit . For colony formation assay of ES cells , 200 cells per well ( 6 well plate ) were plated and Leishman staining was done after 5 days of growth , the cell were grown in media containing a variable concentration of β-mercaptoethanol . For generating knockout in HEK293T , two SgRNAs were designed to target the intron1 and exon1 region of the Atd gene , the sequences of the gRNA are sgRNA1-5’ TTCGTCGTGCCCCGCCTCGTC 3’ and SgRNA2-5’ CAGATCGCGTCGAATTCCCC 3’ . The SgRNA oligos were cloned into the BbsI sites of pU6- ( BbsI ) -CBh-Cas9-T2A-iRFP670 and verified by sequencing . The U6-SgRNA1-gRNA scaffold cassette was amplified and cloned into the XbaI site of the pU6- ( BbsI ) -CBh-Cas9-T2A-iRFP670 plasmid contains the sgRNA2 to generate a dual sgRNA plasmid . These dual SgRNA plasmids were transfected into HEK293T cells , followed by cell sorting to enrich for transfected cells expressing iRFP670 . The sorted cells were cultured and clones were established by dilution cloning . The genomic DNA from replica plates of the clones were genotyped by PCR using primers flanking the sgRNAs regions to detect the deletion in ATD locus . The forward primer 5’ CACTGAGCGCCTTCTACAGAGTTG 3’ and reverse primer 5’ GAAAGTAGAAGGAACTCATAGTGAC 3’ . The ATD knockouts were further validated by sequencing the PCR amplicon and by performing western blot for ATD protein . Sg RNA oligos for mouse ES cell knockouts were designed at the exon1 and intron1 junction of the gene coding for ATD , the sequences of the oligos are 5’ GGCCGATGGAGACGCCGCGG 3’ and 5’ CCCTATCCGCGGAACCGTGC 3’ . The protocol for knockout generation in ES cells is identical to HEK293T cells , except that the plasmid was transfected into the cells using nucleofection . The mES cell colonies were screened using gene-specific primers 5’ CAAAGCTGGTCAATTCCACATCCG 3’ and 5’ ATTCTGAGAAGCGAGATGGCTCAC 3’ which flank the target region . The wild type and ATD knockout HEK293T cells were plated in 12 well culture plates ( 50000 cells/well ) and incubated for 24 hr in CO2 incubator . The following day different concentrations of H2O2 ( 50 , 75 , and 100 µM ) were added to the corresponding wells and incubated for 24 hr . After the incubation time , 0 . 5 mg/ml of MTT solution was added to each well and incubated in dark for 3 to 4 hr . The Formazan crystals were dissolved by adding 500 µl of 10% acidified-SDS , followed by transfer to 96 well plates . The intensity of the color was measured using a spectrophotometer at 562 nm . As a confirmatory step the number of live cells per well were counted under 10X objective , and 10X eyepiece magnification of compound microscope using Neubauer-improved counting chamber ( Paul Marienfeld GmbH and Co . KG , Germany ) , cells were stained using trypan-blue . In general , wild type and knockout cells ( treated and untreated ) are harvested , washed with PBS for 2 times and lysed using Laemmli buffer , boiled and separated on SDS-PAGE and transferred to PVDF membrane , followed by probing with appropriate primary and secondary antibody . In the case of luciferase experiments , wild type and knockout cells were transfected with FlucDM-EGFP constructs and treated with different concentrations of H2O2 . Following the treatment , these cells were lysed using NP-40 containing lysis buffer [50 mM Tris–HCl ( pH 7 . 8 ) , 150 mM NaCl , 1% NP-40 , 0 . 25% sodium deoxycholate , 1 mM EDTA , protease inhibitor cocktail ( Roche ) ] and centrifuged at 12 , 000 g at 4°C for 15 min , the supernatant was used as the soluble fraction and the pellet was boiled in Laemmli buffer that was used as the insoluble fraction and was subjected to immunoblotting . For purifying the reporter protein , GFP , the cells were transfected with pEGFP-N1 empty vector and allowed to grow for 24 hr , followed H2O2 treatment . Cells were initially washed with PBS and followed by lysis using a buffer containing 10 mM Tris ( pH 7 . 8 ) , 150 mM NaCl , 0 . 5 mM EDTA , 2 mM Na3VO4 , 10 mM NaF , 10 mM N-ethylmaleimide , protease inhibitor mixture , and 0 . 5% Nonidet P-40 . The lysates were subjected to centrifugation at 20 , 000 g , 10 min at 4°C , the supernatant was incubated with GFP-Trap beads ( ChromoTek ) for 2 hr at 4°C . The beads were washed with buffer containing ingredients of lysis buffer except for Nonidet P-40 , to remove non-specifically bound proteins . To the GFP bound beads Laemmli sample buffer is added , boiled , and loaded on the SDS-PAGE . After running the PAGE gel followed by coomassie staining , regions containing the protein are sliced and subjected to reduction , alkylation , and in-gel digestion using Trypsin as described by Shevchenko et al . , 2006 . Finally , the peptides were desalted and enriched as per the protocol in Rappsilber et al . , 2007 . The Q Exactive HF ( Thermo Fisher Scientific , Bremen , Germany ) was used to perform HCD mode fragmentation and LC-MS/MS analysis . Samples were fractionated using commercial column PepMap RSLC C18 , 3 μm , 100 Å , 75 μm i . d . ×150 mm . The LCs used were EASY-nLC 1200 systems ( Thermo Fisher Scientific , San Jose , CA ) . The column temperature was maintained at 30°C . The peptides were loaded in solvent A ( 5% Acetonitrile 0 . 1% formic acid in Ultrapure water ) and eluted with a nonlinear gradient of solvent B ( 0 . 1% TFA , 0 . 1% formic acid in 95% acetonitrile ) by a gradient to 25% of solvent B over 35 min and 40% for 10 mins at a constant flow rate of 300 nL/min . Instruments were configured for DDA using the full MS/DD−MS/MS setup . Full MS resolutions were set to 60000 at m/z 200 , and full MS AGC target was 3E6 with an IT of 100 ms . The mass range was set to 400–1650 . AGC target value for fragment spectra was set at 1E5 , and the intensity threshold was kept at 3E4 . Isolation width was set at 1 . 3 m/z . The normalized collision energy was set at 28% . Peptide match was set to preferred , and isotope exclusion was on . The data were analyzed in two software’s PEAKS6 . 0 . and Proteome Discoverer 2 . 2 . Peaks 6 . 0 De novo sequencing and database search analysis parameters: Trypsin , and 3 allowed missed cleavages in one peptide end . The parent mass tolerance of 10 ppm using monoisotopic mass , and fragment ion tolerance of 0 . 05 Da . Carbamidomethylation of cysteine ( +57 . 02 ) as a fixed modification , methionine oxidation ( +15 . 99 ) and deamidation of asparagine and glutamine ( NQ +0 . 98 ) were set as variable modifications . Data was validated using a false discovery rate ( FDR ) method . Peptide identifications were accepted for peptides with –10logP score cut-off was set to >20 to achieve theoretical peptide FDR of 0 . 1% . Proteome Discoverer is an MS data analysis platform provided by Thermo Fisher Scientific for its mass spectrometers . Raw files of GFP control and knockout samples were imported into Proteome Discoverer 2 . 2 , and HTSequest algorithm search was used . In both the software , the sample was searched against the in-house mutant database . The enzyme specificity was set to trypsin , and three missed cleavages were allowed . Carbamidomethylation of cysteine was set as fixed modification and oxidation of methionine as a variable modification . The identified peptide sequence with more than threshold scores ( –10logP score >50 for peaks 6 . 0 ) were used to BLAST against wild type GFP protein sequence to find out mutations . All the sequences are extracted using the Protein-BLAST search and the structure-based sequence alignments are generated using the T-Coffee server ( Notredame et al . , 2000 ) . Phylogenetic trees were constructed using iTOL server ( https://itol . embl . de/ ) . Images of different protein structures are generated using PyMOL . While the tRNA sequences are taken from GtRNAdb ( Chan and Lowe , 2016 ) . | The first animals evolved around 750 million years ago from single-celled ancestors that were most similar to modern-day organisms called the Choanoflagellates . As animals evolved they developed more complex body plans consisting of multiple cells organized into larger structures known as tissues and organs . Over time cells also evolved increased levels of molecules called reactive oxygen species , which are involved in many essential cell processes but are toxic at high levels . Animal cells also contain more types of molecules known as transfer ribonucleic acids , or tRNAs for short , than Choanoflagellate cells and other single-celled organisms . These molecules deliver building blocks known as amino acids to the machinery that produces new proteins . To ensure the proteins are made correctly , it is important that tRNAs deliver specific amino acids to the protein-building machinery in the right order . Each type of tRNA usually only pairs with a specific type of amino acid , but sometimes the enzymes involved in this process can make mistakes . Therefore , cells contain proofreading enzymes that help remove incorrect amino acids on tRNAs . One such enzyme – called ATD – is only found in animals . Experiments in test tubes reported that ATD removes an amino acid called alanine from tRNAs that are supposed to carry threonine , but its precise role in living cells remained unclear . To address this question , Kuncha et al . studied proofreading enzymes in human kidney cells . The experiments showed that , in addition to ATD , a second enzyme known as ThrRS was also able to correct alanine substitutions for threonines on tRNAs . However , reactive oxygen species inactivated the proofreading ability of ThrRS , suggesting ATD plays an essential role in correcting errors in cells containing high levels of reactive oxygen species . These findings suggest that as organisms evolved multiple cells and the levels of tRNA and oxidative stress increased , this led to the appearance of a new proofreading enzyme . Further studies found that ATD originated around 900 million years ago , before Choanoflagellates and animals diverged , indicating these enzymes might have helped to shape the evolution of animals . The next step following on from this work will be to understand the role of ATD in the cells of organs that are known to have particularly high levels of reactive oxygen species , such as testis and ovaries . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"cell",
"biology"
] | 2020 | Genomic innovation of ATD alleviates mistranslation associated with multicellularity in Animalia |
The sequence of events that initiates T cell signaling is dictated by the specificities and order of activation of the tyrosine kinases that signal downstream of the T cell receptor . Using a platform that combines exhaustive point-mutagenesis of peptide substrates , bacterial surface-display , cell sorting , and deep sequencing , we have defined the specificities of the first two kinases in this pathway , Lck and ZAP-70 , for the T cell receptor ζ chain and the scaffold proteins LAT and SLP-76 . We find that ZAP-70 selects its substrates by utilizing an electrostatic mechanism that excludes substrates with positively-charged residues and favors LAT and SLP-76 phosphosites that are surrounded by negatively-charged residues . This mechanism prevents ZAP-70 from phosphorylating its own activation loop , thereby enforcing its strict dependence on Lck for activation . The sequence features in ZAP-70 , LAT , and SLP-76 that underlie electrostatic selectivity likely contribute to the specific response of T cells to foreign antigens .
Signal transduction by the T cell receptor ( TCR ) triggers the activation of three non-receptor tyrosine kinases: Lck , ZAP-70 , and Itk ( Smith-Garvin et al . , 2009; Weiss and Littman , 1994 ) . A notable feature of this pathway is a strict hierarchy in kinase activation , which is accompanied by highly specific phosphorylation of substrates by each kinase ( Figure 1 ) . T cells must mount a strong and sustained response upon encountering a foreign peptide antigen bound to a major histocompatibility complex ( MHC ) molecule on an antigen-presenting cell , without launching an immune reaction against self-antigens . The origin of this selectivity is not well understood , and it cannot be explained by the modest differences in affinities between self and foreign peptide antigens for the T cell receptor ( Palmer and Naeher , 2009 ) . The pattern of tyrosine kinase activity downstream of the T cell receptor is implicated in a kinetic proofreading mechanism that has been posited to underlie the fidelity of the T cell response ( Chakraborty and Weiss , 2014; McKeithan , 1995 ) . In such a mechanism , complexes formed between the T cell receptor and antigens must be sufficiently long-lived to propagate a biochemical signal through the sequential steps of kinase activation and substrate phosphorylation within the T cell . 10 . 7554/eLife . 20105 . 003Figure 1 . T cell receptor-proximal signaling . ( A ) Initiating events in T cell receptor signaling ( based on Au-Yeung et al . , 2009 ) . ( B ) Ordered kinase activation and substrate phosphorylation by Lck and ZAP-70 leads to downstream signaling . Abbreviations: TCR , T cell receptor; MHC , major histocompatibility complex; Lck , lymphocyte-specific kinase; ZAP-70 , ζ-chain associated protein of 70 kilodaltons; Itk , interleukin-2-inducible T cell kinase; ITAMs , immunoreceptor tyrosine-based activation motifs; LAT , linker for the activation of T cells; SLP-76 , SH2-containing leukocyte protein of 76 kilodaltons; and MAPK , mitogen-activated protein kinase . DOI: http://dx . doi . org/10 . 7554/eLife . 20105 . 003 The Src-family kinase Lck initiates intracellular signaling in T cells by phosphorylating the cytoplasmic tails of CD3 γ , δ , and ε , and the ζ subunits of the T cell receptor ( Figure 1 ) . This phosphorylation occurs on immunoreceptor tyrosine-based activation motifs ( ITAMs ) , which are conserved sets of sequences bearing two tyrosines separated by ten to twelve residues ( Love and Hayes , 2010; Reth , 1989 ) . ZAP-70 , a Syk-family kinase with tandem SH2 domains , is recruited to the membrane by binding these doubly-phosphorylated ITAMs ( Chan et al . , 1991; Iwashima et al . , 1994; Wange et al . , 1993 ) . ZAP-70 is auto-inhibited before ITAM binding , and its activation requires that it be phosphorylated by Lck , both on its SH2-kinase linker and on the activation loop , a conserved regulatory segment in the kinase domain ( Deindl et al . , 2007; Yan et al . , 2013 ) . Once activated , ZAP-70 phosphorylates two scaffold proteins , LAT and SLP-76 ( Bubeck Wardenburg et al . , 1996; Au-Yeung et al . , 2009; Zhang et al . , 1998 ) . These scaffolds are phosphorylated at multiple sites , leading to the formation of phosphotyrosine-dependent signaling complexes ( Balagopalan et al . , 2010; Zhu et al . , 2003 ) . In one such complex , the Tec-family kinase Itk binds to phosphorylated SLP-76 through its SH2 domain ( Bunnell et al . , 2000 ) . Itk phosphorylates and activates phospholipase-Cγ1 ( PLCγ1 ) , which is bound to phosphorylated LAT . Signaling by PLCγ1 leads to a cytoplasmic calcium increase , thereby initiating part of the transcriptional response of the T cell to activation by antigens ( Weiss et al . , 1991 ) . The order of these signaling events is now well established ( Figure 1B ) , and is thought to provide tight control over the T cell response . It is not clear , however , how this sequence of kinase reactions is enforced , given that tyrosine kinases are generally considered to be promiscuous and even 'sloppy' enzymes ( Mayer , 2012 ) . In this study , we focus on the key molecular features of ZAP-70 , Lck , and their substrates that dictate the order of kinase activation and substrate phosphorylation in a T cell . Specifically , we examine why the substrate specificity of ZAP-70 is so narrow that it can only phosphorylate the scaffold proteins LAT and SLP-76 efficiently , and why ZAP-70 cannot phosphorylate and activate itself . We also ask why LAT and SLP-76 are not phosphorylated efficiently by Lck . The avoidance of phosphorylation of these two scaffold proteins by Lck is critical for ensuring that the more tightly regulated kinase , ZAP-70 , propagates the signal downstream only after it is activated by Lck . To answer these questions , we utilized a high-throughput mutagenesis and screening platform to measure kinase activity towards a large number of substrates simultaneously . Our approach generates sequence-activity relationships for variants of specific peptide substrates by genetically encoding scanning point-mutagenesis libraries of peptides into a bacterial surface-display system that has been developed previously ( Rice and Daugherty , 2008 ) . Phosphorylation of individual peptides in these libraries is detected by fluorescence-activated cell sorting ( FACS ) , coupled to Illumina DNA sequencing technology ( deep sequencing ) ( Bentley et al . , 2008 ) . We applied this high-throughput platform to measure the effect of proximal mutations on the phosphorylation of several tyrosine residues in LAT and the T cell receptor ζ subunit ( TCRζ ) . Our data show that the presence of multiple negatively-charged residues in the vicinity of tyrosines in LAT , and the exclusion of positively-charged ones , allows those tyrosines to be selected through electrostatic interactions with ZAP-70 . This sequence pattern also prevents phosphorylation of LAT by Lck . Brownian dynamics and molecular dynamics simulations suggest that ZAP-70 detects the negative charge on LAT and SLP-76 by using a substrate binding region on its catalytic domain that is enriched in positively-charged residues to an extent that appears to be unique among non-receptor tyrosine kinases . Our analysis also suggests that the inability of ZAP-70 to undergo robust autophosphorylation is due to steric and electrostatic repulsion that blocks the adoption of the enzyme-substrate complex in which the activation loop of one kinase domain acts as the substrate for another . This feature appears to be a natural consequence of the specialization of ZAP-70 towards LAT and SLP-76 as preferred substrates .
A widely-used approach to determine consensus sequences in kinase substrates relies on peptide libraries with random sequences surrounding a central tyrosine residue . In older versions of this technique , degenerate peptide libraries were treated with a kinase of interest , and the resulting phosphopeptides were separated and sequenced in bulk to determine the distribution of amino acid residues at each position ( Songyang et al . , 1995 ) . Recent iterations of this technique utilize positional scanning peptide libraries where degenerate sub-libraries are created in which single amino acid residues are fixed at various positions flanking the tyrosine . All of the sub-libraries are phosphorylated in parallel and then immobilized on a membrane or microarray for high-throughput detection ( Deng et al . , 2014; Hutti et al . , 2004 ) . While these methods provide a reliable way to determine consensus motifs and can provide insights into kinase specificity ( Begley et al . , 2015 ) , reliance on the analysis of degenerate mixtures prevents a direct comparison of the effects of specific mutations in particular substrate sequences . To directly determine the importance of specific residues in substrates of ZAP-70 and Lck , we developed a high-throughput platform to measure the relative rates of phosphorylation for hundreds of individual peptides simultaneously ( Figure 2 ) . This assay utilizes a previously developed scaffold for bacterial surface-display of peptides , the enhanced circularly permuted OmpX ( eCPX ) protein ( Rice and Daugherty , 2008 ) . This scaffold has been used to measure phosphorylation levels of tyrosine-containing peptides on the surface of E . coli cells upon addition of a tyrosine kinase to the cell suspension , followed by detection using a pan-phosphotyrosine antibody and flow cytometry ( Henriques et al . , 2013 ) . We expanded this technique by applying it to libraries of genetically encoded peptides and coupling it to fluorescence-activated cell sorting ( FACS ) , followed by deep sequencing ( Figure 2 ) . 10 . 7554/eLife . 20105 . 004Figure 2 . A high-throughput assay for tyrosine kinase specificity . Top left panel: E . coli cells are transformed with plasmids encoding a library of peptide variants fused to the bacterial surface-display scaffold , eCPX ( Rice and Daugherty , 2008 ) . Top right panel: Expression of the peptide-scaffold fusions is induced to permit surface-display of the peptides , then the peptides on the extracellular surface of the cells are phosphorylated by the addition of a tyrosine kinase to the cell suspension ( Henriques et al . , 2013 ) . Bottom right panel: Phosphorylated cells are labeled with a fluorescent pan-phosphotyrosine antibody , and cells with a high fluorescence signal are isolated by fluorescence-activated cell sorting . Bottom left panel: DNA from the sorted cells and an unsorted control population is isolated and sequenced by Illumina deep sequencing to determine the enrichment of the DNA sequence encoding each variant in the library after selecting for a high phosphorylation level . DOI: http://dx . doi . org/10 . 7554/eLife . 20105 . 004 In a typical experiment , E . coli cells were transformed with a DNA library encoding peptides fused to the eCPX scaffold . After growth and induction of scaffold expression , the cells were washed , then resuspended in a buffer with a tyrosine kinase , ATP , and Mg2+ . At an early time-point in the reaction , when it was less than 30% complete , the kinase activity was quenched by the addition of EDTA to the suspension . The cells were labeled with a fluorescent pan-phosphotyrosine antibody , and sorted for high phosphotyrosine level . DNA from both unsorted and sorted cells was isolated and deep-sequenced to determine the frequency of each peptide in the library before and after selection for high phosphorylation level . For DNA corresponding to each peptide , an enrichment value was calculated as described previously for a high-throughput binding assay ( McLaughlin et al . , 2012 ) . Briefly , the ratio of the abundance of DNA corresponding to a peptide in the sorted and unsorted samples was determined , and that enrichment ratio was normalized to the enrichment ratio for a reference member of the library . The normalized enrichment ratio for a particular DNA sequence in the library is a measure of the relative efficiency by which the corresponding peptide is phosphorylated by the kinase . To test the validity of our approach , we first generated a small DNA library encoding the wild-type sequences of peptide segments from LAT , SLP-76 , the putative ZAP-70 substrate p38α ( Salvador et al . , 2005 ) , and TCRζ ( see Figure 3A and Figure 3—figure supplement 1 for sequences of the peptides used ) . The peptide segments were 19–22 residues long , and this library was screened against the isolated kinase domains of ZAP-70 and Lck ( Figure 3B; see the appendix for a discussion of variation in peptide surface-display levels ) . The results of the initial screens recapitulated known specificity trends observed in T cells ( Chu et al . , 1996; Isakov et al . , 1996; Williams et al . , 1998 ) . They were also consistent with in vitro measurements of kinetic parameters obtained using purified kinases and peptide substrates ( Figure 3—figure supplement 2 , 3 ) . Specifically , we found that ZAP-70 efficiently phosphorylates LAT and SLP-76 , but not TCRζ ITAMs , while Lck is an efficient kinase for ITAMs , but not for LAT or SLP-76 . Aside from Lck phosphorylation of LAT Tyr 64 , neither kinase readily phosphorylated the first five tyrosine residues on LAT ( Figure 3B ) , which are not known to be phosphorylated in T cells ( Balagopalan et al . , 2010 ) . The relative phosphorylation rates for three of the phosphosites in our initial screen ( Tyr 132 and Tyr 191 of LAT and Tyr 145 of SLP-76 ) have been measured in T cells , and those results are consistent with our peptide-based measurements ( Houtman et al . , 2005 ) . 10 . 7554/eLife . 20105 . 005Figure 3 . Phosphorylation by ZAP-70 or Lck of a variety of peptides based on LAT , SLP-76 , p38α , and TCRζ . ( A ) Sequences surrounding the key tyrosine residues analyzed in this study . The focal tyrosine in each sequence is denoted as residue 0 , and other proximal tyrosines are highlighted by a black box . ( B ) Enrichment of peptides from a library of peptide sequences based on LAT , SLP-76 , p38α , and TCRζ after phosphorylation by ZAP-70 or Lck , followed by selection . Error bars represent standard deviations in enrichment values from two screens . The exact peptide sequences for each member of the library analyzed in panel B , including the locations of several tyrosine-to-phenylalanine mutants , are given in Figure 3—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 20105 . 00510 . 7554/eLife . 20105 . 006Figure 3—figure supplement 1 . List of peptide sequences in the library containing segments of LAT , SLP-76 , p38α , and TCRζ . Tyrosine residues and tyrosine-to-phenylalanine mutations are shown in bold . DOI: http://dx . doi . org/10 . 7554/eLife . 20105 . 00610 . 7554/eLife . 20105 . 007Figure 3—figure supplement 2 . List of purified peptides used for in vitro kinetic assays . DOI: http://dx . doi . org/10 . 7554/eLife . 20105 . 00710 . 7554/eLife . 20105 . 008Figure 3—figure supplement 3 . In vitro phosphorylation kinetics of purified peptides by the ZAP-70 and Lck kinase domains . All peptides were used at a concentration of 500 μM , and both kinases were used at a concentration of 1 μM . DOI: http://dx . doi . org/10 . 7554/eLife . 20105 . 008 The obvious common feature of the known phosphorylation sites for ZAP-70 is that each tyrosine is surrounded by several acidic residues , with only the rare positively-charged residue nearby ( Figure 3A ) . By contrast , Lck substrates typically have both positively- and negatively-charged residues near the tyrosine , and they have a near-neutral net charge . This suggests that high negative charge is likely to be a key determinant of whether a potential target site is phosphorylated by ZAP-70 , but the location and number of negatively-charged residues around each of the tyrosines in LAT and SLP-76 is not conserved . There may also be other roles for the negative charge on LAT and SLP-76 that are unrelated to kinase specificity . For example , it was shown recently that LAT phosphorylation results in clustering , and that the phosphatase CD45 is excluded from these LAT clusters because it is negatively-charged ( Bunnell et al . , 2002; Su et al . , 2016 ) . To define the sequence determinants of efficient phosphorylation by ZAP-70 , we created three scanning point-mutagenesis libraries based on LAT sequences spanning Tyr 127 , Tyr 132 , and Tyr 226 . In these libraries , every possible single mutation in a 20 residue peptide segment was represented with near-equal stoichiometry , as verified by DNA sequencing . Each library was screened against the ZAP-70 kinase domain in triplicate or quadruplicate ( Figures 4 and 5 ) . These data allow us to assess the impact of all possible point mutations at a single site on a given peptide , as shown for substitutions of Asp 225 and Glu 231 in the LAT Tyr 226 peptide ( Figures 4A and B ) . The data also reveal the impact of individually introducing any particular amino acid residue at all sites within that peptide , as shown for lysine substitutions at each position in the LAT Tyr 226 peptide ( Figure 4C ) . The full datasets from each screen are represented as heatmaps , which display the impact of all possible individual substitutions at every site in the peptide ( Figure 5 ) . 10 . 7554/eLife . 20105 . 009Figure 4 . Effect of single amino acid substitutions on the phosphorylation of LAT Tyr 226 by ZAP-70 . ( A ) Impact of all amino acid substitutions at LAT Asp 225 on the phosphorylation of Tyr 226 by ZAP-70 . ( B ) Impact of all amino acid substitutions at LAT Glu 231 on the phosphorylation of Tyr 226 by ZAP-70 . ( C ) Impact of lysine substitutions at all 20 positions ( LAT residues 214–233 ) on the phosphorylation of Tyr 226 by ZAP-70 . Error bars represent standard deviations from an average of four measurements . The enrichment values of each variant were normalized to the enrichment value of the wild-type sequence and presented on a logarithmic scale . Thus , a value of 0 indicates that the substitution did not impact phosphorylation relative to the wild-type sequence , whereas positive and negative values mean that the substitution enhanced or diminished phosphorylation , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 20105 . 00910 . 7554/eLife . 20105 . 010Figure 5 . ZAP-70 phosphorylation of LAT point mutant libraries . ( A ) Average enrichment values from four independent screens for phosphorylation of the LAT Tyr 226 library by ZAP-70 . ( B ) Correlation between the rates of phosphorylation of 11 purified LAT Tyr 226 variants and their enrichment values from the screen shown in panel A . Horizontal error bars represent the standard deviations from three kinetic measurements with purified peptides , and vertical error bars represent the standard deviations from four screens . ( C ) Average enrichment values from three independent screens for phosphorylation by ZAP-70 of the LAT Tyr 127 library in a Y132F background . ( D ) Average enrichment values from three independent screens for phosphorylation by ZAP-70 of the LAT Tyr 132 library in a Y127F background . All enrichment values are log10-transformed and normalized relative to the parent peptide sequence in that screen , which has a value of 0 . DOI: http://dx . doi . org/10 . 7554/eLife . 20105 . 01010 . 7554/eLife . 20105 . 011Figure 5—figure supplement 1 . Correlations between enrichments from four replicates for ZAP-70 phosphorylation of the point mutant library spanning LAT Tyr 226 . DOI: http://dx . doi . org/10 . 7554/eLife . 20105 . 01110 . 7554/eLife . 20105 . 012Figure 5—figure supplement 2 . Phosphorylation by the ZAP-70 kinase domain of peptides spanning LAT Tyr 226 with various charge-altering substitutions . All peptides were used at a concentration of 500 μM , and ZAP-70 was used at a concentration of 1 μM . DOI: http://dx . doi . org/10 . 7554/eLife . 20105 . 01210 . 7554/eLife . 20105 . 013Figure 5—figure supplement 3 . Phosphorylation by the ZAP-70 kinase domain of peptides spanning LAT Tyr 191 with the R189A or R189D substitutions . ZAP-70 was used at a concentration of 1 μM . DOI: http://dx . doi . org/10 . 7554/eLife . 20105 . 01310 . 7554/eLife . 20105 . 014Figure 5—figure supplement 4 . Phosphorylation by the ZAP-70 kinase domain of peptides spanning LAT Tyr 132 with or without the G131D substitution . ZAP-70 was used at a concentration of 1 μM . DOI: http://dx . doi . org/10 . 7554/eLife . 20105 . 01410 . 7554/eLife . 20105 . 015Figure 5—figure supplement 5 . Sequence logos based on an alignment of vertebrate LAT segments surrounding Tyr 132 . Note that Tyr 127 is not conserved outside of mammals , and that the PLCγ SH2-binding motif ( Y[I/L]XV ) is conserved . DOI: http://dx . doi . org/10 . 7554/eLife . 20105 . 015 The results of these screens were highly reproducible ( Figure 5—figure supplement 1 ) , the trends were independent of the specific pan-phosphotyrosine antibody used , and enrichment values showed a strong correlation with measurements of phosphorylation rates for purified peptides ( Figure 5B ) . In all three screens , mutation of the target tyrosine residues ( Tyr 226 , Tyr 127 , and Tyr 132 , respectively ) resulted in a substantial depletion of DNA encoding those peptides after sorting ( negative enrichment value relative to wild-type ) , setting a 'floor' for the detection limit of the assay ( Figure 5A , C and D ) . The relative magnitudes of the kinetic effects and the bacterial display selectivity effects ( the slope of the line in Figure 5B ) depend on the peptide being tested . The data shown in Figure 5B are for the peptide spanning LAT Tyr 226 , a good substrate for ZAP-70 , and individual substitutions only have a modest , albeit measurable , effect on phosphorylation rate . By contrast , for ZAP-70 phosphorylation of a poor substrate , such as the peptide encompassing LAT Tyr 132 ( Figure 5D ) , substitutions at Gly 131 result in as much as a 16-fold enhancement in phosphorylation rate ( Figure 5—figure supplement 4 ) . A striking result from these screens is that the selection of substrates by ZAP-70 is controlled by an electrostatic filter , whereby the presence of a lysine or arginine residue anywhere within seven residues upstream or downstream of the substrate tyrosine severely compromises the efficiency of phosphorylation . A similar , but less dramatic , reduction in phosphorylation efficiency was observed when tyrosine-proximal residues were substituted by histidine . Replacement of native negatively-charged residues by neutral residues had a mild detrimental effect on phosphorylation efficiency , and introduction of acidic residues at neutral positions was often beneficial ( Figure 5 ) . Consistent with these observations , replacement of a single glutamate residue by lysine , either upstream or downstream of the tyrosine , resulted in a 30–40% reduction in the LAT Tyr 226 phosphorylation rate by ZAP-70 in kinase assays using purified peptide substrates , and mutation of multiple glutamates to alanine or lysine residues further reduced phosphorylation ( Figure 5—figure supplement 2 ) . The deleterious effect of arginine and lysine residues at any position also explains the slow phosphorylation of LAT Tyr 191 by ZAP-70 ( Figure 3—figure supplement 3 ) . Replacement of Arg 189 ( at the −2 position ) by alanine or aspartate residues successively improved the efficiency of Tyr 191 phosphorylation by ZAP-70 ( Figure 5—figure supplement 3 ) . In addition to revealing the exclusion rule for arginine and lysine , the screens also emphasize the importance of particular amino acid residues at three specific positions . ZAP-70 has a very strong preference for an aspartate residue at the −1 position relative to the tyrosine residue and a modest preference for a glutamate or hydrophobic residue at the +1 position , consistent with known preferences for the ZAP-70 paralog Syk ( Deng et al . , 2014; Schmitz et al . , 1996; Xue et al . , 2012 ) . There is also a preference for a hydrophobic residue at the +3 position in all three peptides , which is a feature common to other tyrosine kinases ( Bose et al . , 2006; Deng et al . , 2014 ) . Given the strong preference for an aspartate residue at the −1 position , we were intrigued that the residue immediately before Tyr 132 in LAT is a glycine . Phosphorylation of Tyr 132 is essential for T cell differentiation and function , as it generates the docking site for PLCγ1 , which initiates calcium signaling ( Roncagalli et al . , 2010 ) . In our screen for Tyr 132 phosphorylation ( Figure 5D ) , replacement of Gly 131 with virtually any other residue improved the efficiency of phosphorylation by ZAP-70 , and the G131D mutation caused a 16-fold enhancement of phosphorylation by ZAP-70 in kinase assays using purified peptides ( Figure 5—figure supplement 4 ) . Gly 131 in LAT is highly conserved in mammals , and it is also common in other vertebrates ( Figure 5—figure supplement 5 ) . PLCγ1 binding sites on other proteins do not contain a glycine adjacent to the phosphotyrosine , indicating that this is not a requirement for SH2 domain engagement ( Jones et al . , 2006; Leung et al . , 2014 ) . A slow rate of phosphorylation at Tyr 132 in LAT may be important for kinetic proofreading during T cell receptor signaling , the significance of which will be explored in future studies . To understand why Lck readily phosphorylates ITAM peptides but not LAT- or SLP-76-based peptides , we screened three libraries against Lck in our high-throughput platform ( Figure 6 ) . Two libraries were based on the second ITAM in TCRζ , one spanning Tyr 111 in a Y123F background , and the other spanning Tyr 123 in a Y111F background ( Figures 6A and B , respectively ) . The two ITAM libraries have the tyrosine close to one or the other terminus of the peptide . Thus , each library provides information on Lck specificity either downstream or upstream of the tyrosine residue , respectively . We also screened one of the LAT point mutant libraries ( LAT Tyr 127 ) against the Lck kinase domain ( Figure 6C ) . We chose this library because the sequence surrounding LAT Tyr 127 has the canonical features of most ZAP-70 substrates ( many acidic residues , no lysines or arginines , and a −1 aspartate ) , but in vitro measurements indicated that it was still a modest substrate for Lck ( Figure 3—figure supplement 3 ) . 10 . 7554/eLife . 20105 . 016Figure 6 . Lck phosphorylation of TCRζ ITAM and LAT point mutant libraries . ( A ) Average enrichment values from two independent screens for phosphorylation by Lck of the TCRζ Tyr 111 library in a Y123F background . ( B ) Average enrichment values from two independent screens for phosphorylation by Lck of the TCRζ Tyr 123 library in a Y111F background . ( C ) Average enrichment values from two independent screens for phosphorylation by Lck of the LAT Tyr 127 library in a Y132F background . ( D ) Comparison of the effect of lysine mutations on LAT Tyr 127 phosphorylation by ZAP-70 and Lck . Error bars represent standard deviations in enrichment values from three and two screens with ZAP-70 and Lck , respectively . All enrichment values are log10-transformed and normalized relative to the parent peptide sequence in that screen , which has a value of 0 . DOI: http://dx . doi . org/10 . 7554/eLife . 20105 . 01610 . 7554/eLife . 20105 . 017Figure 6—figure supplement 1 . Phosphorylation by the Lck kinase domain of peptides spanning LAT Tyr 226 with various charge-altering substitutions . All peptides were used at a concentration of 500 μM , and Lck was used at a concentration of 1 μM . DOI: http://dx . doi . org/10 . 7554/eLife . 20105 . 017 Analysis of these three libraries revealed three major differences between ZAP-70 and Lck: ( 1 ) In contrast to the lysine/arginine exclusion rule for ZAP-70 , Lck has a modest preference for positively-charged residues , ( 2 ) Lck does not have a requirement for an aspartate residue at the −1 position , and ( 3 ) ZAP-70 recognizes residues both upstream and downstream of the phosphosite , while Lck largely utilizes downstream residues for substrate discrimination . This point is important for understanding the difference in autophosphorylation ability between the two kinases , as discussed below . Lck has a strong preference for a bulky hydrophobic residue at the −1 position relative to the phosphosite , and it does not tolerate an aspartate residue at this position . This preference explains why Lck does not readily phosphorylate tyrosine residues in LAT and SLP-76 that contain an aspartate residue at the −1 position ( Figure 3 ) . It also explains why Lck phosphorylates the first tyrosine residue in each ITAM faster than the second one ( Housden et al . , 2003 ) , as the first tyrosine is typically preceded by a leucine while the second is preceded by a smaller aliphatic or polar residue ( Figure 3A ) . The lysine/arginine exclusion rule observed for ZAP-70 substrate sequences clearly does not apply to Lck . Mutation of acidic residues located upstream of Tyr 127 in LAT to virtually any other residue , including lysine and arginine , resulted in a slight , but significant , enhancement of phosphorylation efficiency by Lck ( Figure 6C ) . Downstream of the tyrosine , lysine and arginine residues were not tolerated at the +1 to +3 positions , but positively-charged residues were preferred at positions located further downstream ( Figure 6D ) . This is consistent with the presence of several lysine and arginine residues located five to ten positions downstream of the phosphosites in TCRζ ITAMs ( Figure 3A ) , and mutation of native lysine residues downstream of TCRζ Tyr 111 to aspartate or glutamate was detrimental for Tyr 111 phosphorylation by Lck ( Figure 6A ) . To validate our conclusions , we measured the phosphorylation kinetics of purified mutant LAT peptides by Lck . Simultaneous mutation to alanine of several glutamates located five to seven residues upstream of LAT Tyr 226 had no measurable effect on phosphorylation by Lck , and mutation of those residues to lysine only reduced the rate of phosphorylation modestly ( Figure 6—figure supplement 1 ) . The same mutations to lysine in LAT reduced the rate of Tyr 226 phosphorylation by ZAP-70 by more than 80% ( Figure 5—figure supplement 2 ) . Mutation of Glu 231 ( +5 position ) to alanine caused a two-fold increase in the Lck-catalyzed rate of LAT Tyr 226 phosphorylation , suggesting that a downstream negative charge on substrates might repel Lck , and this rate was further enhanced by mutation of that same glutamate residue to lysine ( Figure 6—figure supplement 1 ) . By contrast , the E231K mutation had a negative impact on LAT Tyr 226 phosphorylation by ZAP-70 ( Figure 5—figure supplement 2 ) . These observations explain why Lck phosphorylates LAT Tyr 127 with moderate efficiency: this is the only ZAP-70 substrate without a negatively-charged residue downstream of the tyrosine ( Figure 3A ) . Unlike Lck , ZAP-70 has a relatively large positively-charged patch surrounding the binding site for peptide substrates , which provides potential contact points for substrate residues both upstream and downstream of the tyrosine ( Figures 7A and B ) . This region in ZAP-70 has at least ten lysine and arginine residues and only three negatively-charged residues , one of which is required for catalysis ( Asp 461 ) . To identify interactions that might dictate substrate recognition , we ran molecular dynamics simulations of the ZAP-70 kinase domain bound to a peptide spanning LAT residues 219 to 233 ( surrounding Tyr 226 ) . Using the structure of the substrate-bound insulin receptor kinase domain as a guide ( Hubbard , 1997; Parang et al . , 2001 ) , we built a model of the ZAP-70 kinase domain in an active conformation , bound to the LAT peptide , ATP , and two Mg2+ ions ( there are no crystal structures of substrate complexes of ZAP-70 ) . The peptide was modeled into the active site by docking the tyrosine residue and the three residues immediately after it so that they form an antiparallel β sheet with the activation loop , as seen in tyrosine kinase-substrate co-crystal structures ( Bose et al . , 2006; Hubbard , 1997; Levinson et al . , 2006; Parang et al . , 2001; Zhang et al . , 2006 ) . The remaining peptide residues were modeled in an arbitrary conformation , projecting into the solvent and away from the kinase domain ( Figure 7—figure supplement 1 ) . 10 . 7554/eLife . 20105 . 018Figure 7 . Electrostatic features of the ZAP-70 and Lck kinase domains , and peptide binding modes of ZAP-70 . ( A ) The electrostatic surface potential , calculated using APBS ( Baker et al . 2001 ) and displayed using PyMOL ( Schrödinger 2015 ) , of the ZAP-70 kinase domain ( PDB code 1U59 ) and the Lck kinase domain ( PDB code 1QPJ ) . ( B ) Schematic representation ( left ) and an instantaneous structure from a molecular dynamics simulation ( right ) of the ZAP-70 kinase domain bound to a peptide containing LAT Tyr 226 . ( C ) Two different instantaneous structures from molecular dynamics simulations of ZAP-70 bound to the same peptide , highlighting key kinase-substrate interactions . DOI: http://dx . doi . org/10 . 7554/eLife . 20105 . 01810 . 7554/eLife . 20105 . 019Figure 7—figure supplement 1 . Starting structure used for molecular dynamics simulations of ZAP-70 bound to a peptide spanning LAT Tyr 226 . DOI: http://dx . doi . org/10 . 7554/eLife . 20105 . 01910 . 7554/eLife . 20105 . 020Figure 7—figure supplement 2 . Hydrogen bonds between the ZAP-70 activation loop and the peptide +1 and +3 positions during five molecular dynamics trajectories . The presence of a hydrogen bond in an instantaneous structure is noted in the graphs with a dot when the hydrogen bond distance was less than 2 . 5 Å . Side chains in the chemical structure schematic are omitted for clarity . DOI: http://dx . doi . org/10 . 7554/eLife . 20105 . 02010 . 7554/eLife . 20105 . 021Figure 7—figure supplement 3 . The close proximity between the LAT phospho-acceptor tyrosine and ATP ( left ) or the catalytic aspartate in ZAP-70 ( right ) during five molecular dynamics trajectories . A red dot indicates that , in that particular instantaneous structure , the two atoms connected by a dotted red line in the schematic were less than the distance cutoff noted above each graph . DOI: http://dx . doi . org/10 . 7554/eLife . 20105 . 02110 . 7554/eLife . 20105 . 022Figure 7—figure supplement 4 . Ion pairs between LAT Asp 225 ( −1 position ) and lysine residues 504 and 538 on ZAP-70 during five molecular dynamics trajectories . Red dots indicate that the distance between Asp 225 Cγ and Lys 504 Nε was less than 5 Å in that particular instantaneous structure , and blue dots correspond to the same measurement for Asp 225 and Lys 538 . DOI: http://dx . doi . org/10 . 7554/eLife . 20105 . 02210 . 7554/eLife . 20105 . 023Figure 7—figure supplement 5 . The frequency of ion pairs observed during instantaneous structures sampled every nanosecond from each trajectory . Ion pairs were defined as existing when the distance between a LAT Glu Cδ atom was less than 4 . 5 Å from a ZAP-70 Lys Nε atom . DOI: http://dx . doi . org/10 . 7554/eLife . 20105 . 02310 . 7554/eLife . 20105 . 024Figure 7—figure supplement 6 . Ion pairs between LAT Glu 231 ( +5 position ) and Arg 514 on ZAP-70 during five molecular dynamics trajectories . Red dots indicate that the distance between Glu 231 Cδ and Arg 514 Cζ was less than 5 Å in that particular instantaneous structure . DOI: http://dx . doi . org/10 . 7554/eLife . 20105 . 02410 . 7554/eLife . 20105 . 025Figure 7—figure supplement 7 . In vitro phosphorylation kinetics of purified peptides by the wild-type ZAP-70 kinase domain and an FG loop mutant containing the K541A , K542A , and K544A substitutions . All peptides were used at a concentration of 500 μM , and both kinases were used at a concentration of 1 μM . DOI: http://dx . doi . org/10 . 7554/eLife . 20105 . 02510 . 7554/eLife . 20105 . 026Figure 7—figure supplement 8 . In vitro phosphorylation kinetics of purified peptides surrounding LAT Tyr 226 with various upstream charge-altering mutations by the wild-type ZAP-70 kinase domain and an FG loop mutant containing the K541A , K542A , and K544A substitutions . All peptides were used at a concentration of 500 μM , and both kinases were used at a concentration of 1 μM . DOI: http://dx . doi . org/10 . 7554/eLife . 20105 . 02610 . 7554/eLife . 20105 . 027Figure 7—figure supplement 9 . In vitro phosphorylation kinetics of purified peptides surrounding LAT Tyr 226 with the E231A or E231K mutations by the wild-type ZAP-70 kinase domain and mutants containing the R514A or R514E substitutions . All peptides were used at a concentration of 500 μM , and all three kinases were used at a concentration of 1 μM . DOI: http://dx . doi . org/10 . 7554/eLife . 20105 . 027 Using this starting structure , we generated five molecular dynamics trajectories that ran for 500 ns each ( Supplementary files 1 and 2 ) . In all five simulations , the short antiparallel β sheet formed by the peptide and the activation loop persisted for the length of each trajectory ( Figure 7—figure supplement 2 ) . Additionally , the phospho-acceptor tyrosine residue often adopted a configuration in close proximity to the catalytic aspartate residue ( Asp 461 ) and the γ-phosphate of ATP , consistent with the requirements for phospho-transfer ( Figure 7—figure supplement 3 ) . The molecular dynamics simulations revealed the formation of a stable ion pair between the −1 aspartate of the substrate ( Asp 225 ) and Lys 504 , a highly conserved residue in tyrosine kinases ( Figure 7C and Figure 7—figure supplement 4 ) . Located near the active site , and within a substrate-binding site that leads into it , is a loop connecting helices F and G of the kinase domain ( the FG loop , Figure 7B ) . This loop presents four lysine residues towards the substrate ( residues 538 , 541 , 542 , and 544 ) . The −1 Asp frequently formed an interaction with Lys 538 and made occasional contact with Lys 544 ( Figure 7C and Figure 7—figure supplement 4 ) . In Lck , the residues in the FG loop corresponding to Lys 538 and Lys 544 in ZAP-70 are isoleucine and threonine , respectively , which presumably removes the requirement for Asp at the −1 position of the substrate and enforces the preference for hydrophobic residues that has been noted previously ( Figure 6 ) ( Songyang et al . , 1995 ) . The molecular dynamics simulations show that the FG loop in ZAP-70 is also important for recognition of residues located at the −5 to −7 positions with respect to the tyrosine residue ( Figure 7C ) . For the LAT Tyr 226 peptide , the upstream residues Glu 219 , Glu 220 , and Glu 221 formed frequent but transient ion pairs with all four lysine residues on the FG loop as well as with Lys 504 . These ion pairs typically persisted for 5 to 10 ns before breaking and forming again . While any specific ion pair was short-lived , the clusters of negative charges on LAT and positive charges on ZAP-70 ensured that at least one or more ion pairs were formed in at least 50% of the instantaneous structures sampled from each trajectory ( Figure 7—figure supplement 5 ) . The simulations also showed that ZAP-70 recognizes negatively-charged residues located downstream of Tyr 226 . In the molecular dynamics trajectories , Glu 231 , located five residues downstream of Tyr 226 , frequently formed an interaction with Arg 514 in ZAP-70 ( Figure 7C and Figure 7—figure supplement 6 ) . In Lck , the residue corresponding to this arginine is a glycine , which explains why Lck does not select for negative residues at this position in the substrate . The results of the molecular dynamics simulations are consistent with a number of experimental measurements of kinase activity . We purified a ZAP-70 kinase domain construct in which three of the four lysine residues in the FG loop were replaced by alanine ( a 'KKMK-to-AAMA' mutation in the FG loop ) . This mutant ZAP-70 displayed substantially reduced activity against most of the LAT and SLP-76 peptides in kinase assays , and also showed an increased ability to phosphorylate ITAMs , when compared to wild-type ZAP-70 ( Figure 7—figure supplement 7 ) . When upstream glutamate residues in a LAT Tyr 226-containing peptide were replaced by alanine or lysine , there was a marked reduction in phosphorylation efficiency by wild-type ZAP-70 . By contrast , the phosphorylation rate for ZAP-70 bearing the KKMK-to-AAMA mutation was not affected by these charge-altering mutations in LAT ( Figure 7—figure supplement 8 ) . Arg 514 in ZAP-70 interacts with negatively-charged residues downstream of the tyrosine , and the importance of this residue for the ability of ZAP-70 to phosphorylate LAT was validated by mutation of this residue to alanine and glutamate . Both mutations reduced the rate at which ZAP-70 phosphorylated LAT Tyr 226 . Replacing Glu 231 ( +5 position ) in the LAT peptide with lysine reduced phosphorylation by wild-type ZAP-70 , but the presence of the lysine in the substrate peptide had no effect on phosphorylation by ZAP-70 with Arg 514 mutated to alanine or glutamate , consistent with the removal of a repulsive interaction ( Figure 7—figure supplement 9 ) . The E231K mutation in the LAT Tyr 226 peptide results in reduced phosphorylation by ZAP-70 . In contrast , for Lck , this mutation results in increased phosphorylation ( Figure 5—figure supplement 2 and Figure 6—figure supplement 1 ) . This is consistent with the opposite charge preferences of ZAP-70 and Lck downstream of substrate tyrosines ( Figure 6 ) . While the molecular dynamics simulations point to interactions between ZAP-70 and LAT that are important once the peptide is bound , we wondered if long-range electrostatic interactions could also play a role in substrate recognition . We carried out Brownian dynamics simulations to address this question ( Gabdoulline and Wade , 1998; Northrup and Erickson , 1992 ) . In these simulations , the diffusive motions of molecules can be tracked over much longer timescales than in conventional molecular dynamics , because internal motions and the detailed structure of the solvent are neglected ( Figure 8A ) . Brownian dynamics trajectories were calculated using the program SDA ( Martinez et al . , 2015 ) , and the kinase domains and the peptide substrates were both treated as rigid bodies that interact only through electrostatic and van der Waals forces . The effects of water molecules and ions were modeled through the Poisson-Boltzmann equation for continuum electrostatics ( Baker et al . , 2001 ) . 10 . 7554/eLife . 20105 . 028Figure 8 . Brownian dynamics simulations of LAT peptide association with ZAP-70 and Lck kinase domains . ( A ) Schematic representation of Brownian dynamics trajectories . For each kinase-peptide pair , 10 , 000 trajectories were initiated by randomly placing the peptide on the surface of a sphere of radius 150 Å , centered on the center of mass of the kinase domain . Trajectories were terminated when the peptide moved more than 250 Å from the center of mass of the protein . ( B ) Distribution of various LAT Tyr Cα atoms ( shown as light red dots ) within 70 Å of the ZAP-70 or Lck kinase domain center of mass . For visualization purposes , a random sampling of 0 . 1% of all peptide positions within the 70 Å sphere in all 10 , 000 trajectories is displayed . The density of LAT Tyr Cα atoms in each panel reflects the amount of time that peptide spent in the vicinity of the kinase domain . ( C ) Frequency of substrate tyrosine Cα atoms within four quadrants encompassing the ZAP-70 or Lck kinase domain . Only peptide positions within 70 Å of the center of mass of the kinase domain , as shown in panel B , were considered for this analysis . Quadrants are defined based on atoms in the N-lobe or C-lobe of the kinase , proximal or distal to the active site , as shown in the left panel . DOI: http://dx . doi . org/10 . 7554/eLife . 20105 . 02810 . 7554/eLife . 20105 . 029Figure 8—figure supplement 1 . Ionic strength-dependent phosphorylation of LAT Tyr 127 by ZAP-70 . The substrate peptide was fused to SUMO to facilitate blotting , and isolated SUMO was also analyzed as a negative control for phosphorylation ( top row ) . Substrate concentrations were 175 μM , and the ZAP-70 kinase domain was used at a concentration of 500 nM . DOI: http://dx . doi . org/10 . 7554/eLife . 20105 . 029 We generated Brownian dynamics trajectories using crystal structures of the kinase domains of ZAP-70 and Lck ( PDB codes 1U59 ( Jin et al . , 2004 ) and 1QPJ ( Zhu et al . , 1999 ) , respectively ) . Peptides were modeled as rigid and extended β strands of 15 residues , with 7 residues on either side of the phospho-acceptor tyrosine , with sequences corresponding to various LAT phosphosites . For each kinase-peptide pair , 10 , 000 trajectories were initiated by randomly placing the peptide on the surface of a sphere of radius 150 Å , centered on the center of mass of the kinase domain ( Figure 8A ) . A Brownian dynamics trajectory was then generated for each initial configuration of the peptides . Trajectories were terminated when the peptide moved more than 250 Å from the center of mass of the protein . For analysis , coordinates were sampled from the trajectories every 200 ps , leading to ~1 million to ~100 million instantaneous structures for the different peptides ( peptides that tend to stay longer in the vicinity of the protein lead to longer trajectories ) . To visualize the Brownian dynamics trajectories for each peptide , we selected all instantaneous structures for which the Cα atom of the tyrosine residue is within 70 Å of the center of mass of the protein . A random sampling of 0 . 1% of these peptide positions is shown in Figure 8B for six different systems , five for ZAP-70 with peptide segments corresponding to Tyr 127 , Tyr 226 , Tyr 171 , Tyr 110 , and Tyr 191 in LAT , and one for Lck with the peptide corresponding to Tyr 127 in LAT . The first three tyrosine residues are good substrates for ZAP-70 , while Tyr 110 and Tyr 191 are relatively poor ones ( Figure 3 ) . Tyr 127 and Tyr 226 both have a net charge of −6 , with the charges on Tyr 171 , Tyr 110 , and Tyr 191 being −4 , −3 , and −2 , respectively . Comparison of the number of instantaneous structures within the 70 Å sphere for the different peptides is instructive – the greater this number , the longer the peptide stayed in the vicinity of the kinase domain before escaping beyond the 250 Å horizon . By this metric , Tyr 127 , a good ZAP-70 substrate , is enriched around the ZAP-70 kinase domain by a factor of ~150 compared to Tyr 191 , a poor substrate . Also , Tyr 127 is enriched around the ZAP-70 kinase domain by a factor of ~200 relative to the same peptide around Lck . Within the 70 Å sphere , the good substrates for ZAP-70 are preferentially localized around the active site . This effect is particularly striking for Tyr 127 and Tyr 226 ( Figure 8B , top-left and top-middle panels ) . Tyr 191 , which is a poor substrate for ZAP-70 , does not cluster around the ZAP-70 active site ( Figure 8B , bottom middle panel ) . A quantitative analysis of this feature is shown in Figure 8C . Tyr 127 , which is a good substrate for ZAP-70 and is also phosphorylated by Lck , does not cluster around the Lck active site ( Figure 8B , bottom right panel ) . Thus , long-range electrostatic steering appears to be a feature of the ZAP-70-LAT interaction that is not shared by Lck . The importance of electrostatics for substrate recruitment by ZAP-70 is substantiated by the fact that ZAP-70 activity against LAT peptides is strongly dependent on ionic strength ( Figure 8—figure supplement 1 ) . The role of electrostatics in substrate recognition by ZAP-70 contrasts with the previously appreciated role for electrostatics in controlling the activation of Src-family kinases , which is also salt dependent ( Ozkirimli et al . , 2008 ) . Src-family kinases , including Lck , are activated by trans-autophosphorylation of their activation loops ( Cooper and MacAuley , 1988; Hui and Vale , 2014; Moarefi et al . , 1997 ) . ZAP-70 , by contrast , requires Lck or another Src-family kinase for activation ( Chu et al . , 1996; Williams et al . , 1998 ) , as it cannot efficiently phosphorylate its own activation loop or SH2-kinase linker ( Yan et al . , 2013 ) ( and data not shown ) . The molecular basis for activation loop phosphorylation in tyrosine kinases is still not completely understood . Early studies on Src-family kinases demonstrated that they could phosphorylate peptides based on their activation loop sequences , and that the primary sequence of the peptide impacted phosphorylation efficiency; however , activation loop phosphorylation was substantially faster with full-length protein substrates when compared to peptides , indicating a role for the tertiary structure of the substrate in autophosphorylation reactions ( Casnellie et al . , 1982; Hunter , 1982 ) . We carried out a comprehensive manual inspection of all crystal structures of tyrosine kinases in the Protein Data Bank to identify plausible autophosphorylation complexes that might be suggested by crystal lattice packing . We identified one crystal structure of the insulin-like growth factor-1 receptor ( IGF1-R ) kinase domain in which two of the four molecules in the asymmetric unit interact in a manner that appears compatible with activation loop phosphorylation , with one kinase taking the role of the enzyme and the other that of the substrate ( PDB code 3LVP ) ( Nemecek et al . , 2010 ) . Recently , this same autophosphorylation complex was identified independently in an automated survey of the Protein Data Bank ( Xu et al . , 2015 ) . We built homology models for Lck and c-Src autophosphorylation complexes based on the IGF1-R structure ( PDB code 3LVP ) . Several other trans-autophosphorylation complexes have been proposed for tyrosine kinases , including potential structures for activation loop phosphorylation ( Wu et al . , 2008; Xu et al . , 2015 ) and C-terminal tail phosphorylation ( Chen et al . , 2008 ) . We considered these alternative proposals for trans-autophosphorylation complexes but favored the model based on PDB code 3LVP as it was consistent with our mutagenesis experiments ( noted below , in this section and the following one ) . Trans-autophosphorylation complexes have also recently been defined for certain serine/threonine kinases ( Oliver et al . , 2006; Pike et al . , 2008; Zorba et al . , 2014 ) . These structures do not provide suitable models for Lck and ZAP-70 , because the shorter activation loops of the tyrosine kinases do not allow them to form the corresponding complexes . In the Lck and c-Src models based on PDB code 3LVP , there were no clashes between the two kinase domains , and there were salt bridges at the interface that could plausibly stabilize it . The model for the Lck autophosphorylation complex is illustrated in Figure 9A , with one view showing the enzyme-kinase in a near-standard orientation and the other with the substrate-kinase in a similar orientation . In this model , as is also seen for peptide recognition by kinases , there is a short two-stranded antiparallel β sheet formed by the activation loop of the enzyme-kinase and the substrate . An obvious feature of the Lck and c-Src autophosphorylation models is that the FG loop of the enzyme-kinase is at the heart of the docking site of the substrate kinase ( Figure 9A , top panels ) . The FG loop docks on a hydrophobic patch on the substrate-kinase ( Figure 9A , bottom panels ) . This hydrophobic patch is formed between the G and EF helices of the substrate kinase , a known docking site on kinases ( Depetris et al . , 2005 ) . This close packing is possible because the FG loop of the Src-family kinases , like that of IGF1-R , has small hydrophobic residues facing outwards . 10 . 7554/eLife . 20105 . 030Figure 9 . A structural model for activation loop phosphorylation of Lck . ( A ) Schematic cartoons and structural renderings of a model for Lck activation loop trans-autophosphorylation . This model is based on a crystal structure of IGF1-R ( PDB code 3LVP ) ( Nemecek et al . , 2010 ) , in which one kinase domain molecule in the asymmetric unit , deemed the 'substrate-kinase' , presents its activation loop tyrosine into the active site of a second kinase domain molecule in the asymmetric unit , deemed the 'enzyme-kinase' . ( B ) Western blot analysis of in vitro activation loop autophosphorylation reactions with the wild-type ZAP-70 kinase domain and FG loop mutants bearing the K542A substitution alone or simultaneous K541A , K542A , and K544A substitutions ( the 'KKMK' to 'AAMA' mutant ) . ( C ) Western blot analysis of in vitro activation loop autophosphorylation reactions with the wild-type ZAP-70 kinase domain and mutants in which Arg 514 was substituted with alanine or glutamate . Kinase domains were used at a concentration of 5 μM in the reactions shown in panels B and C . DOI: http://dx . doi . org/10 . 7554/eLife . 20105 . 03010 . 7554/eLife . 20105 . 031Figure 9—figure supplement 1 . Activation loop autophosphorylation kinetics for the wild-type Lck kinase domain ( residues 229–509 ) and a mutant with the E448Q substitution . DOI: http://dx . doi . org/10 . 7554/eLife . 20105 . 03110 . 7554/eLife . 20105 . 032Figure 9—figure supplement 2 . Activation loop autophosphorylation kinetics for the Lck and c-Src kinase domains containing substitutions in the FG loop . The kinase domains were used at a concentration of 100 nM . DOI: http://dx . doi . org/10 . 7554/eLife . 20105 . 032 To evaluate if the proposed interface is energetically stable for Src-family kinases , we initiated three molecular dynamics trajectories from this model for Lck . The simulations ranged in length from 30 ns to 120 ns , and the dimer interface remained intact in all of the trajectories . We mutated Glu 448 in Lck , which is involved in a potential intermolecular salt bridge , to glutamine , and observed a concentration-dependent loss of autophosphorylation activity in vitro ( Figure 9—figure supplement 1 ) . Thus , we consider the IGF1-R structure to provide a reasonable model for the enzyme-substrate docking complex that facilitates Lck activation loop autophosphorylation . The FG loop of ZAP-70 bears four lysine residues that are important for determining selectivity for residues upstream of the tyrosine residue in substrates ( Figure 7 ) . This has an obvious consequence for trans-autophosphorylation of the activation loop . The highly charged FG loop in the enzyme-kinase is incompatible with docking on the hydrophobic patch in the substrate-enzyme in the models for the autophosphorylation complex . Additionally , in the models for the autophosphorylation complexes of Lck and c-Src , close contact between the enzyme-kinase and the substrate-kinase is enabled by the presence of a glycine residue in the FG loop ( Gly 443 in Lck ) of the enzyme-kinase ( Figure 9A ) . Gly 443 in Lck is conserved in many tyrosine kinases , including all other Src-family kinases as well as IGF-1R , which our model is based on . Gly 443 in Lck is replaced by Lys 542 in ZAP-70 , one of the positively-charged residues in the FG loop , and Lys 542 cannot be accommodated within the autophosphorylation complex . Two other glycine residues are also important in the substrate-kinase molecule . Gly 415 in the substrate-Lck packs close to the FG loop of the enzyme-kinase ( Figure 9A ) , but this residue is replaced by Arg 514 in ZAP-70 , a residue that is important for determining the specificity for negatively-charged residues downstream of the tyrosine in substrate peptides ( Figure 7 ) . If ZAP-70 were to take the position of a substrate-kinase , then Arg 514 would be incompatible with the autophosphorylation complex , as it would come into close contact with Lys 504 in the FG loop of the ZAP-70 molcule playing the role of the enzyme . Gly 399 in the substrate-Lck is replaced by alanine in ZAP-70 , which the model predicts would also lead to clashes . One of the striking results of the specificity screens with Lck is the absence of strong selection upstream of the tyrosine residue in peptide substrates . This weak upstream selectivity in Lck is correlated with the presence of small , nonpolar residues in the FG loop in Lck and other Src-family kinases , which our modeling predicts is necessary for the formation of the autophosphorylation complex . Thus , the Src-family kinases appear to have traded upstream selectivity in substrates for the ability to undergo efficient autophosphorylation on the activation loop . In contrast , the ability of ZAP-70 to impose strong electrostatic selectivity upstream of the tyrosine residue appears to have negated its ability to undergo efficient autophosphorylation . We tested whether mutations at positions that clearly impact substrate specificity in ZAP-70 would also affect autophosphorylation rates . The KKMK-to-AAMA mutation in the FG loop , which reduced LAT and SLP-76 phosphorylation rates ( Figure 7—figure supplement 7 ) , caused a clear increase in activation loop autophosphorylation of the ZAP-70 kinase domain ( Figure 9B ) . Consistent with this observation , introduction of lysine residues at corresponding positions into the Lck and c-Src kinase domains reduced their autophosphorylation rates ( Figure 9—figure supplement 2 ) . We observed a similar enhancement of ZAP-70 activation loop autophosphorylation when Arg 514 was replaced with alanine or glutamate ( Figure 9C ) . In order to connect our in vitro data to a cellular context , we studied the activities of full-length Lck and ZAP-70 in a reconstituted cell-based system in which human embryonic kidney ( HEK ) 293 cells were transiently transfected with various constructs of Lck or ZAP-70 , together with the full-length substrates LAT or TCRζ , as described previously ( Brdicka et al . , 2005 ) . In this system , Lck shows substantial tyrosine phosphorylation , whereas ZAP-70 is phosphorylated only when it is co-expressed with Lck ( Figure 10A ) . LAT phosphorylation is only seen when Lck is co-expressed with ZAP-70 . Likewise , robust TCRζ phosphorylation is only observed when Lck is present . 10 . 7554/eLife . 20105 . 033Figure 10 . Lck , ZAP-70 , and Syk activation and specificity in a model cell line . ( A ) Co-expression of wild-type full-length kinases and chimeras of Lck and ZAP-70 with LAT or TCRζ . Two chimeric kinases were used: Lck/ZAP-70 refers to a construct with the Lck SH3-SH2 module fused to the ZAP-70 kinase domain . ZAP-70/Lck refers to the ZAP-70 tandem SH2 module fused to the Lck kinase domain . ( B ) Comparison of F-G loop mutations of ZAP-70 and Lck . ZAP-70* refers to the KKMK-to-AAMA mutant , and Lck* refers to the PGMT-to-KKMK mutant . ( C ) Comparison of F-G loop mutations in ZAP-70 and Syk . All experiments were carried out in human embryonic kidney ( HEK ) 293T cells as previously described ( Brdicka et al . , 2005 ) , and interpretation of the results is also based on the analysis in Brdicka et al . 2005 . DOI: http://dx . doi . org/10 . 7554/eLife . 20105 . 03310 . 7554/eLife . 20105 . 034Figure 10—figure supplement 1 . Western blot analysis ZAP-70 and Syk phosphorylation in HEK293 cells with site-specific antibodies . The top blot shows phosphorylation of a tyrosine in the SH2-kinase linker of ZAP-70 or Syk ( Tyr 319 or Tyr 352 , respectively ) . The bottom blot shows phosphorylation of a tyrosine in the kinase domain activation loop of ZAP-70 or Syk ( Tyr 493 or Tyr 526 , respectively ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20105 . 034 We also tested the importance of the adapter domains of Lck and ZAP-70 in determining specificity for LAT and the TCRζ . To do this , we made two chimeric constructs . One ( Lck/ZAP-70 ) has the SH3 and SH2 domains of Lck fused to the kinase domain of ZAP-70 . The other ( ZAP-70/Lck ) has the tandem SH2 domains of ZAP-70 fused to the kinase domain of Lck . The behavior of these constructs is similar to that of the wild-type protein from which the kinase domain is derived , demonstrating that the principal determinant of phosphorylation specificity is the kinase domain ( Figure 10A ) . We also analyzed the ZAP-70 paralog , Syk , as this kinase has previously been shown to have intermediate substrate specificity and auto-activation capability between ZAP-70 and Lck ( Mukherjee et al . , 2013; Tsang et al . , 2008 ) . In these experiments , Syk was phosphorylated in the absence of Lck , unlike ZAP-70 , and it phosphorylated both LAT and TCRζ , unlike ZAP-70 and Lck . We assessed the impact of mutations in the FG loop on the activities of Lck and ZAP-70 in HEK293 cells ( Figure 10B ) . Simultaneous mutation of Lys 541 , Lys 542 , and Lys 544 in ZAP-70 to alanine ( KKMK-to-AAMA mutant , denoted ZAP-70* ) resulted in a reduction of LAT phosphorylation and an increase in TCRζ phosphorylation , even in the absence of Lck , suggesting that this mutant ZAP-70 can auto-activate . Conversely , mutation of the FG loop in Lck ( PGMT-to-KKMK mutant , denoted Lck* ) resulted in a loss of both Lck and TCRζ phosphorylation . No gain in LAT phosphorylation was observed for this protein . The sequence of the FG loop in Syk resembles that of ZAP-70 but contains one glycine residue that is important for autophosphorylation in Lck ( Gly 443 in Lck , Gly 575 in Syk , and Lys 542 in ZAP-70 ) . Mutation of the FG loop sequence of Syk to that of ZAP-70 ( RGMK-to-KKMK ) decreased Syk activation loop autophosphorylation and also decreased the ability of Syk to phosphorylate TCRζ and LAT ( Figure 10C and Figure 10—figure supplement 1 ) . These mutations also made Syk more dependent on activation by Lck . Mutation of the FG loop sequence of ZAP-70 to that of Syk ( KKMK-to-RGMK ) did not result in robust autophosphorylation , presumably because Arg 514 in ZAP-70 , which is a tyrosine in Syk , inhibits formation of the enzyme-substrate complex ( Figure 9C ) . Similar to Syk , however , this ZAP-70 mutant could phosphorylate both LAT and TCRζ . These cell-based data are consistent with the ability of Syk to facilitate T cell receptor signaling in the absence of Lck ( Chu et al . , 1996; Williams et al . , 1998 ) . We asked if the specificity-defining sequence features of ZAP-70 , LAT , and SLP-76 are unique among tyrosine kinases and their substrates . To address this question , we analyzed a curated list of experimentally-determined substrates for human tyrosine kinases from the PhosphoSitePlus database ( Hornbeck et al . , 2015 ) . This dataset comprises roughly 1100 phosphosites for approximately 100 tyrosine kinases . We binned these sequences based on the net charge of the sequence surrounding the tyrosine and the identity of the −1 position with respect to the phosphosite , two key determinants of ZAP-70 specificity . On average , tyrosine kinase substrates have a near-neutral net charge in proximity to the phospho-acceptor tyrosine , and the residues at the −1 position segregate into three groups: bulky-hydrophobic , polar-neutral , and negatively-charged ( Figure 11A ) . Lck substrates , such as the T cell receptor ITAM phosphosites , fall into each of these major categories and are thus representative of typical tyrosine kinase substrates . LAT and SLP-76 phosphosites , however , are distinctive in having both a −1 acidic residue and a substantially negative net charge . Both of these features are conserved in LAT sequences across different vertebrates ( Figure 11B ) . 10 . 7554/eLife . 20105 . 035Figure 11 . Unique sequence features of ZAP-70 and its substrates . ( A ) Analysis of net charge and −1 residue distributions for 1100 tyrosine kinase substrates from the PhosphoSite Plus database ( Hornbeck et al . , 2015 ) . Analyzed sequences span seven residues on either side of the phospho-acceptor tyrosine . Data are represented as a contour plot , which depicts the prevalence of phosphosites in the database with particular combinations of −1 residue identity and net charge . Darker regions of the contour plot indicate that more sequences have that particular combination of properties . ( B ) Sequence logos showing the conservation around five LAT tyrosines in 59 species from fish to mammals . ( C ) Conservation of lysine and arginine residues important for ZAP-70 substrate recognition . The top panel displays a sequence logo showing a lack of conservation at many of these positions in human non-receptor tyrosine kinases ( NRTKs ) . The bottom panel displays high conservation at these positions between ZAP-70 orthologs from fish to mammals . ( D ) Distribution of net charge for the kinase domains of all 32 human non-receptor tyrosine kinases . DOI: http://dx . doi . org/10 . 7554/eLife . 20105 . 035 ZAP-70 accommodates the unique features of LAT and SLP-76 by using a positively-charged region on its kinase domain to engage substrates . We assessed sequence conservation in this region and found that all of the lysine and arginine residues are conserved in ZAP-70 orthologs from fish to mammals . Most of these residues are not conserved in other human non-receptor tyrosine kinases ( Figure 11C ) . The predicted isoelectric point of the ZAP-70 kinase domain is ~9 , and it has a net charge of +7 . There are three other human non-receptor tyrosine kinase domains ( Fer , Tnk1 , and Ctk ) that have similar net charge ( Figure 11D ) , but the positively-charged residues in these kinases do not cluster near the substrate binding region . Thus , the degree and placement of positive charge on its kinase domain makes ZAP-70 an outlier among tyrosine kinases . Our results provide a molecular explanation for the order of initial phosphorylation events during T cell receptor signaling . This sequence of events is largely dictated by the substrate specificities of the ZAP-70 and Lck kinase domains , which we have defined using a high-throughput assay to measure phosphorylation levels for hundreds of mutant peptides based on natural substrates of these kinases . Unlike Lck , and probably most other tyrosine kinases , ZAP-70 preferentially phosphorylates substrates that are enriched in negatively-charged residues , depleted of positively-charged residues , and have an aspartate at the −1 position relative to the phospho-acceptor tyrosine . These features are characteristic of the targets of ZAP-70 in LAT and SLP-76 . In accommodating LAT and SLP-76 , ZAP-70 not only became a highly specialized kinase but also lost the ability to activate itself by trans-autophosphorylation . This coupling between substrate recognition and autophosphorylation in ZAP-70 , along with the electrostatic features of LAT and SLP-76 , insulate these proteins from undesirable cross-talk with other signaling molecules ( Figure 12 ) . 10 . 7554/eLife . 20105 . 036Figure 12 . Schematic depiction of the biochemical insulation of ZAP-70 , LAT , and SLP-76 . Many tyrosine kinases , such as Lck , c-Src , and Itk , can activate themselves through trans-autophosphorylation of their activation loops ( bottom panel , left side ) . These kinases can efficiently phosphorylate typical tyrosine kinase substrates , which have a near-neutral net charge ( bottom panel , right side ) . ZAP-70 activation requires phosphorylation of its activation loop by Lck ( top left panel ) . Once activated , ZAP-70 exclusively phosphorylates LAT and SLP-76 , but not typical tyrosine kinase substrates , and LAT and SLP-76 cannot be phosphorylated by kinases like Lck , c-Src , and Itk ( top right panel ) . Thus , ZAP-70 and its substrates are insulated from other proteins in the T cell receptor signaling pathway . DOI: http://dx . doi . org/10 . 7554/eLife . 20105 . 036 A critical aspect of this electrostatic selection mechanism is that it provides stringent control over kinase-substrate interactions without restricting the diversity of downstream signals . The evolution of ZAP-70 substrates has led to the incorporation of substantial negative charge both upstream and downstream of their phospho-acceptor tyrosines . The resulting electrostatic selection is sufficiently discriminatory to ensure exclusive phosphorylation of these sites in LAT and SLP-76 by ZAP-70 , while allowing each of these phosphosites to have unique functional properties . Not only can each phosphosite on LAT and SLP-76 bind to distinct SH2 domain-containing proteins ( Houtman et al . , 2004 ) , but the rate of phosphorylation at these sites can also be tuned without compromising kinase specificity , as described earlier for LAT Tyr 132 . What evolutionary forces might have shaped the origin of the electrostatic features in LAT and SLP-76 ? They probably evolved not only through positive selection for efficient phosphorylation by ZAP-70 , but also through negative selection against phosphorylation by other tyrosine kinases , including Lck . This type of system-wide negative selection has been noted previously for SH3 domain-ligand interactions in yeast , which are intrinsically promiscuous , but can achieve a remarkable degree of specificity in the cellular context ( Zarrinpar et al . , 2003 ) . In the case of T cell receptor-proximal kinase signaling , negative selection was presumably crucial for establishing an insulated network in which Src-family kinases cannot phosphorylate LAT . Several studies have shown that there is an active pool of Lck in the T cell , even prior to receptor stimulation ( Nika et al . , 2010; Schoenborn et al . , 2011 ) , and the amount of active Lck establishes a threshold for responses to peptide ligands ( Manz et al . , 2015 ) . Both Lck and LAT are membrane-bound , and they may cluster in the same membrane micro-domains ( Zhang et al . , 1998 ) . Given this , if Lck could readily phosphorylate LAT , aberrant downstream signaling would result without T cell receptor stimulation by an appropriate antigen . The tight regulation of ZAP-70 and its narrow substrate specificity are also important for faithful T cell receptor signaling , as they favor a T cell response to peptide antigens bound to MHC molecules . The MHC proteins on antigen-presenting cells are recognized by the CD4 and CD8 co-receptors , and Lck is tethered to these co-receptors at the membrane ( Figure 1A ) . Upon peptide-MHC interaction with the T cell receptor and its co-receptors , Lck is repositioned near its favored substrates , the ITAMs within the cytoplasmic tails of the T cell receptor complex . This ensures not only the appropriate Lck-mediated phosphorylation of ITAMs but also the proper recruitment , phosphorylation , and activation of ZAP-70 by Lck . The strict dependence on Lck for ZAP-70 membrane recruitment and activation dictates that receptor triggering only occurs when peptide antigens are presented by MHC molecules . B cells have an analogous pathway to the T cell receptor pathway that utilizes the Src-family kinase Lyn , the ZAP-70 paralog Syk , and the scaffold protein BLNK ( or SLP-65 ) . Unlike T cells , however , receptor activation in B cells does not require that antigens are presented on MHC molecules . Furthermore , B cell activation can be considered more flexible , because Syk can slowly auto-activate ( Tsang et al . , 2008 ) and it can phosphorylate ITAMs ( Chu et al . , 1996; Mukherjee et al . , 2013 ) , thereby making the Src-family kinase somewhat expendable ( Takata et al . , 1994 ) . Src-family kinase participation in B cell signaling serves to speed up and increase the sensitivity of the response to B cell receptor stimulation ( Mukherjee et al . , 2013 ) . These subtle differences between T and B cells underscore how signaling pathways and their corresponding molecules have evolved to achieve varying degrees of control . The techniques and findings we have presented here provide a framework to characterize the evolution and specialization of T and B cell kinases and to enhance our understanding of the design principles underlying lymphocyte signal transduction pathways .
The cell-based assays for kinase activation and specificity were carried out as described previously ( Brdicka et al . , 2005; Yan et al . , 2013 ) . HEK 293T cells were transiently co-transfected using Lipofectamine and Plus reagents ( Invitrogen ) with expression constructs for full-length LAT or the T cell receptor ζ chain along with different expression constructs for the three kinases , ZAP-70 , Lck , or Syk , as indicated in Figure 10 . Cells were lysed by resuspension in an SDS-PAGE gel loading dye , and cellular debris was removed by ultracentrifugation . The supernatants from the cell lysates were analyzed by western blotting . Phosphorylation of the protein constructs was analyzed with the Millipore 4G10 pan-phosphotyrosine antibody . Expression levels were monitored using specific antibodies for each protein . | A class of enzymes known as tyrosine kinases relay signals in cells by adding phosphate groups onto specific sites ( called 'tyrosine residues' ) in other proteins . Most tyrosine kinases can phosphorylate many targets ( or 'substrates' ) ; they can also phosphorylate and thereby activate themselves , when given the right signal . Many tyrosine kinases select their substrates on the basis of their location; once recruited to and activated at a specific site , these enzymes will typically phosphorylate many nearby proteins . A tyrosine kinase called ZAP-70 is found in immune cells known as T cells . ZAP-70 works together with another kinase called Lck to activate T cells , which enables the cells to mount an immune response when they encounter foreign molecules . This pathway is precisely controlled , with Lck activated first , followed by ZAP-70 . Unlike most other tyrosine kinases , ZAP-70 cannot activate itself , and it will only phosphorylate a narrow range of substrates . The origin of these constraints are not understood , but they are thought to be crucial for ensuring that T cells readily respond to foreign molecules but not to healthy cells . Shah et al . developed a high-throughput technique to investigate which features ZAP-70 and Lck use to select their substrates . First , hundreds of different sequences based on natural substrates were genetically encoded and introduced into bacterial cells , with one type per bacterium . The bacteria displayed these sequence variants on their surface , and Shah et al . then treated the bacteria with either ZAP-70 or Lck . Cell sorting was used to isolate those bacterial cells with variants that were phosphorylated , and high-throughput DNA sequencing was used to identify the phosphorylated sequences . This approach revealed that ZAP-70 was deterred from phosphorylating sites that carry a positive charge and strongly preferred sites that are negatively-charged , such as those found in its two major substrates . Shah et al . also showed that Lck , which behaves like a typical tyrosine kinase , could not phosphorylate the substrates of ZAP-70 because of their substantial negative charge . This lack of cross-reactivity between Lck and the ZAP-70 substrates prevents premature signaling in T cells . Using simulations , Shah et al . went on to show that a positively-charged region on ZAP-70 ( which is more prominent than in other tyrosine kinases ) helps ZAP-70 interact with negatively-charged substrates . This region also deters the kinase from activating itself , making it dependent instead upon Lck for activation . Together , these results identify the distinctive features of ZAP-70 that are important for ensuring that T cells are activated only when they sense foreign molecules on unhealthy cells . The work will lead to future studies exploring the tightly controlled signaling events carried out by tyrosine kinases in T cells in more detail . | [
"Abstract",
"Introduction",
"Results",
"and",
"discussion",
"Materials",
"and",
"methods"
] | [
"structural",
"biology",
"and",
"molecular",
"biophysics",
"immunology",
"and",
"inflammation"
] | 2016 | An electrostatic selection mechanism controls sequential kinase signaling downstream of the T cell receptor |
Bicuspid aortic valve ( BAV ) is the most common congenital cardiovascular disease in general population and is frequently associated with the development of thoracic aortic aneurysm ( TAA ) . There is no effective strategy to intervene with TAA progression due to an incomplete understanding of the pathogenesis . Insufficiency of NOTCH1 expression is highly related to BAV-TAA , but the underlying mechanism remains to be clarified . A comparative proteomics analysis was used to explore the biological differences between non-diseased and BAV-TAA aortic tissues . A microfluidics-based aorta smooth muscle-on-a-chip model was constructed to evaluate the effect of NOTCH1 deficiency on contractile phenotype and mitochondrial dynamics of human aortic smooth muscle cells ( HAoSMCs ) . Protein analyses of human aortic tissues showed the insufficient expression of NOTCH1 and impaired mitochondrial dynamics in BAV-TAA . HAoSMCs with NOTCH1-knockdown exhibited reduced contractile phenotype and were accompanied by attenuated mitochondrial fusion . Furthermore , we identified that mitochondrial fusion activators ( leflunomide and teriflunomide ) or mitochondrial fission inhibitor ( Mdivi-1 ) partially rescued the disorders of mitochondrial dynamics in HAoSMCs derived from BAV-TAA patients . The aorta smooth muscle-on-a-chip model simulates the human pathophysiological parameters of aorta biomechanics and provides a platform for molecular mechanism studies of aortic disease and related drug screening . This aorta smooth muscle-on-a-chip model and human tissue proteomic analysis revealed that impaired mitochondrial dynamics could be a potential therapeutic target for BAV-TAA . National Key R and D Program of China , National Natural Science Foundation of China , Shanghai Municipal Science and Technology Major Project , Shanghai Science and Technology Commission , and Shanghai Municipal Education Commission .
Bicuspid aortic valve ( BAV ) disease is the most common congenital cardiovascular abnormality and is found in nearly 1 . 4% of the general population ( Garg et al . , 2005; Michelena et al . , 2011; Verma and Siu , 2014 ) . BAV arises from incomplete separation or fusion of the aortic valve cusps and is associated with an approximately 40% risk of developing thoracic aortic aneurysm ( TAA ) , namely , bicuspid aortopathy ( Verma and Siu , 2014 ) . BAV-TAA poses a severe health threat to a large population because progressive aneurysmal dilation can potentially develop into lethal dissection or rupture ( Goldfinger et al . , 2014; Olsson et al . , 2006 ) . The current clinical management mainly relies on prophylactic surgical repair of the notably dilated aorta ( Coady et al . , 2010 ) . At present , the understanding of pathophysiological mechanisms of BAV-TAA is incomplete , which leads to the absence of effective pharmaceutical therapy to alleviate aortopathy progression ( Lindeman and Matsumura , 2019 ) . Multiple factors , such as genetics and hemodynamics , are involved in the etiologies of BAV-TAA . In particular , genetic factors are considered to play a pivotal role in the disease progression ( Isselbacher et al . , 2016; Prakash et al . , 2014 ) . NOTCH1 insufficiency has been observed in the population with BAV ( Balistreri et al . , 2018; Harrison et al . , 2019; Malashicheva et al . , 2020; Sciacca et al . , 2013 ) . However , the underlying mechanism through which insufficient NOTCH1 induces aortopathy remains to be explored . Mitochondrial dysfunction has been closely linked to a variety of cardiovascular disorders , such as heart failure and atherosclerosis . Recent studies found that mitochondrial dysfunction was also related to the development of arterial aneurysm formation ( Cooper et al . , 2021; van der Pluijm et al . , 2018; Oller et al . , 2021 ) . A single-cell transcriptome analysis on aneurysmal human aortic tissue suggested that mitochondrial dysfunction and increased chromatin oxidative phosphorylation ( OXPHOS ) were found in TAA tissues and insufficient ATP production might not be sufficient for the contractile activities of human aortic smooth muscle cells ( HAoSMCs ) ( Li et al . , 2020 ) . Particularly , mitochondrial fission and fusion are dynamically balanced to maintain mitochondrial homeostasis and functions; and a shift toward fission event is one of the main causes of mitochondrial dysfunction . In a mice model of abdominal aortic aneurysm ( AAA ) , impaired mitochondrial dynamics was found to play salient roles in disease development , and could be attenuated by the mitochondrial fission inhibitor Mdivi1 ( Cooper et al . , 2021 ) . However , these studies focused on the analysis of abdominal aortic aneurysms and genetic TAA with FBN1 or Fubulin-4 mutation . It has been reported that there was close interaction between NOTCH1 signaling and homeostatic mitochondrial dynamics in the differentiation of cardiomyocytes ( Kasahara et al . , 2013 ) and the survival of breast cancer cells ( Chen et al . , 2018 ) . Therefore , the relationship between NOTCH1 signaling pathway and mitochondrial dynamics in BAV-TAA needs to be clarified . The traditional TAA animal models are frequently applied for the pathogenesis research and pharmaceutical therapy , however , not suitable for studying BAV-TAA . Although Koenig et al . generated NOTCH1-haploinsufficient mice in a preliminary 129S6 background that exhibited aortic root dilation , these mice did not show BAV characteristics ( Koenig et al . , 2017 ) . Therefore , these models may not provide sufficient information of pathogenesis and drug response of BAV-TAA . Owing to the bioengineering advance , microfluidic-based organ-on-chip models have been widely developed to replicates human tissue microenvironment for toxicity analysis , drug screening and disease modeling and thus promotes pharmaceutical translation from preclinical studies to clinical trials ( Zhang et al . , 2018; Ingber , 2016; Park et al . , 2019; Thacker et al . , 2020; Hofemeier et al . , 2021 ) . It provides an opportunity to use a novel platform to study BAV-TAA on a susceptible human genetic background and may fill the gap between animal and human medicine . Here , we engineered an in vitro aorta smooth muscle-on-a-chip model of primary HAoSMCs that emulates the biomechanics of the human aortic wall . We characterized the association between mitochondrial dynamics and NOTCH1 deficiency in BAV-TAA on this platform . Our study provides the first demonstration previously undocumented role of impairment of mitochondrial fusion in bicuspid aortopathy , which may serve as a potential pharmacological target for preventing disease progression .
The structure of the three-layer microfluidic aorta smooth muscle-on-a-chip model was designed using computer-aided design ( CAD ) software ( Autodesk Inc ) . The size of the three layers were 100 mm × 40 mm × 6 mm . The top and bottom layers had microchannels with dimensions of 70 mm × 6 mm × 4 mm , and the middle layer contained a microchannel with dimensions of 70 mm × 6 mm × 6 mm . Molds of the three layers were custom-made using a high-precision computer numerical control ( CNC ) engraving machine ( Jingyan Technology ) . The frame of the molds and the microchannels were carved out of polymethyl methacrylate ( PMMA ) plates , which were then glued on another PMMA plate . Polydimethylsiloxane ( PDMS , Sylgard 184 , Dow Corning ) was polymerized in defined casts at a weight ratio of base to curing agent of 10:1 . The mixed PDMS was poured into the molds and underwent cross-linking at 70°C for 2 hr . Commercialized PDMS membranes were purchased from Hangzhou Bald Advanced Materials . Detailed parameters of the commercialized PDMS membrane were provided as follows: thickness of 200 ± 2 μm , shore A hardness of 50 , Yang's elastic modulus of 1 . 7 MPa , tensile strength of 4 MPa , tear strength of 7 KN/m and light transmittance of 93% . The Young's modulus values were obtained by the following experiments . The tensile stress–strain responses was measured using a tensile testing machine ( Instron ) . Prior to the measurement , the PDMS membrane was cut into a piece of 3 cm length and 3 mm width membrane . The membrane was fixed to the testing machine with a fixture . The sample was automatically stretched in a gradient within the proportional limit . The Young's modulus of the PDMS membrane was calculated by the slope value of the tensile stress–strain curve . Subsequently , the three PDMS layers were peeled off the molds . The bottom layer of the PDMS slab was bonded with one PDMS membrane after oxygen plasma treatment ( Harrick Plasma ) , and the top PDMS layer was bonded to another PDMS membrane in a similar manner . The middle PDMS layer was then sandwiched between the top and bottom membranes; this step was performed under a microscope to guarantee that the upper and lower microchannels fully overlapped with the middle microchannel . The cells were stretched by applying different percentages of rhythmic strain to the PDMS membranes for 24 hr . The vacuum pump was connected to a water-oil separator , which dried the gas to protect the downstream vacuum regulator and solenoid valve . The inlet of the vacuum pumps was then connected to the computer-controlled solenoid system by applying rhythmic stretching at a frequency of 1 Hz and then connected to the gas channel of the aorta smooth muscle-on-a-chip model . The regulator was used to control the vacuum magnitude . The solenoid valve was a voltage-dependent on/off valve used to control the stretching frequency of the PDMS membrane . When the supply voltage was greater than 24 V , the gas in the channel was pumped out , which stretched the PDMS membrane . Otherwise , the gas channel was connected to the atmosphere , and the membrane deformation recovered . The on/off frequency of the solenoid valve was controlled by a microcontroller unit ( MCU ) . Thus , the stretching frequency could be controlled by changing the preset program of the MCU , and the stretching amplitude could be controlled by adjusting the vacuum regulator manually . We used two pressure ranges , 10 kPa ( 7 . 18 ± 0 . 44% , low strain ) and 15 kPa ( 17 . 28 ± 0 . 91% , high strain ) , throughout the experiments . As the control , cells were cultured in aorta smooth muscle-on-a-chip models under static conditions . After 24 hr of rhythmic strain , samples were collected for immunofluorescence , RT-qPCR , western blotting and mitochondrial membrane dynamics analyses . Written informed consent was obtained from all patients before participation . Human aortic specimens were utilized under approvals of Zhongshan Hospital , Fudan University Ethics Committee ( NO . B2020-158 ) in accordance with the Declaration of Helsinki . Human aortic samples were collected from patients who underwent ascending aorta surgery at Zhongshan Hospital , Fudan University . Echocardiography was used to characterize aortic valve morphology and ascending aortic diameter prior to surgery . The tissue samples were immediately frozen in liquid nitrogen and stored at −80°C . Six aortic tissues were obtained from patients with a tricuspid aortic valve but without aortic dilation ( non-diseased; mean age: 62 . 2 years; range: 51–74 years; four males ) , and another six samples were obtained from patients with BAV-related thoracic aortic aneurysm ( BAV-TAA; mean age: 59 . 3 years; range: 43–72 years; four males ) . The patients’ basic information is available in Supplementary file 1a . Primary human aortic smooth muscle cells ( p-HAoSMCs ) were isolated from non-diseased ascending aortic tissues and BAV-TAA aortic tissues ( n = 3 ) . The ascending aortic tissues were washed with phosphate-buffered saline ( PBS , Thermo Fisher Scientific ) . The intima and adventitia layers of the tissues were removed , and the media layer was preserved for the harvesting of p-HASMCs . Subsequently , the media layer was cut into small pieces ( 2–3 mm in length ) and cultured in high-glucose Dulbecco’s modified Eagle’s medium ( DMEM , Gibco ) with 20% fetal bovine serum ( FBS , Gibco ) and 1% penicillin and streptomycin ( p/s , Gibco ) for 2–3 weeks at 37°C and 5% CO2 in a humidified incubator . After approximately 10–12 days , the p-HASMCs started to migrate out of the tissue pieces . When the cells reached approximately 80% confluency , first-passaged cells were rinsed with PBS , digested using 0 . 25% trypsin ( Gibco ) , and replated in smooth muscle cell culture medium ( SMCM , ScienCell ) . The cells were characterized through an immunofluorescence analysis of four different specific markers of smooth muscle cells ( CNN1 , SM22 , MYH11 and α-SMA ) . We used p-HASMCs at a low passage ( P2-P5 ) in all the experiments . In addition to the p-HAoSMCs isolated from aortic tissues , a human aortic smooth muscle cell line ( CRL1999 ) and commercialized p-HAoSMCs ( PCS-100–012 ) were purchased from ATCC ( American Type Culture Collection ) in accordance with their ethical regulations and compliances . p-HAoSMCs and the CRL1999 cell line were cultured in SMCM . Prior to cell seeding , the surface of the cell culture channel was coated with mouse collagen at a concentration of 80 µg/mL ( Sigma ) by incubating for 1 hr at room temperature and drying for 2 hr at 70°C . Afterward , the cell culture channel was washed with PBS , and cells were seeded on the PDMS membranes in a cell culture channel at a density of 2 x 106 cells/mL . The cells were cultured in Dulbecco’s modified Eagle’s medium/nutrient mixture F-12 ( DMEM/F-12 , Thermo Fisher Scientific ) supplemented with 10% FBS in a cell culture channel . After seeding , aorta smooth muscle-on-a-chip models were incubated at 37°C and 5% CO2 in a humidified incubator for 24 hr for cell attachment . The aorta smooth muscle-on-a-chip models were then ready for mechanical stimulation experiments . HAoSMCs cell line ( CRL1999 , Lot number 70019189 , Homo sapiens ) and primary HAoSMCs ( PCS-100–012 , Lot number 80323179 , Homo sapiens ) were purchased from ATCC . The identity has been authenticated by STR analysis and mycoplasma contamination was conformed by sterility test and pathogenic virus test provided by ATCC . The cells have human unique DNA profiles and were negative for mycoplasma contamination . Mdivi-1 ( Sigma ) , an inhibitor of mitochondrial fission , was dissolved in dimethylsulfoxide ( DMSO ) and stored at −20°C before use , and fresh medium was used to obtain a final concentration of 30 μM . Leflunomide and teriflunomide ( Sigma ) , two different activators of mitochondrial fusion , were dissolved in DMSO at appropriate concentrations . Prior to cell treatment , fresh medium was used to obtain a concentration of 75 μM for both drugs . After the cells were fully attached to the PDMS membrane in the cell culture channel , medium containing a mitochondrial fusion activator ( leflunomide and teriflunomide ) or mitochondrial fission inhibitor ( Mdivi-1 ) was added to the aorta smooth muscle-on-a-chip models . After 24 hr of rhythmic strain , the samples were collected for comparative experiments . Immunofluorescence analysis was performed in the microfluidic aorta smooth muscle-on-a-chip model in situ after 24 hr of rhythmic stretch . The medium was aspirated from the cell culture channel , and the cells were washed with PBS , immediately fixed with 4% paraformaldehyde ( Beyotime ) for 30 min at room temperature and permeabilized with 1% ( v/v ) Triton X-100 ( Beyotime ) for 15 min . Afterward , blocking solution with 5% bovine serum albumin ( Sigma ) was applied to the cells to block nonspecific binding sites for 30 min at room temperature , and the cells were incubated overnight at 4°C with primary antibodies . The primary antibodies used in this study and their working concentrations are listed in Supplementary file 1b . After incubation , the cells were washed three times with PBS and incubated with Alexa 594 anti-rabbit secondary antibody ( Thermo Fisher Scientific ) at a dilution of 1:300 for 1 hr at room temperature under dark conditions . The nuclei were counterstained with 4′ , 6-diamidino-2-phenyllindole ( DAPI ) ( Thermo Fisher Scientific ) for 10 min . The aorta smooth muscle-on-a-chip models were then disassembled , and images were acquired with a fluorescence microscope ( Leica DMi8 ) and analyzed using ImageJ software . Cells and aortic tissue samples were lysed using RIPA ( Beyotime ) lysis buffer supplemented with protease inhibitor phenyl methyl sulfonyl fluoride ( PMSF , Beyotime ) . To collect an appropriate concentration of protein for western blotting , we collected and pooled cell protein from three microfluidic aorta smooth muscle-on-a-chip model . For the tissue samples , the intima and adventitia were peeled out , and the middle layers were ground into small pieces . The extracts were incubated for 30 min on ice for complete lysis and centrifuged at 14 , 000 rpm and 4°C for 25 min . The supernatant was collected after centrifugation , and the debris was discarded . The protein concentrations were quantified using a BCA Protein Assay kit ( Thermo Fisher Scientific ) . The extracted proteins were diluted in sample loading buffer and heated for 5 min at 95°C . Ten micrograms of each protein sample were then separated by running on a 10% SDS-PAGE gel and subsequently transferred to 0 . 2 µm polyvinylidene fluoride ( PVDF ) membranes ( Millipore ) . The PVDF membranes were blocked with 5% skimmed milk ( Beyotime ) for 1 hr at room temperature and incubated with different primary antibodies ( Supplementary file 1b ) overnight at 4°C . The membranes were then incubated with horseradish peroxidase-conjugated goat anti-rabbit and goat anti-mouse IgG secondary antibodies ( Cell Signaling Technology ) at 1:6000 dilution for 1 hr at room temperature . Bands were detected using the SuperSignal chemiluminescence reagent substrate ( Millipore ) . The protein expression level was normalized using the housekeeping protein β-actin . Quantitative estimation of the band intensity was performed using Image J software . NOTCH 1-targeted short hairpin RNA ( shRNA ) was designed and synthesized by GeneChem . p-HASMCs and CRL-1999 cells were seeded in six-well plates ( 6 x 105 cells/well ) and cultured in a humidified incubator at 37°C with 5% CO2 . When the cells reached 30% confluency , the cells were divided into three groups: control ( WT ) , negative control shRNA ( NC ) and NOTCH1 knockdown ( NOTCH1-KD ) . The cells in the NC and NOTCH1-KD groups were infected with lentivirus-nonspecific shRNA and lentivirus-shRNA-NOTCH1 at a multiplicity of infection ( MOI ) of 10 according to the manufacturer’s recommended protocol ( GeneChem ) . After 8–12 hr of infection , the virus particles were removed from the respective wells , and fresh SMCM was added . The cells were further cultured for 72 hr in a humidified incubator at 37°C with 5% CO2 . To establish a stable cell line , the cells were treated with puromycin at a concentration of 2 µg/mL for 5 days . When the cells reached approximately 80% confluency , the cells were harvested , and the knockdown efficiency of NOTCH1 was evaluated by qRT-PCR and western blotting assays . For analyses of the mitochondrial morphology and membrane potential , we used different types of fluorescent dyes , including tetramethylrhodamine methyl ester ( TMRM ) , MitoSOX , and MitoTracker ( all from Thermo Fisher Scientific ) , according to the manufacturer’s guidelines . Cells were stretched on aorta smooth muscle-on-a-chip models for 24 hr . An appropriate concentration of each fluorescent dye was added to three different aorta smooth muscle-on-a-chip models and incubated at 37°C for 30 min in the dark . After incubation , the channels were washed three times with PBS , and the nuclei were counterstained with Hoechst ( Sigma ) for 10 min . Representative staining images of all three fluorescent dyes ( under static and strain conditions ) were acquired using a fluorescence microscope ( Leica ) and analyzed using Image J software . The level of ATP production by cells on the aorta smooth muscle-on-a-chip models was determined using an ATP assay kit ( Beyotime ) following the manufacturer’s instructions . In brief , after mechanical stimulation of the cells on the aorta smooth muscle-on-a-chip models , the cells were lysed using ATP lysis buffer , and total protein was collected by centrifugation of the cell lysate at 12 , 000 rpm and 4°C for 5 min . After centrifugation , the supernatant was collected , mixed with ATP detection reagent and incubated for 10 min at room temperature . After incubation , the ATP concentration was measured using a luminometer . An ATP concentration standard curve was then established and used to calculate the ATP concentration of each sample . The aortic samples were minced and lysed with RIPA on ice for 30 min . The extracts were centrifuged at 14 , 000 rpm and 4°C for 25 min , and the supernatant was collected after centrifugation . The protein concentrations were quantified using a BCA Protein Assay kit ( Thermo Fisher Scientific ) . Filter-aided sample preparation ( FASP ) was performed for protein digestion . Before alkylation with 10 mM dithiothreitol ( DTT , Sigma ) and 30 mM iodoacetamide ( IAA , Sigma ) , the proteins were loaded in 10 kDa centrifugal filter tubes ( Millipore ) and treated twice with 50 mM NH4HCO3 ( Sigma ) . The extracts were digested with trypsin at a ratio of 1:50 and incubated at 37°C overnight . Trifluoroacetic acid ( TFA , 0 . 1% ) was added to stop the digestion reaction . One hundred micrograms of peptides containing 0 . 1% TFA was loaded in high-pH reversed-phase fractionation spin columns ( Thermo Fisher Scientific ) . We obtained 10 flow-through fractions , and two fractions were combined to obtain one sample . The resulting five fractions were dried by vacuum centrifugation . The samples were resuspended in 30 μL of solvent A ( A: water with 0 . 1% formic acid; B: ACN with 0 . 1% formic acid ) , separated by nanoLC and analyzed by on-line electrospray tandem mass spectrometry . The experiments were performed using a nanoAquity UPLC system ( Waters Corporation ) connected to a quadrupole-Orbitrap mass spectrometer ( Q Exactive HF ) ( Thermo Fisher Scientific ) equipped with an online nanoelectrospray ion source . Two microliters of peptide sample were loaded onto an analytical column ( Acclaim PepMap C18 , 75 μm x 25 cm ) and subsequently separated with a linear gradient from 5% B to 30% B over 110 min . The column flow rate was maintained at 300 nL/min , and the column temperature was maintained at 45°C . An electrospray voltage of 2 . 2 kV versus the inlet of the mass spectrometer was used . The Q Exactive HF mass spectrometer was operated in the data-dependent mode to switch automatically between MS and MS/MS acquisition . Survey full-scan MS spectra ( m/z 350–1500 ) were acquired with a mass resolution of 60 K . The automatic gain control ( AGC ) was set to 3000000 with a maximum injection time of 50 ms . Fifteen sequential high-energy collisional dissociation ( HCD ) MS/MS scans with a resolution of 15 . 0 K were acquired with the Orbitrap . The intensity threshold was 50 , 000 , and the maximum injection time was 80 ms . The AGC target was set to 100 , 000 , and the isolation window was 1 . 6 m/z . Ions with charge states of 2+ , 3+ , and 4+ were fragmented with a normalized collision energy ( NCE ) of 30% . In all cases , one microscan was recorded using dynamic exclusion of 20 s . In the MS/MS , the fixed first mass was set to 110 . Online peaks were used for the analysis of proteomic data . The precursor mass error tolerance was set to 10 ppm with a fragment mass error tolerance of 0 . 05 Da . In all software programs , carbamidomethylation was set as a fixed modification , and variable modifications of oxidation ( M ) , acetylation ( N-term ) and deamidation ( NQ ) were included . The false discovery rate ( FDR ) for peptide and protein identifications was set to 1% . Total ion chromatography ( TIC ) was used for normalization . The rest of the parameters were set to the default values . The MS/MS spectra were searched using the Andromeda search engine against the Swiss-Prot database ( Taxonomy: Homo sapiens , Release 2020-11-02 ) ( total of 20385 entries ) . The statistical analyses were mainly conducted in R 3 . 6 . 1 . Label-free quantification ( LFQ ) was used for the following analysis flow . Proteins containing more than 50% missing values were removed , and the remaining missing values were inputted by k-nearest neighbor ( kNN ) imputation based on the Euclidean distance using the DMwR package in R . After normalizing the trimmed mean of M-values ( TMM ) , PCA showed no significant batch effect . Differential expression analysis was conducted using the Limma package . The proteins with a p-value threshold of 0 . 05 and fold change > 1 . 5 were identified as differentially expressed proteins and inputted into the IPA . The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE partner repository with the dataset identifier PXD026303 . The experimental results are expressed as the means ± standard deviations ( SDs ) . A minimum of three individual replications of each group were used for the relative analyses . The statistical analyses were performed using GraphPad Prism eight software . Two‐tailed Student’s t tests were used to compare values between two groups , and one‐way or two-way analysis of variance ( ANOVA ) followed by Tukey’s post hoc test was used for multiple-group comparisons . Statistical significance was indicated by *p < 0 . 05 , **p < 0 . 01 , ***p < 0 . 001 , and ****p < 0 . 0001 .
To explore the pathological process involved in BAV-TAA , aortic tissues were collected from six patients with BAV-TAA who underwent ascending aorta replacement and six patients with non-diseased aorta who underwent cardiac surgery . The clinical characteristics of the patients are shown in Supplementary file 1a . Hematoxylin and eosin ( H and E ) staining showed interrupted elastic fibers and thinning of the tunica media in the BAV-TAA aortas compared with non-diseased aortic tissue ( Figure 1a ) . A Western blotting analysis showed that NOTCH1 expression was significantly lower in BAV-TAA aortic tissues than in non-diseased aortic tissues . SM22 and CNN1 expression was significantly reduced in BAV-TAA aortic tissues ( Figure 1b–c ) . We evaluated the expression of mitochondrial fission- and fusion-related proteins in non-diseased and BAV-TAA aortic tissues . The results showed that DRP1 and MFF expression was increased in the BAV-TAA group compared with the non-diseased group , but the differences were not significant probably due to individual patient differences , which resulted in relatively large protein expression differences within each group . However , the protein expression of MFN1 and MFN2 was significantly lower in the BAV-TAA group than in the non-diseased group ( Figure 1c ) . In general , we found that MFN1 and MFN2 were expressed at low levels in tissues with NOTCH1 insufficiency ( Figure 1d ) . The expression of NOTCH1 exhibited a positive correlation with MFN1 and MFN2 in aortic tissues . The expressions of DRP1 and MFF did not show a correlation with the expression of NOTCH1 ( Figure 1e ) . To further explore the biological differences between non-diseased and BAV-TAA aortic tissues , a comparative proteomics analysis of global proteins in aortic tissues was performed by high-performance liquid chromatography tandem mass spectrometry . In total , 70 upregulated proteins and 257 downregulated proteins were identified ( Supplementary file 2 ) . The enriched canonical pathways identified by Ingenuity Pathway Analysis ( IPA ) showed that acute phase response signaling , mitochondrial dysfunction and oxidative phosphorylation pathways were significantly enriched in BAV-TAA aortic tissues ( Figure 1f , Supplementary file 1c ) . Among these pathways , Z-score of oxidative phosphorylation pathway was −2 . 333 , indicating significantly inhibited . Z-score of mitochondria dysfunction was not applicable due to insufficient evidence in the knowledge base for confident activity predictions across datasets . In addition , metabolic signaling pathways affecting mitochondrial function , such as the EIF2 and sirtuin signaling pathways , were also significantly enriched in BAV-TAA . Figure 1g and Supplementary file 1d show the enriched proteins associated with mitochondrial dysfunction , oxidative phosphorylation pathways , and acute phase response signaling . In total , seven upregulated proteins and 11 downregulated proteins were found to be related to acute phase response signaling , and one upregulated protein and 10 downregulated proteins were associated with mitochondrial dysfunction ( Figure 1g , Supplementary file 1d ) . Figure 1h shows a volcano plot of these 11 differentially expressed proteins that allows visualization of the fold change and p-value of all differentially expressed proteins between the two groups . MFN1 , MFN2 , and NOTCH1 were not detected by mass spectrometry analysis , mainly because the aortic tissues contain highly cross-linked extracellular matrix that can be refractory to protein extraction . The publicly available data of the most compressive clinical aortic proteome , up to now , also indicated the missingness of MFN1 , MFN2 , DRP1 , and NOTCH1 protein in aortic specimens ( Herrington et al . , 2018 ) . During cardiac systolic and diastolic cycles , the thoracic aortic wall experiences rhythmic tensile strain . Ascending aorta is the first section of the aorta , which starts from the left ventricle of the heart and extends to the aortic arch . It is connected to the left ventricular outflow track and is the part that pumps oxygenated blood to the body's tissues and organs . Clinical studies have shown that the circumferential strains of the aortic wall range from low values of 7 . 0 ± 2 . 5% to high values of 21 . 5 ± 12 . 4% , and these can be further influenced by age , the aortic diameter , and the presence of aortopathy ( Akazawa et al . , 2016; Bell et al . , 2014 ) . To better simulate the rhythmic tensile strain experienced by HAoSMCs in vivo , we developed a compact microfluidic aorta smooth muscle-on-a-chip model with commercialized highly flexible polydimethylsiloxane ( PDMS ) membranes ( Figure 2a ) . The model was composed of three chambers: ( i ) a top vacuum chamber deforming the upper PDMS membrane; ( ii ) a middle chamber containing the culture medium to maintain cell growth on the PDMS membranes; and ( iii ) a bottom vacuum chamber deforming the lower PDMS membrane . HAoSMCs were cultured on the PDMS membranes in the middle cell culture chamber . The dimensions of the model are shown in Figure 2—figure supplement 1a . The measured Young's elastic modulus values of the commercialized PDMS membrane were 1 . 71 MPa within 25% tensile strain and 1 . 67 MPa within 500% tensile strain ( Figure 2—figure supplement 2 ) and it shows excellent homogeneity and tensile properties ( Figure 2—figure supplement 3 ) . The rhythmic tensile strain was generated by connecting the top and bottom chambers to a vacuum pump that cyclically deformed the PDMS membrane . The rhythm and value of the dynamic negative pressure in the chambers were controlled by a set of apparatuses consisting of a monochip computer , a pressure regulator and a solenoid valve . To quantify the tensile strains of the PDMS membrane generated by negative pressure in a vacuum chamber , we captured the real-time deformations of the PDMS membranes and measured the changes in length . We captured the real-time deformations of the PDMS membranes from a cross-sectional view of the microfluidic model , with vacuum pressures of 0 kPa , 10 kPa , and 15 kPa , and measured the strain magnitude of the PDMS membrane ( Figure 2b ) . The two deformations of the upper and lower PDMS membranes were coincident in terms of amplitudes and rhythms ( Figure 2c ) . A vacuum pressure of 10 kPa induced 7 . 18 ± 0 . 44% strain ( 7 . 09 ± 0 . 18% strain in the lower layer and 7 . 27 ± 0 . 28% strain in the upper layer ) , and 15 kPa induced 17 . 28 ± 0 . 91% strain ( 17 . 29 ± 0 . 62% strain in the lower layer and 17 . 23 ± 0 . 64% strain in the upper layer ) ( Figure 2c and Figure 2—figure supplement 1b ) . We tested three prototypes with varying culturing channels , that is 2 , or 4 , or 6 mm in width ( Figure 2b–c and Figure 2—animations 1–6 ) . Finally , we opted for the largest size to harvest enough cells for protein analysis replication . To replicate different strains on the human aortic wall , we applied 7 . 18 ± 0 . 44% strain induced by a vacuum pressure of 10 kPa as a relatively low strain and 17 . 28 ± 0 . 91% strain induced by a vacuum pressure of 15 kPa as a relatively high strain . To identify the effect of rhythmic tensile strain on the contractility of HAoSMCs , the changes in cellular morphology , alignment and contractile/synthetic phenotypic markers were assessed under rhythmic low/high strain or static conditions ( Figure 3a ) . Cytoskeletal F-actin staining images of HAoSMCs showed a decrease in the cell width and an increase in the cell length in the presence of rhythmic low or high strain ( Figure 3b ) . The results also revealed an increase in the length-to-width ratio from 2 . 33 ± 0 . 82 under static conditions to 2 . 74 ± 1 . 01 under low strain or 3 . 50 ± 1 . 19 under high strain ( Figure 3c ) . Compared with the irregular orientation of the cells observed under static culture , the cells tended to align perpendicularly to the direction of the applied strain . The angle between the directions of the cellular alignment and the applied strain was approximately 90° ( Figure 3d ) . To evaluate the effect of rhythmic tensile strain on the expression of phenotypic markers , the protein levels of SM22 , CNN1 , and OPN in HAoSMCs were measured by western blotting . The results showed that the contractile phenotype markers SM22 and CNN1 were upregulated under either low or high strain ( Figure 3e–f ) . The expression of the synthetic phenotypic marker OPN under both rhythmic low and high strain was lower than that observed under static conditions . These results indicate that the application of rhythmic strain can induce HAoSMCs to spread to longer shapes , align unidirectionally , and exhibit enhanced contractility on the chip model . To clarify the effect of NOTCH1 insufficiency on HAoSMC contractility , we cultured NOTCH1-insufficient cells under rhythmic low or high strain or static conditions and then characterized the expression of phenotypic markers . The schematic workflow of the experimental design is shown in Figure 4a . First , NOTCH1 was knocked down in HAoSMCs using a lentivirus expressing short hairpin RNA ( shRNA ) targeting NOTCH1 ( Shao et al . , 2015 ) . To evaluate the effect of the lentivirus vector on NOTCH1 and phenotypic markers , HAoSMCs without any treatments ( WT ) , HAoSMCs treated with negative control shRNA ( NC ) and HAoSMCs treated with NOTCH1 shRNA ( NOTCH1-KD ) were verified using quantitative real-time PCR and western blotting experiments . An approximately 60% reduction in NOTCH1 mRNA expression was found in the NOTCH1-KD group compared with the WT and NC groups ( Figure 4—figure supplement 1a ) . Under static conditions , the mRNA expression of SM22 and CNN1 was upregulated and that of OPN was downregulated in the NOTCH1-KD group compared with the WT and NC groups ( Figure 4—figure supplement 1a ) . The same tendency was found for the protein expression levels by western blotting analysis ( Figure 4—figure supplement 1b–c ) . NOTCH1-KD and WT HAoSMCs were cultured on aorta smooth muscle-on-a-chip models under rhythmic strain and static conditions to characterize the expression of SM22 and CNN1 . As shown in Figure 4b–c , immunofluorescent staining images showed that SM22 and CNN1 were upregulated in the NOTCH-KD group compared with the WT group under static conditions . However , the opposite results were observed under rhythmic strain conditions: the expression of SM22 and CNN1 in NOTCH-KD HAoSMCs was lower than that in the WT group under rhythmic strain conditions . Western blotting analyses revealed similar alterations in the expression of SM22 and CNN1 ( Figure 4d–e ) . In WT HAoSMCs , rhythmic strain induced the upregulation of SM22 and CNN1 expression compared with the levels observed under static conditions . However , a downregulation of SM22 and CNN1 expression was detected in NOTCH-KD HAoSMCs exposed to rhythmic strain . These results suggested that rhythmic strain induced different effects between WT and NOTCH-KD HAoSMCs . In NOTCH-KD HAoSMCs , SM22 and CNN1 expression were higher under static conditions and lower under rhythmic strain than those in WT HAoSMCs . Furthermore , we assessed the alterations in mitochondrial function and dynamics in NOTCH1-KD HAoSMCs . Specifically , western blotting was performed to detect the protein expression of MFN1 and MFN2 ( mitochondrial fusion-related proteins ) and DRP1 and MFF ( mitochondrial fission-related proteins ) . No significant difference in MFN1 , MFN2 , DRP1 , and MFF protein expression was found between the WT and NC groups under static conditions , which suggested that the expression of these mitochondrial dynamics-related proteins was not affected by the lentivirus vector ( Figure 4—figure supplement 1d–e ) . As shown in Figure 5a–b , the expression of MFN1 and MFN2 was downregulated in NOTCH1-KD HAoSMCs under static conditions , and the expression of MFN2 presented a significant difference between the two groups . DRP1 and MFF expression was significantly increased in NOTCH1-KD HAoSMCs under static conditions . Rhythmic strain induced a decrease in MFN1 expression in the NOTCH1-KD group but an increase in the WT group . MFN2 expression was decreased in both the WT and NOTCH1-KD groups after exposure to rhythmic strain but was lower in the NOTCH1-KD group than in the WT group , particularly under rhythmic high strain conditions . Rhythmic strain induced an increase in DRP1 and MFF expression in both the WT and NOTCH1-KD groups , and DRP1 and MFF were highly expressed in NOTCH1-KD HAoSMCs under both rhythmic strain and static conditions ( Figure 5b ) . The fragmented mitochondrial morphology observed in NOTHC1-KD HAoSMCs under rhythmic strain conditions indicated decreased mitochondrial fusion , increased mitochondrial fission , that is , impaired mitochondrial dynamics . As shown in Figure 5c , MitoTracker staining of the mitochondrial shape indicated elongated and interconnected mitochondrial networks in both WT and NOTCH1-KD groups under static conditions . Low or high strain induced alterations of the mitochondrial shape in both WT and NOTHC1-KD HAoSMCs , and these alterations included both a decrease in long rod-shaped elongated interconnected mitochondrial networks and a significant increase in spheroid-shaped fragmented mitochondrial morphology and were more frequent in NOTHC1-KD HAoSMCs under high strain conditions ( Figure 5c–d ) . Mitochondrial dynamics play an important role in the maintenance of normal mitochondrial function . As shown in Figure 5e–f , the fluorescence intensity of the tetramethylrhodamine methyl ester perchlorate ( TMRM ) staining , which reflects the mitochondrial membrane potential , was lower in NOTHC1-KD HAoSMCs than in the WT group under rhythmic strain . These results indicated loss of mitochondrial membrane potential in NOTCH1-KD HAoSMCs under rhythmic strain . The mitochondrial superoxide ( MitoSOX ) staining of NOTHC1-KD HAoSMCs was significantly higher than that of the WT group under rhythmic strain ( Figure 5g–h ) , which indicated that ROS production was increased in NOTCH1-KD HAoSMCs under rhythmic strain conditions . As an energy source of cells synthesized by mitochondria , the ATP concentration was also evaluated . Rhythmic strain increased the ATP concentration in both WT and NOTHC1-KD HAoSMCs , but under rhythmic strain conditions , a lower ATP concentration was found in NOTCH1-KD HAoSMCs than in WT HAoSMCs ( Figure 5i ) . No significant difference in the TMRM or MitoSOX fluorescence intensity or ATP concentration was found between WT and NOTCH1-KD HAoSMCs under static conditions , which indicated that NOTCH1 insufficiency did not affect mitochondrial function under static conditions . Taken together , these data indicated that NOTCH1 insufficiency could induce mitochondrial dysfunction in HAoSMCs by reducing mitochondrial fusion , inducing loss of mitochondrial membrane potential , increasing ROS production and generating insufficient ATP under rhythmic strain , and these effects are accompanied by an impaired contractile phenotype . These findings were consistent with previous studies showing that imbalanced mitochondrial dynamics could induce VSMC dedifferentiation into the synthetic phenotype ( Salabei and Hill , 2013 ) . To confirm whether the decreased contractile phenotype of NOTCH1-knockdown HAoSMCs can be rescued by inhibition of mitochondrial fission or activation of mitochondrial fusion , we evaluated the phenotypic alterations and mitochondrial dynamics of NOTCH1-insufficient HAoSMCs treated with a mitochondrial fission inhibitor ( Mdivi-1 ) or mitochondrial fusion activators ( leflunomide or teriflunomide ) under rhythmic high-strain conditions . A schematic workflow of the drug screening experiments was shown in Figure 6a . Non-aneurysmal HAoSMCs with NOTCH1 knockdown and BAV-TAA HAoSMCs were used in drug testing experiments . The expression of MFN1 and/or MFN2 was reduced in NOTCH1-knockdown HAoSMCs of non-aneurysmal patient #1 , #2 , #3 and cells of BAV-TAA patient #2 , #3 under rhythmic high-strain condition compared to static condition . Also , the cellular expression of SM22 and/or CNN1 was reduced in non-aneurysmal patient #1 , #2 and BAV-TAA patient #1 , #2 , #3 under rhythmic high-strain condition compared to static condition . ( Figure 6—figure supplement 1 ) . In the model using CRL1999 HAoSMCs ( Figure 6b ) , all three drugs enhanced MFN1 and MFN2 expression to different extents , and the greatest increase in MFN1 expression was obtained with Mdivi-1 . In addition , the three drugs all decreased DRP1 expression to different extents but did not affect MFF expression . The three drugs also increased SM22 and CNN1 expression to different extents , and the maximal expression of SM22 and CNN1 was obtained with leflunomide . In p-HAoSMCs isolated from patients in the non-diseased group ( Figure 6c ) , the three drugs increased MFN1 expression , but no significant difference was found compared with the control group . Teriflunomide significantly increased MFN2 and DRP1 expression and significantly decreased MFF expression . Although the effects of these drugs on mitochondrial dynamics were not desirable , all three drugs increased the expression of SM22 and CNN1 to different extents . In p-HAoSMCs ( ATCC ) ( Figure 6d ) , none of the three drugs exerted an obvious effect on MFN1 , MFN2 , and DRP1 expression , and teriflunomide significantly decreased MFF expression . All three drugs increased the expression of SM22 and CNN1 to different extents . To further evaluate the potential of these drugs for clinical use in the treatment of NOTCH1-insufficient patients with BAV-TAA , we isolated p-HAoSMCs from the aortic tissues of three individual patients with BAV-TAA . The clinical characteristics of three patients with BAV-TAA are shown in Supplementary file 1e . Significantly lower NOTCH1 protein expression was found in p-HAoSMCs from BAV-TAA aortic tissues compared with p-HAoSMCs from non-diseased aortic tissue ( Figure 6—figure supplement 2 ) . The three drugs exerted different effects on mitochondrial dynamics-related proteins ( Figure 6e–g ) . All three drugs enhanced MFN2 expression to different extents , and maximal MFN2 expression was obtained with the treatment of teriflunomide . The three drugs increased MFN1 expression and decreased DRP1 expression to different extents in two lines of BAV-TAA p-HAoSMCs , and the greatest decrease in DRP1 expression was obtained with teriflunomide . The three drugs decreased MFF expression to different extents in the two lines of BAV-TAA p-HAoSMCs and exerted different effects on the expression of the contractile proteins SM22 and CNN1 , and teriflunomide significantly increased SM22 expression in the two lines of BAV-TAA p-HAoSMCs . All three drugs enhanced CNN1 expression to different extents , and Mdivi-1 or teriflunomide significantly increased CNN1 expression in one line of BAV-TAA p-HAoSMCs . Different individual cases exhibited unevenness in drug response . Racial differences may lead to different cellular response to experimental stimuli or to drug treatment . HAoSMCs used in this study were from different human races . One primary HAoSMC line ( ATCC , CRL1999 ) was isolated from a Caucasian donor and another one ( ATCC , PCS-100–012 ) was isolated from an African American donor . Other four primary HAoSMCs were isolated from East Asian patients . Differences in patients’ clinical characteristics , such as age , sex , and aortic diameter , may also have an impact on the experiment results . BAV aortopathy exclusively involves proximal aorta , including aortic root , ascending aorta and aortic arch , but spares the descending aorta and abdominal aorta . The majority of VSMCs at the proximal aorta and descending aorta arise from lineages of neural crest and paraxial mesoderm respectively . The heterogeneity of VSMCs could lead to section-specific microphysiology in the aortic wall and differences in the vulnerability of VSMCs to pathogenic stimuli . Thus , for the purpose of minimizing bias in drug response , HAoSMCs derived exclusively from the proximal aorta should be preferentially utilized in BAV aortopathy models . Overall , the data from BAV-TAA p-HAoSMCs indicated that teriflunomide exerted a more obvious effect on rescuing impaired mitochondrial dynamics and the expression of the contractile phenotype proteins SM22 and CNN1 under rhythmic high-strain conditions . These results indicated that leflunomide , Mdivi-1 , and , in particular , teriflunomide , could serve as drug candidates for ameliorating the disease progression of BAV-TAA .
In this study , protein analyses of human aortic aneurysm tissues suggested the insufficient expression of NOTCH1 in BAV-TAA was associated with the impaired mitochondrial dynamics and OXPHOS . To verify it , we established a microfluidic model of aorta smooth muscle-on-a-chip , inspired by previous organ-on-chips models ( Ribas et al . , 2017; Yasotharan et al . , 2015; Huh et al . , 2010 ) , enabling us to reproducibly generate the rhythmic tensile strain of the native human aortic wall . We showed that this model could apply a biomimetic rhythmic tensile strain with a confined amplitude/magnitude and rhythm to HAoSMCs . We also confirmed that impaired mitochondrial fusion contributed to the attenuated contractile phenotype observed in NOTCH1-KD HAoSMCs . We further identified that MFN1/2 agonists and DRP1 inhibitors were able to reverse the imbalance in mitochondrial dynamics and partly rescue the contractile phenotype in NOTCH1-insufficient HAoSMCs . The majority of studies on the molecular mechanism and pharmacotherapy for TAA have been conducted using conventional cell culture models and animal models . The pathophysiological features of human aortic aneurysm can be induced in animal models by elastase , calcium chloride , angiotensin II or transgenesis , which partially enable investigation of the etiology , pathogenesis , and therapeutic targets of TAA at an early stage ( Eckhouse et al . , 2013; Ikonomidis et al . , 2003; Mao et al . , 2015 ) . However , all these animal models cannot replicate the native characteristics of BAV-TAA . Moreover , the apparent species gap between preclinical animal experiments and clinical studies might also impede pharmaceutical discovery ( Lindeman and Matsumura , 2019 ) . In contrast , the shortcoming of the conventional cell culture model of aortic disease lies in its inability to recapitulate in vivo biomechanical stimuli . Advances in microfabrication and microfluidics have provided novel techniques for building organ-on-a-chip models that integrate rhythmic tensile strains into in vitro cell culture . Differently from conventional two dimensional ( 2D ) and three dimensional ( 3D ) cell culture methods , which are widely used in biological research , organ-on-a-chip models can replicate multicellular architectures , tissue-tissue interfaces , and biomechanical forces that exist in vivo , and precisely pattern cells and manipulate various mechanical and chemical parameters , such as flow rate , stretch , pressure , oxygen , and pH , providing controllable culture conditions not possible with conventional cultures ( Ahadian et al . , 2018 ) . In the case of cardiovascular research , the functionally important cardiomyocytes , endothelial cells and smooth muscle , are constantly subjected to hemodynamic factors in vivo , including blood flow shear stress , rhythmic strain and fluid pressure , which cannot be simulated and given to the cells by conventional cell culture techniques . Organ-on-a-chip models can be broadly defined as any form of microfabricated cell culture device that models organ-specific biochemical and physical microenvironments ( Zhao et al . , 2019; Park et al . , 2019 ) and represent an in vitro platform that is complementary to animal models . The present aorta smooth muscle-on-a-chip model was designed based on the inspiration and reference of pioneering researchers' works , such as progeria-on-a-chip ( Ribas et al . , 2017 ) , artery-on-a-chip ( Yasotharan et al . , 2015 ) , and so on . Previous reported 'artery-on-a-chip' platforms are mainly focused on peripheral vessel , cerebral artery , or other arterial diseases . Yasotharan et al . reported an artery-on-a-chip platform that mimicked in vivo transmural pressure of an outer diameter of 120 μm olfactory artery segment , in which vascular tone and calcium dynamics were simultaneously assessed ( Poussin et al . , 2020 ) . Poussin et al . have established a model of microvessels-on-a-chip under flow using primary human coronary arterial endothelial cells , to measure the adhesion of monocytes to the lumen of perfused microvessels ( Poussin et al . , 2020 ) . Arterioles with a diameter of 15 μm , branches of the large artery , are mainly affected by fluid shear but receive little tensile tension . In contrast , aorta diameter is about 25–35 mm with thick three layers , in which smooth muscle cells are the main component in tunica media and experience high tensile strain . Aortopathies are the degenerative changes in the aortic wall inducing thinning or even rupture of the aorta . Thus , HAoSMCs in tunica media and the tensile strain they receive are essential components for the success of aorta smooth muscle-on-a-chip model . By anatomizing the complexity of the biomechanical microenvironment , this study focused on the physiological rhythmic strain on HAoSMCs in the aortic tunica media , which plays a central role in the pathogenesis of aneurysms . In particular , the strain of the ascending aorta , which was defined as the maximal change in the ascending aorta diameter over a cardiac cycle , can vary diversely from a low value of 7 . 0 ± 2 . 5% to a high value of 21 . 5 ± 12 . 4% depending on age , sex and disease severity ( Akazawa et al . , 2016; Bell et al . , 2014; Morrison et al . , 2009 ) . Under rhythmic strain , healthy HAoSMCs maintain a contractile phenotype , which allows regulation of the vascular tone . While , diseased HAoSMCs dedifferentiate into a synthetic phenotype under pathological conditions , and this dedifferentiation represents the initial step toward an aneurysm pathology ( Petsophonsakul et al . , 2019 ) . The aorta smooth muscle-on-a-chip model developed in this study featured the controlled magnitude and rhythm of mechanical stimuli relevant to the human pathophysiological parameters of aorta biomechanics . In this study , HAoSMCs exhibited longer shapes in morphology , align unidirectionally , and present contractile phenotype on the on-chip model . In static conventional condition , HAoSMCs present more synthetic phenotype , which is oppositely different from the phenotype exist in native normal aortic wall . Most importantly , the on-chip model recapitulated the imbalanced mitochondrial dynamics which was accordant with analyses on tissues from human BAV-TAA and in vivo mouse model of abdominal aortic aneurysm ( Cooper et al . , 2021 ) . In short , organ-on-a-chip models can mimic the biomechanical parameters , which are essential for aortopathy disease development and more fully explore the pathophysiological changes of the cells and their real responsiveness to drugs , in a complementary manner with conventional cell culture and in vivo models . However , both of extracellular matrix and vascular cells , including epithelial cells , VSMCs , fibroblasts , macrophages , contribute to the pathological process of the aortic diseases . The present reductionist on-chip model merely represented HAoSMCs activities and did not recapitulated the entire aortic wall . Future works are expected to incorporate more components , especially the aortic extracellular matrix , into the on-chip model and explore ECM degradation and interactive molecular mechanisms involved in human aortopathies . NOTCH1 has an important role in the maintenance of normal mitochondrial dynamics . Previous studies have reported that the expression of NOTCH1 signaling was attenuated in the aortic tissues of patients with BAVs ( Sciacca et al . , 2013 ) . Additionally , Balistreri et al . found decreased mRNA levels of NOTCH1-4 in aortic aneurysm tissues and decreased protein levels of soluble NOTCH1 in plasma ( Balistreri et al . , 2018 ) . Using clinical samples , our protein analysis also indicated that patients with BAV-TAA presented lower expression of NOTCH1 protein in ascending aortic aneurysms; and the NOTCH1 insufficiency was accompanied with attenuated mitochondrial fusion . This was consistent with the previous studies that reported NOTCH1 regulates mitochondria dynamics in cardiovascular cell differentiation and survival . Kasahara et al . found that mitochondrial fusion regulates the differentiation of myocardial cells through NOTCH1 signaling pathway ( Kasahara et al . , 2013 ) . Here , we found that NOTCH1 insufficiency induced mitochondrial dysfunction in HAoSMCs by reducing mitochondrial fusion , along with loss of mitochondrial membrane potential , increasing ROS production and generating insufficient ATP under rhythmic strain . These results indicated that the relationship between impaired mitochondrial fusion and NOTCH1 deficiency in BAV-TAA . Consistently , we found decreased contractile phenotype in NOTCH1-deficient HAoSMCs under rhythmic tensile strain and in BAV-TAA aortic tissues , accompanied by lower NOTCH1 expression and impaired mitochondrial fusion . NOTCH1 signaling is also involved in the regulation of HAoSMCs differentiation and the expression of contractile proteins . HAoSMCs are highly plastic cells and undergo reversible changes in phenotype in response to environmental stimuli ( Vásquez-Trincado et al . , 2016 ) . In the aorta , HAoSMCs are considered to have the functional roles of both maintaining aortic tone in response to hemodynamic stimuli , and synthesizing and modeling the ECM ( Michel et al . , 2018 ) . Most healthy HAoSMCs in the vascular wall in vivo exhibit a contractile phenotype , which allows them to maintain vascular tone ( Milewicz et al . , 2008 ) . HAoSMCs contractile units associated proteins , such as SM22 , CNN1 , MYH11 , and α-SMA , distribute the force on the aortic wall through regulation of extracellular matrix . Decreases in α-SMA , MYH11 , SM22 , and CNN1 attenuate HAoSMCs contractility unit formation and further disrupt force generation , promoting the development of aortic aneurysm or dissections ( Gillis et al . , 2013 ) . Differentiated HAoSMCs have a contractile phenotype with little proliferation and little secretion of extracellular matrix . HAoSMCs transform from a contractile into a synthetic phenotype and induce loose of vascular tone , which is one of the major pathological process in TAA . Thus , the enhancement of HAoSMCs contractility is indicative of the effectiveness of pharmacotherapy for controlling aortic aneurysm ( Oller et al . , 2021 ) . In this study , we found that NOTCH1 insufficiency in HAoSMCs downregulated the contractile phenotype proteins SM22 and CNN1 under rhythmic strain . Also , the expressions of SM22 and CNN1 were decreased in BAV-TAA aortic tissues compared to non-diseased aortic tissues . Noseda et al . reported that NOTCH1 activation was needed for expression of the contractile phenotype protein αSMA in VSMCs via the NOTCH/CSL axis ( Noseda et al . , 2006 ) . High et al . reported that NOTCH1 promotes the differentiation of the cardiac neural crest into differentiated VSMCs and the expression of αSMA ( High et al . , 2007 ) . However , the specific mechanism by which NOTCH1 leads to a reduced contractile phenotype has not been fully investigated . Based on results obtained from the aorta smooth muscle-on-a-chip model , it comes to the assumption that NOTCH1-insufficiency reduced the contractile phenotype through decreasing mitochondrial fusion . Mitochondrial dynamics itself has been recognized as one of the critical factors that regulate the phenotype switching of vascular smooth muscle cells ( VSMCs ) . Salabei and Hill reported that PDGF-BB induces the dedifferentiation of VSMCs into the synthetic phenotype through overactivation of mitochondrial fission ( Salabei and Hill , 2013 ) . Chen et al . found that the expression of MFN2 is markedly reduced in hyperproliferative VSMCs from spontaneously hypertensive rat arteries ( Chen et al . , 2004 ) . Using the aorta smooth muscle-on-a-chip model , we demonstrated the decreased mitochondrial fusion in NOTCH1-insufficient HAoSMCs , accompanied with decreased contractile phenotypes . Thus , mitochondrial dynamics regulated by NOTCH1 may have a role in maintains of contractile phenotype . In addition , we also found the contractile phenotype was upregulated in NOTCH1-insufficient HAoSMCs after treatment with mitochondrial fusion activators and mitochondrial fission inhibitors , which further verified the hypothesis that contractile phenotype of HAoSMCs was regulated by NOTCH1-mitochondrial fusion axis . In clinical samples , our protein analysis indicated that aortic tissues from patients with BAV-TAA presented the reduced expressions of mitochondrial fusion proteins and contractile phenotype proteins , compared with non-diseased aortic tissues . Further , the mass spectrometry-based proteomic analysis of aortic tissue samples indicated that mitochondrial dysfunction and OXPHOS pathways were significantly altered in BAV-TAA aortic tissues . Consistent with the previous single-cell transcriptome analysis , TAA aortic tissues showed decreased mitochondrial function; and this finding suggests that OXPHOS ATP production might be insufficient for VSMC contractile activities ( Li et al . , 2020 ) . Therefore , maintaining normal mitochondrial dynamics via NOTCH1 could be one of the mechanisms through which NOTCH1 regulates HAoSMCs’s contractility and maintains vascular wall tension . The mitochondria morphology , that is fragmentation or elongation , which is controlled by precisely regulated mitochondrial fusion and fission , has been related to cardiovascular disorders , such as atherosclerosis and myocardial infarction ( Murphy et al . , 2016 ) . Excessive mitochondrial fission induced a reduction in mitochondrial membrane potential and contractile phenotype of vascular VSMCs , and an increase in oxygen species production ( Lim et al . , 2015 ) . These effects induced by mitochondrial fission were prevented by Mdivi-1 , which is an inhibitor of mitochondrial fission related protein DRP1 . Cooper et al . found that the upregulated DRP1 and mitochondrial fission in mouse abdominal aortic aneurysm , associated with impaired mitochondrial function and decreased contractility of mouse VSMCs ( Cooper et al . , 2021 ) . The induction of the contractile to synthetic phenotype switch of VSMCs by platelet-derived growth factor , was also found to be associated with mitochondrial fragmentation/fission and attenuated MFN2 protein levels ( Salabei and Hill , 2013 ) . In this study , the excessive mitochondrial fragmentation of HAoSMCs implied diseased phenotype , while the restoration of mitochondrial homeostasis , i . e . the balance of fragmentation/fission and elongation/fusion , implied the rescue of HAoSMC contractility abnormality . However , the specific mechanism by which reduced mitochondrial fusion causes a reduction in the contractile phenotype requires further investigation . In addition , our proteomic analysis of tissue samples showed that several mitochondrial dysfunction-related pathways , namely , EIF2 signaling and the sirtuin signaling pathway , were also altered in BAV-TAA aortic tissue , which may provide research directions to investigate further mechanisms . Lastly , we demonstrated that MFN1/2 agonists and DRP1 inhibitors reversed the imbalanced mitochondrial dynamics in BAV-TAA p-HAoSMCs with intrinsic NOTCH1 insufficiency or in NOTCH1-KD HAoSMCs on the aorta smooth muscle-on-a-chip model , which provides a prototype for an in vitro drug testing platform . Two FDA-approved drugs , leflunomide or teriflunomide , rescued the contractile phenotype of BAV-TAA p-HAoSMCs and showed their potential to ameliorate disease progression . Zorzano’s group first found that dihydroorotate dehydrogenase inhibitors ( leflunomide , teriflunomide , or brequinar sodium ) could promote mitochondrial elongation through induction of the mitochondrial proteins MFN1 and MFN2 in HeLa cells ( Miret-Casals et al . , 2018 ) . The experimental data obtained in this study encourage further studies on the application of leflunomide and teriflunomide in BAV-TAA . This study has several limitations . Although the phenotypic reversal has observed after treatments of mitochondrial fission inhibitor and fusion agonists based on aorta smooth muscle-on-chip model , in vivo genetically modified mouse experiments are still needed to further confirm the therapeutic effects , that is whether these drugs can halt the progression of the disease . Also , genetically modified mouse aortic aneurysm model is required to further elucidate the genetic mechanism of NOTCH1—mitofusin axis in ascending aortic aneurysm . At the current stage , aorta smooth muscle-on-chip is still insufficient to alone predict clinical success , but it may be complementary with animal models in the sense that , together they are able to provide more comprehensive basis for preclinical assays with greater predictive power . Therefore , more verification tests , including in vitro on-chip tests and in vivo animal validations , are needed before translating the finding into prospective clinical tail . In conclusion , we constructed an aorta smooth muscle-on-a-chip model , which could serve as a complementary tool to the current cell culture system and animal models . Using the aorta smooth muscle-on-a-chip model , we found that NOTCH1 insufficiency in HAoSMCs induced phenotypic switching from a contractile to a synthetic phenotype accompanied by an impairment of mitochondrial fusion , implying its potential role as a therapeutic target for BAV-TAA . At the current stage , in vitro microphysiological models and animal models are complementary to each other in the sense that they are able to provide more comprehensive basis for preclinical assays with greater predictive power . | To function properly , the heart must remain a one-way system , pumping out oxygenated blood into the aorta – the largest artery in the body – so it can be distributed across the organism . The aortic valve , which sits at the entrance of the aorta , is a key component of this system . Its three flaps ( or ‘cusps’ ) are pushed open when the blood exits the heart , and they shut tightly so it does not flow back in the incorrect direction . Nearly 1 . 4% of people around the world are born with ‘bicuspid’ aortic valves that only have two flaps . These valves may harden or become leaky , forcing the heart to work harder . This defect is also associated with bulges on the aorta which progressively weaken the artery , sometimes causing it to rupture . Open-heart surgery is currently the only way to treat these bulges ( or ‘aneurysms’ ) , as no drug exists that could slow down disease progression . This is partly because the biological processes involved in the aneurysms worsening and bursting open is unclear . Recent studies have highlighted that many individuals with bicuspid aortic valves also have lower levels of a protein known as NOTCH1 , which plays a key signalling role for cells . Problems in the mitochondria – the structures that power up a cell – are also observed . However , it is not known how these findings are connected or linked with the aneurysms developing . To answer this question , Abudupataer et al . analyzed the proteins present in diseased and healthy aortic muscle cells , confirming a lower production of NOTCH1 and impaired mitochondria in diseased tissues . They also created an ‘aorta-on-a-chip’ model where aortic muscle cells were grown in the laboratory under conditions resembling those found in the body – including the rhythmic strain that the aorta is under because of the heart beating . Abudupataer et al . then reduced NOTCH1 levels in healthy samples , which made the muscle tissue less able to contract and reduced the activity of the mitochondria . Applying drugs that tweak mitochondrial activity helped tissues from patients with bicuspid aortic valves to work better . These compounds could potentially benefit individuals with deficient aortic valves , but experiments in animals and clinical trials would be needed first to confirm the results and assess safety . The aorta-on-a-chip model developed by Abudupataer et al . also provides a platform to screen for drugs and examine the molecular mechanisms at play in aortic diseases . | [
"Abstract",
"Introduction",
"Materials",
"and",
"methods",
"Results",
"Discussion"
] | [
"medicine"
] | 2021 | Aorta smooth muscle-on-a-chip reveals impaired mitochondrial dynamics as a therapeutic target for aortic aneurysm in bicuspid aortic valve disease |
There is an urgent requirement for safe , oral and cost-effective drugs for the treatment of visceral leishmaniasis ( VL ) . We report that delamanid ( OPC-67683 ) , an approved drug for multi-drug resistant tuberculosis , is a potent inhibitor of Leishmania donovani both in vitro and in vivo . Twice-daily oral dosing of delamanid at 30 mg kg-1 for 5 days resulted in sterile cures in a mouse model of VL . Treatment with lower doses revealed a U-shaped ( hormetic ) dose-response curve with greater parasite suppression at 1 mg kg-1 than at 3 mg kg-1 ( 5 or 10 day dosing ) . Dosing delamanid for 10 days confirmed the hormetic dose-response and improved the efficacy at all doses investigated . Mechanistic studies reveal that delamanid is rapidly metabolised by parasites via an enzyme , distinct from the nitroreductase that activates fexinidazole . Delamanid has the potential to be repurposed as a much-needed oral therapy for VL .
The repurposing of drugs and clinical candidates offers an attractive alternative to de novo drug discovery ( Fischbach and Walsh , 2009; Cragg et al . , 2014; Peters , 2013; Law et al . , 2013; Novac , 2013; Aube , 2012 ) , particularly in terms of reducing research and development costs for neglected diseases of poverty ( Andrews et al . , 2014 ) . Visceral leishmaniasis ( VL ) , a neglected tropical disease resulting from infection with the protozoan parasites Leishmania donovani or L . infantum is a case in point , with the two anti-leishmanial front-line therapies miltefosine and amphotericin B both originally developed for other indications ( Stuart et al . , 2008 ) . In addition , the anti-trypanosomal clinical candidate fexinidazole was recently discovered to have potent activity in a murine VL model ( Wyllie et al . , 2012 ) , resulting in a phase II proof of concept clinical trial ( NCT01980199 ) against VL being conducted in Sudan . There are approximately 50 , 000 reported cases of VL per year , with the vast majority of infections in South America , East Africa and the Indian subcontinent . However , the number of cases is likely to be vastly underreported , with the actual annual incidence estimated to be between 200 , 000 and 400 , 000 ( Alvar et al . , 2012 ) . VL is fatal if untreated and , in the absence of effective vaccines and vector control methods , efficacious chemotherapy is required to combat the disease . Each of the currently available drugs has one or more drawbacks , including the need for hospitalization , prolonged therapy , parenteral administration , high cost , variable efficacy , severe toxic side-effects and resistance ( Croft et al . , 2006 ) . Thus , there is an urgent need for better , safer efficacious drugs that are fit-for-purpose in resource-poor settings . Given the success of repurposing fexinidazole for use in the treatment of VL ( Wyllie et al . , 2012 ) , there is now a renewed interest in the anti-parasitic potential of nitroaromatic drugs . Recently , we demonstrated that the anti-tubercular clinical candidate ( S ) -PA-824 possesses moderate activity against L . donovani parasites both in vitro and in vivo ( Patterson et al . , 2013 ) . Although ( R ) -PA-824 , the enantiomer of the candidate showed superior activity , this compound has not entered pre-clinical development , precluding a rapid move to a VL clinical trial . In addition , a recently reported screen of anti-tubercular nitroimidazoles against L . donovani identified DNDI-VL-2098 as a suitable compound for further preclinical evaluation ( Mukkavilli et al . , 2014; Gupta et al . , 2015 ) . The high degree of structural similarity between delamanid ( Deltyba , OPC-67683 ) and both ( R ) -PA-824 and DNDI-VL-2098 ( Figure 1 ) prompted us to investigate this nitroimidazole , which has recently received conditional approval in Europe for the treatment of multidrug-resistant tuberculosis ( Committee for Medicinal Products for Human Use , 2013; Ryan and Lo , 2014 ) . 10 . 7554/eLife . 09744 . 003Figure 1 . Chemical structures of delamanid ( ( R ) -OPC-67683 ) and the known anti-leishmanial nitroimidazoles DNDI-VL-2098 , ( R ) -PA-824 , fexinidazole and fexinidazole sulfone . Synthetic schemes for the synthesis of delamanid and analogues are described in Figure 1—figure supplements 1–2 . DOI: http://dx . doi . org/10 . 7554/eLife . 09744 . 00310 . 7554/eLife . 09744 . 004Figure 1—figure supplement 1 . Synthetic route towards delamanid ( 7 ) and ( S ) -delamanid ( 13 ) . Reagents and conditions; a ) 2-bromo-4-nitro-1H-imidazole , DIPEA , EtOAc , 65°C , 20 hr; b ) K2CO3 , MeOH , room temp . , 16 hr; c ) MsCl , pyridine , CH2Cl2 , 0°C→room temp . , 16 hr; d ) DBU , EtOAc , room temp . , 16 hr; e ) 6 , NaH , DMF , 0→50°C , 1 . 5–4 hr . DOI: http://dx . doi . org/10 . 7554/eLife . 09744 . 00410 . 7554/eLife . 09744 . 005Figure 1—figure supplement 2 . Synthetic route towards des-nitro-delamanid ( 18 ) . Reagents and conditions; a ) DIPEA , 2-nitroimidazole , EtOAc/MeCN , 77°C , 44 hr , 72%; K2CO3 , MeOH , room temp . , 20 hr , 83%; MsCl , pyridine , CH2Cl2 , 0°C→room temp . , 16 hr; DBU , EtOAc , room temp . , 16 hr , 15% over 2 steps; NaH , DMF , 0→50°C , 5 h , 30% . DOI: http://dx . doi . org/10 . 7554/eLife . 09744 . 005
The life cycle of L . donovani alternates between a flagellated promastigote form residing in the alkaline midgut of the female sandfly vector and an amastigote form that multiplies intracellularly in acidic phagolysosomes of the mammalian host macrophages . Both stages can be cultured axenically; however , intra-macrophage cultures of amastigotes are a more suitable model of mammalian infection for drug discovery . The anti-tubercular drug delamanid and its corresponding S-enantiomer were synthesized ( Appendix 1 and Figure 1—figure supplement 1 ) and assessed for anti-leishmanial activity . The potency of both compounds was determined in vitro against L . donovani ( LdBOB ) promastigotes and against intracellular amastigotes ( LV9 ) in mouse peritoneal macrophages . The ( S ) -enantiomer of delamanid showed promising anti-leishmanial activity against both developmental stages of the parasite ( EC50 values of 147 ± 4 and 1332 ± 106 nM against promastigotes and amastigotes , respectively ) . However , delamanid ( the R-enantiomer ) proved to be an order of magnitude more potent against promastigotes , axenic amastigotes and intracellular amastigotes with EC50 values of 15 . 5 , 5 . 4 and 86 . 5 nM , respectively ( Table 1 ) . Both compounds were found to be inactive ( EC50 >50 µM ) in a counter screen against the mammalian cell line HepG2 ( Table 1 ) . 10 . 7554/eLife . 09744 . 006Table 1 . Activity of delamanid against laboratory and clinical isolates of L . donovani in vitro . EC90 values are calculated from the EC50 , Hill slopes and the molecular weight of delamanid . DOI: http://dx . doi . org/10 . 7554/eLife . 09744 . 006SpeciesDevelopmental stageEC50 , nM ( Hill slope ) EC90 , nMEC90 , ng ml-1Leishmania donovani ( LdBOB ) Promastigote15 . 5 ± 0 . 07 ( 8 . 4 ) 20 . 210 . 8Leishmania donovani ( LdBOB ) Amastigote ( axenic ) 5 . 4 ± 0 . 05 ( 5 . 3 ) 8 . 24 . 4Leishmania donovani ( LV9 ) Amastigote ( in macrophage ) 86 . 5 ± 1 . 7 ( 2 . 3 ) 225120Leishmania donovani ( DD8 ) Amastigote ( in macrophage ) 298 ± 13 ( 2 . 7 ) 672359Leishmania donovani ( BHU1 ) Amastigote ( in macrophage ) 230 ± 10 ( 4 . 1 ) 393210Leishmania donovani ( SUKA001 ) Amastigote ( in macrophage ) 259 ± 7 ( 3 . 6 ) 476254L . infantum ( ITMAP263 ) Amastigote ( in macrophage ) 940 ± 0 . 05 ( 3 . 4 ) 1790955Human ( HepG2 ) N/A>5000-- Future anti-leishmanial therapies will be required to demonstrate a broad spectrum of activity against different Leishmania strains and against drug resistant parasites ( Patterson and Wyllie , 2014 ) . With this in mind , L . donovani and L . infantum clinical isolates were assessed for their sensitivity to delamanid ( Table 1 ) . These included: the Indian WHO reference strain DD8; an Indian antimony resistant isolate BHU1; a recent Sudanese isolate SUKA 001; and the L . infantum strain ITMAP263 from Morocco . These clinical isolates were marginally less sensitive to delamanid than our laboratory strain LV9 from Ethiopia , but at the EC90 varied by only 3-fold ( L . donovani ) or 8-fold ( L . infantum ) ( Table 1 ) . Although not investigated further here , promastigotes of L . major Friedlin , a cause of cutaneous leishmaniasis , were also highly sensitive to delamanid ( EC50 6 . 3 ± 0 . 11 nM , slope factor 2 . 2 ) . The corresponding des-nitro analogue was also synthesized ( Appendix 1 and Figure 1—figure supplement 2 ) and assayed against L . donovani promastigotes . Des-nitro-delamanid was found to be inactive ( EC50 >50 µM ) , which is consistent with the nitro group being involved in the mechanism of action or having a role in the binding of delamanid to its molecular target ( s ) in L . donovani . The plasma protein binding of delamanid was measured and found to be high ( Fu = 0 . 0045 ) , in agreement with that reported previously ( Committee for Medicinal Products for Human Use , 2013 ) . A kinetic solubility assay demonstrated that delamanid possesses sufficient aqueous solubility ( >250 µM in 2 . 5% DMSO ) for use in in vitro assays . The efficacy of delamanid was assessed in a murine model of VL . Groups of infected BALB/c mice ( seven days post infection with ex vivo L . donovani LV9 amastigotes ) were dosed twice-daily , for five consecutive days with an oral formulation of delamanid ( 1 , 3 , 10 , 30 or 50 mg kg-1 ) . On day 14 post-infection , the parasite burdens in the livers of infected mice were determined and compared with those of control animals . The only current oral anti-leishmanial therapy miltefosine ( 30 mg kg-1 , once-daily , 5 days ) was included as a positive control . Both delamanid and miltefosine were well tolerated at these doses , with no mice displaying any overt signs of toxicity . An initial experiment showed that treatment with delamanid at 50 mg kg-1 effectively cured the study mice , with no detectable parasites in the liver smears , whereas control mice dosed with vehicle alone showed a high level of infection ( Figure 2 ) . A second in vivo study with mice dosed twice-daily at 30 , 10 or 3 mg kg-1 suppressed infection in the murine model by 99 . 5% , 63 . 5% and 16 . 0% , respectively , establishing a dose-dependent anti-leishmanial effect within the range of 3–50 mg kg-1 . These results give an estimated ED50 and ED90 of 7 . 3 and 21 . 5 mg kg-1 , respectively ( Figure 2—figure supplement 1 ) . At 30 and 50 mg kg-1 delamanid compares favourably with miltefosine ( 98 . 8–99 . 8% suppression at 30 mg kg-1 ) , which exemplifies the therapeutic potential of delamanid . 10 . 7554/eLife . 09744 . 007Figure 2 . Effects of drug treatment on the parasite burden of mice infected with L . donovani . Groups of mice ( five per group ) infected with L . donovani ( strain LV9 ) were dosed with drug vehicle ( orally ) , miltefosine ( orally ) or delamanid ( twice daily , orally ) on day 7 post-infection and for a total of 5 or 10 days . Two days after the final dose , animals were humanely euthanized and parasite burdens were determined microscopically by examining Giemsa-stained liver smears . Grey bars , 5-day treatment; red bars , 10 day treatment . This graph shows the combined data from six individual animal studies; n = 9 , or 10 for 1 , 3 and 10 mg/kg dosing; for all other experiments n = 5 . These data are available in tabular form in Figure 2—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 09744 . 00710 . 7554/eLife . 09744 . 008Figure 2—source data 1 . Efficacy and PK/PD data from all experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 09744 . 00810 . 7554/eLife . 09744 . 009Figure 2—figure supplement 1 . ED50 determination for delamanid in a mouse model of VL . Results are based on dosing in the range 3 to 50 mg kg-1 b . i . d . oral for 5 days . DOI: http://dx . doi . org/10 . 7554/eLife . 09744 . 009 A third in vivo study with a further reduced delamanid dose of 1 mg kg-1 resulted in a suppression of parasitaemia of 86 . 3% compared with control mice , proving unexpectedly superior to dosing at 3 or 10 mg kg-1 ( Figure 2 ) . A subsequent experiment encompassing a range of doses ( 10 , 3 , 1 mg kg-1 , 5 days ) in a single study showed a similar hormetic effect , with twice daily dosing at 1 mg kg-1 being more efficacious than 10 mg kg-1 . However , this study also demonstrated that there is some variability in the efficacy of delamanid at lower doses ( Figure 2—source data 1 ) . The hormetic effect was also observed in an extended dosing experiment in which delamanid was instead dosed twice-daily for 10 days at 10 , 3 or 1 mg kg-1 , with the suppression of infection being 92 . 3% , 24 . 3% and >99 . 9% , respectively . A second 10-day experiment with a broader range of doses ( 30 , 10 , 3 , 1 , 0 . 3 mg kg-1 ) further confirmed the hormetic effect . In addition , this study demonstrated that further reducing the delamanid dose ( 0 . 3 mg kg-1 ) resulted in a reduction in efficacy comparable to dosing at 3 mg kg-1 , resulting in a biphasic dose response relationship ( Figure 2 ) . It is important to understand the pharmacokinetic and pharmacodynamic ( PK/PD ) behaviour of delamanid in order to optimise the efficacious dosing regimen ( Velkov et al . , 2013 ) . By measuring the change in drug concentration over time in L . donovani-infected mice , two standard PK parameters can be obtained: maximum concentration ( Cmax ) in blood; and the area under the curve ( AUC ) , a measure of total drug exposure over time . The drug concentration over time is measured in order to determine whether the concentration of a drug exceeds the minimum inhibitory concentration ( MIC , EC90 in this case ) and , if so , for how long ( time over MIC , T>MIC ) . Parameters such as Cmax/MIC , AUC/MIC and T>MIC are important for achieving drug efficacy in an in vivo model of disease . Both Cmax and AUC measure the total drug level in blood or plasma; however , only unbound drug molecules are able to bind to their targets ( Bohnert and Gan , 2013 ) . Therefore , the plasma protein binding level ( expressed as the fraction unbound , Fu ) of delamanid was also measured and used to calculate an adjusted EC90 ( assay EC90 × 1/Fu ) for comparison with blood concentration over time . Accordingly , the blood levels of the drug were measured at intervals ( up to 8 hr post dose ) during the in vivo efficacy studies . Data for the first and ninth dose in a 5-day twice daily treatment experiment ( Figure 3A , B ) show a dose-dependent response with accumulation over time . A similar effect was noted in a 10-day study ( 1st and 19th dose; Figure 3C , D ) . More detailed analysis of the combined PK data from five experiments ( including two 10-day treatment studies ) shows a linear relationship between doses of 0 . 3–10 mg kg-1 and peak blood concentration ( Cmax ) or area under the curve ( AUC ( 0-t ) ) with accumulation from day 1 through to day 10 ( Figure 3E , F ) . The 10 and 30 mg kg-1 doses should provide adequate coverage over the EC90 ( 120 ng ml-1 ) as measured for the L . donovani isolate LV9 in macrophages over a 3 day exposure ( Table 1 ) . However , due to high protein binding , the free fraction ( Fu = 0 . 0045 ) cannot account for biological activity in vivo at any dose . An explanation for why the free drug theory ( Bohnert and Gan , 2013 ) is not applicable in this case is presented below . 10 . 7554/eLife . 09744 . 010Figure 3 . Pharmacokinetic behaviour of delamanid in infected mice . ( A and B ) show the blood levels of delamanid following the first oral dose on day 1 ( A ) and the penultimate oral dose on day 5 ( B ) for 1 , 3 , 10 and 30 mg kg-1 b . i . d . ( teal , black , red and blue symbols , respectively ) . Error bars are SEM ( n = 3 for 30 mg kg-1 , n = 6 all other doses ) . ( C and D ) show the blood levels of delamanid from a single VL PK/PD study following the first oral dose on day 1 ( C ) and the penultimate oral dose on day 10 ( D ) for 0 . 3 , 1 , 3 , 10 and 30 mg kg-1 b . i . d . ( grey , teal , black , red and blue symbols , respectively ) . Error bars are SEM ( n = 5 ) . ( E and F ) show the relationship between dose with Cmax or AUC ( 0-8 h ) , respectively , after the first oral dose on day 1 ( black ) , or the penultimate dose on day 5 ( red ) or day 10 ( blue ) . Error bars in ( E ) are SEM ( n ranges from 3–8 depending upon day and dose – see Figure 2 – source data1 ) . Lines in ( E ) and ( F ) are best fits by linear regression . DOI: http://dx . doi . org/10 . 7554/eLife . 09744 . 010 To determine whether delamanid was cytostatic or cytotoxic , mid-log promastigotes were incubated with drug concentrations equivalent to 10 times the EC50 value ( Figure 4A ) . Growth of drug-treated cultures ceased almost immediately with cell numbers declining after 8 hr and no live parasites visible at 24 hr . To determine the actual point where treated cells lost viability , at defined intervals parasites were washed and sub-cultured without drug . No viable parasites could be recovered after 12 hr in the presence of drug , confirming that delamanid is rapidly leishmanicidal . In support of this apparent rapid mechanism of cell killing , EC50 values determined after 24 , 48 and 72 hr were essentially identical ( Figure 4B ) . In addition , the potency ( EC50 value ) of delamanid was found to be dependent on the initial cell density ( Figure 4C ) and on the assay serum concentration ( Figure 4D ) . 10 . 7554/eLife . 09744 . 011Figure 4 . Effects of delamanid on L . donovani promastigotes . ( A ) Delamanid causes rapid cell killing . Promastigotes were exposed to delamanid ( 10 times EC50 ) and samples removed at intervals to determine cell density and cell viability . Black symbols: no inhibitor; red symbols: plus drug; the point of irreversible drug toxicity . ( B ) Drug sensitivity is independent of exposure beyond 24 hr . Black , red and blue symbols are EC50 curves determined after 24 , 48 and 72 hr , respectively . ( C ) Drug sensitivity is cell-density dependent . Black , red and blue symbols are EC50 curves determined after 72 hr , with initial seeding densities of 103 , 104 and 105 cells ml-1 , respectively . ( D ) Drug sensitivity is serum dependent . Black , red and blue symbols are EC50 curves determined after 72 hr in the presence of 5 , 10 and 20% FCS , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 09744 . 011 Many nitroheterocyclics require bio-activation of their nitro groups to become biologically active . In Mycobacterium tuberculosis , delamanid is assumed to be reductively activated by the same unusual deazaflavin ( F420 ) -dependent nitroreductase ( Ddn ) known to activate the closely related nitroimidazo-oxazine drug PA-824 ( Manjunatha et al . , 2006; Singh et al . , 2008; Manjunatha et al . , 2009 ) . In the absence of a Ddn homologue in Leishmania , we assessed whether the reduction of delamanid is catalysed by the NADH-dependent bacterial-like nitroreductase ( NTR ) already shown to activate the nitroimidazoles fexinidazole and nifurtimox in these parasites ( Wyllie et al . , 2012 ) . The potency of delamanid was determined against parasites overexpressing NTR . Increased concentrations of NTR in these transgenic parasites were confirmed by a 13-fold increase in their sensitivity to nifurtimox ( EC50 of 8 . 0 ± 0 . 2 and 0 . 61 ± 0 . 006 μM for WT and transgenic parasites , respectively Figure 5A ) , known to undergo two-electron reduction by NTR ( Hall et al . , 2011 ) . However , overexpression of NTR in promastigotes did not significantly alter their sensitivity to delamanid ( EC50 of 4 . 5 ± 0 . 004 and 4 . 1 ± 0 . 003 nM for WT and transgenic parasites , respectively ) ( Figure 5B ) . To confirm that the same was also true in the amastigote stage of these parasites , metacyclic promastigotes overexpressing NTR were used to infect mouse peritoneal macrophages . The resulting intracellular parasites were found to be just as sensitive to delamanid as WT parasites with EC50 values of 57 . 8 ± 2 . 1 and 55 . 2 ± 4 . 3 nM , respectively ( Figure 5C ) . These findings indicate that NTR does not play a role in the activation of delamanid in L . donovani in either stage of the life cycle and that the mechanism of action of this nitroheterocyclic drug is different from that of fexinidazole . 10 . 7554/eLife . 09744 . 012Figure 5 . Mode of action of delamanid is distinct from nifurtimox . ( A , B and C ) Susceptibility to nifurtimox ( A ) is increased in NTR-overexpressing promastigotes ( red symbols ) , but not to delamanid ( B ) compared to WT cells ( black symbols ) . Data are the mean of triplicate cultures from a single experiment . ( C ) The susceptibility of delamanid is not increased in NTR-overexpressing amastigotes in macrophages ( red symbols ) compared to WT ( black symbols ) . Data are the mean of duplicate cultures from a single experiment . DOI: http://dx . doi . org/10 . 7554/eLife . 09744 . 012 Given that NTR does not activate delamanid in L . donovani promastigotes and the requirement of the nitro group for biological activity , it was important to determine if the drug is metabolised in culture . To address this issue , the concentration of delamanid was determined by UPLC-MS/MS in cultures of promastigotes over a 24 hr period . Delamanid is known to be primarily metabolised in plasma by albumin ( Shimokawa et al . , 2015 ) and to a lesser extent by CYP3A4 , CYP1A1 , CYP2D6 and CYP2E1 ( Sasahara et al . , 2015 ) . Thus , the concentration of delamanid in culture medium without parasites was measured over the same time period as a control . In the presence of medium alone , delamanid decreased linearly in a concentration-dependent manner ( Figure 6A ) . However , in the presence of L . donovani promastigotes the rate of disappearance of delamanid was markedly increased , such that the drug had essentially disappeared by 6 hr ( Figure 6B ) . The net amount of delamanid metabolised by parasites as a function of time is also linear and dependent on the initial concentration in the medium ( Figure 6C ) . Linear regression of these data revealed that the rate of cell metabolism is not saturated up to the top concentration tested ( Figure 6D ) . Analogous experiments using mouse peritoneal macrophages and THP-1 monocytes found no evidence of delamanid metabolism by these host cell lines . Elucidation of the chemical identity of the delamanid metabolite ( s ) , their possible role in parasite killing and the enzyme ( s ) responsible for their biosynthesis will be the focus of future studies . 10 . 7554/eLife . 09744 . 013Figure 6 . Delamanid metabolism in L . donovani promastigotes . ( A ) Medium plus delamanid alone and ( B ) cells incubated in medium plus delamanid . Delamanid concentrations added are 15 , 45 and 150 nM ( black , red and blue , respectively ) . The lines represent best fits by linear regression for all data points in ( A ) and 0 to 5 hr in ( B ) . The dotted line in ( B ) is the best fit by non-linear regression to a single exponential decay . ( C ) Net metabolism of delamanid by cells was obtained by subtraction of ( A ) from ( B ) . Data fitted by linear regression gave correlation coefficients of 0 . 996 , 0 . 991 and 0 . 951 for delamanid concentrations of 15 , 45 and 150 nM , respectively . ( D ) Rates of delamanid metabolism obtained from ( C ) are linear up to 150 nM ( correlation coefficient 0 . 996 , explicit errors used in fit ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09744 . 013
For diseases of poverty such as visceral leishmaniasis there is limited financial incentive to initiate expensive , high risk and time-consuming de novo drug discovery programmes . Consequently , the repurposing of existing drugs has become an attractive approach towards the identification of much needed new treatments for VL and other neglected parasitic diseases ( Andrews et al . , 2014; Wyllie et al . , 2012 ) . The recently approved anti-tubercular drug delamanid ( Ryan and Lo , 2014 ) was deemed to be of particular interest as a number of other nitroimidazoles have been shown to also possess promising anti-leishmanial activities ( Wyllie et al . , 2012; Patterson et al . , 2013; Gupta et al . , 2015 ) . In the current study , we show that delamanid is highly active in vitro against intracellular L . donovani amastigotes ( EC50 0 . 087 µM ) with activity superior to that of both the current VL drug miltefosine ( EC50 3 . 3 µM ) and the active sulfone metabolite of the VL clinical candidate fexinidazole ( EC50 5 . 3 µM ) in the same assay ( Wyllie et al . , 2012 ) . The in vitro anti-leishmanial activity of delamanid shows the same enantiomeric specificity as the delamanid analogue DNDI-VL-2098 in vivo ( Gupta et al . , 2015 ) . Similarly , the R-enantiomer of an analogue of delamanid has been shown to be more potent against M . tuberculosis than its corresponding S-enantiomer ( Sasaki et al . , 2006 ) . However , this contrasts with the closely related nitroimidazole PA-824 , where the enantiomeric specificity for L . donovani and M . tuberculosis is opposite ( Patterson et al . , 2013 ) . Delamanid is rapidly leishmanicidal with a cell-density dependent potency and as expected , demonstrated no observable toxicity in a mammalian cell assay . The observed shift in drug sensitivity with increasing serum concentration is likely due to increased metabolism by albumin to inactive metabolites , rather than changes in free drug concentration due to protein binding . Delamanid is orally bioavailable , well tolerated , shows dose linearity up to 10 mg kg-1 and accumulation with repeated administration in agreement with previous pharmacokinetic ( PK ) studies in mice ( Matsumoto et al . , 2006 ) . However , the pharmacokinetic / pharmacodynamic ( PK/PD ) relationship is not straightforward . First , the PK/PD relationship does not fit with the free drug hypothesis , which states that , in the absence of energy-dependent transport processes , the extracellular and intracellular free drug concentrations are equal after steady-state equilibrium has been achieved , and that only the free drug is able to bind to the target to exert its pharmacological effect ( Bohnert and Gan , 2013; Smith et al . , 2010 ) . Based on the in vitro intra–macrophage amastigote EC90 value ( 120 ng ml-1 ) , the high PPB of delamanid ( Fu 0 . 0045 ) results in an adjusted intra macrophage L . donovani EC90 of 26 , 700 ng ml-1 , a concentration that is not achieved in whole blood at any point in the efficacy study . Therefore , the efficacy of delamanid in the VL animal model is unexpected . However , there are exceptions to the free drug theory , such as drugs that form active metabolites resulting in inactivation ( covalent or otherwise ) of multiple targets ( Smith et al . , 2010 ) . The parasite-specific metabolism presented here is entirely consistent with this exception . Second , the pronounced bi-phasic suppression of parasite burden at high and low doses of delamanid in vivo is highly unusual . This U-shaped dose-response curve is reminiscent of hormesis in toxicology ( Calabrese and Baldwin , 2003; 2002 ) , but , to our knowledge , is unprecedented in microbiology . This effect is not observed in dose response curves in vitro , so must be related to a physiological or metabolic threshold response in the infected drug-treated animal . In vivo delamanid is proposed to undergo primary metabolism with loss of the nitro group , mainly catalysed by albumin ( Committee for Medicinal Products for Human Use , 2013; Shimokawa et al . , 2015 ) . Further metabolism to seven other metabolites is thought to occur via hydrolysis reactions and oxidation by CYP3A4 ( Committee for Medicinal Products for Human Use , 2013; Sasahara et al . , 2015 ) . The cause of this U-shaped dose-response curve is currently not known . One possibility is that a drug metabolite of delamanid antagonises the bio-activation of delamanid , or antagonises the downstream effects in the leishmania parasite . The formation of a putative antagonist metabolite would have to show a saturable sigmoidal dose response , such that at higher concentrations , delamanid , or an active parasite-specific metabolite thereof , are able to displace the antagonist from the bio-activating enzyme , or proteins related to the downstream effect respectively . It should be noted that in the same VL animal model the related nitroimidazole ( S ) -PA-824 was also more efficacious at a lower dose ( 30 vs 100 mg kg-1 ) ( Patterson et al . , 2013 ) . Further studies with ( S ) -PA-824 should be conducted to determine if this compound also displays a hormetic PK/PD relationship and establish if this is a chemotype-related characteristic . Plots of Cmax versus parasite suppression and calculated AUC ( 0–24 hr ) versus parasite suppression ( Figure 7A , B ) suggest that the delamanid blood levels required for cure in the VL model exceed those observed in TB patients receiving the drug . Increasing the dosing duration in the in vivo VL model from 5 to 10 days improved the mean parasite suppression at all investigated doses ( Figure 2 ) and resulted in some mice with no detectable liver parasites when dosed at 1 mg kg-1 ( Figure 2—source data 1 ) . As it known that delamanid is tolerated for up to six months ( Committee for Medicinal Products for Human Use , 2013 ) further extending the duration of the VL model beyond 10 days should be considered . The VL target product profile calls for a treatment regimen of <10 days . However , in the current VL therapy , miltefosine is dosed orally for 28 days , so extended dosing should be clinically acceptable . Note , the study mice dosed twice-daily at 1 mg kg-1 have drug levels lower than that achieved in human TB patients dosed once daily at 100 mg . Given that this low dose is more efficacious than any other dose below 30 mg kg-1 , model studies of extended duration should focus around this dosing level . 10 . 7554/eLife . 09744 . 014Figure 7 . PK/PD relationships in mice . ( A and B ) Mean suppression of parasite burden as a function of Cmax for the penultimate dose ( panel A ) and extrapolated AUC ( 0–24 hr ) for the last day of the 10-day treatment regimen ( panel B ) . The black dotted line in ( A ) is the EC90 value obtained for infected macrophages after 72 hr exposure ( 120 ng ml-1 ) . The red dotted line in ( A ) and ( B ) represents the mean delamanid Cmax ( 375–400 ng ml-1 ) and mean AUC ( 0–24 hr ) ( 7000–8000 h*ng ml-1 ) obtained in 144 TB patients after 14 days treatment with 100 mg , oral , once daily from day 14–56 ( Sasahara et al . , 2015 ) . The data in this graph were derived from a single in vivo study; related aggregated data from previous studies is shown in Figure 7—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 09744 . 01410 . 7554/eLife . 09744 . 015Figure 7—figure supplement 1 . ( A and B ) Mean suppression of parasite burden as a function of Cmax for the penultimate dose and extrapolated AUC ( 0–24 hr ) for the last day of the 5- and 10-day treatment regimens , respectively . Open and closed symbols represent data combined from the 5- and 10-day studies , respectively . This graph represents the aggregated data from 3 separate in vivo studies . The black dotted line in ( A ) is the EC90 value obtained for infected macrophages after 72-hr exposure . The red dotted line in ( A ) and ( B ) represents the mean delamanid Cmax ( 375–400 ng ml-1 ) and mean AUC ( 0–24 hr ) ( 7000–8000 h*ng ml-1 ) obtained in TB patients after 14 days treatment with 100 mg , oral , once daily ( Sasahara et al . , 2015 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09744 . 015 Despite a wealth of pharmacokinetic data in patients and human volunteers ( Committee for Medicinal Products for Human Use , 2013 ) the unusual PK/PD relationship hinders our ability to accurately predict the outcome of delamanid dosing in VL patients . Indeed , careful examination of parasite suppression versus Cmax ( Figure 7—figure supplement 1 ) shows that the mean Cmax in delamanid-treated TB patients corresponds to an efficacy minimum in the VL model . Given that delamanid is rapidly metabolised by leishmania-infected macrophages in vitro , we examined the effect of delamanid with a 5 day exposure with daily drug and medium change . This gave an EC50 value of 28 . 0 ± 1 . 6 nM ( slope 2 . 7 ) from which an EC90 could be calculated ( 62 . 7 nM or 33 . 5 ng ml-1 ) . This concentration is lower than the lowest observed Cmax in the efficacy studies ( Figure 7A ) . Thus , careful design of the dosing regimen for VL patients may avoid the risk that treatment will lack efficacy due to reaching a Cmax and AUC ( 0–24 hr ) within the higher ineffective concentration range . The nature of the parasite-specific metabolising / activating enzyme ( s ) is not known , but is clearly distinct from the deazaflavin-dependent nitroreductase in M . tuberculosis ( Manjunatha et al . , 2006 ) and the nitroreductase in leishmania involved in the activation of fexinidazole metabolites ( Wyllie et al . , 2012 ) . The identification of this target , and the metabolites that it produces , are the focus of our current work . The cell-density dependent potency of delamanid is consistent with the formation of a putative reactive , covalent metabolite . In addition , the rapidly cytocidal activity of delamanid is consistent with the rapid rate of drug metabolism by L . donovani in culture . In terms of drug development the divergent modes of action for fexinidazole and delamanid are advantageous , as the likelihood of cross-resistance developing is reduced , and the potential for their co-administration as a combination therapy is retained . The current practice in the pharmaceutical industry is to avoid developing compounds containing a nitro-aromatic group due to the known liabilities of this class , particularly potential mutagenicity and carcinogenicity . As a result , outside the anti-infectives there are relatively few nitro-aromatic drugs . Indeed , nitro-aromatic moieties are relatively common in drugs or clinical candidates for kinetoplastid diseases compared to chemotherapies for other indications . This over-representation is linked to the mechanism of action of nitro-drugs; selective bio-activation by parasitic bacterial-like NTRs leading to selective anti-parasitic activity . The studies presented herein are consistent with delamanid also being activated by a parasite-specific enzyme absent from host cells . Preclinical studies have demonstrated that delamanid is not mutagenic ( Matsumoto et al . , 2006 ) . Moreover , repeated oral administration in mice or rats for up to 104 weeks showed no evidence of carcinogenicity ( Committee for Medicinal Products for Human Use , 2013 ) . Taken together these points alleviate some of the concerns ordinarily associated with the development of nitro-drugs , although long term safety can only be established after extensive clinical use in relevant populations . While revising this manuscript , Thompson and co-workers reported the structure-activity relationships for an extensive series of bicyclic nitro-compounds , including delamanid ( Thompson et al . , 2016 ) . These authors observed partial efficacy in L . infantum-infected hamsters dosed with delamanid , not inconsistent with our findings . The lower efficacy observed by these authors could be due to one or more of the following: inadequate dosing ( once versus twice per day ) ; different animal model ( hamster versus mouse ) ; or different species of leishmania ( L . infantum is less sensitive than L . donovani ) . Although this paper found VL-2098 to have superior efficacy in a once daily treatment regimen , pre-clinical development of this compound has been abandoned due to testicular toxicity ( http://www . dndi . org/diseases-projects/portfolio/completed-projects/vl-2098/ last accessed 3rd April 2016 ) . Importantly , this reproductive toxicity has not been observed with delamanid ( Committee for Medicinal Products for Human Use , 2013 ) . Delamanid meets many of the criteria specified in the target product profile ( TPP ) for VL [http://www . dndi . org] . It is rapidly cytocidal and thus potentially efficacious in immunocompromised patients such as those co-infected with HIV ( Alvar et al . , 2008 ) particularly since delamanid is not associated with any clinically relevant anti-retroviral drug-drug interactions ( Ryan and Lo , 2014; Blair and Scott , 2015 ) . Due to the prevalence of TB-VL coinfection in Ethiopia ( Hurissa et al . , 2010 ) and Sudan ( El-Safi et al . , 2004 ) the TPP also specifies that any new treatments should be compatible with TB medications , a stipulation met by delamanid ( Blair and Scott , 2015 ) . Delamanid can be administered orally , an important requirement for patients who have limited access to even the most basic of health care facilities , whereas liposomal amphotericin B ( AmBisome ) has to be administered intravenously and requires cold storage for stability . Perhaps the most challenging issue may be cost of goods . Assuming an efficacious dose of 100 mg once-daily for in excess of 10 days and the market cost of delamanid in developed countries ( US$42 per 50 mg tablet ) ( Lessem , 2014 ) , the predicted cost per patient would be over US$840 , significantly higher than that specified in the VL TPP ( <US$10 or <$80 per course ) and predicted to be more expensive than other current treatment strategies ( Meheus et al . , 2010 ) . However , differential pricing or financial arrangements may reduce the cost for this neglected disease of poverty . Additional factors also support the development of delamanid as a VL therapy . Importantly , the results presented here demonstrate that delamanid is active against a number of clinically-relevant field isolates . In addition , we are currently investigating the potential of delamanid to be repurposed for Chagas’ disease . In summary , these data suggest that delamanid has the potential to be repurposed as a VL therapy . Additional VL animal model studies exploring the effect of extended delamanid dosing beyond 10 days should be investigated .
All leishmania strains follow the WHO International Code designating the animal from which the parasite was isolated , country , date of isolation and strain designation ( see International Leishmania Network http://leishnet . net/site/ ? q=node/5 ) . The WHO designations and origins of each Leishmania isolate used are detailed in Table 2 . The clonal Leishmania donovani cell line LdBOB was grown as promastigotes at 24°C in 10% FCS , as previously described ( Goyard et al . , 2003 ) , except when investigating the effect of serum concentration on drug efficacy , in which case 5 , 10 , or 20% FCS was used . Transgenic LdBOB promastigotes expressing the L . major nitroreductase ( LmjF . 05 . 0660 ) enzyme ( Wyllie et al . , 2012 ) were cultured under identical conditions in the presence of nourseothricin ( 100 μg ml−1 ) . L . major promastigotes ( Friedlin strain ) were grown in M199 medium ( Caisson Laboratories , Logan , UT ) with supplements , as previously described ( Oza et al . , 2005 ) . L . donovani ( LV9 strain ) ex vivo amastigotes were used in both in vitro and in vivo drug sensitivity assays . Amastigotes were derived from hamster spleens , as previously described ( Wyllie and Fairlamb , 2006 ) . All other Leishmania clinical isolates ( Table 2 ) were grown in RPMI 1640 ( Sigma , UK ) supplemented with 20% FCS , 100 µM adenine , 5 µM hemin , 20 mM MES , 3 µM 6-biopterin and 1 mM biotin . In all cases the FCS used was certified as mycoplasma free . 10 . 7554/eLife . 09744 . 016Table 2 . Leishmania isolates used in delamanid drug sensitivity studies . DOI: http://dx . doi . org/10 . 7554/eLife . 09744 . 016SpeciesWHO CodeLaboratory CodeOriginYear of IsolationProvided byL . donovaniMHOM/SD/62/1S CL2D aLdBOBSudan1962via Professor Stephen Beverley , Washington UniversityL . donovaniMHOM/ET/67/HU3LV9Ethiopia1967via Professor Jennie Blackwell , Cambridge UniversityL . donovaniMHOM/IN/02/BHU1BHU1India2002via LSHTM , London , UKL . donovaniMHOM/SU/09/SUKA001SUKA001Sudan2009via LSHTM , London , UKL . donovaniMHOM/IN/80/DD8DD8India1980via LSHTM , London UKL . infantumMHOM/MA/67/ITMAP263ITMAP263Morocco1967via LSHTM , London , UKL . majorMHOM/IL/81/FriedlinFriedlinIsrael1981via LSHTM , London , UKa Derived from this strain ( Goyard et al . , 2003 ) Delamanid was prepared as previously described ( Sasaki et al . , 2006; Kiyokawa and Aki , 2005 ) ( Figure 1—figure supplement 1 ) . Modification of the delamanid synthetic route afforded ( S ) -delamanid and des-nitro-delamanid ( Figure 1—figure supplements 1–2 ) . Compound purity was determined by liquid chromatography-mass spectrometry , with all compounds found to be >95% pure . For in vivo experiments , delamanid was further analysed by ultra high-performance liquid chromatography-mass spectrometry ( UPLC-MS ) , with all batches found to be of >98% purity . The optical rotation of delamanid was in close agreement to the published value ( Sasaki et al . , 2006 ) , confirming the optical purity of the material used in this study . Detailed synthetic procedures and analysis of key compounds and intermediates are provided in Appendix 1 . To examine the effects of test compounds on growth , triplicate cultures were seeded with 1 × 105 parasites ml-1 . Parasites were grown in the presence of drug for 72 hr , after which 50 μM resazurin was added to each well and fluorescence ( excitation of 528 nm and emission of 590 nm ) measured after a further 2 hr incubation ( Jones et al . , 2010 ) . Data were processed using GRAFIT ( version 5 . 0 . 13; Erithacus software ) and fitted to a 2-parameter equation , where the data are corrected for background fluorescence , to obtain the effective concentration inhibiting growth by 50% ( EC50 ) :y=1001+ ( [I]EC50 ) m In this equation [I] represents inhibitor concentration and m is the slope factor . Experiments were repeated at least two times and the data is presented as the weighted mean plus weighted standard deviation ( Young , 1962 ) . When investigating the speed of drug-mediated cell killing , parasites were grown in the presence of drug for 24 , 48 , or 72 hr in an otherwise identical assay . The same assay was used to investigate the effect of seeding density upon drug efficacy , except that the number of parasites used to seed the assays was varied to be either 103 , 104 or 105 parasites ml-1 . Delamanid was added to early-log cultures of LdBOB promastigotes ( ~1 × 106 ml-1 ) at concentrations equivalent to 10 times its EC50 value . At intervals , the cell density was determined , samples of culture ( 500 µl ) removed , washed and resuspended in fresh culture medium in the absence of drug . The viability of drug-treated parasites was monitored for up to 24 hr and the point of irreversible drug toxicity determined by microscopic examination of subcultures after 5 days . In-macrophage drug sensitivity assays were carried out using starch-elicited mouse peritoneal macrophages and hamster-derived ex vivo amastigotes ( Wyllie et al . , 2012 ) or metacyclic promastigotes ( Wyllie et al . , 2013 ) , where appropriate . Assays to determine the sensitivity of HepG2 cells to test compounds were carried out precisely as previously described ( Patterson et al . , 2013 ) . HepG2 were obtained from ATCC and routinely tested for mycoplasma contamination by Mycoplasma Experience Ltd . The PPB of delamanid was determined by the equilibrium dialysis method ( Jones et al . , 2010 ) . The aqueous solubility of delamanid was measured using a laser nephelometry-based method ( Patterson et al . , 2013 ) . Groups of female BALB/c mice ( 5 per group ) were inoculated intravenously ( tail vein ) with approximately 2 × 107 L . donovani LV9 amastigotes harvested from the spleen of an infected hamster ( Wyllie and Fairlamb , 2006 ) . From day 7 post-infection , groups of mice were treated with either drug vehicle only ( orally ) , with miltefosine ( 30 mg kg-1 orally ) , or with delamanid ( 1 , 3 , 10 , 30 or 50 mg kg-1 orally ) . Miltefosine was administered once daily for 5 , or 10 days , with vehicle and delamanid administered twice daily over the same period . Drug dosing solutions were freshly prepared each day , and the vehicle for delamanid was 0 . 5% hydroxypropylmethylcellulose , 0 . 4% Tween 80 , 0 . 5% benzyl alcohol , and 98 . 6% deionized water . On day 14 ( for 5 day dosing experiments ) , or day 19 post-infection ( for 10 day dosing experiments ) , all animals were humanely euthanized and parasite burdens were determined by counting the number of amastigotes/500 liver cells ( Wyllie et al . , 2012 ) . Parasite burden is expressed in Leishman Donovan Units ( LDU ) : mean number of amastigotes per 500 liver cells × mg weight of liver ( Bradley and Kirkley , 1977 ) . The LDU of drug-treated samples are compared to that of untreated samples and the percent inhibition calculated . ED50 values were determined using GRAFIT ( version 5 . 0 . 13; Erithacus software ) by fitting data to a 2-parameter equation , as described above . Blood samples ( 10 μl ) from 3 of 5 infected mice ( see in vivo drug sensitivity above ) in each dosing group were collected from the tail vein and placed into Micronic tubes ( Micronic BV ) containing deionized water ( 20 μl ) . Samples were taken following the first dose on the first ( day 7 post-infection ) and last day of dosing ( day 11 , or 16 post-infection ) at 0 . 5 , 1 , 2 , 4 and 8 hr post-dose . Diluted blood samples were freeze-thawed three times prior to bioanalysis . The concentration of delamanid in mouse blood was determined by UPLC-MS/MS on a Xevo TQ-S ( Waters , UK ) by modification of that described previously for the analysis of fexinidazole ( Sokolova et al . , 2010 ) and PK parameters determined using PKsolutions software ( Summit , USA ) . AUC ( 0–24 hr ) was extrapolated from the calculated AUC ( 0-8 hr ) , with second daily dose administered at 8 hr post first daily dose . Rate of metabolism studies were carried out at 15 , 45 and 150 nM delamanid ( equivalent to 1- , 3- and 10-times EC50 ) in culture medium alone and in the presence of wild type L . donovani promastigotes ( 1 × 107 parasites ml-1 ) . At 0 , 0 . 5 , 1 , 2 , 4 , 6 , 8 and 24 hr aliquots were removed , precipitated by addition of a 3-fold volume of acetonitrile and centrifuged ( 1665 × g , 10 min , room temperature ) . The supernatant was diluted with water to maintain a final solvent concentration of 50% and stored at −20ºC prior to UPLC-MS/MS analysis , as described below . UPLC-MS/MS was performed on a Waters Acquity UPLC interfaced to a Xevo TQ-S MS . Chromatographic resolution was achieved on a 2 . 1 × 50 mm Acquity BEH C18 , 1 . 7 µm column which was maintained at 40ºC with an injection volume of 8 µl . The mobile phase consisted of A: deionized water plus 0 . 01% ( v/v ) formic acid and B: acetonitrile plus 0 . 01% ( v/v ) formic acid at a flow rate of 0 . 6 ml min-1 . The initial gradient was 5% B held for 0 . 5 min before increasing to 95% B from 0 . 5–2 min , where it was held from 2–2 . 6 min before decreasing back to 5% B from 2 . 6–3 min . Mass spectra were obtained using electrospray ionization ( ESI ) , in positive ion mode with the following conditions: capillary 3 . 5 kV; desolvation temperature 600ºC; source temperature 150°C; desolvation gas flow ( nitrogen ) 1000 l h-1 and collision gas ( argon ) gas of 0 . 15 ml min-1 . Multiple reaction monitoring ( MRM ) was performed for delamanid using the transition 535 . 02 > 351 . 80 at a cone voltage of 16 V and collision energy of 33 V . Data was processed using the TargetLynx feature of Mass Lynx v4 . 1 . | Better , safer , oral drugs are desperately needed for the treatment of visceral leishmaniasis , a parasitic infectious disease that causes an estimated 40 , 000 deaths a year , predominantly in South America , East Africa and the Indian subcontinent . The parasite that causes visceral leishmaniasis is transmitted between individuals by blood-sucking sandflies , and there are currently no vaccines that protect against the disease . In addition , all currently available drug treatments have serious limitations – they are expensive , toxic , have to be applied over a long period of time ( mainly by injection ) and may become ineffective as the parasites adapt to resist the drug . A cost-effective way to find a new treatment for a disease is to repurpose existing clinically approved drugs that are used to treat other diseases . Patterson , Wyllie et al . now report that a drug called delamanid , which was recently approved for the treatment of tuberculosis , can cure visceral leishmaniasis in mice . The drug worked when applied orally at doses that might be achievable in human patients , and can also kill parasites obtained from human patients . Patterson , Wyllie et al . also provide evidence that suggests that delamanid is processed in the parasites by an unknown enzyme . However , this enzyme is not the one that activates a different class of drugs that are used to treat visceral leishmaniasis . Future studies now need to identify the enzyme that is targeted by delamanid , and could investigate combinations of drugs that slow the emergence of resistant parasites and improve delamanid’s safety and effectiveness . Clinical trials are required to test how well delamanid treats visceral leishmaniasis in humans . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"microbiology",
"and",
"infectious",
"disease"
] | 2016 | The anti-tubercular drug delamanid as a potential oral treatment for visceral leishmaniasis |
Electrophysiological methods , that is M/EEG , provide unique views into brain health . Yet , when building predictive models from brain data , it is often unclear how electrophysiology should be combined with other neuroimaging methods . Information can be redundant , useful common representations of multimodal data may not be obvious and multimodal data collection can be medically contraindicated , which reduces applicability . Here , we propose a multimodal model to robustly combine MEG , MRI and fMRI for prediction . We focus on age prediction as a surrogate biomarker in 674 subjects from the Cam-CAN dataset . Strikingly , MEG , fMRI and MRI showed additive effects supporting distinct brain-behavior associations . Moreover , the contribution of MEG was best explained by cortical power spectra between 8 and 30 Hz . Finally , we demonstrate that the model preserves benefits of stacking when some data is missing . The proposed framework , hence , enables multimodal learning for a wide range of biomarkers from diverse types of brain signals .
Non-invasive electrophysiology assumes a unique role in clinical neuroscience . Magneto- and electophencephalography ( M/EEG ) have an unparalleled capacity for capturing brain rhythms without penetrating the skull . EEG is operated in a wide array of peculiar situations , such as surgery ( Baker et al . , 1975 ) , flying an aircraft ( Skov and Simons , 1965 ) or sleeping ( Agnew et al . , 1966 ) . Unlike EEG , MEG captures a more selective set of brain sources with greater spectral and spatial definition ( Ahlfors et al . , 2010; Hari et al . , 2000 ) . Yet , neither of them is optimal for isolating anatomical detail . Clinical practice in neurology and psychiatry , therefore , relies on additional neuroimaging modalities with enhanced spatial resolution such as magnetic resonance imaging ( MRI ) , functional MRI ( fMRI ) , or positron emission tomography ( PET ) . Recently , machine learning has received significant interest in clinical neuroscience for its potential to predict from such heterogeneous multimodal brain data ( Woo et al . , 2017 ) . Unfortunately , the effectiveness of machine learning in psychiatry and neurology is constrained by the lack of large high-quality datasets ( Woo et al . , 2017; Varoquaux , 2017; Bzdok and Yeo , 2017; Engemann et al . , 2018 ) and comparably limited understanding about the data generating mechanisms ( Jonas and Kording , 2017 ) . This , potentially , limits the advantage of complex learning strategies proven successful in purely somatic problems ( Esteva et al . , 2017; Yoo et al . , 2019; Ran et al . , 2019 ) . In clinical neuroscience , prediction can therefore be pragmatically approached with classical machine learning algorithms ( Dadi et al . , 2019 ) , expert-based feature engineering and increasing emphasis on surrogate tasks . Such tasks attempt to learn on abundant high-quality data an outcome that is not primarily interesting , to then exploit its correlation with the actual outcome of interest in small datasets . This problem is also known as transfer learning ( Pan and Yang , 2009 ) which , in its simplest form , is implemented by reusing predictions from a surrogate-marker model as predictors in the small dataset . Over the past years , predicting the age of a person from its brain data has crystalized as a surrogate-learning paradigm in neurology and psychiatry ( Dosenbach et al . , 2010 ) . First results suggest that the prediction error of models trained to learn age from brain data of healthy populations provides clinically relevant information ( Cole et al . , 2018; Ronan et al . , 2016; Cole et al . , 2015 ) related to neurodegenerative anomalies , physical and cognitive decline ( Kaufmann et al . , 2019 ) . For simplicity , this characteristic prediction error is often referred to as the brain age delta Δ ( Smith et al . , 2019 ) . Can learning of such a surrogate biomarker be enhanced by combining expert-features from M/EEG , fMRI and MRI ? Research on aging has suggested important neurological group-level differences between young and elderly people: Studies have found alterations in grey matter density and volume , cortical thickness and fMRI-based functional connectivity , potentially indexing brain atrophy ( Kalpouzos et al . , 2012 ) and decline-related compensatory strategies . Peak frequency and power drop in the alpha band ( 8–12 Hz ) , assessed by EEG , has been linked to aging-related slowing of cognitive processes , such as the putative speed of attention ( Richard Clark et al . , 2004; Babiloni et al . , 2006 ) . Increased anteriorization of beta band power ( 15–30 Hz ) has been associated with effortful compensatory mechanisms ( Gola et al . , 2013 ) in response to intensified levels of neural noise , that is , decreased temporal autocorrelation of the EEG signal as revealed by flatter 1/f slopes ( Voytek et al . , 2015 ) . Importantly , age-related variability in fMRI and EEG seems to be independent to a substantial degree ( Kumral et al . , 2020 ) . The challenge of predicting at the single-subject level from such heterogenous neuroimaging modalities governed by distinct data-generating mechanisms has been recently addressed with model-stacking techniques ( Rahim et al . , 2015; Karrer et al . , 2019; Liem et al . , 2017 ) . Rahim et al . , 2015 enhanced classification in Alzheimer’s disease by combining fMRI and PET using a stacking approach ( Wolpert , 1992 ) , such that the stacked models used input data from different modalities . Liem et al . , 2017 have then applied this approach to age-prediction and found that combining anatomical MRI with fMRI significantly helped reduce errors while facilitating detection of cognitive impairment . This suggests that stacked prediction might also enable combining MRI with electrophysiology . Yet , this idea faces one important obstacle related to the clinical reality of data collection . It is often not practical to do multimodal assessments for all patients . Scanners may be overbooked , patients may not be in the condition to undergo MRI and acute demand in intensive care units may dominate priorities . Incomplete and missing data is , therefore , inevitable and has to be handled to unleash the full potential of multimodal predictive models . To tackle this challenge , we set out to build a stacking model for predicting age from electrophysiology and MRI such that any subject was included if some data was available for at least one modality . We , therefore , call it opportunistic stacking model . At this point , there are very few multimodal databases providing access to electrophysiology alongside MRI and fMRI . The Leipzig Mind-Brain-Body ( LEMON ) dataset ( Babayan et al . , 2019 ) includes high-quality research-EEG with MRI and fMRI for 154 young subjects and 75 elderly subjects . The dataset used in the present study is curated by the Cam-CAN ( Shafto et al . , 2014; Taylor et al . , 2017 ) and was specifically designed for studying the neural correlates of aging continuously across the life-span . The Cam-CAN dataset is currently the largest public resource on multimodal imaging with high-resolution electrophysiology in the form of MEG alongside MRI data and rich neuropsychological data for more than 650 healthy subjects between 17 and 90 years . The choice of MEG over EEG may lead to a certain degree of friction with the aging-related literature in electrophysiology , the bulk of which is based on EEG-studies . Fortunately , MEG and EEG share the same classes of neural generators , rendering the aging-related EEG-literature highly relevant for MEG-based modeling . On the other hand , the distinct biophysics of MEG and EEG makes both modalities complementary methods . While EEG captures sources of any orientation , MEG preferentially captures tangential but not radial sources . Compared to EEG , MEG benefits from the magnetic transparency of the skull , which facilitates source localization by reducing the risk of errors due to an incorrect head conductivity model , but also by limiting the large-scale mixing of neural sources . This significantly increases the signal-to-noise ratio for MEG in higher frequencies , rendering it a formidable technique for studying cortical oscillatory activity ( Lehtelä et al . , 1997; Gobbelé et al . , 1998 ) . MEG is , therefore , an interesting modality in its own right for developing neuro-cognitive biomarkers while its close link with EEG may potentially open the door to translatable electrophysiology markers suitable for massive deployment with clinical EEG . Our study focuses on the following questions: 1 ) Can MRI-based prediction of age be enhanced with MEG-based electrophysiology ? 2 ) Do fMRI and MEG carry non-redundant clinically relevant information ? 3 ) What are the most informative electrophysiological markers of aging ? 4 ) Can potential advantages of multimodal learning be maintained in the presence of missing values ?
We begin by summarizing the proposed method . To build a model for predicting age from electrophysiology , fMRI and anatomical MRI , we employed prediction-stacking ( Wolpert , 1992 ) . As in Liem et al . , 2017 , the stacked models , here , referred to different input data instead of alternative models on the same data . We used ridge regression ( Hoerl and Kennard , 1970 ) to linearly predict age from high-dimensional inputs of each modality . Linear predictions were based on distinct features from anatomical MRI , fMRI and MEG that have been commonly associated with aging . For extracting features from MEG , in a first step , we drew inspiration from EEG-literature on aging and considered evoked response latencies , alpha band peak frequency , 1/f slope topographies assessed in sensor-space . Previous work on neural development and aging ( Khan et al . , 2018; Gola et al . , 2013 ) and Alzheimer’s disease ( Gaubert et al . , 2019 ) has pointed at the importance of spatial alterations in stationary power spectra which can be exploited using high-dimensional regression techniques ( Fruehwirt et al . , 2017 ) . In this work , we have adapted this reasoning to the more general problem of predicting age while exploiting the advanced source-modeling options supported by the Cam-CAN dataset based on MEG and the individual MRIs . Therefore , it was our principal effort to expose the geometry of stationary power spectra with minimal distortion by using source localization based on the individual head geometry ( Sabbagh et al . , 2019 ) to then perform high-dimensional regression . As a result , we predicted from the spatial distribution of power and bivariate interactions between signals ( connectivity ) in nine frequency bands ( Table 1 ) . For MRI and fMRI , we followed the method established in Liem et al . , 2017 and included cortical thickness , cortical surface area and subcortical volume as well as functional connectivity based on the fMRI time-series . For detailed description of the features , see Table 2 and section Feature extraction in Materials and methods . To correct for the necessarily biased linear model , we then used a non-linear random forest regressor with age predictions from the linear model as lower-dimensional input features . Thereby , we made sure to use consistent cross-validation splits for all layers and automatically selected central tuning-parameters of the linear model and the random forest with nested cross-validation . Our stacked models handle missing values by treating missing value as data , provided there is an opportunity to see at least one modality ( Josse et al . , 2019 ) . We , therefore , call it opportunistic stacking model . Concretely , the procedure duplicated all variables and inserted once a small value and once a very large value where data was initially missing for which we chose biologically implausible age values of −1000 and 1000 , respectively . For an illustration of the proposed model architecture , see Figure 1 section Stacked-Prediction Model for Opportunistic Learning in Materials and methods for a detailed description of the model . Currently , anatomical MRI is the canonical modality for brain age prediction . However , MRI does not access brain dynamics , whereas MEG and fMRI both capture neuronal activity , hence , convey additional information at smaller time-scales . How would they add to the prediction of brain age when combined with anatomical MRI ? Figure 2A depicts a model comparison in which anatomical MRI served as baseline and which tracked changes in performance as fMRI and MEG were both added through stacking ( black boxplot ) . Anatomical MRI scored an expected generalization error of about 6 years ( SD=0 . 6 , P2 . 5 , 97 . 5=[4 . 9 , 7 . 16] ) , whereas expected chance-level prediction was about 15 . 5 years ( SD=1 . 17 , P2 . 5 , 97 . 5=[13 . 26 , 17 . 8] ) based on a dummy-model proposing as prediction the average age of the training-data . MRI performed better than chance-level prediction in every single cross-validation fold . The average improvement over chance-level prediction across folds was at least 9 years ( SD=1 . 33 , P2 . 5 , 97 . 5=[−12 . 073 , −7 . 347] ) . Relative to MRI , age-prediction performance was reduced by almost 1 year on average by adding either MEG ( Pr<MRI=91% , M=−0 . 79 , SD=0 . 57 , P2 . 5 , 97 . 5=[−1 . 794 , 0 . 306] ) or fMRI ( Pr<MRI=94% , M=−0 . 96 , SD=0 . 59 , P2 . 5 , 97 . 5=[−1 . 99 , 0 . 15] ) . Finally , the performance gain was greater than 1 year on average ( Pr<MRI=99% , M=−1 . 32 , SD=0 . 672 , P2 . 5 , 97 . 5=[−2 . 43 , −0 . 16] ) when adding both MEG and fMRI to the model , yielding an expected generalization error of about 4 . 7 years ( SD=0 . 55 , P2 . 5 , 97 . 5=[3 . 77 , 5 . 74] ) . Note that dependable numerical p-values are hard to obtain for paired model comparisons based on cross-validation on the same dataset: Many datasets equivalent to the Cam-CAN would be required . Nevertheless , the uncertainty intervals extracted from the cross-validation distribution suggests that the observed differences in performance were systematic and can be expected to generalize as more data is analyzed . Moreover , the out-of-sample ranking between the different models was stable over cross-validation folds ( Figure 2—figure supplement 1 ) with the full model achieving the first rank 71/100 times and performing at least 80/100 better than the MRI + fMRI or the MRI + MEG model . This emphasizes that the relative importance of MEG and fMRI for enhancing MRI-based prediction of age can be expected to generalize to future data . The improved prediction obtained by combining MEG and fMRI suggests that both modalities carry independent information . If MEG and MRI carried purely redundant information , the random forest algorithm would not have reached better out-of-sample performance . Indeed , comparing the cross-validated prediction errors of MEG-based and fMRI-based models ( Figure 2B ) , errors were only weakly correlated ( rSpearman=0 . 139 , r2=0 . 019 , p=1 . 31×10−3 ) . fMRI , sometimes , made extreme errors for cases better predicted by MEG in younger people , whereas MEG made errors in distinct cases from young and old age groups . When adding anatomical MRI to each model , the errors became somewhat more dependent leading to moderate correlation ( rSpearman=0 . 45 , r2=0 . 20 , p=2 . 2×10−16 ) . This additive component also became apparent when considering predictive simulations on how the model actually combined MEG , fMRI and MRI ( Figure 2—figure supplement 2 ) using two-dimensional partial dependence analysis ( Karrer et al . , 2019; Hastie et al . , 2005 , chapter 10 . 13 . 2 ) . Moreover , exploration of the age-dependent improvements through stacking suggest that stacking predominantly reduced prediction errors uniformly ( Figure 2—figure supplement 3 ) instead of systematically mitigating brain age bias ( Le et al . , 2018; Smith et al . , 2019 ) . These findings demonstrate that stacking allows to enhance brain-age prediction by extracting information from MEG , fMRI and MRI while mitigating modality-specific errors . This raises the question whether this additive information from multiple neuroimaging modalities also implies non-redundant associations with behavior and cognition . The brain ageΔ has been interpreted as indicator of health where positive Δ has been linked to reduced fitness or health-outcomes ( Cole et al . , 2015; Cole et al . , 2018 ) . Does improved performance through stacking strengthen effect-sizes ? Can MEG and fMRI help detect complementary associations ? Figure 3 summarizes linear correlations between the brain ageΔ and the 38 neuropsychological scores after projecting out the effect of age , Equations 6- 8 ( see Analysis of brain-behavior correlation in Materials and methods for a detailed overview ) . As effect sizes can be expected to be small in the curated and healthy population of the Cam-CAN dataset , we considered classical hypothesis testing for characterizing associations . Traditional significance testing ( Figure 3A ) suggests that the best stacking models supported discoveries for between 20% ( 7 ) and 25% ( 9 ) of the scores . Dominating associations concerned fluid intelligence , depression , sleep quality ( PSQI ) , systolic and diastolic blood pressure ( cardiac features 1 , 2 ) , cognitive impairment ( MMSE ) and different types of memory performance ( VSTM , PicturePriming , FamousFaces , EmotionalMemory ) . The model coefficients in Figure 3B depict the strength and direction of association . One can see that stacking models not only tended to suggest more discoveries as their performance improved but also led to stronger effect sizes . However , the trend is not strict as fMRI seemed to support unique discoveries that disappeared when including the other modalities . Similarly , some effect sizes were even slightly stronger in sub-models , for example for fluid intelligence in MRI and MEG . A priori , the full model enjoys priority over the sub-models as its expected generalization estimated with cross-validation was lower . This could imply that some of the discoveries suggested by fMRI may suffer from overfitting , but are finally corrected by the full model . Nevertheless , many of the remaining associations were found by multiple methods ( e . g . fluid intelligence , sleep quality assessed by PSQI ) whereas others were uniquely contributed by fMRI ( e . g . depression ) . It is also noteworthy that the directions of the effects were consistent with the predominant interpretation of the brain age Δ as indicator of mental or physical fitness ( note that high PSQI score indicate sleeping difficulties whereas lower MMSE scores indicate cognitive decline ) and directly confirm previous findings ( Liem et al . , 2017; Smith et al . , 2019 ) . Note that the results were highly similar when performing deconfounding jointly via multiple regression ( Equation 9 , Figure 3—figure supplement 1 ) instead of predicting age-residualized neuropsychological scores , and when including additional predictors of non-interest , that is gender , handedness and head motion ( Equation 10 , Figure 3—figure supplement 2 ) . More elaborate confounds-modeling even seemed to improve SNR as suggested by an increasing number of discoveries and growing effect sizes . These findings suggest that brain age Δ learnt from fMRI or MEG carries non-redundant information on clinically relevant markers of cognitive health and that combining both fMRI and MEG with anatomy can help detect health-related issues in the first place . This raises the question of what aspect of the MEG signal contributes most . Whether MEG or EEG-based assessment is practical in the clinical context depends on the predictive value of single features , the cost for obtaining predictive features and the potential benefit of improving prediction by combining multiple features . Here , we considered purely MEG-based age prediction to address the following questions: Can the stacking method be helpful to analyze the importance of MEG-specific features ? Are certain frequency bands of dominating importance ? Is information encoded in the regional power distribution or more related to neuronal interactions between brain regions ? Figure 4A compares alternative MEG-based models stacking different combinations of MEG-features . We compared models against chance-level prediction as estimated with a mean-regressor outputting the average age of the training data as prediction . Again , chance-level was distributed around 15 . 5 years ( SD=1 . 17 , P2 . 5 , 97 . 5=[13 . 26 , 17 . 80] ) . All models performed markedly better . The model based on diverse sensor space features from task and resting state recordings showed the lowest performance around 12 years MAE ( SD=1 . 04 , P2 . 5 , 97 . 5=[9 . 80 , 13 . 52] ) , yet it was systematically better than chance ( Pr<Chance=98 . 00% , M=−4 , SD=1 . 64 , P2 . 5 , 97 . 5=[−7 . 11 , −0 . 44] ) . All models featuring source-level power spectra or connectivity ( ‘Source Activity , Source Connectivity’ ) performed visibly better , with expected errors between 8 and 6 . 5 years and no overlap with the distribution of chance-level scores . Models based on source-level power spectra ( ‘Source Activity’ , M=7 . 40 , SD=0 . 82 , P2 . 5 , 97 . 5=[6 . 01 , 9 . 18] ) and connectivity ( ‘Source Connectivity’ , M=7 . 58 , SD=0 . 90 , P2 . 5 , 97 . 5=[6 . 05 , 9 . 31] ) performed similarly with a slight advantage for the ‘Source Activity’ model . The best results were obtained when combining power and connectivity features ( ‘Full’ , M=6 . 75 , SD=0 . 83 , P2 . 5 , 97 . 5=[5 . 36 , 8 . 20] ) . Adding sensor space features did not lead to any visible improvement of ‘Full’ over ‘Combine Source’ with virtually indistinguishable error distributions . The observed average model-ranking was highly consistent over cross-validation testing-splits ( Figure 4—figure supplement 1 ) , suggesting that the relative importance of the different blocks of MEG features was systematic , hence , can be expected to generalize to future data . The observed ranking between MEG models suggests that regional changes in source-level power spectra contained most information while source-level connectivity added another portion of independent information which helped improve prediction by at least 0 . 5 years on average . A similar picture emerged when inspecting the contribution of the Layer-I linear models to the performance of the full model in terms of variable importance ( Figure 4B ) . Sensor space features were least influential , whereas top contributing features were all related to power and connectivity , which , upon permutation , increased the error by up to 1 year . The most informative input to the stacking model were ridge regression models based on either signal power or the Hilbert analytic signal power concatenated across frequency bands ( Pcat , Ecat ) . Other noteworthy contributions were related to power envelope covariance ( without source leakage correction ) as well as source power in the beta ( 15–30 Hz ) and alpha ( 8–15 Hz ) band frequency range . The results suggest that regional changes in power across different frequency bands are best summarized with a single linear model but additional non-linear additive effects may exist in specific frequency bands . The observed importance rankings were highly consistent with importance rankings obtained from alternative methods for extraction of variable importance ( Figure 4—figure supplement 2 ) , emphasizing the robustness of these rankings . Moreover , partial dependence analysis ( Karrer et al . , 2019; Hastie et al . , 2005 , chapter 10 . 13 . 2 ) suggested that the Layer-II random forest extracted non-linear functions ( Figure 4—figure supplement 3 ) . Finally , the best stacked models scored lower errors than the best linear models ( Figure 4—figure supplement 4 ) , suggesting that stacking achieved more than mere variable selection by extracting non-redundant information from the inputs . These findings show that MEG-based prediction of age is predominantly enabled by power spectra that can be relatively easily accessed in terms of computation and data processing . Moreover , the stacking approach applied to MEG data helped improve beyond linear models by upgrading to non-linear regression . One important obstacle for combining signals from multiple modalities in clinical settings is that not all modalities are available for all cases . So far , we have restricted the analysis to 536 cases for which all modalities were present . Can the advantage of multimodal stacking be preserved in the absence of complete data or will missing values mitigate prediction performance ? To investigate this question , we trained our stacked model on all 674 cases for which we had the opportunity to extract at least one feature on any modality , hence , opportunistic stacking ( see Figure 1 and Table 3 in section Sample in Materials and methods ) . We first compared the opportunistic model with the restricted model on the cases with complete data Figure 5A . Across stacking models , performance was virtually identical , even when extending the comparison to the cases available to the sub-model with fewer modalities , for example MRI and fMRI . We then scored the fully opportunistic model trained on all cases and all modalities and compared it to different non-opportunistic sub-models on restricted cases ( Figure 5A , squares ) . The fully opportunistic model always out-performed the sub-model . This raises the question of how the remaining cases would be predicted for which fewer modalities were available . Figure 5B shows the performance of the opportunistic split by subgroups defined by different combinations of input modalities available . As expected , performance degraded considerably on subgroups for which important features ( as delineated by the previous results ) were not available . See , for example , the subgroup for which only sensor-space MEG was available . This is unsurprising , as prediction has to be based on data and is necessarily compromised if the features important at train-time are not available at predict-time . One can , thus , say that the opportunistic model operates conservatively: The performance on the subgroups reflects the quality of the features available , hence , enables learning from the entire data . It is important to emphasize that if missing values depend on age , the opportunistic model inevitably captures this information , hence , bases its predictions on the non-random missing data . This may be desirable or undesirable , depending on the applied context . To diagnose this model-behavior , we propose to run the opportunistic random forest model with the observed missing values as input and observations from the input modalities set to zero . In the current setting , the model trained on missing data indicators performed at chance level ( Pr<Chance=30 . 00% , M=0 . 65 , SD=1 . 68 , P2 . 5 , 97 . 5=[−2 . 96 , 3 . 60] ) , suggesting that the missing values were not informative of age .
Our results have revealed complementary effects of anatomy and neurophysiology in age-prediction . When adding either MEG or fMRI to the anatomy-based stacking model , the prediction error markedly dropped ( Figure 2A ) . Both , MEG and fMRI helped gain almost 1 year of error compared to purely anatomy-based prediction . This finding suggests that both modalities access equivalent information . This is in line with the literature on correspondence of MEG with fMRI in resting state networks , highlighting the importance of spatially correlated slow fluctuations in brain oscillations ( Hipp and Siegel , 2015; Hipp et al . , 2012; Brookes et al . , 2011 ) . On the other hand , recent findings suggest that age-related variability in fMRI and EEG is independent to a substantial degree ( Kumral et al . , 2020; Nentwich et al . , 2020 ) . Interestingly , the prediction errors of models with MEG and models with fMRI were rather weakly correlated ( Figure 2B , left panel ) . In some subpopulations , they even seemed anti-correlated , such that predictions from MEG or fMRI , for the same cases , were either accurate or extremely inaccurate . This additional finding suggests that the improvements of MEG and fMRI over anatomical MRI are due to their access to complementary information that helps predicting distinct cases . Indeed , as we combined MEG and fMRI in one common stacking model alongside anatomy , performance improved on average by 1 . 3 years over the purely anatomical model , which is almost half a year more precise than the previous MEG-based and fMRI-based models . These results strongly argue in favor of the presence of an additive component , in line with the common intuition that MEG and fMRI are complementary with regard to spatial and temporal resolution . In this context , our results on performance decomposition in MEG ( Figure 4 ) deliver one potentially interesting hint . Source power , especially in the α ( 8-15Hz ) and β ( 15-26Hz ) range were the single most contributing type of feature ( Figure 4A ) . However , connectivity features , in general , and power-envelope connectivity , in particular , contributed substantively ( Figure 4B , Table 4 ) . Interestingly , applying orthogonalization ( Hipp et al . , 2012; Hipp and Siegel , 2015 ) for removing source leakage did not notably improve performance ( Table 4 ) . Against the background of research on MEG-fMRI correspondence highlighting the importance of slow fluctuations of brain rhythms ( Hipp and Siegel , 2015; Brookes et al . , 2011 ) , this finding suggests that what renders MEG non-redundant with regard to fMRI are regional differences in the balance of fast brain-rhythms , in particular in the α-β range . While this interpretation may be enticing , an important caveat arises from the fact that fMRI signals are due to neurovascular coupling , hence , highly sensitive to events caused by sources other than neuronal activity ( Hosford and Gourine , 2019 ) . Recent findings based on the dataset analyzed in the present study have shown that the fMRI signal in elderly populations might predominantly reflect vascular effects rather than neuronal activity ( Tsvetanov et al . , 2015 ) . The observed complementarity of the fMRI and MEG in age prediction might , therefore , be conservatively explained by the age-related increase in the ratio of vascular to neuronal contributions to the fMRI signal , while MEG signals are directly induced by neuronal activity , regardless of aging . Nevertheless , in the context of brain-age prediction these mechanisms are less important than the sensitivity of the prediction , for instance , regarding behavioral outcomes . In sum , our findings suggest that electrophysiology can make a difference in prediction problems in which fast brain rhythms are strongly statistically related to the biomedical outcome of interest . In this study , we have conducted an exploratory analysis on what might be the cognitive and health-related implications of our prediction models . Our findings suggest that the brain age Δ shows substantive associations with about 20–25% of all neuropsychological measures included . The overall big-picture is congruent with the brain age literature ( see discussion in Smith et al . , 2019 for an overview ) and supports the interpretation of the brain age Δ as index of decline of physical health , well-being and cognitive fitness . In this sample , larger values of the Δ were globally associated with elevated depression scores , higher blood pressure , lower sleep quality , lower fluid intelligence , lower scores in neurological assessment and lower memory performance . Most strikingly , we found that fMRI and MEG contributed additive , if not unique information ( Figure 3 ) . For example , the association with depression appeared first when predicting age from fMRI . Likewise , the association with fluid intelligence and sleep quality visibly intensified when including MEG . This extends the previous discussion in suggesting that MEG and fMRI are not only complementary for prediction but also with regard to characterizing brain-behavior mappings . In this context , it is worwhile considering that predicting biomedical outcomes from multiple modalities may reduce susceptibility to ‘modality impurity’ as often observed in modeling of individual differences in cognitive abilities ( Friedman and Miyake , 2004; Miyake et al . , 2000 ) . In the present study , it was remarkable that cardiac measures were exclusively related to fMRI-based models and vanished as MEG was included . This may not be entirely surprising as the fMRI signal is a combination of , both , vascular and neuronal components ( Hosford and Gourine , 2019 ) and aging affects both of them differently , which poses an important challenges to fMRI-based studies of aging ( Geerligs et al . , 2017; Tsvetanov et al . , 2016 ) . It is imaginable that the cardiac measures were not associated with brain age estimates from fMRI when combined with the modalities as vascular components may have enhanced the SNR of neuronal signals through deconfounding ( for extensive discussion on this topic , see Tsvetanov et al . , 2019 ) . Which neuronal components might explain the enhanced brain-behavior links extracted from the multimodal models ? It is enticing to speculate that the regional power of fast-paced α and β band brain rhythms captures fast-paced components of cognitive processes such as attentional sampling or adaptive attention ( Gola et al . , 2013; Richard Clark et al . , 2004 ) , which , in turn might explain unique variance in certain cognitive facets , such as fluid intelligence ( Ouyang et al . , 2020 ) or visual short-term memory ( Tallon-Baudry et al . , 2001 ) . On the other hand , functional connectivity between cortical areas and subcortical structures , in particular the hippocampus , may be key for depression and is well captured with fMRI ( Stockmeier et al . , 2004; Sheline et al . , 2009; Rocca et al . , 2015 ) . Unfortunately , modeling such mediation effects exceeds the scope of the current work , although it would be worth being tested in an independent study with a dedicated design . Could one argue that the overall effect sizes were too low to be considered practically interesting ? Indeed , the strength of linear association was below 0 . 5 in units of standard deviations of the normalized predictors and the target . On the other hand , it is important to consider that the Cam-CAN sample consists of healthy individuals only . It , thus , appears as rather striking that systematic and neuropsychologically plausible effects can be detected . Our findings , therefore , argue in favor of the brain age Δ being a sensitive marker of normative aging . The effects are expected to be far more pronounced when applying the method in clinical settings , that is , in patients suffering from mild cognitive impairment , depression , neurodevelopmental or neurodegenerative disorders . This suggests that brain age Δ might be used as a screening tool for a wide array of clinical settings for which the Cam-CAN dataset could serve as a normative sample . One critical factor for application of our approach in the clinic is the problem of incomplete availability of medical imaging and physiological measurements . Here , we addressed this issue by applying an opportunistic learning approach which enables learning from the data available at hand . Our analysis of opportunistic learning applied to age prediction revealed viable practical alternatives to confining the analysis to cases for which all measurements are available . In fact , adding extra cases with incomplete measurements never harmed prediction of the cases with complete data and the full multimodal stacking always outperformed sub-models with fewer modalities ( Figure 5A ) . Moreover , the approach allowed maintaining and extending the performance to new cases with incomplete modalities ( Figure 5B ) . Importantly , performance on such subsets was explained by the performance of a reduced model with the remaining modalities . Put differently , opportunistic stacking performed as good as a model restricted to data with all modalities . In practice , the approach allows one to improve predictions case-wise by including electrophysiology next to MRI or MRI next to electrophysiology , whenever there is the opportunity to do so . A second critical factor for translating our findings into the clinic is that , most of the time , it is not high-density MEG that is available but low-density EEG . In this context , our finding showed that the source power was the most important feature , which is of clear practical interest . This is because it suggests that a rather simple statistical object accounts for the bulk of the performance of MEG . Source power can be approximated by the sensor-level topography of power spectra which can be computed on any multichannel EEG device in a few steps and only yields as many variables per frequency band as there are channels . Moreover , from a statistical standpoint , computing the power spectrum amounts to estimating the marginal expectation of the signal variance , which can be thought of as main effect . On the other hand , connectivity is often operationalized as bivariate interaction , which gives rise to a more complex statistical object of higher dimensionality whose precise , reproducible estimation may require far more samples . Moreover , as is the case for power envelope connectivity estimation , additional processing steps each of which may add researcher degrees of freedom ( Simmons et al . , 2011 ) , such as the choice between Hilbert ( Brookes et al . , 2011 ) versus Wavelet filtering ( Hipp et al . , 2012 ) , types of orthogonalization ( Baker et al . , 2014 ) , and potentially thresholding for topological analysis ( Khan et al . , 2018 ) . This nourishes the hope that our findings will generalize and similar performance can be unlocked on simpler EEG devices with fewer channels . While clinical EEG may not well resolve functional connectivity it may be good enough to resolve changes in the source geometry of the power spectrum ( Sabbagh et al . , 2020 ) . On the other hand , source localization may be critical in this context as linear field spread actually results in a non-linear transform when considering the power of a source ( Sabbagh et al . , 2019 ) . However , in practice , it may be hard to conduct high-fidelity source localization on the basis of low-density EEG and frequently absent information on the individual anatomy . It will , therefore , be critical to benchmark and improve learning from power topographies in clinical settings . Finally , it is worthwhile to highlight that , here , we have focused on age in the more specific context of the brain age Δ as surrogate biomarker . However , the proposed approach is fully compatible with any target of interest and may be easily applied directly to clinical end points , for example drug dosage , survival or diagnosis . Moreover , the approach presented here can be easily adapted to work with classification problems , for instance , by substituting logistic regression for ridge regression and by using a random forest classifier in the stacking layer . We have provided all materials from our study in form of publicly available version-controlled code with the hope to help other teams of biomedical researchers to adapt our method to their prediction problem . For the present study , we see four principal limitations: availability of data , interpretability , non-exhaustive feature-engineering and potential lack of generalizability due to the focus on MEG . The Cam-CAN is a unique resource of multimodal neuroimaging data with sufficient data points to enable machine learning approaches . Yet , from the point of view of machine learning , the Cam-CAN dataset is a small dataset . This has at least two consequences . If the Cam-CAN included many more data points , for example beyond 10–100 k subjects , the proposed stacking model might possibly be of limited advantage compared to purely non-linear models , for example random forests , gradient boosting or deep learning methods ( Bzdok and Yeo , 2017 ) . At the same time , the fact that the Cam-CAN has been unique so far , hinders generalization testing to equivalent multimodal datasets from other sites based on alternative scanning methodologies , protocols and devices ( Engemann et al . , 2018 ) . This also renders computation of numerical hypothesis tests ( including p-values ) more difficult in the context of predictive modeling: The majority of data points is needed for model-fitting and metrics derived from left-out cross-validation splits , for example , predictions of brain age , lack statistical independence . This breaks essential assumptions of inferential statistics to an arbitrary and unknown degree . Our inferences were , therefore , predominantly based on estimated effect-sizes , that is the expected generalization error and its uncertainty assessed through cross-validation . Second , at this point , statistical modeling faces the dilemma of whether inference or prediction is the priority . Procedures optimizing prediction performance in high dimensions are not yet supported by the in-depth understanding required to guarantee formal statistical inferences , whereas models with well-established procedures for statistical inference lack predictive capability ( Bzdok et al . , 2018; Bzdok and Ioannidis , 2019 ) . Forcing interpretation out of machine learning models , therefore , often leads to duplicated analysis pipelines and model specifications , which is undesirable in terms of methodological coherence ( for example Hoyos-Idrobo et al . , 2019; Haufe et al . , 2014; Biecek , 2018 ) . In the present work , we refrained from conducting fine-grained inferential analysis beyond the model comparisons presented , in particular inspection of layer-1 weightmaps whose interpretation remains an ongoing research effort . We hope , nevertheless , that the insights from our work will stimulate studies investigating the link between MEG , fMRI and MRI across the life-span using an inference-oriented framework . Third , the MEG-features used in the present study were non-exhaustive . Based on the wider MEG/EEG-literature beyond the neuroscience of aging , many other features could have been included . Instead , feature-engineering was based on our aging-specific literature review constrained by biophysical considerations . In particular , the distinction between sensor-space and source-space features was purely descriptive and not substantive . From an empirical perspective , mirroring all features in sensor-space and source-space could have yielded more specific inferences , for example regarding the role of source-power . On the other hand , biophysical prior knowledge implies that oscillatory peak frequencies and evoked response latencies are not modified by source localization , whereas source localization or data-driven approximations thereof are essential for predicting from M/EEG power spectra ( Sabbagh et al . , 2019 ) . It is also fair to admit that , in the present paper , our passion was preferentially attracted by source modeling of neural power spectra . However , one could imagine that with equal investment of resources , more information could have been extracted from the sensor-level features ( see Gemein et al . , 2020 for approaches to tackle the important methodological issue of unbalanced investment of development-time ) . Related , the current work has strongly benefited from expertise on modeling of MEG power spectra under the assumption of stationary as captured by global power spectra , covariance or connectivity . Recent findings suggest that non-stationary analyses focusing on transient electrophysiological events may uncover clinically relevant information on cognitive brain dynamics ( Barttfeld et al . , 2015; Baker et al . , 2014; Vidaurre et al . , 2018; Van Schependom et al . , 2019 ) . It is , therefore , important to highlight that our proposed framework is open and readily enables integration of additional low- or high-dimensional inputs related to richer sensor-level features or non-stationary dynamics , beyond MEG as input modality . Finally , while MEG and EEG share the same types of neural generators , their specific biophysics render these methods complementary for studying neuronal activity . At this point , unfortunately , there is no public dataset equivalent of the Cam-CAN including EEG or , both , EEG and MEG . Such a data resource would have enabled studying the complementarity between MEG with EEG as well as generalization from stacking with MRI and MEG to stacking models with MRI and EEG . We hope that our method will help other scientists to incorporate the multimodal features related to their domain expertise into their applied regression problems .
We included MEG ( task and rest ) , fMRI ( rest ) , anatomical MRI and neuropsychological data ( cognitive tests , home-interview , questionnaires ) from the CAM-Can dataset ( Shafto et al . , 2014 ) . Our sample comprised 674 ( 340 female ) healthy individuals between 18 ( female = 18 ) to 88 ( female = 87 ) years with an average of 54 . 2 ( female = 53 . 7 ) and a standard deviation of 18 . 7 ( female = 18 . 8 ) years . We included data according to availability and did not apply an explicit criterion for exclusion . When automated processing resulted in errors , we considered the data as missing . This induced additional missing data for some cases . A summary of available cases by input modality is reported in Table 3 . For technical details regarding the MEG , fMRI , and MRI data acquisition , please consider the Cam-CAN reference publications ( Shafto et al . , 2014; Taylor et al . , 2017 ) . Feature extraction was guided by the perspective of predictive modeling . For the goal of enhancing prediction performance as opposed to statistical inference ( Bzdok and Ioannidis , 2019 ) , we emphasized on differences between modalities , hence , chose modality-specific methods and optimizations at the risk of sacrificing direct comparability between features used for MEG , fMRI and MRI . The selection of features was guided by our literature review on the neuroscience of aging presented in the introduction . For MEG , we analyzed sensor space features related to timing ( Price et al . , 2017 ) , peak frequency ( Richard Clark et al . , 2004 ) and temporal autocorrelation ( Voytek et al . , 2015 ) . Source space features included the power of source-level signals ( Sabbagh et al . , 2019 ) and envelopes and their bivariate interactions ( Khan et al . , 2018 ) in nine frequency bands ( see Table 1 , adapted from the Human Connectome Project , Larson-Prior et al . , 2013 ) . The inclusion of power envelopes was theoretically important as the slow fluctuations of source power and their bivariate interactions have been repeatedly linked to fMRI resting state networks ( Hipp and Siegel , 2015; Brookes et al . , 2011 ) . On the other hand , we specifically focused on the unique capacity of MEG to access spatial information induced by fast-paced brain rhythms emerging from regional sources ( King and Dehaene , 2014; Stokes et al . , 2015 ) . For extracting features from MRI and fMRI , we adapted the approach established by Liem et al . , 2017 . For fMRI , we computed bivariate functional connectivity estimates . For MRI , we focused on cortical thickness , cortical surface area and subcortical volumes . An overview on all features used is presented in Table 2 . In the remainder of this section , we describe computation details . We used the stacking framework ( Wolpert , 1992 ) to build our predictive model . However , we made the important specification that input models were regularized linear models trained on input data from different modalities , whereas stacking of linear predictions was achieved by a non-linear regression model . Our model can be intuitively denoted as follows: ( 1 ) y=f ( [X1β1…Xmβm] ) Here , each Xjβj is the vector of predictions y^j of the target vector y from the jth model fitted using input data Xj: ( 2 ) {y=X1β1+ϵ1 , … , y=Xmβm+ϵm} We used ridge regression as input model and a random forest regressor as a general function approximator f [Ch . 15 . 4 . 3] ( Hastie et al . , 2005 ) . A visual illustration of the model is presented in Figure 1 . To explore the cognitive implications of the brain age Δ , we computed correlations with the neurobehavioral score from the Cam-CAN dataset . Table 5 lists the scores we considered . The measures fall into three broad classes: neuropsychology , physiology and questionnaires ( ‘Type’ columns in Table 5 ) . Extraction of neuropsychological scores sometimes required additional computation , which followed the description in Shafto et al . , 2014 , ( see also ‘Variables’ column in scores ) . For some neuropsychological tasks , the Cam-CAN dataset provided multiple scores and sometimes the final score of interest as described in Shafto et al . , 2014 , had yet to be computed . At times , this amounted to computing ratios , averages or differences between different scores . In other scores , it was not obvious how to aggregate multiple interrelated sub-scores , hence , we extracted the first principal component explaining between about 50% and 85% of variance , hence offering reasonable summaries . In total , we included 38 variables . All neuropsychology and physiology scores ( up to #17 in Table 5 ) were the scores available in the ‘cc770-scored’ folder from release 001 of the Cam-CAN dataset . We selected the additional questionnaire scores ( #18-23 in Table 5 ) on theoretical grounds to provide an assessment of clinically relevant individual differences in cognitive functioning . The brain age Δ was defined as the difference between predicted and actual age of the person ( 5 ) BrainAgeΔ=agepred−age , such that positive values quantify overestimation and negative value underestimation . A common problem in establishing brain-behavior correlations for brain age is spurious correlations due to shared age-related variance in the brain age Δ and the neurobehavioral score ( Smith et al . , 2019 ) . To mitigate confounding effects of age , we computed the age residuals as ( 6 ) scoreresid=score−scoreage , where score is the observed neuropsychological score and scoreage is its prediction from the following polynomial regression: ( 7 ) scoreage=ageβ1+age2β2+age3β3+ϵ , The estimated linear association between the residualized score and the brain age Δ was given by β1 in ( 8 ) scoreresid=BrainAgeΔβ1+ϵ , To obtain comparable coefficients across scores , we standardized both the age and the scores . We also included intercept terms in all models which are omitted here for simplicity . It has been recently demonstrated , that such a two-step procedure can lead to spurious associations ( Lindquist et al . , 2019 ) . We have , therefore , repeated the analysis with a joint deconfounding model where the polynomial terms for age are entered into the regression model alongside the brain age predictor . ( 9 ) score=BrainAgeΔβ1+ageβ2+age2β3+age3β4+ϵ . Finally , the results may be due to confounding variable of non-interest . To assess the importance of such confounders , we have extended the model ( Equation 9 ) to also include gender , handedness ( binarized ) and a log Frobenius norm of the variability of motion parameters ( three translation , three rotation ) over the 241 acquired images . ( 10 ) score=BrainAgeΔβ1+genderβ2+handbinaryβ3+log ( norm ( motion ) ) β4+ageβ5+age2β6+age3β7+ϵ . Note that motion correction was already performed during preprocessing of MRI and fMRI . Likewise , MEG source localization took into account individual head geometry as well as potentially confounding environmental noise through whitening with the noise covariance obtained from empty room recordings . Following the work by Liem et al . , 2017 , we included total grey matter and total intracranial volume as important features of interest among the MRI-features . We share all code used for this publication on GitHub: https://github . com/dengemann/meg-mri-surrogate-biomarkers-aging-2020 . ( Engemann , 2020; https://github . com/elifesciences-publications/meg-mri-surrogate-biomarkers-aging-2020 ) Our stacked model architecture can be compactly expressed using the StackingRegressor class in scikit-learn ( Pedregosa et al . , 2011 ) as of version 0 . 22 . | How old are you ? What about your body , and your brain ? People are used to answering this question by counting the years since birth . However , biological age could also be measured by looking at the integrity of the DNA in cells or by measuring the levels of proteins in the blood . Whether one goes by chronological age or biological age , each is simply an indicator of general health – but people with the same chronological age may have different biological ages , and vice versa . There are different imaging techniques that can be used to study the brain . A method called MRI reveals the brain’s structure and the different types of tissue present , like white and grey matter . Functional MRIs ( fMRIs for short ) measure activity across different brain regions , while electrophysiology records electrical signals sent between neurons . Distinct features measured by all three techniques – MRI , fMRI and electrophysiology – have been associated with aging . For example , differences between younger and older people have been observed in the proportion of grey to white matter , the communication between certain brain regions , and the intensity of neural activity . MRIs , with their anatomical detail , remain the go-to for predicting the biological age of the brain . Patterns of neuronal activity captured by electrophysiology also provide information about how well the brain is working . However , it remains unclear how electrophysiology could be combined with other brain imaging methods , like MRI and fMRI . Can data from these three techniques be combined to better predict brain age ? Engemann et al . designed a computer algorithm stacking electrophysiology data on top of MRI and fMRI imaging to assess the benefit of this three-pronged approach compared to using MRI alone . Brain scans from healthy people between 17 and 90 years old were used to build the computer model . The experiments showed that combining all three methods predicted brain age better . The predictions also correlated with the cognitive fitness of individuals . People whose brains were predicted to be older than their years tended to complain about the quality of their sleep and scored worse on memory and speed-thinking tasks . Crucially , Engemann et al . tested how the algorithm would hold up if some data were missing . This can happen in clinical practice where some tests are required but not others . Positively , prediction was maintained even with incomplete data , meaning this could be a useful clinical tool for characterizing the brain . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"tools",
"and",
"resources",
"neuroscience"
] | 2020 | Combining magnetoencephalography with magnetic resonance imaging enhances learning of surrogate-biomarkers |
We do not know how or why multicellularity evolved . We used the budding yeast , Saccharomyces cerevisiae , to ask whether nutrients that must be digested extracellularly select for the evolution of undifferentiated multicellularity . Because yeast use invertase to hydrolyze sucrose extracellularly and import the resulting monosaccharides , single cells cannot grow at low cell and sucrose concentrations . Three engineered strategies overcame this problem: forming multicellular clumps , importing sucrose before hydrolysis , and increasing invertase expression . We evolved populations in low sucrose to ask which strategy they would adopt . Of 12 successful clones , 11 formed multicellular clumps through incomplete cell separation , 10 increased invertase expression , none imported sucrose , and 11 increased hexose transporter expression , a strategy we had not engineered . Identifying causal mutations revealed genes and pathways , which frequently contributed to the evolved phenotype . Our study shows that combining rational design with experimental evolution can help evaluate hypotheses about evolutionary strategies .
Multicellular organisms have evolved from a unicellular ancestor at least 25 times ( Grosberg and Strathmann , 2007 ) , but we know little about what selected for the simplest form of multicellularity: an undifferentiated clump of cells produced by the repeated division of a single cell . Two driving forces have been proposed , protection from a variety of factors ( including predation [Kessin et al . , 1996] , environmental stress [Smukalla et al . , 2008] , and phagocytosis [Boraas et al . , 1998] ) and more efficient nutrient usage ( Dworkin , 1972; Pfeiffer and Bonhoeffer , 2003; Koschwanez et al . , 2011; Alegado et al . , 2012 ) . In earlier work , we showed that sharing public goods favors clumps over isolated cells and proposed that sharing could have selected for simple multicellularity ( Koschwanez et al . , 2011 ) . The budding yeast , Saccharomyces cerevisiae , utilizes sucrose by secreting invertase ( Dodyk and Rothstein , 1964; Carlson et al . , 1981 ) . Over 95% of this enzyme remains in the cell wall ( Esmon et al . , 1987; Tammi et al . , 1987 ) , where it hydrolyzes sucrose into glucose and fructose , which are imported into the cell by a variety of hexose transporters ( Meijer et al . , 1996; Reifenberger et al . , 1997 ) . Lab yeast strains cannot grow from low density in low concentrations of sucrose because of diffusion: each cell captures only a small fraction of the sugars that sucrose hydrolysis releases , and the molecules released by other , distant cells are at very low concentration . As a result , cells cannot capture enough glucose and fructose to grow . Forming multicellular clumps overcomes this failure; cells in a clump can capture glucose and fructose diffusing from their neighbors and grow in concentrations of sucrose where low concentrations of individual cells cannot . Speculating on evolution based on experiments with engineered strains is problematic . How well is the ease of engineering a strategy correlated with its evolutionary accessibility ? Are multiple mutations required ? Do these mutations reduce fitness in other environmental conditions ? Are other strategies more accessible ? Do certain combinations of strategies outcompete single strategies ? And finally , how many different strategies does a set of parallel cultures adopt ? Experimentally evolving populations , characterizing their phenotypes , and finding the mutations responsible for evolution allows us to address these questions . We compared engineered and evolved solutions to the problem of growing on low sucrose concentrations . We engineered and tested three strategies that allow yeast cells to grow on low sucrose: forming multicellular clumps , importing sucrose before hydrolyzing it , and increasing invertase expression . We also evolved unengineered , laboratory strains to grow in the same conditions . All but one of the 12 clones selected from 10 independent populations form multicellular clumps as a result of incomplete cell separation , showing that this selection efficiently selects for multicellularity . In addition , 10 of the clones elevated invertase expression and 11 elevated hexose transporter expression , a strategy that we failed to anticipate , but none showed evidence of sucrose import . We combined bulk segregant analysis ( Michelmore et al . , 1991; Brauer et al . , 2006; Segrè et al . , 2006; Birkeland et al . , 2010; Magwene et al . , 2011 ) with whole genome sequencing to identify putative causal mutations . We recreated two of the evolved clones , one with five mutations and one with eight mutations , to show that mutations we had identified were indeed causal . Finally , we competed the evolved clones against their ancestor and found that adaptation in sucrose severely reduces fitness in high glucose .
We began by asking whether undifferentiated multicellularity was the only strategy that allowed yeast cells to grow on low sucrose . We hypothesized that there were two alternative strategies: increasing invertase expression , and importing sucrose and then hydrolyzing it inside the cell ( Figure 1 ) . Having previously tested undifferentiated multicellularity ( Koschwanez et al . , 2011 ) , we engineered the other two strategies and tested whether they allow cells to grow from low densities in low sucrose concentrations . Population growth requires both cell growth and cell proliferation . For simplicity , we refer to the combination of these properties as growth . 10 . 7554/eLife . 00367 . 003Figure 1 . Three engineered strategies for growth in low sucrose . Strategy 1 , form multicellular clumps , was previously verified ( Koschwanez et al . , 2011 ) . The results of testing strategy 2 , make more invertase , and strategy 3 , import sucrose , are shown in Figure 2 . All three strategies outcompete wild-type strains when the sole carbon source is 1 mM sucrose ( Table 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00367 . 003 Increased invertase expression will increase the rate of sucrose hydrolysis at the cell wall . Although cells will still lose most of the fructose and glucose to the environment , the increased hydrolysis rate will increase the monosaccharide concentration at the plasma membrane and lead to a higher rate of sugar import . We increased invertase expression by replacing the promoter of the invertase gene ( SUC2 ) with the GAL1 promoter in a yeast strain that is unable to utilize galactose ( see supplementary file 3 for all strains used in this study; Ingolia and Murray 2007 ) . As a result , galactose serves as a gratuitous inducer: it induces Suc2 but cannot itself be metabolized . Single cells were placed in each microwell of a 96-well plate , and incubated with a range of sucrose and galactose concentrations . Figure 2A shows that increasing invertase expression allows growth from a single cell in low sucrose concentrations; increasing inducer concentrations leads to better growth . 10 . 7554/eLife . 00367 . 004Figure 2 . Two strategies for growth from low sucrose concentrations . ( A ) Strong expression of secreted invertase allows growth from a single cell at low sucrose concentrations . All GAL1 promoter induction data is from the same yeast strain yJHK312 in which transcription of SUC2 is driven by the GAL1 promoter . Galactokinase ( GAL1 ) is deleted from this strain so that galactose acts as an inducer and not as a carbon source , and the Gal regulon has been engineered to produce a graded rather than a bistable response to increased galactose concentrations by overexpressing GAL3 from the ACT1 promoter ( Ingolia and Murray , 2007 ) . ( B ) Sucrose import allows growth from a single cell in low sucrose concentrations . The ‘SUC2 , import' strain yJHK372 expresses SUC2 from the SUC2 promoter and MAL11 from the ACT1 promoter . The ‘SUC2 , no import' strain yJHK222 expresses SUC2 from the SUC2 promoter . The ‘suc2-1cyt , import' strain yJHK373 expresses cytoplasmic invertase from the SUC2 promoter and MAL11 from the ACT1 promoter . The ‘suc2Δ , import' strain yJHK374 has SUC2 deleted and expresses MAL11 from the ACT1 promoter . For both ( A ) and ( B ) , single cells were inoculated by fluorescence activated cell sorting ( FACS ) into 150 µl wells at the given sugar and galactose concentration and grown without shaking for 85 hr at 30°C and the results shown are totals of three experiments; each experiment used one plate per sugar concentration , and each plate used 24 wells per strain or galactose concentration . In both figures , 2 mM glucose + 2 mM fructose is used as a positive control , and error bars refer to 95% binomial confidence interval using the adjusted Wald method . FRU is fructose , GLC is glucose , and SUC is sucrose . DOI: http://dx . doi . org/10 . 7554/eLife . 00367 . 004 An alternative strategy is importing sucrose before hydrolyzing it since some invertase molecules are retained in the cytoplasm rather than being exported by protein secretion . Mal11 is a maltose importer that can also import sucrose ( Stambuk et al . , 1999 ) , but is not expressed in our yeast strains ( Brown et al . , 2010 ) . We expressed Mal11 from a strong , constitutive promoter ( PACT1 ) in three different strains: a standard lab strain ( SUC2 ) , a strain that does not secrete invertase but produces cytoplasmic invertase ( suc2-1cyt ) ( Koschwanez et al . , 2011 ) , and a strain that lacks invertase ( suc2Δ ) . Figure 2B shows that both the SUC2 and suc2-1cyt strains that make the maltose importer can grow from single cells on as little as 1 mM sucrose . The nearly identical growth of the SUC2 and suc2-1cyt strains shows that the importer makes extracellular sucrose digestion dispensable , and the failure of the suc2Δ strains to grow shows that sucrose utilization after import still requires invertase . Thus three different strategies , clumping , increased invertase expression , and sucrose import , each allows yeast to grow from low cell density at low sucrose concentrations . We competed each of the three strategies against an unmodified strain to ask whether they could invade an ancestral population . Derivatives of the two competing strains , each expressing a different fluorescent protein , were mixed together and passaged on 1 mM sucrose . In each passage , the cells were grown together for almost eight generations and then diluted 200-fold into fresh medium . The competition was assessed by the number of passages required to eliminate the less fit strain , or the ratio between the two strains at the end of the sixth passage . The three engineered strains all outcompeted the ancestral wild-type lab strain in 1 mM sucrose ( Table 1 ) , although the strain engineered to express increased invertase ( EngHiInvertase ) was a much worse competitor than the other two ( EngClumpy , EngSucImport ) . 10 . 7554/eLife . 00367 . 005Table 1 . Fitness of engineered strains , evolved clones , recreated strains , and reverted strainsDOI: http://dx . doi . org/10 . 7554/eLife . 00367 . 005Strain 1Strain 21 mM sucrose1 mM glucose + 1 mM fructose80 mM glucoseEngClumpyWild-type lab+++00EngHiInvertaseWild-type lab+00EngSucImportWild-type lab+++00EvoClone1Ancestor++++−−−−−−−EvoClone2Ancestor++++0−−EvoClone3Ancestor++++−−−−−−−EvoClone4Ancestor+++++−−−EvoClone5Ancestor+++++−−−EvoClone6Ancestor+++++−−EvoClone7AAncestor++++−−−−−EvoClone7BAncestor+++++−EvoClone7CAncestor++++−−−−−−−EvoClone8Ancestor++++−−−−−EvoClone9Ancestor++++−−−−EvoClone10Ancestor++++−−−−−Recreated2Ancestor++++−−Recreated9Ancestor++++0−EvoClone2Recreated2++0EvoClone9Recreated9+−−−−EvoClone2Reverted2+++++−EvoClone9Reverted9++++++EvoClone2EngClumpy++0−−ace2ΔAncestor+++00gin4-W19* irc8-G57V mck1-G227Vfs249Ancestor+++−−++++Strain 1 eliminates strain 2 in 1–2 growth cycles . +++ Strain 1 eliminates strain 2 in 3–4 growth cycles . ++ Strain 1 eliminates strain 2 in 5–6 growth cycles . + Strain 1 > 75% of population after 6 growth cycles . 0 Neither strain is >75% of population after 6 growth cycles . − Strain 2 > 75% of population after 6 growth cycles . −− Strain 2 eliminates strain 1 in 5–6 growth cycles . −−− Strain 2 eliminates strain 1 in 3–4 growth cycles . −−−− Strain 2 eliminates strain 1 in 1–2 growth cycles . Growth cycle numbers are averaged over three independent experiments . Having shown that engineering can produce three different strategies for growth on low sucrose , we asked , ‘What would evolution do ? ' We experimentally evolved multiple , parallel cultures to grow well on low sucrose . Ten independent populations of budding yeast were serially diluted in minimal , 1 mM sucrose-containing medium . The starting strain for each population was a haploid , non-clumpy , prototrophic strain ( Figure 3A ) that constitutively expressed YFP and carried a DNA polymerase mutation ( POL3-L523D ) that elevated its mutation rate roughly 100-fold ( Jin et al . , 2005 ) . We used a mutator strain to increase the speed of adaptation . At each growth cycle , 5×105 cells were inoculated into 50 ml of medium ( Figure 3B ) . Over 25–35 serial dilutions , the time it took the culture to become cloudy fell from 2 weeks to 3 days , and all 10 populations ( named EvoPopulation1–10 ) displayed a clumpy phenotype ( Figure 3C ) . There was no loss of constitutive YFP expression . 10 . 7554/eLife . 00367 . 006Figure 3 . Evolved populations show a clumpy phenotype . ( A ) An ancestor derivative ( yJHK111 ) after growth in 1 mM sucrose . ( B ) Schematic of experimental evolution . Cells were inoculated in 1 mM sucrose media , grown to high density , and then 105 cells were reinoculated into fresh media for a total of 25–35 cycles . A sample was frozen down at each serial dilution . ( C ) Samples taken from the last time point of the evolved populations . Representative DIC images were taken with a 40× objective in a glass-bottomed , 96-well plate . All scale bars are 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 00367 . 00610 . 7554/eLife . 00367 . 007Figure 3—figure supplement 1 . Eleven of twelve clones show a clumpy phenotype . EvoPopulation7 had three morphologically distinct clones , named EvoClone7A , 7B , and 7C . For each of the remaining populations , we selected one clone from a group of eight morphologically indistinguishable clones . Representative DIC images were taken with a 40× objective in a glass-bottom 96-well plate . Scale bars are 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 00367 . 00710 . 7554/eLife . 00367 . 008Figure 3—figure supplement 2 . Size distribution of evolved clones . Sizes were measured on a Multisizer 3 Coulter counter . The thick line in the center of the box is the median size of the cells . The left and right side of the box correspond to the first and third quartiles . The lines extending from the boxes ( the whiskers ) correspond to the values within 1 . 5 times the inter-quartile range ( IQR ) . Clumps or cells with sizes outside the whiskers are shown as dots . Cells were grown in the media shown for at least 12 hr , and sizes were measured during log-phase growth . Note that the x-axis is log-scale . DOI: http://dx . doi . org/10 . 7554/eLife . 00367 . 00810 . 7554/eLife . 00367 . 009Figure 3—figure supplement 3 . EvoClone9 morphology changes in different media . Representative 40× DIC images of EvoClone9 cells that were grown to log phase in ( A ) 1 mM sucrose , ( B ) 80 mM glucose and ( C ) 1 mM glucose + 1 mM fructose . Scale bars are 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 00367 . 009 We asked which strategies individual populations had adopted . Eight clones were selected from each population and their morphology and growth on sucrose were examined . For nine of the ten populations ( all but EvoPopulation7 ) , all the clones were morphologically identical; one clone was selected from each of these populations and used for further studies . EvoPopulation7 produced three different phenotypes: one had medium-sized clumps ( EvoClone7A ) , one was not clumpy ( EvoClone7B ) , and one had large clumps ( EvoClone7C ) . Images of the clones are shown in Figure 3—figure supplement 1 , and size distributions of each clone are shown in Figure 3—figure supplement 2 . We tested the ability of the evolved clones to compete with their ancestor . Each of the evolved clones strongly outcompeted their ancestor , eliminating it from cultures within two passages on 1 mM sucrose ( Table 1 ) . We conclude that efficient use of public goods , liberated through extracellular hydrolysis , selects for the evolution of undifferentiated multicellularity . We asked how the evolved clones formed clumps . Wild yeast isolates can form multicellular clumps in two ways: flocculation ( Guo et al . , 2000 ) , in which separate cells stick to each other via cell wall-bound adhesins , or failure to separate daughters from their mothers because the cell wall that joins them is not digested after cytokinesis ( Yvert et al . , 2003 ) . To find which method the evolved clones used , we co-cultured two different colored versions of each evolved clone , using constitutive expression of different fluorescent proteins to mark the two versions . If the clumps formed by flocculation , many clumps would contain cells of both colors; if cells fail to separate , each clump would contain only one color since it would arise by the repeated division of a single cell . All the multicellular clones are the result of incomplete separation; each clump contains cells of only one or the other color . Representative images from two evolved clones are shown in Figure 4 ( images from the remaining clones are shown in Figure 4—figure supplement 1 ) . We checked our method by using two control strains: a flocculating strain that expresses a high level of Flo1 , a known adhesin , ( Smukalla et al . , 2008 ) and a lab strain that contains a wild isolate allele of AMN1 ( AMN1-RM11 ) , which prevents cell separation after many of the cell divisions ( Yvert et al . , 2003 ) . Figure 4 shows that the flocculating clumps from the strain expressing Flo1 contain both colors , whereas clumps from the AMN1-RM11 strain contain only one color . 10 . 7554/eLife . 00367 . 010Figure 4 . Clumpiness is due to failure to separate and not flocculation . Each image shows two genetically identical strains that are labeled with different fluorescent proteins , shown as magenta and green in the image . The strains were grown together from low density in 1 mM sucrose . ( A ) Lab strain with constitutively expressed FLO1 . Flocculation is evident from the mix of colors in a single clump . ( B ) Lab strain with the RM11 allele of AMN1 . ( C ) EvoClone2 . ( D ) EvoClone9 . The clumps in the AMN1-RM11 strain and the evolved clones are uniform in color , showing that clumpiness is due to failure to separate after cell division . Representative fluorescent images were taken with a 20× objective in a glass-bottomed , 96-well plate . All scale bars are 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 00367 . 01010 . 7554/eLife . 00367 . 011Figure 4—figure supplement 1 . Clumpiness is due to failure to separate and not flocculation . See Figure 4 for a description of the experiment . Images of EvoClone2 and EvoClone9 in Figure 4 are cropped versions of the images shown here for EvoClone2 and EvoClone9 ( cropped area outlined in yellow ) . Scale bars are 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 00367 . 011 We used RNA sequencing to examine the pattern of gene expression in the evolved clones . We isolated and sequenced RNA from the 12 evolved clones and two independent ancestor derivatives , all grown in 1 mM sucrose to log phase . In 10 of the 12 clones invertase ( SUC2 ) expression was elevated between 3- and 21-fold above their ancestor ( Table 2 ) ; EvoClone7C and EvoClone8 were the only clones with no significant change in SUC2 expression . 10 . 7554/eLife . 00367 . 012Table 2 . Significant changes in invertase and hexose transporter expressionDOI: http://dx . doi . org/10 . 7554/eLife . 00367 . 012Strain nameSignificant increases in invertase ( SUC2 ) expressionSignificant increases in hexose transporter expressionEvoClone15XHXT1 42XHXT2 5XHXT3 108XHXT4 91XEvoClone27XHXT4 9XEvoClone37XHXT4 4XEvoClone49XHXT1 4XHXT3 32XHXT4 14XHXT9 5XHXT11 4XEvoClone57XHXT1 4XHXT2 8XHXT3 6XHXT4 21XHXT6 7XHXT7 6XEvoClone69XHXT2 9XHXT4 12XEvoClone7A3XHXT2 7XHXT3 2XHXT4 92XEvoClone7B3XHXT2 10XHXT3 22XHXT4 103XEvoClone7CNot significantHXT1 3XHXT2 4XHXT3 10XHXT4 52XHXT6 4XEvoClone8Not significantHXT2 6XHXT3 5XHXT4 13XEvoClone94XHXT2 5XEvoClone1021XNoneSee supplementary file 1C for list of all genes that were significantly changed and their change in each evolved clone . RNA sequencing revealed another change that likely led to better growth in 1 mM sucrose: hexose transporter expression was elevated in 11 of 12 evolved clones . Yeast encodes at least 16 different hexose transporters , most encoded by members of the HXT gene family . The expression of HXT4 , which encodes a high-affinity glucose transporter , was elevated in 10 of these evolved clones ( Table 2 ) ; EvoClone10 was the only strain without increased expression of any hexose transporter . The effect of evolution on genes with known roles in cell separation is shown in Table 3 . Four genes involved in cell separation , AMN1 , CTS1 , DSE2 , and SCW11 , had significantly reduced levels in all 11 clumpy evolved clones , and three other genes implicated in cell separation showed reduced expression in 10 ( DSE1 , SUN4 ) , or 9 ( DSE4 ) of the clumpy clones . These seven genes are not decreased in expression in the non-clumpy evolved clone ( EvoClone7B ) or the control strain . This supports the argument that the clumps form by failure to separate . Supplementary file 1A lists genes whose expression was elevated or reduced in nine or more evolved clones ( and unchanged in the control strain ) . We suspect many of these genes may show differences in expression because the evolved clones grow much more rapidly than their ancestors in low sucrose medium . Supplementary file 1B lists the genes whose expression was increased or decreased at least tenfold in each of the ten evolved clones we examined . All of the approximately 1500 genes with significant expression level changes are listed in supplementary file 1C along with their level of change in each evolved clone . 10 . 7554/eLife . 00367 . 013Table 3 . Cell separation genes whose expression fell significantlyDOI: http://dx . doi . org/10 . 7554/eLife . 00367 . 013GeneReduced in multicellular clonesReduced in single cell cloneFunctionAMN111/110/1Cell separation proteinCTS111/110/1Cell separation , chitinaseDSE211/110/1Cell separation , possible glucanaseSCW1111/110/1Cell separation , possible glucanaseDSE110/110/1Cell separation , protein of unknown functionSUN410/110/1Cell separation , possible glucanaseDSE49/110/1Cell separation , possible glucanase Forming multicellular clumps and increasing hexose transporter expression both suggest that evolved clones are still hydrolyzing sucrose extracellularly . But since we had shown that engineering sucrose import allowed growth on low sucrose , we needed to rule out the possibility that some of the evolved clones were using this strategy . If an evolved clone depended on sucrose import , keeping it from secreting invertase would have little effect on its growth in sucrose . To test for this possibility , we removed the signal sequence ( Kaiser and Botstein , 1986; Perlman et al . , 1986 ) that directs Suc2's secretion from each of the evolved clones ( now named EvoCloneX-suc2-1cyt ) and competed these derivatives against the corresponding , unmodified , evolved clone . In each case , the evolved clone quickly outcompeted the version that did not secrete invertase ( Table 4 ) , demonstrating that growth of all the evolved clones depends on secreted invertase . To verify that this growth defect was not due to reduction in invertase expression , we measured , using reverse transcription followed by quantitative PCR ( RT-qPCR ) , the SUC2 expression of the two evolved clones ( EvoClone2 and EvoClone9 ) and their suc2-1cyt counterparts grown in 1 mM glucose . In both cases , expression of SUC2 was slightly greater in the suc2-1cyt strain but statistically insignificant over three independent trials . 10 . 7554/eLife . 00367 . 014Table 4 . Fitness of evolved clones after removal of SUC2 signal sequenceDOI: http://dx . doi . org/10 . 7554/eLife . 00367 . 014Strain 1Strain 21 mM sucrose1 mM glucose + 1 mM fructoseEvoClone1EvoClone1-suc2-1cyt++++0EvoClone2EvoClone2-suc2-1cyt+++0EvoClone3EvoClone3-suc2-1cyt++++0EvoClone4EvoClone4-suc2-1cyt+++0EvoClone5EvoClone5-suc2-1cyt++++0EvoClone6EvoClone6-suc2-1cyt+++++EvoClone7AEvoClone7A-suc2-1cyt+++0EvoClone7BEvoClone7B-suc2-1cyt+++0EvoClone7CEvoClone7C-suc2-1cyt+++++EvoClone8EvoClone8-suc2-1cyt+++++EvoClone9EvoClone9-suc2-1cyt++++0EvoClone10EvoClone10-suc2-1cyt+++++wtwt-suc2-1cyt+++0wt-suc2-1cyt-importerwt-suc2-1cyt++++0See Table 1 for definition of fitness measurements . We used two control competitions: in the first , a standard lab strain outcompeted a lab strain with a missing SUC2 secretion signal sequence ( suc2-1cyt ) ; in the second , a suc2-1cyt strain with MAL11 expressed from the ACT1 promoter outcompeted the suc2-1cyt strain that did not express a sucrose importer . Did evolving to grow faster on low sucrose affect the ability of the evolved clones to grow on other carbon sources ? Other experiments have resulted in antagonistic pleiotropy , where improved fitness in one environment often corresponds to reduced fitness in another ( Lenski , 1988; Wenger et al . , 2011 ) . We therefore determined the fitness of the evolved clones in three environments: low ( 1 mM ) sucrose , low monosaccharide ( 1 mM glucose plus 1 mM fructose , the hydrolysis products of 1 mM sucrose ) , and high ( 80 mM ) glucose . Each evolved clone quickly outcompeted the ancestor in low sucrose and lost to its ancestor in high glucose ( Table 1 ) . Four of the evolved clones also lost quickly on low monosaccharide; the remaining clones had approximately the same fitness as the ancestor , suggesting that most of the mutations acquired in low sucrose were selectively neutral in the equivalent monosaccharide concentration . The size distributions of each evolved clone was similar in all three conditions ( Figure 3—figure supplement 2 ) except for EvoClone9 , which formed much larger clumps in sucrose than it did in the other two media ( Figure 3—figure supplements 2 and 3 ) . We began the genetic characterization of the evolved clones by looking for the causal mutations . Because we used a strain that produced roughly one mutation per cell cycle , we predicted that neutral or slightly deleterious mutations would substantially outnumber the mutations that made cells grow faster in low sucrose concentrations . The neutral and nearly neutral mutations accumulate because they occur in lineages that were lucky enough to have strongly beneficial mutations; in the absence of sex , these mutations thus hitchhike during selection . To identify the causal mutations , we used bulk segregant analysis , which uses sexual reproduction to separate causal from hitchhiking mutations . The evolved strain was crossed to its ancestor , put through meiosis , and a large population of haploid spores was selected for the evolved phenotype . Mutations that confer a strong advantage on low sucrose will be present in almost all the selected spores , whereas those that do not will be present in roughly half the spores . We monitored the allele frequency in the selected pool by preparing DNA and sequencing it to roughly 100-fold coverage ( Figure 5 ) . We found a total of 80 putative causal mutations in the twelve evolved clones out of 1521 mutations , confirming that most mutations are non-causal ( Table 5 ) . To track the spread of the putative causal mutations through the evolved populations , we used Sanger sequencing to measure the allele frequency ( Gresham et al . , 2008 ) over time . Figure 6 shows the spread of putative causal mutations through two of the populations , EvoClone2 and EvoClone9; Figure 6—figure supplements 1 and 2 show the remaining populations . 10 . 7554/eLife . 00367 . 015Figure 5 . Schematic of bulk segregant analysis and evolved clone reconstruction . ( A ) A clone is selected from the population and then backcrossed to a derivative of its ancestor . The resulting diploid is sporulated , allowing the mutant alleles to randomly segregate among the haploid progeny . When the haploid progeny are selected for growth in low sucrose , only those cells with causal alleles ( red triangles ) remain; non-causal alleles ( blue diamond , square , and circle ) segregate randomly and are present in about half of the spores . ( B ) The ancestor , evolved clone , and pool of selected progeny are sequenced . Comparing the genome sequences of the ancestor and evolved clone reveals mutations . The allele frequency in the selected spores can then be estimated from the frequency of the reads in the pool of selected progeny . We classified any mutant allele present in >90% of the reads as a putative causal mutation ( Table 5 ) . ( C ) The wild-type alleles in the ancestor were replaced with the putative causal mutations to recreate the evolved clone ( Figure 7 ) . ( D ) Growth of the recreated strain was tested in low sucrose ( Table 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00367 . 01510 . 7554/eLife . 00367 . 016Figure 5—figure supplement 1 . Protocol for replacing alleles in yeast . ( 1 ) design the region of the gene to be included in the plasmid as follows: ( A ) . Determine the ‘usable' gene , or the promoter plus the open reading frame ( ORF ) plus the terminator . ( B ) Figure out which end of the usable gene is closer to the mutation . This is the side that will not be truncated . ( C ) Find a cut site >300 bp away from the mutation . If the new mutation is needed after transformation and before loopout , place the cut site toward far end of gene . This is the end of the gene that will be truncated . If old function is needed , place the cut site toward near end of gene . The region between the mutation and the cut site is now the ‘plasmid' region . The distance between the cut site and the mutation is necessary because of recision . ( D ) Extend the plasmid region at least 300 bp onto the opposite side of the cut site . This distance ensures efficient transformation . ( E ) Extend the plasmid region to the nearest end of the usable gene . This will ensure the gene is intact after transformation . ( F ) Find the distance from the mutation to each end of the plasmid region . Extend the plasmid region so that the length that does not include the cut site is larger than the length that does include the cut site . This ensures that loopout is more likely to result in the new mutation . ( 2 ) PCR amplify the plasmid region of the gene , treat it with polynucleotide kinase , and blunt-end ligate it into a digested URA3 plasmid backbone that has been treated with Antarctic phosphatase . Only pertinent regions of the plasmid are shown in the diagram . An additional yeast drug marker is useful in the plasmid backbone to verify that the insert has looped out . ( 3 ) Transform the yeast with the cut plasmid and select on—URA . This will integrate the linearized plasmid into the chromosome . PCR amplify and sequence the new allele to verify insertion; one primer should be outside the included region . ( 4 ) Grow the transformed yeast strain overnight in YPD to allow the insert to loop out through homologous recombination and plate on 5FOA . Cells will only grow on 5FOA if the URA3 marker has looped out . PCR amplify and sequence the new allele with both primers inside the included region to verify that the new allele is the only copy of the gene remaining . Replica plate to YP 2% acetate to eliminate petite mutants . DOI: http://dx . doi . org/10 . 7554/eLife . 00367 . 01610 . 7554/eLife . 00367 . 017Table 5 . Putative causal mutations in each evolved cloneDOI: http://dx . doi . org/10 . 7554/eLife . 00367 . 017Strain nameNominal generationsNumber of mutationsNon-synonymous and promoter mutations segregating at evolved allele frequency >90%Nucleotide changeAmino acid changeMutant allele reads/total readsEvoClone130771ACE2703 G→TE235*77/79IRA18987C→GT2996S99/102PHO87196 G→CA66P55/55RGT13157C→TQ1053*72/74SAN11707C→AN569K76/77SIN4383 G→AG128D106/106UBR11916T→AL639*42/44EvoClone2273115ACE2968T→AL323*110/110CSE2100_100delTS35Rfs54102/107IRA17657A→TS2553C72/75MTH1459_459delCH154Tfs15693/100UBR1524 G→AC175Y104/112EvoClone3301252IRA28081C→AS2694Y75/81IRC8365T→CL122P84/88NAT11782 G→AW594*84/87SYP11376C→TT459I90/96EvoClone422995ACE2670 G→TE224*93/93RGT12494_2495insTL832Ffs834135/139SIN4382 G→AG128S104/105UBR11916T→AL639*56/61EvoClone5237120ACE2565C→TQ189*149/152ARO2371C→GA124G93/101MCK138_38delGG14Dfs22152/158SNF271 G→TR24I94/100SNF31235T→AV412E148/148EvoClone6232110ACE2507_507delTN169Kfs177142/146GCN2892A→GN298D115/117GPB2235 G→TE79*144/145MTH1152_152delGS51Ifs56110/112NRG1371C→AS124*92/95RAD6191C→AP64H127/135EvoClone7A242105ACE21901C→AS634*84/87RAD61225T→AN75K97/97EvoClone7B24294GCN3176C→AS59Y79/87IRA27049_7049delCA2350Gfs2354129/132RAM1566T→AL189Q133/137SAN11464C→GN488K97/100SNF3692C→AA231D87/90EvoClone7C242115GCR2533T→AL178Q148/154IRA27049_7049delCA2350Gfs2354166/169PDR12527 G→AD843N145/159PUF41960C→TQ654*184/202EvoClone8253122ACE2−379 G→APromoter103/103AXL2700T→CS234P87/87ERG1427 G→AE143K160/163HXK193A→TE31D200/217IFM11724T→AI575N130/134MIT1188 G→AW63*131/142SKS11311C→GY437*152/152SNF31237 G→AE413K101/101UBC5443A→GD118G137/141UBR13859_3859delGG1287Dfs1345170/179EvoClone9265126ARE1−10 G→TPromoter127/127GCN24582A→CI1528L125/137GIN457 G→AW19*79/81IRC8170 G→TG57V118/119MCD1524C→TS175L115/115MCK1675_675delGG227Vfs249114/116MED11009C→GL337V104/1131465 G→TE489*118/125UBR13148_3148delCL1050Yfs1063141/145EvoClone10242196ACE21874A→TQ625L121/122AXL2432_432delCY145Mfs154161/164BPH12369C→AS790Y144/144DNF22351T→CF784S90/97ECM53466 G→AD1156N127/128ENP21129T→AF377I127/132GAC1−7T→APromoter106/117HTZ1−369T→CPromoter69/73KEM12268 G→AM756I148/148MCD1−28 G→TPromoter153/153MPT52409T→AL590*146/146MRPS17325 G→AD109N141/147NUT12582C→AS861*166/169PRC1−283 G→APromoter138/148RGT12060G→TG687V91/91SAC61736A→TK542M125/137TOP31679T→CV560A100/103UBR156T→AL19Q130/140WHI2187 G→TE63*138/139WTM2−297T→APromoter125/129Nomenclature based on ( den Dunnen and Antonarakis 2000 ) : Mx→y: nucleotide change from x to y at base M , starting at base 1 ( negative indicates promoter region ) . M_Ndelx: Deletion ( Insertion:ins ) of nucleotide x from base M to N . XNY: amino acid change from X to Y at codon N . * indicates stop codon . XNYfsN: as above , plus a frame shift mutation that results in stop codon at N . The following mutations likely hitchhiked and were not included in this table: EvoClone3: atg4 with sin4; EvoClone7A: slx4 promoter with ace2; EvoClone7B: thi3 with ram1 , crt10 with ira2 , EvoClone8: crh1 promoter with ubr1; EvoClone9: brr1 with MED1 , nsp1 promoter with irc8; EvoClone10: pri2 promoter with rgt1 , ino80 with nut1 . This claim is based on the genetic linkage between the two alleles and the lower allele frequency of the mutation we argue is hitch-hiking relative to the putative causal mutation . The following mutations are not shown in the time courses in Figure 6 and Figure 6—supplements 1 and 2 because they were present at frequencies of less than 5% of the final population: EvoClone7C: puf4-Q654*; EvoClone8: mit1-W63*; EvoClone9 mcd1-S175L; EvoClone10: top3-V560A The following mutations were in the original , time zero strain and are not included in this table even through they segregated at > 90%: EvoClone2: ira1-F664I; EvoClone8: phm8-I97N , rpl37a ( −52T→G ) , vta1-A247V , yor1-E393D; EvoClone9: vta1-A247V; EvoClone10: aim32-E241G , irs4-N257S , nnt1 ( −427T→C ) , prp9-N155S , yrb1-N120I . 10 . 7554/eLife . 00367 . 018Figure 6 . Putative causal mutation frequency at time points during the evolution . The alleles at the indicated time points were sequenced using Sanger sequencing , and frequencies were estimated from peaks in the trace plots . See Figure 6—figure supplements 1 and 2 for the other evolved populations . See Table 5 for amino acid and nucleotide changes . DOI: http://dx . doi . org/10 . 7554/eLife . 00367 . 01810 . 7554/eLife . 00367 . 019Figure 6—figure supplement 1 . Putative causal mutation frequency at time points during the evolution for EvoClone 1 , 3 , 4 , 5 , 6 , 7A , 7B , and 7C . See Figure 6 . DOI: http://dx . doi . org/10 . 7554/eLife . 00367 . 01910 . 7554/eLife . 00367 . 020Figure 6—figure supplement 2 . Putative causal mutation frequency at time points during the evolution for EvoClone 8 and 10 . See Figure 6 . DOI: http://dx . doi . org/10 . 7554/eLife . 00367 . 020 Over all 12 clones , the 80 putative causal mutations lie in or near 53 genes . Two genes , ACE2 and UBR1 , were mutated in at least half the clones , and groups of genes in three pathways , involved in glucose sensing ( the RGT1 group ) , growth regulation ( the IRA1/IRA2 group ) , and transcription ( the Mediator group ) were also mutated in many clones . We suspect that most of the putative causal mutations are loss of function mutations . Many alleles of the two most frequently mutated genes , ACE2 and UBR1 were nonsense mutations . In addition , amongst the 39 genes that were only mutated once , 12 ( 31% ) of the mutations were nonsense mutations . This suggests that many of the mutations in the remaining 27 genes are loss of function mutations: for two genes that have been extensively studied , URA3 and CAN1 , 31% of strong loss of function mutations are nonsense mutations ( Lang and Murray , 2008 ) . ( Because loss-of-function mutations are typically recessive , we use lower case italic nomenclature to indicate the putative causal allele ( e . g . , irc8-G57V ) , and upper case italics to indicate the wild-type allele ( e . g . , IRC8 ) . See the notes for Table 5 for a description of the mutation nomenclature . Roman text ( e . g . , Irc8 ) indicates the protein ) . Supplementary file 2 is a summary of the mutations in each gene pathway . Ace2 activates the transcription of enzymes that degrade the septum that connects mother to daughter cell ( Colman-Lerner et al . , 2001; Sbia et al . , 2008 ) . ACE2 was mutated in eight evolved clones; six of these mutations were nonsense mutations scattered through the open reading frame . All six of these clones had at least tenfold reductions in the expression of three genes known to aid in cell separation ( CTS1 [chitinase] , DSE1 , and DSE2 [supplementary file 1B] ) in agreement with the original finding that ace2Δ mutants are clumpy ( Dohrmann et al . , 1992 ) . Ubr1 is an E3 ubiquitin ligase in the N-end rule pathway . Ubr1 targets proteins with certain N-terminal amino acids , which are exposed by proteolytic cleavage or removal of the N-terminal methionine , for degradation ( Bartel et al . , 1990 ) . UBR1 was mutated in 6 of the 12 clones; four mutations were nonsense mutations , strongly suggesting that the remaining two are also a loss of function mutations . Rad6 , the ubiquitin-conjugating enzyme that forms a heterodimer with Ubr1 ( Dohmen et al . , 1991 ) , had a missense mutation in a seventh evolved clone . The Rgt1 pathway controls cells' response to external glucose concentrations . In high glucose , Rgt1 , a transcription factor , represses the expression of high affinity hexose transporters such as HXT4 ( Ozcan and Johnston , 1999 ) . In low glucose , Snf3 , a membrane-spanning glucose sensor , induces degradation of Mth1 , a protein that is needed for Rgt1 to extert its repressive effects ( Broach , 2012 ) . One of the genes in the Rgt1 pathway was mutated in 8 of the 12 clones: SNF3 ( 3 missense ) , MTH1 ( 2 nonsense ) , and RGT1 ( 2 nonsense , 1 missense ) ; no clones had two mutations in the pathway . Because Mth1 and Rgt1 work together to repress genes involved in growth on poor carbon sources , we expect that the mutations we isolated are loss of function mutations . MTH1 loss of function mutations have been selected by others in glucose-limited chemostats ( Kao and Sherlock , 2008; Kvitek and Sherlock , 2011 ) . On the other hand , because Snf3's activity leads to the induction of glucose-repressed genes , we speculate that the SNF3 mutations may result in gain of function . One of our mutations , snf3-A231D , lies near a known dominant mutation , SNF3-R229K ( Ozcan et al . , 1996 ) , that elevates hexose transporter expression . The Ras-cAMP pathway regulates cell growth in response to nutrients ( Broach , 2012 ) . Ras1 and Ras2 are small G proteins whose activity responds to nutrient sensing and who stimulate the activity of adenyl cyclase , which produces cAMP . Ira1 and Ira2 both encode GTPase activating proteins that inactivate Ras by stimulating its intrinsic GTPase activity ( Tanaka et al . , 1989 , 1990 ) . IRA1 or IRA2 were mutated in four unrelated evolved clones ( 3 missense , 1 nonsense ) . Mutations that increase cAMP-dependent protein kinase activity are known to prevent starvation-induced cell cycle arrest ( Matsumoto et al . , 1983 ) . By increasing Ras activity , loss of function mutations in IRA1 and IRA2 will elevate cAMP , which may allow cells to grow at lower external sugar concentrations than their ancestors . Mediator is a multiprotein , global regulator of eukaryotic transcription . We found mutations in four of the at least 20 proteins that make up Mediator ( Myers and Kornberg , 2000 ) : SIN4 ( 2 missense mutations ) , CSE2 ( 1 nonsense mutation ) , MED1 ( 1 nonsense mutation ) , and NUT1 ( 1 nonsense mutation ) . Both mutations in SIN4 occurred in residue 128 ( G128D , G128S ) and one , sin4-G128D , was also identified in a screen for suppressors of a deletion in SWI6 . Swi6 associates with Swi4 to form SBF ( SCB [Swi4-Swi6 cell cycle box] binding factor ) , a complex that regulates transcription early in the yeast cell cycle ( Li et al . , 2005 ) . The SIN4 mutations thus may cause increased expression of a gene normally activated by SBF . The hexose transporter gene , HXT3 , ( Iyer et al . , 2001 ) is a known target of SBF and its was strongly elevated ( 108-fold and 32-fold ) in the two clones ( EvoClone1 and EvoClone4 ) carrying SIN4 mutation ( Table 2 and 5 ) . We analyzed read depth across all evolved clones and saw three major duplication or deletion events: ( 1 ) chromosome 3 in EvoClone10 was duplicated in the region between 152 and 172 kb from the left end of the chromosome . This doubling of read depth was reduced to a 50% increase in the segregated pool , indicating that the duplication was not causal . ( 2 ) The number of P-Type ATPases at the ENO1/2/5 locus was reduced from 3 to 2 in EvoClone6 and EvoClone10 . This reduction was also present in the backcrossed strain , indicating that it may have been causal . ( 3 ) The number of hexose transporter gene repeats at the HXT6/7 locus increased from two in the ancestral strain to three in EvoClone7C . This amplification was also present in the backcrossed strain , indicating that it also may have been causal . Amplification of the HXT6/7 locus has been found in other experimental evolutions ( Brown et al . , 1998; Gresham et al . , 2008; Kao and Sherlock , 2008 ) . Are the putative causal mutations really responsible for the evolved phenotypes ? We addressed this question by engineering the candidate mutations from two clones , EvoClone2 and EvoClone9 , into their ancestor and asking if this manipulation reproduced the behavior of these clones . To recreate the sets of putative causal mutations , we replaced the five ancestral alleles in the ancestor with the five putative causal mutations in EvoClone2 to make Recreated2 and the eight putative causal mutations in EvoClone9 to make Recreated9 . Figure 7 and Figure 3—figure supplement 2 shows that the morphology and the clump size distribution of the recreated strains are similar to that of the evolved clones . 10 . 7554/eLife . 00367 . 021Figure 7 . Engineering in alleles can recreate the evolved and ancestral morphologies . The ancestral strain was converted to the evolved morphology by converting ancestral alleles to those of the putative causal mutations and the evolved strains were converted to the ancestral morphology by converting the putative causal mutations to their ancestral alleles . The strains were grown separately in 1 mM sucrose and then mixed . The top row shows EvoClone2 strains and the bottom row shows EvoClone9 strains . The ancestor constitutively expresses mCherry and is shown in yellow; the evolved clone constitutively expresses YFP and is shown in green . The recreated strain ( left ) and the reverted strain ( right ) constitutively express CFP and are shown in magenta ( recreated evolved ) and cyan ( reverted to ancestral ) . Representative confocal fluorescent ( left ) and brightfield ( right ) images were taken with a 60× objective in a glass-bottomed , 96-well plate . All scale bars are 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 00367 . 021 To assess the fitness of the reconstructed strains , we competed them against their ancestor and the evolved clones . Both the evolved and recreated clones outcompeted their ancestor in 1–2 passages ( Table 1 ) showing that they are much fitter on low sucrose . Both recreated strains were slightly less fit their evolved counterparts: there were fewer recreated than evolved cells after six passages ( 46 generations ) . This fitness defect has three possible causes: ( 1 ) the recreated clones are missing minor causal alleles , which we failed to find; ( 2 ) detrimental mutations were introduced during the multiple transformations required to make the recreated strains; or , ( 3 ) the evolved clone ( which is a mutator , unlike the recreated strain ) continued to evolve and adapt during the competition . To confirm that the putative causal alleles accounted for the evolved phenotype , we reverted these alleles in the evolved clones to their ancestral state . We replaced each of the mutant alleles in EvoClone2 and EvoClone9 with the ancestral allele . Both the resulting strains , Reverted2 and Reverted9 , grew poorly in 1 mM sucrose , mostly existed as single cells , and were quickly outcompeted by the evolved clones ( Table 1 , Figure 7 , and Figure 3—figure supplement 2 ) . We used bulk segregant analysis to ask whether individual alleles were required for the evolved phenotype . We crossed the recreated strains to their ancestors , isolated haploid spores , selected this population for growth on 1 mM sucrose , and measured the frequency of the evolved alleles in the selected population . All five mutations in Recreated2 were strongly selected for , six of the eight mutations in Recreated9 were strongly selected for , and one mutation in Recreated9 was moderately selected for , confirming that all but one of the thirteen putative causal mutations were causal for improved growth on low sucrose ( Figure 8 ) . We suspect that the eighth mutation in Recreated9 , gcn2-I1528L , may be a false positive . In the original spore selection , the mutant allele segregated at only 91% , the low end of our threshold , and the mutation changed a poorly conserved isoleucine to leucine , suggesting it is unlikely to have a major effect on the protein's activity . In sum , our recreation experiments show that the putative causal alleles , with a single exception , are both necessary and sufficient to produce the evolved phenotype . 10 . 7554/eLife . 00367 . 022Figure 8 . Bulk segregant analysis with the recreated strains verifies causal alleles and shows that alleles responsible for clumpiness are selected in low sucrose and not low monosaccharide . The recreated strains were backcrossed , sporulated , and selected in three different media: 1 mM sucrose ( low sucrose ) , 80 mM glucose ( high glucose ) , and 1 mM glucose plus 1 mM fructose ( low monosaccharide ) . The mutant allele fraction was estimated from Sanger sequencing across the allelic variants . The size of the data point ( small , medium , or large ) for each allele and media combination refers to one of three independently derived diploids . ( A ) Recreated2 allele segregation . ( B ) Recreated9 allele segregation . ( C ) Ancestor strain with ACE2 deleted ( ace2Δ ) has a clumpy phenotype . The ACE2 mutation in Recreated2 , a likely loss of function mutation that caused the clumpiness in EvoClone2 , was selected for in low sucrose and was not selected for in low monosaccharide . ( D ) Ancestor strain that has wild type alleles of IRC8 , MCK1 , and GIN4 replaced with the EvoClone9 alleles has a clumpy phenotype . All three mutant alleles were selected for in low sucrose and were not selected for in low monosaccharide . Representative DIC images were taken with a 40× objective in a glass-bottomed , 96-well plate . Scale bars are 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 00367 . 02210 . 7554/eLife . 00367 . 023Figure 8—figure supplement 1 . Change in HXT4 and SUC2 expression in various allelic combinations of Recreated2 compared to the ancestor . Change in RNA expression was measured using RT-qPCR using RNA isolated from a culture grown in 1 mM sucrose minimal media . Each point is the mean change in expression from three independent trials . The total length of the error bar is twice the size of the standard deviation . Ancestor alleles are shown in upper case and mutant alleles are shown in lower case . ( A ) Change in HXT4 expression . Allelic combinations containing mth1 have higher expression than those combinations with MTH1 . ( B ) Change in SUC2 expression . Note that the axes are different in the two plots . DOI: http://dx . doi . org/10 . 7554/eLife . 00367 . 023 We studied the role of the causal mutations in different growth conditions . Do the mutations selected on low sucrose increase , decrease , or have no effect on fitness on other carbon sources ? To ask this question , we selected spores from crosses between ancestral and recreated clones in low monosaccharide and high glucose . Several mutations were selected for in low sucrose but not in low monosaccharide: mutations in MTH1 , CSE2 , and ACE2 in Recreated2; and in MCK1 , IRC8 , and GIN4 in Recreated9 . The primary selective force for four of these six mutations is likely to be their ability to produce multicellularity: an ace2Δ strain is morphologically similar to EvoClone2 and a mck1-G227Vfs249 irc8-G57V gin4-W19* strain is morphologically similar to EvoClone9 ( Figure 8 ) . Each of these two strains also outcompetes the ancestor strain in low sucrose but not in low monosaccharide ( Table 1 ) . This shows that multicellular clumps were specifically selected by growth in low sucrose , rather than being a general response to low sugar concentrations . But clumpiness alone was not sufficient to match the growth of the evolved strains in sucrose—the engineered clumpy strain ( EngClumpy ) was not as fit in 1 mM sucrose as one of the evolved clones that had similar clump size ( EvoClone2 ) ( Table 1 ) . Our evolved clones are less fit than their ancestor on high glucose ( Table 1 ) . There are two explanations for this observation . The first is a direct result of selection: some mutations that improve growth on low sucrose reduce fitness on high glucose . The second appeals to the large number of mutations that hitchhiked with the causal mutations: some of the hitchhikers are neutral on low sucrose , but slow cell growth on high glucose . We distinguished these two possibilities by competing the evolved and recreated strains against their ancestor in high glucose . For both EvoClone2 and EvoClone9 , the recreated strains performed as well as or better than the evolved clones , but worse than the ancestor , suggesting that both causal alleles and hitchhiking alleles contribute to the evolved clones' reduced fitness in high glucose . We identified the causal alleles that impair growth on high glucose through bulk segregant analysis . When the progeny of the cross between ancestor and recreated strains are grown on high glucose , one of the causal mutations ( ubr1-C175Y ) in Recreated2 is selected against , three are approximately neutral , and one ( ira1-S2553C ) is strongly selected for . For Recreated 9 , one mutation ( gin4-W19* ) is very strongly selected against , one ( ubr1-L1050Yfs1063 ) is weakly selected against , three are roughly neutral , two are weakly selected for ( mck1-G227Vfs249 and are1 ( -10 G→T ) ) , and one ( mcd1-S175L ) is very strongly selected for in high glucose ( Figure 8 ) . Note that the mutations in UBR1 and GIN4 have opposite effects in low monosaccharide: ubr1 mutations are selected for in both low sucrose and low monosaccharide , whereas the gin4-W19* mutation is selected against in low monosaccharide . To analyze the effect of individual mutations on expression of HXT4 and SUC2 in EvoClone2 , we made seven strains , each with a different allelic combination of the five causal mutations . We then measured HXT4 and SUC2 expression using RT-qPCR in the seven new strains and in Recreated2 . Figure 8—figure supplement 1 shows the results . The increase in HXT4 expression is clearly a result of the MTH1 loss of function mutation ( mth1-H154Tfs156 ) : all combinations containing this allele have higher HXT4 expression than those containing the ancestral MTH1 allele . This result confirms the role of MTH1 in hexose transporter repression . The gradient in SUC2 expression across the allelic combinations indicates that SUC2 expression is a complex phenotype , which is under the quantitative control alleles at several genes . We also checked the clumpiness of each strain by manual observation using a microscope: the strains with the wild type ACE2 allele were not clumpy , while each strain with ace2-L323* was clumpy . This result confirms that the clumpiness in EvoClone2 is caused by the loss of function mutation in ACE2 .
In earlier work , we showed that clumps of yeast could grow in concentrations of sucrose where single cells could not . We speculated that the sharing of public goods ( in this case , the hydrolysis products of sucrose ) could select for multicellularity . Here , we confirmed this hypothesis: 11 of 12 clones evolved in low sucrose formed multicellular clumps caused by cells failing to separate after cell division ( Figure 3 and Figure 3—figure supplements 1–3 ) . RNA sequencing revealed two additional strategies that most clones used: 10 elevated invertase expression , a strategy we had engineered , and 11 elevated hexose transporter expression , a strategy we had not engineered . We found no evidence of another strategy we had engineered: sucrose import . We cannot eliminate the possibility that the clones import some sucrose , but when we kept cells from secreting invertase , their fitness in low sucrose was severely reduced , showing that their growth still depends on invertase secretion . We identified the mutations that led to increased growth on low sucrose and showed that these were necessary and sufficient to recreate the evolved phenotype in two of the evolved clones . The genes that were mutated identified pathways that control cell growth and nutrient utilization . The evolved cells clump because they fail to separate after cell division , rather than by flocculating ( Figure 4 ) . A loss of function mutation in ACE2 was likely the primary contributor to the clumpy phenotype in at least 6 of the 11 multicellular clones . We were able to recreate the clumpy morphology of EvoClone9 with mutations in MCK1 , IRC8 , and GIN4; note that EvoClone3 , one of the two other clumpy clones with wild type ACE2 , also had a mutation in IRC8 . Mutations in ACE2 are likely to be frequent because inactivation of this gene simultaneously reduces the expression of multiple genes needed for cell separation; mutations in the genes that Ace2 regulates would be likely to have smaller effects . None of our clones became clumpy by restoring function to AMN1 , the gene whose loss of function allele was selected for during laboratory domestication of budding yeast ( Yvert et al . , 2003 ) . This is not surprising since the target size for mutations that restore AMN1 to its wild type function is much smaller than the target size for inactivating ACE2 . Was the clumpy phenotype selected because it confers some other advantage , such as faster settling in unstirred cultures ( Ratcliff et al . , 2012 ) , rather than because it improves growth in low sucrose ? Three lines of evidence show that this is not true . First , none of the evolved clones had a strong fitness advantage over the ancestor in 1 mM glucose plus 1 mM fructose , the monosaccharide equivalent of 1 mM sucrose ( Table 1 ) showing that these clones were specifically adapted to sucrose . Second , a strain that was engineered to be clumpy ( EngClumpy ) outcompeted the ancestor in low sucrose , but not low monosaccharide ( Table 1 ) , demonstrating that low sucrose concentrations specifically selects for clumps , reinforcing our earlier findings on engineered multicellularity ( Koschwanez et al . , 2011 ) . Third , the alleles that recreate the clumpy morphology ( ace2-L323* in EvoClone2 and the combination of mck1-G227Vfs249 , gin4-W19* , and irc8-G57V in EvoClone9 ) are strongly selected for in low sucrose and are not selected for in low monosaccharide ( Figure 8 and Table 1 ) . Because the starting pool of segregants and the selection protocol was identical in both media , this result shows that more efficient sucrose utilization , rather than some other phenotype , selected for the clumpy phenotype . Analyzing gene expression and the role of individual mutations demonstrated that we had selected for more than multicellularity . Most ( 10 out of 12 ) of the evolved clones had increased invertase expression , even though the strain we had engineered to have high invertase expression had only a small advantage over a wild-type strain ( Table 1 ) . The advantage of increased invertase expression in a shaken culture of single cells is likely to be small since all cells share the monosaccharides that escape from the small , unstirred volume that surrounds each cell . But high invertase expression makes sense in conjunction with the other two strategies we observed , multicellularity and elevated monosaccharide import . In a clump , multiple cells share a larger unstirred volume and have preferential access to the monosaccharides their neighbors release ( Koschwanez et al . , 2011 ) . Increasing invertase expression will increase the rate of sucrose hydrolysis , and making more hexose transporters will increase the fraction of the monosaccharides that are imported rather than escaping to the bulk medium . Thus we expect that increasing invertase expression , making cell clumps , and making more hexose transporters will act together to increase fitness on low sucrose . Because low extracellular glucose induces the expression of both high affinity hexose transporters and invertase , it is likely that single mutations can increase levels of both types of proteins . In addition , clumpiness may indirectly elevate invertase expression: because secreted invertase expression peaks at roughly 0 . 5 mM external glucose ( Koschwanez et al . , 2011 ) , clumps may increase invertase expression because concentrating the cells in space collectively raises the local glucose concentration . Many of the mutations we selected are likely to affect the linked processes of sugar harvesting and the control of cell growth . As the level of potential nutrients in their environment falls , cells must make choices: how much of their resources they invest in trying to harvest nutrients from their environment , and how much of the imported nutrients they store to meet future challenges and how much they use to maximize their current growth rate . As an example , wild strains may stop growing and dividing at low external glucose levels because this decision improves their ability to survive if glucose levels continued to fall , even though they could support cell growth at the current glucose concentration if they expressed very high levels of hexose transporters and stored no sugar as glycogen . Influences on this type of decision are likely to explain repeated mutations in one gene ( UBR1 ) and three groups of genes , which control glucose's effect on transcription ( the RGT1 group ) , how nutrient availability controls cell growth ( the IRA group ) , and general transcription ( the Mediator group ) . UBR1 was mutated in half of the evolved clones . Four of the six mutations are nonsense mutations , suggesting that we selected for inactivation of this protein . We suspect that prevention of N-end rule mediated protein degradation increases the expression of genes that promote growth in carbon limited cells: in both evolved clones that we recreated , the evolved allele of UBR1 was selected for in sucrose and low monosaccharide and selected against in high glucose ( Figure 8 ) . The RGT1 group of genes gives external glucose concentrations the ability to control the expression of genes needed for growth in low glucose concentrations . Genes in this pathway were causally mutated in 8 of 12 evolved clones . The pattern of mutations we observe in these genes is consistent with selection to reduce the repression mediated by Rgt1 and Mth1 . We also found , by using RT-qPCR on one of the reconstructed clones , that a loss-of-function mutation in MTH1 was responsible for the increase in HXT4 expression ( Figure 8—figure supplement 1 ) . Mutations in one of the components of the Mediator complex , Sin4 , had previously been shown to be involved in genes required for passage through Start , the transition in the yeast cell cycle that is most sensitive to nutritional signals . We recovered two mutations in this gene and found they were both associated with a strong increase in expression of Hxt3 , a low affinity glucose transporter . We saw mutations in IRA1 and IRA2 in four independent lineages . These mutations are likely to reduce the activity of Ira1 and Ira2 and thus increase the activity of Ras at low sugar concentrations , promoting growth . A mutant of IRA1 in EvoClone2 was one of only two mutants of 13 in the recreated strains that was strongly selected for in all three conditions , and mutants of IRA1 have been found in a yeast experimental evolution done in low glucose ( Kao and Sherlock , 2008 ) and rich media ( Lang , personal communication , October 2012 ) . We suspect that Ira1 and Ira2 reduce the activity of Ras and thus the growth rate of cells on suboptimal carbon sources ( low sugar concentrations , ethanol ) . Because we used batch cultures , even the high glucose cultures went through an environmental shift after they had fermented all the glucose to ethanol . Disruption of either IRA1 or IRA2 increases sensitivity to heat shock ( Tanaka et al . , 1989 , 1990 ) . We suggest that we have selected for mutants than ignore pathways designed to induce cells exposed to unfavorable conditions to reduce their growth rate and induce programs that will protect them if conditions worsen . We took an integrated approach to understanding evolution . Experimental evolution has long been used to study nutrient adaptation in yeast ( Paquin and Adams , 1983; Ferea et al . , 1999 ) and bacteria ( Levin et al . , 1977; Lenski et al . , 1991; Barrick et al . , 2009 ) . Although evolved bacteriophage isolates were sequenced in the 1990s ( Bull et al . , 1997 ) , the cost of cellular genome sequencing has only recently dropped to a level where it is possible to find single nucleotide polymorphisms in an evolved microbe ( Barrick et al . , 2009; Araya et al . , 2010 ) . We combined engineering , experimental evolution , bulk segregant analysis , and genetic reconstruction to build a picture of how budding yeast evolve in response to a specific challenge: the need to grow on a nutrient that is released into the local environment by extracellular hydrolysis . Testing hypothetical strategies by engineering allowed us to look for their phenotypic signatures , such as the increased expression of hydrolytic enzymes , in evolved populations . Having a method to sift the minority of causal mutations from the majority of hitchhikers allowed us to use mutators and increase the speed of evolution . The ability to swap evolved and ancestral alleles made it possible to verify our methods for identifying causal mutations and show that we had identified mutations that were either generalists , promoting growth across all the environments we tested , or specialists showing antagonistic pleiotropy by being strongly beneficial in one environment and deleterious in another . Finally , performing multiple experiments in parallel allowed us to identify mutations in particular genes and pathways as contributing strongly to evolutionary success . We speculate that unicellular organisms have evolved a combination of three strategies to access nutrients that are digested extracellularly: increasing secretion of an extracellular hydrolase , increasing the import of its products , and forming multicellular clumps . An increase in nutrient transport seems the least costly , but its effect as a sole strategy is limited if the local concentration of the hydrolysis products remains unchanged . An increase in enzyme secretion will increase the level of the hydrolysis products , but enzyme production is costly and allows unrelated cells from the same or different species to ‘cheat' and steal simple nutrients without incurring the cost . The formation of multicellular clumps allows the hydrolysis products to be shared among direct kin and reduces the risk of cheating ( Koschwanez et al . , 2011 ) . We have shown that yeast , when selected for growth in low sucrose media , used a mixture of all three strategies . The combination of laboratory engineering and evolutionary analysis shown here and in our previous work supports the speculation that the sharing of public goods was the initial selection for undifferentiated multicellularity . For organisms with rigid cell walls like yeast , algae , and bacteria , cells must secrete enzymes to separate mother from daughter after cytokinesis . The target size for mutations in these enzymes and the control circuits that regulate their production will be large , making it easy to evolve undifferentiated multicellularity . Although lab strains of yeast were selected to be unicellular during their domestication ( Mortimer and Johnston , 1986 ) , the level of clumpiness varies in wild isolates of yeast , and undifferentiated multicellularity may have been recurrently selected for its advantage in harvesting nutrients or for protection from stress ( Smukalla et al . , 2008 ) . Organisms without a cell wall , such as animal cells , would need another innovation , such as cell-adhering cadherins ( Abedin and King , 2008 ) , in order to evolve multicellular clumps . But even in these organisms , cells must adhere to each other during sexual reproduction , implying the existence of adhesive proteins that could be modified to allow the evolution of multicellular clumps . If undifferentiated multicellularity is so easy to evolve in eukaryotes , then why did it take so long for fossils of differentiated multicellular organisms to appear in the fossil record ? Indeed , there is suggestive evidence of undifferentiated eukaryotic multicellularity 2 billion years ago ( El Albani et al . 2010 ) , but the radiation of more complex multicellularity is not seen for over a billion years later . This disparity has received a large amount of attention in paleontology , particularly in the context of the seemingly rapid radiation of the animals at the base of the Cambrian . Possible explanations for this ‘Cambrian explosion' include novel selective forces due to the environmental changes that occurred shortly before the Cambrian or the possibility that evolutionary arms races among a few early species drove the radiation ( Marshall , 2006; Peters and Gaines , 2012 ) . Alternatively , the evolution of differentiated multicellularity may have been slow and gradual and we simply lack the fossil evidence . Whatever the case , it is interesting to ask why the first differentiation might have arisen . From our data , we speculate that the effectiveness of clumping as a strategy could be improved if a limited number of cells increase their enzyme secretion . Such a simple division of labor may have been the precursor to complex multicellularity in the lineages that lead to animals and other complex multicellular species .
Unless otherwise noted , the experimental evolution and all experiments were conducted in minimal ( no amino acids or nucleotides ) synthetic media with the following two exceptions: ( 1 ) the auxotrophic strains expressing high levels of Flo1 required the addition of leucine , histidine , and lysine , and ( 2 ) strains undergoing galactose induction in Figure 2 were pregrown in YP 2% glycerol ( 10 g/l yeast extract , 20 g/l peptone , and 2% [v/v] glycerol ) plus the indicated concentration of galactose . The minimal synthetic media was made with refrigerated sugar stocks and refrigerated 10× yeast nitrogen base ( YNB ) that was based on Wickerman's recipe ( Wickerham , 1951 ) but without riboflavin , folic acid , and inositol ( see [Koschwanez et al . , 2011] for the recipe ) . 200 units/ml Penicillin and 0 . 2 mg/ml streptavidin were added to the experimental evolution media to prevent bacterial contamination . Unless otherwise noted , all chemicals used in this research were purchased from Sigma-Aldrich ( www . sigmasldrich . com ) . Each of the 10 evolved cultures was derived from a haploid MATa W303 prototrophic strain that expressed a yellow fluorescent protein ( YFP ) variant from the constitutive ACT1 promoter . See supplementary file 3 for a description of the starting strains and their construction . To start the evolution , a single colony from each strain was inoculated into 80 mM glucose media and grown for 7–20 generations , and 5×105 cells were then inoculated into a 125 ml flask containing 50 ml of 1 mM sucrose media . The flasks were placed at 30°C on a shaker rotating at 120 rpm . When there was visible growth in a culture ( >106 cells/ml ) , it was passaged as follows . First , the culture was spun down and resuspended in 1 ml of yeast nitrogen base ( YNB ) to ensure that non-sucrose carbon sources , created by invertase activity or cellular metabolism , were not inoculated into the new cultures . Second , a sample of resuspended cells were mixed with glycerol and frozen at −80°C . Third , the cell density was measured with a Coulter counter ( www . beckmancoultermedia . com ) . Finally , the resuspended cells were vortexed and 5×105 cells were inoculated into a new 125 ml flask containing 50 ml of warm 1 mM sucrose media . When the cells became too clumpy to count accurately in the Coulter counter , the cultures were instead diluted at a 500:1 ratio into fresh media . To pick clones , eight replicates of each culture were serial diluted 2:1 twelve times in 1 mM sucrose media in 96-well plates such that the last dilution contained an estimated average of less than one cell per well for each replicate . For each population , there were thus eight parallel sets of dilutions . The plates were incubated on a plate shaker at 30°C for 3 days , and each well was examined using a microscope for growth in sucrose and for morphology . A well was assumed to contain a clonal population at the lowest starting cell density where growth was seen if there was a lower density where growth was not seen in any of the eight wells . Each of the selected clones was verified to grow from low density in 1 mM sucrose within 2 days . URA3 was deleted from each YFP-labeled MATa evolved clone . The ura3Δ strain was mated with yJHK519 , a MATα , trp1Δ::kanMX4 ura3::PSTE2-URA3 derivative of the ancestor labeled with CFP . In this strain , the endogenous URA3 promoter is replaced with the STE2 promoter , which is only induced in MATa cells , making it possible to select for MATa spores after meiosis . Mating was performed by mixing cells from the two strains together on a YPD plate with a toothpick and growing overnight at 30°C . The mating mixtures were then plated onto G418 , -TRP plates , and a diploid strain was selected from a colony on the plate . To sporulate the diploid strains , cultures were grown to saturation in YPD , and then diluted 1:50 into YP 2% acetate . The cells were grown in acetate for 12–24 hr and then pelleted and resuspended in 2% acetate . After 4–5 days of incubation on a roller drum at 25°C , sporulation was verified by observing the cells in a microscope . To digest ascii , 1 ml of the sporulated culture was pelleted and resuspended in 50 µl 10% Zymolyase ( www . zymoresearch . com ) for 1 hr at 30°C . 400 µl of water and 50 µl of 10% Triton X-100 were added , and the digested spores were sonicated for 5–10 s to separate the tetrads . Tetrad separation was verified by observing the cells in a microscope . The separated tetrads were then spun down slowly ( 6000 rpm ) and resuspended in growth media . To select haploid spores , the entire digested spore culture was added to 50 ml of 1 mM glucose + 1 mM fructose minimal media and growth to saturation . This media selected for TRP1 , haploid MATa cells: neither haploid MATα nor diploid MATa/MATα cells can express URA3 from the STE2 promoter . Each culture was selected for growth in 1 mM sucrose through four passages ( inoculation to saturated growth ) as follows: one 100:1 dilution , and three 500:1 dilutions ( ≈34 generations total ) . The cells were spun down and resuspended in YNB during each dilution in order to eliminate the non-sucrose carbon sources from the media . Genomic DNA was made from the final saturated culture . Bulk segregant analysis of Recreated2 and Recreated9 followed the same procedure with one exception: the cells were not spun down in between dilutions . Instead , 100 µl of culture was added directly to 50 ml of fresh media for a 500:1 dilution . To prepare genomic DNA , the culture was pelleted and resuspended in 50 µl of 1% Zymolyase in 0 . 1 M , pH 8 . 0 NaEDTA . The cells were incubated for 30 min at 37°C to digest the cell wall , and then the cells were lysed by adding 50 µl 0 . 2 M NaEDTA , 0 . 4 M pH 8 . 0 Tris , 2% SDS and incubated at 65°C for 30 min . 63 µl of 5 M potassium acetate was added , and the mixture was incubated for 30 min on ice . The insoluble residue was then pelleted , and 750 µl of ice-cold ethanol was added to 300 µl of the supernatant to precipitate the DNA . The DNA was pelleted , and the pellet was resuspended in 0 . 2 mg/ml RNAase A . After 1 hr of incubation at 37°C , 2 µl of 20 mg/ml Proteinase K was added , and the solution was incubated for an additional 2 hr at 37°C . The DNA was again precipitated by adding 130 µl isopropanol . The DNA was pelleted , briefly washed with 70% ethanol , repelleted , and resuspended in 100 µl 10 mM Tris , pH 8 . 0 . To isolate RNA , cells in log-phase growth were fixed by adding 6 ml of culture to 9 ml of ice-cold methanol and then incubating at −20°C for 10 min . Cells were pelleted at 4°C , resuspended in 1 ml of RNAase-free ice-cold water in a 2 ml cryogenic storage vial , and then repelleted at 4°C . RNA was then isolated with acidic phenol using a published protocol ( Collart and Oliviero , 2001 ) . Isolated RNA was treated with DNase I ( Thermo Scientific EN0525 , www . thermofisher . com ) according to manufacturer's instructions . Intact RNA was verified by observing two sharp rRNA bands using agarose gel electrophoresis . cDNA was made from 100–200 ng of RNA using Thermo Scientific Maxima First Strand cDNA Synthesis Kit for RT-qPCR ( K1641 ) . The 20 µl reaction was diluted tenfold and 11 µl of the diluted sample was used for real time qPCR using Thermo Scientific Maxima SYBR Green/ROX qPCR Master Mix ( K0222 ) . Amplification efficiency of the primers was verified by generating standard curves with four serial dilutions . The following primers were used: SUC2 ( GCTTTCCTTTTC-CTTTTGGCTGG , TCATTCATCCAGCCCTTGTTGG ) ; HXT4 ( TTGGGTTACTGTACAAACTACG , TGTCA-TACCACCAATCATAAAC ) ; ALG9 ( GTTTAATCCGGGCTGGTTCCAT , TAGACCCAGTGGACAGATAGCG ) . ALG9 was used for normalization ( Teste et al . , 2009 ) . The ΔΔCT method was used to find change in RNA expression ( Livak and Schmittgen , 2001 ) . Three independent trials were performed for each reported RT-qPCR result and the data reported is the mean difference in expression between two strains . DNA and RNA libraries were prepared for sequencing using the Illumina TruSEQ kit ( www . illumina . com ) and were sequenced on an Illumina HiSeq 2000 . Mean coverage across the genome was as follows: ancestor DNA—56× , evolved clone DNA—10× minimum , sporulated pools—90× minimum , RNA—40× minimum . Single end , 50 bp reads were used for the ancestor DNA , evolved clone DNA , and RNA . Paired end , 100 bp reads were used for the sporulated pool DNA . Sequencing data is deposited in the Sequence Read Archive . DNA sequences were aligned to the S288C reference genome r64 ( downloaded from the Saccharomyces Genome Database , www . yeastgenome . org ) using the Burrows-Wheeler Aligner ( bio-bwa . sourceforge . net ) ( Li and Durbin , 2009 ) . The resulting SAM ( Sequence Alignment/Map ) file was converted to a BAM ( binary SAM ) file , sorted , indexed , and made into a pileup format file using the samtools software ( samtools . sourceforge . net ) ( Li et al . , 2009 ) . Indels were realigned locally using GATK ( www . broadinstitute . org/gatk/ ) ( McKenna et al . , 2010 ) , and variants were called from the pileup file using the Varscan software ( varscan . sourceforge . net ) ( Koboldt et al . , 2012 ) . To perform the segregation analysis , we wrote a custom sequencing pipeline in Python ( www . python . org ) , using the Biopython ( biopython . org ) and pysam ( code . google . com/p/pysam/ ) modules , that finds sequence variants between the ancestor and clone , classifies each variant as a nonsynonymous coding region , synonymous coding region , or promoter mutation , and ranks each mutation by its segregation frequency . All software written for this analysis is publicly available at https://github . com/koschwanez . We used the following criteria to select putative causal mutations: ( 1 ) the evolved clone was mutated relative to the strain used to inoculate the time zero culture , ( 2 ) the mutation caused a nonsynonymous substitution in a coding region or changed the promoter sequence ≤500 bp upstream of the coding region start site , and ( 3 ) the mutation appeared in over 90% of the reads in backcrossed pool . Putative causal mutations were manually verified by looking at aligned reads . RNA sequences were aligned to the S288C reference genome using TopHat ( Trapnell et al . , 2009 ) , and significant differences in expression between the ancestor and the evolved clone were called using the default setting in Cufflinks ( Trapnell et al . , 2010 ) . The Cufflinks package uses the log of the ratio of expression in two conditions together with an estimate of the gene's variance to generate a t-value that it uses in a Students t-test . We used one ancestor derivative , yJHK110 , as the reference strain to compare expression with the evolved clones , and the other ancestor derivative , yJHK111 , as a control strain in the comparison . Unless otherwise noted , data analysis was performed in the R programming language ( www . r-project . org ) and plots were generated using the R library ggplot2 ( Wickham , 2009 ) . The Adjusted Wald method of calculating 95% binomial confidence intervals ( Agresti and Coull , 1998 ) was used in Figure 2 because a low number of samples were used to generate a binomial mean . Cartoons were made using Adobe Illustrator CS5 ( www . adobe . com ) . Commercial Sanger sequencing returns trace plots , chromatograms that indicate the relative frequency of each base at each position in the sequenced DNA . Trace plots were used in two analyses . First , we estimated the spread of mutations through a population by Sanger sequencing time points from the evolution . Second , we estimated the segregation of alleles in the backcrossed , recreated strains by Sanger sequencing the putative causal alleles in the final , selected population . The fraction of mutant alleles in the population was assumed to be the height of the mutant allele peak divided by the height of the mutant allele peak plus the ancestor allele peak . In the time course analysis , values below 5% ( the approximate background level ) were assumed to be zero , and values above 95% were assumed to be 100% . In the segregation analysis , actual values , not corrected for background , are shown . We note that many of the mutations appear to rise with very similar time courses to each other . This could reflect the modest time resolution of our measurements and the insensitivity of using Sanger sequencing to estimate allele frequencies , but we suspect it reflects the strong advantages that accrue to those rare lineages where two or three beneficial mutations occur in quick succession ( Lang et al . , 2011 ) . To compare the fitness of the recreated and evolved strains , we developed a fitness assay that was suitable for strains with non-uniform morphologies . We could not use standard , quantitative methods of measuring fitness , such as FACS-based competitions ( Desai et al . , 2007 ) , because we were not able to accurately count the number of cells per clump in a large population . We therefore used a qualitative microscopy-based fitness assay to compare growth of a YFP-labeled population and a CFP-labeled population . We started the competitions by growing each population separately to saturation in 1 mM glucose + 1 mM fructose . Equal volumes of each strain were mixed together , and then checked under a microscope to verify that both strains were equally present . 50 µl of the mixed culture was then inoculated into 10 ml of the test media . When the culture reached visible density ( >106 cells/ml ) , three blindly chosen images of a 20 µl sample were taken on the microscope with a 20× objective . Extermination was defined as zero cells in one strain and more than 30 total cells of the other strain . If both strains were still present , 50 µl of the culture was inoculated into 10 ml of fresh media . After six growth cycles , a winner was declared if there was a clear majority ( >75% ) of one strain . Otherwise , the competition was declared a tie . The number of cycles until extermination was averaged across three independent experiments , and qualitative values ( ++++ through −−−− ) were assigned as described in Table 1 . + or − was assigned if one of the strains won two out of the three independent competitions that lasted through all six growth cycles . The evolved clones already constitutively expressed YFP , and a constitutively expressed CFP version of each clone was made by deleting the PACT1-ymCitrine-URA3 construct from the URA3 locus and then transforming with a ura3Δ::PACT1-yCerulean-URA3 construct . The two versions of each strain were grown separately in 1 mM glucose + 1 mM fructose media to saturation , and then the two versions were vortexed or sonicated ( both gave identical results ) for 10 s and then combined at low density in fresh 1 mM sucrose media . The mixed cultures were grown to visible density at 30°C and then pipetted into a well of a glass-bottomed 96-well plate and examined by microscope . In each of the three independent experiments , at least 50 clumps were checked and three representative images were taken for each strain . Images except those in Figure 7 were taken in a 96-well glass-bottomed plate ( greiner bio-one , www . gbo . com ) on a Nikon Ti inverted microscope ( www . nikoninstruments . com ) with a Photometrics CoolSnap HQ camera ( www . photometrics . com ) and MetaMorph software ( www . metamorph . com ) . Figure 7 images were taken on a Nikon Ti inverted microscope with a Yokogawa spinning disc confocal unit ( www . yokogawa . com ) , 447 , 515 , and 594 nm lasers ( www . spectraloptics . com ) , a Hamamatsu Orca camera ( www . hamamatsu . com ) , and MetaMorph software . Images were converted to 8-bit , projected , adjusted for contrast , and annotated with scale bars using the Fiji distribution of ImageJ ( Schindelin et al . , 2012 ) . Contrast was changed for visibility only; our results do not depend on illumination levels and are not affected by the change in contrast . Alleles were replaced in both the evolved and ancestor strains by transforming with a URA3 plasmid that contained a portion of the targeted gene with the desired mutation and was digested at a single cut site within the target gene ( Rothstein , 1991 ) . See Figure 5—figure supplement 1 for the complete protocol . Transformants were selected on—URA plates , colonies that grew were streaked out on—URA plates , and the target sequence was amplified and sequenced to verify insertion . The colony was grown overnight in YPD ( 10 g/l yeast extract , 20 g/l peptone , and 2% [w/v] dextrose ) and selected on 5FOA plates to identify cells in which the URA3 construct had looped out ( Boeke et al . , 1984 ) . Colonies that grew were streaked out on 5FOA , replica plated to YP 2% Acetate ( 20 g/l agar , 10 g/l yeast extract , 20 g/l peptone , and 2% [w/v] potassium acetate ) to eliminate petites , and the target region was amplified and sequenced to ensure that only the desired allele was present . Restriction enzymes , DNA polymerase , polynucleotide kinase , and ligation enzymes were purchased from New England Biolabs ( www . neb . com ) . | Life first appeared on Earth more than 3 billion years ago in the form of single-celled microorganisms . The diverse array of complex life forms that we see today evolved from these humble beginnings , but it is not clear what triggered the evolution of multicellular organisms from single cells . One of the simplest multicellular eukaryotes is the yeast , Saccharomyces cerevisiae—a fungus that has been used for centuries in baking and brewing and , more recently , as a model organism in molecular biology . Yeast cells feed on sugar ( sucrose ) , but are unable to absorb it directly from their surroundings . Instead they secrete an enzyme called invertase , which breaks down the sucrose into simpler components that cells can take up with the help of sugar transporters . However , single yeast cells living in a low-sucrose environment face a problem: most of the simple sugars that they produce diffuse out of reach . To overcome this difficulty , the cells could form multicellular clumps , which would enable each cell to consume the sugars that drift away from its neighbours . Alternatively , the cells could increase their production of invertase , or they could begin to take up sucrose directly . Using genetic engineering , Koschwanez et al . produced three strains of yeast , each with one of these traits , and confirmed that all three strategies do indeed help fungi to grow in low sucrose . But could any of these traits evolve spontaneously ? To test this possibility , Koschwanez et al . introduced wild-type yeast cells into a low-sucrose environment and studied any populations of cells that managed to survive . Of 12 that did , 11 had acquired the ability to form multicellular clumps , while 10 had increased their expression of invertase . Surprisingly , none had evolved the ability to import sucrose . However , 11 of the populations that survived also displayed an adaptation that the researchers had not predicted beforehand: they all expressed higher levels of the sugar transporters that take up sucrose breakdown products . The work of Koschwanez et al . suggests that the benefits of being able to share invertase and , therefore , simple sugars , may have driven the evolution of multicellularity in ancient organisms . Moreover , their use of rational design ( engineered mutations ) combined with experimental evolution ( allowing colonies to grow under selection pressure and studying the strategies that they adopt ) offers a new approach to studying evolution in the lab . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"evolutionary",
"biology"
] | 2013 | Improved use of a public good selects for the evolution of undifferentiated multicellularity |
The permeability barrier of nuclear pore complexes ( NPCs ) controls bulk nucleocytoplasmic exchange . It consists of nucleoporin domains rich in phenylalanine-glycine motifs ( FG domains ) . As a bottom-up nanoscale model for the permeability barrier , we have used planar films produced with three different end-grafted FG domains , and quantitatively analyzed the binding of two different nuclear transport receptors ( NTRs ) , NTF2 and Importin β , together with the concomitant film thickness changes . NTR binding caused only moderate changes in film thickness; the binding isotherms showed negative cooperativity and could all be mapped onto a single master curve . This universal NTR binding behavior – a key element for the transport selectivity of the NPC – was quantitatively reproduced by a physical model that treats FG domains as regular , flexible polymers , and NTRs as spherical colloids with a homogeneous surface , ignoring the detailed arrangement of interaction sites along FG domains and on the NTR surface .
In eukaryotic organisms , the nuclear envelope separates the nucleoplasm from the cytoplasm and encloses most of the genetic material in the cell . The ordered course of gene expression requires selective transport through this double membrane . This function is provided by nuclear pore complexes ( NPCs ) , large membrane-spanning protein complexes that perforate the nuclear envelope ( Fahrenkrog and Aebi , 2003; Fernandez-Martinez and Rout , 2012; Floch et al . , 2014; Gorlich and Kutay , 1999; Grossman et al . , 2012; Macara , 2001 ) . Although small molecules up to 20–40 kDa ( i . e . , up to roughly 5 nm in diameter ) can diffuse freely through the NPC , the passage of larger macromolecules is impeded unless they are bound to nuclear transport receptors ( NTRs ) ( Gorlich and Kutay , 1999; Keminer and Peters , 1999; Mohr et al . , 2009; Yang and Musser , 2006 ) . The permeability barrier in the 30 to 50 nm diameter central transport channel of NPCs is formed by domains of NPC proteins ( nucleoporins ) that are rich in phenylalanine-glycine motifs ( FG domains; single-letter code is used throughout ) and that are grafted to the channel walls at a high density ( Bui et al . , 2013 ) . The FG domains are thought to be highly flexible and behave like natively-unfolded proteins ( Denning et al . , 2003; Denning and Rexach , 2007; Denning et al . , 2002; Hough et al . , 2015 ) . As such , they do not have a defined secondary or higher-order protein structure . However , depending on their intramolecular and intermolecular interactions , these proteins can organize into supramolecular assemblies such as protein meshworks , brushes or scaffolds ( Dyson and Wright , 2005; Uversky and Dunker , 2010 ) . There is a broad consensus that FG domains interact with NTRs and facilitate their passage through NPCs . In addition , there is an attractive ( cohesive ) interaction between FG domains . This promotes the formation and determines the properties of FG domain phases ( Eisele et al . , 2013; Frey and Gorlich , 2007; Frey et al . , 2006; Patel et al . , 2007; Schmidt and Gorlich , 2015 ) , and is also essential for the formation of a functional permeability barrier ( Frey et al . , 2006; Hulsmann et al . , 2012 ) . However , the nature of these interactions remains controversial ( Peters , 2009 ) , both for the interactions between FG domains , and for the interactions between FG domains and NTRs . As a consequence , there remains a large uncertainty about the morphology of the permeability barrier ( Frey and Gorlich , 2007; Frey et al . , 2006; Lim et al . , 2007; Peters , 2005; Rout et al . , 2000; Yamada et al . , 2010 ) and about how it is influenced by the substantial enrichment of NTRs in the NPC conduit ( Frey and Gorlich , 2009; Kapinos et al . , 2014; Lowe et al . , 2015; Mohr et al . , 2009 ) . Because of the low degree of order in the FG domain meshwork and its spatial confinement within the NPC , it has been difficult to address these questions using traditional biochemical assays and structure determination methods . Alternatively , computational models can provide valuable insights into collective behavior of FG domains , but are affected by the size and complexity of the NPC , and in particular by the experimental uncertainty on protein distributions and interactions ( see Osmanovic et al . , 2013a for a review ) . To obtain a more comprehensive understanding of the interactions between FG domains and NTRs in the context of nucleo-cytoplasmic transport , we have employed a novel approach that combines measurements of the uptake of NTRs by well-defined nanoscale assemblies of FG domains with ( coarse-grained ) computational modeling for a quantitative interpretation of the experimental results in terms of FG domain morphology and intermolecular interactions . Specifically , we produce films of end-grafted purified FG domains that are similar to the protein meshwork in NPCs , both in their thickness and in their FG motif density ( Eisele et al . , 2010; 2013 ) . With such planar model systems it is possible to quantify NTR binding and investigate NTR-induced thickness changes ( Eisele et al . , 2010; 2012; Schoch et al . , 2012 ) . Binding curves of several NTR/FG domain systems have been shown to deviate from an ideal Langmuir isotherm , suggesting that the binding avidity of NTRs to FG domain films strongly depends on the concentration of NTRs in the film ( Eisele et al . , 2010; Kapinos et al . , 2014; Schleicher et al . , 2014; Wagner et al . , 2015 ) and on the proportion of FG motifs that are occupied by NTRs . Here , we describe the use of this system to identify common features and obtain a more quantitative understanding of these interactions , by analyzing and comparing the binding of two different NTRs , nuclear transport factor 2 from Homo sapiens ( NTF2 ) and Importin β from Saccharomyces cerevisiae ( Impβ ) , to plane-grafted FG domain films that each are generated from one of three different FG domains: the FG domain of Nsp1 from S . cerevisiae ( that has FxFG and just FG motifs ) , a glycosylated FG domain of Nup98 from Xenopus tropicalis ( Nup98-glyco; with primarily GLFG and just FG motifs ) and an artificial , regular repeat with exclusively FSFG motifs ( reg-FSFG ) . The two transport receptors differ in size ( 29 . 0 kDa for the functional NTF2 homodimer and 95 . 2 kDa for Impβ ) and in the number and distribution of binding sites for FG domains . Two identical sites are located between the subunits of NTF2 ( Bayliss et al . , 2002 ) , whereas for mammalian Impβ two different sites have been identified by crystallography ( Bayliss et al . , 2000 ) and molecular dynamics simulations have suggested there may be up to nine sites spread over the Impβ surface ( Isgro and Schulten , 2005 ) . Recent crystallography work revealed eight binding sites on the exportin CRM1 ( Port et al . , 2015 ) , suggesting that the dispersal of binding pockets across the protein surface is a common feature of the larger NTRs . The FG domains employed in this study differ in prevalent FG motif types , FG domain size , abundance of FG motifs relative to FG domain size ( Table 1 ) , as well as in the distribution of FG motifs along the peptide chains and the composition of the spacer regions between FG motifs ( Table 1—source data 1 ) ( Labokha et al . , 2013; Radu et al . , 1995; Rout and Wente , 1994 ) . 10 . 7554/eLife . 14119 . 003Table 1 . Properties of employed FG domain constructs . See Table 1—source data 1 for the full amino acid sequences of these constructs . DOI: http://dx . doi . org/10 . 7554/eLife . 14119 . 00310 . 7554/eLife . 14119 . 004Table 1—source data 1 . Amino acid sequence of employed FG domain constructs . FG domains are shown in black letters , His tags in blue letters , and remaining parts ( i . e . , TEV cleavage sites , Cys tags and spacers ) in grey letters . FxFG motifs are marked in yellow , GLFG motifs in green , other FG motifs in purple . Nup98-glyco features O-GlcNAc on ~30 of the S and T residues . DOI: http://dx . doi . org/10 . 7554/eLife . 14119 . 004FG domainamino acids asequenceFG motifsFG motifs/amino acidsFxFGGLFGOtherNsp1615irregular , natural190140 . 054Nup98-glyco496irregular , natural38280 . 079reg-FSFG315regular , artificial16000 . 051a Excluding the His tags but including all other auxiliary amino acids ( TEV cleavage sites , Cys tags and spacers ) . Our approach has enabled us to explore the universality/diversity of NTR binding to FG domains , to quantify the binding and to interpret it in terms of NTR distribution in and on FG domain assemblies , while also demonstrating how we can benchmark parameters in computational simulations to a well-defined experimental model . From the quantitative comparison between experiment and computational modeling , we learn about the levels of structural and chemical detail and heterogeneity that are required to effectively model and understand NTR uptake by FG domain assemblies , and gain new insights into the physical mechanisms – largely related to collective low-affinity interactions and the formation of a phase ( Hyman and Simons , 2012 ) of FG domains and NTRs – that determine NPC transport selectivity .
Selected FG domains , i . e . , Nsp1 , Nup98-glyco and reg-FSFG , were purified ( Figure 1—figure supplement 1 ) and anchored stably and specifically to planar surfaces , through their His tags ( Figure 1—figure supplement 2 ) . We monitored the formation of FG domain films and their interaction with NTF2 and Impβ by spectroscopic ellipsometry ( SE ) and quartz crystal microbalance ( QCM-D ) , simultaneously and on the same sample ( Figure 1—figure supplement 3 ) , to quantify areal protein densities , Γ ( i . e . , amounts of protein per unit area , expressed as pmol/cm2; 1 pmol/cm2 equals 0 . 6 molecules per 100 nm2 ) , and effective film thicknesses , d , respectively . The FG domain grafting density was tuned to range between 4 and 11 pmol/cm2 ( i . e . , between 2 . 4 and 6 . 6 molecules per 100 nm2 ) , by varying the FG domain solution concentration and incubation time . This range covers and extends around the estimated grafting density in a yeast NPC that is thought to be 5 . 2 to 6 . 9 pmol/cm2 ( i . e . , 3 . 1 to 4 . 1 molecules per 100 nm2 ) ; this estimate is based on the assumption of a cylindrical channel of 35–40 nm in diameter and 30–35 nm in length ( Yang et al . , 1998 ) , and of ~136 FG domains per channel ( Rout et al . , 2000; Strawn et al . , 2004 ) . It is also a range ( ≥5 pmol/cm2 for Nup98-glyco ) over which the FG domain films maintain a homogeneous appearance , as previously verified by atomic force microscopy ( Eisele et al . , 2013 ) . NTRs were titrated into FG domain films over a range covering three orders of magnitude in NTR concentration . This range includes the typical cellular concentrations of NTRs , e . g . , 0 . 5 μM NTF2 homodimer in X . laevis eggs ( Kirli et al . , 2015 ) , 0 . 3 μM NTF2 homodimer in HeLa cells ( Gorlich et al . , 2003 ) , and 3 to 5 μM Impβ in X . laevis ( Kirli et al . , 2015; Wuhr et al . , 2014 ) . The highest concentration in our experiments ( 10 μM ) is comparable to the total concentration of NTRs found in cells ( Hahn and Schlenstedt , 2011; Kirli et al . , 2015; Wuhr et al . , 2014 ) . Figure 1 summarizes the experimental data at equilibrium as a function of NTR concentration , cNTR , in solution . A set of controls confirmed that NTF2 and Impβ bound specifically to the immobilized FG domains ( Figure 1—figure supplements 4 and 5 ) , and that binding equilibriums were indeed achieved ( Figure 1—figure supplement 3B ) . Irrespective of the FG domain and NTR types , NTR binding and unbinding was rapid , i . e . , largely determined by mass transport to and from the surface upon changes in NTR concentration ( Figure 1—figure supplement 3B ) . This is consistent with reports on the kinetics of NTRs interacting with individual FG motifs ( Milles et al . , 2015 ) , FG domains ( Hough et al . , 2015 ) , and FG domain assemblies ( Eisele et al . , 2010; Frey and Gorlich , 2007 ) , which all found binding to be exceptionally rapid . Taking these observations together , we conclude that we measure genuine interactions between NTRs and supramolecular assemblies of FG domains . 10 . 7554/eLife . 14119 . 005Figure 1 . Isotherms of NTR binding ( top row , log-log presentation ) and FG domain film thickness evolution ( bottom row , lin-log presentation ) for NTF2 and Impβ binding to different FG domains ( see labels at top ) at selected FG domain grafting densities ( visualized by distinct symbols and colors ) . Error bars are shown for all data points in the binding isotherms , and for three selected data points ( indicating the trends ) per curve in the thickness isotherms . The data for Impβ binding to the 10 . 0 pmol/cm2 Nsp1 film were reproduced from Eisele et al . ( 2010 ) ; this data was acquired with Nsp1 carrying a His tag at the opposite end ( N-terminus ) compared to the other Nsp1 data in this study , in a separate SE measurement and no simultaneously recorded thickness data are available . Full experimental details are available in ‘Materials and methods’ and Figure 1—figure supplements 1–5; tabulated results are available in Figure 1—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 14119 . 00510 . 7554/eLife . 14119 . 006Figure 1—source data 1 . Tables of data shown in Figure 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 14119 . 00610 . 7554/eLife . 14119 . 007Figure 1—figure supplement 1 . Quality of purified recombinant proteins used in this study . FG domains with His tag were dissolved in formamide and diluted 1:3 in SDS sample buffer . 1 . 5 to 2 . 5 μg of each protein construct were resolved by SDS-PAGE and stained with Coomassie G250 . The band corresponding to reg-SSSG runs much higher than the molecular weight expected; this is usually the case for very hydrophilic proteins ( Shirai et al . , 2008 ) . All preparations contain more than 95% full length protein . DOI: http://dx . doi . org/10 . 7554/eLife . 14119 . 00710 . 7554/eLife . 14119 . 008Figure 1—figure supplement 2 . FG domains are anchored specifically and stably through their terminal His tag . ( A ) Binding and elution of reg-FSFG was monitored by SE , on a silica surface previously functionalized with a supported lipid bilayer ( SLB; 7% tris-NTA ) . Arrows on top of the graph indicate the start and duration of incubation with different sample solutions; during remaining times , the surface was exposed to working buffer . Most of the reg-FSFG remains stably bound upon rinsing in working buffer . reg-FSFG was fully eluted after the imidazole treatment ( grey shaded area , not monitored ) , demonstrating that binding is specific through the His tag . ( B-D ) Site-specific and stable anchoring of Nsp1 to His tag capturing QCM-D sensors , as well as Nsp1 and Nup98-glyco to SLBs ( 10% bis-NTA ) on silica is demonstrated by QCM-D; the data are reproduced from Figure 1 in Eisele et al . ( 2012 ) and Figure S2 in Eisele et al . ( 2013 ) , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 14119 . 00810 . 7554/eLife . 14119 . 009Figure 1—figure supplement 3 . Schematic illustration of the experimental approach and representative data . ( A ) Schematic illustration of the experimental approach . ( B–C ) Representative data for combined SE/QCM-D measurements . Areal protein densities ( top ) are obtained through optical modeling of SE data ( see Materials and methods ) . QCM-D responses ( bottom ) are recorded in parallel , and normalized frequency shifts Δfi/i and dissipation shifts ΔDi for a selected overtone ( i = 3 ) displayed here . Film thickness is obtained through viscoelastic modeling of QCM-D data ( see Materials and methods ) . Δf = ΔD = 0 corresponds to the functionalized surface before FG domain grafting . ( B ) Monitoring of FG domain film formation . Nsp1 was exposed to a silica substrate previously functionalized with an SLB ( 7% tris-NTA ) . The final grafting density is 4 . 9 pmol/cm2 . Minor perturbations in Δf and ΔD between 33 and 43 min are due to transient variations in the solution temperature in contact with the QCM-D sensor during the rinsing with buffer , and do not represent changes of the Nsp1 film . ( C ) Time-resolved data for the titration of NTF2 into this Nsp1 film . The NTF2 solution concentration was increased in 12 steps ( 0 . 0025 , 0 . 01 , 0 . 05 , 0 . 1 , 0 . 2 , 0 . 5 , 1 , 2 , 3 , 5 , 7 . 5 and 10 μM ) , and then decreased in 16 steps ( ( 2/3 ) j × 10 μM , with j = 1 , … , 16 ) , followed by continuous rinsing with buffer solution to remove all NTF2 from the bulk solution . The rapid binding and unbinding of NTF2 observed here is representative for all titration measurements performed , and binding equilibriums could thus be readily attained . Moreover , binding of NTF2 was largely reversible , with less than 7% of the maximal binding remaining following rinsing in buffer for any given measurement . For Impβ , more than 75% were readily eluted from 5 pmol/cm2 Nsp1 films , and we had previously shown this NTR to elute close to completely from 10 pmol/cm2 Nsp1 films ( Eisele et al . , 2012; Eisele et al . , 2010 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14119 . 00910 . 7554/eLife . 14119 . 010Figure 1—figure supplement 4 . Controls for the binding of NTRs to His tag capturing surfaces monitored by QCM-D . ( A ) NTF2 on His tag capturing QCM-D sensor; ( B ) NTF2 on SLB ( 7% tris-NTA ) ; ( C ) Impβ on SLB ( 4% tris-NTA ) . The vertical scales were chosen such that the full range would approximately cover the magnitude expected for a full monolayer of NTRs . DOI: http://dx . doi . org/10 . 7554/eLife . 14119 . 01010 . 7554/eLife . 14119 . 011Figure 1—figure supplement 5 . NTF2 binds all FG domains predominantly through its primary binding site . ( A ) Ratio of equilibrium bound amounts of NTF2 W7A mutant and wild type NTF2 as a function of FG domain type . NTF2 and NTF2 W7A mutant were sequentially exposed to FG domain films , and binding was quantified by SE . Mean and standard errors of two to four independent measurements per FG domain type with FG domain surface densities ranging between 5 and 13 pmol/cm2 are presented . The tryptophan at position 7 is known to be important for the binding of NTF2 through a structurally defined ( Bayliss et al . , 2002 ) binding site to FG motifs ( Bayliss et al . , 1999 ) . For all tested FG domain types , binding of the W7A mutant was reduced by more than 80% compared to native NTF2 . ( B ) Native NTF2 did not bind to reg-SSSG ( here at 8 . 3 pmol/cm2 ) , confirming that binding to reg-FSFG requires the FSFG motif . These data confirm that the NTF2 in our experiments binds specifically to the immobilized FG domains through its primary FG motif binding site . DOI: http://dx . doi . org/10 . 7554/eLife . 14119 . 011 Interestingly , the shape of the binding isotherms ( i . e . , the areal NTR density in the film , ΓNTR , eq , versus cNTR; Figure 1 , top row ) remained largely unchanged with FG domain type and grafting density . This common shape prompted a more detailed analysis , including the use of phenomenological models ( see Figure 2A for a selected measurement; all other measurements led to similar conclusions , see Figure 2B ) . For cNTR ≤ 0 . 05 μM , the slope in the log-log binding isotherms is one ( Figure 1; and Figure 2A , main plot , dashed line ) , as expected from the low-concentration limit of a Langmuir isotherm . This indicates that – at low concentrations – individual NTR molecules bind to the FG domain film independently . In this concentration range , the ratio ΓNTR , eq / cNTR = PC × d was constant , with partition coefficients PC between 103 and 105 ( Figure 2—figure supplement 1A ) , implying that NTRs are strongly enriched in the FG domain films compared to their concentration in solution . 10 . 7554/eLife . 14119 . 012Figure 2 . Quantitative analysis of the binding isotherms . ( A ) A selected data set ( NTF2 binding to 6 . 1 pmol/cm2 reg-FSFG; symbols ) with fits to simple binding models ( lines ) . Data at low NTR concentration ( cNTR ≤ 0 . 05 μM ) display a close-to-linear relation ( dashed line with slope 1 . 0 in the log-log plot ) , as expected for independent binding , yet the Langmuir isotherm ( inset , dashed line in lin-log plot ) fails to reproduce the data over the full range of NTR concentrations . The Hill equation provides a good description of the data in the high-concentration range ( 0 . 05 μM ≤ cNTR ≤ 10 μM; solid line ) . ( B ) By normalizing the areal densities and NTR concentrations to ΓNTR , max and K0 . 5 , respectively , all data could be overlaid on a single master curve , where the effective maximal binding ΓNTR , max and the half-maximal binding K0 . 5 were determined from fits with the Hill equation ( see main text and Figure 2—figure supplement 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14119 . 01210 . 7554/eLife . 14119 . 013Figure 2—figure supplement 1 . Quantitative analysis of the binding isotherms in Figure 1 . Analysis as a function of NTR and FG domain types ( visualized by distinct symbols and colors , as indicated in A ) and grafting density ΓFG . ( A ) Partition coefficients PC were obtained from a linear fit to the data at cNTR≤ 0 . 05 μM . ( B-D ) Results of fits with the Hill equation , i . e . , the Hill coefficient α , the concentration for half maximal binding K0 . 5 and the effective maximal binding ΓNTR , max . In a first step , seven data sets where selected for which data covered the full range of 0 . 05 μM ≤ cNTR ≤ 10 μM , and fitted over 0 . 05 μM ≤ cNTR ≤ 10 μM with α , K0 . 5 and ΓNTR , max as adjustable parameters . From this analysis , a mean value of α = 0 . 71 ± 0 . 04 was determined . In a second step , an equivalent analysis was performed for the remaining three data sets , for which the experimental data did not cover the full concentration range ( i . e . , NTF2 on 5 . 4 pmol/cm2 Nup98-glyco , NTF2 on 8 . 3 pmol/cm2 Nsp1 , and Impβ on 5 . 1 pmol/cm2 Nsp1 ) , with α = 0 . 71 fixed . By thus fixing α , we prevented a scatter of the fit parameters that would only be due to the limited input data . ( E ) Ratio of NTR bound per FG motif at cNTR = 10 μM; the three data points not directly measured were obtained through extrapolation using the Hill equation . DOI: http://dx . doi . org/10 . 7554/eLife . 14119 . 013 For higher concentrations , however , the Langmuir isotherm ( i . e . , ΓNTR , eq = ΓNTR , max × cNTR / ( K0 . 5 + cNTR ) , with ΓNTR , max the maximum areal density of bound NTRs and K0 . 5 the concentration for half-maximum binding ) failed to faithfully describe the data ( Figure 2A , inset , dashed line ) . This is in line with earlier observations ( Eisele et al . , 2010; Wagner et al . , 2015 ) . For a quantitative comparison between different curves , we fitted the experimental data with the Hill equation ( Figure 2A , main plot , solid line ) , i . e . , ΓNTR , eq=ΓNTR , max×cNTRα/ ( K0 . 5α+cNTRα ) ( Weiss , 1997 ) . The Hill coefficients for all curves lie within the narrow range α = 0 . 71 ± 0 . 04 ( Figure 2—figure supplement 1B ) . This narrow spread in α in the Hill fits and the small variations ( typically less than a factor of two ) in K0 . 5 for the different FG domain types and grafting densities ( Figure 2—figure supplement 1C ) , confirm the uniformity of the binding isotherms noted above . Unsurprisingly , there was more variation in the effective maximal binding ΓNTR , max as determined from the Hill fits ( Figure 2—figure supplement 1D ) . The uniformity of the binding isotherms can be further articulated by plotting normalized areal densities ( ΓNTR , eq/ΓNTR , max ) versus normalized NTR concentrations ( cNTR/K0 . 5 ) , with ΓNTR , max and K0 . 5 determined from the Hill fit , to show that this reduces all data to a single master curve ( Figure 2B ) . The agreement in curve shape is remarkable , given that the used FG domains provide a large spread of FG motif types and FG motif arrangement along the peptide chains ( Table 1 ) and that the two tested NTR types differ both in size and in the number of binding sites for FG motifs . Since α is smaller than one , the Hill fits indicate that NTR binding is negatively cooperative in the physiologically relevant concentration range , i . e . , the average binding strength decreases as the FG domain assembly becomes enriched with NTRs . This finding is in line with recent reports that propose a modulation of NTR binding by the presence of other NTRs ( Kapinos et al . , 2014; Schleicher et al . , 2014; Wagner et al . , 2015 ) . In this context , it is worth noting that the areal density of bound NTR represents only a small fraction of the FG motif density available in the films . The total number of FG motifs per FG domain is 33 for Nsp1 , 39 for Nup98-glyco and 16 for reg-FSFG ( Table 1 ) . Our data illustrate that , at cNTR = 10 μM , the films contained at least 10 and 50 FG motifs per bound NTF2 dimers and per bound Impβ , respectively ( Figure 2—figure supplement 1E ) . With two binding sites for FG motifs per NTF2 dimer , and up to nine binding sites per Impβ , this implies that no more than 20% of the FG motifs were simultaneously engaged in NTR binding . Importantly , NTR binding does not correlate with the total abundance of FG motifs: for example , Nsp1 binds more than twice the number of NTF2 per FG motif compared to Nup98-glyco ( Figure 2—figure supplement 1E ) , consistent with its binding NTF2 more strongly ( Clarkson et al . , 1997 ) . Taken together , the analysis of binding isotherms demonstrates that NTRs are substantially enriched in FG domain films , that the accumulation of NTRs in FG domain films progressively reduces the strength of NTR binding , and that the NTR binding behavior has universal features that are independent of the detailed chemical and structural features of the FG domains and NTRs . Variations in film thickness d ( Figure 1 , bottom row ) following NTR binding were generally moderate . At NTF2 solution concentrations up to 1 μM , the thickness remained virtually unchanged for Nup98-glyco and reg-FSFG , and decreased marginally ( by up to 7% ) for Nsp1 . At higher concentrations , the thickness gradually increased , by between 5 and 35% at 10 μM compared to the pristine FG domain film , depending on the FG domain type and grafting density . For Impβ binding to Nsp1 , there was a moderate and gradual thickness increase up to 25% at 10 μM . In all cases , the increase in film thickness was smaller than or comparable with the dimensions of the NTRs . These findings are in clear disagreement with the film collapse by more than 50% , reported by Lim et al . on nanoscale islands of FG domain assemblies ( Lim et al . , 2007 ) , and the 'nanomechanical collapse' model proposed based on those data . Instead , our data are in full agreement with other thickness measurements on similar systems ( Eisele et al . , 2010; Kapinos et al . , 2014; Wagner et al . , 2015 ) , which consistently did not give any indications for such a collapse , but rather indicate that the global morphology of FG domain films remains preserved irrespective of the concentration and type of NTR . In our experiments , the shape of the binding isotherms was independent of the detailed chemical and structural features of the FG domains and NTRs . We therefore hypothesized that it must arise from generic physical features of the FG domain / NTR system , among which are the nature of FG domains as flexible polymers and of NTRs as globular colloids , as well as the mean , overall interactions between FG domains and NTRs . To test this hypothesis , we adapted a previously developed computational model ( Osmanovic et al . , 2012; 2013b ) to planar surfaces ( Figure 3A ) . The model treats polymers as beads on a chain , where the bead diameter is set to twice the contour length of an amino acid , to reproduce the flexibility of unfolded peptide chains ( see Materials and methods ) ; the interactions between FG domains are essentially smeared out over the whole ( homogeneous ) chain thus effectively including interactions between FG motifs , but potentially also with other parts of the FG domain chains . The model explicitly considers the confinement through grafting , the size , the flexibility , the geometrical excluded volume and the cohesiveness of FG domains , the concentration and geometrical excluded volume of NTRs , and the attraction between FG domains and NTRs . The polymer and the colloid surface are homogeneous , and two adjustable parameters regulate the interaction strengths ( see Materials and methods ) : εpp the cohesiveness ( Eisele et al . , 2013 ) of polymer segments , and εpc the attraction between a polymer segment and a colloid . From the computed density maps ( Figure 3—figure supplement 1 ) for appropriate polymers and colloids , physical parameters such as average film thickness and binding isotherms were extracted ( Figure 3—figure supplements 2–5 ) and compared with the experimental data . 10 . 7554/eLife . 14119 . 014Figure 3 . Computational model . ( A ) Schematic illustration of the computational model . FG domains are represented as end-grafted polymers anchored at 5 . 5 pmol/cm2 ( i . e . , 3 . 3 molecules per 100 nm2 ) to the bottom of a 100 nm diameter cylinder , and modeled as strings of beads , where each bead has equal bond length and diameter ( two amino acids , 0 . 76 nm ) . The number of polymer beads was set to match the length of experimentally used FG domains . NTF2 dimers and Impβ are represented as spherical colloids of 4 . 0 and 6 . 0 nm diameter , respectively . ( B ) Matching of the computational model with experimental data for FG domain films in the absence of NTRs . Horizontal lines represent the experimentally determined film thickness per amino acid for different FG domains ( black line - Nsp1 at 4 . 9 pmol/cm2; blue line - Nup98-glyco at 5 . 4 pmol/cm2; orange line - reg-FSFG at 6 . 1 pmol/cm2 ) , with shaded areas in matching colors indicating confidence intervals . Symbols represent the thickness as predicted by the computational model as a function of εpp for the different FG domains ( at 5 . 5 pmol/cm2; colors match experimental data ) . The data points and the upper and lower ends of the vertical lines refer to the effective thicknesses where the densities have dropped to 5% , 1% and 10% of the maximal densities in the film , respectively . Symbols for Nsp1 and Nup98-glyco are translated along the x axis by +0 . 1 kBT and -0 . 1 kBT , respectively , to improve their visibility . Dashed lines through the symbols are cubic interpolations ( the black dashed line is for Nsp1 and Nup98-glyco ) . Full computational details are available in ‘Materials and methods’ and Figure 3—figure supplements 1–5 . DOI: http://dx . doi . org/10 . 7554/eLife . 14119 . 01410 . 7554/eLife . 14119 . 015Figure 3—figure supplement 1 . Scheme illustrating how computational modeling data is presented in the form of maps of the polymer and colloid packing fractions . Maps are cross-sections along the axis of the modeled cylinder ( with polymers grafted at the bottom; cf . Figure 3A ) . The bottom part of the cylinder at full width ( 100 nm ) is shown , with horizontal and vertical dimensions to scale . The left half of each map shows the polymer packing fraction as a heat map ( with scale bar in the bottom right ) and the colloid packing fraction as iso-density lines; the right half shows the colloid packing fraction as a heat map and the polymer packing fraction as iso-density lines . DOI: http://dx . doi . org/10 . 7554/eLife . 14119 . 01510 . 7554/eLife . 14119 . 016Figure 3—figure supplement 2 . Computational modeling data for a polymer length equivalent to Nsp1 and colloids of 4 . 0 nm diameter ( equivalent to NTF2 homodimers ) . ( A ) Maps of the polymer and colloid packing fractions at 10 μM colloid in solution , presented as described in figure supplement 1 , for selected sets of εpc ( rows , as indicated on left side ) and εpp ( columns , as indicated on bottom ) . ( B-C ) Areal density of colloids in the film and film thickness , respectively , as a function of colloid concentration in solution for selected sets of εpp ( as indicated on top ) and εpc ( as indicated in the graphs ) . The data covers the full parameter space computed . The lines and upper and lower ends of the vertical bars in ( C ) correspond to effective thicknesses where the densities have dropped to 5% , 1% and 10% of the maximal densities in the film , respectively ( see Figure 3 and Materials and methods ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14119 . 01610 . 7554/eLife . 14119 . 017Figure 3—figure supplement 3 . Computational modeling data for a polymer length equivalent to Nup98-glyco and colloids of 4 . 0 nm diameter ( equivalent to NTF2 homodimers ) . Data are displayed analogous to Figure 3—figure supplement 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 14119 . 01710 . 7554/eLife . 14119 . 018Figure 3—figure supplement 4 . Computational modeling data for a polymer length equivalent to reg-FSFG and colloids of 4 . 0 nm diameter ( equivalent to NTF2 homodimers ) . Data are displayed analogous to Figure 3—figure supplement 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 14119 . 01810 . 7554/eLife . 14119 . 019Figure 3—figure supplement 5 . Computational modeling data for a polymer length equivalent to Nsp1 and colloids of 6 . 0 nm diameter ( equivalent to Impβ ) . Data are displayed analogous to Figure 3—figure supplement 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 14119 . 019 First , we considered the FG domain films without NTRs . Figure 3B displays the predicted film thickness normalized by the number of amino acids in the protein . As expected , the film thickness decreases ( i . e . , the film condenses ) with increasing cohesiveness εpp . At any given εpp for the polymers representing Nsp1 and Nup98-glyco , the normalized thickness is identical , as would be expected based on mean field theory for polymer brushes , irrespective of the degree of cohesiveness ( Zhulina et al . , 1990 ) . The somewhat larger values for the roughly two-fold shorter chain representing reg-FSFG are thus likely to reflect finite-size effects . To obtain estimates for the cohesiveness parameter εpp for all FG domains , we compared the computational predictions ( symbols in Figure 3B ) to the experimental thickness data ( horizontal bands in Figure 3B ) . Table 2 shows εpp as a function of FG domain type , obtained via a cubic interpolation ( dashed line in Figure 3B ) between the εpp values for which computational data were available . Because the interface between the film and the bulk solution is not ideally sharp , it was unclear a priori which computational density threshold ( see Materials and methods ) should match the effective thickness measured by QCM-D most accurately . Reassuringly , however , the results were only weakly influenced by the precise definition of film thickness that was used in the computational model , i . e . , density thresholds of 1% or 10% instead of 5% typically altered the estimates for εpp by less than 10% . From the comparison of computation and experiment , it was clear that εpp for Nup98-glyco exceeded that of Nsp1 , in good agreement with our previous study ( Eisele et al . , 2013 ) , in which we also found Nup98-glyco to be more cohesive than Nsp1 . Nup98-glyco and reg-FSFG have similar cohesiveness . Using similar computational modeling in a NPC-mimicking pore geometry of 50 nm diameter , we had previously estimated εpp ≈ 0 . 05 kBT from nanomechanical studies on intact NPCs ( Bestembayeva et al . , 2015 ) . This is close to the εpp values found here , though slightly higher , probably because the presence of NTRs was not taken into account in the computational model to which the nanomechanical data were matched . 10 . 7554/eLife . 14119 . 020Table 2 . Interaction parameters determined based on comparison of experiment and computational model . DOI: http://dx . doi . org/10 . 7554/eLife . 14119 . 020FG domainεpp ( kBT ) εpc ( kBT ) NTF2εpc ( kBT ) ImpβNsp10 . 024 ± 0 . 0010 . 34 ± 0 . 020 . 40 ± 0 . 02Nup98-glyco0 . 030 ± 0 . 0020 . 36 ± 0 . 02-reg-FSFG0 . 030 ± 0 . 0050 . 40 ± 0 . 02- Next , we analyzed the binding isotherms . For a given cohesiveness parameter εpp , we found that the precise setting of the NTR•FG domain interaction εpc strongly influenced the amount of bound NTRs for any given NTR concentration , by orders of magnitude for 0 . 1 kBT changes in εpc , in the explored parameter range of 0 . 1 to 0 . 5 kBT . It also strongly affected the overall shape of the binding isotherms ( Figure 3—figure supplements 2B , 3B , 4B and 5B ) . Figure 4 ( top row ) shows computational data for parameter sets of εpp and εpc that best match the experimental results for the different FG domain and NTR types , where εpp was determined by the film thickness measurement in the absence of NTRs ( see Figure 3B ) . Taking into account that these are fits with a single free parameter ( εpc ) over several orders of magnitude in bound NTR and in NTR concentration , the agreement with the experimental data is remarkably good . With εpp estimated from the thickness in the absence of NTR ( Figure 3B ) and εpc from a comparison to the binding isotherms ( Figure 4 , top row ) , the performance of the model was further validated via the film thickness as a function of NTR concentration in solution . There is good agreement between the experimental data and the computational results ( Figure 4 , bottom row ) . Table 2 summarizes the results of this analysis , with the estimates of εpc varying less than ~20% between the different FG domains and NTRs . Taken together , the measured binding of NTRs to FG domain films was accurately modeled by our simplified description of the relevant interactions in terms of the two key parameters εpp and εpc , where we treat all amino acids in the FG domain chains identically and the NTR surfaces as homogeneous . 10 . 7554/eLife . 14119 . 021Figure 4 . Matching of experimental and computational data . The top row shows binding isotherms and the bottom row the concomitant film thickness evolution . The grafting density was set to 5 . 5 pmol/cm2 in all computations , and the experimental data with the closest FG domain grafting densities are reproduced from Figure 1 and visualized by black symbols . Computational data are shown as green lines . The solid lines represent the best match to the experiment , and the corresponding εpp and εpc are indicated . The best match of the binding isotherms was determined by minimization of the least square differences of log ( ΓNTR , eq ) over the range 0 . 025 μM ≤ cNTR ≤ 10 μM , where the experimental data was interpolated and extrapolated using a linear fit for cNTR < 0 . 05 μM , and the Hill equation for cNTR > 0 . 05 μM , as shown in Figure 2 and Figure 2—figure supplement 1 . Dashed lines and dotted lines in the top row correspond to a change in εpc by -0 . 01 kBT and +0 . 01 kBT , respectively , with εpp unchanged . The lines and upper and lower ends of the vertical bars in the bottom row correspond to effective thicknesses where the densities have dropped to 5% , 1% and 10% of the maximal densities in the film , respectively ( see Figure 3 and Materials and methods ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14119 . 021 It should be emphasized that these interaction strengths represent effective , smeared-out affinities between the polymer beads and colloids in our model ( Figure 3A ) . To relate εpc to a rough estimate for the binding energy of an NTR in the FG domain film , one may assume the NTR colloid to be surrounded by polymer beads at the maximum polymer packing in our calculations ( ~20% ) , yielding at most ~26 and ~53 polymer beads in contact with the colloidal surface , for the NTF2- and Impβ-mimicking colloids , respectively . Hence for NTF2 , the corresponding binding energy is ≲26 × εpc; and εpc between 0 . 3 and 0 . 4 kBT implies a binding energy of ≲10 kBT per NTF2 homodimer . Similarly , εpc ~ 0 . 4 kBT implies a binding energy of ≲20 kBT per Impβ . Hence our results would correspond to a few kBT binding energy per FG-binding site on the NTRs , in reasonable agreement with the millimolar affinities per FG motif observed with Impβ ( Milles et al . , 2015 ) . With the values for εpp and εpc constrained by the comparison to the experimental data , the computational model makes predictions about the distribution of FG domains and NTRs along the surface normal . These are shown in Figure 5 ( top row ) for cNTR = 10 μM , i . e . , in the physiological range of total NTR concentrations . They reveal that , given parameter settings that best match the experimental system ( Table 2 ) , the NTRs effectively penetrate and fill all FG domain films . 10 . 7554/eLife . 14119 . 022Figure 5 . NTRs favor the penetration of and binding into FG domain films , but only just so . Computed packing fraction profiles ( polymer – blue solid line , colloid – red dashed line ) in the presence of 10 μM NTR , as a function of distance from the grafting surface . The top row shows the predictions for the parameter sets of εpp and εpc that match the experimental data best ( cf . Figure 4 and Table 2 ) . The bottom row shows predictions with εpp increased by 33% compared to the best match . Schemes ( insets ) illustrate the distinct distributions of NTRs with these two parameter choices . DOI: http://dx . doi . org/10 . 7554/eLife . 14119 . 022 Figure 5 ( bottom row ) demonstrates that a relatively small change in the FG domain cohesiveness can have a dramatic effect on the NTR distribution . For example , with the NTF2•reg-FSFG interaction maintained at εpc = 0 . 4 kBT , an increase in inter-FG domain attraction by 33% , from εpp = 0 . 03 kBT to 0 . 04 kBT , essentially impaired NTF2 accumulation inside the film , and the NTF2 was enriched instead at the film-solution interface – that is to say , on the film rather than in the film . A similar trend is observed for all other combinations of NTRs and FG domains tested , albeit to a lesser extent for the least cohesive FG domains ( Nsp1 ) with the smaller NTR ( NTF2 ) . The model also indicated that a minor reduction in the colloid•FG domain interaction strength drastically reduces binding ( Figure 3—figure supplements 2 , 3 and 5 ) . Figure 6 illustrates this for a selected colloid concentration ( 1 . 4 μM , in the range of individual NTR concentrations in the cell ) in the solution phase , and shows that by reducing εpc by only 25% compared to the best match for the respective NTRs and FG domains , there is a reduction in colloid binding by more than one order of magnitude . Such a dramatic effect is remarkable: For comparison , a reduction of less than 25% would be expected for simple Langmuir-type one-to-one binding . Collectively , these results suggest that the native system is tuned to operate within a rather narrow parameter space in εpc and εpp that facilitates the strong enrichment of NTRs within the FG domain film , whereas similarly sized proteins with weaker binding strength are effectively excluded . 10 . 7554/eLife . 14119 . 023Figure 6 . Colloid binding depends sharply on colloid•FG domain interaction strength εpc . Computed colloid binding as a function of εpc is shown , for a colloid concentration in solution of 1 . 4 μM and with the FG domain cohesiveness εpp set to the values that best match the experimental data ( cf . Table 2 ) . The blue vertical lines indicate the εpc values giving the best match to the experimental data for the indicated FG domains and NTRs ( cf . Table 2 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14119 . 023
We have measured reconstituted assemblies of FG domains in planar geometry and at grafting densities that are similar to those in the NPC . This yields laterally homogeneous films ( Eisele et al . , 2013 ) , which may be qualified as polymer brushes ( Israelachvili , 1991 ) . As pointed out previously ( Eisele et al . , 2013 ) , this qualification per se does not imply a distinction between a 'brush-like , entropic' scenario ( Lim et al . , 2007 ) of the NPC transport barrier on one side , and a hydrogel scenario ( Frey et al . , 2006 ) on the other . The main distinguishing factor between these opposing scenarios is the level of cohesiveness of the FG domains , which in our computational model is parameterized by εpp , and defines the compactness of the FG domain phase ( Eisele et al . , 2013 ) . From the comparison between experimental and computational results for the thickness of the different FG domain films ( in the absence of NTRs , Figure 3B ) , it appears that the FG domain assemblies are compacted by a factor two to four compared to a film of perfectly non-cohesive ( εpp = 0 , equivalent to a self-avoiding random walk model ) and flexible peptide chains , as follows from an extrapolation of the computational data ( dashed lines in Figure 3B ) . In spite of this condensation , the measured film thickness ( Figure 1 ) for Nsp1 is larger than the inner radius of the S . cerevisiae NPC ( 18 to 20 nm [Alber et al . , 2007; Yang et al . , 1998] ) , and that for Nup98-glyco more than 60% of the inner radius of the X . tropicalis NPC ( ~25 nm [Eibauer et al . , 2015] ) . This would be consistent with the FG domains forming a pore-filling cohesive meshwork or condensed polymer brush , as implied by the selective phase model ( Frey and Gorlich , 2007; Frey et al . , 2006; Hulsmann et al . , 2012; Ribbeck and Gorlich , 2001 ) , at least for Nsp1 in S . cerevisiae , and possibly for Nup98-glyco in X . tropicalis taking into account the nanopore confinement ( Osmanovic et al . , 2012 ) . On exposing the FG domain films to NTF2 and Impβ , we find a remarkable quantitative similarity ( Figure 2 ) in the binding isotherms for Nsp1 , Nup98-glyco , and reg-FSFG , in spite of their chemical diversity . Future studies should test if a different behavior is found for less cohesive FG domains and/or other NTRs . Our analysis in terms of the Hill equation indicates negative cooperativity for NTR binding to the FG domain assemblies . This implies that the free energy of binding per NTR decreases with an increasing amount of bound NTR in the film , and that this decrease is more pronounced than would be expected for the one-to-one binding of NTRs to independent and uncorrelated binding sites ( i . e . , the Langmuir isotherm ) . Similar observations have been made previously ( Kapinos et al . , 2014; Schleicher et al . , 2014; Wagner et al . , 2015 ) . Key findings of the present work are that this negative cooperativity appears robust against variations in NTR and in the spread of FG motifs between Nsp1 , Nup98-glyco , and reg-FSFG , and that it can be reproduced , over orders of magnitude in NTR concentration , by a model that considers the FG domains as homogeneous polymers and the NTRs as featureless spheres , i . e . , ignoring chemical and structural heterogeneity . Our analysis suggests that NTR binding is largely determined by generic effects such as overall binding energy , crowding and excluded-volume interactions , and by the entropic costs of NTR absorption that , in turn , reduce the conformational freedom of the grafted and flexible FG domains . Such effects may well be expected since , at physiological NTR concentrations , the absorption of NTRs roughly doubles the total density of the films ( Figure 5 , top row ) . In our computational model , these effects are included , and the amount of NTR binding follows from collective low-affinity interactions – the balance between the overall cohesiveness of the FG domains ( εpp ) and the smeared-out , average NTR•FG domain interactions ( εpc ) . In this context , it is important to emphasize that enrichment of NTRs in an FG domain phase – in our model films but also in the NPC – vastly differs from NTR•FG motif binding under dilute conditions . This can be illustrated , for example , by comparing the enrichment of Nsp1 films for NTF2 and Impβ: the higher total Impβ•Nsp1 binding energy ( integrated over the Impβ surface ) does not translate into a higher enrichment than for NTF2 , even in the limit of low NTR concentrations ( Figures 1 and 4 , top rows ) . The binding energy is balanced by excluded-volume interactions and polymer cohesiveness and entropy , i . e . , by generic physical effects . At equilibrium , this balance is such that , in spite of the high binding energy between NTRs and FG domains , NTRs can exchange between the NPC and the nucleus/cytoplasm at minimal cost , thus facilitating cargo uptake and release . Remarkably , this overall balance appears rather finely tuned in several respects . Firstly , a small reduction in εpc produces a dramatic decrease in the uptake of colloids by the FG domain films ( Figure 6 ) . This is of major functional importance: by its strong dependence on εpc , the variable colloid uptake explains how FG domain assemblies in the NPC greatly favor the uptake of NTRs over other cytosolic proteins , which have been inferred to also bind FG domains ( Hough et al . , 2015 ) , albeit more weakly . This then explains how , because of the selective uptake , NTRs can efficiently translocate across the NPC while more weakly binding proteins cannot . We anticipate that this predicted fine tuning can be experimentally validated by systematically adjusting protein affinity to FG domain assemblies , which awaits further studies . In addition , from our comparison between experiment and computational modeling , the NTRs appear to favor the penetration of and binding into the FG domain assemblies , but only just so: penetration is inhibited for slightly larger FG domain cohesiveness ( i . e . , larger εpp ) , which results in NTRs preferentially binding on top of ( and not into ) the FG domain assembly ( Figure 5 ) . This is consistent with the remarkable selectivity exhibited by NPC transport and suggests that the cohesiveness of the FG domains is tuned to be as tight as possible to optimize exclusion of inert proteins by their size ( Eisele et al . , 2013 ) , while still sufficiently loose to facilitate penetration by NTRs . These features also lend themselves to further experimental validation: the model predicts how NTRs are distributed in FG domain assemblies and how these distributions depend on the interaction strength between NTRs and FG domains , which can be verified in future neutron or X-ray reflectometry measurements on FG domain films . Extrapolating our results on planar assemblies to the pore geometry of the NPC , we note that large structural changes can occur within the here determined parameter range of εpp: In analogous calculations for FG domains in an NPC-mimicking pore geometry of 50 nm diameter ( Osmanovic et al . , 2012; 2013b ) , we observed a transition ( Osmanovic et al . , 2012 ) between , on one hand , a central and pore-occluding condensate of FG domains in the NPC conduit , and on the other hand , a more open state with FG domains localized closer to the pore wall ( see , e . g . , Figure 7 in Osmanovic et al . , 2013b ) . Such large and collective transitions are required to facilitate transport of larger cargo•NTR complexes , with a size comparable to the nuclear pore diameter , through the NPC . Depending on the interaction strengths , FG domains assemblies may thus adopt qualitatively different behaviors , e . g . , absorption versus adsorption of NTRs and different types of polymer condensation . The overall FG domain interactions and NTR•FG affinity appear to be tuned close to the boundaries that separate these types of behaviors . The observed sensitivity to the values assigned to parameters offers an explanation of why different modeling approaches thus far have led to qualitatively different predictions of FG domain behavior in the NPC ( Ando et al . , 2013; 2014; Gamini et al . , 2014; Ghavami et al . , 2014; Miao and Schulten , 2009; Mincer and Simon , 2011; Moussavi-Baygi et al . , 2011a; 2011b; Opferman et al . , 2012; 2013; Osmanovic et al . , 2012; 2013b; Popken et al . , 2015; Tagliazucchi et al . , 2013; Wolf and Mofrad , 2008 ) , where most models succeed in capturing at least some aspects of experimental data on NPCs . The here observed sensitivity to parameter settings indicates that it is critical to calibrate computational models and their parameters against well-controlled experiments . We propose that experimental data obtained on well-defined FG domain assemblies ( such as those provided and used here ) , possibly complemented by structural data of isolated FG domains in solution ( Yamada et al . , 2010 ) , could serve as reference for such calibration . It would be desirable that identical sets of reference data are used by the computational modelers , as this would enable rigorous comparison between different computational approaches . To facilitate this effort , we provide data files of the NTR binding and thickness isotherms ( Figure 1—source data 1 ) . Our results have the advantage that they allow for a quantitative comparison between computational simulations and the experimental ( model ) system . For the entire NPC , such rigorous testing of computational models is presently complicated by experimental uncertainties in the locations of different FG domains inside the NPC , by the difficulties in accurately validating interaction parameters for the ensemble of FG domains in the NPC , and by the predicted bistable behavior of polymers grafted in nanopore geometries ( see , e . g . , Peleg et al . , 2011 and Osmanovic et al . , 2012 ) . That said , given the observed insensitivity to chemical heterogeneity of the FG domains , one can estimate the level of detail that will need to be included for building an appropriate model and understanding of mechanisms of selective transport in the NPC . The results presented here indicate that it is essential to take into account the flexible nature and cohesion of FG domains , as well as the crowding of NTRs that bind to the FG domain assemblies , but that heterogeneity at the scale of amino acids may only be of minor importance . In summary , we have used a bottom-up nanoscale system for a quantitative study of how NTRs interact with FG domains from the NPC . Highly similar binding isotherms were found for NTF2 binding to assemblies of Nsp1 from S . cerevisiae , of Nup98-glyco from X . tropicalis , and of an artificially designed regular FSFG construct; and for a different NTR , Impβ , binding to Nsp1 . This similarity suggests that – while the overall balance of interactions is essential – the detailed chemical and structural heterogeneity of the FG domains is not a critical factor for how NTRs interact with FG domain assemblies and thus with the NPC . This conclusion is supported by the good agreement – over several orders of magnitude of NTR concentration – between the experimental data and a physical model that treats the FG domains as chemically and structurally homogeneous polymers and the NTRs as spherical colloids . These results imply that the enrichment of NTRs into the FG domain phase is determined by generic physical effects – the flexible nature and spatial confinement of FG domains and the NTR size – and by the overall balance of a collection of low-affinity inter-FG-domain and NTR•FG domain interactions . Moreover , our computational data show that moderate changes in this overall balance cause remarkably large changes in the protein uptake by the FG domain assemblies , an observation that is fully consistent with the transport selectivity of the NPC . Given the success of our model in replicating NTR binding behavior in a nanoscale mimic for the NPC , we therefore propose that a similar approach may be viable to describe NPC transport selectivity , i . e . , in terms of generic polymer models , without necessarily taking into account the full amino acid sequences of FG domains and NTRs . However , our results also show that computational models need to be carefully calibrated to experimental data – such as has been done in this work – if they are to provide a meaningful contribution to the NPC field , since small ( ~30% ) changes in interaction parameters can result in qualitatively different behaviors .
We used the following FG domains: Nsp1 ( 64 . 1 kDa ) , amino acids 2 to 601 of Nsp1 with a C-terminal His10 tag; Nup98-glyco ( 58 . 3 kDa ) , amino acids 1 to 485 of Nup98 with ~30 O-GlcNAc modified S and T residues per chain ( Eisele et al . , 2013; Labokha et al . , 2013 ) and an N-terminal His14-TEV tag; reg-FSFG ( 34 . 1 kDa ) , an artificially designed regular FSFG domain with 16 repetitions of the sequence STPAFSFGASNNNSTNNGT and an N-terminal His14-TEV tag; reg-SSSG ( 32 . 2 kDa ) , a polypeptide identical to reg-FSFG but with phenylalanines replaced by serines . All FG domains were purified as described earlier ( Eisele et al . , 2010; 2013; Frey et al . , 2006; Labokha et al . , 2013 ) and stored at a concentration of 10 mg/mL in 50 mM Tris pH 8 . 0 , 6 M guanidine hydrochloride ( GuHCl ) at –80°C . We used the following NTRs: NTF2 from H . sapiens ( NTF2 , amino acids 1 to 127; 29 . 0 kDa for the homodimer ) ; and Impβ from S . cerevisiae ( 95 . 2 kDa ) . NTF2 , the W7A mutant of NTF2 and Impβ were expressed and purified as previously described ( Bayliss et al . , 1999; Eisele et al . , 2010 ) and stored at a concentration of 100 µM in working buffer ( 10 mM Hepes , pH 7 . 4 , 150 mM NaCl ) at −80°C . Before use , all protein constructs were diluted in working buffer to desired concentrations . For all our measurements , the residual concentration of GuHCl in the final solution was below 60 mM . The purity of proteins was verified by SDS-PAGE ( Figure 1—figure supplement 1 ) . The formation of FG domain films and the binding of NTRs to FG domain films were simultaneously followed by SE and QCM-D on the same surface and in a liquid environment ( Figure 1—figure supplement 3A ) ( Richter et al . , 2013 ) . To this end , we used a custom-built cuvette-like open fluid cell , placed in a Q-Sense E1 system ( Biolin Scientific AB , Västra Frölunda , Sweden; providing QCM-D data ) and mounted on a spectroscopic rotating-compensator ellipsometer ( M2000V , J . A . Woollam Co . , Lincoln , NE; providing SE data ) , as described in detail elsewhere ( Carton et al . , 2010 ) . In the SE measurements , ellipsometric angles ( Δ and ψ ) were acquired over a wavelength range of λ = 380 to 1000 nm at 70° angle of incidence and about 5 s time resolution . In the QCM-D measurements , frequency and dissipation shifts ( Δfi and ΔDi ) were acquired for six overtones ( i = 3 , 5 , . . . , 13; corresponding to resonance frequencies of fi ≈ 15 , 25 , . . . , 65 MHz ) with a time resolution better than 1 s . Prior to each measurement , the walls of the cuvette were passivated by incubation with a buffer solution containing 10 mg/mL of bovine serum albumin ( BSA; Sigma ) for 30 min . The cuvette was rinsed with buffer , ultrapure water and blow-dried with nitrogen . For the measurement , the cuvette was filled with ~2 mL working buffer , continuously stirred and held at a temperature of 23°C . Samples were injected directly into the buffer-filled cuvette at desired concentrations . To remove samples , the cuvette content was diluted by repeated addition of excess buffer and removal of excess liquid until the concentration of soluble sample , estimated from the dilution rate , was below 10 ng/mL . For measurements with NTF2 and Nsp1 , we used His tag capturing QCM-D sensors ( QSX340; Biolin Scientific AB ) . These sensors are coated with a thin layer of poly ( ethylene glycol ) ( PEG ) that exposes Cu2+ ions for the capture of His tagged molecules , and could be readily used as provided for FG domain film formation . We previously demonstrated that His tag capturing QCM-D sensors are suited to create dense monolayers of site-specifically anchored Nsp1 , and that such Nsp1 films have comparable properties to Nsp1 films formed on functionalized supported lipid bilayers ( SLBs ) ( Eisele et al . , 2012 ) . For measurements with Nup98-glyco , with reg-FSFG , and with Impβ and Nsp1 , we used SLBs as immobilization platform instead as this provided improved binding specificity . These measurements were performed on silica-coated QCM-D sensors that are optimized for combined QCM-D/SE experiments ( QSX335; Biolin Scientific AB ) . The sensors were cleaned by immersion in a 2% sodium dodecyl sulfate solution for 30 min , rinsed with ultrapure water , blow-dried with nitrogen , and exposed to UV/ozone ( BioForce Nanosciences , Ames , IA ) for 30 min . We mounted the cleaned sensors in the combined SE/QCM-D and functionalized their surface with supported lipid bilayers ( SLBs ) exposing Ni2+ ions for the capture of His tagged molecules , as described previously ( Eisele et al . , 2010; 2013 ) . Briefly , we used sonication to prepare small unilamellar lipid vesicles ( SUVs ) containing dioleoylphosphatidylcholine ( DOPC; Avanti Polar Lipids , Alabaster , AL ) and 3 to 10 mol-% of lipid analogs with headgroups comprising two or three Ni2+-chelating nitrilotriacetic acid moieties ( bis-NTA or tris-NTA ) ( Beutel et al . , 2014; Lata et al . , 2006 ) . SLBs were spontaneously formed by injecting SUVs ( at 50 µg/mL final concentration ) with NiCl2 ( at 10 µM final concentration ) into the buffer-filled SE/QCM-D fluid cell . SLB formation was monitored by SE and QCM-D and only SLBs of good quality ( i . e . , showing low QCM-D dissipation shifts , ΔD < 0 . 5 × 10–6 , and high frequency shifts , |Δf| > 25 Hz ) were used for further measurements . We formed FG domain films by injecting the FG domains directly into the SE/QCM-D cuvette equipped with a functionalized sensor . FG domain film formation was monitored , and FG domain concentration ( up to 2 . 9 µM ) and incubation time ( up to 90 min ) modulated to obtain FG domain films of desired grafting density ( Figure 1—figure supplement 3B ) . NTRs were titrated in discrete steps , first increasing and then decreasing , followed by at least 5 min of continuous rinsing with working buffer to remove NTRs from the solution phase . Incubation times were 5 to 30 min for titration steps with increasing NTR concentrations and 5 min for decreasing concentrations , i . e . , sufficiently long for equilibrium to be reached , as verified from the SE/QCM-D curves ( Figure 1—figure supplement 3C ) . We determined the thickness of FG domain films by fitting the QCM-D data to a continuum viscoelastic model , as described in detail previously ( Eisele et al . , 2012 ) . Briefly , we used the software QTM ( Johannsmann ) ( option “small load approximation” ( Johannsmann , 1999; 2008 ) ) . The FG domain films were modeled as homogeneous viscoelastic films with a storage modulus ( G’ ) and a loss modulus ( G” ) that depend on frequency in the form of a power law . The film density was fixed based on the areal mass density ( determined by SE , see below ) and partial specific volume of proteins , and the density of water . The semi-infinite bulk solution was assumed to be a Newtonian fluid with the density and viscosity of water . The interface between the FG domain film and the bulk solution is not ideally sharp , and the definition of film thickness thus not trivial . The viscoelastic model neglects the fuzzy interface and assumes a homogeneous film . However , because the acoustic contrast in polymer materials is generally high ( Johannsmann , 2008 ) , the ( acoustic ) thickness measured by QCM-D is expected to include a substantial part of the interfacial region that has a relatively low polymer density ( Domack et al . , 1997 ) . The specified errors represent a confidence level of one standard deviation ( 68% ) . In previous studies ( Eisele et al . , 2010; 2012; 2013 ) , we found that the thickness results obtained by QCM-D for FG domain films are comparable to within the specified confidence levels to those obtained with other techniques ( atomic force microscopy and SE ) , thus confirming that the thickness determination is robust . SE data were fitted to a model of multiple optically homogeneous layers , implemented in the software CompleteEASE ( J . A . Woollam Co . ) , to quantify protein surface densities . The fitting methods for QSX335 and QSX340 sensor substrates are described in detail in refs . ( Carton et al . , 2010 ) and ( Eisele et al . , 2012 ) , respectively . Irrespective of the substrate , the FG domain film was treated as a transparent Cauchy film with an effective optical thickness dSE and a wavelength-dependent refractive index n ( λ ) . Protein surface densities Γ where obtained through de Fejter’s equation ( De Feijter et al . , 1978 ) , i . e . , Γ = dSEΔn / ( MW × dn/dc ) , where Δn is the difference in refractive index between the FG domain film and the buffer solution ( assumed to be wavelength independent ) and MW the protein molecular mass . We used dn/dc = 0 . 18 cm3/g as refractive index increment for our protein films ( Richter et al . , 2013 ) . The resolution in Γ × MW was typically 0 . 5 ng/cm2 . Among the optical mass-sensitive techniques , SE is particularly suited to quantify the areal mass density of organic films up to a few 10 nm thick , because mass determination is virtually insensitive to the distribution of material within the film ( Richter et al . , 2013 ) . To model the experimental systems , we adapted a classical density functional theory approach that was derived , validated against Monte Carlo simulations , and applied in previous studies of polymers in a cylindrical confinement ( Osmanovic et al . , 2012; 2013b ) . We described the FG domain assembly as a collection of one-end-grafted polymers anchored homogeneously to the bottom of a cylinder of 100 nm in diameter and 120 nm in height ( Figure 3A ) . Each polymer was modeled as a string of physically and chemically identical beads that can rotate freely with respect to each other . The bead diameter and ( fixed ) bond length of the polymers were both taken to be 0 . 76 nm , such that each bead is in direct contact with its nearest neighbors on the string . Specifically , this yields a Kuhn length of 0 . 76 nm , i . e . , twice the contour length of an amino acid , and a persistence length of 0 . 38 nm , mirroring the flexibility of unfolded polypeptide chains ( Stirnemann et al . , 2013 ) . The number of beads per polymer chain was chosen to be 300 , 260 and 155 to represent Nsp1 , Nup98-glyco and reg-FSFG , respectively . These numbers match half the number of amino acids ( except the His tag used for anchorage ) of the respective FG domain constructs being modeled , and thus the contour length , to within 5% ( Table 1 ) . The polymer grafting density was set to 5 . 5 pmol/cm2 , i . e . , 3 . 3 polymers per 100 nm2 , to approximately match the FG domain grafting density in the NPC and the majority of the experimental data sets presented here . It was kept identical for the different FG domains to facilitate comparison and cross-validation of the computational results . For significantly lower grafting densities , lateral heterogeneity became increasingly important – as also observed experimentally ( Eisele et al . , 2013 ) – partly invalidating our approach in that regime . On the other hand , for significantly higher grafting densities , excluded-volume interactions were such that the computational convergence was much harder to achieve . NTF2 was represented as a colloidal sphere of 4 . 0 nm diameter , where the latter value was based on the diameter of a sphere that approximately includes the atomistic NTF2 dimer structure ( Bayliss et al . , 1999 ) . Similarly , Impβ was represented as a sphere of 6 . 0 nm , roughly matching the atomistic Impβ structure ( Forwood et al . , 2010 ) , and 50% larger than NTF2 , in agreement with a difference in diameter expected based on the mass difference between Impβ and NTF2 . The colloidal spheres could freely diffuse into and out of the cylinder with a chemical potential corresponding to the molar concentration , as described previously ( Osmanovic et al . , 2013b ) . The affinity between two polymer beads was parameterized by εpp , and the interaction between a polymer bead and a colloid by εpc , where εpp and εpc refer to the energy gain on bringing two polymer beads , and a polymer bead and a colloid , respectively , from infinite separation to hard contact . The attractive interactions were of the generic , exponential form with a decay length of 1 nm . εpp and εpc were varied to map the morphological space of the FG domain/NTR films as a function of the balance between inter and intra FG domain interactions on the one hand , and FG domain-NTR interactions on the other . We assumed that there was no attraction between the colloids . Rotational symmetry was assumed around the central axis of the cylinder ( perpendicular to the grafting surface ) , thus reducing the system to two dimensions for computational efficiency . For consistency with our previously developed algorithms , zero-density boundary conditions were imposed for both polymers and colloids at the side walls of the cylinder . At the cylinder top and bottom , zero-density boundary conditions were used for the polymers and periodic boundary conditions for the colloids . Local ( and a priori radially dependent ) film thicknesses were defined via iso-density profiles corresponding to a defined fraction ( 1% , 5% and 10% ) of the maximal density within the film . For comparison to the experimental data , these thicknesses were averaged across the cylinder ( taking into account the 2πr weighting factor in cylindrical coordinates ) . The results were verified for robustness against the exact threshold setting ( 1% , 5% and 10% , see Results section ) , and found to typically correspond to the thickness that includes ≳90% of the total film material . Amounts of bound colloids were expressed as areal densities averaged over the full surface ( of 100 nm diameter ) of the cylinder . To test for boundary-related artifacts in the computational results , areal colloid densities were also computed as averages across the central part ( of 50 nm diameter ) of the cylinder . Within the range of parameters relevant for comparison with experiment , the results agreed to within less than a factor of two . Polymer and colloid density profiles were determined from respective packing fractions versus the axis that is normal to the grafting surface . Analysis of a profile was performed on the average of profiles for different radial positions in the cylinder . | The cells of animals , plants and other eukaryotic organisms contain a compartment called the nucleus that contains most of the cell's genetic material . Proteins and other molecules – collectively known as cargos – can enter and exit the nucleus via tiny channels in the membrane that surrounds and protects it . Receptor proteins – called nuclear transport receptors – bind to potential cargos and shuttle them through the channels . This selective transport process relies on the nuclear transport receptors being attracted to flexible , spaghetti-like proteins that are anchored to the walls on the inside of each channel . However , because of their flexible and disordered nature , these so-called FG domains are difficult to study , and the details of the transport process are poorly understood . Zahn , Osmanović et al . decided to study how the FG domains behave and what happens when they interact with nuclear transport receptors by using ultrathin films made of just the FG domains . This is a good model system because the films are easier to study than the whole channels , but are likely to retain the essential properties of the real barrier formed in the nuclear envelope . Zahn , Osmanović et al . compared the binding of two nuclear transport receptors of different sizes , taken from humans and yeast , to FG domain films made from one of three different FG domains . The experiments showed that the different nuclear transport receptors bind to the different FG domains in very similar ways . Zahn , Osmanović et al . then used a computational model that essentially represented the FG domains as sticky spaghetti and the nuclear transport receptors as perfectly round meatballs . This sticky-spaghetti-with-meatballs model reproduced the experimental data , implying that the exact chemical make-up and structure of the molecules may not be critical for controlling the transport of cargo across the nuclear envelope . Future studies will test whether the generic physical features of nuclear transport receptors and FG domains can indeed explain how the cargo molecules pass through the nuclear envelope . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"physics",
"of",
"living",
"systems"
] | 2016 | A physical model describing the interaction of nuclear transport receptors with FG nucleoporin domain assemblies |
Overproduced yeast ribosomal protein ( RP ) Rpl26 fails to assemble into ribosomes and is degraded in the nucleus/nucleolus by a ubiquitin-proteasome system quality control pathway comprising the E2 enzymes Ubc4/Ubc5 and the ubiquitin ligase Tom1 . tom1 cells show reduced ubiquitination of multiple RPs , exceptional accumulation of detergent-insoluble proteins including multiple RPs , and hypersensitivity to imbalances in production of RPs and rRNA , indicative of a profound perturbation to proteostasis . Tom1 directly ubiquitinates unassembled RPs primarily via residues that are concealed in mature ribosomes . Together , these data point to an important role for Tom1 in normal physiology and prompt us to refer to this pathway as ERISQ , for excess ribosomal protein quality control . A similar pathway , mediated by the Tom1 homolog Huwe1 , restricts accumulation of overexpressed hRpl26 in human cells . We propose that ERISQ is a key element of the quality control machinery that sustains protein homeostasis and cellular fitness in eukaryotes .
Protein quality control ( PQC ) has emerged as a major mechanism for maintaining protein homeostasis and cellular fitness . Defects in the cellular machinery that governs PQC cause multiple human diseases including multisystem proteinopathy ( Brandmeir et al . , 2008; Watts et al . , 2004 ) and Amyotrophic Lateral Sclerosis ( ALS ) ( Johnson et al . , 2010; Kabashi and Durham , 2006 ) . Other diseases , such as cancer , can exhibit heightened dependency on PQC pathways , which underlies the hypersensitivity of multiple myeloma cells to proteasome inhibitors ( Cenci et al . , 2012; Deshaies , 2014 ) . Therefore , a deeper understanding of PQC will advance our understanding of both normal physiology and pathological states , and may enable novel approaches to treat multiple diseases . Ribosome biogenesis is an intricate process involving many chaperones and assembly factors ( Kressler et al . , 2010; Warner , 1999 ) . Ribosomal proteins made in excess over rRNA and other ribosomal proteins are among the most rapidly degraded proteins in eukaryotic cells ( Abovich et al . , 1985; Dephoure et al . , 2014; Gorenstein and Warner , 1977; Torres et al . , 2010 , 2007; Warner , 1977 ) , suggesting that proper coordination of synthesis and assembly is critical . Newly-synthesized human ribosomal proteins are subject to degradation by the proteasome in the nucleolus ( Lam et al . , 2007 ) , and we recently found that overexpressed yeast ribosomal proteins that fail to assemble are conjugated with ubiquitin and degraded by the proteasome in the nucleus ( Sung et al . , 2016 ) . Insoluble material that accumulates upon transient inhibition of the proteasome in yeast is strongly enriched for ribosomal proteins ( Sung et al . , 2016 ) , pointing to PQC of unassembled ribosomal proteins as a major pathway of proteostasis . However , the PQC pathway that mediates ERISQ remains unknown – an important gap in our understanding of PQC that we set out to address .
We evaluated 115 mutant yeast strains , each lacking a different non-essential ubiquitin-proteasome system ( UPS ) gene , for those that accumulated non-essential ribosomal protein Rpl26a tagged with a FLAG epitope ( Rpl26aFLAG ) upon its overexpression from the GAL10 promoter . Accumulation of Rpl26aFLAG in most mutants was similar to wild type ( WT ) and well below the level detected in rpl26a∆rpl26b∆ ( Figure 1—figure supplement 1A and B ) , which accumulated overexpressed Rpl26aFLAG due to lack of competition from endogenous Rpl26 ( Sung et al . , 2016 ) . Notably , Rpl26aFLAG accumulated to high levels in tom1∆ and ubc4∆ cells ( Figure 1A and Figure 1—figure supplement 1A and B ) . 10 . 7554/eLife . 19105 . 003Figure 1 . Ubc4/5 and Tom1 are the E2 and E3 enzymes responsible for ERISQ . ( A ) Rpl26aFLAG accumulates in tom1∆ and ubc4∆ . Accumulation of Rpl26aFLAG upon galactose induction in WT , tom1∆ and ubc4∆ cells was evaluated by SDS-PAGE and immunoblotting with the indicated antibodies . n = 3 biological replicates . ( B ) Rpl26aFLAG ubiquitination depends on Ubc4/Ubc5 . Rpl26aFLAG was induced in cells of the indicated genotypes and cell lysates were prepared and subjected to pull-down with UBA domain resin . Input and bound proteins were evaluated as in ( A ) . n = 3 biological replicates . ( C ) Rpl26aFLAG accumulates in tom1CA cells . As in ( A ) except that the Tom1 ligase-dead ( tom1CA ) mutant was used . n = 3 biological replicates . ( D ) Rpl26aFLAG ubiquitination depends on Tom1 . As in ( B ) except that cells expressing WT Tom1 or Tom1CA were treated with bortezomib for 45 min after addition of galactose . n = 3 biological replicates . ( E ) Rpl26aFLAG binds 3xHATom1 . Anti-HA immunoprecipitates from cells expressing 3×HATom1 and Rpl26aFLAG were immunoblotted with antibodies to HA , FLAG , and hexokinase . n = 3 biological replicates . ( F ) In vitro ubiquitination of Rpl26aFLAG by Tom1 . Rpl26aFLAG retrieved in 3xHATom1CA immunoprecipitates was supplemented or not with E1/E2/ubiquitin/ATP ( Ubi ) and Tom1 retrieved from untagged ( NT ) , 3xHATOM1 ( WT ) , or 3xHATOM1CA ( CA ) cells , as indicated . See detailed methods in Material and methods . n = 3 biological replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 19105 . 00310 . 7554/eLife . 19105 . 004Figure 1—figure supplement 1 . Identification of ERISQ defect in tom1∆ and ubc4∆ . ( A ) One hundred fifteen different knockout mutant strains each containing a plasmid that expressed Rpl26aFLAG from the GAL10 promoter were used . Rpl26aFLAG induced in rpl26a∆rpl26b∆ cells was used as a positive control . n = 1 biological replicate . ( B ) Quantification of data in ( A ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19105 . 00410 . 7554/eLife . 19105 . 005Figure 1—figure supplement 2 . Characterization of tagged and ligase-dead Tom1 . ( A ) Protein level of 3×HATom1 in cells expressing Tom1 ( NT ) , 3×HATom1 ( WT ) and 3×HATom1CA ( CA ) . n = 2 biological replicates . ( B ) Protein level of overexpressed Rpl26aFLAG induced in cells expressing Tom1 ( NT ) , Tom1HA , 3×HATom1 and tom1∆ . Note that a C-terminal tag on Tom1 compromises function , allowing for greater accumulation of galactose-induced Rpl26aFLAG . n = 2 biological replicates . ( C ) Cells of the indicated genotypes were spotted on SC-TRP and incubated at 30°C or 34°C for 2 days . n = 2 biological replicates . ( D ) As in ( C ) , except that cells of the indicated genotypes were spotted on YPD and incubated at the indicated temperatures for 2 days . n = 2 biological replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 19105 . 00510 . 7554/eLife . 19105 . 006Figure 1—figure supplement 3 . Tom1 mediates ubiquitination of overexpressed Rpl26a . ( A ) Polyubiquitination of Rpl26aFLAG . Rpl26aFLAG was induced in cells of the indicated genotypes and cell lysates were prepared and subjected to pull-down with UBA domain resin . Input and bound proteins were fractionated by SDS-PAGE and detected by immunoblot with the indicated antibodies . n = 2 biological replicates . ( B ) In vitro ubiquitination of Rpl26aFLAG by Tom1 . Lysates of cells expressing Rpl26aFLAG and the indicated allele of 3xHATom1 ( NT is untagged control ) were subjected to pull-down with anti-HA followed by addition of E1/E2/ubiquitin/ATP ( Ubi ) and incubation at 30°C for 1 hr . Reaction products were evaluated by immunoblot with the indicated antibodies . *indicates unmodified Rpl26aFLAG . **indicate ubiquitinated Rpl26aFLAG . Note the increase in ubiquitinated Rpl26aFLAG ( ** ) and the loss of unmodified Rpl26aFLAG ( * ) . n = 2 biological replicates . ( C ) Additional immunoblots of samples in Figure 1F . n = 3 biological replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 19105 . 006 Ubc4 is an ubiquitin-conjugating enzyme ( E2 ) that is paralogous to and functionally redundant with Ubc5 ( Seufert and Jentsch , 1990 ) . Thus , subsequent experiments were performed with ubc4∆ubc5∆ mutants . To test whether Ubc4/Ubc5 promoted ubiquitination of unassembled ribosomal proteins , we examined ubiquitin conjugates of overexpressed Rpl26aFLAG that accumulated in proteasome-deficient pre9∆ cells ( Sung et al . , 2016 ) . Ubiquitinated Rpl26aFLAG was detected in pre9∆ but not in ubc4∆ubc5∆pre9∆ cells ( Figure 1B ) , indicating that Ubc4/Ubc5 promote ubiquitination of excess Rpl26a . Tom1 is an E3 ubiquitin ligase of the HECT ( homologous to E6AP C terminus ) family . To investigate Tom1 function , we constructed tom1CA strains in which the endogenous TOM1 locus was mutated such that the catalytic cysteine3235 was changed to alanine ( tom1CA ) . We also appended a 3×HA epitope sequence to the 5’ end of both TOM1 and tom1CA , and confirmed that the 3xHATom1 and 3xHATom1CA proteins were expressed equivalently ( Figure 1—figure supplement 2A ) and the 3xHATom1 was functional ( Figure 1—figure supplements 2B–D ) . Using these strains , we established that Tom1 E3 activity was required for repression ( Figure 1C ) and ubiquitination ( Figure 1D and Figure 1—figure supplement 3A ) of overexpressed Rpl26aFLAG . Rpl26aFLAG was co-immunoprecipitated with 3×HATom1 ( Figure 1E ) . Upon addition of an ubiquitination cocktail , immunoprecipitations of wild type but not mutant 3×HATom1 ubiquitinated co-precipitated Rpl26aFLAG ( Figure 1—figure supplement 3B ) . Importantly , the activity defect of the 3×HATom1CA immunoprecipitate was complemented by adding 3×HATom1 but not 3×HATom1CA prior to the ubiquitination reaction ( Figure 1F and Figure 1—figure supplement 3C ) . To identify the population of Rpl26aFLAG targeted by Tom1 , we performed sucrose gradient fractionation . Mutant tom1CA cells , like cells treated with the proteasome inhibitor bortezomib ( Sung et al . , 2016 ) , accumulated unassembled Rpl26aFLAG that co-fractionated with 3×HATom1CA ( Figure 2A; note that 3×HATom1 and 3×HATom1CA fractionated similarly ) . Co-immunoprecipitation of 3×HATom1 or 3×HATom1CA with Rpl26aFLAG was only detected in these low MW fractions ( Figure 2B ) . Moreover , ubiquitinated Rpl26aFLAG detected in low MW fractions from bortezomib-treated cells was almost entirely lost from tom1CA cells ( Figure 2B ) . Consistent with the reported localization of Tom1 ( Huh et al . , 2003 ) , Rpl26aFLAG or Rpl26aGFP that accumulated upon their transient overexpression in tom1CA cells were found in the nucleus and nucleolus ( Figure 2C ) . Taken together , these data provide strong evidence that overexpressed Rpl26a failed to assemble into ribosomes and was directly bound and ubiquitinated by Tom1 in the nuclear/nucleolar compartments . 10 . 7554/eLife . 19105 . 007Figure 2 . Tom1 functions in non-ribosomal fractions . ( A ) Sucrose gradient fractionation behavior of 3xHATom1 and Rpl26aFLAG upon galactose induction of Rpl26aFLAG in 3×HATOM1 or 3×HATOM1CA cells . T indicates total extract . n = 2 biological replicates . ( B ) Tom1 is required for ubiquitination of unassembled Rpl26aFLAG . Left: experimental scheme . Right: cells were treated with bortezomib for 30 min after induction of Rpl26aFLAG with galactose and then lysed and fractionated as in panel A prior to being processed as depicted in panel B . n = 2 biological replicates . ( C ) Rpl26a accumulates in the nucleus of tom1CA cells . Left: Subcellular fractionation of Rpl26aFLAG induced in WT and tom1CA cells . Histone H3 and Hexokinase were used as nuclear and cytoplasmic markers , respectively . CA refers to Tom1-Cys3235Ala . Right: Fluorescence microscopy of Rpl26aGFP induced in WT and tom1CA cells . Nop56-RFP marks nucleoli . Shown at far right is the percentage of GFP positive cells . n = 2 biological replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 19105 . 007 To address whether Tom1 might have a broader role in promoting degradation of excess ribosomal proteins other than Rpl26a , we evaluated accumulation of a set of eight ectopically overexpressed ribosomal proteins in tom1∆ and WT cells . Similar to what we observed with bortezomib ( Sung et al . , 2016 ) , deletion of TOM1 enabled increased accumulation of at least seven of them ( Figure 3—figure supplement 1A ) . We next sought to test whether Tom1 promoted degradation of unassembled ribosomal proteins in cells in which they were not deliberately overexpressed . We reasoned that if this is the case , Tom1 should directly associate with ribosomal proteins . Mass spectrometry of 3xHATom1 immunoprecipitates from bortezomib-treated cells revealed enrichment for several ribosomal proteins , including Rpl26b ( Figure 3—figure supplement 1B and Supplementary file 3A ) . Ribosomal proteins are commonly identified in purified ubiquitin conjugates ( Mayor et al . , 2007 , 2005; Peng et al . , 2003 ) or in ubiquitination site mapping experiments that rely on purification of the GlyGly dipeptide that remains attached to a lysine side chain following digestion of an ubiquitin conjugate with trypsin ( Kim et al . , 2011; Lesmantavicius et al . , 2014; Porras-Yakushi and Hess , 2014; Porras-Yakushi et al . , 2015; Sarraf et al . , 2013; Swaney et al . , 2013; Udeshi et al . , 2013b; Wagner et al . , 2011 ) . Thus , we reasoned that if Tom1 plays a broad role in PQC of unassembled ribosomal proteins as suggested by the experiments shown in Figure 3—figure supplement 1A and B , perhaps it accounts for the frequent recovery of ribosomal proteins in prior global ubiquitin conjugate profiling efforts . To address this possibility , we performed quantitative GlyGly profiling of tom1∆ and TOM1 cells using SILAC ( Figure 3—figure supplement 1C ) to identify changes in the level of ubiquitination of specific lysines that occur upon loss of Tom1 . Analysis of three biological replicates ( Figure 3A and Figure 3—figure supplement 2A and Supplementary file 3B ) revealed 1980 unique ubiquitination sites in 920 distinct proteins , of which 972 unique sites in 532 proteins were quantified . All three tom1∆ biological replicates exhibited lower overall ubiquitination than wild type , suggesting a major role for Tom1 in PQC . Of the 141 sites that exhibited a ≥two-fold decrease in ubiquitination in tom1∆ , 51 ( 36% ) were in ribosomal proteins ( Figure 3—figure supplement 2B ) . Moreover , of the ubiquitinated peptides derived from ribosomal proteins , >50% ( 51 of 101 ) decreased in abundance in tom1∆ . By comparison , of 837 non-ribosomal sites identified , only 11% decreased in abundance in tom1∆ . SILAC analysis of the unfractionated cell lysates indicated that the reduction in ribosomal ubiquitin conjugates in tom1∆ was not due to reduction in total ribosomal protein levels ( Figure 3B ) . Gene ontology analysis of the GlyGly profiling data confirmed that ubiquitination of ribosomal proteins ( Figure 3C ) , particularly those of the large ( 60S ) subunit ( Figure 3D ) , was disproportionately impacted by loss of Tom1 . These trends are clearly evident from a plot of the top 25 Tom1-dependent modification sites within all proteins ( Figure 3—figure supplement 2C ) or just ribosomal proteins ( Figure 3E ) . To address whether the strong effects on ribosomal protein ubiquitination seen in tom1∆ cells were due specifically to loss of Tom1’s E3 activity , the GlyGly SILAC analysis was repeated with WT and tom1CA cells . Quantitative analysis of the data confirmed a disproportionate loss of ribosomal protein ubiquitination ( Figure 3—figure supplement 2D ) . 10 . 7554/eLife . 19105 . 008Figure 3 . Diminished ubiquitination and accumulation of insoluble ribosomal proteins in tom1 cells . ( A ) Diminished ubiquitination of ribosomal proteins in tom1∆ . Scatter plot of the SILAC ratios ( tom1∆/WT ) for GlyGly-modified peptides identified in biological replicate 1 versus 2 . Sites with the largest decrease in ubiquitination are annotated . The other pairwise comparisons are in Figure 3—figure supplement 2A . n = 3 biological replicates . ( B ) Column scatter plot representing the distribution of ratios ( tom1∆/WT ) for proteins of the large ( circles ) and small ( square ) ribosome subunits . A red bar indicates the mean . ( C ) Violin plot of gene ontology analysis of ubiquitinated proteins that had one or more ubiquitination site that decreased by ≥ 2-fold . The most strongly affected categories are shown . The number in parentheses refers to the disproportionate enrichment for the category in the top 10% of identifications and is the Benjamini and Hochberg corrected p-value from a Fisher Exact test . ( D ) Violin plot representing the distribution of ubiquitin site occupancy ratios ( tom1∆/WT ) for the large ( blue ) and small ( green ) ribosome subunits , and non-ribosomal proteins ( red ) . ( E ) The 25 ribosomal ubiquitination sites with the largest decrease in ubiquitin occupancy in tom1∆ . +++p<0 . 001; ++p<0 . 01; +p<0 . 05 . Each site was observed in at least two of the three biological replicates . The error bars represent 95% confidence intervals . Note that ubiquitination at K37 and K69 in Rpl26b was decreased by 2 . 4-fold and 1 . 6-fold in tom1∆ , respectively ( Supplementary file 3B ) . ( F , G ) Insoluble ribosomal proteins accumulate in tom1∆ . ( F ) Detergent-insoluble pellet fractions isolated from lysate ( Total ) of indicated cells were analyzed by SDS-PAGE and staining with Coomassie blue or immunoblotting with the indicated antibodies . The pellet fraction is overloaded 20-fold compared to the total and supernatant fractions . n = 2 biological replicates . ( G ) Scatter plot representing ∆iBAQ of biological replicate B vs . A for insoluble proteins in tom1∆ mutants . Ribosomal proteins with the largest increase in the pellet fraction upon TOM1 deletion , and Rpl1 and Rpl3 are annotated . Pearson’s r-value is indicated on top of the plot . The other pairwise comparisons are in Figure 3—figure supplement 3B . n = 3 biological replicates . ( H ) Gene ontology analysis of proteins exhibiting increased insolubility in tom1∆ . Analysis is the same as for panel C . DOI: http://dx . doi . org/10 . 7554/eLife . 19105 . 00810 . 7554/eLife . 19105 . 009Figure 3—figure supplement 1 . Tom1 targets a broad range of overexpressed and endogenous ribosomal proteins . ( A ) Relative levels of the transiently overexpressed , indicated ribosomal proteins ( all tagged with a His6-HA-protein A ZZ domain epitope ) in WT and tom1∆ mutants . n = 2 biological replicates . *indicates the expected size of protein . ( B ) Left: Cells expressing untagged-Tom1 or 3×HATom1 were treated with bortezomib for 1 hr . Total cell extracts were adsorbed to HA resin and the bound fractions were analyzed by mass spectrometry . 80 proteins that increased >1 . 5 fold in 3×HATom1 samples versus untagged samples were categorized into ribosomal or non-ribosomal proteins . Right: The number of peptides derived from the 9 proteins with the highest fold changes in 3×HATom1 vs . untagged samples are shown . n = 1 biological replicate . ( C ) Schematic diagram of SILAC-based quantitative K-ε-GlyGly mass spectrometry ( MS ) strategy to identify Tom1-depdendent ubiquitination sites . DOI: http://dx . doi . org/10 . 7554/eLife . 19105 . 00910 . 7554/eLife . 19105 . 010Figure 3—figure supplement 2 . Quantitative GlyGly proteomic analyses of tom1 mutants . ( A ) Scatter plots of the SILAC ratios ( tom1∆/WT ) of biological replicate 3 vs . 2 ( left ) and biological replicate 3 vs . 1 ( right ) for GlyGly-modified peptides in tom1∆ and WT cells . Sites with the largest decrease in ubiquitination are annotated . These data accompany Figure 3A . n = 3 biological replicates . ( B ) Histogram of tom1∆/WT ubiquitination site ratios for ribosomal proteins of the large and small subunits , and non-ribosomal proteins . For each protein category , the fraction of total with a given ratio is plotted . ( C ) The 25 ubiquitination sites with the largest decrease in ubiquitin occupancy in tom1∆ . +++p<0 . 001; ++p<0 . 01: +p<0 . 05 . Each site was observed in at least two of the three biological replicates . The error bars represent 95% confidence intervals . ( D ) Left: same as ( B ) , except that a tom1CA mutant was used instead of tom1∆ . Right: violin plot representing the distribution of ubiquitin site occupancy ratios for ribosomal proteins of the large ( blue ) and small ( green ) subunit , and non-ribosomal proteins ( red ) . n = 3 biological replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 19105 . 01010 . 7554/eLife . 19105 . 011Figure 3—figure supplement 3 . Endogenous ribosomal proteins accumulate as insoluble species in tom1∆ mutants . ( A ) Accumulation of insoluble proteins in tom1 mutant cells is independent of lysis method or buffer . Cells of the indicated genotypes were lysed with glass beads in the presence of 3 different lysis buffers as indicated below gel image , and fractionated into detergent-soluble and insoluble fractions . Samples were separated by SDS-PAGE and stained with Coomassie Blue . The pellet fraction is overloaded 10-fold compared to the total and supernatant fractions . n = 2 biological replicates . T , S and P indicates total , soluble and pellet fractions , respectively . Note that results shown here are qualitatively similar to results in Figures 3F and 6D–G , even though the method employed to generate those figures employed lysis of spheroplasts in an EDTA-containing buffer . The reason for the higher background in this panel relative to the others is that the pellet fractions were not washed prior to analysis . ( B ) Scatter plots representing the ∆iBAQ of biological replicate C vs . A ( left ) and biological replicate C vs . B ( right ) for insoluble proteins in tom1∆ mutants . Pearson’s r-value is indicated on top of the plot . These data accompany Figure 3G . n = 3 biological replicates . ( C ) The 20 proteins with the largest increase in the pellet fraction upon TOM1 deletion . Bars represent the average ∆iBAQ values with error bars indicating the standard error of the mean ( SEM ) . ( D ) Same as ( C ) , except the top 20 ribosomal proteins are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 19105 . 011 If Tom1 mediates degradation of unassembled ribosomal proteins in unperturbed cells , there should not only be a decrease in ribosomal ubiquitin conjugates in tom1 mutants , but a commensurate increase in unassembled ribosomal proteins that fail to be degraded . Since preliminary sucrose gradient fractionations did not reveal high levels of unassembled ribosomal proteins in tom1∆ cells ( data not shown ) , we reasoned that over time , undegraded excess ribosomal proteins might aggregate and collect in insoluble deposits . To investigate this matter , we prepared detergent-insoluble fractions from WT cells treated with or without bortezomib ( btz ) and tom1∆ cells , and evaluated them for their content of ribosomal proteins . Detergent-insoluble proteins , including Rpl1 and Rpl3 , were greatly increased in tom1∆ cells compared to WT cells ( Figure 3F , Pellet ) . High accumulation of insoluble proteins in tom1∆ cells was evident regardless of the method or buffer employed for cell lysis ( Figure 3—figure supplement 3A ) . This observation was confirmed and extended by mass spectrometry coupled with label-free absolute quantification using iBAQ ( intensity-Based Absolute Quantification ) ( Geiger et al . , 2012 ) . The insoluble proteins that exhibited the largest increase in tom1∆ cells were ribosomal proteins including Rpl1 and Rpl3 ( Figure 3G , Figure 3—figure supplement 3B , and Supplementary file 3C ) . Gene ontology analysis ( Figure 3H ) and a plot of the top 20 detergent-insoluble proteins ( Figure 3—figure supplement 3C ) indicated that ribosomal proteins , including those of both the 60S and 40S subunits ( Figure 3—figure supplement 3D ) comprise the major class of aggregating proteins in tom1∆ cells . GlyGly profiling and analysis of insoluble proteins in tom1 mutants both pointed to a broad role of Tom1 in ubiquitinating and degrading excess , unassembled ribosomal proteins . This raised a critical question that is common to all PQC pathways yet is poorly understood: how does Tom1 ubiquitinate so many different ribosomal proteins , yet manage to maintain some level of specificity for unassembled forms ? To begin to address this question , we constructed a mutant Rpl26a that did not bind to rRNA . Two positively-charged clusters in Rpl26 – RRKARK ( amino acids 12–17 ) and a patch formed by R27 , R28 , R51 , and R52 – mediate binding to 5 . 8S rRNA and assembly into ribosomes ( Babiano et al . , 2012 ) ( Figure 4A ) . We mutated various combinations of these residues to glutamate and observed that some mutants exhibited even less accumulation than overexpressed WT Rpl26aFLAG ( Figure 4B ) . 10 . 7554/eLife . 19105 . 012Figure 4 . A short stretch of positively-charged residues in Rpl26a that mediates rRNA binding promotes association with Tom1 . ( A ) Top: The first 54 amino acids of Rpl26a . Sequences adjacent to rRNA are boxed . Arginine residues targeted for mutation are in red . Bottom: Relative positions of arginines and rRNA based on the atomic model of the yeast 80S ribosome ( PDB files 3U5D and 3U5E ) . Orange and pink ribbons correspond to 25S and 5 . 8S rRNA , respectively . Blue ribbon corresponds to Rpl26 . Predicted regions ( #1 and #2 ) for rRNA binding are highlighted in yellow and boxed . ( B ) Differential accumulation of WT and mutant Rpl26aFLAG upon galactose induction . n = 2 biological replicates . ( C ) Top: Ribosome assembly of WT Rpl26aFLAG or Rpl26a-4EFLAG induced in rpl26a∆rpl26b∆ cells . Bottom: Same as above except that MG132 was added 30 min after addition of galactose . T indicates total extract . n = 2 biological replicates . ( D ) Polyubiquitination of Rpl26a-4EFLAG . Assay was performed as described for Figure 1D . Samples in ‘+’ lanes were treated with deubiquitinating enzyme Usp2 prior to processing for SDS-PAGE , to demonstrate that high MW species were modified with ubiquitin . n = 2 biological replicates . ( E ) The Rpl26-4E mutation disrupts binding to Tom1 . Lysates from cells of the indicated genotypes were subjected to pull-down with anti-HA followed by SDS-PAGE and immunoblotting for the indicated proteins . n = 2 biological replicates . ( F ) Protein level of Rpl26aFLAG mutants upon galactose induction in WT and tom1∆ cells . n = 2 biological replicates . ( G ) Fluorescence images of Rpl26a4E-GFP induced in WT cells . Nop56-RFP marks nucleoli . Dashed circles indicate nuclear region as judged by DAPI staining . n = 2 biological replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 19105 . 01210 . 7554/eLife . 19105 . 013Figure 4—figure supplement 1 . Rpl26a-4E mutant is unstable and degraded by Doa10 in the nucleus/nucleolus . ( A ) Similar accumulation of Rpl26aFLAG and Rpl26a-4EFLAG upon galactose induction in the presence of bortezomib . Total cell lysates were evaluated by SDS-PAGE and immunoblotting with the indicated antibodies . n = 1 biological replicate . ( B ) Differential accumulation of WT and Rpl26a-4E upon galactose induction in WT and known PQC mutants was evaluated by SDS-PAGE and immunoblotting with the indicated antibodies . n = 1 biological replicate . ( C ) Fluorescence images of Rpl26a-4EGFP induced in doa10∆ cells . n = 2 biological replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 19105 . 013 Unlike WT Rpl26aFLAG , Rpl26-4EFLAG ( R12E , R13E , R16E , K17E ) did not accumulate or assemble into ribosomes in rpl26a∆rpl26b∆ cells ( Figure 4C , top panels ) . However , upon inhibition of the proteasome with MG132 or bortezomib , Rpl26a-4EFLAG accumulated ( Figure 4—figure supplement 1A ) and was detected in the low MW fractions of a sucrose gradient ( Figure 4C , bottom panels ) , where it was ubiquitinated , albeit to a lesser extent than unassembled WT Rpl26aFLAG ( Figure 4D ) . Strikingly , Rpl26a-4EFLAG exhibited poor association with 3xHATom1 ( Figure 4E ) and showed only very weak accumulation in tom1∆ ( Figure 4F ) , suggesting that it is not primarily a Tom1 substrate but is re-directed to another QC pathway . Consistent with this idea , Rpl26a-4EFLAG accumulated in doa10 mutants ( Figure 4—figure supplement 1B ) . The failure of Rpl26a-4EFLAG to be targeted by Tom1 was not due to a defect in its nuclear localization ( Figure 4G and Figure 4—figure supplement 1C ) . We suggest that upon its import into the nucleus , Rpl26a-4E becomes a substrate for Doa10 that is localized to the inner nuclear membrane ( Deng and Hochstrasser , 2006 ) . Taken together , our results suggest that residues of Rpl26a that mediate interactions with rRNA and are buried when incorporated into the ribosome are also required for ubiquitination by Tom1 when Rpl26a fails to assemble . To pursue this observation further and probe its potential generality , we focused on Rpl4 , because assembly of Rpl4 has been studied in some depth ( Stelter et al . , 2015 ) . Upon its synthesis , Rpl4 binds the dedicated chaperone Acl4 . The Rpl4–Acl4 then recruits the karyopherin Kap104 , for import into the nucleus . Importantly , a crystal structure is available for Rpl4–Acl4 ( F . H . and A . H . , submitted ) , and the binding site of Kap104 on the complex has been mapped ( [Stelter et al . , 2015]; F . H . and A . H . , submitted ) . In our GlyGly profiling efforts , measurements were obtained for three ubiquitination sites on Rpl4: K55 , K308 , and K338 ( 10- , 8- , and 1 . 6-fold decrease in tom1∆ , respectively ) ( Supplementary file 3B ) . The crystal structure of the ribosome indicates that K55 and K308 , whose ubiquitinations exhibited the strongest dependence on Tom1 , contact rRNA and are not accessible for modification in the mature ribosome ( Ben-Shem et al . , 2011 ) . Interestingly , the crystal structure of the Acl4–Rpl4 complex revealed that Acl4 conceals K55 , and a structural model of Kap104 docked to Acl4–Rpl4 indicates that it protects K308 and K338 . Upon import of Rpl4–Acl4–Kap104 into the nucleus , Kap104 is dissociated through the action of Ran-GTP ( Kressler et al . , 2012 ) ( F . H . and A . H . , submitted ) . To test if this exposes the C-terminal region of Rpl4 to Tom1 ( as would be the case if assembly of Rpl4 was delayed following nuclear import and release of Kap104 ) , we performed an in vitro ubiquitination assay with purified substrates . 3xHATom1 immunoprecipitates readily ubiquitinated Rpl4 in binary Acl4–Rpl4 complexes but not in ternary Acl4–Rpl4–ctKap104 complexes , despite the fact that at least eight lysines of Rpl4 should remain exposed in the ternary complex ( red circle , Figure 5A ) . Ubiquitination of Rpl4 within the binary complex required its extended C-terminus because it was eliminated when the C-terminal region was truncated at residue 276 ( Figure 5B ) . These data , along with those on Rpl26a , suggest that Tom1 selectively recognizes and ubiquitinates ribosomal proteins via residues that are only accessible in the unassembled state . Notably , pulse-chase labeling of yeast cells revealed that Rpl4∆63–87 , which lacks the loop region that binds Acl4 , transiently associates with Tom1 ( see Figure 3B of Stelter et al . , 2015 ) . However , it is unclear if loss of Tom1 during the chase was due to degradation or incorporation of Rpl4∆63–87 molecules into ribosomes . 10 . 7554/eLife . 19105 . 014Figure 5 . Tom1 acts through residues that are normally inaccessible in the structure of the mature ribosome . ( A ) Structure of Rpl4 within the mature ribosome ( PDB ID 4V88 ) . Lysine residues are colored blue , with K55 , K308 , and K338 colored red . Areas involved in binding Acl4 and ctKap104 are indicated . The exact boundaries of the Kap104 binding site are not known . The globular central domain , which is fully exposed in the ternary Acl4–Rpl4–Kap104 complex but is not ubiquitinated , is circled in red . ( B ) Ubiquitination of Acl4–Rpl4 by 3xHATom1 . Anti-HA immunoprecipitates from untagged ( NT ) , 3xHATOM1 ( WT ) , and 3xHAtom1CA ( CA ) cells were supplemented or not with E1/E2/ubiquitin/ATP ( Ubi ) and purified Acl4–FLAGRpl4 , Acl4-FLAGRpl4∆ext and Acl4–FLAGRpl4–ctKap104 proteins . Samples were analyzed by SDS-PAGE and staining with Coomassie blue or immunoblotting with the indicated antibodies . WT , ∆ , and K refer to Acl4–FLAGRpl4 , Acl4-FLAGRpl4∆ext and Acl4–FLAGRpl4–ctKap104 , respectively . See detailed methods in Material and methods . n = 3 biological replicates . ( C ) Tom1 preferentially targets lysines that are inaccessible in mature ribosomes . Lysine residues shown are those from large subunit ribosomal proteins in Figure 3E that are incorporated in the model for the structure of the yeast ribosome ( pdb 4V88 ) . The structure of a HECT domain–donor ubiquitin complex ( pdb 4LCD ) predicts that a gap of radius 25 Å must be present for Tom1 to access a lysine for ubiquitination . Two of the sites ( Rpl10 K30 and Rpl24 K69 ) are accessible in the 60S large subunit but become inaccessible upon formation of the 80S ribosome . DOI: http://dx . doi . org/10 . 7554/eLife . 19105 . 014 To assess more generally if Tom1 targets lysines that are inaccessible in the mature ribosome ( pdb 4V88 ) , we examined the disposition of the major Tom1-dependent ubiquitination sites on large subunit proteins reported in Figure 3E . For this analysis , we used the structure of the HECT domain of Rsp5 covalently conjugated to both a donor ubiquitin and a substrate acceptor ( pdb 4LCD ) . We asked whether the epsilon amino group of a given lysine within the mature large subunit could conceivably make contact with the active site cysteine of Rsp5 . Of the 18 lysines that could be observed in the 4V88 structure , 13 in the free 60S and 15 in the assembled 80S were not accessible to the probe ( Figure 5C ) . Taken together , our data suggest strongly that Tom1-dependent ubiquitination events generally occur on ribosomal proteins prior to their assembly into the ribosome , on residues that normally are either buried in the ribosome , engage in salt bridges , or are otherwise shielded from contact . We next turned our attention to the phenotypic effects of Tom1 deficiency . If a limited capacity to degrade excess ribosomal proteins contributes to the temperature-sensitive growth defect of tom1 mutants ( Utsugi et al . , 1999 ) ( Figure 1—figure supplement 2C and D ) , we reasoned that conditions that foster imbalances in the production of ribosome components should exacerbate this defect . To test this , we performed three different perturbations . First , tom1CA cells ( but not WT cells ) were extremely sensitive to constitutive overexpression of RPL26A ( Figure 6A ) but not RPL26A-4E ( Figure 6—figure supplement 1A ) from the GAL10 promoter . Second , we created a situation in which ribosomal proteins as a group are made in excess of rRNA via depletion ( using the auxin-inducible degron ( AID ) [Morawska and Ulrich , 2013] ) of proteins involved in rRNA synthesis including Rrn3 ( transcription factor for RNA polymerase I ) , Rpa190 ( RNA polymerase I largest subunit ) and Hmo1 ( regulator of transcription by RNA polymerase I ) . Each of these depletions caused a synthetic growth defect when combined with tom1CA ( Figure 6—figure supplement 1B ) . The effect of combining hmo1∆ and tom1CA mutations was even more severe ( Figure 6B ) . Third , we manipulated cells such that assembly of a single ribosomal protein was impaired , via deletion of the Rpl4-selective chaperone Acl4 ( Stelter et al . , 2015 ) . Inactivation of TOM1 in an acl4∆ background caused a substantial synthetic growth defect ( Figure 6C ) . 10 . 7554/eLife . 19105 . 015Figure 6 . Defective ribosome assembly homeostasis and proteostatic collapse in tom1 mutant cells . ( A–C ) Hypersensitivity of tom1CA cells to imbalances in ribosome components . ( A ) Cells of the indicated genotypes were spotted in serial 10-fold dilutions on glucose or galactose medium and incubated at 30°C for 2 days . ev refers to empty vector . n = 2 biological replicates . ( B , C ) As in ( A ) except that cells of the indicated genotypes were spotted on YPD . n = 2 biological replicates . ( D–G ) Massive accumulation of insoluble proteins in tom1 mutant cells . Cells of the indicated genotypes were lysed and fractionated into detergent-soluble and insoluble fractions , which were separated by SDS-PAGE and stained with Coomassie Blue . The pellet fraction is overloaded 20-fold compared to the total and supernatant fractions . n = 2 biological replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 19105 . 01510 . 7554/eLife . 19105 . 016Figure 6—figure supplement 1 . Tom1 is required for maintaining proteostasis . ( A ) Cells of the indicated genotypes were spotted in serial 10-fold dilutions on glucose or galactose medium and incubated at 30°C for 2 days . ev refers to empty vector . n = 2 biological replicates . ( B ) Cells of the indicated genotypes were spotted as serial 10-fold dilutions on YPD with or without 1 mM auxin and incubated at 30°C for 2 days . AID refers to auxin-inducible degron and CA refers to the Cys3235Ala mutation in TOM1 . n = 2 biological replicates . ( C ) List of genetic interactions with tom1∆ as reported in the Saccharomyces Genome Database ( www . yeastgenome . org ) . ( D ) Synthetic growth defects of rps29a∆tom1∆ double mutants . Cells of the indicated genotypes were spotted in serial 10-fold dilutions on YPD and incubated at 30°C for 2 days . n = 2 biological replicates . ( E ) same as panel B except medium was supplemented or not with 2 mM paromomycin . DOI: http://dx . doi . org/10 . 7554/eLife . 19105 . 016 If the synthetic growth defects described above arose from a catastrophic failure of proteostasis , we reasoned that this might manifest itself in the detergent-insoluble fractions of double mutant cells . Strikingly , there was a massive increase in detergent-insoluble proteins when RNA Pol I transcription was diminished ( hmo1∆tom1CA; Figure 6D ) or Rpl4 assembly was perturbed ( acl4∆tom1CA; Figure 6E ) in a tom1CA background . Identification of Tom1 as a key mediator of ERISQ may rationalize numerous genetic interactions that have been reported for tom1∆ . In addition to suppressors that map to stress response/chaperone pathways and genes involved in ribosome protein expression ( PKA pathway ) , deletions in 36 different genes encoding ribosomal proteins exhibit synthetic negative genetic interaction with tom1∆ ( Costanzo et al . , 2010 ) ( Figure 6—figure supplement 1C ) . Based on the effects of the acl4∆ mutant ( Figures 6C , E ) , deletions of one copy of duplicated ribosomal protein genes are predicted to create an imbalance in ribosomal proteins resulting in severe growth and proteostasis defects in a tom1 background . Consistent with this prediction , deletion of one of the two copies of RPS29 led to a synthetic growth defect ( Figure 6—figure supplement 1D ) and enormous accumulation of insoluble proteins in a tom1∆ mutant ( Figure 6F ) . Given the major role of ribosome production in the cellular economy , we evaluated the relative impact of the Tom1-dependent ERISQ pathway on overall proteostasis by comparing the amount of insoluble proteins in well-characterized PQC mutants including doa10∆ and hrd1∆ ( ERAD; ER-associated degradation; [Vembar and Brodsky , 2008] ) , ltn1∆ ( RQC; ribosome bound QC; [Bengtson and Joazeiro , 2010; Brandman et al . , 2012; Defenouillere et al . , 2013; Verma et al . , 2013] ) , san1∆ ( nuclear PQC; [Gardner et al . , 2005] ) and ubr1∆ ( ERAD and cytoplasmic PQC; [Eisele and Wolf , 2008; Heck et al . , 2010] ) . Under unperturbed conditions , tom1∆ cells contained significantly greater amounts of insoluble proteins compared to these other PQC mutants ( Figure 6G , Pellet ) , suggesting that ERISQ is a major player in maintaining proteostasis in yeast . A prior study in human cells demonstrated that a significant fraction of newly-synthesized ribosomal proteins imported into the nucleus is degraded by the UPS ( Lam et al . , 2007 ) , suggesting that a PQC mechanism for unassembled ribosomal proteins is conserved in higher eukaryotes . To test this possibility , we evaluated transient expression of human Rpl26FLAG ( hRpl26FLAG ) in T-REx-293 cells treated with or without MG132 or bortezomib . Cells treated with these proteasome inhibitors accumulated greater amounts of overexpressed hRpl26FLAG ( Figure 7A , Figure 7—source data 1 ) , consistent with the existence of a UPS pathway that degrades overexpressed ribosomal proteins . The closest human homolog of Tom1 is Huwe1 . Knockdown of Huwe1 by shRNA ( Thompson et al . , 2014 ) in both T-REx-293 and HeLa cells ( Figure 7B , Figure 7—source data 2 ) and knockout of HUWE1 in HEK293T cells ( Choe et al . , 2016 ) ( Figure 7C , Figure 7—source data 3 ) enabled transient overexpression of hRpl26FLAG . Importantly , a cycloheximide chase experiment indicated that hRpl26FLAG overexpressed in HUWE1 knockout cells was stable ( Figure 7D , Figure 7—source data 4 ) . 10 . 7554/eLife . 19105 . 017Figure 7 . ERISQ is conserved in human cells . ( A ) Proteasome inhibition enables overexpression of human Rpl26 . Left: transiently expressed hRpl26FLAG in T-REx-293 cells treated with 10 µM MG132 or 1 µM bortezomib ( btz ) for 3 hr . Right: quantification of blots . Values are the mean of three independent experiments and error bars indicate standard deviations . Asterisks indicate significant differences ( two-tailed student's t-test , ***p<0 . 0001 , compared with DMSO treatment ) . n = 3 biological replicates . Source data are available in Figure 7—source data 1 . ( B ) Depletion of HUWE1 enables overexpression of human Rpl26 . As in ( A ) except that T-REx-293 ( left ) and HeLa ( right ) cells were induced with doxycycline for 3 days to express stably integrated shControl or shHUWE1 . The relative ratio of hRpl26FLAG/GAPDH is shown below each lane . n = 3 biological replicates . Source data are available in Figure 7—source data 2 . ( C ) Knockout of HUWE1 enables overexpression of human Rpl26 . Upper: as in ( A ) except that wild type and HUWE1 knockout HEK293T cells were used . Bottom: quantification of blots . Values are the mean of three independent experiments and error bars indicate standard deviations . Asterisks indicate significant differences ( two-tailed student's t-test , ***p<0 . 0001 , compared with WT cells ) . n = 3 biological replicates . Source data are available in Figure 7—source data 3 . ( D ) Overexpressed human Rpl26 is stable in HUWE1 knockout cells . Upper: wild type and HUWE1 knockout HEK293T cells transiently expressing hRpl26FLAG were treated with 10 μg/ml cycloheximide ( CHX ) . Bottom: quantification of blot . n = 1 biological replicate . Source data are available in Figure 7—source data 4 . ( E ) HUWE1 promotes ubiquitination of overexpressed human Rpl26 . As in ( B ) except that HAubiquitin was co-expressed with Rpl26FLAG and MG132 was added 3 hr prior to cell lysis . Total cell extract prepared under denaturing condition was adsorbed to FLAG resin and the bound fraction was immunoblotted with antibodies against FLAG and HA . n = 2 biological replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 19105 . 01710 . 7554/eLife . 19105 . 018Figure 7—source data 1 . Quantification of hRpl26-FLAG and GAPDH levels from three biological replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 19105 . 01810 . 7554/eLife . 19105 . 019Figure 7—source data 2 . Quantification of hRpl26-FLAG and GAPDH levels from one biological replicate . DOI: http://dx . doi . org/10 . 7554/eLife . 19105 . 01910 . 7554/eLife . 19105 . 020Figure 7—source data 3 . Quantification of hRpl26-FLAG and GAPDH levels from three biological replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 19105 . 02010 . 7554/eLife . 19105 . 021Figure 7—source data 4 . Quantification of hRpl26-FLAG and GAPDH levels from one biological replicate . DOI: http://dx . doi . org/10 . 7554/eLife . 19105 . 021 To test if Huwe1 was required for ubiquitination of transiently overexpressed hRpl26FLAG , control or Huwe1-depleted T-REx cells were co-transfected with plasmids encoding hRpl26FLAG and HAubiquitin and then treated with MG132 to induce accumulation of ubiquitin conjugates . IP/Western blot analysis performed under denaturing conditions revealed that hRpl26FLAG was modified by HAubiquitin in control but not Huwe1-depleted cells ( Figure 7E ) . Consistent with this result , prior analysis of Huwe1-deficient cells by GlyGly profiling revealed reductions in the ubiquitination of multiple ribosomal proteins ( Thompson et al . , 2014 ) .
For ERISQ to work , two major challenges must be met . First , Tom1 has to detect many different ribosomal proteins . Second , it has to be able to distinguish their assembled versus unassembled forms . Although the exact structural basis for this discrimination remains to be determined , our analysis reveals that for Tom1 to act upon ribosomal proteins , residues that are normally concealed in the mature ribosome must be accessible . This was shown in vivo for the model substrate Rpl26 and in vitro for Rpl4 . In addition , mapping of the major Tom1-dependent ubiquitination sites on the structure of the ribosome large subunit revealed that nearly 83% ( 15 of 18 ) of the sites on 12 different ribosomal proteins are no longer available to Tom1 following incorporation into the 80S ribosome . This suggests a very simple kinetic competition between Tom1 and rRNA for binding to ribosomal proteins newly arrived in the nucleolus . If the kinetic parameters for this race normally favor rRNA , the correct outcome would dominate and only those proteins that fail to assemble would be targeted . Tom1 has also been implicated in the degradation of unassembled histones ( Singh et al . , 2009 ) , in which case a similar kinetic competition could apply . Although the domain employed by Tom1 to bind its substrates remains unknown , we note that the previously-described Tom1 substrates Dia2 ( Kim and Koepp , 2012 ) , Hht2 ( Singh et al . , 2009 ) and Yra1 ( Iglesias et al . , 2010 ) have pI values ranging from 9 . 3–12 . Thus , Tom1 , which is overall acidic ( pI 4 . 8 ) may have a negatively-charged region that interacts electrostatically with basic substrates . Seminal studies from Warner indicated that human cells , like yeast , are unable to accumulate ribosomal proteins made in excess over rRNA due to rapid turnover ( Warner , 1977 ) , and subsequent proteomic studies revealed that a substantial fraction of newly-synthesized human ribosomal proteins are rapidly degraded by the proteasome ( Lam et al . , 2007 ) . To test whether the mechanism we described in yeast also operates in human cells , we first demonstrated that the human ortholog ( hRpl26 ) of yeast Rpl26 fails to accumulate upon transient overexpression , and then established that both ubiquitination and degradation of excess hRpl26 require Huwe1 , which is the closest human homolog of Tom1 . A recent proteome-wide profiling of ubiquitination sites that exhibit diminished occupancy upon depletion of Huwe1 revealed that , excluding Huwe1 itself , six of the top ten affected sites and seventeen of fifty-six that exhibited ≥two-fold decreased occupancy were from ribosomal proteins ( Thompson et al . , 2014 ) . These data point to a very general role for Huwe1 in ubiquitination of ribosomal proteins , similar to what we show here for yeast Tom1 . Cells lacking Tom1 exhibit a variety of phenotypes besides ERISQ including cell cycle arrest , nucleolar fragmentation , defective mRNA export from the nucleus , reduced Ngg1/Ada3-dependent transcription , reduced polysomes , reduced rate of rRNA processing/maturation ( Duncan et al . , 2000; Saleh et al . , 1998; Tabb et al . , 2001; Utsugi et al . , 1999 ) , and sensitivity to paromomycin ( Figure 6—figure supplement 1E ) . On the one hand it is possible that all but one of these phenotypes are mainly secondary consequences of the primary defect . On the other hand , they may all arise independently from stabilization of different substrates . We can rule out that ERISQ is an indirect consequence of cell cycle arrest , nucleolar fragmentation , defective mRNA export , or a reduced rate of rRNA processing , because the defect in ERISQ is observed at temperatures that are permissive for growth of tom1 mutants , whereas the other defects are only seen ( or in the case of rRNA processing rate , is strongly enhanced ) at the non-permissive temperature and thus are more likely to arise indirectly from exacerbation of a primary defect upon imposition of heat stress . It remains unclear to what extent a defect in ERISQ , when coupled to heat stress , might underlie these other phenotypes . Some of the tom1 phenotypes noted above are suggestive of a potential positive role for Tom1 in the assembly of functional ribosomes . However , this does not conflict with our observation that Tom1 plays a direct role in ubiquitinating and promoting the degradation of ribosome proteins that are overexpressed relative to their assembly partners . A speculative possibility is that Tom1 ubiquitinates a broad range of ribosomal proteins to promote their assembly ( much as fusion to ubiquitin promotes assembly of Rps31; [Finley et al . , 1989] ) , but if they fail to assemble within a given time interval , the ubiquitin conjugated by Tom1 serves to initiate degradation . However , the observation that genetic reductions of protein kinase A activity ( which are predicted to diminish synthesis of ribosome proteins ) suppress tom1∆ ( Figure 6—figure supplement 1C ) seems inconsistent with the idea that Tom1 plays a direct positive role in ribosome assembly . A phenotype of tom1CA cells that we describe here that is particularly notable is that they are exquisitely sensitive to perturbations that alter the balance between production of individual ribosomal proteins or total ribosomal proteins and rRNAs . Specifically , tom1CA mutants are unable to sustain growth upon overexpression of RPL26 , are sensitive to loss of the Rpl4-specific chaperone Acl4 ( Stelter et al . , 2015 ) , are sensitive to reduction in expression ( through deletion of one of two alleles ) of a single ribosomal protein , and are hypersensitive to reduction of function in three different proteins involved in transcription by RNA polymerase I . Moreover , tom1 mutants accumulate high levels of insoluble ribosomal proteins ( which is consistent with their inability to degrade excess ribosomal proteins that may exist at any given point in time ) , and combining tom1 with a mutation that causes an imbalance in ribosome production ( e . g . tom1CA acl4∆ ) leads to an enormous increase in insoluble protein , suggestive of a collapse in cellular proteostasis . Notably , even the tom1∆ single mutant accumulates more insoluble protein than any other PQC mutant that we examined , including the ERAD mutants doa10∆ and hrd1∆ ( Vembar and Brodsky , 2008 ) , the ribosome QC mutant ltn1∆ ( Bengtson and Joazeiro , 2010; Brandman et al . , 2012; Defenouillere et al . , 2013; Verma et al . , 2013 ) , the nuclear QC mutant san1∆ ( Gardner et al . , 2005 ) , or the cytosolic QC mutant ubr1∆ ( Eisele and Wolf , 2008; Heck et al . , 2010 ) . Taken together with our observations that Tom1 directly binds and ubiquitylates unassembled ribosome proteins , these data point to a direct and critical role for Tom1 in the cellular proteostasis network . The hypersensitivity of tom1CA cells to stoichiometric imbalances in ribosome components raises interesting questions as to why unassembled ribosomal proteins would be toxic , and whether feedback mechanisms exist to monitor and respond to failures in ERISQ . Regarding the first question , given that ribosomal proteins are highly expressed , positively charged nucleic acid binding proteins , their accumulation might interfere with RNA biology . Regarding feedback mechanisms , it is interesting to note that in human cells , reduction of rRNA expression by low-dose actinomycin D treatment results in poor assembly of several ribosomal proteins including Rpl5 ( Dai and Lu , 2004 ) , Rpl11 ( Lohrum et al . , 2003; Zhang et al . , 2003 ) , Rpl26 ( Zhang et al . , 2010 ) , Rps7 ( Chen et al . , 2007 ) and Rps14 ( Zhou et al . , 2013 ) . These unassembled ribosomal proteins bind to and titrate ubiquitin ligase Mdm2 , which leads to stabilization and accumulation of the Mdm2 substrate p53 . This provides a sensitive feedback loop to reduce cell growth in response to stresses that impede ribosome assembly . It will be of great interest to determine whether a similar feedback mechanism operates in yeast , and how Huwe1 activity relates to the Mdm2–p53 feedback pathway described in human cells . Dysfunctional ribosomal proteins and ribosome biogenesis has been linked to many diseases ( Freed et al . , 2010; Narla and Ebert , 2010 ) . Particularly , given that one of the characteristic features of cancer is an increase in the overall rate of protein synthesis , it is clear that regulation of ribosome biogenesis is closely associated with tumor progression . Consistent with this expectation , RNA Pol I activity is highly elevated in many cancer cells and it leads to the enlargement of the nucleolus , which has been used as a marker for cancer for over 100 years ( Derenzini et al . , 2009 ) . Furthermore , two major human tumor suppressor proteins , pRB and p53 , have been shown to repress the production of rRNA and the loss of these factors cause an up-regulation of ribosome biogenesis in cancer tissues , consistent with a close relationship between cancer and ribosome synthesis ( Montanaro et al . , 2008 ) . Up-regulation of ribosome production in cancer cells implies an enhanced dependency on QC mechanisms that survey ribosome assembly . However , the molecular pathway that mediates ERISQ in human cells has , up to now , remained unknown . Interestingly , several observations link Huwe1 function to cancer . Huwe1 is overexpressed in multiple human tumors , is essential for proliferation of a subset of tumors ( Adhikary et al . , 2005 ) , and is required for activation of Myc-inducible target genes including ribosomal proteins in colon carcinoma cells ( Peter et al . , 2014 ) . Our data suggest the intriguing hypothesis that accumulation of unassembled ribosomal proteins would titrate Huwe1 , resulting in reductions in both Myc activity and transcription of genes that encode ribosomal proteins . Insights into the detailed molecular basis underlying ERISQ afforded by the discovery reported here will enable investigations into the biological relevance of ERISQ in human pathologies , including cancer , which may lead to novel concepts for therapy .
All yeast strains used in this study ( listed in Supplementary file 1 ) were derived from BY4741 ( MATa his3∆1 leu2∆0 met15∆0 ura3∆0 ) or W303a ( MATa leu2-3 , 112 trp1-1 can1-100 ura3-1 ade2-1 his3-11 , 15 ) . All transformants were verified by auxotrophic selection or genomic PCR . Yeast was grown at 30°C in YPD or appropriate synthetic complete ( SC ) drop-out media . For ectopic expression of proteins from the GAL1 , 10 promoter , cells grown in raffinose containing medium were treated with 2% galactose for 60–90 min . We note that experiments were initiated with cells at OD600 3 . 0 , because the ERISQ pathway was more prominent in cells at mid-log phase ( OD600 ~3 . 0 ) compared to early-log phase ( OD600 ~1 . 0 ) ( data not shown ) . Yeast transformation was performed by lithium acetate method ( Gietz and Schiestl , 2007 ) . For several strains , PCR products were generated by the ‘gene splicing by overlap extension’ method ( Horton et al . , 1989 ) . T-REx-293 ( Thermo Fisher Scientific , Waltham , MA ) , HeLa ( Thompson et al . , 2014 ) , HEK293 ( Choe et al . , 2016 ) cells were grown in Dulbecco’s modified Eagle’s medium ( DMEM ) supplemented with 10% fetal bovine serum ( Atlanta Biologicals , Flowery Branch , GA ) , penicillin , and streptomycin ( Invitrogen , Carlsbad , CA ) at 37°C in 5% CO2 . Mycoplasma contamination has been tested negative by MycoAlert Mycoplasma detection kit ( Lonza , Switzerland ) . Cells used in this study were not in the database of cross-contaminated or mis-identified curated by the International Cell Line Authentication Committee ( ICLAC ) . We haven’t authenticated cells by a third party . Transient transfections were performed using transfection reagents FuGENE HD ( Promega , Madison , WI ) according to manufacturer’s instructions . For lentiviral production , T-REx-293 cell line ( ThermoFisher ) was transfected with the lentiviral construct along with helper plasmids . Forty eight hours after transfection , media supernatant containing the lentivirus was collected . The lentivirus-containing medium supplemented with polybrene , was used to transduce the target cells . The doxycycline-inducible shRNA expression constructs ( pLKO-Tet-ON vector ( Thompson et al . , 2014; Wiederschain et al . , 2009 ) ) containing the control ( RDB3142; CAA CAA GAT GAA GAG CAC CAA ) and shHUWE1 ( RDB3143; TGC CGC AAT CCA GAC ATA TTC ) sequences were used ( Thompson et al . , 2014 ) . shHUWE1 described previously as si5635 ( Zhong et al . , 2005 ) was used . Transduced T-REx-293 cells transduced with the control or HUWE1 shRNA constructs were selected in the presence of 4 μg/ml of puromycin . All plasmids used in this study are listed in Supplementary file 2 . To construct pESC ( HIS ) -PGAL10-RPL26A ( mutants ) -FLAG used in Figure 4 , site-directed mutagenesis was performed using QuikChange Site-Directed Mutagenesis Kit ( Agilent Technologies; 200519 , Santa Clara , CA ) according to manufacturer’s instructions . To construct N-terminally 3×HA-tagged Tom1 , the ~1800 bp PCR product including the KAN selection marker and RFA1 promoter was obtained using pKanMX6–PRFA1–9Myc–AID* ( Morawska and Ulrich , 2013 ) as a template , forward primer 5′-GAG AGG AAA AGA AGA AAA GGT AAA ACA ACG AAT ATT TTT CCG GAT CCC CGG GTT AAT TAA-3′ and reverse primer 5′-TCT TGT AAG TAT AAT CTG GTC TTC T-3′ , and the ~180 bp PCR product encoding the 3×HA tag was obtained using pRS304-3×HA-TOM1 plasmid ( Duncan et al . , 2000 ) as a template , forward primer 5′- AGA AGA CCA GAT TAT ACT TAC AAG AAT GGA ATT CGG CCG CAT CTT TTA CC-3′ and reverse primer 5′- GTT TCT CCT TTC TTG CCT TTT CAC ACC GAG TAA AAA GCA CAG ATC TGC ACT GAG CAG CGT-3′ . With the two PCR products as templates , the ~2000 bp PCR product was obtained using forward primer 5′- GAG AGG AAA AGA AGA AAA GGT AAA ACA ACG AAT ATT TTT CCG GAT CCC CGG GTT AAT TAA-3′ and reverse primer 5′- GTT TCT CCT TTC TTG CCT TTT CAC ACC GAG TAA AAA GCA CAG ATC TGC ACT GAG CAG CGT-3′ . The obtained final PCR product was used for transformation , generating cells expressing 3×HATom1 from the RFA1 promoter . To construct tom1CA mutants , the ~145 bp PCR product including the 3’ end region of TOM1 with a C3235A mutation was obtained using pRS304-3×HA-TOM1CA plasmid ( Duncan et al . , 2000 ) as a template , forward primer 5′-TGA TTT TGG TTC ATC AGA AAG ACT ACC ATC ATC ACA TAC C-3′ and reverse primer 5′-CAA AAG CAG AGA GGC GCG CCT CAG GCA AGA CCA AAC CCT TCA TGC-3′ , and the ~1700 bp PCR product including the KlURA3 was obtained using pFA6a-GFP-KlURA3 plasmid ( Sung et al . , 2008 ) as a template , forward primer 5′-GCA TGA AGG GTT TGG TCT TGC CTG AGG CGC GCC TCT CTG CTT TTG-3′ and reverse primer 5′-CAT GGC GCT ATA AAT TTA CAC GAA AAA TGA CGT CAT TGG TTC TGG AGG AAG TTT GAG-3′ . With the two PCR products as templates , the ~1850 bp PCR product was obtained using forward primer 5′-TGA TTT TGG TTC ATC AGA AAG ACT ACC ATC ATC ACA TAC C-3′ and reverse primer 5′-CAT GGC GCT ATA AAT TTA CAC GAA AAA TGA CGT CAT TGG TTC TGG AGG AAG TTT GAG-3′ . The obtained final PCR product was used for transformation , generating tom1CA strains . Anti-Rpl1 and anti-Rpl3 were generous gifts from Jonathan Warner . Anti-FLAG ( F1804; RRID:AB_262044; 1:10 , 000 dilution ) was from Sigma ( St . Louis , MO ) , anti-Hexokinase ( H2035-02; 1:10 , 000 dilution ) was from USBiological ( Salem , MA ) , anti-HUWE1 ( A300-486A; RRID: AB_2615536; 1:1 , 000 dilution ) was from Bethyl laboratories ( Montgomery , TX ) , anti-HA ( SC-7392; RRID:AB_627809; 1:5 , 000 dilution ) was from Santa Cruz ( Dallas , TX ) , anti-myc ( MMS-150R; RRID: AB_291325; 1:5 , 000 dilution ) was from Covance ( San Diego , CA ) , anti-Ubiquitin ( 05–944; RRID: AB_441944; 1:5 , 000 dilution ) and anti-GAPDH ( MAB374; RRID: AB_2107445; 1:5 , 000 dilution ) were from EMD Millipore , and anti-His6 ( 200-332-382; RRID: AB_10704645; 1:5 , 000 dilution ) was from Rockland ( Limerick , PA ) . For secondary antibody , HRP-conjugated anti-rabbit IgG ( A6154; RRID: AB_258284; 1:10 , 000 dilution ) and HRP-conjugated anti-mouse IgG ( M8770; RRID: AB_260711; 1:10 , 000 dilution ) were from Sigma , IR680RD conjugated anti-rabbit ( 926–68071; RRID: AB_10956166; 1:10 , 000 dilution ) and IR800CW conjugated anti-mouse ( 926–32210; RRID: AB_621842; 1:10 , 000 dilution ) were from LI-COR Biosciences ( Lincoln , NE ) . Yeast cells grown in raffinose-containing SC medium at 30°C ( OD600 ≤ 1 . 0 ) were induced with galactose for 1 hr to express Rpl26aGFP and placed in 384-well glass-bottom microplates ( Whatman , UK ) pretreated with concanavalin A ( Sigma; L7647 ) to ensure cell adhesion . Fluorescence images were taken using a Zeiss Axiovert 200M Inverted Microscope with an FITC filter set ( excitation band pass filter , 450–490 nm; beam splitter , 510 nm; emission band pass filter , 515–565 nm ) and a Rhodamine filter set ( excitation band pass filter , 546 nm; beam splitter , 580 nm; emission long pass filter , 590 nm ) . We analyzed at least 50 cells and subcellular localization of GFP-fused proteins was reconfirmed by co-localization assay as described previously ( Huh et al . , 2003 ) . For denatured samples , yeast or mammalian cells were harvested , washed two times with PBS , and boiled in 2× SDS-containing sample buffer for 5 min followed by brief bead beating or sonication , respectively . For non-denatured samples , yeast cells were harvested and disrupted by bead beating in lysis buffer ( 50 mM Tris-HCl pH 7 . 5 , 150 mM NaCl , 1% Triton X-100 andprotease inhibitor cocktail ( EDTA-free; Roche , Switzerland ) ) . Cell debris was removed by centrifuging at 3000 rpm for 5 min , and the remaining cell extract was centrifuged at 12 , 000 rpm for 10 min in an Eppendorf Centrifuge 5430R . For mammalian cells , harvested cells were washed twice with PBS , and then incubated with RIPA buffer ( 50 mM Tris-HCl pH 7 . 5 , 150 mM NaCl , 1% IGEPAL , 0 . 5% sodium deoxycholate , 0 . 1% SDS and protease inhibitor cocktail ( EDTA-free; Roche ) ) for 10 min . After centrifugation at 12 , 000 rpm for 10 min in an Eppendorf Centrifuge 5430R , the supernatant was transferred to a new tube and mixed with SDS-PAGE sample buffer . Hexokinase and GAPDH were used as an internal control . Myc-GFP was used as a control for transfection efficiency . Protein levels were quantified using Odyssey software . Acl4-FLAGRpl4 , Acl4-FLAGRpl4∆ext and Chaetomium thermophilum Kap104 were expressed in E . coli BL21-CodonPlus ( DE3 ) -RIL cells ( Stratagene , San Diego , CA ) grown in LB media supplemented with appropriate antibiotics . Protein expression was induced at an OD600 of approximately 0 . 6 with 0 . 5 mM isopropyl β-D-thiogalactoside ( IPTG ) for ~18 hr at 18°C ( Acl4-FLAGRpl4 , Acl4-FLAGRpl4∆ext ) or 23°C ( ctKap104 ) . Cells were harvested by centrifugation and resuspended in a buffer containing 20 mM Tris-Base pH 8 . 0 , 500 mM NaCl , 5 mM β-mercaptoethanol ( Sigma ) , 2 μM bovine lung aprotinin ( Sigma ) , and complete EDTA-free protease inhibitor cocktail ( Roche ) , and subsequently flash frozen in liquid nitrogen . Thawed cells were lysed with a cell disrupter ( Avestin , Germany ) and the lysate was centrifuged for 1 hr at 40 , 000 ×g . Cleared lysate of ctKap104 expression was applied to a glutathione sepharose column equilibrated in buffer containing 20 mM Tris-Base pH 8 . 0 , 100 mM NaCl , and 5 mM DTT ( GE Healthcare , Pasadena , CA ) and eluted via a glutathione gradient . Pooled fractions were cleaved with PreScission protease ( GE Healthcare ) for 12 hr . Cleared lysate of Acl4-FLAGRpl4 , Acl4-FLAGRpl4∆ext expression was applied to a Ni-NTA column equilibrated in buffer containing 20 mM Tris-Base pH 8 . 0 , 500 mM NaCl , and 5 mM β-mercaptoethanol and eluted via an imidazole gradient . Pooled fractions were cleaved with ubiquitin-like-specific protease 1 ( Ulp1 ) for 12 hr . Cleaved proteins were bound to a HiTrapQ HP ( GE Healthcare ) column equilibrated in buffer containing 20 mM Tris-Base pH 8 . 0 , 100 mM NaCl , and 5 mM DTT and eluted via a linear NaCl gradient , concentrated , and injected onto a HiLoad 16/60 Superdex 200 column equilibrated in 20 mM Tris-Base pH 8 . 0 , 100 mM NaCl , and 5 mM DTT . The Acl4-FLAGRpl4-ctKap104 complex was assembled by stoichiometric incubation for 1 hr at 4°C and injection onto a HiLoad 16/60 Superdex 200 column equilibrated in 20 mM Tris-Base pH 8 . 0 , 100 mM NaCl , and 5 mM DTT . Five hundred OD600 units of cells expressing 3×HATom1 ( WT ) and 3×HATom1CA were grown in raffinose medium and then induced to express Rpl26aFLAG in galactose medium for 45 min followed by bortezomib ( 50 µM ) treatment for an additional 45 min . Cells were harvested and disrupted by bead beating in 2 . 5 ml lysis buffer ( 50 mM Tris-HCl pH 7 . 5 , 150 mM NaCl , 0 . 5% NP-40 andprotease inhibitor cocktail ( EDTA-free; Roche ) ) . Cell debris was removed by centrifuging at 3 , 000 rpm for 5 min , and the remaining cell extract was centrifuged at 12 , 000 rpm for 10 min in an Eppendorf Centrifuge 5430R . Total lysates were incubated overnight with 150 µl anti-HA magnetic beads . Beads were washed three times with the same lysis buffer and then twice with 1× ubiquitin reaction buffer ( 50 mM Tris-HCl pH 8 . 0 , 10 mM MgCl2 , 0 . 2 mM CaCl2 , 1 mM DTT and 5 µM MG132 ) . Beads were incubated with 2 mM ATP , 166 nM E1 ( Ube1; BostonBiochem; E-305 ) , 1 µM E2 ( UbcH5a; BostonBiochem; E2-616 ) and 20 µg of His6-ubiquitin ( Ubiquitin; BostonBiochem , Cambridge , MA; U-530 ) at 30°C for 1 hr . For Rpl4 ubiquitination , 10 µg of purified substrates ( Acl4-FLAGRpl4 , Acl4-FLAGRpl4∆ext and Acl4-FLAGRpl4-ctKap104 ) were used . One-fifth volume of 5× SDS-containing sample buffer was added to stop the reactions and boiled for 5 min . Immunoprecipitation of ubiquitin conjugates was performed as described with some modifications ( Verma et al . , 2013 ) . TUBE2-UBA resin ( BostonBiochem; AM-130 ) was used to bind polyubiquitinated substrates . One hundred OD600 units of cells were harvested and disrupted by bead beating in 500 µl lysis buffer ( 20 mM Tris-HCl pH 7 . 5 , 150 mM NaCl , 1 mM EDTA , 10% Glycerol , 5 mM NEM , 1% Triton X-100 andprotease inhibitor cocktail ( EDTA-free; Roche ) ) . Cell debris was removed by centrifuging at 3000 rpm for 5 min , and the remaining cell extract was centrifuged at 12 , 000 rpm for 10 min in an Eppendorf Centrifuge 5430R . TUBE2-UBA resin ( 30 µl ) incubated overnight was washed three times with buffer ( 20 mM Tris-HCl pH 8 . 0 , 150 mM NaCl , 0 . 5% Triton X-100 ) . Fifty µl of 2× SDS-containing sample buffer was added to the resin and boiled for 5 min . Note that unmodified proteins can also potentially bind due to indirect interactions with ubiquitinated proteins . For Usp2 treatment , TUBE2-UBA resin prepared as described above was washed twice with 1× ubiquitin reaction buffer ( 50 mM Tris-HCl pH 8 . 0 , 10 mM MgCl2 , 0 . 2 mM CaCl2 and 1 mM DTT ) and mixed with 1 µM Usp2 ( BostonBiochem; E-504 ) at 30°C for 1 hr . Subcellular fractionation was performed as described ( Keogh et al . , 2006 ) . One hundred OD600 units of cells grown in rich medium ( OD600≤1 . 0 ) were collected by centrifugation and then treated with 200 units Zymolyase for 1 hr at 30°C in 1 ml SB buffer ( 1 M Sorbitol , 20 mM Tris-HCl pH 7 . 5 , 10 mM β-mercaptoethanol ) . Spheroplasts collected by centrifugation ( 2000 rpm for 5 min at 4°C ) were washed twice with SB buffer , and then resuspended in 500 µl EBX buffer ( 20 mM Tris-HCl pH 7 . 5 , 100 mM NaCl , 0 . 25% Triton X-100 , 15 mM β-mercaptoethanol and protease inhibitor cocktail ( EDTA-free; Roche ) ) . An aliquot was taken and used as a total cell extract , and the remainder of the lysate was layered over 1 ml NIB buffer ( 20 mM Tris-HCl pH 7 . 5 , 100 mM NaCl , 1 . 2 M Sucrose , 15 mM β-mercaptoethanol and protease inhibitor cocktail ( EDTA-free; Roche ) ) and centrifuged ( 12 , 000 rpm for 15 min at 4°C in an Eppendorf Centrifuge 5430R ) . A sample of the upper soluble fraction was taken and used as cytosol and the rest of the supernatant discarded . The glassy white nuclear pellet was suspended in 500 µl EBX buffer and kept on ice for 10 min with gentle mixing and an aliquot taken and used as the nuclear fraction . 2× SDS-PAGE loading buffer was added to each fraction and samples were incubated at 95°C for 5 min and then subjected to SDS-PAGE and Western analyses . Sucrose gradient and polysome profiling were performed as described ( Verma et al . , 2013 ) . Yeast cells were grown to logarithmic phase in rich medium supplemented with glucose or raffinose at 30°C , and treated with cycloheximide ( 100 µg/ml ) for 15 min before cell lysis to stabilize polysomes . One hundred OD600 units of cells were harvested and disrupted by bead beating in lysis buffer ( 0 . 5 mM DTT , 100 µg/ml cycloheximide , 200 µg/ml heparin , 20 mM Tris-HCl pH 7 . 5 , 140 mM KCl , 5 mM MgCl2 andprotease inhibitor cocktail ( EDTA-free; Roche ) ) . Cell debris was removed by centrifuging at 3000 rpm for 5 min , and the remaining cell extract was centrifuged at 12 , 000 rpm for 10 min in an Eppendorf Centrifuge 5430R . Twenty five A260 units of cell lysate layered on 7%~47% discontinuous sucrose gradient prepared in buffer ( 1 mM DTT , 140 mM KCl , 20 mM Tris-HCl pH 7 . 5 and 5 mM MgCl2 ) were centrifuged in SW55Ti rotor for 90 min at 50 , 000 rpm . For polysome profiling analysis , samples were fractionated while continuously recording the absorbance at 254 nm with a UV detector ( ISCO , Lincoln , NE ) . For Western blot , 0 . 2 ml fractions collected from the top were treated with 0 . 02% sodium deoxycholate for 30 min on ice and then precipitated by adding TCA to 10% final concentration for 1 hr . Pellets were washed with ice-cold acetone , and then resuspended in 2× SDS-containing sample buffer . Isolation of protein aggregates from yeast cells was performed as described previously ( Koplin et al . , 2010 ) with slight modifications . One hundred OD600 units of exponentially growing cells were harvested , and cell pellets were frozen in liquid N2 . The cell pellets were resuspended in 1 ml lysis buffer ( 20 mM Na-phosphate pH 6 . 8 , 10 mM DTT , 1 mM EDTA , 0 . 1% Tween , 1 mM PMSF , protease inhibitor cocktail and 100 units/ml zymolyase ) and incubated at 30° C for 30 min . Chilled samples were treated by tip sonication ( 20% , 10 sec , twice ) and centrifuged for 20 min at 600 g at 4°C . Supernatants were adjusted to identical protein concentrations , and aggregated proteins were pelleted at 16 , 000 g for 20 min at 4°C . After removing supernatants , insoluble proteins were washed once with Wash I buffer ( 20 mM Na-phosphate pH 6 . 8 , 500 mM NaCl , 5 mM EDTA , 2% NP-40 , 1 mM PMSF , and protease inhibitor cocktail ) , and centrifuged at 16 , 000 g for 20 min at 4°C . Insoluble proteins were washed twice with Wash II buffer ( 20 mM Na-phosphate pH 6 . 8 ) and sonicated ( 10% , 10 s , twice ) in 40 μl of Wash II buffer . Pellets were processed either as described below or solubilized in 10 µl SDS sample buffer . 1X of the total cell lysate ( T ) and soluble ( S ) fractions , and 20X of the isolated pellet ( P ) fraction were separated by SDS-PAGE , and analyzed by Coomassie Blue staining and immunoblotting . For the experiment in Figure 3—figure supplement 3A , the cells were lysed by agitation with glass beads as described by Kaganovich et al . , ( 2008 ) , in the presence of 3 different lysis buffers: 1 ) 100 mM Tris-Cl , 1% Triton X-100 , 150 mM KCl , 5 mM MgCl2 and protease inhibitor; 2 ) 100 mM HEPES , 1% Triton X-100 , 300 mM NaCl and protease inhibitor ( Lu et al . , 2014 ) ; 3 ) 30 mM HEPES , 0 . 5% Triton X-100 , 150 mM NaCl , 1% glycerol , 1 mM DTT and protease inhibitor ( Kaganovich et al . , 2008 ) . Mass spectrometry analyses of protein aggregates were performed as described . Insoluble protein pellets were solubilized in an 8 M Urea buffer ( 40 mM Tris , 65 mM DTT , 100 mM Ammonium biocarbonate ) containing cOmplete Protease Inhibitor Cocktail ( Roche ) and sonicated for 10 s at 10% of maximum amplitude using a Branson Digital Sonifier . Samples were digested and prepared for mass spectrometry as described in ( Pierce et al . , 2013 ) . One hundred fifty ng of digested peptides from tom1Δ cells and equal volume of peptides from WT cells were analyzed using an EASY-nLC 1000 coupled to an Orbitrap Fusion . Spectra were analyzed by MaxQuant ( v 1 . 5 . 3 . 30 ) . Digested peptides were loaded onto a 26-cm analytical HPLC column ( 75 μm ID ) packed in-house with ReproSil-Pur C18AQ 1 . 9 μm resin ( 120 Å pore size , Dr . Maisch , Ammerbuch , Germany ) . After loading , the peptides were separated with a 120 min gradient at a flow rate of 350 nl/min at 50°C ( column heater ) using the following gradient: 2–6% solvent B ( 7 . 5 min ) , 6–25% B ( 82 . 5 min ) , 25–40% B ( 30min ) , 40–100% B ( 1 min ) , and 100% B ( 9 min ) where solvent A was 97 . 8% H2O , 2% ACN , and 0 . 2% formic acid and solvent B was 19 . 8% H2O , 80% ACN , and 0 . 2% formic acid . The Orbitrap Fusion was operated in data-dependent acquisition ( DDA ) mode to automatically switch between a full scan ( m/z=350–1500 ) in the Orbitrap at 120 , 000 resolving power and a tandem mass spectrometry scan of Higher energy Collisional Dissociation ( HCD ) fragmentation detected in the ion trap ( using TopSpeed ) . AGC target of the Orbitrap and ion trap was 400 , 000 and 10 , 000 respectively . | Ribosomes are the molecular machines in cells that produce proteins . The ribosomes themselves are composed of almost 80 different proteins that are held together by scaffolds made from molecules of RNA . Each protein is present in one copy , and so equal numbers of all proteins are needed to assemble a ribosome . However , because it takes many steps to produce a protein and biological processes are inherently imprecise , it is essentially impossible for a cell to produce exactly the same number of copies of all the proteins in a ribosome . Much research suggests that , to overcome these issues , a cell will make more of certain ribosomal proteins than it needs , and then degrade the leftovers that are not used . However , it was not clear how this happens , nor was it known what are the consequences of failing to degrade the leftovers . Now , Sung et al . show that yeast cells use an enzyme named Tom1 to attach a protein-marker called ubiquitin to ribosomal proteins that are made in excess and not assembled into ribosomes . The ubiquitin serves as a tag that marks proteins for degradation , and yeast cell that lack Tom1 fail to degrade any excess ribosomal proteins . Consequently , the mutant yeast become sensitive to any factors that alter the balance of the protein and RNA building blocks used to assemble ribosomes . The human equivalent of Tom1 is known as Huwe1 , and the data of Sung et al . suggest that this enzyme acts in a similar pathway . Further experiments are now needed to explore the role of Huwe1 in greater depth , and investigate if problems with this enzyme are associated with any human diseases . Finally , working out the exactly how Tom1 recognizes unassembled ribosomal proteins will be another important challenge for future studies . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"cell",
"biology"
] | 2016 | A conserved quality-control pathway that mediates degradation of unassembled ribosomal proteins |
How organ-shaping mechanical imbalances are generated is a central question of morphogenesis , with existing paradigms focusing on asymmetric force generation within cells . We show here that organs can be sculpted instead by patterning anisotropic resistance within their extracellular matrix ( ECM ) . Using direct biophysical measurements of elongating Drosophila egg chambers , we document robust mechanical anisotropy in the ECM-based basement membrane ( BM ) but not in the underlying epithelium . Atomic force microscopy ( AFM ) on wild-type BM in vivo reveals an anterior–posterior ( A–P ) symmetric stiffness gradient , which fails to develop in elongation-defective mutants . Genetic manipulation shows that the BM is instructive for tissue elongation and the determinant is relative rather than absolute stiffness , creating differential resistance to isotropic tissue expansion . The stiffness gradient requires morphogen-like signaling to regulate BM incorporation , as well as planar-polarized organization to homogenize it circumferentially . Our results demonstrate how fine mechanical patterning in the ECM can guide cells to shape an organ .
Animal organs have a bewildering variety of distinctive forms that are critical for their functions . Although originating in a genetic program , morphogenesis of organs ultimately depends on physical forces , and specifically on their imbalances , to drive shape change ( Thompson , 1917 ) . A central question of morphogenesis is how such force imbalances are created by mechanical anisotropy that is generated within an organ’s components . Current paradigms derive from archetypes of morphogenetic processes such as tissue elongation , and elegant studies have revealed conserved mechanisms that drive elongation across many species . In the Drosophila embryo , planar cell polarized ( PCP ) myosin contractility at the cell cortex generates junctional rearrangements that extend the germband , whereas in vertebrate embryos , PCP actin-based protrusions drive cell movements that extend the neural plate ( Guillot and Lecuit , 2013; Heisenberg and Bellaïche , 2013; Vichas and Zallen , 2011; Walck-Shannon and Hardin , 2014 ) . In these textbook examples of morphogenesis , as in others such as gastrulation and epiboly , the force anisotropies that instruct shape are generated within the tissue’s cells . In theory , asymmetric organs could be generated not only by spatially varying forces produced within cells , but also by spatially varying tissue properties that differentially resist uniformly applied forces . In epithelial organs , morphogenetic forces include not only tension between cells that can cause intercellular rearrangements , but also expansion of luminal contents normal to the epithelial plane; resistance to these forces is mediated by cells and by the extracellular matrix ( ECM ) , including the basement membranes ( BMs ) that line all epithelia . In comparison to the action of cellular forces , the role of non-cellular influences on morphogenesis is poorly understood . A comprehensive study of morphogenetic mechanics requires a tissue that is subject to both cellular and extracellular influences . The Drosophila egg chamber ( or ‘follicle’ ) is such a tissue ( Figure 1A and Figure 1—figure supplement 1 ) and undergoes robust elongation during its development ( Spradling , 1993 ) . Each follicle is a simple tube-like organ consisting of just two cell types , with a somatic epithelium of ‘follicle cells’ ( FCs ) encasing an interconnected cyst of germ cells . The epithelium also produces an underlying BM that surrounds the entire follicle . The organ is initially spherical and grows throughout oogenesis , expanding ~5000 fold in volume over ~3 days . Expansion for the first 35 hr is isotropic , but subsequently becomes anisotropic as the follicle elongates >2-fold specifically along the anterior–posterior ( A–P ) axis to form the distinctively shaped oval egg ( Figure 1A ) . Much of this elongation takes place without cell division . Genes and cell behaviors that are required for egg elongation have been identified , but the mechanical environment that actually shapes the tissue is not known ( Bilder and Haigo , 2012; Cetera and Horne-Badovinac , 2015 ) . 10 . 7554/eLife . 24958 . 003Figure 1 . A mechanical stiffness gradient in the follicle basement membrane . ( A ) Elongation of the Drosophila follicle during oogenesis involves three components: the luminal germline , a surrounding epithelium , and an encasing basement membrane ( BM ) ( see also Figure 1—figure supplement 1 ) . Aspect ratios of stage 3 , 5 , and 7 egg chambers stained for DAPI ( blue ) and phalloidin ( red ) , along with ColIV–GFP ( green ) , are shown . ( B ) Atomic Force Microscopy ( AFM ) measurement of BM stiffness in living follicles . Absence of stroma and external position of BM allow direct access of the AFM probe . ( C ) Follicles are probed at different regions along the A–P axis , including the poles via Polydimethylsiloxane ( PDMS ) ‘egg cartons’ . Stiffness measurements are derived from the first 50 nm of force–extension curves . ( D ) BM stiffness in the follicle center increases during development . Collagen digestion but not F-actin network disruption eliminates nearly all AFM-measured stiffness . ( cf Figure 1—figure supplement 1 ) . ( E ) Regional BM stiffness along the follicle A–P axis; color intensity matches position as in ( C ) . WT follicles develop an A–P symmetrical gradient of mechanical anisotropy . Anterior and posterior poles are not distinguished . ( F ) fat2- and msn-depleted follicle BMs do not increase stiffness during development and remain mechanically isotropic . Scale bar: 25 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 24958 . 00310 . 7554/eLife . 24958 . 004Figure 1—figure supplement 1 . Isotropic mechanical properties of cells in the Drosophila ovary . ( A , B ) Cross-section of the acinus-like Drosophila follicle; planar epithelial and luminal expansionary forces , as well as basement membrane-based resistance , are diagrammed . ( C–E ) As compared to the elongation of growing WT follicles ( C ) , growth of follicles following ablation of epithelium ( tjGAL4 GAL80ts>Diptheria toxin A chain [D] ) is isotropic . Note the absence of epithelium-produced BM ( inset ) . The aspect ratio is quantitated in ( E ) . ( F ) Assessment of cortical tension in follicle epithelial cells using laser nanodissection ( red ) and Myo:GFP localization ( blue ) . ( G ) Severing of A–P and circumferential cell junctions at anterior , center , and posterior positions results in comparable recoil velocities . Example A–P and circumferential cuts in G’ and G’’ are shown . ( H ) Junctional non-muscle MyoII ( Myo:GFP ) localization is equivalent along A–P and circumferential cell junctions . Representative example in H’ . Scale bars: 25 µm in ( A–C ) , 5 µm inset in ( G’ , H’ ) . DOI: http://dx . doi . org/10 . 7554/eLife . 24958 . 00410 . 7554/eLife . 24958 . 005Figure 1—figure supplement 2 . AFM elasticity measurement method . ( A ) Follicles are indented to generate extension–deflection curves . Only the first 50 nm of deflection ( plus 50 nm pre-contact ) are used to fit for the Young’s modulus . Four curves are generated per position , averaged , then compared between A–P positions . ( B ) Indentation piezo ( extension ) speed optimization . Reduced stiffness at high speeds ( 0 . 8–1 . 0 µm/s ) are indicative of viscoelasticity . 0 . 4 µm/s provided optimal elasticity accuracy and measurement speed . DOI: http://dx . doi . org/10 . 7554/eLife . 24958 . 00510 . 7554/eLife . 24958 . 006Figure 1—figure supplement 3 . Validation of pharmacological and hypertonic shock treatments for BM stiffness . ( A ) Collagenase treatment of follicles prior to AFM does not disrupt cell–cell junctions , as monitored by E-cadherin–GFP . ( B ) Latrunculin A treatment of follicles prior to AFM effectively displaces Myo:GFP . ( C ) Hypertonic shrinkage of WT stage 7–8 follicles causes a significant reduction in size , but no significant loss of BM stiffness . Scale bar: 20 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 24958 . 006 Here we use biophysical tools to measure the mechanical conditions present in elongating follicles . Surprisingly , we find no evidence for differential cell-intrinsic forces within the organ , but instead document a robust spatial gradient in stiffness within the BM . Direct BM manipulation indicates that this mechanical gradient is instructive for tissue elongation . Fine mechanical patterning within the BM , generated by independent mechanisms along both the A–P and circumferential axes , endows the BM with anisotropic resistance to tissue expansion that deforms the growing tissue . These results highlight a new parameter of developmental mechanics by uncovering an unappreciated sophistication in BM mechanical properties that can directly impose organ shape .
To understand the conditions that drive elongation of the Drosophila follicle , we first searched for mechanical anisotropy in the organ’s two distinct cell populations . In these assays as well as others below , we examined follicles at stage 8 and earlier , when they display a regular and A–P symmetric morphology . Previous genetic mosaic experiments with several ‘round egg’ mutations exclude the germline as a site of action ( Frydman and Spradling , 2001; Wieschaus et al . , 1981; Viktorinová et al . , 2009 ) , while stripping of epithelium in Heteropeza results in round rather than elongated follicles ( Went and Junquera , 1981 ) . Similarly , we genetically ablated the Drosophila follicle epithelium ( as well as its underlying BM ) , and found that germline growth resulted in a nearly spherical follicle at stages when elongation would normally have initiated ( Figure 1—figure supplement 1 ) . Together , these data suggest that the germline is not an intrinsic source of mechanical anisotropy . To assess whether the follicle epithelium showed PCP cortical contractility , we laser-ablated cellular junctions at different positions along the A–P axis and measured the recoil . In elongating epithelia including the Drosophila ectoderm and wing , this technique reveals differential tension along A–P and dorsal–ventral ( D–V ) axes , an anisotropy associated with polarized Myosin II accumulation ( Bosveld et al . , 2012; Etournay et al . , 2015; Fernandez-Gonzalez et al . , 2009; Rauzi et al . , 2008 ) . However , in the elongating follicle epithelium , dissection of junctions resulted in equivalent retraction of A–P and circumferentially oriented junctions; polarized accumulation of Myo:GFP was not observed ( Figure 1—figure supplement 1 ) . These results suggest that neither follicle cell type intrinsically generates anisotropic physical forces . To identify the source of mechanical anisotropy , we therefore turned to a non-cellular component of the organ: the ECM , specifically the BM . The Drosophila follicle is enclosed by a BM that , like classic vertebrate BMs , is ~150 nm thick and contains Collagen IV , laminin , and perlecan ( Haigo and Bilder , 2011; Isabella and Horne-Badovinac , 2015; Spradling , 1993 ) . BMs and surrounding ECM are known to have important influences on animal organogenesis ( Daley and Yamada , 2013; Morrissey and Sherwood , 2015 ) , but discovery of their mechanical roles has been impeded by the difficulty of measuring these directly in vivo . In the Drosophila follicle , the external position of the BM , the absence of a cellular stroma ( Figure 1A and Figure 1—figure supplement 1 ) and the ability to develop in culture provided an unprecedented opportunity to assess the mechanical properties of an intact BM , in living tissue under physiological conditions . We utilized Atomic Force Microscopy ( AFM ) to measure BM stiffness , calculating the Young’s modulus from the deflection of a cantilevered probe indenting into the basal follicle surface ( Figure 1B , C , Figure 1—figure supplement 2 ) . Treatment of follicles with purified collagenase decreased stiffness by 97% without detectable changes to epithelial junctions , whereas disruption of the cellular actomyosin network with Latrunculin A induced no significant change in the AFM measurements . Furthermore , reducing the turgor pressure of the follicle with a hypertonic solution ( 2000 mOsm sorbitol media ) does not have an effect on the BM stiffness ( Figure 1D and Figure 1—figure supplement 3 ) . These controls indicate that the quantified stiffness predominantly derives from the BM . AFM measurements at the center of staged wild-type ( WT ) follicles showed that the BM gradually stiffens as oogenesis proceeds , increasing from ~30 KPa at stage 3 to ~40 KPa at stage 5 and ~70 KPa at stage 7 ( Figure 1D ) . Interestingly , although stiffness was highly consistent ( >5% variance ) around the circumferential axis at a given position Figure 4F , it significantly varied along the A–P axis ( Figure 1E ) . At stages 3 and 5 , poles were ~50% softer than the central or terminal regions ( see Figure 1D for definitions ) . This difference persisted into later stages , and the central regions further became ~30% stiffer than the terminal regions . Thus , AFM analysis reveals a symmetrical gradient of BM stiffness along the A–P axis of the follicle . If the BM stiffness gradient is functionally important for organ elongation , it should be perturbed in conditions where elongation fails . We analyzed two distinct genotypes in which follicle elongation is defective: mutants for fat2 , which encodes an atypical cadherin that controls basal PCP organization in the follicle epithelium ( Viktorinová et al . , 2009 ) , and RNAi-depleting mutants for misshapen ( msn ) , which encodes a kinase that negatively regulates integrin-mediated adhesion ( Lewellyn et al . , 2013 ) . We carried out AFM on staged fat2 follicles and found that , unlike WT follicles , BM stiffness did not increase from stage 5 to stage 7 ( Figure 1F ) . Strikingly , fat2 follicles showed no significant differences between the central , terminal , and polar regions at any stage . An isotropic and softer BM was also seen in msn-depleted follicles , despite their elevated integrin levels ( Lewellyn et al . , 2013 ) ( Figure 1F ) . The lack of a BM stiffness gradient in non-elongating follicles is consistent with an important role for this mechanical property in organ elongation . The data described above suggest the hypothesis that BM stiffness is in fact the anisotropic mechanical property that drives organ shape , deforming the growing tissue . An alternative hypothesis is that BM stiffness is instead an indirect consequence of organ shape , passively reflecting undetected changes in cell-intrinsic properties . To distinguish between these possibilities , we directly manipulated BM components . We then measured effects on BM mechanics and subsequent tissue elongation , including manipulations in which the A–P stiffness gradient was either eliminated or preserved . The follicle epithelium produces most of its own BM , which can be altered by follicle-wide RNAi or by overexpression driven by tj-Gal4 ( Figure 2I ) ( Haigo and Bilder , 2011; Isabella and Horne-Badovinac , 2015; Van De Bor et al . , 2015 ) . AFM measurements on follicles depleted for SPARC , a factor involved in early BM incorporation of Collagen IV , showed that BM stiffness was ~80% of WT levels in the central regions , but a gradient with increased elasticity was preserved at both terminal regions and poles; elongation of these follicles was indistinguishable from that in WT follicles ( Figure 2A , B ) ( Isabella and Horne-Badovinac , 2015; Martinek et al . , 2008; Pastor-Pareja and Xu , 2011 ) . These follicles are distinct from those that are uniformly depleted of Collagen IV , which are homogenously soft and defective in elongation , resembling fat2- and msn-depleted follicles ( Figure 2C–E ) ( Haigo and Bilder , 2011; Isabella and Horne-Badovinac , 2015 ) . By contrast , uniform overexpression of EHBP1 , which elevates Collagen IV fibril deposition , leads to ~15% increased central stiffness with a ~20% increased anisotropic gradient , and results in organ hyperelongation ( Figure 2F ) ( Isabella and Horne-Badovinac , 2016 ) . 10 . 7554/eLife . 24958 . 007Figure 2 . Manipulating the BM stiffness gradient alters organ shape . For each follicle genotype , AFM-measured positional stiffness at stages 7–8 is shown above and degree of elongation is shown below . Manipulations in ( A–F ) alter gene expression uniformly via tjGAL4 ( I ) or homozygous genotype , whereas those in ( G , H ) alter gene expression regionally using centrally expressed mirGAL4 or terminally expressed fruGAL4 ( J , K ) . Compared to WT ( A ) , depletion of SPARC ( B ) softens the BM but preserves the anisotropic gradient; follicles elongate comparably to WT . Depletion of Collagen IV ( ColIV ) throughout the epithelium ( C ) creates a uniformly soft follicle with severe elongation defects , resembling mutants in which msn is depleted ( D ) or fat2 mutants ( E ) . EHBP1 overexpression ( F ) increases stiffness while retaining an anisotropic gradient , and follicles hyperelongate . Depletion of Col IV in the central region alone ( G ) flattens the gradient while leaving terminal stiffness intact; this results in elongation defects . EHBP1 overexpression in the terminal regions alone ( H ) also flattens the gradient and results in elongation defects . ( L ) Aspect ratio vs stiffness anisotropy ( defined as the ratio of central stiffness to the mean stiffness throughout the A–P axis ) for genotypes ( A–H ) and for tj>DomeDN and tj>PerlOE . DOI: http://dx . doi . org/10 . 7554/eLife . 24958 . 007 We then turned to spatially restricted GAL4 drivers that allow manipulation of BM components in subsets of the gradient . We depleted Collagen IV specifically in the central FCs ( using mirr-GAL4 , Figure 2J ) , where BM stiffness is normally maximal . AFM measurements showed that this manipulation eliminated stiffness differences between the central and terminal regions , and these follicles show significant elongation defects ( Figure 2G ) . To complement this manipulation , we overexpressed EHBP1 locally in the terminal regions ( using fru-GAL4 , Figure 2K ) . This also equilibrated stiffness between the central and terminal regions , and again led to rounder follicles ( Figure 2H ) . The data overall ( Figure 2L ) indicate that a spatially varying gradient in BM stiffness is essential for elongation , with absolute BM stiffness playing a lesser role . Importantly , direct manipulation of AFM-measured BM stiffness , associated with predictable changes to follicle morphogenesis , argues that the stiffness gradient is instructive for organ shape . To functionally test whether soft or stiff and isotropic or anisotropic BMs can indeed resist tissue expansion differentially , we adapted an organ-swelling assay ( Pastor-Pareja and Xu , 2011 ) . We immersed live follicles in deionized water , creating osmotic stress that leads to water influx into the follicle ( Figure 3A , B , Video 1 ) . Acute expansion of the organ challenges the BM , resulting in bursting which can be monitored by live imaging . This assay measures BM rather than epithelial failure because the follicle epithelium is disrupted well before bursting and Latrunculin A treatment does not accelerate bursting ( Figure 3C , D ) . We hypothesized that the frequency and speed at which the BM bursts would reflect its overall stiffness , whereas the position at which it bursts could indicate the location of a weak point . Consistent with the former hypothesis , WT follicles at stage 8 were more resistant to bursting than those at stage 5 ( Figure 3C , D ) . All collagenase-treated follicles burst instantly . Uniformly depleting Collagen IV or SPARC also induced strong increases in bursting frequency , whereas depleting Collagen IV in the central FCs alone did not ( Figure 3F ) . fat2 and msn-depleted follicles showed a phenotype similar to that caused by directly weakening the BM , and burst more frequently and rapidly than WT follicles ( Figure 3C , D , F; Videos 2 and 3 ) , whereas EHBP1-overexpressing follicles were completely resistant to bursting ( Figure 3F; Video 3 ) . Consistent with the latter hypothesis , WT follicles burst most frequently at polar regions , although bursting in collagenase-treated follicles showed no such preference , and fat2 follicles burst more frequently than WT follicles in non-polar regions ( Figure 3E ) . Other BM manipulations also resulted in bursting phenotypes consistent with the hypothesis ( Figure 3F , G ) . For instance , depletion of Collagen IV in the central FCs ( Mirr>CoIVKD ) relocalized swelling and bursting to this region ( Video 3 ) . Soft follicles generally burst more frequently and more rapidly , whereas mechanically isotropic follicles swelled more isotropically before bursting ( Figure 3F , G ) . Overall , the organ-swelling experiments support the hypothesis that the WT gradient in BM stiffness provides differential resistance to organ expansion that is greatest along the central meridian , and smallest at the poles where most elongation occurs . 10 . 7554/eLife . 24958 . 008Video 1 . WT Follicle swelling in H2O . Bursting of WT follicles when placed in water as shown in Figure 4B . Follicle nuclei are visualized using histone–mRFP , and BM is labeled with ColIV–GFP fluorescence ( green ) . DOI: http://dx . doi . org/10 . 7554/eLife . 24958 . 00810 . 7554/eLife . 24958 . 009Video 2 . fat2 follicle swelling in H2O . Rapid bursting of fat2 follicles when placed in water as quantified in Figure 4D . The follicle is visualized using FM4-64 , and BM is labeled with ColIV–GFP fluorescence ( green ) . DOI: http://dx . doi . org/10 . 7554/eLife . 24958 . 00910 . 7554/eLife . 24958 . 010Video 3 . Osmotic bursting of manipulated genotypes placed in water . As quantified in Figure 4E–G , compared to WT , fat2 follicles burst rapidly and often not at the poles , whereas follicles uniformly overexpressing EHBP1 ( tj>EHBP1 ) swell anisotropically and do not burst at all . Overexpressing EHBP1 in poles ( fru>EHBP1 ) induces generally isotropic swelling but also prevents bursting . Depleting Coll IV in the central region ( mirr>Col IV KD ) cause isotropic swelling and central bursting . DOI: http://dx . doi . org/10 . 7554/eLife . 24958 . 01010 . 7554/eLife . 24958 . 011Figure 3 . The BM stiffness gradient creates anisotropic resistance to organ expansion . ( A ) Design of osmotic-swelling experiments . Immersion in water causes influx ( blue arrow ) into the follicle ( diagrammed in cross-section ) , resulting in increased turgor pressure ( red arrows ) that is resisted by the BM ( green ) as the organ swells . ( B ) WT follicle expressing ColIV–GFP , 1 min and 24 min after immersion ( cf . Video 1 ) . Position of the BM breach is indicated by the yellow arrowhead . ( C ) Frequency of follicle BM failure by stage and genotype , along with timing ( D ) of failure . WT BMs accommodate expansion with increasing efficiency as development proceeds in a manner independent of cellular F-actin; fat2 and collagenased follicles burst frequently and rapidly . ( E ) Position of BM failure: WT BMs breach most frequently at the poles , whereas fat2 and collagenased follicles also breach in other regions . ( F ) Frequency of BM failure in manipulated stage 7–8 follicles and ( G ) aspect ratio immediately before bursting . Scale bar: 25 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 24958 . 011 In what elements does the stiffness gradient lie , and how is it generated ? Previous work has suggested that the follicle is shaped by a ‘molecular corset’ , resulting from the PCP organization of cytoskeletal elements or BM fibril-like structures ( Bilder and Haigo , 2012; Cetera and Horne-Badovinac , 2015; Gutzeit et al . , 1993; Isabella and Horne-Badovinac , 2016; Tucker and Meats , 1976 ) . We used the ‘tissue flattening’ image analysis tool ImSAnE ( Heemskerk and Streichan , 2015; Chen et al . , 2016 ) to analyze follicle BM comprehensively , including BM around the entire A–P and circumferential axes of the organ ( Figure 4A ) . In addition to PCP fibril organization , this approach revealed two unappreciated features . 10 . 7554/eLife . 24958 . 012Figure 4 . Uniform circumferential mechanics in elongating follicles . ( A ) ‘Unrolling’ of organ surface by ImSAnE allows quantitation of BM components along both A–P and circumferential axes . Image taken from Chen et al . ( 2016 ) . ( B ) Analysis of BM fibril PCP shows WT polarity when Perl or DomeDN are overexpressed or when SPARC is depleted , contrasting with altered polarity in fat2 and absence of polarity in Col IV-depleted mutants . ( C , D ) Unrolling reveals increased variance in circumferential Col IV levels in fat2 as compared to those in WT or Perl-overexpressing follicles . The heat map indicates lowest ( blue ) to highest ( red ) intensities over equivalent ~35% circumferential segments . ( E , F ) AFM analysis along the circumferential axis of a follicle at a single central meridian . fat2 mutant follicles show high variability in BM stiffness , compared to the consistent values of WT or Perl-overexpressing follicles . Scale bars: 5 µm ( B ) and 10 µm ( C ) . DOI: http://dx . doi . org/10 . 7554/eLife . 24958 . 012 First , around the circumferential axis , ImSAnE quantitation showed that WT follicles display a fairly uniform distribution of Collagen IV fibrils , suggesting a regular supracellular network . By contrast , in fat2 mutant follicles , ImSAnE documented not only the loss of BM fibril polarity but also discontinuous and variable distribution of Collagen IV , with regions of high and low deposition ( Figure 4B–D ) . These phenotypes were shared by follicles depleted for msn . Strikingly , in both fat2- and msn-depleted follicles , AFM measurements around the circumference at a single A-P position ( Figure 4E ) revealed a four-fold increase in the variability of stiffness when compared to the highly consistent stiffness of WT follicles ( Figure 4F ) . The data raise the possibility that uniform circumferential mechanical properties , dependent on tissue rotation , may also be required for elongation . Second , along the A–P axis , we noted intriguing A–P differences in BM component levels . During elongation , Collagen IV levels are increased in central regions and taper toward the poles ( Figure 5A ) . Perlecan levels , by contrast , are lower at anterior and central regions than elsewhere ( Figure 5B ) . Finally , Laminin levels are fairly uniform but are low at the anterior ( Figure 5C ) . We extended the analysis of Collagen IV , which is a major contributor to BM stiffness ( Morrissey and Sherwood , 2015 ) . Quantitation using ImSAnE documented a significant increase of Collagen IV levels in central regions as compared to anterior and posterior terminal regions ( Figure 5H , I ) . This pattern is not solely transcriptional as Collagen IV subunit gene expression is not elevated in the central region ( Van De Bor et al . , 2015 ) ( Figure 5F ) , and uniform ectopic expression of ColIagen IV subunits ( via ‘FLPout GAL4’ ) results in non-uniform incorporation of ColIagen IV into the BM , with enhanced levels in the follicle center ( Figure 5G ) . 10 . 7554/eLife . 24958 . 013Figure 5 . Morphogen-like signaling creates the stiffness gradient . Expression of GFP protein traps in BM components , assessed in WT stage 7–8 follicles that are physically flattened for visualization: ( A ) ColIV , ( B ) aminin B1 , and ( C ) Perlecan . Heat maps indicate lowest ( blue ) to highest ( red ) intensities . The A–P ColIV pattern is disrupted in stage 7–8 follicles mutant for fat2 ( [D] , cf . Figure 5—figure supplement 1 ) or with inhibited JAK/STAT signaling ( tj>domeDN , [E] ) ( cf . Figure 5—figure supplement 2 ) . ( F ) Col IV transcription ( ColIV-LacZ reporter expression ) is not elevated in the central follicle . ( G ) Uniform production of ColIV ( via hsFLP; act>y+>GAL4 UAS-myr-RFP ) throughout the follicle ( G’ ) results in elevated central incorporation . ( H ) ImSAnE ‘unrolling’ of the ColIV–GFP expressing follicle surface allows quantitation of intensity along the entire A–P axis; note the shorter axis of ‘round’ genotypes . ( I ) Along the A–P axis , ColIV levels are significantly elevated in the central region of WT and Perl-overexpressing follicles but not of fat2 or domeDN-expressing follicles . ( J ) Elongation failure is induced by inhibition of JAK-STAT signaling or by overexpression of Perl in follicles . ( K ) AFM reveals that follicles with inhibited JAK-STAT signaling or Perl overexpression do not develop an A–P stiffness gradient; Perl overexpressing follicles are softer than WT follicles . ( L ) Perl-overexpressing follicles burst easily under osmotic challenge , whereas follicles with inhibited JAK-STAT signaling are more similar to WT . Scale bars: 25 µm ( A–G’ ) and 10 µm ( H ) . DOI: http://dx . doi . org/10 . 7554/eLife . 24958 . 01310 . 7554/eLife . 24958 . 014Figure 5—figure supplement 1 . fat2KO phenocopies other fat2 null alleles . ( A ) Loss of basal actin PCP , ( B ) elongation defects , ( C ) Collagen IV–GFP pattern ( compare to Figure 6D ) and ( D ) bursting frequency in distilled water are indistinguishable between fat2KO and the well-characterized EMS-generated null allele fat258D . Scale bars: 10 µm ( A ) and 20 µm ( C ) . DOI: http://dx . doi . org/10 . 7554/eLife . 24958 . 01410 . 7554/eLife . 24958 . 015Figure 5—figure supplement 2 . STAT reporter in fat2 mutants . ( A ) fat2 loss does not disrupt A–P patterning as detected by the 10XSTAT–GFP reporter . Scale bar: 20 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 24958 . 01510 . 7554/eLife . 24958 . 016Figure 5—figure supplement 3 . BM stiffness and active follicle rotation . Depletion of actin regulator Abi following stage 5 ( A ) halts rotation but ( B ) does not soften BM at stage 7–8 nor ( C ) alter bursting characteristics at stage 7–8; ( D ) follicles show normal elongation at stage 7–8 ( cf . Cetera and Horne-Badovinac , 2015 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 24958 . 016 We investigated how these A–P differences in BM composition are regulated . Regional variance in BM stiffness will result from a combination of transcriptional and post-transcriptional regulation ( including secretion , incorporation , and higher-order modification ) of Collagen IV along with other BM components . We asked whether any of these processes are controlled by an organizer-like activity that exists at the follicle poles , in which secretion of a cytokine signal activates JAK/STAT to distinguish cell fates along the A–P axis ( Xi et al . , 2003 ) . Interestingly , inhibition of JAK/STAT signaling ( via expression of a dominant negative receptor ) eliminated the differential A–P distribution of Collagen IV without affecting fibril polarity , and this manipulation gave rise to round follicles and eggs ( Figures 4B and 5J ) . Importantly , AFM measurements demonstrated that these follicles showed relatively high but isotropic BM stiffness ( Figure 5K ) . We conclude that morphogen-like signaling results in BM mechanical patterning that drives elongation . How do the various mechanical properties described above integrate to shape the organ ? ‘Molecular corset’ models derive in part from analysis of follicles mutant for fat2 , the prototypical egg elongation regulator , and their mispolarization of PCP elements such as BM fibrils ( Figure 4B ) . However , fat2 mutant follicles also fail to achieve an even distribution of BM around the follicle circumference ( Figure 4C , D ) . Additionally , ImSAnE quantitation reveals that they have perturbed A–P Collagen IV pattern , although no changes in A–P signaling are seen ( Figure 5H , I , Figure 5—figure supplement 2 ) . Finally , fat2 mutant follicles fail to undergo a whole-tissue rotation event associated with elongation ( Haigo and Bilder , 2011; Viktorinová and Dahmann , 2013 ) . To assess the role of active rotation , we depleted the actin regulator Abi at stage 5 , which results in rotation arrest as elongation initiates ( Cetera et al . , 2014 ) . These follicles stiffened comparably to WT , showed bursting response comparable to WT , and also elongated normally ( Figure 5- Figure 1—figure supplement 3 , Video 4 ) . Conversely , elongation is prevented without disrupting rotation in several genotypes ( see below , Video 4 ) , confirming that phenomena other than active rotation are required to manipulate follicle shape . Nevertheless , the altered tissue-wide distributions of Collagen IV in fat2 mutants complicate interpretations that BM fibril PCP forms the molecular corset . 10 . 7554/eLife . 24958 . 017Video 4 . Follicle rotation in manipulated genotypes . Rotation of tj>Perl and tj>Dome-DN is comparable to that of WT , whereas tj>abi-RNAi initiated at stage 5 blocks rotation . Scale bar: 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 24958 . 017 We were unable to identify manipulations that independently disrupted follicle PCP and the circumferentially continuous BM distribution . Therefore , to investigate the role of BM fibril polarity per se in generating elongation-driving mechanical anisotropy , we uniformly overexpressed Perlecan , which antagonizes the constrictive properties of Collagen IV BMs and can induce round eggs ( Isabella and Horne-Badovinac , 2015; Pastor-Pareja and Xu , 2011 ) . This manipulation did not change the A–P levels , PCP , or circumferential distribution of Collagen IV fibrils ( Figures 4B , D and 5H , I ) . However , AFM analysis revealed that it did create a softer BM in which the anisotropic gradient has been eliminated , and the enclosed follicles fail to elongate ( Figure 5J , K ) . In osmotic stress experiments , follicles overexpressing perlecan swelled more isotropically and burst more rapidly than WT follicles ( Figure 5L ) . Thus , despite the fact that neither the levels , local PCP , or supracellular organization of Collagen IV fibrils are altered in Perlecan-overexpressing follicles , the BM of these follicles had mechanical deficits similar to those of follicles completely lacking a BM . By contrast , follicles deficient for STAT-dependent A–P signaling also fail to elongate but show normal fibril polarity and organization , and are significantly more resistant to bursting ( Figures 4B and 5J–L ) . Together , these data support a requirement for a circumferentially even distribution of PCP fibrils in elongation . However , they also reveal that PCP fibrils alone are insufficient to resist tissue growth anisotropically; the organ-shaping stiffness gradient requires patterned A–P BM levels .
Organ elongation is a fundamental developmental process , and is generally considered to be driven by cell-intrinsic polarized mechanical forces that actively deform tissues . Here , we demonstrate that an elongating tissue can rely instead on mechanical anisotropy patterned into the BM . Our data indicate that this asymmetric resistance within the extracellular environment , rather than asymmetric force generation within the cells , plays the dominant role in molding the follicle , prescribing subsequent morphogenetic cell behaviors . These results direct increased attention to fine BM spatial organization in creating the mechanical environment that shapes each tissue , and may fill the gap between the limited repertoire of cell-based morphogenetic mechanisms and the immense diversity of organ shapes . Stromatic ECMs and BMs surround most animal organs , but their full roles in morphogenesis remain unresolved . Long regarded as an inert scaffold , ECM is known to influence tissue biology through actively regulating ligand availability and adhesion signaling; local BM deposition and degradation also play key roles in the branching morphogenesis of several mammalian organs ( Daley and Yamada , 2013; Harunaga et al . , 2014; Morrissey and Sherwood , 2015; Pastor-Pareja and Xu , 2011; Varner and Nelson , 2014 ) . However , analysis of the mechanical properties of vertebrate BMs in vivo is hampered by surrounding cellular stroma , whose removal necessitates non-physiological manipulations . Because of this , only exceptionally robust BMs , such as those of the eye , have been analyzed following denuding protocols ( Ali et al . , 2016 ) . By contrast , fly follicles lack a cellular stroma , and their topology allows direct access of AFM probes to the BM of an intact living tissue . Our in vivo biophysical measurements of this native BM reveal an unappreciated degree of tissue-level mechanical patterning . Within each follicle , BM stiffness develops reliably and with spatial properties that are carefully regulated along both the A–P and circumferential axes . Along the A–P axis , a stiffness gradient is built that increases ~300% along a ~13-cell , 100 μm arc at stage 8 . Perpendicular to this axis , stiffness around the circumference varies by less than 5% across the same distance . Our data reveal that both axes are critical for organ shaping , and merit a significant revision of the ‘molecular corset’ model previously proposed to mediate elongation ( Bilder and Haigo , 2012; Cetera and Horne-Badovinac , 2015; Gutzeit et al . , 1993; Isabella and Horne-Badovinac , 2016; Tucker and Meats , 1976 ) . Hypotheses of corset structure have focused on the PCP organization of the basal actin network , the microtubule cytoskeleton , or the fibril-like BM . However , manipulations that preserve PCP alignment but nevertheless result in round follicles demonstrate that mechanical anisotropy at the length scale of individual BM fibrils is not sufficient to drive elongation . Instead , they suggest that consistent circumferential stiffness , probably associated with the supracellular BM fibrillar network generated by whole-tissue rotation , is a key element of corset effectiveness . Moreover , manipulations that flatten a pole-derived A–P signaling gradient also flatten the A–P stiffness gradient , and create isotropic organs . Thus , to drive elongation , the corset must also be anisotropic on a ‘global’ tissue-wide scale , in a manner that depends on morphogen-regulated mechanical properties . The direct manipulations of BM components presented here , which lead to predicted tissue shape outcomes , argue that BM mechanics themselves are instructive for morphogenesis . Flattening the stiffness gradient in several ways , including by locally restricted BM alteration , prevents elongation , whereas hyperelongating follicles have an enhanced stiffness gradient . Although we cannot rule out undetected roles of these manipulations in altering cell behaviors via classical intercellular signaling , we see no evidence for such changes in the underlying epithelium . Instead , our results indicate that elongation is imposed by isotropic tissue growth meeting the anisotropic resistance fashioned within the BM . Consistent with this model , manipulations that alter absolute stiffness but preserve a relative gradient still result in tissue elongation . The extent to which ovarian cells respond compliantly or through well-characterized mechanical feedback mechanisms remain to be determined , but the data point to physical properties of the BM as the dominant influence . Our results reveal a tissue elongation mechanism that is conceptually different from cell-intrinsic force asymmetries . Construction of mechanically patterned resistance in an ECM , along both axes orthogonal to its tissue interface , generates a force imbalance that imposes a specific shape on the growing organ , without necessitating spatially restricted localization of force generators within cells . Emerging examples point to the influence that substantial changes in exogenous physical forces can have in organ morphogenesis ( Aigouy et al . , 2010; Behrndt et al . , 2012; Etournay et al . , 2015; Harunaga et al . , 2014; rayRay et al . , 2015; Shyer et al . , 2013 ) as well as in tumor growth ( Kaushik et al . , 2016 ) . The discovery of precise organ-sculpting resistance within a BM motivates the development of tools and assays to explore , on a fine scale , true in vivo BM mechanical properties in both physiological and pathological contexts .
The GAL4 drivers used were tjGAL4 , mirrGAL4 and fruGAL4 ( Borensztejn et al . , 2013 ) ; tubGAL80ts was used to control expression temporally by shifting flies to 29° . The Drosophila genome contains two Collagen IV subunit-encoding genes: ColIVα1 ( Flybase: Cg25c ) and ColIVα2 ( Flybase: vkg ) . For ease , both are referred to in the text and figures as Collagen IV; detailed genotypes for all experiments are listed in Supplementary file 1 . Overexpression constructs UAS-DT-A , UAS-Perlecan ( Flybase: Trol ) , UAS-DomeDN , and UAS-EHBP1 ( Giagtzoglou et al . , 2012 ) ; RNAi constructs against Abi , SPARC , ColIVα1 , ColIVα2 and msn; and GFP protein traps in Collagen IVα2 and perlecan were obtained from the Bloomington stock center . Fosmids carrying LanB1–GFP ( Sarov et al . , 2016 ) were obtained from VDRC . Myo–GFP ( Flybase: sqh ) was provided by Dan Kiehart . Strains showing ectopic expression of ColIV–GFP ( UAS–GFP–ColIVα1 + UAS–GFP–ColIVα2 ) were provided by S . Noselli ( Van De Bor et al . , 2015 ) , and utilized hsFLP; act>y+>GAL4; UAS-myrRFP , activated by a 30 min heat shock at 37° and immediately imaged with RFP signal to confirm uniform expression . fat2KO , kindly provided by Mike Simon , is a null allele generated by ends-out gene replacement ( Maggert et al . , 2008 ) into the first exon and phenocopies other fat2 null alleles ( Figure 5—figure supplement 1 ) . Ovary preparations for fixed and live imaging were performed as previously described ( Chen et al . , 2016 ) . Phalloidin-staining of fixed follicles used 20 nM phalloidin ( Sigma ) . Latrunculin A 50 µM ( Sigma ) , FM4-64FX 5 µg/mL ( Thermo ) , and purified Collagenase 1000 U/mL ( Worthington LS005273 ) were diluted in Schneider’s complete media ( 10 mg/mL insulin , FBS and pen/strep ) for live imaging . The measured osmolarity of the standard media was 300 mOsm . Hypertonic shrinking was performed in standard media supplemented with 1M D-sorbitol ( Sigma ) to 2000 mOsm . Fixed follicles were mounted with tape spacers , except for flattened preparations ( which lacked spacers ) and ImSAnE preparations ( which were mounted in a depression slide ) . Single-plane confocal images were acquired on a Zeiss LSM700 using a Plan Apochromat 20x/NA 0 . 8 lens or a LD C-Apochromat 40x/NA 1 . 1 water-immersion lens and processed in Fiji software ( Schindelin et al . , 2012 ) . Representative images were isolated and assembled into figures using Adobe Photoshop and Illustrator CS6 . For cortical MyoII planar polarity quantification ( Munjal et al . , 2015 ) , IMSAnE ( Heemskerk and Streichan , 2015 ) was used to ‘unroll’ the follicle epithelia as previously described by Chen et al . , ( 2016 ) but with modifications . Apical surfaces of interest ( SOI ) of the epithelia were identified by Sqh-GFP signal . Multilayered cylinder projections of the apical-lateral membranes from the apical-most SOI plus minus 2–2 . 5 μm were generated by IMSAnE class CylinderMeshWrapper . Maximum intensity projections were background subtracted with the Fiji plugin ‘subtract background’ . A–P and circumferential junctions were categorized by 60–90° and 0–30° degrees , respectively , relative to the A–P axis . Cortical Sqh–GFP was selected manually with line tools ( width 8px ) on >30 junctions of each type; the mean ratio was plotted . For ColIV–GFP intensity measurements , in toto images were collected with pixel width of 0 . 17 μm and voxel depth of 0 . 50 μm without Z-intensity correction . Follicle SOI was identified using basal F-actin signal and generalized sinusoidal projections were generated by the IMSAnE class spherelikeFitter . Maximum intensity projections from multilayered pullbacks ±3 μm from the basal epithelia were generated . To measure A–P intensity , five 1-pixel-wide lines were drawn within a 10 μm wide stripe at the central meridian , where the pullbacks have minimal distortion . To measure circumferential intensity , five circumferential 1-pixel-wide lines were drawn within a 10 μm wide stripe along the circumferential meridian . Intensities were standardized to follicle length , then compared across follicles . Variance was calculated for each follicle using the Excel var . p formula . Profile plots were generated in Fiji software . Ecad–GFP follicles were dissected in medium and placed in a glass-bottomed dish . A pulsed Mai-Tai two-photon on a Zeiss LSM 510 confocal microscope was used to sever A–P or circumferential junctions at anterior , central , and posterior positions on the follicle . At 708 nm and 90% power , the ablation time was less than 1 s and the resulting junction relaxation distances were measured within 300 ms . Analysis was executed manually in Fiji software normalizing the relaxed distance to the original junction length . Similar results were obtained using a UV Micropoint laser at 50% power and a Nikon Ti-E inverted microscope with a Yokogawa X1 confocal spinning disk head , with images continuously collected ( 500 ms/frame ) . BM stiffness was measured ( Figure 1—figure supplement 2 ) using either a Bruker Catalyst AFM controlled by Nanoscope 8 . 10 software or a custom-built AFM controlled by LAbview software , both mounted on an inverted Zeiss AxioObserver Z1 microscope . MLCT-C cantilevers ( Bruker ) with a pyramidal tip and a nominal spring constant of 10 pN/nm were used in all experiments . The actual spring constant of each cantilever was determined by thermal calibration in air . Measurements were done in fluid . Approach velocity was optimized as 0 . 4 µm/sec to ensure the fastest rate of elastic measurement without viscoelastic deformation . Sample rate of deflection was 2048 . Retraction speed , which does not affect elasticity measurements , was set to 20 µm/sec . Follicles were prepared as for live imaging; the cantilever was positioned at the desired position by brightfield microscopy . Each positional measurement was taken four times without moving the cantilever in XY and averaged . Young’s Modulus of elasticity was calculated by fitting the cantilever deflection versus piezo extension curves to the modified Hertz model as described ( Rosenbluth et al . , 2006 ) , using a custom-written algorithm in MATLAB ( Mathworks ) . Only the first 50 nm of indentation were used to isolate elasticity from just the basement membrane ( BM ) . For pole measurements , PDMS egg holders were created using custom-made molds , coated first with poly-D lysine and then treated with complete growth media . Follicles were gently mounted in PBS which was subsequently replaced with media . Follicles dissected in complete media were adhered to a poly-D lysine glass-bottomed dish ( MatTek ) before replacing the medium twice with dH20 . Images were collected at 15 s or 30 s intervals on a Zeiss Axioimager with a Plan-Neofluor 10x/0 . 38NA objective . Data were analyzed and displayed using Microsoft Excel . All error bars represent standard errors and centers represent means . At least three biological replicates were undertaken for each experiment and the results are given in Supplementary file 1 . All acquired data were included with the exception of the AFM experiments . For these , only follicles in which all three lateral positions could be quantified were used . Statistical analysis for all data used two-tailed t-tests with p-value thresholds of *p<0 . 05 , **p<0 . 01 , and ***p<0 . 001 . | All organs have specific shapes and architectures that are necessary for them to work properly . Many different factors are responsible for arranging the right cells into the correct positions to make an organ . These include physical forces that act within and around cells to pull them into the right shape and location . A structure called the extracellular matrix surrounds cells and provides them with support; it can also guide cell movements . It is not clear whether the extracellular matrix plays only a passive role or a more active , instructive role in shaping organs , in part , because it is difficult to measure the physical forces within densely packed cells . The ovaries of the fruit fly Drosophila melanogaster provide a simple system in which to study how organs take their shape . Crest et al . developed a method to measure forces in the fly ovary as it changes from being an initially spherical group of cells to its final elongated tube shape . The results revealed that , during this process , the extracellular matrix becomes gradually stiffer from one end of the ovary to the other . This change is the main factor responsible for the cell rearrangements that shape the developing organ . This work reveals that , along with providing structural support to cells , the mechanical properties of the matrix also actively guide how organs form . In the future , these findings may aid efforts to grow organs in a laboratory and to regenerate organs in human patients . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology",
"cell",
"biology"
] | 2017 | Organ sculpting by patterned extracellular matrix stiffness |
NMDA receptor ( NMDAR ) -dependent forms of synaptic plasticity are thought to underlie the assembly of developing neuronal circuits and to play a crucial role in learning and memory . It remains unclear how NMDAR might contribute to the wiring of adult-born granule cells ( GCs ) . Here we demonstrate that nascent GCs lacking NMDARs but rescued from apoptosis by overexpressing the pro-survival protein Bcl2 were deficient in spine formation . Insufficient spinogenesis might be a general cause of cell death restricted within the NMDAR-dependent critical time window for GC survival . NMDAR loss also led to enhanced mushroom spine formation and synaptic AMPAR activity throughout the development of newborn GCs . Moreover , similar elevated synapse maturation in the absence of NMDARs was observed in neonate-generated GCs and CA1 pyramidal neurons . Together , these data suggest that NMDAR operates as a molecular monitor for controlling the activity-dependent establishment and maturation rate of synaptic connections between newborn neurons and others .
New neurons are continuously generated in the hippocampus of the adult mammalian brain . These neurons become granule cells ( GCs ) , the principal neurons in the dentate gyrus ( DG ) of the hippocampus , and they functionally integrate into the hippocampal circuitry ( van Praag et al . , 2002; Toni et al . , 2008 ) . This extreme form of structural remodeling , similar to many other forms of experience-dependent plasticity , requires activation of NMDA receptors ( NMDARs ) ( Platel and Kelsch , 2013 ) . Blockade of NMDARs rapidly increases the proliferation of neural precursor cells , whereas stimulation of NMDARs promotes neuronal fate specification ( Cameron et al . , 1995; Deisseroth et al . , 2004 ) . Moreover , deletion of the NMDAR subunit NR1 reduces the survival rate of adult-born GCs ( Tashiro et al . , 2006a ) . Other than these studies , the role of NMDAR in circuit assembly during new neuron development has so far received little attention , even though NMDAR has been primarily known for its involvement in synapse organization and synaptic plasticity ( Constantine-Paton , 1990; Malenka and Nicoll , 1993 ) . The wiring of new neurons in mature circuits involves a coordinated series of events , from the initial cell contact to the final maturation of functional synapses . Not only the cells themselves but also the connections between newly recruited members and their old counterparts are survivors of a selection process depending upon neuronal activity patterns . The vast majority of excitatory inputs are received by small bulbous protrusions residing on the dendrites of glutamatergic neurons , called dendritic spines . In newborn GCs , dendritic spines first appear around 16 days after neuronal birth . The spine density increases sharply before cells reach 4 weeks of age and continues to increase at a slower pace until reaching a plateau at 8 weeks ( Zhao et al . , 2006 ) . Notably , the NMDAR-dependent survival/death of adult-born GCs is restricted to the time window of 2–3 weeks after neuronal birth , shortly after formation of the first dendritic spines ( Tashiro et al . , 2006a ) . This temporal overlap suggests that the survival of newborn GCs may be related to the state of spinogenesis or synaptogenesis . Indeed , cell death can be induced by non-innervation at the peak of synaptogenesis during embryonic and postnatal development ( Naruse and Keino , 1995 ) . Since activation of NMDARs has been shown to support new spine formation ( Maletic-Savatic et al . , 1999; Kwon and Sabatini , 2011 ) , we hypothesize that NMDAR is required for initial spine gain on dendrites of newborn GCs and that insufficient spine growth may be the underlying cause of the cell death associated with the genetic deletion of the NMDAR subunit NR1 in newborn GCs . There is a positive correlation between spine volume and the number of AMPA ( α-amino-3-hydroxy-5-methyl-4-isoxazole propionic acid ) receptors ( AMPARs ) or , more generally , the synaptic strength ( Matsuzaki et al . , 2001 ) , supporting the model that spine outgrowth and enlargement are tightly coupled to the formation and maturation of glutamatergic synapses ( Zito et al . , 2009 ) . Subject to activity-dependent modifications , spines are highly dynamic in their number , shape and size . Long-term potentiation ( LTP ) and long-term depression ( LTD ) at mature synapses , expressed by synaptic insertion and removal of AMPARs , respectively , are associated with NMDAR-dependent enlargement and shrinkage of dendritic spines ( Yuste and Bonhoeffer , 2001; Matsuzaki et al . , 2004; Zhou et al . , 2004 ) . Stimuli that induce LTP or LTD may also result in rapid outgrowth or loss of spines , and these changes can be prevented by NMDAR blockers ( Engert and Bonhoeffer , 1999; Toni et al . , 1999; Nagerl et al . , 2004 ) . The structural plasticity of spines is thought to complement functional plasticity ( e . g . , LTP and LTD ) and play a central role in learning and memory in mature animals ( Bourne and Harris , 2007 ) . In contrast , emerging evidence suggests that , during early postnatal development , AMPARs can be delivered to spines independently of NMDAR signaling . NMDARs actually restrict AMPAR trafficking to the postsynaptic density and limit synapse maturation ( Ultanir et al . , 2007; Adesnik et al . , 2008; Gray et al . , 2011 ) . Due to variables in experimental designs , the observed opposite effects of NMDARs on AMPAR recruitment in developing vs mature neurons need further examination . Adult-born neurons undergo a long process of maturation resembling embryonic development ( Esposito et al . , 2005 ) ; however , adult neurogenesis happens in mature circuits that differ substantially from the developing brain . We aimed to characterize NMDAR functions during integration of new GCs into the circuit at distinct stages of their development using the tool of single-cell gene deletion and labeling . To address the above issues , we performed an analysis of spine morphogenesis in immature NR1 knockout ( KO ) GCs that either survived naturally or were rescued by the apoptosis regulator Bcl-2 . Both groups of cells were found to be deficient in spine formation . In parallel , we observed an elevation in mushroom spine density and in synaptic AMPAR activity in the absence of NR1 . Furthermore , NMDAR loss initiated at a later stage of GC development , similar to NR1 KO in fully mature GCs or CA1 pyramidal neurons , resulted in enhanced functional synapses without affecting total spine numbers . Thus , NMDAR appears to play two distinct roles during GC development . First , it promotes the initial spine formation and its presence is required for the survival of immature GCs . Second , the receptors monitor spine enlargement and the recruitment of AMPAR once spines are formed . Both aspects of NMDAR function contribute to experience-driven construction of circuits formed by new neurons , even though NMDAR signaling might make less of a contribution to the control of overall spine density upon neuronal maturation .
To determine the role of NMDAR in circuit assembly of new neurons in the adult brain , we first examined the morphology of NR1 KO GCs at 4 weeks of age . A retrovirus encoding Cre recombinase and GFP was developed for inducible knockout of the floxed Grin1 alleles in adult mice ( rv GFP-ires-cre; Figure 1—figure supplement 1 ) . When tested in the ROSA-lacZ reporter mice , rv GFP-ires-cre induced recombination in 97% of GFP+ cells at 6 days post infection ( dpi ) and in all GFP+ cells at 14 and 28 dpi . We then injected rv GFP-ires-cre together with a control retroviral vector expressing mCherry only ( rv CAG-mCherry ) into the Grin1 floxed mice ( Tsien et al . , 1996; Tashiro et al . , 2006a ) ( Figure 1A ) . To assess NMDAR activity in virus-transduced cells and confirm the cell-specific knockout of the Grin1 gene via Cre/loxP recombination , we performed perforated whole-cell patch-clamp recordings at a holding potential of −70 mV and +40 mV to monitor synaptic responses mediated by AMPA and NMDARs , respectively . In control adult-born GCs ( mcherry+GFP− ) at 28 dpi , both AMPA and NMDA currents could be readily evoked by perforant path stimulation ( Figure 1B ) . In contrast , there was only a DNQX-sensitive AMPA component in age-matched GFP+ neurons infected by rv GFP-ires-cre ( Figure 1B ) , suggesting that Cre-mediated recombination successfully removed the floxed Grin1 gene fragment from the mouse genomic DNA . 10 . 7554/eLife . 07871 . 003Figure 1 . NR1 KO cells display decreased spine growth but enhanced spine maturation and AMPAR activity at 4 weeks of age . ( A ) Co-injection of rv CAG-mcherry ( red ) and CAG-GFP-ires-cre ( green ) for the simultaneous labeling of wild-type ( WT ) and NR1 KO newborn granule cells ( GCs ) . ( B ) Left: mCherry+ newborn GCs respond to perforant path stimulation in the absence ( upper panel ) and presence ( lower panel ) of the AMPAR antagonist DNQX . Right: GFP+ Cre-expressing newborn GCs respond to perforant path stimulation in the absence ( upper panel ) but not in the presence ( lower panel ) of DNQX . ( C , D ) Representative images of WT ( C ) and NR1 KO ( D ) newborn GCs at 4 weeks of age . ( E ) Representative images of dendritic processes of newborn WT ( GFP ) and NR1 KO ( cre ) GCs in the outer molecular layer . ( F ) Total spine density is decreased in NR1 KO newborn GCs . ( G ) Comparison of the percentage of each spine type relative to total spine numbers in adult-born WT and NR1 KO GCs . ( H ) Mushroom spine density is increased in NR1 KO newborn GCs . ( I ) Cumulative plot of spine size in NR1 WT and KO GCs . ( J ) Representative traces of AMPAR-mediated miniature excitatory postsynaptic currents ( mEPSCs ) in mCherry+ WT and GFP+ NR1 KO newborn GCs . ( K , L ) Quantitative analysis of mEPSCs by amplitude ( K ) and frequency ( L ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07871 . 00310 . 7554/eLife . 07871 . 004Figure 1—figure supplement 1 . Retrovirus rv CAG GFP-ires-cre was delivered to the dentate gyrus ( DG ) of ROSA-lacZ mice and the recombination efficiency was examined by the expression of β-gal ( red ) in Cre-expressing cells ( GFP+ , green ) . Insets on the right side represent GFP− and β-gal-expressing cells , respectively . Scale bar: 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 07871 . 004 Because the fluorescent signal produced by rv CAG-mCherry labeling was not sufficient for optimal image acquisition and analysis of dendritic spines , we then injected the control CAG-GFP or GFP-ires-cre retrovirus into Grin1f/f mice to compare the morphology of NR1 wild-type ( WT ) and KO cells . There was no obvious difference in overall cell morphology between NR1 KO and WT cells ( Figure 1C , D ) . Dendritic tracing with the ICL TRACE ( http://synapses . clm . utexas . edu/tools/trace/trace . stm ) showed that WT and KO cells were similar in both dendritic length ( WT: 601 . 8 ± 36 . 6 , n = 47 frames , KO: 515 . 1 ± 29 . 3 , n = 38 frames , p = 0 . 08 ) and branching points ( WT: 5 . 92 ± 0 . 33 , KO: 5 . 68 ± 0 . 37 , p = 0 . 64 ) . These data indicate that gross development of dendrites does not require NMDARs . However , detailed analyses of the dendritic segments in the outer third of the molecular layer revealed significant differences between WT and KO cells ( Figure 1E ) . According to the criteria described by Harris and Yuste ( Harris et al . , 1992; Parnass et al . , 2000 ) , all dendritic protrusions were classified into four categories: filopodia , stubby , thin and mushroom spines . Briefly , ‘stubby’ were neckless spines whose head diameters were about equal to their lengths . Spines were defined as ‘filopodia’ if they were long , thin and did not have a head . In contrast , ‘thin’ spines had long , thin necks and obvious heads . ‘Mushroom’ spines were similar to ‘thin’ spines in shape , but with larger heads ( see ‘Materials and methods’ ) . As shown in Figure 1F , total spine density was significantly decreased in NR1 KO cells ( WT: 2 . 14 ± 0 . 08 , n = 59 frames , KO: 1 . 53 ± 0 . 09 , n = 37 frames , p < 0 . 0001 ) . Surprisingly , the percentage of mushroom type relative to total spine numbers was drastically enhanced , whereas the other spine types remained unchanged ( p < 0 . 0001; Figure 1G ) . In line with this observation , mushroom spine density was increased by more than twofold in NR1 KO cells ( WT: 0 . 026 ± 0 . 004 , KO: 0 . 064 ± 0 . 010 , p = 0 . 0001; Figure 1H ) . A cumulative probability graph of the size of all measured spines showed that the spine head area in NR1 KO cells was bigger than that in control cells ( p < 0 . 0001 , Kolmogorov–Smirnov test; Figure 1I ) . The size of the spine head has been positively correlated with synaptic AMPAR level ( Matsuzaki et al . , 2001 ) . Since NR1 KO cells displayed increased mushroom spines , we postulated that NMDAR KO cells might have more synaptic AMPAR activity . Therefore , we measured AMPAR-mediated miniature excitatory postsynaptic currents ( mEPSCs ) . While mEPSCs were infrequent in mCherry+GFP− NR1 WT cells , they were evident in GFP+ NR1 KO cells ( Figure 1J ) . Both the amplitude and frequency of mEPSCs were significantly increased in neurons lacking NR1 as compared to control cells ( WT: n = 5 cells , KO: n = 6 cells , p < 0 . 001 , Kolmogorov–Smirnov test; Figure 1K , L ) . These data confirmed that NR1 KO cells had enhanced functional glutamatergic synapses . Utilizing serial immuno-electron microscopy for GFP , we consistently observed that GFP+ dendritic spines were associated with GFP− axon terminals containing presynaptic vesicles ( Figure 2A ) , verifying that newborn NR1-null GCs could form normal synapses . We also noted that all stubby , thin and mushroom spines were asymmetric and presumably excitatory ( Figure 2A ) . While the axo-dendritic synapses located on dendritic shafts had a typical morphology of symmetric or GABAergic synapses ( Figure 2A ) , they could be initially excitatory in newly born GCs due to the high intracellular chloride concentration ( Ge et al . , 2006 ) . Total head volumes of randomly selected spines averaged 0 . 035 ± 0 . 005 µm3 ( n = 130 spines ) and 0 . 078 ± 0 . 014 µm3 ( n = 74 spines ) in WT and NR1 KO GCs , respectively , indicating a significant difference between these two groups ( p = 0 . 0008; Figure 2B ) . Furthermore , we found that this difference was mainly due to increased volume of big or mushroom spines in NR1 KO neurons , whereas the volumes of small spines ( presumably filopodia or thin ) were roughly the same across groups . These results suggest that changes in the amount of depolarization-induced Ca2+ influx may affect the survival of nascent neurons , although the role of a possible re-distribution of Ca2+ entry through different spine categories cannot be excluded . 10 . 7554/eLife . 07871 . 005Figure 2 . Electron microscopic description of dendritic spines . ( A ) Electron micrograph illustrating dendrites of newborn NR1 KO neurons . Panels show examples of symmetric axo-dendritic synapses ( left panels , arrows ) , filopodia ( middle left panels ) , thin spines ( middle right panels ) and mushroom spines ( right panel ) . Darkly immunolabeled GFP+ dendritic spines are each contacted by GFP− axon terminals ( asterisks ) containing numerous presynaptic vesicles . Scale bars: 1 μm . ( B ) Comparison of total spine volumes in NR1 KO and WT cells . DOI: http://dx . doi . org/10 . 7554/eLife . 07871 . 005 Given that most NR1 KO GCs die before reaching 4 weeks of age ( Tashiro et al . , 2006a ) , it is questionable whether the observed abnormal spine morphogenesis in the few surviving NR1 KO cells represents a general phenomenon in NR1-null neurons . To clarify this issue , we engineered a new retroviral vector expressing a fusion protein of the pro-survival gene Bcl2 and GFP in addition to Cre recombinase ( rv GFPBcl2-ires-cre; Figure 3—figure supplement 1 ) to prevent NR1 KO cells from dying . The logic behind this set of experiments is: NR1 deletion in new GCs results in certain defects that eventually lead to cell death . Since Bcl2 is an important anti-apoptotic protein ( Tsujimoto et al . , 1984; Cleary et al . , 1986 ) , it may suppress apoptosis and drive NR1 KO cells to survive even though they have deficiencies . Therefore , neurons rescued by Bcl2 expression should exhibit features distinct from those surviving normally , and these differences presumably reflect defects caused by NR1 loss . We first tested the virus efficiency of cell death regulation in WT C57Bl/6 mice . In line with prior studies ( Dayer et al . , 2003; Tashiro et al . , 2006b ) , within 28 dpi of rv CAG-GFP , we found a significant decrease in the quantity of GFP+ GCs in C57Bl/6 mice ( GFP , relative cell number to that of 14 dpi , 14 dpi: 1 . 00 ± 0 . 11 , n = 5 mice , 28 dpi: 0 . 30 ± 0 . 22 , n = 4 mice , p = 0 . 019; Figure 3A ) . In contrast , GCs infected by rv GFPBcl2-ires-cre did not die ( GFPBcl2icre , 14 dpi: 1 . 00 ± 0 . 29 , n = 6 mice , 28 dpi: 1 . 00 ± 0 . 14 , n = 6 mice; Figure 3A ) . More importantly , the dramatic death of new neurons observed in Grin1f/f mice injected with rv GFP-ires-cre was not found in Grin1-floxed animals injected with rv GFPBcl2-ires-cre ( GFPicre , 7 dpi: 1 . 00 ± 0 . 37 , n = 5 mice , 21 dpi: 0 . 05 ± 0 . 03 , n = 5 mice , p = 0 . 032; GFPBcl2icre , 7 dpi: 1 . 00 ± 0 . 31 , n = 5 mice , 21 dpi: 1 . 31 ± 0 . 39 , n = 5 mice , p = 0 . 5; Figure 3B ) , suggesting that the expression of Bcl2 did prevent the death of NR1 KO cells . 10 . 7554/eLife . 07871 . 006Figure 3 . NR1 KO cells rescued by expression of the pro-survival gene Bcl2 had an unusually low spine density . ( A ) Bcl2 expression was able to rescue newborn GCs from naturally occurring cell death in C57Bl/6 mice . ( B ) Bcl2 expression was able to rescue NR1 KO newborn GCs from cell death . ( C ) For morphological analysis of GFPBcl2-ires-cre-targeted cells , rv GFPBcl2-ires-cre was co-injected with rv GFP . Bcl2-ires-cre-targeted cells were identified by immunohistochemistry using an antibody against the cre recombinase . The arrow and asterisk show GCs transfected by both rv GFPBcl2-ires-cre and rv GFP . Insets on the right side represent GFP− ( green ) and Cre-expressing ( red ) GCs , respectively . ( D–G ) Analysis of dendritic growth of newborn GCs at 28 dpi . Bcl2-rescued newborn GCs had a decreased total dendritic length in both C57Bl/6 and Grin1f/f mice ( D , F ) , whereas branching points were not changed ( E , G ) . ( H ) Representative images of dendritic processes in the outer molecular layer of newborn GCs at 28 dpi in C57Bl/6 mice labeled by rv GFP , GFP-ires-cre and GFPBcl2-ires-cre . ( I , J ) Total spine density was similar in GFP and GFP-ires-cre targeted newborn GCs but significantly decreased in GFPBcl2-ires-cre targeted newborn GCs . ( K ) The density of mushroom spines did not change significantly in newborn GCs targeted by the three retroviruses . ( L ) Comparison of the percentage of each spine type relative to total spine numbers in new GCs of C57Bl/6 mice infected by the three retroviruses . ( M ) Representative images of dendritic processes in the outer molecular layer of newborn GCs at 28 dpi in Grin1f/f mice labeled by rv GFP , GFP-ires-cre and GFPBcl2-ires-cre . ( N , O ) Total spine density was significantly lower in surviving NR1 KO cells ( Cre ) and further decreased in Bcl2-rescued cells . ( P ) Mushroom spine density was increased in NR1 KO cells ( Cre ) but not in Bcl2-rescued cells . ( Q ) Comparison of the percentage of each spine type relative to total spine numbers in new GCs of Grin1f/f mice infected by the three retroviruses . Scale bars: 20 µm ( C ) and 2 µm ( H , M ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07871 . 00610 . 7554/eLife . 07871 . 007Figure 3—figure supplement 1 . GCs infected by rv GFP Bcl2-ires-cre in ROSA-lacZ reporter mice . Green—GFP , red—β-gal . Insets on the right side represent GFP− and β-gal-expressing cells , respectively . Scale bar: 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 07871 . 007 Next we assessed the morphology of NR1 KO cells that were rescued by Bcl2 expression . Because the fluorescent signal from GFPBcl2 fusion protein was not strong enough for morphological analyses , rv GFPBcl2-ires-cre was injected together with rv CAG-GFP . In this case , we identified rv GFPBcl2-ires-cre targeted cells by Cre immunostaining and imaged Cre+GFP+ cells when the newborn cells were 4 weeks old ( Figure 3C ) . GFP-labeled GCs from animals injected with pure rv CAG-GFP or rv GFP-ires-cre were imaged in parallel for control purposes . In C57Bl/6 mice , Bcl2-expressing cells displayed decreased dendritic length ( GFP: 378 . 2 ± 32 . 0 , n = 58 frames , cre: 351 . 5 ± 45 . 6 , n = 33 frames , Bcl2: 283 . 8 ± 22 . 0 , n = 93 frames , p = 0 . 013 Bcl2 vs GFP; Figure 3D ) . The number of dendritic branching points was not significantly different between the samples ( GFP: 5 . 2 ± 0 . 3 , cre: 4 . 9 ± 1 . 5 , Bcl2: 4 . 4 ± 0 . 2 , p = 0 . 056 Bcl2 vs GFP , Figure 3E ) . Similarly , in Grin1f/f mice , Bcl2-expressing cells displayed a significant decrease in dendritic length ( GFP: 342 . 4 ± 32 . 9 , n = 55 frames , cre: 302 . 7 ± 22 . 8 , n = 85 frames , Bcl2: 256 . 0 ± 18 . 0 , n = 133 frames , p = 0 . 014 Bcl2 vs GFP; Figure 3F ) but not in branching points ( GFP: 5 . 4 ± 0 . 4 , cre: 5 . 1 ± 0 . 3 , Bcl2: 4 . 7 ± 0 . 3 , p = 0 . 14 Bcl2 vs GFP , Figure 3G ) . Since NR1 KO neurons did not display impairment of dendrite length and complexity ( Figure 1C , D ) , we infer from these data that Bcl2 expression promoted survival of a mixed population of nascent GCs , including both WT and NR1 KO cells . Detailed analyses of the dendritic processes in the outer molecular layer showed that Bcl2-expressing cells in C57Bl/6 mice had much lower total spine density ( GFP: 2 . 13 ± 0 . 11 , n = 28 frames , cre: 2 . 22 ± 0 . 11 , n = 17 frames , Bcl2: 1 . 39 ± 0 . 14 , n = 31 frames , p = 0 . 0001 Bcl2 vs GFP; Figure 3H , I ) . A closer examination of the data identified a unique population of dendritic branches with low spine density ( <1 . 5 spines/µm ) in the Bcl2-expressing group but not in samples expressing GFP or Cre alone , although the cumulative fractions of spine counts in these groups were not statistically different ( p > 0 . 1 , Kolmogorov–Smirnov test; Figure 3J ) . However , neither the mushroom spine density ( GFP: 0 . 015 ± 0 . 006 , cre: 0 . 013 ± 0 . 004 , Bcl2: 0 . 017 ± 0 . 006 ) nor the percentage of mushroom spines ( GFP: 0 . 80 ± 0 . 27 , cre: 1 . 06 ± 0 . 30 , Bcl2: 1 . 08 ± 0 . 34 ) relative to total number of protrusions showed any difference between samples in C57Bl/6 mice ( Figure 3K , L ) . These results suggest that WT adult-born GCs selected for death had fewer spines , specifically fewer thin spines ( p < 0 . 05; Figure 3L ) , as compared to those destined for survival . In Grin1f/f mice , deletion of NMDAR in newborn GCs led to a decrease in total spine density , and the spine density was even lower in Bcl2-rescued cells ( GFP: 1 . 96 ± 0 . 06 , n = 68 frames , cre: 1 . 47 ± 0 . 11 , n = 35 frames , Bcl2: 1 . 06 ± 0 . 08 , n = 45 frames , p = 0 . 0054 cre vs Bcl2; Figure 3M , N ) , whereas the cumulative distribution of total spine density did not exhibit significant differences between each pair ( p > 0 . 09 , Kolmogorov–Smirnov test; Figure 3O ) . We noted that , in comparison with WT GFP-expressing cells , Bcl2-expressing GCs did not show dramatically increased density ( GFP: 0 . 028 ± 0 . 004 , cre: 0 . 053 ± 0 . 011 , Bcl2: 0 . 032 ± 0 . 007 , p = 0 . 0092 cre vs GFP ) or percentage ( GFP: 1 . 40 ± 0 . 21 , cre: 4 . 76 ± 0 . 66 , Bcl2:2 . 13 ± 0 . 39 , p < 0 . 0001 cre vs GFP ) of mushroom-like spines as naturally surviving NMDAR KO cells ( Figure 3P–Q ) . However , Bcl2-rescued group displayed a strong trend of mushroom spine enhancement in Grin1f/f mice ( Bcl2 vs GFP: p = 0 . 08 , Figure 3Q ) , but not in WT mice ( Bcl2 vs GFP: p = 0 . 5 , Figure 3L ) , suggesting that NR1 KO GCs accounted for a portion of neurons rescued by Bcl2 and they had more mushroom spines than those dying of reasons other than genetic ablation of NMDARs . Taken together , these data support that NMDAR was critical for spine outgrowth and that newborn GCs destined to die without the protection of Bcl2 were defective in spine formation . Since Bcl2-rescued GCs exhibited a total spine reduction , but not a mushroom spine enhancement ( Figure 3H–Q ) , it seems unlikely that the increases in spine volume and AMPA currents associated with NMDAR loss ( Figure 1I–L ) simply represent a homeostatic change in response to the decrease of total synaptic drive . However , two scenarios may explain the increase in mushroom spines found in surviving NR1 KO cells ( Figure 1 ) . The alteration of spine head diameter could be a direct consequence of NMDAR loss . Alternatively , the survival selection was biased towards neurons that happened to bear lots of mushroom spines or mature synapses . To test these possibilities , we developed a cre-ER retrovirus in which the expression of GFP and the inducible Cre recombinase creERt2 was bridged by the 2A sequence from Picornavirus ( Szymczak et al . , 2004 ) ( rv CGS-creER; Figure 4—figure supplement 1 ) to induce NMDAR deletion in newborn GCs after their critical time window for survival . We injected rv CGS-creER or control rv CAG-GFP into Grin1f/f and Grin1f/+ mice , respectively , and injected oil or tamoxifen ( 180 mg/kg , daily for 4 days ) into the animals at 28 dpi . Mouse brains were collected at 56 dpi ( 28 days after the induction with tamoxifen ) for morphological analyses . We first quantified GFP+ cell number from Grin1f/f mice injected with rv CGS-creER and found no significant difference between oil and tamoxifen treatment ( oil: 84 ± 48 , n = 4 mice , tamo: 117 ± 15 , n = 4 mice , p = 0 . 54; Figure 4A–C ) . Therefore , NMDAR activity was not required for the survival of newborn GCs that were 4 weeks old or older . Although there was no statistically significant decrease in total spine density in NMDAR KO cells ( oil/creER: 2 . 20 ± 0 . 10 , n = 20 frames , tamo/creER: 1 . 91 ± 0 . 12 , n = 14 frames , tamo/GFP: 2 . 29 ± 0 . 11 , n = 21 frames; Figure 4D , E ) , there appeared to be a strong trend of increased mushroom spine density ( oil/creER: 0 . 133 ± 0 . 016 , tamo/creER: 0 . 188 ± 0 . 027 , tamo/GFP: 0 . 131 ± 0 . 015 , p = 0 . 054 comparing tamo/creER with tamo/GFP; Figure 4D , F ) . In particular , the percentage of mushroom-shaped spines was markedly increased by 62% ( oil/creER: 6 . 38 ± 0 . 91 , tamo/creER: 10 . 35 ± 1 . 68 , tamo/GFP: 6 . 17 ± 0 . 92 , p = 0 . 024 comparing tamo/creER with tamo/GFP; Figure 4G ) . These data indicate that enhanced spine maturation was unlikely to be a compensatory effect from the stress of NMDAR-dependent cell survival . 10 . 7554/eLife . 07871 . 008Figure 4 . Using inducible cre to bypass the critical NMDAR-dependent cell survival . ( A , B ) Representative images of newborn GCs in Grin1f/f mice that were infused with rv CAG GFP-t2A-creER and treated with oil ( A ) or tamoxifen ( B ) . ( C ) Deletion of the Grin1 gene initiated in 4-week-old newborn GCs did not affect cell survival . ( D ) Representative images of dendritic processes in the outer molecular layer of rv GFP-t2A-creER-targeted newborn GCs treated with oil and tamoxifen , and of rv GFP-targeted newborn GCs treated with tamoxifen . ( E–G ) Quantification of total spine density ( E ) , mushroom spine density ( F ) and mushroom spine percentage ( G ) in rv GFP-t2A-creER- and rv GFP-targeted newborn GCs ( *p < 0 . 05 ) . Scale bars: 50 µm ( A , B ) and 5 µm ( D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07871 . 00810 . 7554/eLife . 07871 . 009Figure 4—figure supplement 1 . The recombination efficiency of rv CAG GFP-t2A-creER was tested in the ROSA-lacZ mice . ROSA-lacZ reporter mice received stereotaxic delivery of retrovirus into the DG and subsequently tamoxifen daily for 5 days starting at 2 weeks after virus injection . Reporter gene expression was analyzed at 2 weeks after the first dose of tamoxifen . Insets on the right side represent GFP− and β-gal-expressing cells , respectively . Scale bar: 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 07871 . 009 Since NR1 KO newborn GCs display increased mushroom spine density at both immature and relatively mature stages during GC development , we next examined whether NMDARs functioned similarly in GCs generated during embryonic development . To delete NR1 in mature GCs , we engineered a lentivirus CAG-GFP-ires-cre ( lv GFP-ires-cre; Figure 5—figure supplement 1 ) and injected the lv GFP-ires-cre virus or a control lv GFP virus into 8-week-old Grin1f/+ and Grin1f/f mice . Lv-transduced GCs were analyzed at 28 dpi . A brief visual inspection of the sections revealed no obvious difference in the number of lv GFP-ires-cre-transduced cells in f/+ and f/f cells , consistent with the previous observation that NMDAR was not required for the survival of mature GCs ( Figure 5A , B ) . Electrophysiological recordings showed that Cre-targeted cells did not respond to perforant path stimulation in the presence of the AMPAR blocker DNQX , confirming that Cre-expressing cells had no functional NMDARs ( Figure 5C , D ) . There was no difference in total spine density between WT and NMDAR KO cells ( Grin1f/+ GFP: 2 . 97 ± 0 . 14 , n = 29 frames , Grin1f/+ cre: 3 . 09 ± 0 . 18 , n = 28 frames , Grin1f/f GFP: 3 . 15 ± 0 . 10 , n = 71 frames , Grin1f/f cre: 2 . 97 ± 0 . 08 , n = 75 frames; Figure 5E , F ) . However , mature NMDAR KO GCs displayed significantly more mushroom spines than any other group ( Grin1f/+ 0 . 128 ± 0 . 018 , Grin1f/+ cre: 0 . 089 ± 0 . 013 , Grin1f/f GFP: 0 . 117 ± 0 . 009 , Grin1f/f cre: 0 . 193 ± 0 . 010 , p < 0 . 0001; Figure 5G ) . Correspondingly , both the amplitude and the frequency of AMPAR-mediated mEPSCs were increased in NMDAR KO mature cells ( WT: n = 7 cells , KO: n = 6 cells , p < 0 . 01 , Kolmogorov–Smirnov test; Figure 5H–K ) . These data demonstrated that spine maturation and synaptic AMPAR activity were enhanced in mature GCs in the absence of NMDAR . 10 . 7554/eLife . 07871 . 010Figure 5 . The effect of NR1 KO in mature GCs . ( A , B ) Representative images of mature GCs in Grin1f/+ ( A ) and Grin1f/f ( B ) mice targeted by lv CAG-GFP-ires-cre . ( C , D ) Electrophysiological recordings of mature GCs in lv CAG-GFP-ires-cre-targeted mice . GFP− and GFP+ cells represent NR1 WT and KO GCs , respectively . ( E ) Representative images of dendritic processes in the outer molecular layer of GFP+ cells targeted by lv CAG-GFP and GFP-ires-cre . ( F ) NR1 KO mature GCs display similar total spine density as wild type GCs . ( G ) Mushroom spine density was increased in NR1 KO mature GCs ( *p < 0 . 0001 ) . ( H , I ) Sample traces of AMPAR-mediated mEPSCs in GFP− and GFP+ mature GCs in Grin1f/f mice targeted by lv CAG GFP-ires-cre . ( J , K ) Cumulative plots of mEPSC amplitude ( J ) and frequency ( K ) confirmed that AMPAR-mediated activity was enhanced in NR1 KO mature GCs . Scale bars: 100 µm ( A , B ) and 5 µm ( E ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07871 . 01010 . 7554/eLife . 07871 . 011Figure 5—figure supplement 1 . The recombination efficiency of lv CAG GFP-ires-cre was tested in ROSA-lacZ reporter mice 4 weeks after virus delivery . Insets on the right side represent GFP− and β-gal-expressing cells , respectively . Scale bar: 100 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 07871 . 011 Although adult-born GCs only constitute a small population of GCs in the DG and most lentivirus-labeled cells should be mature GCs , we could not completely rule out the possibility that our conclusions about NMDAR-dependent inhibition on spine maturation and AMPAR activity in mature GCs might be confounded by the small population of newborn cells possibly included in our analyses . To resolve this concern , and also to determine whether NMDAR-mediated inhibition of spine maturation was specific to GCs , we examined lv GFP-ires-cre-targeted CA1 pyramidal cells in Grin1f/f mice at 28 dpi ( Figure 6A , B ) . Since no neurogenesis occurs in the CA1 area , the pyramidal cells we examined in this area should be exclusively mature cells . Images of the apical and basal dendrites of CA1 pyramidal cells were obtained from stratum lacunosum-moleculare and stratum oriens , respectively , and were analyzed separately . NMDAR KO pyramidal cells displayed a decreased total spine density in the apical dendrites but not the basal dendrites ( apical GFP: 2 . 63 ± 0 . 12 , n = 38 frames , cre: 2 . 12 ± 0 . 17 , n = 15 frames , p = 0 . 023 , basal GFP: 2 . 59 ± 0 . 11 , n = 28 frames , cre: 2 . 44 ± 0 . 15 , n = 15 frames , p = 0 . 44; Figure 6C ) . Hippocampal CA1 pyramidal neurons have distinct compartments with differential inputs , plasticity characteristics and mechanisms crucial for integrative function ( Spruston , 2008 ) . It is possible that NMDAR subunit composition , distribution and associated signaling pathways are different in apical and basal domains . Therefore , the spine turnover rate may be low and not sufficient to amount to a visible difference in total spine number in basal but not apical dendritic branches . For example , 96% of spines in the adult mouse visual cortex remained stable through weeks of live imaging ( Grutzendler et al . , 2002 ) . Consistent with the data in lentivirus-transduced GCs , mushroom spine density was significantly higher in both apical and basal dendrites of NMDAR KO pyramidal cells ( apical GFP: 0 . 105 ± 0 . 012 , cre: 0 . 192 ± 0 . 034 , p = 0 . 0037 , basal GFP: 0 . 050 ± 0 . 012 , cre: 0 . 125 ± 0 . 030 , p = 0 . 0077; Figure 6D ) . Therefore , spine maturation was also enhanced in CA1 pyramidal cells in the absence of NMDAR . 10 . 7554/eLife . 07871 . 012Figure 6 . Mushroom spine density was increased in CA1 pyramidal cells in response to NR1 deletion in adult mice . ( A ) Representative image of CA1 pyramidal cells labeled by lv GFP-ires-cre . Dotted boxes indicate typical apical and basal dendrite segments for spine analysis . ( B ) Representative images of apical ( A ) and basal ( B ) dendrites in wild type and NR1 KO CA1 pyramidal cells . ( C ) Total spine density was decreased in the apical but not basal dendrites in CA1 pyramidal cells in response to NR1 deletion ( *p < 0 . 05 ) . ( D ) Mushroom spine density was increased in both apical and basal dendrites in NR1 KO pyramidal cells ( *p < 0 . 01 ) . Scale bars: 50 µm ( A ) and 2 µm ( B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07871 . 012
Using retro- and lentiviral vectors that encode the Cre recombinase , we deleted the Grin1 gene in newborn GCs or mature neurons ( GCs and CA1 pyramidal cells ) to examine the impact of NMDAR loss on spine morphology . We found that NMDAR activity was required for initial spinogenesis and that a low level of spine density correlated with a defect in cell survival during the 3–4 weeks after new GCs were born , although NMDAR signaling did not appear to have a major impact on total spine numbers in mature neurons , whether they were neonate- or adult-generated . However , NMDAR was not critical for the maturation of dendritic spines once spines were formed . On the contrary , the density of mushroom spines was increased in NR1 KO GCs , accompanied by increased synaptic AMPAR activity . The latter phenomenon was observed not only in newborn GCs but also in mature GCs and CA1 pyramidal cells . Taken together , our data suggest that NMDARs serve to regulate the genesis of dendritic spines in developing neurons . Furthermore , NMDARs restrict synapse maturation and control spine morphology after spines have been formed , independently of the developmental stage . We reported previously that NMDAR was required for the survival of newborn GCs in the adult mouse hippocampus ( Tashiro et al . , 2006a ) . Through detailed analyses , we found that NMDAR KO newborn GCs that were forced to survive with the expression of the pro-survival gene Bcl2 had defects in spine growth and synaptogenesis . Because the time window of initial spine growth and synapse formation correlated with that of NMDAR-dependent cell survival , we deduced from these observations that the failure of synapse formation may be the underlying mechanism of cell death in NMDAR KO newborn GCs . In addition , we analyzed newborn GCs forced to survive through Bcl2 expression in WT mice and found that the overall population had lower total spine density compared to the cells that were selected to survive under natural conditions . This finding suggests that natural death of WT adult-born GCs might also be a result of the failure of spine formation in certain cell populations . Both voluntary exercise and enriched environment ( EE ) housing increased cell survival ( van Praag et al . , 1999; Tashiro et al . , 2007; Muotri et al . , 2009; Zhao et al . , 2014 ) . However , unlike the forced survival through Bcl2 expression , increased survival of GCs through physical exercise or EE did not lower the average spine density of newborn GCs ( Zhao et al . , 2014 ) , suggesting that exercise and EE likely promote cell survival through enhancing spine formation in newborn GCs . Many of the functional properties of NMDARs are highly sensitive to their subunit composition . Early expression of NR2A in organotypic hippocampal slices reduces the number of synapses and the volume and dynamics of spines , whereas overexpression of NR2B increases spine motility , indicating that the ratio of NR2B over NR2A controls spine motility and synaptogenesis ( Gambrill and Barria , 2011 ) . Furthermore , overexpression of NR3A reduces spine density over time by increasing spine elimination and decreasing spine stability ( Kehoe et al . , 2014 ) . In adult-born GCs , there is a switch in synaptic NMDAR subunit composition from predominantly NR2B to NR2A during their development ( Ge et al . , 2007 ) . Although the expression pattern of NR3A is unclear in nascent GCs , it is reasonable to speculate that NMDAR may be required for spine stabilization in addition to spine formation , depending on developmentally regulated synaptic expression of NMDARs containing a specific type of subunit . Alternatively , NMDARs are located not only at synapses but also at extrasynaptic sites . The roles played by extrasynaptic NMDARs are generally still elusive . It has been shown that filopodia-like protrusions or spines appear de novo after exogenous glutamate application or LTP induction ( Engert and Bonhoeffer , 1999; Maletic-Savatic et al . , 1999 ) , raising a possibility that extrasynaptic NMDARs participate in the regulation of new spine formation . Given that newborn GCs display high levels of NMDARs before the formation of glutamatergic synapses ( Schmidt-Salzmann et al . , 2014 ) , it is possible that the distinct roles of NMDARs during different stages of GC development may be attributed to different locations of NMDARs . Further study is needed to find out whether NMDAR location or subunit composition , or both , could be the determining factor of NMDAR's functional diversity . NR1 deletion leads to decreased numbers of total spines and increased mushroom spines in developing cortical neurons ( Ultanir et al . , 2007 ) . Here we revealed a very similar impact of NR1 loss on neurons born in the adult hippocampus and have provided additional evidence that adult neurogenesis follows a pattern resembling early development . However , the observation that NMDAR KO newborn GCs , irrespective of their age , displayed higher mushroom spine density and more synaptic AMPAR activity was unexpected . It contradicted the notion that NMDAR activity is required for spine enlargement and AMPAR recruitment ( Matsuzaki et al . , 2004; Wang and Kriegstein , 2008 ) and cannot be interpreted simply by differences between developing and mature synapses as previously speculated ( Ultanir et al . , 2007 ) . Several possible mechanisms could account for enhanced spine maturation in the absence of NMDAR . First , the surviving cell population could be selected to survive because those cells have more mushroom spines or functional synapses . This possibility was ruled out because we observed a similar phenomenon in newborn GCs when NR1 was deleted at a later stage , after the NMDAR-dependent critical time window for cell survival . Furthermore , mature GCs and CA1 pyramidal cells also displayed increased mushroom spine density and synaptic activity in the absence of NMDAR , supporting our conclusion that enhanced synapse maturation is not a compensatory effect of NMDAR-dependent cell survival in newborn GCs . Alternatively , elevated mushroom spines could be potentially explained by a synaptic scaling mechanism , namely , NR1 KO cells might increase AMPAR activity in compensation for the loss of NMDAR activity . Indeed , we observed that AMPAR activity was significantly enhanced in both frequency and amplitude in the absence of the functional NMDAR . However , it has been reported that NMDAR activity itself is required for synaptic scaling ( Turrigiano et al . , 1998; Wang et al . , 2011 ) . Therefore , it is unlikely that enhanced spine maturation was a result of synaptic scaling . In an effort to resolve the discrepancy between our current observation and the traditional notion that NMDAR activity is required for AMPAR recruitment and spine enlargement , we tried to compare the systems used in our studies and others . Among the many differences between study systems , we found that one difference could potentially account for the different observations . In studies that indicated an inhibitory role for NMDAR on spine maturation and AMPAR recruitment , NMDAR activity was lost due to the deletion of the Grin1 gene ( Ultanir et al . , 2007; Adesnik et al . , 2008 ) , whereas in studies that supported NMDAR-dependent spine enlargement and AMPAR recruitment , NMDAR activity was blocked by antagonists ( Engert and Bonhoeffer , 1999; Maletic-Savatic et al . , 1999; Zhu et al . , 2000 ) . It was previously pointed out that a genetic approach might offer certain advantages over pharmacological manipulations ( Ultanir et al . , 2007 ) . Here we propose that NMDAR in its inactive form inhibits spine maturation and AMPAR recruitment . The activation of NMDAR by binding of the neurotransmitter glutamate reverses its inhibitory function and allows the recruitment of AMPAR and spine enlargement . When NMDAR activity was blocked by antagonists , NMDAR was kept in its inactive state and thus prevented spine enlargement and AMPAR recruitment . However , when NMDAR was absent due to the loss of its essential subunit NR1 , there was no longer inhibition from the inactive NMDAR , which resulted in uncontrolled spine enlargement and AMPAR recruitment . This hypothesis is in line with the critical role of NMDAR in LTP and reconciles the discrepancies observed when different methods were used to modulate NMDAR activity . Another possibility—not mutually exclusive—is that lack of NMDARs triggers expression of calcium-permeable AMPARs in newborn neurons and these receptors confer upon dendritic spines the ability to potentiate and to grow in the absence of NMDARs; this hypothesis awaits experimental clarification . Experience constructs the neural network in the form of activity-dependent spine morphogenesis , which is central to memory formation and other adaptive changes of the brain . Although the current study has been mainly focused on newborn neurons in the adult brain , our findings suggest that the major neuronal activity detector , NMDAR , regulates the connectivity of specific neurons by common mechanisms in developing and mature nervous systems , including facilitation of new spine formation and control of the pace of existing spine maturation . By preventing precocious synapse maturation , NMDAR participates in ‘neoteny’ , or the extension of the immature state of the brain , which is critical for subsequent and more complete maturation .
Grin1 floxed mice ( Tsien et al . , 1996 ) and the ROSA-lacZ reporter mice ( B6;129S4-Gt ( ROSA ) 26Sortm1sho ) were maintained as homozygous in the Salk mouse facility . For some of the experiments , Grin1f/f mice were bred to Grin1f/+ mice so that littermate Grin1f/+ mice were used as controls . Mice aged 5–7 weeks were used to examine the role of NR1 in newborn GCs , and those 8 weeks old or older were used to examine the role of NR1 in mature neurons . C57Bl/6 female mice , 6 weeks old , were used to test the retrovirus GFPBcl2-ires-cre . The animal protocols were all approved by the Salk Institutional Animal Care and Use Committee . To examine stage-specific roles of the NMDAR , we developed four new cre-expressing viruses . Firstly , retrovirus CAG-GFP ires cre ( rv GFP-ires-cre ) was constructed for the purpose of gene deletion and the simultaneous visualization of the newborn cell morphology ( Figure 1—figure supplement 1 , 28 dpi ) . In the rv GFP-ires-cre vector , the Cre cDNA was placed after the internal ribosomal entry site to minimize the expression of Cre and potentially reduce cre-associated vector recombination in bacteria . When tested in the ROSA-lacZ reporter mice , rv GFP-ires-cre induced recombination in 97% of GFP+ cells at 6 dpi and in all GFP+ cells at 14 and 28 dpi . Secondly , a retrovirus CAG-GFPBcl2-ires-cre ( rv GFPBcl2-ires-cre ) was constructed so that the Cre-targeted cells also expressed the fusion protein of GFP and Bcl2 ( Figure 3—figure supplement 1 , 14 dpi ) . This virus allowed the deletion of the Grin1 gene in newborn GCs and the expression of the pro-survival protein Bcl2 in the same cells so that NR1 KO cells would be prevented from going through apoptosis . The recombination efficiency of rv GFPBcl2-ires-cre was 100% at 7 dpi . Thirdly , an inducible retrovirus CAG-GFP-t2A-creER ( rv GFP-t2A-creER ) was generated so that deletion of the Grin1 gene could be initiated at the desired age to bypass NR1-depedent cell survival ( Figure 4—figure supplement 1 , induction at 14 dpi ) . To test the recombination efficiency of rv GFP-t2A-creER virus in vivo , we gave tamoxifen ( 180 mg/kg , daily for 4–5 days ) to virus-injected ROSA26-lacZ reporter mice starting from 14 and 28 days after virus injection . At 14 dpi , we observed 16% background recombination with oil control and 85% recombination in GFP+ cells . At 28 dpi , recombination efficiency was 97% . Lastly , a Cre-expressing lentivirus CAG-GFP-ires-cre was developed for deletion of the Grin1 gene in both mature and newborn cells to examine the function of NR1 in mature neurons ( Figure 5—figure supplement 1 , lv GFP-ires-cre , 29 dpi ) . Because mature cells greatly outnumbered immature cells , the Cre-targeted cells largely represented mature cells . The recombination efficiency of the lv CAG-ires-cre in ROSA reporter mice was 85% at 29 dpi . Recombinant retroviruses and lentiviruses were prepared in 293T cells as described before ( Zhao et al . , 2006; Tashiro et al . , 2006b ) . Mouse brain sections of 40 µM thickness were prepared with a sliding microtome as described in detail ( Zhao et al . , 2006 ) . Brain sections of one-in-six series were selected for DAPI staining . GFP+ cells in the GC layer were visualized and counted manually with a Nikon E800 microscope ( Melville , NY , United States ) . The total number of labeled GFP+ cells per DG was then estimated by multiplying the number by 6 . All images were acquired through the Bio-Rad R2100 confocal system ( Berkeley , CA , United States ) or the Zeiss 710/780 confocal system ( Germany ) . Images of the whole cell morphology of GFP+ cells were taken with a 40× objective ( Bio-Rad R2100 ) or a 25× W objective ( Zeiss 710/780 ) . GFP+ cells with at least one dendritic process terminating at the outer molecular layer were randomly picked for imaging ( every nth cell depending on the labeling efficiency ) . If the number of labeled cells from the one-in-six series was too low to allow for 5–10 cells to be imaged , more sections were sampled to obtain enough cells from an individual mouse or until all sections were used . For spine analyses , dendritic processes of GFP+ cells in the outer molecular layer were imaged with a 60× oil objective ( NA 1 . 4 , Nikon , on Bio-Rad R2100 ) or with a 63× oil objective ( NA 1 . 4 , Zeiss , on Zeiss 710 ) . The raw confocal image files were subjected to 10 iterations of deconvolution ( AutoDeblur , AutoQuant , Troy , NY , United States ) . Dendrite measuring and spine analyses have been described in detail before ( Zhao et al . , 2006 ) . The categorization of dendritic spine shape was based on qualitative criteria ( Harris et al . , 1992; Parnass et al . , 2000 ) . For classification of mushroom spines , major and minor axes of each spine head were measured with NeuronStudio program . A spine was judged to be of mushroom type if the head area ( estimated with the function ¼ × π × Dmajor × Dminor ) was ≥0 . 4 μm2 . The absolute numbers of each spine type for the NR1 KO experiment ( Figure 1G ) and the Bcl2 rescue experiment ( Figure 3L , Q ) are summarized in Tables 1–3 . 10 . 7554/eLife . 07871 . 013Table 1 . Total number of spines evaluated in Figure 1GDOI: http://dx . doi . org/10 . 7554/eLife . 07871 . 013Spine classesWTKOStubby742391Mushroom4497Thin38841643Filopodia20212110 . 7554/eLife . 07871 . 014Table 2 . Total number of spines evaluated in Figure 3LDOI: http://dx . doi . org/10 . 7554/eLife . 07871 . 014Spine classesGFPcreBcl2Stubby306164248Mushroom231224Thin254516351835Filopodia655910410 . 7554/eLife . 07871 . 015Table 3 . Total number of spines evaluated in Figure 3QDOI: http://dx . doi . org/10 . 7554/eLife . 07871 . 015Spine classesGFPcreBcl2Stubby572208222Mushroom10110860Thin690119842342Filopodia16078164 Electrophysiological recordings of NR1 WT and KO cells were performed using a protocol that was described in detail in a previous study ( Mu et al . , 2011 ) . Specifically , mice injected with retrovirus- or lentivirus-expressing GFP-ires-cre were anesthetized by isoflurane inhalation . Mouse brains were immediately removed and placed in ice-cold dissection buffer ( in mM choline chloride 110 , KCl 2 . 5 , NaH2PO4 1 . 3 , NaHCO3 25 . 0 , CaCl2 0 . 5 , MgCl2 7 , glucose 20 , Na-ascorbate 1 . 3 , and Na-pyruvate 0 . 6 ) . Horizontal slices ( 200 μm thick ) were prepared using a Leica VT1000S vibrotome ( Germany ) and incubated for at least 1 hr at room temperature before recording in standard ACSF ( in mM NaCl 125 , KCl 2 . 5 , NaH2PO4 1 . 3 , NaHCO3 25 , CaCl2 2 , MgCl2 1 . 3 , Na-ascorbate 1 . 3 , Na-pyruvate 0 . 6 , and glucose 10 ) that was saturated with 95% O2 and 5% CO2 . Whole-cell perforated patch recordings were obtained from GCs visualized using an upright microscope ( BX51WI; Olympus ) with infrared differential interference contrast optics ( Japan ) . The Cre-targeted cells were visually identified by their green fluorescence . The following drugs were used for blocking certain activities: 50–100 µM picrotoxin to block GABAergic synaptic transmission , 10 µM DNQX to block AMPAR-mediated activity , 25 µM APV to clock NMDAR-mediated activity and 1 µM TTX to block action potentials . All experiments were performed at room temperature . A bipolar tungsten electrode was used for extracellular stimulation of the perforant path , and GCs were held at −70 mV in voltage-clamp mode unless stated otherwise . Tissue processing for electron microscopy was performed as described previously ( Toni et al . , 2008 ) . Briefly , mice were transcardially perfused with 4% paraformaldehyde and brains were cut coronally at a thickness of 50 μm . Sections were cryoprotected , briefly freeze-thawed four times in liquid nitrogen and treated with 0 . 3% hydrogen peroxide . After a block with 0 . 5% bovine serum albumin , slices were incubated 40 hr with rat anti-GFP antibody ( 1:500 , Chemicon ) at 4°C on a shaker . After washing , sections were incubated for 5 hr at 5°C in biotinylated secondary antibody ( goat anti-rabbit , Fac fragment , 1:500 , Chemicon ) . Sections were then incubated in avidin biotin peroxidase complex ( ABC Elite , Vector laboratories ) , followed by 3 , 3′-diaminobenzidine tetrachloride for 10–20 min to obtain a dark residue in labeled cells . Sections were next postfixed in 2 . 5% glutaraldehyde , followed by 1% osmium tetroxide , dehydrated in ascending concentrations of ethanol and then acetone , and finally embedded in Epoxy resin . Sections of a thickness of 50 nm were contrasted with uranyl acetate followed by lead citrate and observed on a Philips CM10 electron microscope ( Hillsboro , OR , United States ) , at a magnification of 13 , 500× . Synapses were defined by the presence of at least three presynaptic vesicles within 50 nm of the presynaptic membrane , a clearly defined synaptic cleft and a postsynaptic density . Spines were serially sectioned and the surface area of each segment was measured on every image . Volumes were obtained by multiplying the surface area , the section thickness and the number of sections . All data were presented as mean ± standard error . Statistic comparisons were done using unpaired two-tailed t-test , except that the Kolmogorov–Smirnov test was used for data analyses on AMPAR mEPSCs . | The brain contains billions of cells called neurons . Although most neurons have already formed by the time we are born , part of the brain called the hippocampus produces new neurons throughout our life . These new neurons are thought to be important for learning and forming new memories . Neurons send signals to each other across connections called synapses . Small protrusions called spines stick out of the neuron and each tends to have one synapse that receives a signal from another neuron . Via these connections , the neurons are organized into networks and circuits that control how different parts of the brain work . Therefore , once new neurons are made , they also need to be connected to the correct neurons . The NMDA receptor is found in the surface of neurons , and mutated neurons that lack this receptor often die shortly after birth . The NMDA receptor is also known to be important for organizing synapses . Exactly how NMDA receptors help new neurons to survive and integrate into circuits has not been investigated in detail . Mu , Zhao et al . now address this issue by using mice in which a gene called NR1 , which produces one of the proteins that makes up the NMDA receptor , can be deleted at specific stages of neuron development . Analyzing brain slices from the mice showed that deleting NR1 from newly-formed neurons caused them to die within two or three weeks . When these neurons were forced to survive , they had fewer spines than normal . By contrast , deleting NR1 from neurons that has already survived for longer than four weeks did not alter how many spines the neurons had . Instead , the synapses on the spines worked better . Mu , Zhao et al . therefore suggest that NMDA receptors have different roles at different stages of a neuron's development . Initially , NMDA receptors help the neurons to survive and form spines . The receptors then help to ensure the spines become the correct size , and enable the neurons to connect into the right neural circuits by helping to control the strength of synapses . Mu , Zhao et al . theorize that the mere presence of NMDA receptors suppresses spine maturation . Furthermore , this inhibitory effect is only released when the NMDA receptor is activated , or when the NMDA receptor is absent due to the deletion of the NR1 gene . Further studies will be needed to validate this hypothesis . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2015 | Distinct roles of NMDA receptors at different stages of granule cell development in the adult brain |
Humans and other animals often violate economic principles when choosing between multiple alternatives , but the underlying neurocognitive mechanisms remain elusive . A robust finding is that adding a third option can alter the relative preference for the original alternatives , but studies disagree on whether the third option’s value decreases or increases accuracy . To shed light on this controversy , we used and extended the paradigm of one study reporting a positive effect . However , our four experiments with 147 human participants and a reanalysis of the original data revealed that the positive effect is neither replicable nor reproducible . In contrast , our behavioral and eye-tracking results are best explained by assuming that the third option’s value captures attention and thereby impedes accuracy . We propose a computational model that accounts for the complex interplay of value , attention , and choice . Our theory explains how choice sets and environments influence the neurocognitive processes of multi-alternative decision making .
In recent years , studying choices between multiple ( i . e . , more than two ) alternatives has attracted growing interest in many research areas including biology , neuroscience , psychology , and economics ( Berkowitsch et al . , 2014; Chau et al . , 2014; Chung et al . , 2017; Cohen and Santos , 2017; Gluth et al . , 2017; Hunt et al . , 2014; Landry and Webb , 2017; Lea and Ryan , 2015; Louie et al . , 2013; Mohr et al . , 2017; Spektor et al . , 2018a; Spektor et al . , 2018b ) . In such choice settings humans and other animals often violate the independence from irrelevant alternatives ( IIA ) principle of classical economic decision theory ( Rieskamp et al . , 2006 ) . This principle states that the relative preference for two options must not depend on a third ( or any other ) option in the choice set ( Luce , 1959; Marschak and Roy , 1954 ) . Violations of IIA have profound implications for our understanding of the neural and cognitive principles of decision making . For example , if choice options are not independent from each other , we cannot assume that the brain first calculates each option’s value separately before selecting the option with the highest ( neural ) value ( Vlaev et al . , 2011 ) . Instead , dynamic models in which option comparison and decision formation processes are intertwined appear more promising as they account for several violations of IIA ( Gluth et al . , 2017; Hunt and Hayden , 2017; Hunt et al . , 2014; Mohr et al . , 2017; Roe et al . , 2001; Trueblood et al . , 2014; Tsetsos et al . , 2012; Usher and McClelland , 2004 ) . The underling mechanisms of multi-alternative choice and the conditions under which decision makers exhibit different forms of violations of IIA are a matter of current debate ( Chau et al . , 2014; Frederick et al . , 2014; Louie et al . , 2013; Spektor et al . , 2018a; Spektor et al . , 2018b ) . Two recent studies by Louie and colleagues ( Louie et al . , 2013; henceforth Louie2013 ) and by Chau and colleagues ( Chau et al . , 2014; henceforth Chau2014 ) investigated whether and how the value of a third option influences the relative choice accuracy between two other options ( i . e . , the probability of selecting the option with the higher value ) . Both studies reported violations of IIA which , however , contradicted each other: Louie2013 found a negative relationship , so that better third options decreased the relative choice accuracy ( i . e . , the probability of choosing the better of the two original options ) . They attributed this to divisive normalization of ( integrated ) value representations in the brain: To keep value coding by neural firing rates in a feasible range , each value code could be divided by the sum of all values . Thus , a third option of higher value implies a higher division of firing rates and reduces the neural discriminability between the other two options . In contrast , Chau2014 reported a positive relationship , that is , third options of higher values increased relative choice accuracy . They predicted this effect by a biophysical cortical attractor model ( Wang , 2002 ) . Briefly , this model assumes pooled inhibition between competing attractors that represent accumulated evidence for the different options . A third option of higher value would increase the level of inhibition in the model and thus lead to more accurate decisions ( because inhibition reduces the influence of random noise ) . The goal of the present study was to resolve these opposing results by exploiting the multi-attribute nature of Chau2014’s task and extending it with respect to the paradigm and the analysis of the behavioral data . Notably , the paradigms of Louie2013 and Chau2014 differ in several aspects , the most critical being the type of choice options: In Louie2013 , participants decided between different food snacks whereas in Chau2014 , participants chose between rectangles that represented gambles with reward magnitude and probability conveyed by the rectangles’ color and orientation , respectively ( Figure 1A and B ) . Our initial hypothesis for the discrepancy between the studies’ results was that the explicit presentation of the magnitude and probability attributes in the latter induced IIA-violating ‘context effects’ that emerge through multi-attribute comparison processes ( Gluth et al . , 2017; Pettibone and Wedell , 2007; Roe et al . , 2001; Trueblood et al . , 2014; Usher and McClelland , 2004; we elaborate on this and on further hypotheses in the Materials and methods ) . To test this hypothesis , we added a set of novel trials in which the high-value option ( HV ) , the low-value option ( LV ) , and the unavailable distractor ( D ) were positioned in the two-dimensional attribute space in a way to allow a rigorous discrimination between the various multi-attribute context effects , divisive normalization , and the biophysical cortical attractor model ( Figure 1C and D ) . Note that henceforth , by ‘divisive normalization’ we refer to the model proposed by Louie2013 which assumes that normalization occurs at the level of integrated values ( i . e . , after combining attribute values such as magnitude and probabilities into a single value ) . Other models that instantiate hierarchical or attribute-wise normalization can make qualitatively different predictions ( Hunt et al . , 2014; Landry and Webb , 2017; Soltani et al . , 2012 ) . Also , we do not address the role of divisive normalization as a canonical neural computation ( Carandini and Heeger , 2011 ) beyond its conceptualization by Louie2013 . Foreshadowing the results , our data from three behavioral and one eye-tracking experiments with a total of 147 participants are at stark odds with both the findings of Louie2013 and Chau2014 . We did not observe a positive impact of D’s value on relative choice accuracy as reported by Chau2014 in the same task , nor a negative impact of D’s value as predicted by Louie2013 . In other words , violations of IIA did not occur in the Chau2014 paradigm when participants made decisions under time pressure . Furthermore , a reanalysis of the original Chau2014 data suggested that the reported effect is a statistical artifact ( see Materials and methods ) . In contrast to the absence of any effects on relative choice accuracy , however , we consistently found a negative impact of D’s value on absolute choice accuracy ( i . e . , the overall probability of choosing the best option ) , which is not a violation of IIA . We argue that our behavioral and eye-tracking results as well as the results of the original study are best accounted for by value-based attentional capture , that is , by assuming that options capture attention proportional to their value . Value-based attentional capture is a comparatively novel concept in attention research , which has repeatedly been demonstrated in humans and non-human primates ( Anderson , 2016; Anderson et al . , 2011; Grueschow et al . , 2015; Le Pelley et al . , 2016; Peck et al . , 2009; Yasuda et al . , 2012 ) . With respect to the Chau2014 task , it implies that a higher-value D draws more attention and thereby interferes with the choice process . To explain the results from our behavioral and eye-tracking experiments , we integrate the concept of value-based attentional capture into the well-established framework of evidence accumulation in decision making ( Bogacz et al . , 2006; Gluth et al . , 2012; Gluth et al . , 2015; Gold and Shadlen , 2007; Heekeren et al . , 2008; Smith and Ratcliff , 2004 ) . Our model predicts that only absolute but not relative choice accuracy will be affected by the value of the third option ( i . e . , no violation of IIA; Figure 1D ) . It provides a novel and cognitively plausible mechanism of the complex interplay of value and attention on multi-alternative decision making .
Experiment 1 ( with n1 = 31 participants ) aimed at comparing different model predictions with respect to the novel set of trials we introduced ( Figure 1B and C ) and at establishing the core finding of Chau2014 by replicating it . In the decision-making task as introduced by Chau2014 ( Figure 1A ) , two available options of different expected value , HV and LV , are presented in two randomly selected quadrants of the screen and participants are asked to choose the option with the higher expected value . In half of the trials , the remaining two quadrants are left empty , in the other half , a third distractor option , D , is shown in one quadrant but indicated as unavailable after 0 . 1 s . Then , participants have another 1 . 5 s to make their choice . The central behavioral analysis by Chau2014 was a logistic regression of relative choice accuracy ( i . e . , whether HV or LV was chosen while excluding all trials with other responses such as choosing D or the empty quadrant , or being too slow ) on the value difference between the two available options , HV-LV , the sum of their values , HV +LV , the value difference of HV and D , HV-D , the interaction of value differences , ( HV-LV ) × ( HV-D ) , and whether a distractor was present or not , D present . Most importantly , the authors reported a significantly negative regression coefficient of HV-D indicating that higher values of D increased choice accuracy . To test whether our results replicate the findings of Chau2014 , we analyzed decisions made in the ( non-novel ) trials that were identical to those used by Chau2014 . Note that the choice sets in Chau2014 were generated by sampling magnitude and probabilities for HV , LV , and D until HV-LV and HV-D shared less than 25% variance . The options’ average expected values in the resulting trials were 5 . 13 for HV ( SD = 3 . 08; min = 0 . 5; max = 10 . 5 ) , 3 . 51 for LV ( SD = 3 . 02; min = 0 . 25; max = 9 ) , and 4 . 32 for D ( SD = 2 . 11; min = 0 . 25; max = 9 ) . Although the overall choice performance in our data was strikingly similar to the original data ( Table S1 in Supplementary file 1 ) , the negative effect of HV-D on relative choice accuracy could not be replicated . Instead , the average coefficient was positive but not significant ( t ( 30 ) = 1 . 40 , p = . 171 , Cohen’s d = 0 . 25; left panel of Figure 2A; complete results of all regression analyses for all experiments are reported in Tables S2-S5 ) . Interestingly , analyzing absolute choice accuracy ( i . e . , including all trials ) resulted in a significant positive regression coefficient of HV-D ( t ( 30 ) = 5 . 14 , p < . 001 , d = 0 . 92; right panel of Figure 2A ) . This suggests that a higher value of D lowered choice accuracy when all trials were taken into account . To better understand why D’s value has this negative impact on the probability of choosing HV , we analyzed the different possibilities of making errors ( i . e . , choosing LV , choosing D , choosing the empty quadrant , and being too slow ) by testing whether they were predicted by D’s value . The value of D had a significant effect on the probability of choosing D ( t ( 28 ) = 5 . 92 , p < . 001 , d = 1 . 11 ) and also on being too slow ( t ( 29 ) = 2 . 66 , p = . 013 , d = 0 . 49 ) , whereas the probability of choosing LV ( t ( 30 ) = −0 . 47 , p = . 645 , d = −0 . 08 ) or the empty quadrant ( t ( 23 ) = 1 . 07 , p = . 294 , d = 0 . 22 ) were unaffected ( Figure 2B ) . Notably , higher values of D also slowed down response times ( RT ) for HV and LV choices ( t ( 30 ) = 6 . 17 , p < . 001 , d = 1 . 11; Figure 2C ) . We then looked at choice accuracy in the novel trials that we added to differentiate between specific context effects and the models proposed by Louie2013 and Chau2014 . Performance was higher in trials with Ds that were dominated by either HV or LV ( Figure 2C ) , which is the opposite of what is predicted by the Chau2014 model ( Figure 1D ) . This effect of Dominance was significant with respect to both relative choice accuracy ( F ( 1 , 30 ) = 5 . 85; p = . 022 , ηp2 = . 16 ) and absolute choice accuracy ( F ( 1 , 30 ) = 18 . 89; p < . 001 , ηp2 = . 39 ) , the former being consistent with divisive normalization , the latter being consistent with value-based attentional capture ( see Figure 1D; but note that–inconsistent with divisive normalization–we could not replicate the effect on relative choice accuracy in Experiments 2 and 3; Figure 5 , Table S6-S7 for complete results ) . Overall , behavior in the novel trials supported the notion that higher-valued Ds impair multi-alternative decision making . In summary , we could not replicate the results on relative choice accuracy and the IIA violations reported by Chau2014 but obtained opposite effects . At first glance , these results seem to support the divisive normalization account by Louie2013 as an appropriate mechanistic explanation . Note , however , that this account specifically predicts changes in relative choice accuracy , for which we found only weak evidence ( i . e . , no effect of HV-D in the regression , no effect of D’s value on LV-errors ) . On the other hand , the effects on absolute choice accuracy were stronger . Accordingly , we reasoned that the results are better explained by value-based attentional capture: This account states that attention is modulated by value such that value-laden stimuli attract attention and impair goal-directed actions , even if those stimuli are irrelevant to the task ( Anderson , 2016; Anderson et al . , 2011; Grueschow et al . , 2015; Le Pelley et al . , 2016; Peck et al . , 2009; Yasuda et al . , 2012 ) . With respect to Chau2014’s paradigm , value-based attentional capture predicts that high-value Ds draw attention to a greater extent than low-value Ds , which leaves less cognitive resources to focus on the decision between HV and LV . Consequently , decisions slow down , leading to higher RT and to more too-slow errors , and choices of D increase , but the relative probability of choosing HV over LV is unaffected–implying no violation of IIA . In Experiment 2 and 3 , we sought to establish value-based attentional capture as the underlying mechanisms by testing different predictions that this explanation ( but not divisive normalization ) makes . If attentional capture is the driving force behind the performance decrease in the Chau2014 task , then increasing available attentional capacity should improve choice accuracy and reduce the detrimental effects of D . In contrast , divisive normalization effects do not seem to require the imposition of time pressure ( see Louie2013 ) . Based on this rationale , we conducted a second behavioral experiment in which we compared two groups: A high-time pressure ( HP ) Group ( n2 , HP = 25 ) did exactly the same task as in Experiment 1 , but a second , low-time pressure ( LP ) Group ( n2 , LP = 24 ) was given more time to decide ( 6 s instead of 1 . 5 s ) . According to the attentional-capture account , we should replicate the results of Experiment one under high time pressure , but under low time pressure the negative influences of D on ( absolute ) choice accuracy should disappear . For the novel trial sets , we expected a choice accuracy pattern in line with multi-attribute context effects ( see Figure 1D ) , given that these effects are known to become more prominent with longer deliberation time ( Dhar et al . , 2000; Pettibone , 2012; Trueblood et al . , 2014 ) . Consistent with Experiment 1 , we could not replicate the HV-D effect on relative choice accuracy reported in Chau2014 . Again , our results showed a tendency in the opposite direction ( t ( 24 ) = 0 . 50 , p = . 622 , d = 0 . 10 ) . With respect to absolute choice accuracy , there was a strong and significantly positive effect of HV-D in Group HP of Experiment 2 ( t ( 24 ) = 5 . 05 , p < . 001 , d = 1 . 01 ) . In contrast and as predicted , this effect was much weaker and did not reach significance in Group LP ( t ( 23 ) = 1 . 85 , p = . 079 , d = 0 . 38; but note that the difference between groups was not significant: t ( 47 ) = 0 . 61 , p = . 544 , d = 0 . 17; Figure 3A ) . Similarly , only participants of Group HP were less accurate in trials with D present compared to trials without D ( Group HP: t ( 24 ) = -5 . 97 , p < . 001 , d = -1 . 19; Group LP: t ( 23 ) = -0 . 06 , p = . 956 , d = -0 . 01; group difference: t ( 47 ) = -4 . 07 , p < . 001 , d = -1 . 16; Figure 3A ) . Choices of D were significantly more frequent in Group HP ( 5 . 8% vs . 0 . 9% of trials; t ( 47 ) = 4 . 15 , p < . 001 , d = 1 . 20; Figure 3B ) . With respect to the novel trial set , we could not replicate the main effect of Dominance on relative choice accuracy in Group HP that we had found in Experiment 1 ( F ( 1 , 24 ) = 0 . 25; p = . 622 , ηp2= . 01; left panel of Figure 3C ) . In Group LP , there was a significant main effect of Similarity ( F ( 1 , 23 ) = 9 . 20; p = . 011 , ηp2 = . 25; right panel of Figure 3C ) that was absent in any of our experiments conducted under high time pressure . This ( IIA-violating ) effect was due to a higher relative choice accuracy when D was more similar to HV than to LV in the two-dimensional attribute space and is consistent with a combination of an attraction effect ( Huber et al . , 1982 ) and a phantom-decoy effect ( Pettibone and Wedell , 2007; Pratkanis and Farquhar , 1992; see Figure 1D and Materials and methods ) . Taken together , while Group HP replicated the results of Experiment one in most aspects , Group LP showed that the negative impact of D disappeared when time pressure was alleviated , lending further support for a value-based attentional capture account . In addition , only Group LP exhibited an IIA-violating choice pattern in the novel trials consistent with specific context effects , supporting the notion that such effects require longer deliberation times to emerge ( Dhar et al . , 2000; Pettibone , 2012; Trueblood et al . , 2014 ) . Value-based attentional capture predicts an effect of the value of D on choice accuracy that is mediated by attention . In other words , a higher value of D leads to more attention to D , which in turn impairs absolute choice accuracy . This effect has been referred to as value-based oculomotor capture and is thought to underlie the ( behavioral ) effect of value-based attentional capture ( Failing et al . , 2015; Le Pelley et al . , 2015; Pearson et al . , 2016 ) . To directly test whether value-based oculomotor capture drives the negative influence of D on decision making in the Chau2014 paradigm , we obtained eye-movement data as a measure of attention in Experiment three with n3 = 23 participants using the same paradigm as in Experiments 1 and 2 ( Group HP ) . The behavioral results of this experiment were in line with our previous experiments ( see Tables S2-S7 for statistical results ) : The negative HV-D effect on relative choice accuracy reported in Chau2014 could not be replicated , but its effect on absolute choice accuracy was significantly positive; the value of D led to more choices of D and slowed down choices of HV and LV; there was a main effect of Dominance on absolute ( but not relative ) choice accuracy in the novel trial set . With respect to the eye-movement results , we first tested whether options of higher value received more attention , defined as relative fixation duration ( see Materials and methods ) . Thereto , we calculated within each participant the correlation between expected value and relative fixation duration separately for HV , LV , and D . These correlations were consistently higher than zero ( HV: t ( 22 ) = 4 . 79 , p < . 001 , d = 1 . 00; LV: t ( 22 ) = 3 . 84 , p < . 001 , d = 0 . 80; D: t ( 22 ) = 6 . 25 , p < . 001 , d = 1 . 30; Figure 4A ) . Second , we asked whether the dependency of attention to D on D’s value mediated the negative influence of the latter on choice accuracy . A path analysis confirmed this prediction ( Figure 4B ) : In the path model , the value of D had a positive effect on attention to D ( t ( 22 ) = 6 . 30 , p < . 001 , d = 1 . 31 ) , which in turn had a negative effect on choice accuracy ( t ( 22 ) = −6 . 79 , p < . 001 , d = −1 . 42 ) . In fact , the influence of D’s value on choice accuracy was fully mediated by attention with the direct path being not significant ( t ( 22 ) = −0 . 76 , p = . 455 , d = −0 . 16 ) . Finally , we tested whether participants whose attention was affected by D’s value to a greater extent also showed a stronger negative influence of D’s value on absolute choice accuracy . This was confirmed by a significantly negative correlation across participants between the coefficient quantifying the correlation between D-value and D-fixations and the coefficient quantifying the negative impact of D-value on absolute choice accuracy ( r ( 21 ) = −0 . 61; p = . 002; Figure 4C ) . Altogether , the eye-movement results strongly support value-based attentional and oculomotor capture as being the underlying mechanism of sub-optimality in the Chau2014 paradigm ( see Materials and methods , Figure 4—figure supplement 1 and Figure 7—figure supplement 3 for further eye-tracking results ) . We conducted a fourth experiment , for which we used the exact trial sequences of Chau2014 ( provided to us by the authors ) and omitted the novel trials . Besides replicating the effects of value-based attentional capture , another goal of this Experiment 4 was to test whether we could find the HV-D effect on relative choice accuracy when making the experiment almost identical to Chau2014 . The statistical results of this experiment with n4 = 44 participants are provided in Tables S2-S5 and are in line with Experiment 1 to 3: The negative HV-D effect on relative choice accuracy could not be replicated , there was a significantly positive HV-D effect on absolute choice accuracy , and higher values of D led to significantly more choices of D and to longer RT when choosing HV or LV . Figure 5 summarizes the behavioral results collapsed over all experiments conducted under high time pressure . This summary provides clear evidence that there were no robust effects on relative choice accuracy ( and thus no violations of IIA ) but strong effects on absolute choice accuracy consistent with the value-based attentional capture account . Notably , most of the behavioral data analyses in Chau2014 and our study relied on the assumption that participants can integrate magnitude and probability to compute the expected value ( EV = probability × magnitude ) of each option and to choose the option with the larger EV . We tested this assumption by comparing a simple EV-based choice model with two other models that assumed that participants focused on only one attribute ( i . e . , either magnitude or probability ) as well as an expected utility ( EU ) model ( e . g . , Von Neumann and Morgenstern , 1947 ) and prospect theory ( Kahneman and Tversky , 1979; Tversky and Kahneman , 1992 ) ; see Materials and methods ) . In brief , we found that the EV model explained the choice data better than the single-attribute models , but that the EU model provided the best account of the data ( Figure 5—figure supplement 1A ) . The additional complexity of prospect theory compared to EU was not justified by an increased model fit . For reasons of simplicity and comparability to Chau2014 , we used the EV-based value estimates for most statistical and modeling analyses . As a robustness check , however , we reanalyzed our main behavioral tests and replaced EVs by EU-based subjective values . The results were largely unaffected by this adaptation ( Figure 5—figure supplement 1B ) . We also reanalyzed the data of Chau2014 to look for further evidence of value-based attentional capture . As in our own experiments , we found that HV-D was positively linked to absolute choice accuracy ( t ( 20 ) = 4 . 53 , p < . 001 , d = 0 . 99 ) , and that higher values of D led to both more erroneous choices of D ( t ( 19 ) = 4 . 67 , p < . 001 , d = 1 . 05 ) and to higher RT when choosing HV or LV ( t ( 20 ) = 2 . 94 , p = . 008 , d = 0 . 64 ) . Thus , the data of Chau2014 also supported value-based attentional capture . In addition , we show with multiple additional analyses that are detailed in Materials and methods and in Figure 5—figure supplement 2 to Figure 5—figure supplement 5 that the originally reported negative effect of HV-D on relative choice accuracy is a statistical artifact that is due to an incorrect implementation of the interaction term ( HV-LV ) × ( HV-D ) in the performed regression analysis . After correcting this error , the effect of HV-D disappears . Although value-based attentional capture is a well-established empirical finding ( Anderson , 2016; Le Pelley et al . , 2016 ) , to the best of our knowledge it has never been implemented into a decision-making model so far . In the following , we propose and test the Mutual Inhibition with Value-based Attentional Capture ( MIVAC ) model that accounts for the complex interplay of value and attention on choice ( Figure 6; details are provided in Methods ) . MIVAC is an extended mutual inhibition ( MI ) model , a sequential sampling model that assumes a noisy race-to-bound mechanism of separate , leaky , and mutually inhibiting accumulators for each choice option ( Bogacz et al . , 2006; Usher and McClelland , 2001 ) . Notably , the MI model ( without the extensions we propose for MIVAC ) is equivalent to a mean-field reduction of the cortical attractor model that was applied by Chau2014 ( Bogacz et al . , 2006; Wang , 2002; Wong and Wang , 2006 ) , and it is indeed capable of predicting a positive effect of the value of D on relative choice accuracy ( see Figure 6—figure supplement 1 ) . MIVAC assumes an accumulator for each choice option , so four accumulators in the case of the Chau2014 paradigm ( for HV , LV , D , and the empty quadrant ) . A choice is made as soon as an accumulator reaches an upper boundary that collapses in time ( to account for the time limit of the task; for example Gluth et al . , 2012; Gluth et al . , 2013a; Gluth et al . , 2013b; Hutcherson et al . , 2015; Murphy et al . , 2016 ) . Based on our behavioral and eye-movement results , we propose three additional mechanisms . First and foremost , value-based attentional capture is implemented by assuming that the probabilities of fixating particular options depend on their expected values . In other words , more valuable options receive more attention . Second , the input to the accumulator of the currently fixated option is enhanced , consistent with the influence of attention on choice reported in previous work ( Cavanagh et al . , 2014; Krajbich and Rangel , 2011; Krajbich et al . , 2010; Shimojo et al . , 2003 ) and observed in our own data ( see Figure 4—figure supplement 1 and Materials and methods ) . Finally , D can be identified as unavailable , in which case the expected value of D is assumed to be 0 , and its accumulation rate and fixation probability are adjusted accordingly . This feature is a specific requirement for the Chau2014 paradigm in which participants are instructed not to choose D ( this feature can be omitted for other experimental paradigms ) . We propose MIVAC to explain the central behavioral findings across all experiments . Furthermore , we illustrate that simpler models without the added mechanisms of MIVAC ( in particular , without value-based attentional capture ) cannot explain these findings . Thereto , we conducted rigorous quantitative model comparisons in which MIVAC was compared to three models , each of them missing one of the three novel components ( i . e . , without VAC = without value-based attentional capture; without AE = without attentional enhancement; without DD = without distractor detection ) and a baseline MI model , without any of these components . When fitting the models to trials in which a distractor D was present , we found a clear advantage of MIVAC compared to the simplified versions in terms of both average model fit and best model fit per participant ( Figure 7A ) . We then performed a generalization test ( Busemeyer and Wang , 2000 ) by using the models’ estimated parameters from the trials with D to predict behavior in the trials without D . Again , MIVAC outperformed the simpler alternatives ( Figure 7B ) . This generalization test provides strong support for a genuinely better description of the data by MIVAC and rules out an overfitting problem . Qualitatively , MIVAC predicts choice proportions of all potential actions very accurately , and it does so better than the simplified versions without value-based attentional capture with respect to both trials with D present and D absent ( upper and middle panels of Figure 7C ) . The most important test for MIVAC was whether the model accurately predicts the observed increase in choices of D as a function of the value of D . As can be seen in the lower panel of Figure 7C , MIVAC predicts this pattern , and the alternative models either fail to do so or overpredict the frequency of choices of D . MIVAC also reproduces all RT effects reported in Figure 5 , that is , negative effects of the difference and the sum of values of HV and LV , and positive effects of the value of D and the presence of D ( Figure 7D; compare with Figure 5C ) . This is particularly remarkable because MIVAC was fitted only to the choice but not to the RT data . Finally , generalizing MIVAC to the novel trials shows that the model exhibits the observed qualitative patterns for both relative and absolute choice accuracy ( Figure 7E; compare with Figure 5D ) . All these results hold when replacing the EV-based input to the accumulators by EU-based subjective values ( Figure 7—figure supplement 2 ) or when simulating the model with fixations drawn from the empirical fixation duration distributions ( Figure 7—figure supplement 3 ) .
When choosing between multiple alternatives , humans and other animals often violate IIA , which has far-reaching consequences for our understanding of the neural and cognitive principles of decision making ( Hunt and Hayden , 2017; Rieskamp et al . , 2006; Vlaev et al . , 2011 ) . The purpose of our study was to shed light on the unresolved debate of whether the value of a third option leads to violations of IIA in the sense that it either decreases or increases relative choice accuracy between the other two options . Strikingly , we obtained strong evidence that neither violation of IIA is likely to occur when making decisions under time pressure , but that value-based attentional capture leads to a general performance decline ( i . e . , a decline in absolute but not relative choice accuracy representing no violation of IIA ) and slows down the decision process . MIVAC , a computational model that we propose to explain the findings , is based on the assumptions that value drives attention ( Anderson et al . , 2011 ) and attention in turn affects the accumulation of evidence ( Krajbich et al . , 2010 ) . We found that MIVAC reproduced the central behavioral findings for choice accuracy and RT with remarkable precision . Specific characteristics of the Chau2014 task design are likely to have facilitated an influence of value-based attentional capture . Having to choose between four potential actions within about 1 . 5 s puts participants under severe time pressure . This forces them to make use of implicit stimulus-reward associations for identifying attractive options as quickly as possible ( Krebs et al . , 2011; Serences , 2008 ) . Using such implicit associations requires options that are distinguishable via low-level perceptual features , such as color or orientation ( Anderson , 2016 ) , which is exactly what was used in Chau2014 . In fact , the Chau2014 paradigm closely resembles the visual search tasks used to study value-based attentional capture ( Anderson et al . , 2011 ) , in which participants also have 1 . 5 s to identify a target out of several alternatives , while a distractor is characterized by a specific color ( that was previously associated with low or high rewards ) . Yet , there are crucial differences between the two paradigms . Whereas the typical visual search tasks are framed as perceptual tasks , the Chau2014 paradigm is a value-based task . For example , choosing the LV option in the Chau2014 task can still lead to a reward , whereas choosing a non-target in the visual search task is treated as an error . Perceptual and preferential tasks have been shown to elicit different behavior ( Dutilh and Rieskamp , 2016 ) and to rely on partially distinct neural mechanisms ( Polanía et al . , 2014 ) . Such differences likely explain why we did not only find an influence of the value of D on RT but also on choice accuracy , which has been reported in only a minority of studies on value-based attentional capture ( Itthipuripat et al . , 2015; Moher et al . , 2015 ) . An important role of attention in value-based decision making has been established in recent years ( Cavanagh et al . , 2014; Cohen et al . , 2017; Krajbich and Rangel , 2011; Krajbich et al . , 2010; McGinty et al . , 2016; Shimojo et al . , 2003 ) . With some exceptions ( Itthipuripat et al . , 2015; Towal et al . , 2013 ) , research on the interaction of attention and choice has focused on the bias that the former exerts on the latter: Individuals are more likely to choose options that receive comparatively more attention . We replicated this effect in our eye-tracking experiment ( Figure 4—figure supplement 1 ) . However , several other eye-movement patterns in our data differed from previous findings . First , participants were more likely to look at more valuable options ( Figure 4A ) , and this effect was significant even for the first fixation in a trial ( Materials and methods ) . Second , the duration of first fixations was not shorter but longer compared to middle fixations ( Figure 7—figure supplement 1 ) . Third , the effect of attention on choice was not modulated by value ( in the sense that the effect would be more pronounced for options of high compared to low value; Figure 4—figure supplement 1 ) , which is in line with some ( Cavanagh et al . , 2014 ) but not with other previous findings ( Krajbich and Rangel , 2011; Krajbich et al . , 2010 ) . Correspondingly , the MIVAC model differs from previous attention-based instantiations of sequential sampling models , in particular the attentional Drift Diffusion Model ( aDDM; Krajbich and Rangel , 2011; Krajbich et al . , 2010 ) , in at least two aspects . First , attention is not distributed randomly across choice options but depends on the options’ values ( i . e . , value-based attentional capture ) , and second , the accumulation of evidence for attended options is enhanced additively ( i . e . , independently of value ) . The latter feature is more in line with the results and model presented in Cavanagh et al . ( 2014 ) . Interestingly , this study also used abstract choice stimuli as compared to the food snacks used in the studies by Krajbich and colleagues , and participants were put under at least mild time pressure ( i . e . , choices had to be made within 4 s ) . We conclude that attention and its influence on decision making can depend to some extent on the experimental design ( Spektor et al . , 2018b ) . An alternative explanation for the interplay of attention , value , and choice could be that attention does not influence choice , but that choice intention influences attention . Stated differently , when having the intention to choose an option this option will receive more attention which then leads to longer fixation durations , so that high-value options are looked at longer only because they are more likely to be chosen . This reversed interpretation of the role of eye movements is particularly plausible for later stages of an ongoing decision because the last fixation is often directed at the eventually chosen option ( Shimojo et al . , 2003; but see Krajbich et al . ( 2010 ) , for an account of this ‘gaze-cascade effect’ within the aDDM framework ) . However , our analyses of first fixations and first-fixation durations are incompatible with such an interpretation ( see Materials and methods ) . As stated above , the probability to fixate an option first depended on that option’s value , consistent with value-based attentional capture . Importantly , this effect remained significant after controlling for the eventual choice , suggesting that it does not simply reflect the intention to choose the first-fixated option . Furthermore , we found that not only the first fixation and its duration but also the value of the first-fixated option contributed to predicting the eventual choice . In our view , this combined contribution of gaze time and value is best accounted for by assuming that people accumulate evidence for choosing an option based on both the option’s value and the amount of attention spent on it . Therefore , we implemented these mechanisms into MIVAC accordingly . At first glance , an effect opposite to what was observed by Chau2014 appears as support for the divisive normalization account by Louie2013 . However , our results refute such an interpretation because the relative choice proportions between HV and LV were unaffected by the value of D . As with the different eye-movement results discussed above , the absence of a ‘divisive normalization’ effect in our data might be related to dissimilarities between the task paradigms . First of all , participants in Chau2014 ( but not Louie2013 ) made decisions under time pressure , which promoted value-based attentional capture effects but suppressed violations of IIA . Furthermore , the decoy was highlighted as unavailable in Chau2014 , whereas in Louie2013 , the decoy was simply defined as the option with the lowest value . Finally , Chau2014 used abstract two-dimensional stimuli , whereas Louie2013 used concrete food snacks , for which it is currently debated whether and how specific attributes are taken into account ( Rangel , 2013 ) . Louie2013 did not address the distinction between single- and multi-attribute decisions , but their model should predict a negative influence of D’s value on relative choice accuracy in both cases ( because it assumes normalization to occur at the level of the integrated value signal ) . Although it goes beyond the scope of the current study , it will be critical to test to what extent effects of value-based attentional capture can be generalized to different implementations of multi-alternative decisions in future research . Notably , the principle of normalization can be implemented into sequential sampling models , albeit in a different way compared to the mutual inhibition mechanism of MIVAC . According to Teodorescu and colleagues ( 2016 ) , normalization acts on the input level of the accumulation process such that the input to one accumulator is reduced by the input to all other accumulators ( a mechanism also referred to as ‘feed-forward inhibition’; see Bogacz et al . , 2006 ) , whereas mutual inhibition acts on the accumulator level such that the accumulation of evidence for one option is reduced by how much evidence has been accumulated for the competing options . Separating between these different instantiations of inhibition is best achieved by taking their RT predictions into account ( Teodorescu et al . , 2016 ) . Generally , we did not obtain any reliable value-dependent violation of IIA in the standard version of the Chau2014 paradigm , but only when we gave participants more time to decide . In line with previous research ( Dhar et al . , 2000; Pettibone , 2012; Trueblood et al . , 2014 ) , these findings demonstrate the time-dependency of context effects: Being under time pressure or not determines whether effects related to value-based attentional capture or multi-attribute context effects can be expected to occur . One reason for this dependency could be that people adaptively select decision strategies based on the current decision context ( Gluth et al . , 2014; Payne et al . , 1988; Rieskamp and Otto , 2006 ) , and not all strategies are prone to the same contextual or attentional biases . Furthermore , multi-attribute context effects could be the consequence of dynamic choice mechanisms that require comparatively long deliberation times to exert a measurable influence on decisions ( Roe et al . , 2001; Trueblood et al . , 2014 ) . Importantly , even though the negative influence of D on absolute choice accuracy is not a violation of IIA , our results cannot be well accounted for by standard economic choice models , such as the multinomial logit model ( e . g . , McFadden , 2001 ) . We compared three variants of multinomial logit models with MIVAC ( see Materials and methods and Figure 7—figure supplement 5 ) . Two of these models either exclude or include D as a regular choice option . Excluding D leads to the prediction that D does not decrease absolute choice accuracy at all . Including D leads to the prediction that D is chosen as a function of its value relative to HV and LV and thus to the prediction that D is chosen in too many trials . A third alternative is to assume that choices are based on a combination of the options’ EVs ( as signaled by the rectangles’ colors and orientations ) with a subjective value assigned to the colors of the frames that signal whether an option is a target or a distractor . Although this version of a multinomial logit model is able to predict that D is chosen in only a minority of trials , its predictions are clearly worse than those of MIVAC , as the logit model predicts too few choices of HV and too many choices of LV ( Figure 7—figure supplement 5 ) . Furthermore , standard economic choice models do not take RT into account and thus cannot predict that high-value D options slow down the choice process ( Figure 5C ) such that the probability of making too-slow errors is increased ( Figure 5B ) . A dynamic component , like the accumulation of evidence as implemented in MIVAC , is required to explain these RT-related effects . Also note that even though the current version of our model does not predict violations of IIA ( consistent with our data under high time pressure ) , it is straightforward to extend MIVAC to allow such violations . In Materials and methods , we describe one possible extension of MIVAC and demonstrate how it allows the model to predict the IIA violation in our data under low time pressure . Methodologically , the current study is an eminent example of the importance of replication attempts for the advancement of empirical science ( Munafò et al . , 2017 ) , which is particularly timely given the current debate on replicability of research in psychology , neuroscience and other fields ( Camerer et al . , 2016; Open Science Collaboration , 2015; Poldrack et al . , 2017 ) . Our initial hypothesis for explaining the IIA violation reported in Chau2014 was based on the assumption of a robust and reliable effect . However , we could not replicate this effect ( see Materials and methods and Figure 5—figure supplement 2 to Figure 5—figure supplement 5 for additional analyses that challenge both the replicability and the reproducibility of the effect proposed by Chau2014 ) . When developing novel ideas and experiments on the basis of past findings , it is important that these findings are reliable and have been replicated , to avoid leading research fields into scientific cul-de-sacs , in which time and resources are wasted ( Munafò et al . , 2017 ) . In our case , only the ( unsuccessful ) attempt to replicate an original study allowed us to identify truly robust behavioral effects , which favor an entirely different mechanistic explanation than originally proposed: Attention can be captured by irrelevant but value-laden stimuli , which slows down the goal-directed choice process and impairs decision making under time pressure . Crucially , neither our study nor replication studies in general are destructive . We gained novel insights about the role of attention in multi-alternative decision making , and we will gain similarly important insights when attempting to replicate other studies .
Thirty-one participants ( 21 female , age: 20 – 47 , M = 27 . 71 , SD = 6 . 59 , 29 right-handed ) completed Experiment 1 . A total of 51 participants signed up for Experiment 2 . Due to computer crashes , the data of two participants ( one from Group HP and one from Group LP ) were incomplete and excluded from the analyses , resulting in a final sample of 49 participants , 25 were in Group HP ( 13 female , age: 20–46 , M = 26 . 88 , SD = 6 . 62 , 24 right-handed ) and 24 were in Group LP ( 11 female , age: 19–35 , M = 23 . 96 , SD = 3 . 43 , 20 right-handed ) . Participants were assigned randomly to the two groups . Thirty participants signed up for the eye-tracking Experiment 3 . One participant was excluded for not passing the training-phase criterion , and additional six participants were excluded due to incompatibility with the eye-tracking device ( for further details see section Eye-tracking procedures and pre-processing ) , resulting in a final sample of 23 participants ( 14 female , age: 18–54 , M = 25 . 70 , SD = 8 . 66 , 19 right-handed ) . Forty-seven participants signed up for Experiment 4 . Due to failing the training-phase criterion , three participants were excluded , resulting in a final sample of 44 participants ( 36 female , age: 18–46 , M = 23 . 70 , SD = 5 . 74 , 40 right-handed ) . The sample size for Experiment 1 was based on the assumption that testing 1 . 5 times as many participants as in the original study ( Chau2014 ) would suffice to replicate its main results . In Experiment 4 , we tested more participants to ensure a statistical power of >0 . 90 for replicating the effect of HV-D on relative choice accuracy ( note that the effect size in Chau2014 was d = −0 . 495 ) . The sample size for Experiments 2 and 3 were based on observing strong effects ( i . e . , d ≥ 0 . 8 ) of value-based attentional capture in Experiment 1 ( but it should be noted that no formal power analysis was conducted ) . All participants gave written informed consent , and the study was approved by the Institutional Review Board of the Department of Psychology at the University of Basel . All experiments were performed in accordance with the relevant guidelines and regulations . Data of all participants included in the final samples are made publicly available on the Open Science Framework at https://osf . io/8r4fh/ . The paradigm was very similar for all four experiments . Participants repeatedly chose between either two ( binary trials ) or three ( distractor trials ) two-outcome lotteries ( gambles ) , each yielding an outcome of magnitude X in Swiss Francs ( CHF ) with probability p or 0 otherwise . The gambles were represented by rectangles , each shown in a random quadrant of the screen , whereby the rectangles’ colors represented outcomes X and the angles represented the probabilities p . Outcomes ranged from CHF 2 to CHF 12 in steps of CHF 2 and were represented by colors ranging from either green to blue or blue to green . Probabilities ranged from 1/8 to 7/8 in steps of 1/8 and were represented by orientation angles ranging from 0° to 90° or from 90° to 0° in steps of 15° ( see Figure 1B for all colors and orientations ) . Associations between colors/outcomes and orientations/probabilities were counter-balanced between participants . In binary trials , participants saw the two options for 100 ms before orange frames appeared around each option ( pre-decision phase ) . After the frames appeared , participants had up to 1 . 5 s to make a choice by pressing 7 , 9 , 1 , or 3 on the numeric keypad for upper left , upper right , lower left , or lower right quadrant , respectively ( participants belonging to Group LP of Experiment 2 had 6 s after appearance of the frames to decide ) . Distractor trials were similar to binary trials: All of the options were presented for 100 ms before frames appeared around them . Contrary to the binary trials , one of the options had a magenta frame ( the distractor ) , signaling that it could not be chosen . Choosing the distractor resulted in a screen telling that the option was not available after which a new trial began . Similarly , choosing an empty quadrant resulted in a screen showing that the quadrant was empty after which a new trial began . If a valid choice was registered , the trial continued with a choice-feedback phase for 1 to 3 s , in which the chosen option was highlighted in a dark red color . To ensure that participants paid attention to all available options , participants had to complete a ‘match’ trial before the choice-feedback phase in 15% of all trials . On match trials , one of the options from the decision phase ( including the distractor , if distractor trial ) was presented in the middle of the screen . Participants had up to 2 s ( 6 s in Experiment 2 , Group LP ) to press the key corresponding to the option’s quadrant . If correct , participants saw a screen saying ‘correct’ and an extra CHF 0 . 10 were added to the participant’s account , otherwise they saw a screen saying ‘wrong’ . After every match trial , the trial continued with the choice-feedback phase . After the choice-feedback phase , participants received feedback about the outcomes of the gambles for 1 to 3 s . The frames’ colors changed to grey if the option did not yield a reward ( i . e . , the outcome was CHF 0 ) or to golden yellow if the option yielded a reward . Participants also received feedback about the distractor’s outcome on distractor trials . After this outcome-feedback phase , a new trial began with an inter-trial interval of 1 . 5 to 3 s in which a fixation cross was shown . After giving informed consent and filling out the demographic questionnaire , participants received detailed instructions about the task and were familiarized with the outcome and probability associations by making six judgments for each dimension in a paired comparison . In all experiments , participants completed a training phase and an experimental phase . In the training phase , participants encountered up to 210 trials , half of which were distractor trials . These trials were randomly generated with the boundary condition that 2/3 of the trials were not dominant ( i . e . , HV did not have a higher probability and a higher outcome than LV ) and the rest were dominant . The training phase continued until participants encountered at least 10 dominant binary trials , and chose HV in at least 70% of the last 10 encountered dominant binary trials . The training phase ended when this criterion was reached and the experimental phase began . If the participants finished all 210 training trials without passing the criterion , the experiment ended . As reported above , participants who did not pass the criterion were excluded from the analysis . The experimental phase consisted of either 412 ( Experiment 1 to 3 ) or 300 ( Experiment 4 ) trials . The 300 trials used in Experiment four were shared across all experiments and are those used by Chau2014 . In Experiment 4 , all trials were presented in exactly the same orders as in Chau2014’s experiment , whereas in the other experiments , the distractor trials were presented in the order provided to us by the authors of Chau2014 with the randomized binary trials interleaved . In addition to these trials , Experiments 1 to 3 included 56 novel distractor trials and , correspondingly , the 56 binary trials belonging to these novel trials ( details are provided in the next two sections ) . Throughout the experiment , participants had the opportunity to make four breaks . After completing the experiment , participants received their show-up fee ( CHF 5 for 15 min ) , the average reward of the chosen option ( distractor and empty quadrant choices counted as no reward ) , and the accumulated match bonuses . If participants reached the experimental phase , the experiments took approximately 75 min . Before conducting our experiments , we assumed that the positive relationship between the value of D and relative choice accuracy as reported by Chau2014 was a robust effect . To explain the apparent contradiction with the findings of Louie2013 , we reasoned that the explicit presentation of two attributes ( i . e . , magnitude X and probability p of reward ) in the Chau2014 task led people to compare the options on those attributes directly ( i . e . , a multi-attribute decision between attributes X and p instead of a decision between expected values , EV ) . Importantly , certain attribute-wise comparison processes are known to produce IIA violations , so-called ‘context effects of preferential choice’ ( Berkowitsch et al . , 2014; Gluth et al . , 2017; Mohr et al . , 2017; Pettibone and Wedell , 2007; Roe et al . , 2001; Trueblood et al . , 2014; Usher and McClelland , 2004 ) . More specifically , our initial hypothesis was that individuals can recognize that D is either better or worse than LV and/or HV with respect to each attribute ( e . g . , D might dominate LV with respect to probability ) . Critically , since LV is per definition worse than HV , it is more likely that D dominates LV than that it dominates HV on some attribute . However , this is only true as long as D has not very low attribute values ( leading to a low EV overall ) . Thus , the dominance relationship between D and LV/HV may help participants to identify the option with the highest EV . The positive relationship between D and relative choice accuracy would then be an epiphenomenon of this dominance relationship mechanism . To give an example , let us assume that HV , LV , and D are specified as follows: HV: p = 5/8; X = CHF 8 → EV = CHF 5 LV: p = 3/8; X = CHF 4 → EV = CHF 1 . 5 D: p = 6/8; X = CHF 6 → EV = CHF 4 . 5 In this case , D is superior to HV and LV with respect to probability . With respect to magnitude , however , D is superior to LV but inferior to HV . Hence , by counting the number of times D is superior to LV and HV on the attributes ( i . e . , two for LV vs . one for HV ) , a decision maker could correctly identify the option with the highest EV . Critically , this information is not helpful anymore when we replace D by a distractor D* of lower EV: D*: p = 6/8; X = CHF 2 → EV = CHF 1 . 5 In this new case , the distractor is inferior to both HV and LV with respect to magnitude and ( still ) superior to both with respect to probability ( i . e . , the count is one for LV vs . one for HV ) . Thus , the option with the highest EV cannot be identified anymore based on the dominance relationship alone . This demonstrates that the high-value D might better support choosing between HV and LV than the low-value D* . Notably , the idea that the relative ranking of options influences decision making has also been suggested by others ( Howes et al . , 2016; Stewart et al . , 2006; Tsetsos et al . , 2016 ) . Besides this dominance relationship hypothesis , the paradigm of Chau2014 might also involve other context effects that influence behavior . We considered the attraction effect ( Huber et al . , 1982 ) and the phantom-decoy effect ( Pettibone and Wedell , 2007; Pratkanis and Farquhar , 1992 ) . According to the attraction effect , the preference between two options ( in our case between HV and LV ) can be changed by adding a third option ( in our case D ) that is similar but clearly inferior to only one of the two options ( in our case , if D is similar to HV and worse than it , HV should be preferred; if D is similar to LV and worse than it , LV should be preferred ) . The phantom-decoy effect predicts that if an option is similar but worse than D ( and D is unavailable , as in the Chau2014 task ) , it is more likely to be chosen . According to a combination of attraction and phantom-decoy effects , independent of whether D is worse or better than the similar option , the option being more similar to D should be more likely be chosen . In contrast to the context effects , the Chau2014 model , and the divisive normalization account , value-based attentional capture does not predict any influence of D on relative choice accuracy , but a negative effect on absolute choice accuracy ( see Figure 1D ) . As described in the following section , we sought to distinguish between all these different context effects and the model proposed by Chau2014 by implementing a novel set of trials . In the novel set of trials used to dissociate predictions from various models and context effects , the HV and LV options were arranged such that i . ) HV was superior to LV with respect to one attribute ( magnitude or probability ) but ii . ) inferior with respect to the other attribute , and iii . ) D could be placed such that it either fully dominated HV/LV or it was fully dominated by HV/LV ( for an example , see Figure 1C in the main text ) . In the Chau2014 task , there are 14 possible combinations of HV and LV that fulfil these criteria . For each of these 14 combinations , D was placed directly ‘above’ or ‘below’ HV or LV resulting in four trials per combination and 56 novel trials in total ( we also added 56 binary trials without D , so that Experiment 1 to 3 had 112 trials more than the original study ) . The ( qualitative ) predictions of the models and context effects with respect to these novel trials are outlined in Figure 1D . Chau2014’s biophysical cortical attractor model predicts a positive effect of the value of D on relative choice accuracy . Thus , the performance should be higher in trials with dominant distractors ( i . e . , D > HV and D > LV ) . The divisive normalization model by Louie2013 predicts higher accuracy for low-value Ds , or in other words , it predicts the opposite of Chau2014’s model . Our initial dominance relationship hypothesis predicts higher choice accuracy if HV dominates D ( i . e . , D < HV ) , or if LV is dominated by D ( i . e . , D > LV ) , because in these cases HV is better than D on more attributes than LV . The combination of attraction and phantom-decoy effects predicts more accurate choices when D is dominated by HV ( i . e . , D < HV ) or dominates HV ( i . e . , D > HV ) , or in other words , when D is more similar to HV than to LV . Importantly , all these predictions refer to relative choice accuracy ( for which we do not find any robust effects in the Chau2014 task with short deliberation time; see Figure 5D ) . On the contrary , value-based attentional capture predicts no effect on relative choice accuracy , but a reduction of absolute choice accuracy when D has a high value ( i . e . , D > HV and D > LV ) . Note that the different predictions can also be formulated in terms of main effects and interactions of an ANOVA with the factors Dominance ( D dominates or is dominated by HV/LV ) and Similarity ( D is more similar to HV or to LV ) . Within this ANOVA , the cortical attractor model predicts a main effect of Dominance . The divisive normalization model also predicts this main effect but in the opposite direction . The dominance relationship hypothesis predicts an interaction effect of Dominance and Similarity . The combined attraction/phantom-decoy effect predicts a main effect of Similarity . Value-based attentional capture predicts a main effect of Dominance in the same direction as the divisive normalization model but on absolute ( not relative ) choice accuracy . In each trial , there was always a higher-valued option HV and a lower-valued option LV . We used two different dependent measures , relative and absolute choice accuracy . Relative choice accuracy refers to the proportion of HV choices among choices of HV and LV only , whereas absolute choice accuracy refers to the proportion of HV choices among all choices ( including choices of D , choices of the empty quadrant , and missed responses due to the time limit ) . Importantly , an influence of the value of D on relative choice accuracy implies a violation of IIA , but an influence on absolute choice accuracy does not necessarily imply this . For each of the two dependent variables , we estimated intra-individual logistic regressions and tested the regression coefficients between subjects against 0 using a two-sided one-sample t-test with an α level of . 05 . We used the set of predictor variables reported in Chau2014 , which consisted of the difference in EV of the two available options , HV-LV , the sum of their EVs , HV+LV , the EV difference between HV and D , HV-D , the interaction between HV-LV and HV-D , ( HV-LV ) × ( HV-D ) , and whether it was a binary or distractor trial , D present . In the binary trials , the predictors HV-D and ( HV-LV ) × ( HV-D ) were kept constant ( i . e . , replaced by the mean values in the distractor trials in the regression analysis ) . The predictor variables HV-LV , HV+LV , and HV-D were standardized . Importantly , standardization was conducted before generating the interaction term ( HV-LV ) × ( HV-D ) in order to avoid nonessential multicollinearity ( Aiken and West , 1991; Mahwah et al . , 2003; Dunlap and Kemery , 1987; Marquardt , 1980 ) . The interaction term itself and the binary predictor D present were not standardized ( note that this would not have changed any statistical inferences but only the absolute values of coefficients ) . In addition , we analyzed the influence of the ( standardized ) value of D on the tendency to choose D ( logistic regression ) and on the RT of HV and LV choices ( linear regression ) . The RT analysis included additional predictor variables with a significant influence on RT ( i . e . , HV-LV , HV+LV , D present ) . Note that only the 300 trials that were also used in the Chau2014 paradigm ( but not our 112 novel trials ) were included in these regression analyses . The novel trial sets used to dissociate different model predictions were analyzed by a 2 ( Dominance ) x 2 ( Similarity ) ANOVA ( for details see above ) . Because participants received feedback after each decision , we tested whether improvements over time affected the results of relative or absolute choice accuracy by re-analyzing the regressions with an additional predictor variable that coded for the ( standardized ) trial number . Although we found a small learning effect on absolute choice accuracy ( t ( 122 ) = 2 . 12 , p = . 036 , d = 0 . 19 ) , this did not affect any other effects qualitatively . When adding the trial number predictor to the analysis of the influence of D’s value on the tendency to choose D , we found a strong learning effect ( t ( 122 ) = −10 . 18 , p < . 001 , d = −0 . 92 ) , suggesting that participants improved in avoiding to choose D over the course of the experiment ( but the effect of D’s value remained significant ) . Hence , future instantiations of our model could incorporate a dynamic component to accommodate this learning effect . Experiment 3 was conducted while participants’ gaze positions were recorded using an SMI RED500 eye-tracking device . The experimental procedure was adapted to make it suitable for an eye-tracking experiment . Participants completed the experiment on a 47 . 38 × 29 . 61 cm screen ( 22" screen diagonal ) with a resolution of 1680 × 1050 pixels . During the inter-trial interval , participants were instructed to look at the fixation cross and the random duration of the inter-trial interval was removed . Instead , there was a real-time circular area of interest ( AOI ) with a diameter of 200 pixels around the fixation cross . Participants’ gazes had to ( continuously ) stay within this AOI for 1 s for the trial to begin . This was done to make sure that participants were indeed looking at the fixation cross and to check the calibration at every trial . If this criterion was not reached within 12 s , the eye tracker was re-calibrated . This procedure was explained to the participants by the experimenter . In case these re-calibrations happened too frequently ( e . g . , three times in a row within the same trial , or at least three times in ten trials ) , the sampling frequency was reduced from the initial 500 Hz to 250 Hz . If the issues continued until the lowest possible frequency of 60 Hz was reached , the experiment was aborted and participants received their show-up fee and decision-based bonuses accumulated until then . The first calibration took place just before the training phase and the eye tracker was re-calibrated after each of the four breaks . This experiment took approximately 90 min to complete . The raw gaze positions were re-coded into events ( fixations , saccades , and blinks ) in SMI’s BeGaze2 software package using the high-speed detection algorithm and default values . AOIs were defined around the positions where the frames of the options were , and all fixations inside the frame were counted towards the option within that quadrant . Fixations at empty quadrants as well as all fixations outside of the pre-defined AOIs were counted as empty gazes . Fixations within a trial were collapsed and summed to form the dependent variables relative fixation duration and number of fixations . We report results based on the relative fixation duration ( i . e . , the sum of the duration of all fixations on a specific quadrant divided by the sum of the duration of all fixations on any quadrant ) . Note that this measure is highly correlated with the number of fixations , which yielded similar results . To test the hypothesis that the negative influence of D on absolute choice accuracy is driven by value-based attentional/oculomotor capture , we conducted the following three analyses: i . ) We tested for the dependency of relative fixation duration of HV , LV , and D on their respective EVs ( Figure 4A ) . Thereto , the correlations between the options’ relative fixation durations and EVs were calculated for each participant , and the individual Fisher z-transformed correlation coefficients were subjected to one-sample t-test against 0 on the group level ( after checking for normality assumptions via the Kolmogorov-Smirnoff test at p < . 1 ) . ii . ) We conducted a path analysis ( within each participant ) in which the influence of the value of D on absolute choice accuracy was hypothesized to be mediated by the relative fixation duration on D ( Figure 4B ) . iii . ) We run an ( across-participant ) correlation between the behavioral regression coefficients representing the influence of the value of D on absolute choice accuracy and the regression coefficients representing the influence of the value of D on relative fixation duration on D ( Figure 4C ) . To inform our computational model , we tested whether there is evidence for a direct effect of attention on choice ( i . e . , an option that receives relatively more attention is more likely to be chosen–independent of the option’s value ) as reported by Cavanagh and colleagues ( Cavanagh et al . , 2014 ) , or whether the effect of attention is modulated by value ( i . e . , only options with high value that receive more attention are more likely to be chosen ) as proposed by Krajbich and colleagues ( Krajbich and Rangel , 2011; Krajbich et al . , 2010 ) . Thereto , we conducted a logistic regression analysis with absolute choice accuracy as dependent variable and the following six predictor variables: the expected value of HV , the expected value of LV , the relative fixation duration of HV , the relative fixation duration of LV , the interaction of the value of HV and the relative fixation duration of HV , and the interaction of the value of LV and the relative fixation duration of LV . The first three predictors were standardized , and the interaction variables were generated based on standardized variables to avoid nonessential multicollinearity ( Marquardt , 1980 ) . A direct attention model predicts a positive effect of the relative fixation duration of HV and a negative effect of the relative fixation duration of LV . A value-dependent attention model predicts positive and negative effects of the interaction terms for HV and LV , respectively . The results are shown in Figure 4—figure supplement 1 . Besides the positive effect of HV-LV , there were significantly positive and negative effects of the relative fixation duration of HV ( t ( 22 ) = 9 . 40 , p < . 001 , d = 1 . 96 ) and LV ( t ( 22 ) = −8 . 20 , p < . 001 , d = −1 . 71 ) , respectively . Of the interaction terms , only LV was significantly negative ( t ( 22 ) = −4 . 09 , p < . 001 , d = 0 . 85 ) , but HV was not significantly positive ( t ( 22 ) = −1 . 21 , p = . 237 , d = −0 . 25 ) . Thus , we found unequivocal evidence for a direct effect of attention on choice ( Cavanagh et al . , 2014 ) but not for an interaction of attention and value ( Krajbich and Rangel , 2011; Krajbich et al . , 2010 ) . Consequently , we implemented the influence of attention on choice in the computational model by a value-independent , additive increase of the input signal to the accumulator of the currently attended option ( see Equation 9 below ) . A central and often-tested prediction of the aDDM ( Krajbich and Rangel , 2011; Krajbich et al . , 2010 ) is that the direction of the initial fixation is independent of value ( Konovalov and Krajbich , 2016; Krajbich and Rangel , 2011; Krajbich et al . , 2010 ) . However , the results of our eye-tracking experiment are inconsistent with this prediction . We found that the probability to fixate HV , LV , D first was positively linked to the respective option’s value ( HV: t ( 22 ) = 6 . 03 , p < . 001 , d = 1 . 26; LV: t ( 22 ) = 3 . 84 , p < . 001 , d = 0 . 80; D: t ( 22 ) = 3 . 57 , p = . 002 , d = 0 . 74 ) . Importantly , the effects for HV and D remained significant when controlling for the eventual choice ( HV: t ( 22 ) = 4 . 43 , p < . 001 , d = 0 . 92; LV: t ( 22 ) = 0 . 62 , p = . 539 , d = 0 . 13; D: t ( 21 ) = 2 . 70 , p = . 013 , d = 0 . 58 ) , indicating that the effect was not solely driven by an intention to choose a particular option . Instead , the effect can be explained by assuming that value-based oculomotor capture biases early competition on the saccade map ( Pearson et al . , 2016 ) . In addition , we tested whether the first fixation and its duration was predictive of choice ( when controlling for HV-LV and D ) , which would be consistent with previous findings ( Cavanagh et al . , 2014; Krajbich and Rangel , 2011; Krajbich et al . , 2010 ) and suggest a biasing influence of attention on evidence accumulation . Indeed , the probability to choose HV , LV or D was predicted by whether the first fixation was made to the respective option ( HV: t ( 22 ) = 6 . 11 , p < . 001 , d = 1 . 27; LV: t ( 22 ) = 4 . 98 , p < . 001 , d = 1 . 04; D: t ( 20 ) = 7 . 02 , p < . 001 , d = 1 . 53 ) . Similarly , the duration of the first fixation predicted the choice of all three options ( HV: t ( 22 ) = 7 . 26 , p < . 001 , d = 1 . 51; LV: t ( 22 ) = 6 . 18 , p < . 001 , d = 1 . 29; D: t ( 20 ) = 2 . 92 , p = . 008 , d = 0 . 64 ) . Yet , the value difference HV-LV remained being predictive of the choice of HV and LV ( HV: t ( 22 ) = 8 . 09 , p < . 001 , d = 1 . 69; LV: t ( 22 ) = −11 . 49 , p < . 001 , d = −2 . 40 ) , and the value of D remained being predictive of the choice of D ( t ( 20 ) = 4 . 88 , p = . 003 , d = 1 . 07 ) . Thus , the first fixation and its duration contributed to the probability of choosing an option but was not fully predictive of it . In our view , these results are best explained by assuming that gaze time biases the evidence accumulation towards the fixated option as implemented in the MIVAC model . Note that the analysis of first fixations included trials from the training phase . Many analyses in Chau2014 and our study , such as the regression analysis of relative choice accuracy , relied on the assumption that participants were able to integrate magnitude and probability information to calculate the expected value ( EV ) of each option and to choose the option with the largest EV . We tested this assumption by comparing five simple choice models with respect to how well they predicted whether participants chose HV or LV ( note that in contrast to MIVAC , we do not propose any of these models to account for all behavioral findings but instead treat them only as auxiliary models to test the assumption that participants relied on EV when making decisions ) . The first two models assumed that participants relied only on magnitude information ( ‘OM’ ) or only on probability information ( ‘OP’ ) . The third model assumed that participants relied on EV ( ‘EV’ ) . The fourth model assumed that participants relied on subjective values given by expected utility theory ( e . g . Von Neumann and Morgenstern , 1947 ) ( ‘EU’ ) , specified as follows: ( 1 ) SVEU=EUx , p=ux×pux=xαwhere α is a free parameter ( α > 0 ) modulating the curvature of the utility function . The last model assumed that participants relied on subjective values given by prospect theory ( Kahneman and Tversky , 1979; Tversky and Kahneman , 1992 ) ( “PT” ) , specified as follows:SVPT=V ( x , p ) =v ( x ) ×w ( p ) vx=xαw ( p ) =pτ ( pτ+[1−p]τ ) 1τwhere α is again a free parameter ( α >0 ) modulating the curvature of the utility function and where τ is an additional free parameter ( τ >0 ) modulating the shape of the probability-weighting function . For each of these models , the probability of choosing HV was given by the logistic/soft-max choice function:PHV=11+e-δ× ( SV[HV]Model-SV[LV]Model ) where δ is a free parameter ( δ >0 ) modulating the sensitivity to value differences , and SV[HV]Model and SV[LV]Model refer to the model-specific values of options HV and LV , respectively . We used the same minimization algorithm and model-comparison approach as for MIVAC ( see section Modeling procedures for MIVAC III: parameter estimation and nested model comparison ) . Models were compared on the basis of the choice data from the 300 trials that were part of the initial choice set of Chau2014 and from the 123 participants of our study who conducted the Chau2014 paradigm under time pressure . Because the EU model provided the most parsimonious account of the data ( Figure 5—figure supplement 1 ) , we reanalyzed the central behavioral tests of our study ( shown in Figure 5A–5C ) and replaced the options’ EVs by the EU-based subjective values estimates . Furthermore , we reran the MIVAC model to test whether using EU-based subjective values as input to the model’s accumulators would lead to qualitatively different predictions ( Figure 7—figure supplement 2; details are provided in section Modeling procedures for MIVAC IV: MIVAC with subjective values as inputs ) . Note that the analyses and predictions for the novel trial sets ( Figures 1D and 5D ) do not depend on whether EV or EU is assumed as the underlying choice model . The computational model we propose , MIVAC , is an extended version of the mutual inhibition ( MI ) model ( Bogacz et al . , 2006; Usher and McClelland , 2001 ) . The MI model belongs to the sub-class of accumulator models within the framework of sequential sampling models ( Ratcliff and Smith , 2004; Smith and Ratcliff , 2004 ) . Sequential sampling models assume that evidence for choice options is accumulated over time until a decision threshold is reached after which a choice is made . Accumulator models use separate evidence accumulators for each choice option . MIs further assume mutual ( or lateral ) inhibition of these accumulators , that is , the more evidence for one choice option has been accumulated , the stronger this accumulator inhibits the competing accumulators ( in contrast to feedforward inhibition models that assume inhibition between the inputs to accumulators ) . In addition , accumulators are leaky , that is , they tend to decay toward the starting point ( i . e . , 0 ) . One reason why we chose this framework ( and not , for instance , the aDDM ) is that the MI is equivalent to a mean-field reduction of the biophysical attractor model that was applied by Chau2014 ( Bogacz et al . , 2006; Wang , 2002; Wong and Wang , 2006 ) . Accordingly , without a value-based attentional capture mechanism , the MI model is able to predict a positive effect of the value of the distractor ( D ) on choice accuracy similar to the model proposed by Chau2014 ( see Figure 6—figure supplement 1 ) . In general , we begin with a baseline MI and then add additional assumptions ( about the interplay of value , attention , and choice ) that are necessary to explain the observed behavioral findings . Formally , the MI assumes four accumulators representing the four choice options high-value option ( HV ) , low-value option ( LV ) , D , and the empty quadrant . The n by t matrix A representing all n ( =4 ) accumulators is updated every time step Δt according to:At+1=S⋅At+ It+Etwhere S represents an n by n leakage and inhibition matrix with leakage parameter δ and inhibition parameter ϕ as on- and off-diagonal elements , respectively , It represents an n by t input matrix that–in the case of the baseline MI–contains the EV of each choice option ( which is 0 for the empty quadrant ) , and Et represents an n by t noise matrix with each element being drawn from a normal distribution with mean 0 and standard deviation σ . As soon as accumulator Ai , t reaches an upper threshold θt ( i . e . , Ai , t ≥ θt ) , option i is chosen ( at time point t ) . Similar to previous work ( Gluth et al . , 2012; Gluth et al . , 2013a; Gluth et al . , 2013b; Hutcherson et al . , 2015; Murphy et al . , 2016 ) , we assume that the decision threshold decreases ( linearly ) with time ( note that we found the empirical response-time [RT] distribution in this task to lack the typical right-skewed shape , which is possibly due to the strict time limit and would be in line with a decreasing decision threshold; see Hawkins et al . , 2015 ) . In contrast to the leaky competing accumulator model ( Usher and McClelland , 2001 ) but in line with suggestions of Bogacz et al . , 2006 , we allowed A to take on negative values . Potentially , the baseline MI could have six free parameters: the leakage parameter δ , the inhibition parameter ϕ , the standard deviation σ of the noise vector , the height of the decision threshold at the start and end of the decision period θ0 and θtmax , and a non-decision time parameter τnon-dec . However , because we sought to extend this model by three additional components and because predictions of the MI cannot be derived analytically but must be simulated , we drastically constrained the model by fixing all but one of the aforementioned parameters ( it should be noted at this point that we had to make several simplifying assumptions to simulate the model in a reasonable amount of time ) . Leakage and inhibition were set to δ = 0 . 96 and ϕ = −0 . 036 , respectively , the start and end points of the decision threshold were set to θ0 = 1’000 and θtmax = 200 , respectively , and the non-decision time was set to τnon-dec = 200 ms ( so that tmax – τnon-dec = 1’400 ms ) . Thus , only σ remained as a free parameter ( this parameter controls the stochasticity of the decision process and allows to account for performance differences between participants ) . Importantly , fixed parameter values were chosen carefully so that the model made various sensible predictions ( i . e . , the relative choice ratio between HV and LV , the average RT , and the proportion of misses due to no threshold crossings until tmax were ensured to lie all within the range of the empirical observations ) . For the MIVAC model , we propose three additional mechanisms that are required to account for the behavioral data and that describe the interplay of value , attention , and decision making as observed in the eye-tracking experiment . First , we implemented value-based attentional capture by modelling fixations to the choice options and by assuming that the probability of fixating a particular option depends on the option’s value ( relative to the other options ) . Similar to a recent study on reinforcement learning and attention ( Leong et al . , 2017 ) , we used a logistic function to describe the relationship of value and fixation probability: ( 8 ) Fi=eγ×nEV ( i ) ∑jeγ×nEV ( j ) where Fi is the probability of fixating option i , γ is a free parameter , and nEVi is the expected value of option i normalized with respect to all currently presented options ( normalization was used to avoid that value-based attentional capture effects depend too much on the sum of all values ) . We assumed that every 200 ms , a new fixation is made , and with probability Fi it is made to option i ( with the possibility to re-fixate the same option ) . This time window roughly corresponds to the average fixation duration of 230 ms in our data . We chose this implementation to maintain a high speed of model simulations and to make more conservative model predictions , as a more data-driven implementation of using empirical fixation patterns ( e . g . , Cohen et al . , 2017; Krajbich and Rangel , 2011; Krajbich et al . , 2010 ) may improve model fit but limit generalizability . In the section Modeling procedures for MIVAC V: MIVAC with fixation durations drawn from empirical distributions , we show that using empirical fixation durations ( instead of a fixed duration of 200 ms ) does not change the models’ predictions and the model comparison results qualitatively . The second additional mechanism is the enhanced input to the currently attended accumulator . This assumption is based on previous work showing that options which receive more attention are more likely to be chosen ( Cavanagh et al . , 2014; Krajbich and Rangel , 2011; Krajbich et al . , 2010; Shimojo et al . , 2003 ) , which we replicated in our eye-tracking experiment . Note , however , that we found evidence for a direct effect of attention on choice rather than for a modulatory effect that depends on the value of the attended option ( see section Eye-tracking analysis II: direct vs . value-dependent influences of attention on choice and Figure 4—figure supplement 1 ) . Accordingly , we assumed that the input Ia , t of the currently attended option a is enhanced additively rather than multiplicatively: ( 9 ) Ia , t=EV ( a ) +βwhere EV ( a ) is the expected value of a , and β is a free parameter representing the attention-based enhancement of accumulation . Together with value-based attentional capture , this means that MIVAC assumes that options of higher value receive more attention , and that the evidence for an attended option is accumulated faster ( but MIVAC does not assume that there is an interaction of value and attention , in the sense that the increase of accumulation due to attention is stronger for higher- compared to lower-valued options ) . Finally , we assumed that participants can detect that the distractor is unavailable but occasionally fail to do so . This assumption was necessary because participants usually followed the instructions of the Chau2014 paradigm and avoided choosing D in most of the trials but still picked it more often than the empty quadrant: In trials with D present , there were 6 . 7% choices of D and only 0 . 1% choices of the empty quadrant ( t ( 122 ) = 12 . 17; p<0 . 001 ) ; in trials without D , there were only 3 . 1% choices of both empty quadrants together . Therefore , we assumed that after every 100 ms , D can be identified as being unavailable with probability π , which is a free parameter ( restricting the identification of D to every 100 ms was again necessary to maintain fast simulations of the model; note that 100 ms also corresponds to the point in time at which participants were told which of the options D is ) . If D is identified , the input to its accumulator is set to 0 ( plus noise ) just as for the empty quadrant . The probability to fixate D is also affected by its identification ( i . e . , nEVD is set to 0 in Equation 8 ) . Taken together , MIVAC has four free parameters , the standard deviation σ of accumulation from the baseline MI model , the value-based attentional capture parameter γ , the attention-based enhancement β of accumulation , and the probability π to identify D as being unavailable ( see Table S8 in Supplementary file 1 for estimated parameter values ) . In addition to the baseline MI and MIVAC , we tested three simplified versions of MIVAC that each lacked one of the additional features ( i . e . , either γ , β , π , or all of them were set to 0 ) . It should be noted that the proposed model does not predict violations of IIA , such as the attraction or the phantom-decoy effect ( for which we found evidence in the experiment with long deliberation time ) . We refrained from implementing this because our central goal was to model the Chau2014 paradigm with short deliberation time , for which we did not find any IIA violations , and because we wanted to maintain a high speed of model simulations . Importantly , however , MIVAC can be extended to allow predicting IIA violations straightforwardly , for instance by implementing a stage of attribute-wise comparisons that modify the input It ( Gluth et al . , 2017; Hunt et al . , 2014; Roe et al . , 2001; Trueblood et al . , 2014; Usher and McClelland , 2004 ) . In the section Modeling procedures for MIVAC VI: Extending MIVAC to allow for violations of IIA , we provide one example of how such an extension could work , and show that it allows accounting for the IIA violation observed in the experiment with long deliberation time . Thus , in principle MIVAC is compatible with the rich literature on violations of IIA in value-based and perceptual decision making ( Berkowitsch et al . , 2014; Gluth et al . , 2017; Huber et al . , 1982; Mohr et al . , 2017; Rieskamp et al . , 2006; Spektor et al . , 2018a; Spektor et al . , 2018b; Trueblood et al . , 2013; Tsetsos et al . , 2012 ) . As mentioned , parameter estimation required time-consuming simulations of the models . Similar to our previous work ( Gluth et al . , 2013b ) , we chose a step-size Δt of 10 ms for fast simulations , and approximated the likelihood by simulating every trial 100 times . Trial-wise predictions were truncated to a minimum probability of . 01 and a maximum probability of . 99 ( Gluth et al . , 2013b; Nassar and Frank , 2016; Rieskamp , 2008 ) . Note that the estimation of parameters was based on predicting choices but not RT , because our main goal was to explain the choice data , and because predicting RT would have required more simulations per trial and estimating the non-decision time ( and possibly also the decision threshold ) , which would have prolonged computations to an infeasible amount of time . As shown in Figure 7D , however , the model reproduces all of the observed RT effects ( i . e . , higher value differences and sums of HV and LV decrease RT , higher values of D and the presence of D increase RT ) . Parameter estimation was realized by combining an initial grid search algorithm to obtain reasonable starting values , which were then passed on to a constrained simplex minimization algorithm ( Nelder and Mead , 1965 ) as implemented MATLAB’s fminsearchcon to obtain the final estimates . For the grid search , 31 steps per parameter were used , equally spaced within the following bounds: σ between 5 and 50 , γ between −1 . 12 and 3 . 08 , β between −2 . 1 and 18 . 9 , π between 0 and 1 ( the same values were used as constrains during the simplex minimization ) . The simplex algorithm was run 16 times in total , 3 times with the parameter set obtained from the best grid search model fit , 9 to 13 times with a parameter set randomly selected from the best 1% of grid search model fits ( 20% in case of the baseline MI ) , and 0 to 4 times with the best parameter set obtained from the simplex search of nested models ( e . g . , all other models are nested in MIVAC , so the algorithm for MIVAC was started 4 times with the four other models’ best parameters as starting values ) . The final estimates of parameters were taken from the simplex search with the best model fit . Only trials of the Chau2014 set with D present were used to estimate parameters . This allowed testing a generalization ( Busemeyer and Wang , 2000 ) of the models to the trials without D ( Figure 7B ) and to the novel trials that we added in Experiment 1 to 3 ( Figure 7E ) . The central goal of computational modeling was to compare MIVAC against simpler versions that lacked either one or all of the novel mechanisms . Thus , model comparison comprised the baseline MI , MIVAC , and three versions of MIVAC without either γ , β , or π ( which were set to 0 for the respective model ) . Since all other models are nested within MIVAC , they could be compared using a likelihood ratio test ( Lewandowsky and Farrell , 2011 ) . However , we used the Bayesian Information Criterion ( BIC; Schwarz , 1978 ) to compare the models , because it is more conservative when comparing the most complex model against simpler ones , and because the BIC allows to estimate participant-wise model evidence: For each participant , we classified model evidence as ‘weak’ , ‘positive’ , ‘strong’ , or ‘very strong’ , when the BIC difference x between the best and the second-best model was x < 2 , 2 < x < 6 , 6 < x < 10 , or x > 10 , respectively ( Raftery , 1995 ) . Following our previous work ( Gluth et al . , 2017 ) , we tested the generalizability of the models by comparing their deviances for the trials without D and for the novel trials , which were not used for parameter estimation . We also compared the models on a qualitative level with respect to i . ) how well they predict choices in D present trials , ii . ) how well their predictions can be generalized to trials without D , and iii . ) how well they account for the dependency of choosing D on the value of D ( Figure 7C ) . Note , that all quantitative and qualitative model comparisons are based on out-of-sample predictions by running 10’000 new simulations per trial , participant , and model using the estimated parameters . A simple parameter recovery analysis ( in which we generated synthetic data using MIVAC’s estimated parameters of each participant and the 150 trials from the Chau2014 paradigm and re-estimated parameters for these synthetic data ) confirmed that the model is identifiable with the set of trials used in Chau2014 and our study ( for all parameters , correlations between data-generating and re-estimated parameter values were significantly positive at p<0 . 001; see Figure 7—figure supplement 1 ) . Similar to our behavioral analyses , MIVAC assumes that people are able to integrate magnitude and probability information to estimate an option’s EV ( since the EV serves as the basic input signal to each accumulator; see Equation 9 ) . Because our analyses of the choice data suggested that an EU-based choice model provided a more accurate account of the data than an EV-based model ( see Figure 5—figure supplement 1 ) , we reran MIVAC and replaced the EVs as inputs to the options’ accumulators by the EU-based subjective value estimates . Because the range of EU-based values heavily depend on the power utility parameter α ( if 0 < α < 1 , the range is narrower than the range of EVs; if α > 1 , the range is wider than the range of EVs ) , we ensured that subjective values were kept in the same range as the EVs by the following transformation ( see Berkowitsch et al . , 2015 ) : ( 10 ) EUtransformed=minEV+EU-minEU*maxEV-minEVmaxEU-minEU We simulated responses from all models 10’000 times per trial and participant with the previously estimated parameters . As can be seen from Figure 7—figure supplement 2 , the predictions of the EU-based MIVAC model are very similar to the EV-based MIVAC model . As stated in the main text , we did not draw fixations from empirical fixation distributions ( as has been done , for instance , by Krajbich et al . , 2010 ) when estimating parameters of MIVAC for two reasons . First , we sought to avoid overfitting MIVAC by equipping it with actual fixation patterns: If the main behavioral effects had been solely due to differences in fixation patterns across options , then using those patterns would have allowed MIVAC to make accurate predictions , even though the model’s proposed features ( e . g . , value-based attentional capture ) would not have been critical . Second , without using empirical fixation distributions the model can be faster simulated and thereby fitted to the data in a feasible amount of time . Here , we show that using empirical fixation patterns does not alter the predictions of MIVAC and its simplified competitor models qualitatively . Thereto , we first analyzed mean fixation durations by splitting fixations according to the factors Option Type ( i . e . , HV , LV , D , empty quadrant ) and Fixation Type ( i . e . , first vs . middle fixations; see Krajbich et al . , 2010; Figure 7—figure supplement 3A ) . A 4×2 repeated-measures ANOVA revealed significant main effects for both Option Type ( F ( 3 , 18 ) = 29 . 51 , p < . 001 , ηp2 = . 83 ) and Fixation Type ( F ( 1 , 20 ) = 25 . 70 , p < . 001 , ηp2 = . 56 ) and a significant interaction ( F ( 3 , 18 ) = 18 . 24 , p < . 001 , ηp2 = . 75 ) . Accordingly , we approximated the empirical fixation-duration distributions by estimating the means and standard deviations of log-normal distributions ( Krajbich et al . , 2010 ) for each option and fixation type separately ( Figure 7—figure supplement 3B ) . Finally , we simulated responses from all models 10’000 times per trial and participant with the previously estimated parameters . We sampled each fixation duration from the log-normal distribution with the respective means and standard deviations per option and fixation type ( e . g . , when the first fixation in a trial was on HV , then the fixation duration was drawn from a log-normal distribution with its mean and standard deviation taken from the fit to the HV/first distribution ) . As can be seen from Figure 7—figure supplement 3C , the predictions of MIVAC do not change qualitatively when empirical fixation durations are used . Moreover , using these fixation durations does not allow the simplified competitor models to catch up with MIVAC ( Figure 7—figure supplement 3D ) . As stated in the main text , MIVAC does not predict violations of IIA but can be extended straightforwardly to do so . Here , we provide one example of such an extension and show that it allows to predict the IIA violation we obtained in the experiment with long deliberation time ( Experiment 2 , Group LP ) . For the extension , we followed the assumption made by many multi-attribute sequential sampling models , including the Multialternative Decision Field Theory ( Roe et al . , 2001 ) , the Leaky Competing Accumulator model ( Usher and McClelland , 2004 ) , and the Multi-attribute Linear Ballistic Accumulator model ( MLBA; Trueblood et al . , 2014 ) , namely that the input to the accumulator of an option i is based on comparisons of this option with all other options . In our notation , the input of option i thus becomes ( 11a ) Ii , t=∑jVi , j+I0+βif i is currently fixated , ( 11b ) Ii , t=∑jVi , j+I0if i is currently not fixated , where I0 is a free parameter of the model ( I0 ≥0 ) , representing the baseline input to each accumulator , and Vi , j represents the output of a comparison process between option i and option j . In specifying Vi , j , we follow the MLBA , according to which this comparison process is the weighted sum of the attribute-wise differences:Vi , j=wi , jM*uiM-ujM+wi , jP*uiP-ujPwhere is wi , jM is the weight given to attribute M ( for ‘magnitude’ ) in the comparison between options i and j , and uiP is the subjective representation of attribute P ( for ‘probability’ ) for option i . In the full MLBA , the subjective representations can be subject to non-linear transformations . For the sake of simplicity , however , we used the objective magnitudes and probabilities as uM and uP , respectively . The weight given to an attribute X is a function of the similarity between the options with respect to that attribute X:wi , jX=e-λ1*uiX-ujXif uiX ≥ujX , wi , jX=e-λ2*uiX-ujXif uiX <ujX , where λ1 and λ2 are two free parameters of the model that modulate the decay of attribute weights when the distance between the options’ attribute values increases . Note that this means that perceived similarity is not symmetrical: A can be perceived as more similar to B than B to A . As stated above , the outputs from this attribute-comparison layer are the inputs It to the accumulation process ( Equation 9 ) . For our simulations , we set I0 to 20 , λ1 to 3 , and λ2 to 0 . 05 . While the value for I0 was chosen simply to avoid negative inputs to accumulators , we chose to set λ1 > λ2 because then the weight wi , jX of attribute X in the comparison between option i against option j will be higher when i is better than j with respect to X compared to when it is worse , with this difference being largest at close distances ( because of the decay function that links λ to w ) . As a consequence , option i will receive an enhanced input in the presence of a similar , inferior alternative , but it will not receive a reduced input in the presence of a similar , superior alternative . Such a disproportionate attention weight is a default assumption of MLBA and predicts both the attraction and the phantom-decoy effect , that is , an increased relative choice accuracy when ( an inferior or superior ) D is more similar to HV than to LV . All other parameter values were taken from the fit of the non-extended MIVAC model . The deliberation time was set to 6’000 ms . We ran 10’000 simulations per participant . As shown in Figure 7—figure supplement 4 , the extended MIVAC is able to predict the pattern of IIA violation that we observed in Experiment 2 Group LP , namely that the probability of choosing HV is increased in the presence of a similar ( inferior or superior ) D option . We also compared MIVAC to three different multinomial logit models as representatives of standard economic choice models ( e . g . , McFadden , 2001 ) ; see Figure 7—figure supplement 5 ) . The first logit model ( ‘Exclude D’ ) assumes that participants are perfectly able to identify that D should not be chosen . Thus , irrespective of whether D is present or absent , the probability of choosing HV becomes a logistic function of the expected value ( EV ) difference between HV and LV: ( 14 ) PHV=11+e-δ× ( EV ( HV ) -EV ( LV ) ) where δ is a free parameter reflecting the participant’s sensitivity to EV differences . The second logit model ( ‘Include D’ ) assumes that participants are completely unable to identify that D should not be chosen . Thus , the probability of choosing an option i is:Pi=eδ×EV ( i ) ∑jeδ×EV ( j ) with D being part of the choice set and where the sum of EVs is taken for all choice options . The third logit model ( ‘EV and Frame Value’ ) assumes that D is treated as a regular choice option ( as in the ‘Include D’ model ) , and that choices are based on an additive combination of the options’ EVs with additional subjective values that are assigned to the differently colored frames ( which indicate whether an option is a target or a distractor; see Figure 1A ) . Formally , the overall subjective value ( SV ) of an option i is: ( 16 ) SV ( i ) EV and Frame Value=EV ( i ) +FV ( i ) where FV ( i ) is the ‘Frame Value’ of option i . We assumed that the Frame Value for distractors is 0 and the Frame Value for targets is a free parameter ( restricted to positive values ) . The choice probability is then given by Equation 15 with EV replaced by SV . Maximum likelihood estimates of the models’ parameter ( s ) were computed on the basis of trials with D present , using the same minimization algorithm as for MIVAC . Because the three models and MIVAC used different sets of data ( i . e . , the ‘Exclude D’ model used only trials in which either HV or LV were chosen , the other two logit models also included trials in which D was chosen , MIVAC further included trials in which the empty quadrant was chosen or no decision was made ) , the models could not be compared via quantitative criteria . Instead , we compared their ability to predict the observed choice proportions of HV , LV , and D in trials with D present , and of HV and LV in trials with D absent ( Figure 7—figure supplement 5 ) . The authors of Chau2014 provided us with the behavioral data of their fMRI experiment that included 21 participants . We reanalyzed the data to check for evidence of effects related to value-based attentional capture: We tested for an effect of HV-D on absolute choice accuracy , for an effect of D’s value on the propensity to choose D , and for an effect of D’s value on RT . The results of all of these analyses were in line with predictions of a value-based attentional capture account ( see Results ) . When trying to reproduce the results of the original study , we also found that the HV-D effect on relative choice accuracy ( i . e . , the central behavioral result of Chau2014 ) was only significantly negative when the interaction term ( HV-LV ) × ( HV-D ) in the same regression analysis was generated on the basis of uncentered HV-LV and HV-D variables . In multiple regression analyses , centering ( i . e . , subtracting each entry by the mean ) or standardizing ( i . e . , subtracting the mean and dividing by the standard deviation ) predictors before generating their interaction term is critical to avoid nonessential multicollinearity ( Aiken and West , 1991; Mahwah et al . , 2003; Dunlap and Kemery , 1987; Marquardt , 1980 ) . This is particularly important when one or both of the predictors do not have a meaningful 0 ( i . e . , are always positive or always negative ) . This is the case for HV-LV which is always positive by definition . Accordingly , the uncentered interaction term ( HV-LV ) × ( HV-D ) is correlated substantially with HV-LV ( r = . 287 ) and correlates particularly strong with HV-D ( r = . 862 ) ( upper row of Figure 5—figure supplement 2 ) . In contrast , the mean-centered interaction term is only weakly correlated with HV-LV ( r = . 004 ) and HV-D ( r = − . 064 ) ( lower row of Figure 5—figure supplement 2 ) . To test how susceptible the original results of HV-D are to the generation and implementation of the interaction term , we conducted several control analyses . First of all , we repeated the regression analysis on relative choice accuracy , but used the correct interaction term ( i . e . , the term generated on the basis of standardized HV-LV and HV-D predictors ) . Strikingly ( but in line with our own results ) , the effect of HV-D in the Chau2014 data was positive and not significant ( t ( 20 ) = 1 . 56 , p = . 134; see also Figure 5—figure supplement 3 , which shows the effect sizes of HV-D in all experiments when the correct vs . incorrect interaction term is used in the analysis ) . Second , we took the interaction term out of the regression analysis , which yielded similar results ( t ( 20 ) = 0 . 67 , p = . 510 ) . Third , we reanalyzed the original data in the following way: We separated the dataset by each level of HV-LV , which yielded 12 bins , and conducted the regression analysis on relative choice accuracy for each bin separately . This procedure allowed us to take out the interaction term ( together with the predictor variable HV-LV itself ) , and to test for a ‘cleaner’ effect of HV-D in each bin . As shown in Figure 5—figure supplement 4 a significantly negative effect of HV-D on relative choice accuracy is seen in only 1 of the 12 bins , but there is also a significantly positive effect in another bin . Conducting this analysis with our own data yielded very similar results . Fourth , we simulated data from hypothetical decision makers who decide solely on the basis of HV-LV ( via a logistic/soft-max choice function [see Equation 14 above] for which we set the choice-sensitivity parameter to 0 . 5 to roughly match the choice accuracy observed in the data ) . For these simulated participants , we know that their behavior is not influenced by HV-D . We conducted 1’000 simulations of 21 participants ( the sample size of Chau2014 ) and performed four logistic regression analyses , either with or without the interaction term , and either with standardized or uncentered HV-LV and HV-D predictors . We found a spurious negative influence of HV-D on relative choice accuracy when including an interaction term with uncentered predictors ( t ( 999 ) = −3 . 43 , p < . 001; upper row of Figure 5—figure supplement 2 ) . This influence was absent when including the interaction term with standardized predictors ( t ( 999 ) = 0 . 36 , p = . 717 ) or when excluding the interaction term with either standardized or uncentered predictors ( both: t ( 999 ) = −0 . 89 , p = . 375; lower row of Figure 5—figure supplement 2 ) . Taken together , the negative effect of HV-D on relative choice accuracy is a statistical artifact of the incorrect implementation of the interaction term ( HV-LV ) × ( HV-D ) into the logistic regression analysis . The reanalysis of Chau2014 raises strong doubts on the reproducibility of the original results . Yet , theoretically there might still be a true negative HV-D effect that in principle could have been found in our own four experiments ( while being absent in the original study due to its low statistical power ) . To test for the ‘replicability’ of this effect without reverting to the ( equivocal ) effect size of the original study , we conducted a recently proposed test of detectability that is independent of the reported effect size of the original study ( Simonsohn , 2015 ) . This approach uses only the sample size ( and the statistical design ) of the original study to specify the ( hypothetical ) effect size that would have given the study only 33% statistical power , which can be regarded as undoubtedly insufficient . If the effect size of the study that attempts to reproduce the effect of the original study is significantly below this d33% threshold , then one can conclude that the studied effect is not large enough to have been detectable with the original sample size . The results of this test for the HV-D effect on relative choice accuracy show that the 95% confidence intervals of the effect sizes for each of our Experiments 1 , 2 HP , 3 , and four as well for all these experiments combined and for the original study itself ( when using the correct implementation of the interaction term ) were indeed closer to 0 than the d33% threshold for the sample size of 21 participants used in Chau2014 ( Figure 5—figure supplement 3 ) . But even when using the incorrect implementation of the interaction term ( that biases the results toward more negative HV-D effects ) , the confidence intervals of the effect size for our four experiments combined are closer to 0 than the d33% threshold . Finally , we conducted a Bayesian analysis of the HV-D effects on relative ( and absolute ) choice accuracy to be able to quantify the evidence in favor of the null hypothesis ( Wagenmakers , 2007 ) . Specifically , we applied the ttestBF function of the R package BayesFactor ( with default settings ) to obtain Bayes Factors ( BF ) in favor of the null ( or alternative ) hypothesis that the HV-D regression coefficients of relative and absolute choice accuracy are equal to ( or deviate from ) 0 . We included the 123 participants from Experiments 1 , 2 HP , 3 , and 4 who performed the task under high time pressure . For the HV-D effect on relative choice accuracy we obtained a BF01 in favor of the null hypothesis of 20 . 05 when testing against the alternative hypothesis of a negative HV-D effect ( Figure 5—figure supplement 5 , upper panel ) . This means the data are about twenty times more likely to have occurred under the null hypothesis , which is also seen as strong evidence for the null hypothesis ( Kass and Raftery , 1995 ) . When testing against the alternative hypothesis of a positive HV-D effect ( which would be the opposite of what Chau2014 had reported and in line with the divisive normalization account of Louie et al . , 2013 ) , we obtained a BF01 in favor of the null hypothesis of 3 . 14 , which can be considered as positive evidence for the null hypothesis ( Kass and Raftery , 1995 ) . For the HV-D effect on absolute choice accuracy we obtained a BF10 in favor of the alternative hypothesis of 4 . 07 × 1013 , which is seen as decisive evidence ( Figure 5—figure supplement 5 , lower panel ) . | A man in a restaurant is offered a choice between apple or blueberry pie , and chooses apple . The waiter then returns a few moments later and tells him they also have cherry pie available . “In that case” , replies the man , “I’ll have blueberry” . This well-known anecdote illustrates a principle in economics and psychology called the independence principle . This states that preferences between two options should not change when a third option becomes available . A person who prefers apple over blueberry pie should continue to do so regardless of whether cherry pie is also on the menu . But , as in the anecdote , people often violate the independence principle when making decisions . One example is voting . People may vote for a candidate who would not usually be their first choice only because there is also a similar but clearly less preferable candidate available . Such behavior provides clues to the mechanisms behind making decisions . Studies show , for example , that when people have to choose between two options , introducing a desirable third option that cannot be selected – a distractor – alters what decision they make . But the studies disagree on whether the distractor improves or impairs performance . Gluth et al . now resolve this controversy using tasks in which people had to choose between rectangles on a computer screen for the chance to win different amounts of money . Contrary to a previous study , their four experiments showed that a high-value distractor did not change how likely the volunteers were to select one of the two available options over the other . Instead , the distractor slowed down the entire decision-making process . Moreover , volunteers often selected the high-value distractor despite knowing that they could not have it . One explanation for such behavior is that high-value items capture our attention automatically even when they are irrelevant to our goals . If a person likes chocolate cake , their attention will immediately be drawn to a cake in a shop window , even if they had no plans to buy a cake . Eye-tracking data confirmed that volunteers in the above experiments spent more time looking at high-value items than low-value ones . Those volunteers whose gaze was distracted the most by high-value items also made the worst decisions . Based on the new data , Gluth et al . developed and tested a mathematical model . The model describes how we make decisions , and how attention influences this process . It provides insights into the interplay between attention , valuation and choice – particularly when we make decisions under time pressure . Such insights may enable us to improve decision-making environments where people must choose quickly between many options . These include emergency medicine , road traffic situations , and the stock market . To achieve this goal , findings from the current study need to be tested under more naturalistic conditions . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2018 | Value-based attentional capture affects multi-alternative decision making |
Sarcoidosis is a complex systemic granulomatous disease of unknown etiology characterized by the presence of activated macrophages and Th1/Th17 effector cells . Data mining of our RNA-Seq analysis of CD14+monocytes showed enrichment for metabolic and hypoxia inducible factor ( HIF ) pathways in sarcoidosis . Further investigation revealed that sarcoidosis macrophages and monocytes exhibit higher protein levels for HIF-α isoforms , HIF-1β , and their transcriptional co-activator p300 as well as glucose transporter 1 ( Glut1 ) . In situ hybridization of sarcoidosis granulomatous lung tissues showed abundance of HIF-1α in the center of granulomas . The abundance of HIF isoforms was mechanistically linked to elevated IL-1β and IL-17 since targeted down regulation of HIF-1α via short interfering RNA or a HIF-1α inhibitor decreased their production . Pharmacological intervention using chloroquine , a lysosomal inhibitor , decreased lysosomal associated protein 2 ( LAMP2 ) and HIF-1α levels and modified cytokine production . These data suggest that increased activity of HIF-α isoforms regulate Th1/Th17 mediated inflammation in sarcoidosis .
Sarcoidosis is a systemic granulomatous disease of unknown etiology that is characterized by extensive local inflammation and granuloma formation in different organs with an increase in T-helper type 1 ( Th1 ) mediated cytokine production ( Hunninghake et al . , 1994; Iannuzzi et al . , 2007; Miyara et al . , 2006; Rastogi et al . , 2011 ) . Pulmonary involvement in sarcoidosis is the leading cause of morbidity and mortality . In the lungs , the presence of activated macrophages and the expansion of oligoclonal T and B cells suggest sustained activation of inflammatory pathways in this disease ( Fazel et al . , 1992; Iannuzzi et al . , 2007 ) . Activated macrophages , monocytes , and T cells in sarcoidosis produce a plethora of cytokines including TNF-α , IL-1β , interferon ( IFN ) gamma , IL-17 , and others ( Facco et al . , 2011; Müller-Quernheim , 1998; Rastogi et al . , 2011; Talreja et al . , 2016 ) . Previously , we have shown that sarcoidosis bronchoalveolar lavage ( BAL ) cells and alveolar macrophages ( AMs ) , unlike those from healthy controls , exhibit high constitutively active p38 and lack dual specificity phosphatase ( DUSP1 or MKP-1 ) . The sustained p38 activation directly controls expression of several cytokines in sarcoidosis AMs and monocytes and the modulation of p38 regulates T cell responses ( Rastogi et al . , 2011; Talreja et al . , 2016 ) . Recently , we performed RNA-sequencing ( RNA-seq ) in sarcoidosis monocytes and identified altered gene expression profiles and cellular pathways ( Talreja et al . , 2017 ) . These were: metabolic including glycolysis and lipolysis , phagocytosis , inflammation , oxidative phosphorylation , and HIF signaling pathways ( Talreja et al . , 2017 ) . Among differentially expressed genes in sarcoidosis monocytes , we found a large number of genes containing hypoxia response elements ( HREs ) in their regulatory regions and , by pathway analysis , enrichment of hypoxia inducible factor signaling pathways . Furthermore , in an independent study applying 1H nuclear magnetic resonance ( NMR ) -based analysis , we identified metabolic and mitochondrial alterations in sarcoidosis ( Geamanu et al . , 2016 ) . Based on these observations , we hypothesize that HIF-isoform expression plays an important role in the maintenance of inflammation ( Rastogi et al . , 2011; Talreja et al . , 2016 ) , metabolic imbalance , and mitochondrial dysfunction in sarcoidosis ( Geamanu et al . , 2016 ) . The oxygen-sensitive transcription factors HIF-1α and HIF-2α are key transcriptional regulators of hypoxia-associated genes to adapt to decreased availability of O2 ( Semenza , 2011; Wang and Green , 2012 ) . In the presence of O2 , cytosolic HIF-α isoforms are hydroxylated by prolyl-hydroxylases ( PHD ) through an iron dependent mechanism , which prevents heterodimerization with HIF-1β ( ARNT ) and consequent nuclear translocation as an active transcription factor ( Palazon et al . , 2014; Semenza , 2003; Semenza , 2011 ) . HIF transcription factors alter the expression of various genes involved in metabolism , cell differentiation , proliferation , and angiogenesis in hypoxic tissues . Although the role of HIF-α isoforms in hypoxia and cancer is well studied , there is a knowledge gap regarding their role in regulating immune cells under normoxic conditions . The role of HIF-1α in sarcoidosis has not been studied . In the current study , we applied a combination of transcriptional and functional approaches to investigate the role of HIF-1α in mediating the inflammatory immune response in AMs , monocytes , and PBMCs of sarcoidosis patients as compared to healthy controls . Because sarcoidosis predominantly affects the lungs , we carried out the functional studies using AMs to determine the lung immune responses , while monocytes and PBMCs were used to assess peripheral immunity . Under normoxic conditions we found enhanced expression and activity of HIF-1α in sarcoidosis AMs and monocytes . Furthermore , HIF-1α expression was directly correlated with IL-1β production in AMs and PBMCs . Down regulation of HIF-1α expression via short interfering RNA ( siRNA ) decreased IL-1β in sarcoidosis AMs , while decreased HIF-1α expression in PBMCs decreased IL-1β and IL-17 in response to anti-CD3 challenge .
Patients ( Table 1 and Materials and methods ) were ambulatory outpatients who were not hypoxic . Differentially expressed ( DE ) genes between sarcoidosis monocytes and healthy monocytes previously determined ( Talreja et al . , 2017 ) were subjected to pathway analysis . The pathway analysis showed impaction of metabolic pathways , including oxidative phosphorylation , purine and pyruvate metabolism in sarcoidosis . Because most of genes in these pathways showed the presence of hypoxia response elements ( HREs ) , we further focused on interrogation of the HIF-pathway . Figure 1A shows the heat map of HIF signaling genes in monocytes . There are clear differences in the intensity and expression of genes related to the HIF pathway in monocytes of healthy controls and sarcoidosis subjects . Next , we compared the expression of selected genes related to HIF transcription factor activity . The transcription factor aryl hydrocarbon receptor nuclear translocator ( ARNT , also known as HIF-1β ) heterodimerizes with HIF-1α to form a transcriptional active complex ( Wolff et al . , 2013 ) . The gene count between sarcoidosis and healthy control subjects demonstrate significantly higher ARNT gene expression in sarcoidosis monocytes ( Figure 1B ) . Endothelial PAS domain protein 1 ( EPAS1 ) , also known as HIF-2α , is a hypoxia inducible transcription factor ( Hu et al . , 2003; Thompson et al . , 2014 ) . The EPAS1 gene count between sarcoidosis and healthy control subjects demonstrates significantly higher EPAS1 expression in sarcoidosis monocytes ( Figure 1C ) . EP300 is a co-activator important for transcriptional activity of HIFs ( Palazon et al . , 2014 ) . Similarly , we found higher p300 gene expression in sarcoidosis monocytes as compared to healthy controls ( Figure 1D ) . However , there were no differences in HIF-1α gene transcripts between the two groups . Since HIF-1α is known to be predominantly regulated through modification of its protein stability ( Lee et al . , 2004; Salceda and Caro , 1997 ) , we evaluated HIF-1α and HIF-2α protein abundance in AMs and monocytes of sarcoidosis patients , isolated as described in Materials and methods . AMs or monocytes were cultured ex vivo under normoxic conditions . Western analysis of cell lysates probed with antibody against HIF-1α showed increased HIF-1α protein expression in sarcoidosis AMs and monocytes ( Figure 2A and B ) . Similar results were seen for HIF-2α protein expression ( Figure 2C and D ) . Since HIFα heterodimerizes with ARNT ( also known as HIF-1β ) , translocates to the nucleus , and recruits transcriptional coactivator p300 to transactivate target genes containing hypoxia-responsive elements ( HREs ) ( Semenza , 2003; Talwar et al . , 2017a; Talwar et al . , 2017b ) , we also examined their protein expression . Sarcoidosis AMs also show a higher expression of ARNT ( Figure 2E and F ) and p300 ( Figure 2E and G ) . Similarly , we evaluated the HIF-1α protein abundance in isolated monocytes from sarcoidosis subjects and healthy controls and found significantly higher HIF-1α expression ( Figure 2H and I ) . However , in contrast to increased HIF-2α gene transcripts , we did not detect HIF-2α in either sarcoidosis or control monocytes at the protein level . Because the lack of detection could have been due to low protein abundance in monocytes or lower sensitivity of antibody epitope , we compared the HIF-1α and −2α expression by flow cytometry . Figure 2J shows FSC-A/SSC-A gating . FACS analysis of PBMCs double stained for CD14 and HIF-1α or HIF-2α shows that in healthy controls 5–9% of PBMCs are CD14+HIF-1α+ , whereas in sarcoidosis 20% to 35% of PBMCs are CD14+ HIF-1α+ . Analysis of CD14+ monocytes based on the expression of HIF-1α shows 25–60% HIF-1α+ CD14+ monocytes in controls , whereas in sarcoidosis HIF-1α+ CD14+ monocytes are 64–96% ( K ) . Interestingly , in healthy controls 0–0 . 1% of PBMCs are CD14+ HIF-2α+ , whereas in sarcoidosis 1–3% of PBMCs are CD14+ HIF-2α+ . It shows that in healthy controls the percentage of HIF-2α+ CD14+ monocytes is negligible , whereas in sarcoidosis there is higher percentage of HIF-2α+ CD14+monocytes ( 5–9% ) ( Figure 2L ) . Thus , these results show that sarcoidosis AMs and peripheral monocytes exhibit increased expression of HIF isoforms compared to healthy controls . These data suggest a different protein expression profile of HIF-2α in lung macrophages versus peripheral monocytes with low abundance in monocytes versus AMs . To further confirm increased expression of HIF-1α protein in sarcoidosis and to determine whether HIF-1α accumulates in the nucleus , we immunostained AMs using specific an antibody against HIF-1α . Images were analyzed by immunofluorescent microscopy ( AX10 , Zeiss ) . We quantitated the percentage of cells showing HIF-1α expression ( Figure 3A and B ) in sarcoidosis . The staining is representative of one out of the five patients . It shows that about 60–90% of AMs express HIF-1α . Images ( Figure 3C–H ) were analyzed by confocal laser scanning microscopy ( CLSM-310 , Zeiss ) . Confocal microscopy images show nuclei stained with DAPI ( blue ) in a single AM ( C ) and a multinucleated giant cell ( D ) , nuclear and cytoplasmic accumulation of HIF-1α in green ( E and F ) , overlay image shows nuclear co-localization of HIF-1α ( G and H ) . We saw enhanced expression and accumulation of HIF-1α in the cytoplasm and nuclei of sarcoidosis AMs , both in a single AM ( Figure 3E ) and in a multinucleated giant cell ( Figure 3F ) that are known to be characteristic cells in sarcoidosis granuloma . HIF-1α is highly expressed and overlay images show that HIF-1α accumulates in nuclei ( Figure 3G and H ) as compared to cytoplasm . To further explore the expression seen in sarcoidosis AMs , we assessed the presence of HIF-1α in lung biopsies of patients with sarcoidosis . Positive immunostaining was seen in multinucleated giant cells of granulomas as well as macrophages ( Figure 3I and J , thick arrow ) , whereas fibroblasts and normal lungs lack HIF-1α expression . Negative staining was done by using isotype control antibody ( Figure 3K ) . Similarly , we observed increased HIF-1α immunostaining signal in sarcoidosis liver and skin tissue samples . These results further confirmed that HIF-1α accumulates in sarcoidosis granulomatous tissues . HIF-1α is a critical transcription factor regulating metabolic reprogramming during inflammation , in part through upregulation of the SLC2A1 gene encoding glucose transporter ( Glut ) 1 ( Chen et al . , 2001 ) . HIF-1α and Glut1 upregulation contribute to production of several pro-inflammatory cytokines including IL-1β ( Talwar et al . , 2017a; Talwar et al . , 2017b; Tannahill et al . , 2013 ) . Therefore , we evaluated the expression of Glut1 and pro-IL-1β at baseline in AMs and monocytes from sarcoidosis and control subjects . Sarcoidosis AMs exhibited a variable amount of Glut1 and pro-IL-1β ( 18/18 patients ) but only 1 out of 10 healthy controls showed expression ( Figure 4A and B ) . We found similar results for pro-IL-1β in monocytes ( Figure 4C and D ) . Furthermore , increased pro-IL-1β expression directly correlated with Glut1 and HIF-1α expression in sarcoidosis AMs ( Figure 4E ) . To determine whether increased pro-IL-1β expression in sarcoidosis leads to released IL-1β , we measured secreted IL-1β in the conditioned media of AMs and monocytes cultured in the absence or presence of LPS via ELISA . The results showed that unstimulated and LPS-stimulated cultured sarcoidosis AMs and monocytes secrete higher IL-1β as compared to healthy controls ( Figure 4F and G ) . These data suggest that increased expression of HIF-1α leads to increased IL-1β production in sarcoidosis patients . The interleukin one receptor antagonist ( IL-1Ra ) is mainly secreted by monocytes , macrophages , and neutrophils . IL-1Ra ( IL-1RII ) competitively binds to IL-1β and forms a nonsignaling complex IL-1Ra to the surface receptors for IL-1β and inhibits the effect of IL-1β on cells ( Arend , 2000; Janson et al . , 1991 ) . Since the sarcoidosis AMs produced significantly high levels of IL-1β , we assessed the conditioned media for the secreted IL-1Ra . Figure 4H shows that sarcoidosis AMs produced significantly high levels of IL-1Ra as compared to control AMs . Similarly , sarcoidosis PBMCs ( Figure 4I ) produced high levels of IL-1Ra as compared to control PBMCs . IL-1β is regulated at the transcriptional level through expression of several transcription factors including Signal Transducer and Activator of Transcription ( STAT ) 3 , HIF-1α , and others ( Samavati et al . , 2009; Talwar et al . , 2017b ) . To determine the relative contribution of increased HIF-1α in IL-1β production in sarcoidosis AMs , we transiently transfected sarcoidosis AMs with either non-targeted siRNA or HIF-1α targeted siRNA . After 24 hr of transfection , cells were treated with LPS ( 100 ng/mL ) . Targeted downregulation of HIF-1α via siRNA of sarcoidosis AMs led to a significant reduction ( about 50% ) in HIF-1α ( Figure 5A and B ) and pro-IL-1β ( Figure 5C and D ) protein expression . To determine the specificity of targeted downregulation of HIF-1α on other cytokines , we assessed the conditioned medium for IL-1β and IL-10 production and found significantly decreased IL-1β production ( Figure 5E ) . However , HIF-1α inhibition did not inhibit IL-10 production ( Figure 5F ) . Similar to AMs , the targeted down regulation of HIF-1α in sarcoidosis PBMCs resulted in decreased production of IL-1β in response to LPS ( Figure 6A ) ; the effect of HIF-1α inhibition was specific for IL-1β since there was no significant effect on IL-10 production ( Figure 6B ) . These results clearly show that HIF-1α expression regulates IL-1β production in sarcoidosis AMs and PBMCs . Recent work has shown that the HIF transcription factors are key elements in the control of immune cell metabolism and function in macrophages , B-cells , and T-cells ( Palazon et al . , 2014; Wang and Green , 2012 ) . T helper 17 cells ( Th17 ) represent a lineage of effector T cells critical in host defense and autoimmunity . It is has been shown that Th1 and Th17 cells contribute to sarcoidosis pathology ( Ramstein et al . , 2016 ) . Based on this , we hypothesize that the HIF-1α inhibition may also modulate IL-1β and IL-17 production in response to anti-CD3 challenge . First , we assessed the effect of anti-CD3 activation on the production of IL-1β and IL-17 in healthy controls and sarcoidosis PBMCs . PBMCs were treated with anti-CD3 for 24 hr and the conditioned media were assessed for IL-1β and IL-17 production . Sarcoidosis PBMCS were seen to produce significantly higher levels of IL-1β ( Figure 6C ) and IL-17 ( Figure 6D ) . To investigate the contribution of HIF-1α in Th1/Th17 cytokine production , we investigated the effect of targeted downregulation of HIF-1α in PBMCs in response to anti-CD3 challenge on the production of various inflammatory cytokines . Inhibition of HIF-1α by siRNA significantly decreased the production of anti-CD3 induced IL-1β ( Figure 6E ) , IL-17 ( Figure 6F ) , and IL-6 ( Figure 6G ) . However , inhibition of HIF-1α did not decrease IFN-γ ( Figure 6H ) and IL-10 ( Figure 6I ) production . These results suggest that HIF-1α specifically regulates IL-1β and IL-17 in sarcoidosis . To confirm our results , we used echinomycin , a small molecule inhibitor of HIF-1α that has been shown to inhibit HIF-1α DNA binding activity ( Tang and Yu , 2013; Vlaminck et al . , 2007 ) . We evaluated the effect of echinomycin HIF-1α inhibition on anti-CD3-induced IL-1β and IL-17 production and T cell activation in sarcoid PBMCs . To do so , cultured sarcoidosis PBMCs were pre-treated with echinomycin in vitro , then activated with anti-CD3 in the presence of rIL-2 , followed by determination of activated CD4+CD25+ T-cells by flow cytometry and measurement of cytokines by ELISA . Our results showed that PBMCs of patients with sarcoidosis ( n = 23 ) exhibit higher expression for activated CD4+CD25+T cells ( mean ± SEM , 11 . 08 ± 5 . 32% as compared to healthy ( n = 7 ) controls ( mean ± SEM , 5 . 16 ± 2 . 71% , p < 0 . 05 ) . Figure 7A shows that PBMCs of a patient with sarcoidosis exhibited higher expression for activated CD4+CD25+T cells ( 10% ) , further increasing to 50% in response to anti-CD3 stimulation ( Figure 7B ) . Pre-treatment of PBMCs with echinomycin decreased the number of activated T cells ( 3% ) at base line ( Figure 7C ) and in response to anti-CD3 stimulation to 15% ( Figure 7D ) . Furthermore , pretreatment with echinomycin significantly decreased both baseline and anti-CD3 induced IL-1β production ( Figure 7E ) . Similarly , pretreatment with echinomycin significantly decreased anti-CD3 induced IL-17 ( Figure 7F ) and IL-6 ( Figure 7G ) production in sarcoidosis PBMCs . Chloroquine ( CHQ ) is an anti-malarial drug and remains an integral treatment for systemic inflammatory diseases such as systemic lupus erythematosus and sarcoidosis ( Lee et al . , 2011; Morse et al . , 1961 ) . CHQ inhibits lysosomal degradation/autophagy either by altering lysosomal acidification or inhibiting the levels of lysosomal associated proteins ( LAMP ) ( He et al . , 2011; Ma et al . , 2012; Rubinsztein et al . , 2007 ) . We hypothesized that CHQ modulates LAMP2 , HIF-1α , and HIF-2α levels and cytokine production in sarcoidosis AMs and PBMCs . To examine this hypothesis , isolated AMs were pre-treated with CHQ and then activated with LPS . Interestingly , CHQ decreased LAMP2 levels and both HIF-1α ( by approximately 50% ) and HIF-2α protein expression ( by approximately 65% ) in sarcoidosis AMs after LPS stimulation ( Figure 8A–D ) . Furthermore , CHQ significantly decreased ( 70% ) the expression of pro-IL-1β ( Figure 8D and E ) . Similarly , measurement of released IL-1β in conditioned medium was significantly decreased both at baseline and in response to LPS stimulation ( Figure 8F ) . To assess the effect of CHQ on IL-1β and IL-17 production by sarcoidosis PBMCs , cultured PBMCs were pre-treated with CHQ in vitro and then activated with anti-CD3 . CHQ significantly decreased anti-CD3 induced IL-1β ( Figure 8G ) and IL-17 ( Figure 8H ) production in sarcoidosis PBMCs ( p < 0 . 05 ) .
Sarcoidosis is a chronic granulomatous disease with aberrant immune response to undefined environmental or infectious triggers ( Iannuzzi et al . , 2007 ) . How specific antigens lead to a sustained granulomatous inflammation in sarcoidosis is largely unknown . Our novel RNA-seq data showed aberrant metabolic pathways and enrichment of DE genes for HIF pathways in monocytes of sarcoidosis patients ( Talreja et al . , 2017 ) , confirming our previous metabolomics data showing aberrant metabolic pathways including increased glycolysis and malfunctional tricarboxylic acid ( TCA ) cycle in sarcoidosis ( Geamanu et al . , 2016; Talreja et al . , 2017 ) . In the current study , we investigated the role of HIF-isoforms in sarcoid alveolar macrophages and blood monocytes as well as PBMCs . Alveolar macrophages and monocytes have a central role in the maintenance of immunological homeostasis in response to pathogens providing an important host-defense ( Aberdein et al . , 2013 ) . In sarcoidosis , both cell types are in an activated state and produce spontaneous ex vivo cytokines and chemokines including , IL-1β , TNF-α , IL-6 , IL-18 , and others ( Gracie et al . , 2003; Müller-Quernheim , 1998; Rastogi et al . , 2011; Rolfe et al . , 1993 ) . Our current study confirms our previous findings that IL-1β plays an important role in sarcoidosis ( Rastogi et al . , 2011; Talreja et al . , 2016 ) . In addition , we find increased IL-1Ra in sarcoidosis AMs and PBMCs , suggesting activation of the IL-1 pathway . IL-1Ra is a member of the IL-1 family , whose production is stimulated by many substances including cytokines and bacterial or viral components; it has been suggested to act as a decoy receptor and is a natural inhibitor for the biologically active IL-1β ( Lang et al . , 1998 ) ; ( Arend , 2000; Santarlasci et al . , 2013 ) . In several inflammatory diseases , including lupus and Crohn's disease ( CD ) , elevated IL-1β production is associated with IL-1Ra ( Cominelli and Pizarro , 1996 ) . Our data are in line with previous studies showing increased IL-1Ra in sarcoidosis ( Mikuniya et al . , 2000; Rolfe et al . , 1993 ) . Further studies need to delineate the clinical role of IL-1Ra in sarcoidosis . Here , we show that sarcoidosis AMs and monocytes in normoxic ex vivo culture conditions and without any stimulation exhibit constitutively active HIF-1α and HIF-1β ( ARNT ) along with its coactivator , p300 . Furthermore , in situ HIF-1α immune staining of sarcoidosis lung biopsies demonstrated HIF-1α abundance in the center of granulomatous tissue and in multinucleated giant cells . We found that a higher percentage of CD14+ monocytes express HIF-1α and HIF-2α in sarcoidosis subjects as compared to controls . Our data show that the increased HIF-1α expression is coupled to increased Glut1 protein levels , and enhanced IL-1β , IL-6 and IL-17 production . Downregulation of HIF-1α via siRNA or chemical inhibitors in sarcoidosis PBMCs leads to a decrease in IL-6 and IL-17 production at baseline and in response to anti-CD3 stimulation . In sarcoid subjects HIF-2α was predominantly expressed in the lung macrophage population whereas sarcoidosis monocytes showed lower levels of HIF-2α . HIF-2α downregulation had no significant effect on IL-1β and IL-17 production in sarcoidosis ( data not shown ) . We speculate that HIF-2α regulates other macrophage functions such as phagocytosis and cell metabolism . Classically , sarcoidosis granulomas feature activated antigen presenting cells initiating adaptive immune responses with an increase in activated CD4+T-cells and Th1 mediated cytokines . Recently , it has been shown that Th17+/CD4+T cells are increased in sarcoidosis granulomatous tissue and peripheral blood ( Facco et al . , 2011; Ostadkarampour et al . , 2014; Ramstein et al . , 2016 ) . Recent studies indicated that IL-1β plays a critical role in regulation of Th1/Th17 cells in response to commensal microbes ( Duhen and Campbell , 2014 ) . IL-1β promotes Th17 differentiation from naive CD4+ T cells by enhancing IL-1 receptor expression ( Lee et al . , 2010 ) . Furthermore , IL-1 synergizes with IL-6 to regulate Th17 differentiation and effector Th17 cell function through regulation of transcription factors , including IRF4 and RORγt ( Chung et al . , 2009 ) . Thus , in sarcoidosis increased IL-1β and IL-6 explains Th17 differentiation . Previously , our group and other showed increased IL-6 production in AMs and PBMCs of sarcoidosis subjects at baseline and in response to TLR or NLR ligands ( Rastogi et al . , 2011; Talreja et al . , 2016 ) . Levels of IL-6 may be important in progression of fibrotic lung changes in sarcoidosis ( Le et al . , 2014 ) . Our data indicate that downregulation of HIF-1α via siRNA or chemical inhibitor reduces IL-6 production by sarcoid PBMCs . HIF-1α and HIF-2α are two critical transcription factors that regulate an array of genes involved in inflammation , angiogenesis , metabolic reprogramming , mitochondrial function , T-cell differentiation and Th17 development ( Cummins et al . , 2016; Dang et al . , 2011; Nizet and Johnson , 2009; Palazon et al . , 2014; Phan and Goldrath , 2015 ) . Upregulation of HIF isoform plays a critical role in providing metabolic reprogramming in myeloid cells that is required to develop trained immunity for a robust immune response ( Cheng et al . , 2014 ) . It has been shown that mice with a myeloid cell-specific defect in HIF-1α were unable to mount a trained immune response against bacterial sepsis ( Cheng et al . , 2014; Netea et al . , 2016 ) . Trained immunity is associated with profound metabolic reprogramming in macrophages ( Yao et al . , 2018 ) , dendritic cells , and natural killer cells ( Netea et al . , 2016 ) . New mounting evidence indicates that metabolic reprogramming , including upregulation of glycolysis and depression of the TCA cycle , is a required metabolic switch for the development of innate memory , which in turn leads to upregulation of inflammatory cytokines including IL-1β and IL-17 . Similar to cancer metabolism , during inflammation aerobic glycolysis ( Warburg effect ) plays an important role in the maintenance of cellular energy supply ( Koppenol et al . , 2011; Warburg , 1956 ) . Sarcoidosis AMs and monocytes exhibit a phenotype resembling the Warburg effect or trained immunity exhibiting an abundance of HIF isoforms , higher expression for Glut1 , and higher production of IL-1β and IL-17 . Glut1 is regulated by HIF-1α transcriptional activity and its elaboration is an important step in the metabolic switch from oxidative phosphorylation to glycolysis ( Chen et al . , 2001 ) . Fluorodeoxyglucose positron emission tomography ( FDG PET ) scans are commonly used to identify metabolic activity in cancer and PET scans have been shown to be useful in active sarcoidosis ( Avril , 2004; Ben-Haim and Ell , 2009 ) . Increased Glut1 levels may explain the observed increased FDG uptake in PET/CT scans in active sarcoidosis ( Sobic-Saranovic et al . , 2013 ) . Despite the importance of HIF signaling , the role of HIF-1α and HIF-2α in lung diseases has not been established and only a few studies addressed the role of HIFs in primary human immune cells . One previous study reported increased HIF-1α mRNA in lymphocytes of peripheral blood but a decreased mRNA level in HIF-1α BAL cells . In contrast to our study one prior study reported decreased HIF-1α mRNA and protein expression in sarcoidosis tissue biopsies ( Tzouvelekis et al . , 2012 ) , although the same study reported increased expression of VEGF , which is directly regulated by HIF ( Tzouvelekis et al . , 2012 ) . The discrepancy of the results may be due to stages of the disease or evaluation of heterogeneous cell populations . Several pathways including the PI3 kinase , mTOR , MEK/ERK , GSK3β , and p38 pathways have been proposed to regulate LPS mediated HIF-1α expression and stabilization ( Palazon et al . , 2014; Peyssonnaux et al . , 2007; Talwar et al . , 2017a; Talwar et al . , 2019 ) . Previously , we have shown that sustained p38 activation directly controls expression of several cytokines in sarcoid AMs ( Rastogi et al . , 2011 ) . The increased p38 phosphorylation in sarcoidosis was associated with lack of mitogen activated protein kinase phosphatase ( MKP ) -1 expression in sarcoidosis AMs and monocytes ( Rastogi et al . , 2011 ) . Furthermore , p38 MAPK regulates IL-17 production by Th17 cells through regulation of various transcription factors ( Huang et al . , 2015; Noubade et al . , 2011 ) . Interestingly , our recent study showed that macrophages derived from MKP-1 deficient mice exhibited higher HIF-1α and IL-1β expression and higher ROS production in response to LPS; in addition , p38 inhibition decreased HIF-1α expression in MKP-1 deficient macrophages and modified cytokine production ( Talwar et al . , 2017a ) . In our current work , we found significantly higher HIF-1α expression in sarcoidosis AMs and PBMCs . This can be partly explained by a constitutively active p38 in macrophages of sarcoidosis subjects ( Rastogi et al . , 2011; Talreja et al . , 2016 ) . We observed that a p38 inhibitor ( SB203580 ) partly decreased the expression of HIF-1α ( data not shown ) and cytokine levels in sarcoidosis . Activation of TLR4 and TLR2 by a variety of pathogen-derived molecules as well as environmental toxins has been shown to induce and stabilize HIF-1α expression ( Frede et al . , 2007; Liao et al . , 2014; Palazon et al . , 2014 ) . Abundance of HIF-1α in sarcoidosis also implies aberrant degradation by proteasomal or/and lysosomal pathways . Autophagy and the ubiquitin-proteasome system ( UPS ) are two major pathways involved in the degradation of proteins . It has been shown that there is a compensatory interaction between these two pathways and inhibition of one pathway leads to activation of the other ( Wang et al . , 2013 ) . Our RNA sequencing data showed upregulation of lysosomal pathways , confirming previous findings by other investigators ( Talreja et al . , 2016; Tomita et al . , 1999 ) . LAMP2 , along with LAMP1 , comprise about 50% of lysosomal proteins . In sarcoidosis we observed upregulation of LAMP2 both at the gene and protein level . CHQ is an ancient drug that in addition to its anti-malaria activity has been used for autoimmune diseases , including sarcoidosis ( Morse et al . , 1961 ) . Therefore , we determined the effect of CHQ on LAMP2 and HIF-α isoform expression . Surprisingly , we found that CHQ inhibits the increased levels of LAMP2 , HIF-α isoforms , and cytokine production in sarcoidosis . We speculate that in sarcoidosis inhibition of lysosomal function by CHQ leads to increased proteasome degradation of HIF-α isoforms leading to subsequent inhibition of IL-1β and IL-17 cytokines production . Environmental factors , altered metabolism , and inflammation can be linked to epigenetic changes such as methylation and acetylation that may contribute to HIF-1α expression and stability in sarcoidosis ( Watson et al . , 2010 ) . How HIF signaling in the absence of a hypoxic trigger regulates metabolic reprogramming and influences inflammation in chronic inflammatory diseases , especially respiratory diseases including sarcoidosis , has not been well illuminated . Our report identifies a role for HIF signaling in sarcoidosis granulomatous inflammation . The identification of the mechanisms underlying the aberrant regulation of HIF-1α and HIF-2α leading to persistent inflammation and Th1/Th17 pathology in sarcoidosis should open new avenues in rational drug discovery , not only for this disease but also for other inflammatory diseases .
Chemicals were purchased from Sigma Chemical ( St . Louis , MO ) unless specified otherwise . LPS and chloroquine was purchased from InvivoGen ( San Diego , CA ) . Antibodies against HIF-1α ( # bs0737 ) and HIF-2α ( #bs1477 ) were purchased from Bioss Inc ( Woburn , MA ) , P300 ( sc-585 ) was from Santa Cruz Biotechnology ( Santa Cruz , CA ) , Glut1 ( PA1-46152 ) from Thermofisher Scientific ( Waltham , MA ) . The antibody for pro-IL-1β ( # AF-201-NA ) was purchased from R and D Systems ( Minneapolis , MN ) , and β-actin ( #ab8227 ) was purchased from Abcam ( Cambridge , MA ) . Horseradish peroxidase–conjugated anti-mouse IgG ( #7076S ) and anti-rabbit IgG ( #7074S ) antibodies and antibody for ARNT ( #5531 ) were purchased from Cell Signaling Technology ( Beverly , MA ) . Horseradish peroxidase–conjugated anti-goat IgG ( sc-2033 ) was purchased from Santa Cruz Biotechnology , The anti-human antibodies used for flow cytometry were CD4-FITC ( #340133 ) , CD25-PE ( #341009 ) , CD14-PerCPCy5 . 5 ( #561116 ) and purified CD3 ( #555337 ) , purchased from BD Biosciences ( San Jose , CA ) . The secondary antibody used for immunostaining Alexa 488 ( #A11070 ) was purchased from Molecular Probes ( Grand Island , NY ) . CellTiter 96 AQueous One Solution Cell Proliferation Assay was purchased from Promega ( Madison , WI ) . The Committee for Investigations Involving Human Subjects at Wayne State University approved the protocol for obtaining alveolar macrophages by bronchoalveolar lavage ( BAL ) and blood by phlebotomy from control subjects and patients with sarcoidosis . The IRB number for this study is 055208MP4E . All methods were performed in accordance with the relevant guidelines and regulations . Informed consent was obtained from all subjects enrolled for the study . Sarcoidosis diagnosis was based on the ATS/ERS/WASOG statement ( Hunninghake et al . , 1999 ) . The criteria for enrollment in the diseased group were: ( i ) a compatible clinical/radiographic picture consistent with sarcoidosis , ( ii ) histologic demonstration of non-caseating granulomas on the tissue biopsy , and ( iii ) exclusion of other diseases capable of producing a similar histologic or clinical picture , such as fungus or mycobacteria . Subjects excluded were: ( i ) smokers , ( ii ) individuals receiving immune suppressive medication ( defined as corticosteroid alone and/or in combination with immune modulatory medications ) , ( iii ) individuals with positive microbial culture in routine laboratory examinations or viral infection; or ( iv ) individuals with known hepatitis or HIV infections or any immune suppressive condition . The criteria for enrollment in the control group were: ( i ) absence of any chronic respiratory diseases , ( ii ) lifetime nonsmoker , ( iii ) absence of HIV or hepatitis infection , and ( iv ) negative microbial culture . A total of 51 patients with sarcoidosis and 23 controls participated in this study . The medical records of all patients were reviewed , and data regarding demographics , radiographic stages , pulmonary function tests , and organ involvements were recorded . BAL was collected during bronchoscopy after administration of local anesthesia and before tissue biopsies ( Rastogi et al . , 2011; Talreja et al . , 2016 ) . Briefly , a total of 150 to 200 mL of sterile saline solution was injected via fiberoptic bronchoscopy; the BAL fluid was retrieved , measured , and centrifuged . Cells recovered from the BAL fluid were filtered through a sterile gauze pad and washed three times with phosphate-buffered saline ( PBS ) , resuspended in endotoxin-free RPMI 1640 medium ( HyClone ) supplemented with L-glutamine ( Life Technologies ) , penicillin/streptomycin ( Life Technologies ) , and 1% fetal calf serum ( HyClone ) , and then counted . BAL cells were cultured on adherent plates for 1 hr at 37°C in air containing 5% CO2 . Non-adherent cells were removed by aspiration; adherent cells were washed three times and used as AMs . Viability of these populations was routinely about 97% and by morphologic criteria the adherent cells were in excess of 99% AMs ( Rastogi et al . , 2011; Talreja et al . , 2016 ) . PBMCs were isolated from heparinized blood using Ficoll-Histopaque ( Sigma , St . Louis , MO ) density gradient separation followed by washing twice with PBS . Cell suspension was made in endotoxin-free RPMI 1640 medium ( HyClone ) supplemented with L-glutamine ( Life Technologies ) , penicillin/streptomycin ( Life Technologies ) , and 10% fetal calf serum ( HyClone ) . Cells were cultured in 12-well plates for further experiments ( Rastogi et al . , 2011; Talreja et al . , 2016 ) . CD14+ monocytes were purified from PBMCs by using the MACS monocyte isolation kit ( Miltenyl Biotech , San Diego , CA ) according to the manufacturer’s instructions . The purity of enriched monocytes was evaluated by flow cytometry using PerCPCy5 . 5-conjugated CD14 antibody ( #561116 , BD Biosciences ) ; the purity of monocytes was about 95% . Isolated AMs or PBMCs were transiently transfected with non-specific silencer siRNA ( NS siRNA , 200 pM ) or targeted HIF-1α silencer siRNA ( 200 pM , Thermofisher-Scientific ) in the presence of lipofectamine 2000 ( Invitrogen ) . The sequence of siRNA used: sense ( 5’−3’ ) GGAACCUGAUGCUUUAACUtt and antisense AGUUAAAGCAUCAGGUUCCtt . After 24 hr of transfection , cells were activated with either LPS ( 100 ng/mL ) or anti-CD3 ( 1 µg/mL ) . Viability of cells was assessed after siRNA treatment by MTS assay and 95% of cells were viable . Cell viability was assessed using MTS assay [CellTiter 96 AQueous One Solution Cell Proliferation Assay] ( Promega , Madison , WI ) following the manufacturer’s instructions . Briefly , cells equivalent to 1 × 104/well were seeded in 96-well culture plate and incubated for 24–48 hr with different treatments . After incubation , 20 µl of CellTiter 96 AQueous One Solution Reagent was added per well for 2 hr and the absorbance was measured at 490 nm using a 96-well plate reader . The levels of IL-1β , IL-1Ra , IL-17 , IL-10 , IL-6 , and IFN-γ in the conditioned medium were measured by ELISA according to the manufacturer's instructions ( ELISA DuoKits; R and D Systems , Minneapolis , MN ) . PBMCs from subjects with sarcoidosis were isolated , cultured , and after appropriate treatment were stained for cell surface markers using fluorescent labelled antibodies for CD4-FITC ( #340133 , BD Biosciences ) , and CD25-PE ( #341009 , BD Biosciences ) . Intracellular staining of PBMCs was done for HIF-1α and HIF-2α . Briefly , PBMCs were first surface stained for CD14 using CD14-PerCPCy5 . 5 antibody and then fixed using 100 μl of 1% paraformaldehyde for 30 min and then permeabilized with permeabilization buffer ( 0 . 5% saponin ) for 15 min at room temperature . Cells were centrifuged and resuspended in 100 μl of permeabilization buffer and stained with HIF-1α ( bs0737 , Bioss Inc ) or HIF-2α ( bs1477 , Bioss Inc ) antibody for 30 min . Cells were washed and stained with the Alexa 488 secondary antibody ( #A11070 , Molecular Probes ) . After 30 min cells were washed twice , resuspended in staining buffer , and analyzed for CD14+HIF-1α+ and CD14+HIF-2α+ monocytes by flow cytometry . PBMCs were not stained specifically to exclude DC contamination . Flow cytometry was performed on a BD LSR II SORP and data analysis was performed using FlowJo software ( FlowJo , LLC , Ashland , OR ) ( Talreja et al . , 2016 ) . Samples were gated on cells using FSC/SSC and doublet discrimination was performed to identify singlets using SSC-W vs . SSC-A . The flowcytometry work was done at the Microscopy , Imaging and Cytometry Resources ( MICR ) Core at the Karmanos Cancer Institute , Wayne State University . Intracellular expression of HIF-1α in sarcoidosis AMs was visualized by immunofluorescence staining . AMs ( 1 × 105 ) were allowed to adhere overnight on chamber slides . The cells were washed briefly with PBST and fixed with 3 . 7% paraformaldehyde . Cells were washed and permeabilized with 0 . 1% Triton X-100 , blocked ( 10% FCS ) , and then incubated with anti-HIF-1α ( bs0737 , Bioss Inc ) overnight at 4°C . The secondary antibody used was Alexa-fluor 488- conjugated goat anti-rabbit antibody . The next day cells were washed three times with PBS for 5 min , the slide was mounted with a drop of ProLong Gold antifade reagent with DAPI ( Invitrogen ) . The slide was examined using an Axiovert 40 CFL fluorescence microscope ( Carl Zeiss MicroImaging , Inc ) . Total cellular proteins were extracted by addition of RIPA buffer containing a protease inhibitor cocktail and antiphosphatase I and II ( Sigma Chemicals ) . Protein concentration was measured with the BCA assay ( Thermo Scientific , CA ) . Equal amounts of proteins ( 10–25 μg ) were mixed 1/1 ( v/v ) with 2x sample buffer ( 20% glycerol , 4% sodium dodecyl sulfate , 10% 2-βME , 0 . 05% bromophenol blue , and 1 . 25 M Tris-HCl , pH 6 . 8 ) , and fractionated on a 10% sodium dodecyl sulfate–polyacrylamide gel . Proteins were transferred onto a polyvinylidene difluoride membrane ( Bio-Rad ) for 60 min at 18 V using a SemiDry Transfer Cell ( Bio-Rad ) . The polyvinylidene difluoride membrane was blocked with 5% nonfat dry milk in TBST ( Tris-buffered saline with 0 . 1% Tween 20 ) for 1 hr , washed , and incubated with primary Abs ( 1/1000 ) overnight at 4°C . The blots were washed with TBST and then incubated for 1 hr with horseradish peroxidase–conjugated secondary anti-IgG Ab using a dilution of 1/10 , 000 in 5% nonfat dry milk in TBST . Membranes were washed four times in TBST . Immuno-reactive bands were visualized with a chemiluminescent reagent ( Amersham GE , PA ) . Images were captured on Hyblot CL film ( Denville; Scientific , Inc; Metuchen , NJ ) using a JPI automatic X-ray film processor model JP-33 . Optical density analysis of signals was performed using Image J software ( Rastogi et al . , 2011; Talreja et al . , 2016 ) . Tissue sections from the sarcoidosis transbronchial lung biopsy samples were selected for immunostaining after review of the glass slides that had been previously prepared using the routine hematoxylin-eosin protocol on paraffin-embedded sections . Additional fixed slides were cut , subjected to peroxide block protocol , pretreated , and then incubated first with primary antibody ( anti-HIF-1α , bs0737 , Bioss Inc ) and then with a secondary conjugated polymer; each incubation step was done for 30 min at room temperature . Negative staining was done by using an isotype control antibody . After another incubation step with the chromogen ( 5 min at room temperature ) , the sections were counterstained with hematoxylin and dehydrated with ethanol and xylene prior to mounting . Images were analyzed by microscopy ( BX40 , Olympus ) . A Student t-test or one-way analysis of variance and post hoc repeated measure comparisons ( least significant difference ) were performed to identify differences between groups . ELISA results were expressed as mean ± SEM . For all analyses , two-tailed p values of less than 0 . 05 were considered to be significant . | Sarcoidosis is a rare disease that is characterized by the formation of small lumps known as granulomas inside the body . These lumps are made up of clusters of immune cells , and are commonly found in the skin , lung or eye . Other organs of the body can also be affected , and symptoms will vary depending on where in the body lumps form . There is currently no specific treatment for sarcoidosis , as the direct cause of the disease is unknown . The disease is often treated with drugs that suppress the immune system . However , this type of treatment can lead to significant side effects and patients will respond to these drugs in different ways . Patients with sarcoidosis have a heightened immune response to microbes that can cause infections , and rather than providing protection , this heightened response causes damage and inflammation to the body’s organs . Now , Talreja et al . have identified which genes and proteins control this inflammatory response in immune cells from the lungs and blood of sarcoidosis patients . Immune cells in the lungs of sarcoidosis patients were found to have higher levels of hypoxia inducible factor ( HIF ) – a gene-regulating protein that controls the uptake and metabolism of oxygen in mammals . In addition , lung tissue affected with granulomas also expressed increased levels of a specific version of HIF known as HIF-1 . Talreja et al . showed that the increased expression of HIF in the immune cells of sarcoidosis patients was mechanistically linked to the production of several molecules that promote inflammation . Inhibiting HIF-1 led to a decrease in the production of these inflammatory molecules , indicating that increased activity of HIF-1 causes inflammation in sarcoidosis patients . It remains unclear what causes this abundance of HIF-1α . It is possible that specific modifications of this factor prevent it from degrading , resulting in higher levels . By identifying a link between HIF-1 and inflammation , these findings open up potential new avenues of the treatment for sarcoidosis patients . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"medicine",
"immunology",
"and",
"inflammation"
] | 2019 | HIF-1α regulates IL-1β and IL-17 in sarcoidosis |
Although critical for brain function , the physiological values of cerebral oxygen concentration have remained elusive because high-resolution measurements have only been performed during anesthesia , which affects two major parameters modulating tissue oxygenation: neuronal activity and blood flow . Using measurements of capillary erythrocyte-associated transients , fluctuations of oxygen partial pressure ( Po2 ) associated with individual erythrocytes , to infer Po2 in the nearby neuropil , we report the first non-invasive micron-scale mapping of cerebral Po2 in awake , resting mice . Interstitial Po2 has similar values in the olfactory bulb glomerular layer and the somatosensory cortex , whereas there are large capillary hematocrit and erythrocyte flux differences . Awake tissue Po2 is about half that under isoflurane anesthesia , and within the cortex , vascular and interstitial Po2 values display layer-specific differences which dramatically contrast with those recorded under anesthesia . Our findings emphasize the importance of measuring energy parameters non-invasively in physiological conditions to precisely quantify and model brain metabolism .
To understand the relationship between brain oxygenation and diseases associated with hypoxia or ischemia , it is important to first determine what fixes the resting value of tissue Po2 , that is , the concentration of oxygen in the interstitium that bridges oxygen delivery from erythrocytes to oxygen consumption by mitochondria . Numerous methods have been used to monitor brain oxygenation , and the most spatially resolved approaches have long relied on fine Clark-type electrodes ( for review see Ndubuizu and LaManna , 2007 ) , which have been used to report resting-state Po2 and local oxygen consumption in various brain regions ( Lecoq et al . , 2009; Masamoto et al . , 2003; Offenhauser et al . , 2005; Thompson et al . , 2003 ) during neuronal activation . However , these electrodes are invasive , do not faithfully report Po2 in vessels and cannot easily be used to determine Po2 in physiological conditions , that is , in awake , unstressed animals , avoiding the use of anesthetics . As anesthetics affect resting and evoked neuronal and astrocyte activity , arterial blood pressure and cerebral blood flow , the physiological values of cerebral interstitial Po2 and their relationship to blood flow parameters in capillaries remain unknown . Recently , a two-photon phosphorescent probe PtP-C343 has been generated ( Finikova et al . , 2007 , 2008 ) and two-photon phosphorescence lifetime microscopy ( 2PLM ) has been used to obtain depth-resolved , micron-scale measurements of Po2 in the anesthetized rodent brain ( Devor et al . , 2011; Lecoq et al . , 2011; Parpaleix et al . , 2013; Sakadzić et al . , 2010; Sakadžić et al . , 2014 ) . In addition , by detecting single red blood cells ( RBCs ) during Po2 measurement , we demonstrated the possibility of simultaneously monitoring blood flow and Po2 in capillaries ( Lecoq et al . , 2011 ) and of detecting erythrocyte-associated transients ( EATs ) , Po2 fluctuations associated with each individual erythrocyte flowing in capillaries , which were first reported in mesentery capillaries ( Golub and Pittman , 2005 ) . We showed that in olfactory bulb glomeruli of anesthetized mice , one parameter of EATs , the Po2 in between two red blood cells ( Po2InterRBC ) , is at equilibrium with , and thus reports , the Po2 in the nearby neuropil ( Parpaleix et al . , 2013 ) . This result implied that measurements of Po2InterRBC could provide a powerful tool to non-invasively map local interstitial oxygen concentration in the brain of awake animals . Here , we report that in both the olfactory bulb glomerular layer and the somatosensory cortex of unstressed , awake , resting mice , the interstitial Po2 ( equivalent to Po2InterRBC ) has the same mean value of ~23 mm Hg , spanning over a range of about 40 mm Hg . This contrasts with the large differences of capillary hematocrit and RBC flow values observed between the two brain regions . In addition , we find that in the cortex capillary and interstitial Po2 values display layer-specific differences , being lower in layer I than in layer II/III or layer IV . We also find that hemoglobin in brain capillaries is highly saturated with oxygen . Finally , we show that in both brain regions , the interstitial Po2 is much lower during wakefulness than under isoflurane anesthesia .
To ensure that we measured Po2 in real physiological conditions , that is , in awake , unstressed animals , each animal was habituated to all the conditions present during 2PLM Po2 measurements for several weeks prior to the experiments ( see methods for detailed training procedures ) . In brief , over the course of 2–3 days , each mouse was habituated to handling , and trained to run on a treadmill placed in its cage . A titanium bar was then surgically attached to the cranium and then a cranial window implanted over the region of interest , either the olfactory bulb or the somatosensory cortex . Then , over 2–4 weeks , the mouse was progressively habituated to being head-fixed , via the attached bar , in the dark , below the objective of the two-photon microscope , and with the limbs and body free to move on the treadmill . Habituation was achieved when the animal remained calm for periods >1 hr in the set-up with short bouts of running ( ~30 s ) . On the day of recording , the animal was briefly anesthetized ( 2% isoflurane , <5 min ) and the oxygen sensor PtP-C343 was injected intravenously . The animal was returned to its home cage , and after a delay of 90–120 min , Po2 recordings sessions of 1–3 hr commenced . Each animal underwent 1–3 recording sessions per day over the course of 2–7 days , with breaks of at least several hours between each session . Note that similar Po2 values were obtained from one day to the next and between sessions occurring the same day ( without reinjection of PtP-C343 ) . Using our previous approach ( Lecoq et al . , 2011; Parpaleix et al . , 2013 ) , we characterized EATs in 38 capillaries ( n = 5 animals ) from the glomerular layer of awake resting mice ( Figure 1 ) . Po2 measured at the RBC border ( Po2RBC ) was significantly larger than at mid-distance between two RBCs ( Po2InterRBC ) . Po2Mean , which was intermediate between these two values , is the average Po2 measured in a capillary without taking into account the existence of EATs , and is the only capillary Po2 value that has commonly been reported in the brain ( Sakadzić et al . , 2010; Vovenko , 1999 ) . Several measurements were made in each capillary but the average values of Po2RBC , Po2InterRBC and Po2Mean were similar whether calculated on all measurements or on all capillaries ( 262 measurements: Po2RBC = 60 . 5 ± 0 . 9 mm Hg; Po2InterRBC = 23 . 4 ± 0 . 5 mm Hg; Po2Mean = 32 . 8 ± 0 . 7 mm Hg . 38 capillaries: Po2RBC = 60 . 6 ± 2 . 3 mm Hg; Po2InterRBC = 23 ± 1 . 5 mm Hg; Po2Mean = 32 . 7 ± 1 . 9 mm Hg ) ( Figure 1B ) . The standard deviation of measurements ( SD ) made in a given capillary , during single or multiple recording sessions were modest for Po2InterRBC ( mean SD = 2 . 6 ± 0 . 3 mm Hg ) and slightly larger for Po2RBC ( mean SD = 5 . 4 ± 0 . 5 mm Hg ) . Figure 1C illustrates that average Po2 values masked the large span of all values measured . This was particularly true for Po2RBC , for which values frequently exceeded 70 mm Hg . Overall , these data show that in the glomerular layer of the awake resting mouse , interstitial Po2 ranges from 15 to 35 mm Hg in about 82% of our measurements . We then investigated whether the range and fluctuations of Po2values depend on two vascular parameters , RBC blood flow and hematocrit , in the same capillaries . 10 . 7554/eLife . 12024 . 003Figure 1 . Erythrocyte-associated transients ( EATs ) in the olfactory bulb glomerular layer of the awake mouse . ( A ) Left panel , schematic diagram of the organisation of the olfactory bulb . OSN: olfactory sensory neuron , PG: periglomerular neuron , M: mitral cell , ONL: olfactory nerve layer , GL: glomerular layer , EPL: external plexiform layer , MCL: mitral cell layer . Right panel , top , schematic illustrating the 2PLM Po2 measurement procedure in capillaries . Bottom , diagram showing Po2 values extracted from EATs . The continuous trace represents the Po2 profile relative to the RBC border in one selected capillary: Po2 at the RBC border ( Po2RBC , in this case 47 . 2 mm Hg ) , Po2 at distance from a RBC ( Po2InterRBC , in this case 8 . 6 mm Hg ) which gives an estimate of Po2 in the interstitium of the glomerular layer , and average Po2 in the capillary ( Po2Mean , in this case 19 mm Hg ) . ( B ) Multiple ( ~4–8 ) measurements were made in each capillary . The Po2InterRBC is significantly lower than both Po2Mean and Po2RBC , whether calculated on all measurements ( the mean value is the average of all measurements pooled from all capillaries assessed , n = 262 , left panel ) , or on the mean values from each capillary ( the mean value is the average of the single mean values for each of the capillaries , n=38 , right panel ) . SD ( the average of the SD values for each capillary , presented as mean standard deviation ± SEM ) illustrates the fluctuations of Po2 values in each capillary across the multiple measurements . Data presented as mean ± SEM . *p<0 . 05 , ***p<0 . 001 , Kruskal-Wallis test with 2-tailed Dunn's multiple comparison post-hoc test . ( C ) Frequency distributions of all measurements of Po2RBC ( left panel ) , Po2InterRBC ( middle panel ) , and Po2Mean ( right panel ) . 5 mm Hg bin width . For all plots n = 5 mice . DOI: http://dx . doi . org/10 . 7554/eLife . 12024 . 003 Mean capillary RBC flow and hematocrit values were 30 . 6 ± 2 . 6 cells/s and 34 . 6 ± 1 . 8% , respectively ( Figure 2A ) . Both RBC flow and hematocrit displayed a wide range of values ( Figure 2B ) with a positively skewed frequency distribution of RBC flow . Simultaneous measurements of Po2 and blood flow parameters revealed that although interstitial Po2 ( Po2InterRBC ) is correlated with both RBC flow and hematocrit ( r = 0 . 3091 , p<0 . 0001 and r = 0 . 4365 , p<0 . 0001 , Spearman r correlation , respectively ) , these relationships are non-linear . In particular , Po2InterRBC is mostly independent of both RBC flow below 60 cells/s and hematocrit from 20 to 50% ( Figure 2C and E ) , increasing only at high values of both parameters . This stable region covers 82 . 4% of our measurements . In contrast , Po2RBC increased with RBC flow and hematocrit at low values , becoming stable above 20 cells/s and 30% , respectively ( Figure 2C and E ) . Note that Po2Mean increased with both RBC flow and hematocrit values ( RBC flow: r = 0 . 5116 , p<0 . 0001; hematocrit: 0 . 6752 , p<0 . 0001 , Spearman r correlation , Figure 2D and F ) . 10 . 7554/eLife . 12024 . 004Figure 2 . Relationships of capillary blood flow and hematocrit to Po2 values , in the olfactory bulb glomerular layer of the awake mouse . ( A ) RBC flow and hematocrit calculated from the mean values of each capillary ( n = 38 ) . SD ( the average of the SD values for each capillary , presented as mean standard deviation ± SEM ) illustrates the fluctuations of RBC flow and hematocrit in each capillary . ( B ) Frequency distributions of RBC flow and hematocrit ( 5% bin ) . ( C ) Distribution of all Po2RBC and Po2InterRBC measurements as a function of RBC flow . Note that the Po2InterRBC is independent of RBC flow below 60 cells/s while Po2RBC increases with RBC flow below 40 cells/s . ( D ) Po2Mean as a function of RBC flow . ( E ) Distribution of Po2RBC and Po2InterRBC as a function of hematocrit . Po2InterRBC is independent of hematocrit from 20 to 50% . Po2RBC increases with hematocrit at low values . ( F ) Po2Mean as a function of hematocrit . Bar graph data presented as mean ± SEM . *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 . Kruskal-Wallis test with 2-tailed Dunn's multiple comparison post-hoc test . For all plots n = 5 mice . DOI: http://dx . doi . org/10 . 7554/eLife . 12024 . 004 Isoflurane is a volatile anesthetic that is commonly used in the study of brain activation and metabolism . It differently affects regional cerebral blood flow in humans ( Ramani and Wardhan , 2008 ) , and modulates neurovascular coupling in a concentration-dependent fashion ( Masamoto et al . , 2009 ) as well as the relationship between spontaneous or evoked neuronal activity with BOLD signal ( Aksenov et al . , 2015 ) . However , its effects on brain oxygenation have only been investigated using approaches with low spatial resolution and which did not allow simultaneous measurement of blood flow ( Liu et al . , 1995; Ortiz-Prado et al . , 2010 ) . We performed paired measurements of EATs and flow parameters in a set of capillaries , both when the animals were awake and when they were anesthetized with isoflurane ( ~0 . 75% as measured at the animal’s snout , delivered in air with no supplementary O2 ) . As can be seen from Figure 3A , isoflurane significantly increases all capillary Po2 values ( Po2Mean , Po2RBC and Po2InterRBC ) . This effect was present in all but one of the capillaries tested . Although reduced neuronal activity ( and hence O2 consumption ) in the isoflurane anesthetized state ( Aksenov et al . , 2015 ) is likely to play a role , it appears that this elevation of Po2 largely resulted from an increase in RBC flow , as the increase in Po2 values observed is in accordance with that which would be predicted from the observed increase in RBC flow based on the relationship presented in Figure 2C and D . This increase in capillary RBC flow rate is related to isoflurane’s vasodilatory effects on large vessels ( Figure 3B ) ( Koenig et al . , 1994 ) . Consequently , cellular processes , including neuronal activity in response to odor , occur at a much higher oxygen concentration during isoflurane anesthesia than in the awake , resting state , and this difference is likely to be exacerbated when isoflurane is delivered in gas mixtures where [O2] is greater than 21% ( as in many other studies ) . Since the olfactory bulb glomerular layer has a specific neuronal and vascular organisation that could be a main determinant of Po2 values , we extended our investigation of tissue oxygenation to the cerebral cortex , due to its importance to higher cognitive functions and its use as an ischemic model . 10 . 7554/eLife . 12024 . 005Figure 3 . Isoflurane changes the brain oxygenation state . Po2 and RBC flow were compared in the same sets of olfactory bulb glomerular layer capillaries when the animal was awake , and when the animal was anesthetized with isoflurane ( 0 . 75% , delivered in air , no oxygen added ) . ( A ) Isoflurane anesthesia increased all RBC flow and Po2 values as compared to the awake state ( n = 142 measurements in each condition , 16 capillaries from 3 mice ) . Bar graph data presented as mean ± SEM . , ***p<0 . 001 , paired 2-tailed Wilcoxon signed rank test . ( B ) Isoflurane anesthesia induced a large dilation of pial vessels ( a: artery , v: vein ) . See also Figure 3—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 12024 . 00510 . 7554/eLife . 12024 . 006Figure 3—figure supplement 1 . Isoflurane alters oxygenation in the somatosensory cortex . ( A ) As in the glomerular layer , isoflurane anesthesia ( 0 . 75% delivered in 21% O2 air ) increases Po2 and RBC flow in layer I capillaries in the somatosensory cortex relative to the awake state . ( B ) This isoflurane-induced increase in Po2 was associated in a right-shift in the frequency distribution of measured Po2Mean , Po2InterRBC and Po2RBC along with that of estimated So2 . *p<0 . 05 , paired , 2-tailed , Wilcoxon signed rank test . n = 36 measurements under isoflurane and 22 in the awake state , from five capillaries . DOI: http://dx . doi . org/10 . 7554/eLife . 12024 . 006 We made measurements of all Po2 and blood flow parameters in capillaries distributed from the cortical surface to a depth of 410 µm in the fore- and hind-limb regions of the somatosensory cortex . Due to the greater inter-capillary distance in the cortex than in the glomeruli ( Lecoq et al . , 2009; Sakadžić et al . , 2014 ) , and as Po2 gradients have previously been observed around blood vessels in anesthetized animals ( Devor et al . , 2011; Sakadzić et al . , 2010; Sharan et al . , 2008 ) , it is likely that there are tissue regions , in areas far from the nearest capillary , in which the Po2 is lower than what is reported by Po2InterRBC . Nonetheless , based on our previous results ( Parpaleix et al . , 2013 ) in olfactory bulb glomeruli and on theoretical predictions ( Lücker et al . , 2014 ) , we expect that Po2InterRBC reports steady-state tissue Po2 up to a radius of approximately 10 µm from a capillary . Furthermore , recent work , that models hematocrit distribution in large microcirculatory networks and accurately replicates physiological RBC distribution , predicts that the distribution of oxygen in tissue volumes supplied by these networks is largely homogeneous ( Gould and Linninger , 2015 ) . We thus consider that our measured values of Po2InterRBC represent a significant proportion of the range of Po2 values present in the cortical interstitium . Our Po2 measurements revealed that the averages and ranges of Po2 values in cortical capillaries ( Figure 4A and B ) were similar to those measured in the olfactory bulb glomerular layer ( Po2RBC = 66 . 3 ± 1 . 6 mm Hg , Po2InterRBC = 23 . 3 ± 1 . 1 mm Hg , and Po2Mean = 36 . 3 ± 1 . 3 mm Hg ) , indicating that at rest in the awake state , these two brain areas have similar levels of both RBC , capillary and pericapillary oxygenation . In contrast , the average RBC flow and hematocrit values ( 41 . 9 ± 1 . 8 cells/s and 47 . 3 ± 0 . 9% , respectively ) were significantly higher than those found in the olfactory bulb glomerular layer ( Figure 4B ) with the hematocrit values being higher than previously reported levels in the cerebral cortex of anesthetized animals ( Hudetz , 1997 ) . The lower hematocrit levels in glomerular layer capillaries could be related to differences in the bulb cerebrovascular supply which , in contrast to the cortex ( Blinder et al . , 2013 ) , is poorly established ( Coyle , 1975 ) . 10 . 7554/eLife . 12024 . 007Figure 4 . The relationship of Po2 to RBC flow and hematocrit in the somatosensory cortex of the awake mouse . ( A ) Average values and SD of Po2 parameters in somatosensory cortex capillaries , calculated from the mean values from each capillary ( 81 capillaries , 528 measurements , SD = the average of the SD values for each capillary , presented as mean standard deviation ± SEM ) . ( B ) Distribution of all capillary Po2 values averaged in ( A ) . Frequency distribution histogram of local tissue Po2 ( Po2InterRBC ) in inset . 5 mm Hg bin . ( C ) Average values and SD of RBC flow and hematocrit calculated from the mean values from each capillary . ( D ) Distribution of all Po2RBC and Po2InterRBC measurements as a function of RBC flow . Note that for most values ( from 20 to 60 cells/s ) , both Po2InterRBC and Po2RBC increase with RBC flow . ( E ) Po2Mean as a function of RBC flow . ( F ) Distribution of all Po2RBC and Po2InterRBC measurements as a function of hematocrit . For most values ( from 20 to 60% ) , both Po2InterRBC and Po2RBC increase with hematocrit . ( G ) Po2Mean as a function of hematocrit . Bar graph data presented as Mean ± SEM . *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 . Kruskal-Wallis test with 2-tailed Dunn's multiple comparison post-hoc test . DOI: http://dx . doi . org/10 . 7554/eLife . 12024 . 007 As in the olfactory bulb , both Po2RBC and Po2InterRBC were correlated with RBC flow rate and hematocrit ( Po2RBC with RBC Flow and hematocrit: r = 0 . 3824 , p<0 . 0001 and r = 0 . 2315 , p<0 . 0001 , Spearman r correlation , respectively; Po2InterRBC with RBC Flow and hematocrit: r = 0 . 4283 , p<0 . 0001 and r = 0 . 5018 , p<0 . 0001 , Spearman r correlation , respectively ) . Po2InterRBC ( Figure 4D and F ) increased with RBC flow , from 20 to 60 cells/s , that is , over the majority of the measurements , becoming stable thereafter . It also increased with hematocrit . Similarly , Po2RBC increased with both RBC flow and hematocrit . Note that Po2Mean , was correlated with and increased through the entire range with both RBC flow and hematocrit ( RBC flow: r = 0 . 4670 , p<0 . 0001; hematocrit: 0 . 5945 , p<0 . 0001 , Spearman r correlation , Figure 4E and G ) . Several studies , using polarographic electrodes or 2PLM in anesthetized rodents , have reported that vascular and interstitial Po2 varies with cortical depth ( Devor et al . , 2011; Masamoto et al . , 2003; Sakadzić et al . , 2010 ) . As anesthetics could differently modulate synaptic activity and blood flow in layers I to IV , it is difficult to predict the extent to which Po2 depth variations occur in the awake animal . We thus first measured Po2 in descending and ascending large vessels and then compared Po2and blood flow parameters of capillaries from layers I to IV ( from the surface to 410 µm in depth ) . Capillaries less than 60 μm below the surface were considered to be in layer I , those from 90 to 260 μm below the surface were classified as layer II/III capillaries and those deeper than 340 μm were considered layer IV capillaries ( capillaries from 260 to 340 µm were not considered due the ambiguity of their location ) . Penetrating vessels ( arterioles and venules ) were traced from their point of descent below the surface down along their main trunk until they ramified into smaller vessels ( or descended below 410 µm , which was the maximum depth at which we made measurements ) . Note that in these vessels EATs were not detectable due to the close apposition of RBCs . The mean Po2 was 69 . 2 ± 1 . 4 mm Hg for arterioles and 39 . 9 ± 1 . 3 mm Hg for venules and , in contrast to what was reported in anesthetized animals ( Devor et al . , 2011; Sakadzić et al . , 2010 ) , no gradient was observed with depth ( Figure 5A and B ) . 10 . 7554/eLife . 12024 . 008Figure 5 . Depth profiles of vascular and local tissue oxygenation in the somatosensory cortex of the awake mouse . ( A ) Left panel , maximum intensity projection of superficial portion of the vasculature of the somatosensory cortex , with boxed area highlighting a penetrating arteriole shown in the XZ projection ( right panel ) . ( B ) Left panel , Po2 values in penetrating arterioles ( red ) and venules ( blue ) as a function of depth from the cortical surface . Each line represents a single vessel . Right panel , mean of all Po2 values recorded from vessels as a function of depth . Note the absence of Po2 gradients with depth ( 50 µm bin size , 136 measurements in 11 arterioles , 148 measurements in 14 venules , from 6 mice . ) Data presented as mean ± SD ( C ) Comparison of Po2 values ( left panel ) , RBC flow and hematocrit ( right panel ) in layers I , II/III and IV . Note that capillary Po2Mean and Po2InterRBC are higher in layers II/III and IV than in layer I , although there are no significant differences in either blood flow parameter . Data presented as mean ± SEM . * p<0 . 05 , **p<0 . 01 , Kruskal-Wallis test with 2-tailed Dunn's multiple comparison post-hoc test . n = Layer I: 113 measurements in 17 capillaries , Layer II/III: 230 measurements in 41 capillaries , Layer IV: 151 measurements in 15 capillaries ( D ) Po2RBC and Po2InterRBC as a function of hematocrit in layers I , II/III and IV . DOI: http://dx . doi . org/10 . 7554/eLife . 12024 . 008 Laminar analysis of capillary blood flow and Po2 values revealed some specific differences: although the hematocrit and blood flow values were similar in all three layers ( Figure 5C , right panel ) , all Po2 values were lower in layer I than in layer II/III ( Figure 5C , left panel ) . The low Po2InterRBC values 14 . 7 ± 1 . 7 mm Hg suggests that interstitial Po2 is correlated with the capillary density which is lower in layer I than II/III ( Blinder et al . , 2013; Sakadžić et al . , 2014 ) . The correlated analysis of Po2 with hematocrit revealed that low Po2InterRBC values were present at a wide distribution of hematocrit levels ( Figure 5D , left panel ) . In addition , even though the average Po2RBC was higher in layer II/III than in layer IV , it was independent of hematocrit in both layers ( Figure 5D ) . Finally , the effects of isoflurane in the cortex ( Layer I , Figure 3—figure supplement 1 ) were similar to those observed in the olfactory bulb . Thus , in addition to its direct effects on neurons , isoflurane increases oxygen delivery to the entire brain . Po2InterRBC values show a wide distribution in both the olfactory bulb glomerular layer and the somatosensory cortex ( Figure 1C and 4B ) . Notable in both structures is the presence of measurements of Po2InterRBC , and hence local tissue Po2 , that were <15 mm Hg . In the olfactory bulb glomerular layer these low Po2InterRBC capillaries ( n=5 ) were found to have a large range of RBC flow rates but low hematocrit levels ( <25% , Figure 6A and 6B ) . In contrast , in somatosensory cortex capillaries ( n = 13 ) in which the Po2InterRBC value was <15 mm Hg , neither RBC flow rates nor hematocrit levels were notably low ( Figure 6C and 6D ) . Instead , these low Po2InterRBC values were mostly found in capillaries in layer I ( 10 of 13 capillaries ) , suggesting that the presence of low tissue Po2 values results from several factors . 10 . 7554/eLife . 12024 . 009Figure 6 . In the awake mouse , low interstitial Po2 in the olfactory bulb glomerular layer is associated with low hematocrit capillaries . Po2 values from all glomerular layer capillaries with average Po2InterRBC values of <15 mm Hg ( n = 24 measurements in capillaries ) are plotted as a function of RBC flow ( A ) and hematocrit ( B ) . The RBC flow rates in these capillaries were distributed across a wide range , whereas in all cases capillary average hematocrit was <25% , suggesting that , in the glomerular layer , areas of low interstitial Po2 are supplied by capillaries with relatively low hematocrit values ( hematocrit of these capillaries = 18 . 6 ± 1 . 4% , n = 5; hematocrit of all other capillaries = 37 ± 1 . 7% , n = 33; p = 0 . 002 , unpaired t-test . mean ± SEM ) . Conversely , cortical capillaries with average Po2InterRBC <15 mm Hg ( n = 13 capillaries , 101 measurements ) had wide ranges of both RBC flow ( C ) and hematocrit ( D ) . However , the majority ( 10 of 13 ) of these capillaries were located in layer I . In all plots , single measurement values and mean ± SD of all measurements in each capillary shown . DOI: http://dx . doi . org/10 . 7554/eLife . 12024 . 009 Given that Po2RBC should be representative of the Po2 level inside RBCs , we used known values of the Hill coefficient and P50 for mouse hemoglobin ( Uchida et al . , 1998 ) to estimate hemoglobin saturation ( So2 ) in capillaries . In both the olfactory bulb glomerular layer and the somatosensory cortex ( Figure 7A and B ) , measured So2 is >50% in the vast majority of cases ( 94 . 7% and 98 . 5% of measurements in the glomerular layer and the cortex , respectively ) . This shows that in the resting brain , the majority of hemoglobin exists as oxyhemoglobin . Furthermore the presence of numerous measurements of So2 >85% in both structures suggests that hemoglobin saturation is nearly maximal in a significant proportion of capillaries ( ~10% of capillaries in the glomerular layer , ~20% in the somatosensory cortex ) . Note that , if So2 were to be inappropriately estimated from Po2Mean values , a very different distribution of hemoglobin saturation would be derived ( Figure 7C ) . This demonstrates that in capillaries , an accurate measurement of Po2RBC is a prerequisite for estimating So2 . 10 . 7554/eLife . 12024 . 010Figure 7 . In the awake mouse , the majority of hemoglobin in cerebral capillaries is oxygenated . ( A ) Left panel , frequency distribution of Po2RBC values measured in the glomerular layer of awake mice , which were used to compute So2 values ( right panel ) . ( B ) Equivalent Po2RBC ( left panel ) and So2 distributions ( right panel ) from the cerebral cortex . In both structures > 90% of the measurements have So2 values of >50% ( 94 . 7% and 98 . 5% of measurements in the glomerular layer and the cortex respectively ) with ~60% of So2 values in the cortex being >75% . ( C ) Left panel , frequency distribution of Po2Mean values from cerebral cortex capillaries , with the So2 distribution that would be computed were these lower Po2 values considered to represent those present at the hemoglobin molecules in RBCs ( right panel ) . Glomerular layer: n = 262 measurements in 38 capillaries . Somatosensory cortex: n = 528 measurements from 81 capillaries . Bin size 5 mm Hg and 5% for Po2 and So2 respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 12024 . 010
In the olfactory bulb and somatosensory cortex of awake resting mice , we report similar average values of interstitial Po2 ( ~23 mm Hg ) . In the awake rodent , brain oxygenation has previously only been investigated with lower resolution approaches . Liu et al . ( Liu et al . , 1995 ) used electron paramagnetic resonance ( EPR ) oximetry with a lithium phthalocyanine crystal being implanted in the cerebral cortex of rats 24 hr prior to the measurements . They reported that restrained and untrained rats have a cerebral Po2 of about 34 mm Hg , which , surprisingly , was reduced by isoflurane anesthesia ( 1% in 21% O2 ) to about 24 mm Hg . This resting value ( 34 mm Hg ) is significantly higher than our average interstitial Po2 value , most probably due to the effects of the stress resulting from restraint in the absence of habituation . The effect of isoflurane , which decreased Po2 is intriguing , and indicates that the main effect of isoflurane in this case was to abolish the Po2 augmenting effects of stress . Interestingly , increasing the delay from crystal implantation to 7 days and introducing some habituation to restraint reduced cerebral Po2 to 27 mm Hg ( Dunn et al . , 2000 ) . The best demonstration of physiological measurements of Po2 has been performed in unrestrained rats , in which brain Po2was measured with an implanted fiber optic probe measuring quenching of an oxygen sensor ( Ortiz-Prado et al . , 2010 ) . Although the probe was very invasive , it reported an average bulk tissue Po2 value of 25 mm Hg , a value very similar to ours . We are thus confident that Po2InterRBC is an excellent reporter of the interstitial Po2 , that our extensive habituation procedures are efficient , and that this study gives the first non-invasive and physiological values of Po2 , hematocrit and capillary blood flow in the rodent brain . Our results show a number of differences from those previously reported with high-resolution measurement techniques in anesthetized animals . Tissue Po2 values have previously been reported to be in the range of ~5–100 mm Hg , with the higher values occurring in regions close to pial arterioles ( Devor et al . , 2011; Sakadzić et al . , 2010 ) . Our values of tissue Po2 ( Po2InterRBC ) lie in the lower end of this wide range , do not include measurements from the capillary-free peri-arteriolar regions , and so are likely to be reflective of the Po2 levels that exist in the majority of the tissue , away from large arterioles . Considering intravascular Po2 , our measurements of capillary Po2Mean ( average ≈36 mm Hg in the cortex ) seem to be broadly in agreement with those reported by Sakadzić et al . , 2010 ( ~25–35 mm Hg ) , but our recorded values of capillary RBC Po2 ( Po2RBC , average ≈66 mm Hg in the cortex ) , are dramatically higher than those published by Sakadzić et al . , 2014 , where the most frequent values were ~25–35 mm Hg . A prominent discrepancy between the results of previous studies and our findings in awake , resting animals relates to changes in cortical oxygenation with depth . Many previous studies report drops in interstitial Po2 with increasing depth in the cortex ( Cross and Silver , 1962; Devor et al . , 2011; Masamoto et al . , 2003; Nair et al . , 1975; Ndubuizu and LaManna , 2007; Whalen et al . , 1970 ) , with Po2 in layer I being higher than in underlying layers . In contrast , our measures of Po2InterRBC suggest that interstitial Po2 is lower in layer I than in either layer II/III or IV . These conflicting patterns of laminar variations of Po2 also exist in comparisons of the mean vascular Po2 values ( Po2Mean ) , where previous 2PLM studies have reported decreases in mean vascular Po2 in concert with decreases in penetrating arteriole and venule Po2 ( Devor et al . , 2011; Sakadzić et al . , 2010 ) with increasing depth in the cortex . In the present study , we see an increase in capillary Po2Mean from layer I to deeper layers , and no gradient in penetrating vessel Po2 with depth . It is possible that the deviation from normal physiological conditions inherent in anesthesia and acute surgical preparation , which we avoid with our approach , leads to the emergence of these disagreements . Although average Po2InterRBC values in both the cortex and glomerular layer were ~23 mm Hg , in both structures a number of capillaries were found to have Po2InterRBC values <15 mm Hg , indicating the existence of regions in both structures where the interstitial Po2 is close to reported values of Po2 below which cellular respiratory rate is strongly dependent on Po2 ( ~10 mm Hg ) ( Kasischke et al . , 2011 ) . Furthermore , examples were found in both brain regions where the Po2InterRBC was below the value of ~3 . 4 mm Hg that has previously been reported as the critical Po2 in brain tissue ( Kasischke et al . , 2011 ) . Similarly low tissue Po2 measurements have previously been observed ( Ndubuizu and LaManna , 2007; Whalen et al . , 1973 ) , but their existence has been interpreted as being related to disruption of normal tissue physiology in the experimental preparations ( Wilson et al . , 2012 ) . However , the presence of such low values in our awake preparations indicates that , surprisingly , at least some small regions of the brain can subsist at very low Po2 values . Our finding of high Po2RBC and corresponding capillary So2 ( estimated using the parameters used in Sakadžić et al . , 2014 ) values in both the olfactory bulb and the somatosensory cortex ( Figure 7 ) differs from recently published findings in the anesthetized , ventilated mouse ( Sakadžić et al . , 2014 ) . We find that capillary So2 in both structures is generally high , indicating that the majority of hemoglobin in these vessels exists as oxyhemoglobin , and thus that capillaries are capable of supplying very significant quantities of oxygen to support neural function . In conclusion , the present study establishes , for the first time , accurate and precise values of physiological Po2 in the vasculature and interstitium of mouse cerebral grey matter . As it is known that O2 concentration is a critical parameter in determining the properties of neuronal function ( Huchzermeyer et al . , 2008 , 2013; Ivanov and Zilberter , 2011 ) , and neurovascular interactions ( Gordon et al . , 2008 ) , these values provide a standard on which future research can rely to provide relevant , physiologically accurate conditions of oxygenation in which to investigate such processes . | Brain cells need a constant supply of oxygen to fuel their activities . This oxygen is delivered by the flow of blood through the vessels in the brain . If the blood flow to brain tissue is cut off as happens in stroke , or if an individual stops breathing , the brain becomes deprived of oxygen and brain cells will be damaged and die . To better understand how the brain works in health and disease , scientists need to learn how much oxygen the blood must deliver to the brain tissue to adequately support the activities of brain cells . Many studies have measured oxygen levels in the brain . However , these studies have looked only roughly and taken measurements from large areas of the brain , or they have involved animals receiving anesthesia , which can alter blood flow and oxygen use in the brain . Recently , scientists discovered that they could measure oxygen concentration at high detail in the brain of anesthetized rodents with a specialized microscope , by using molecules that emit light at a rate that depends on the local oxygen concentration . Now , Lyons et al . have shown that this same technique can be used in mice that are awake . First , a piece of the skull was replaced with glass to create a small transparent window . Then , the animals were allowed to recover for a few weeks , and were trained to get them used to how they would be handled during the experiments . After this period , the oxygen concentrations and blood flow in different parts of the mouse brains were measured in fine detail using the microscope while the animals were awake and relaxed . The experiments showed that oxygen levels in awake resting mice are actually lower than in anesthetized mice , and that oxygen levels differ between different parts of the mouse brain . This first detailed look at oxygen levels in the brain of awake animals will likely lead to more studies . For example , future studies may look at how quickly the brain uses oxygen under normal conditions and what happens in the brain during disease . | [
"Abstract",
"Introduction",
"Results",
"Discussion"
] | [
"neuroscience"
] | 2016 | Mapping oxygen concentration in the awake mouse brain |
Cell surface reception of Sonic hedgehog ( Shh ) must ensure that the graded morphogenic signal is interpreted accordingly in neighboring cells to specify tissue patterns during development . Here , we report endocytic sorting signals for the receptor Patched1 ( Ptch1 ) , comprising two ‘PPXY’ motifs , that direct it to degradation in lysosomes . These signals are recognized by two HECT-domain ubiquitin E3 ligases , Smurf1 and Smurf2 , which are induced by Shh and become enriched in Caveolin-1 lipid rafts in association with Ptch1 . Smurf-mediated endocytic turnover of Ptch1 is essential for its clearance from the primary cilium and pathway activation . Removal of both Smurfs completely abolishes the ability of Shh to sustain the proliferation of postnatal granule cell precursors in the cerebellum . These findings reveal a novel step in the Shh pathway activation as part of the Ptch1 negative feedback loop that precisely controls the signaling output in response to Shh gradient signal .
The secreted Sonic hedgehog ( Shh ) protein specifies spatial tissue patterns during development by providing positional cues embedded in its concentration gradient ( Jiang and Hui , 2008; Robbins et al . , 2012; Ryan and Chiang , 2012 ) . During embryogenesis , neighboring progenitor cells in a developing field are able to discern incremental changes in the Shh signal strength and adopt their respective fate accordingly ( Ribes and Briscoe , 2009; Balaskas et al . , 2012 ) . This ability requires a cell surface reception system that can transform the graded Shh signal into different levels of signaling output , but how this is accomplished is poorly understood . In the adult , Shh plays a crucial role in guiding the differentiation of tissue-specific stem cells ( Jaks et al . , 2008; Shin et al . , 2011; Arwert et al . , 2012 ) , and inappropriate activation of Shh signaling could be the culprit that underlines neoplastic growth in the gut epithelium ( Nielsen et al . , 2004 ) or lead to outright cancers ( Scales and de Sauvage , 2009; Stecca and Ruiz , 2010; Northcott et al . , 2012 ) . At the cell surface , whereas a network of membrane proteins , including Hip1 ( Chuang et al . , 2003 ) , Gas1 ( Lee et al . , 2001 ) , Boc/iHog , and Cdo/Boi ( Okada et al . , 2006; Tenzen et al . , 2006; Yao et al . , 2006; Beachy et al . , 2010 ) , bind Shh and control the range and competence of its receiving cells , the core of Shh signal reception consists of Patched1 ( Ptch1 ) , a 12-pass membrane receptor that acts negatively on Smoothened ( Smo ) , a G-protein-coupled , receptor-like signal transducer ( Rohatgi and Scott , 2007b ) . Binding of Shh to Ptch1 alleviates the Ptch1 inhibition of Smo , allowing the signal to propagate to three Gli proteins , the transcriptional effectors of the pathway , and activate the expression of target genes , including pathway components Ptch1 and Gli1 themselves . Since Gli1 is a potent activator of Shh target genes , its induction by the ligand ensures that pathway activation will attain the intended effect in a positive feedback loop . On the other hand , induction of the inhibitory Ptch1 amounts to a negative feedback control , which was regarded crucial to the interpretation of the Shh gradient signal ( Ribes and Briscoe , 2009 ) . In effect , Ptch1 serves two roles in Shh signaling: first , it acts cell autonomously in suppressing the downstream pathway , and second , the excessive Ptch1 induced by Shh acts as a sink in limiting the spread of the ligand , thereby affecting the neighboring cells in a non-cell autonomous fashion ( Chen and Struhl , 1996; Torroja et al . , 2004 ) . However , it is not clear what counteracts the induction of Ptch1 to achieve the precision of the regulation . For many years , Ptch1 and Smo have been seen in punctate intracellular vesicles in both Drosophila and mammalian cells ( Capdevila et al . , 1994; Ramirez-Weber et al . , 2000; Zhu et al . , 2003; Li et al . , 2012 ) , and their trafficking between the cytoplasmic membrane and intracellular vesicles found to be crucial to the activation of the Hedgehog pathway ( Denef et al . , 2000; Incardona et al . , 2000; Zhu et al . , 2003; Nakano et al . , 2004; Lu et al . , 2006; Milenkovic et al . , 2009; Li et al . , 2012 ) . It is known that ligand engagement of Drosophila receptor Ptc triggers its internalization and membrane presentation of Smo , but membrane trafficking of Ptch1 and Smo in mammalian cells has an added complexity in that Shh signals through the primary cilium ( Huangfu et al . , 2003; Corbit et al . , 2005; Goetz and Anderson , 2009 ) , a microtubule-based membrane protrusion that emanates from the interphase centrioles ( Lefebvre and Rosenbaum , 1986; Pazour and Witman , 2003; Nachury et al . , 2010 ) . The prevailing model for mammalian Shh activation entails Ptch1 exiting from and Smo translocating into the primary cilium ( Rohatgi et al . , 2007a; Kovacs et al . , 2008 ) . Some data suggest that Smo trafficking through membranous compartments is controlled by small lipids and the sterol-sensing domain of Ptch1 ( Martin et al . , 2001; Bijlsma et al . , 2006; Corcoran and Scott , 2006; Yavari et al . , 2010 ) . Since the structural framework of Ptch1 resembles that of bacterial amino acid transporters ( Carstea et al . , 1997 ) , it is conceivable that Ptch1 controls Smo activity or trafficking through such a small molecular intermediate . However , little evidence is available to account for how Ptch1 internalization through endocytosis is regulated , and it is unclear whether ciliary trafficking and endocytosis are obligatorily coupled ( Nachury et al . , 2010 ) . Receptor endocytosis plays crucial roles in coordinating the strength and duration of many cell signaling systems ( Piddini and Vincent , 2003; Polo and Di Fiore , 2006 ) . At various steps of the endocytic pathway , from the plasma membrane to the endosomes , receptors can be sorted to the proteolytic lumens of lysosomes , leading to desensitization , or back to the plasma membrane for a rapid recovery of cellular responsiveness . In addition to the classical Clathrin-mediated endocytosis , recent advances indicate that membrane receptors are also internalized through lipid rafts ( Le Roy and Wrana , 2005; Lajoie and Nabi , 2010 ) , which are specialized membrane domains enriched in cholesterol and sphingomyelin and stabilized by Caveolin 1 ( Cav-1 ) ( Allen et al . , 2007 ) . Unlike the Clathrin-mediated endocytosis , cargos of caveolae were shown to be unloaded to late endosomes , thereby bypassing early endosomes ( Quirin et al . , 2008; Hayer et al . , 2010; Sandvig et al . , 2011 ) . A major forward endocytic sorting signal is ubiquitination ( Hicke and Dunn , 2003; Mukhopadhyay and Riezman , 2007; Hayer et al . , 2010 ) , and many HECT-domain E3 ligases have been implicated in the Ubiquitin control of endocytosis , including Smurf2 ( Di Guglielmo et al . , 2003; Metzger et al . , 2012 ) , which was first identified as a negative regulator of TGF-β/BMP signaling ( Kavsak et al . , 2000; Zhang et al . , 2001 ) . Here , we present evidence that Smurf1 and Smurf2 are the Ubiquitin E3 ligases that promote Ptch1 movement from lipid rafts to late endosomes for subsequent degradation in lysosomes . This movement is essential for Ptch1's clearance from primary cilia , Shh pathway activation , and the role of Shh in sustaining the proliferation of cerebellar granule cell precursors . In light of the negative feedback control of Shh signaling by Ptch1 , this destruction system would allow the level of signaling output to be set precisely according to the level of the Ptch1 protein .
The C-terminal tails of Drosophila Ptc and mouse Ptch1 play an important role in determining its membrane distribution and stability , possibly through the highly conserved ‘PPXY’ motif ( Lu et al . , 2006; Kawamura et al . , 2008 ) , which is recognized by the WW domain frequently found in HECT-domain E3 ligases ( Metzger et al . , 2012 ) . Mammalian Ptch1 contains an evolutionarily conserved C-terminal ‘PPXY’ motif and a second one in the third intracellular loop ( Figure 1—figure supplement 1 ) , whereas Ptch2 does not and is quite stable ( Kawamura et al . , 2008 ) . Under a confocal microscope and in transfected murine embryonic fibroblasts ( MEFs ) , Ptch1-GFP was primarily detected in punctate vesicles ( Figure 1A ) , consistent with what was reported in COS and HeLa cells ( Incardona et al . , 2000; Karpen et al . , 2001 ) ; a large proportion of these Ptch1-GFP vesicles were likely to be endosomes ( see below and in Martin et al . , 2001; Incardona et al . , 2002 ) . In light of the ubiquitination control of endocytosis , we suspected that the ‘PPXY’ motifs of Ptch1 might be the signal that regulates its turnover through endosomes and lysosomes . To test this hypothesis , we sought to determine how Ptch1 engages the endocytic pathway by focusing our attention at the rim of the plasma membrane , where treatment with conditioned medium ( CM ) from HEK293 cells expressing the N-terminal signaling fragment of Shh ( ShhN ) for 1 hr rendered some of the Ptch1-GFP vesicles also positive for Cav-1 ( Figure 1A ) , a specific marker of lipid rafts . To quantify the colocalization , we sampled 10 randomly selected rim areas from different cells imaged for each data point and calculated the colocalization coefficient . The results indicated that ShhN almost doubled the colocalization coefficient between Ptch1-GFP and endogenous Cav-1 from 0 . 37 ± 0 . 04 to 0 . 68 ± 0 . 04 ( Figure 1A , B ) . Blocking late endosome/lysosome passage with chloroquine ( Chlq ) and concanavalin A ( ConA ) or lysosomal proteolysis with leupeptin ( Leu ) showed similar effects ( Figure 1A , B , Figure 1—figure supplement 2 ) . In contrast , the mutant Ptch1Δ2PY-GFP that lacks both ‘PPXY’ motifs exhibited a higher level of colocalization with Cav-1 than its wildtype counterpart even without ShhN treatment ( Figure 1A , B ) . Some of the Ptch1-GFP vesicles at the plasma rim were also positive for Clathrin heavy chain that marks the Clathrin-coated pits , but in contrast to the ligand-inducible enrichment in Cav-1 lipid rafts , the fraction of Ptch1-GFP in Clathrin-coated pits was affected neither by ligand treatment nor deletion of the ‘PPXY’ motifs ( Figure 1C , D ) . To complement the confocal imaging experiments , we conducted a co-sedimentation experiment in a discontinuous sucrose density gradient and found that both Ptch1-FLAG and Ptch1Δ2PY-FLAG co-sedimented with Cav-1 in 20% and 25% buoyancy fractions ( Figure 1E ) , indicating that when expressed exogenously , Ptch1-FLAG can find its way into Cav-1 positive lipid rafts even without Shh induction . Thus , both deletion of the ‘PPXY’ motifs and blocking endocytosis cause Ptch1 to accumulate in lipid rafts . 10 . 7554/eLife . 02555 . 003Figure 1 . The PPXY motifs define sorting signals from lipid rafts to late endosomes . ( A ) Confocal images showing colocalization of exogenously expressed Ptch1-GFP or Δ2PY ( green ) with native Cav-1 ( red ) at the rim of the plasma membrane , and ( B ) calculation of the colocalization coefficients in ( A ) in transfected MEFs . ShhN-CM and Chlq were added 1 hr prior to fixation . The chamber slides were chilled at 4°C for 20 min and then shifted to 37°C for another 20 min before fixation with 4% paraformaldehyde in PBS . The cells were then permeabilized with 0 . 5% Triton X-100 and stained with an antibody for Cav-1 . ( C ) Representative images and ( D ) quantification of colocalization between Ptch1-GFP and clathrin heavy chain . MEFs were transfected with Ptch1-GFP or PtchΔ2PY-GFP , then treated with ShhN or Ctrl medium for 1 hr before fixation . ( E ) Western analyses of sucrose gradient fractions showing Ptch1-FLAG co-sedimented with Smurf2CG-Myc and Cav-1 . Δ2PY was inefficient in bringing Smurf2CG-Myc into Cav-1 positive sedimentation fractions . ( F ) Western blot analyses of stabilities of Ptch1 and the ‘PPXY’ motif mutants in MEFs . Chlq or MG132 treatment was carried out for 4 hr . The confocal images were taken with a 63x objective , and the insets in 1A were digitally magnified . Bars represent mean ± standard deviation ( SD ) . Statistical analyses were performed by two-tail Student's t test . ***p<0 . 001 , and n . s . , not statistically significant ( p>0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02555 . 00310 . 7554/eLife . 02555 . 004Figure 1—figure supplement 1 . Position and sequence alignment of ‘PPXY’ motifs . ( A ) Schematic representation of Ptch1 constructs ( left ) and sequence alignments ( right ) of Drosophila and vertebrate Ptch1 surrounding the two evolutionarily conserved PPXY motifs . DOI: http://dx . doi . org/10 . 7554/eLife . 02555 . 00410 . 7554/eLife . 02555 . 005Figure 1—figure supplement 2 . Lysosomal inhibitors cause Ptch1-GFP to accumulate in lipid rafts . ( A ) Representative confocal images showing accumulation of Ptch1-GFP in Cav-1 positive lipid rafts after blocking endocytosis with lysosomal inhibitors Leu and ConA . ( B ) Calculation of colocalization coefficients in ( A ) . The confocal images were taken with a 63x oil lens , and the insets were digitally magnified . Bars represent mean ± standard deviation ( SD ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02555 . 005 Since an end point of endocytosis is degradation in lysosomes , we further asked if wildtype Ptch1 and ‘PPXY’ motif mutants accumulate differently in the presence of proteasomal or lysosomal blocker . When expressed in MEFs , Ptch1-GFP was an unstable protein; the bulk of which appeared to turnover via proteasomes as Ptch1-GFP accumulated to a very high level in the presence of MG132 ( Figure 1F , compare lanes 1 and 3 ) . A small portion of Ptch1-FLAG appeared to turnover via lysosomes as indicated by the moderate level of accumulation in the presence of lysosomal inhibitor Chlq ( Figure 1F , lanes 1 and 2 ) . In contrast , Ptch1 mutants lacking either one of or both ‘PPXY’ motifs were relatively stable when expressed in MEFs , although they could be induced to accumulate further by MG132 but not Chlq ( Figure 1F , lanes 4–12 ) . These results suggest that Shh promotes turnover of at least a portion of ectopically expressed Ptch1 via endosomes and lysosomes , but the entry point is likely the Cav-1 positive lipid rafts rather than the conventional clathrin-coated pits . To ascertain if the ‘PPXY’ motifs are the actual signal that sorts Ptch1 from lipid rafts to endosomes/lysosomes , we asked if Ptch1-GFP or Δ2PY-GFP could be identified in early endosomes , late endosomes , or lysosomes , which are marked by Rab5-RFP , Rab7-RFP , or Lamp1-RFP , respectively . In the absence of ShhN , Ptch1-GFP and Rab7-RFP could be readily detected together in punctate vesicles , and ShhN treatment drastically increased that colocalization as indicated by colocalization coefficient , which increased from 0 . 29 ± 0 . 03 to 0 . 51 ± 0 . 02 ( Figure 2A , B ) . Similar colocalization between Ptch1-GFP and endogenous Rab7 was also observed under ShhN treatment using specific antibodies ( Figure 2—figure supplement 1 ) . We could not detect vesicles marked positively with both Ptch1-GFP and Lamp1-RFP or Ptch1-GFP and endogenous Lamp1-RFP without blocking lysosomal enzymes by leupeptin ( Figure 2—figure supplement 2A ) , but colozalization between Ptch1-GFP and endogenous Lamp1 was revealed with the use of leupeptin ( Figure 2C ) . We did not see Ptch1-GFP colocalizing with either transfected Rab5-RFP ( Figure 2—figure supplement 2B ) or endogenous Rab5 ( Figure 2—figure supplement 3 ) without or with ShhN treatment . These observations are consistent with the notion that endocytic cargos of caveolae are unloaded directly to late endosomes , bypassing early endosomes ( Quirin et al . , 2008; Hayer et al . , 2010; Sandvig et al . , 2011 ) . In contrast to Ptch1-GFP , Δ2PY-GFP was never found together with any of the three endosome/lysosome markers and ShhN treatment caused no statistically significant change thereof ( Figure 2A–C , Figure 2—figure supplements 1 and 2 ) , indicating that Shh is not able to induce Δ2PY to move beyond lipid rafts to enter late endosomes . 10 . 7554/eLife . 02555 . 006Figure 2 . PPXY motifs are required for Shh-induced endocytosis of Ptch1 . ( A ) Confocal images showing colocalization of Ptch1-GFP or Δ2PY ( green ) with Rab7-RFP ( red ) , and ( B ) calculation of the colocalization coefficients in ( A ) in transfected MEFs . ( C ) Confocal images showing localization of Ptch1-GFP or Δ2PY ( green ) in vesicles marked anti-Lamp1 ( red ) in the presence of 1 mg/ml leupeptin . ( D ) Confocal imaging and ( E ) calculation of colocalization coefficient of Ptch1-GFP and Rab7-RFP in Kif3a−/− and control MEFs . ShhN treatment was for 1 hr and the cells were processed as in Figure 1A . Statistical analyses were performed by two-tail Student's t test . ***p<0 . 001 , and n . s . , not statistically significant ( p>0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02555 . 00610 . 7554/eLife . 02555 . 007Figure 2—figure supplement 1 . Shh promotes colocalizaiton of Ptch1-GFP with endogenous Rab7 in late endosomes . Representative confocal images showing ShhN treatment promotes colocalization of Ptch1-GFP in late endosomes visualized by anti-Rab7 . Transfected MEFs were treated with ShhN-CM or control conditioned medium for 1 hr , followed by incubations at 4°C for 20 min and 37°C for 20 min . The close-up images were digitally amplified . DOI: http://dx . doi . org/10 . 7554/eLife . 02555 . 00710 . 7554/eLife . 02555 . 008Figure 2—figure supplement 2 . Lack of colocalization of Ptch1-GFP or Δ2PY-GFP with exogenous Rab5-RFP and Lamp1-RFP without blocking lysosomal turnover . Representative Confocal images and quantification of colocalization coefficients showing that Ptch1-GFP or Δ2PY-GFP does not colocalize with Lamp1-RFP ( red ) ( A ) or Rab5-RFP ( B ) . Transfected MEFs were treated ShhN or control conditioned medium without Leupeptin for 2 hr , and then the cells were processed as in Figure 2A . DOI: http://dx . doi . org/10 . 7554/eLife . 02555 . 00810 . 7554/eLife . 02555 . 009Figure 2—figure supplement 3 . Ptch1-GFP or Δ2PY-GFP was not found in early endosomes marked by anti-Rab5 immunofluorescence staining . Representative Confocal images showing Ptch1-GFP or Δ2PY-GFP and endogenous Rab5 in non-overlapping green or red channel , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 02555 . 009 The current paradigm stipulates that Shh induces Ptch1 exit from the primary cilium during signaling ( Rohatgi et al . , 2007a; Kovacs et al . , 2008 ) . This prompted us to ask if ciliary export or its structural integrity is prerequisite to endocytosis of Ptch1 by comparing the abilities of Ptch1-GFP to associate with Rab7-RFP in Kif3a−/− or otherwise isogenic control MEFs . Although Kif3a−/− MEFs do not make cilia ( Chen et al . , 2011 ) , Ptch1-GFP could still proceed to late endosomes/lysosomes under the influence of ShhN unabatedly ( Figure 2D , E ) , implying that Ptch1 endocytosis is downstream from or independent of ciliary trafficking . Based on results from the above several lines of investigation , we conclude that the ‘PPXY’ motifs constitute sorting signals that direct Ptch1 to move into late endosomes for turnover in lysosomes . This sorting event likely takes place in Cav-1 positive lipid rafts since Δ2PY accumulates there in the absence of this signal . Ptc or Ptch1 endocytosis has been observed in cells from Drosophila to mammals for some time ( Denef et al . , 2000; Incardona et al . , 2000 , 2002; Martin et al . , 2001; Torroja et al . , 2004; Lu et al . , 2006 ) , but its role was primarily attributed to ligand sequestration or clearing ( Incardona et al . , 2000; Torroja et al . , 2004 ) . In Drosophila , the role of Ptc in ligand sequestration has been shown to be separable from that of signaling based on analyses of certain mutants ( Chen and Struhl , 1996; Torroja et al . , 2004 ) . However , we observed that when re-expressed in Ptch1−/− MEFs , the ‘PPXY’ motif mutants accumulated in the primary cilium , in contrast to their wildtype counterpart; ShhN treatment effectively forced Ptch1-GFP to exit the primary cilium , but it was less effective against these mutants ( Figure 3A , B ) . Ciliary accumulation of the ‘PPXY’ motif mutants is likely a consequence of their inability to endocytose , rather than a specific defect of ciliary export , since these mutants also accumulate in lipid rafts ( Figure 1A , B ) and blocking endocytosis with high concentration of leupeptin showed a similar effect without or with ShhN treatment ( Figure 3—figure supplement 1 ) . Combined with results from the stability experiment ( Figure 1F ) , this observation indicated that these two ‘PPXY’ motifs play an equivalent role in regulating Ptch1 function in cilia . To support this notion , we made temporal measurements of endogenous Smo translocating into the primary cilium , which is an obligatory early event of Shh signaling and was reported as concurrent to the exit of Ptch1 therefrom ( Rohatgi et al . , 2007a ) . In Ptch1−/− MEFs , immunofluorescence staining showed that Smo was constitutively present in the primary cilium ( Figure 3C ) , as expected ( Corbit et al . , 2005; Rohatgi et al . , 2007a; Kovacs et al . , 2008 ) . Re-introducing Ptch1-GFP cleared Smo out of the primary cilium , but ShhN treatment allowed Smo to move back in to nearly its full extent within 4 hr ( Figure 3C , D ) . Conversely , ShhN treatment triggered the ciliary export of Ptch1 at a rate comparable to that of Smo import ( Figure 3C , and compare Figure 3D , E ) . Re-introducing Δ2PY , on the other hand , only allowed a substantially lower level of Smo to be imported back into cilia after ShhN treatment and Δ2PY was itself resistant to Shh-induced export ( Figure 3C , E ) . 10 . 7554/eLife . 02555 . 010Figure 3 . The ‘PPXY’ motifs regulate the opposing movements of Ptch1 out of and Smo into the primary cilium . ( A ) Representative confocal images and ( B ) distribution of GFP fluorescence showing accumulation of the ‘PPXY’ motif mutants of Ptch1 in primary cilia in the absence or presence of ShhN . Two-tail Student's t test was used for statistical analysis . ***p<0 . 001 , n . s . , not significant ( p>0 . 05 ) . ( C ) Immunofluorescence of GFP as well as endogenous Smo ( red ) and acetylated tubulin ( blue ) staining in Ptch1−/− MEFs transfected with Ptch1-GFP or Δ2PY . ( D ) Quantification of anti-Smo staining and ( E ) GFP fluorescence as in ( C ) . Only transfected GFP positive cells were counted for the ciliary localization of endogenous Smo . In all of the above experiments , transfected cells were grown to confluence and then serum-starved for 24 hr to allow for ciliogenesis . ShhN-CM treatment was for 24 hr , or as indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 02555 . 01010 . 7554/eLife . 02555 . 011Figure 3—figure supplement 1 . Inhibition of Lysosomal turnover dampens Shh-induced ciliary exit of Ptch1-GFP . Representative confocal images and calculations thereof showing Ptch1-GFP fluorescence accumulated in primary cilia . ShhN and leupeptin ( 1 mg/ml ) were added to the WT MEFs for 2 hr . Two-tail Student's t test was used for statistical analysis . **p<0 . 01 , n . s . , not significant ( p>0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02555 . 011 As a ligand-binding and inhibitory receptor , the functions of Drosophila Ptc are twofold; first acting through Smo , Ptc negatively regulates downstream pathway signaling cell autonomously , and second , through ligand sequestration Ptc suppresses Hh signaling in neighboring cells . To determine if Ptch1 endocytosis impinges on downstream pathway activation , we measured the ability of Ptch1-GFP or Δ2PY to confer Shh inducibility to the 8xGliBS-luc reporter in Ptch1−/− MEFs . When co-transfected with Ptch1-GFP , the 8xGliBS-luc reporter showed a robust inductive response to ShhN , resulting in a dose–response curve typical of a substrate-enzyme relationship; however , this reporter was barely induced by ShhN when it was co-transfected with Δ2PY ( Figure 4A ) . The Shh signaling blockade imposed by Δ2PY could be by-passed by siRNA-mediated knockdown of Sufu ( Figure 4B ) , a downstream negative regulator , suggesting that the blockade is pathway-specific and occurs upstream of Sufu function . So far all our evidence points to inability of the ‘PPXY’ motif mutants to undergo Shh-induced endocytosis rather than a defect in their intrinsic activity . Indeed , in Ptch1−/− MEFs , these mutants were equally effective as wildtype Ptch1 or cyclopamine in suppressing 8xGliBS-luc reporter independent of the Shh ligand ( Figure 4C ) . Finally to address the effect of the ‘PPXY’ motifs deletion on the non-cell-autonomous function of Ptch1 , we designed a ‘mixing’ experiment , in which Ptch1−/− MEFs re-expressing wildtype Ptch1-GFP or Δ2PY-GFP were mixed at 5 to 1 ratio with a line of stable NIH3T3 cells harboring the genomically integrated 8xGliBS-luc reporter ( Chen et al . , 2011 ) . In the presence of limiting amount of ShhN ( 1:64 dilution of the conditioned medium ) , Δ2PY showed a robust inhibition of the ligand-induced reporter activity in the neighboring cells; however this effect was nullified at high ShhN concentration ( 1:16 dilution ) ( Figure 4D ) . 10 . 7554/eLife . 02555 . 012Figure 4 . The ‘PPXY’ motifs are required for eliciting both cell and non-cell autonomous transcriptional responses to Shh . ( A ) Luciferase assays for Ptch1 and the Δ2PY mutant in Ptch1−/− MEFs that were transfected together with the 8xGliBS-luc reporter construct . Each data point was obtained in triplicate and the error bars denote the standard error . ( B ) Rescuing Shh induction blockade imposed by Δ2PY using siSufu in Ptch1−/− MEFs . The experiment was set up as in ( A ) except that Sufu was knocked down by siRNA at the same time as cDNA transfection and 1:16 dilution of ShhN-CM was used . ( C ) Relative activities of the GliBS-luc reporter that was co-expressed with Ptch1 or the ΔPY mutants in Ptch1−/− MEFs without ShhN-CM treatment . The ΔPY mutants displayed same inhibitory effect as WT Ptch1 . ( D ) Non-cell autonomous inhibition of GliBS-luc reporter in neighboring cells . Ptch1−/− MEFs transfected with Ptch1-FLAG , Δ2PY , or the vector control were mixed at 5:1 ratio with NIH3T3:GliBS-luc reporter cells . The cells were given ShhN-CM for 24 hr , and two-tail Student's t test was used for statistical analyses . *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 , and n . s . , not significant ( p>0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02555 . 012 In summary , our data indicate that whereas Ptch1 engagement to the ligand may have a nominal effect of internalizing Shh , it can be also regarded as an interaction that allows Shh to induce Ptch1 clearance from the primary cilium , the site of Shh signaling , and this regulation equally impinges on both cell and non-cell autonomous signaling functions of Ptch1 . Previously , the C-terminal domain of Drosophila Ptc was shown to be recognized by Nedd4 HECT-domain E3 ligase ( Lu et al . , 2006 ) . We expressed mouse Nedd4 and Nedd4l together with Ptch1-FLAG in HEK293 cells , and found that neither one promoted Ptch1 degradation , and several other HECT-domain E3 ligases including Wwp1 , Wwp2 , Huwe1 , Herc1 , Herc3 , Herc4 , Herc6 , Hecw1 , and Hecw2 also showed no effect , but co-expression of Smurf1 or Smurf2 did ( Figure 5A , B ) . Consistent with a specific role , the ligase deficient Smurf1CA and Smurf2CG mutants failed to influence Ptch1 stability ( Figure 5B ) . Since the ‘PPXY’ motif mutants accumulated in cilia , we asked if knockdown of either Smurf or both with siRNAs could augment the ciliary localization of Ptch1-GFP . We found this was the case in NIH3T3 cells without ( Figure 5C , D ) or with ShhN treatment ( Figure 5—figure supplement 1 ) . Because Smurf2 is known to direct the TGF-β type I receptor and the μ opioid neuropeptide receptor to endocytic turnover ( Di Guglielmo et al . , 2003; Henry et al . , 2012 ) , we posited that Smurf1 and Smurf2 might be the enzymes that control Ptch1 endocytosis and chose them for further analysis . 10 . 7554/eLife . 02555 . 013Figure 5 . Smurf1 and Smurf2 are E3 ligases required for Shh signaling . ( A ) Western analyses of Ptch1-FLAG in HEK293 cells that were co-transfected with cDNAs encoding a battery of HECT-domain E3 ligases as indicated , and ( B ) ligase deficient Smurf mutants . β-actin was used as a loading control . ( C ) Representative confocal images and ( D ) calculations of Ptch1-GFP fluorescence accumulated in primary cilia as the result of siRNA knockdown of Smurf1 , Smurf2 , or both in NIH3T3 cells . Primary cilia were marked by acetylated tubulin ( red ) . ( E ) RT-PCR detection of Gli1 , Smurf1 , and Smurf2 mRNAs in wildtype ( WT ) , Smurf1−/− , and Smurf2−/− MEFs transfected with non-silencing ( NS ) or Smurf-specific siRNAs as indicated . HPRT mRNA was used as an internal control . A representative gel image is shown here . ( F ) RT-qPCR quantification of fold induction of Gli1 mRNA from an experiment as in ( E ) . Fold induction was calculated using Gli1 mRNA level normalized against that of Hprt for even loading and then against the normalized Gli1 mRNA level from cells transfected with NS siRNA and without ShhN treatment . ( G ) RT-qPCR analysis of relative levels of Smurf1 and Smurf2 mRNAs from the experiment in ( F ) . ( H ) RT-qPCR detection of endogenous Gli1 mRNAs in Smurf1−/−;Smurf2fl/fl MEFs that were infected with either Ad-GFP ( mock ) or Ad-Cre for 12 hr , and then treated with either control or ShhN conditional medium for 72 hr . ( I ) Western analyses of endogenous Smurf2 in Smurf1−/−;Smurf2fl/fl MEFs from the experiment in ( H ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02555 . 01310 . 7554/eLife . 02555 . 014Figure 5—figure supplement 1 . Knockdown of Smurf1 and Smurf2 simultaneously dampens Shh-induced ciliary exit of Ptch1-GFP . Representative confocal images and calculations thereof showing Ptch1-GFP fluorescence accumulated in primary cilia . NIH3T3 cells were transfected with siRNAs specific for Smurf1 and Smurf2 , and then the cells were treated with control or ShhN conditioned medium before Ptch1-GFP was visualized in cilia and quantified . Two-tail Student's t test was used for statistical analysis . *p<0 . 05 , ***p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 02555 . 014 Smurf1 and Smurf2 share redundant functions during development as individually knockout Smurf1−/− or Smurf2−/− mice are healthy and fertile , but the embryos lacking both genes were not able to develop to term ( Yamashita et al . , 2005; Narimatsu et al . , 2009; Blank et al . , 2012 ) . To assess the role of Smurfs in Shh signaling , we quantified the transcriptional responses of endogenous Gli1 by RT-PCR ( Figure 5E ) and RT-qPCR ( Figure 5F ) in MEFs with different Smurf genetic background , and found that silencing Smurf1 and Smurf2 simultaneously in wildtype MEFs completely abolished Shh induction of Gli1 ( Figure 5E , F ) . MEFs that lack one of the two Smurf genes still mounted a considerable Gli1 transcriptional response to ShhN; however , silencing the remaining Smurf2 allele in Smurf1−/− or Smurf1 allele in Smurf2−/− MEFs , respectively , led to marked curtailment of Gli1 activation ( Figure 5E , F ) . Expression of Smurfs showed a compensatory upregulation in response to the loss of the other Smurf in these MEFs as reported ( Yamashita et al . , 2005; Tang et al . , 2011 ) , but surprisingly , ShhN induced expression of both Smurfs ( Figure 5E , G ) . During the course of this investigation , we generated Smurf1−/−;Smurf2fl/fl mice , which will be described in detail elsewhere . In Smurf1−/−;Smurf2fl/fl MEFs , Ad-cre infection-mediated ablation of conditional Smurf2fl alleles severely dampened the Gli1 transcriptional response to ShhN ( Figure 5H , I ) . Similarly , two other Shh signaling responses , namely Shh-induced ciliary import of Smo and Gli3 , were also affected ( Figure 6A–C ) . Since we could rescue Shh induction of GliBS-luc responses in Ad-cre infected Smurf1−/−;Smurf2fl/fl MEFs ( Smurfs null ) by reintroducing wildtype Smurf1 or Smurf2 but not mutant Smurf1CA or Smurf2CG cDNA ( Figure 6D ) , or by siRNA-mediated knockdown of Suppressor of fused ( Sufu ) ( Figure 6E ) , an essential downstream negative regulator of Shh signaling , the observed defects of GliBS-luc induction have to be Smurfs and Shh pathway specific . Taken together , the above results show that simultaneous inactivation of both Smurf genes and removal of the ‘PPXY’ motifs of Ptch1 have congruent effects on various Shh signaling events , and indicate that a common Smurf function is required at a step upstream from the control of the ciliary import of Smo . 10 . 7554/eLife . 02555 . 015Figure 6 . Smurf1 and Smurf2 are required For Shh signaling . ( A ) Representative confocal images of Smo and Gli3 immunofluorescence staining in cilia of wildtype ( WT ) , Smurf1−/−;Smurf2fl/fl , or Smurf1−/−;Smurf2fl/fl MEFs infected with Ad-Cre viruses . ( B ) Quantification of Smo and ( C ) Gli3 immunofluorescence staining in cilia of ( A ) . In the above experiments , ShhN treatment was carried out for 24 hr , and the means and standard deviation were calculated from two independent experiments ( n = 20 ) . ( D ) GliBS-luc assays in Smurf1−/−;Smurf2fl/fl MEFs showing the deficiency of Shh induction associated with genomic ablation of both Smurfs can be rescued by re-introducing either wildtype Smurf1 or Smurf2 but not their corresponding mutants . ( E ) GliBS-luc reporter assays for the ability of siSufu to by-pass the requirement of Smurfs in Shh signaling . Smurf1−/−;Smurf2fl/fl MEFs were infected with Ad-cre and then transfected with siSufu or ns control . The cells were then treated with a series of dilutions of ShhN-CM before luciferase activities were assayed . Error bars denote standard deviations . Statistical analyses were performed by two-tail Student's t test . *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 , and n . s . , not significant ( p>0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02555 . 015 If Smurfs are the E3 ligases that recognize the endocytic sorting signals of Ptch1 , these proteins should physically interact in lipid rafts . A number of evidence demonstrates that this is the case . First , in non-permeabilizing MEFs , we found exogenously expressed Ptch1-RFP colocalized with the ligase deficient , GFP-tagged Smurf2CG mutant in Cav-1 positive lipid rafts at the rim of the plasma membrane ( Figure 7A ) . Although first identified as modulators of TGF-β/BMP signaling , Smurfs are preferentially localized in the nucleus ( Kavsak et al . , 2000 ) and play a crucial function in maintaining genomic stability ( Blank et al . , 2012 ) . Serendipitously , we found that treatment with ShhN ligand or co-expression with Ptch1-RFP each caused Smurf2GFP to move from the nucleus to the cytoplasm ( Figure 7B , Figure 7—figure supplement 1 ) . In light of the Shh induction ( Figure 5E , F ) , these results indicate that Shh signaling could increase the cytoplasmic pool of Smurfs . Third , fluorescence resonance energy transfer ( FRET ) analysis showed that Ptch1-CFP was localized in close proximity with Smurf1-YFP or Smurf2-YFP at punctate intracellular vesicles in MEFs ( Figure 7C , D ) , and ShhN treatment enhanced this colocalization ( Figure 7E ) . However , Δ2PY-CFP failed to generate FRET with Smurf2-YFP ( Figure 7C , D ) . Theses result were further corroborated in the discontinuous sucrose gradient sedimentation experiment described earlier , in which the ligase-deficient Smurf2CG-Myc co-sedimented in the Cav-1-containing 20–25% sucrose fractions readily with Ptch1-FLAG , whereas Δ2PY was inefficient in bringing Smurf2CG-Myc into these fractions ( Figure 1E ) . Finally , using co-immunoprecipitation , we demonstrated that Ptch1 specifically binds either Smurf1 or Smurf2 , and Ptch1 mutants lacking either PY-1 or PY-2 motif can still bind Smurfs , albeit with reduced affinity; however , Δ2PY completely lacks affinity for either Smurf1 or Smurf2 ( Figure 7F ) . 10 . 7554/eLife . 02555 . 016Figure 7 . Colocalization and interaction between Ptch1 and Smurfs in Cav-1 positive lipid rafts . ( A ) Confocal images showing colocalization of GFP-Smurf2CG and Ptch1-RFP in Cav-1 positive lipid rafts . The cells were not permeabilized before they were stained with anti-Cav-1 , and the images were taken with a 63x oil lens . ( B ) Quantification of nuclear and cytoplasmic distribution of Smurf2GFP as in Figure 7—figure supplement 1 . The percentage of mostly nuclear ( N > C ) , even distribution ( N = C ) , or mostly cytoplasmic ( N < C ) of Smurf2GFP pattern cells was calculated based images of 40 cells at each data point . ( C ) FRET analysis of Ptch1-CFP or Δ2PY-CFP interaction with Smurf1-YFP or Smurf2-YFP in transfected MEFs . Representative images of CFP , YFP , FRET fluorescence , and N-FRET are shown . ( D ) Quantification of N-FRET values using the sensitized emission method , which is expressed as means plus SD in the bar graph . ( E ) FRET analysis of Ptch1-CFP interaction with Smurf2-YFP in transfected MEFs that were treated with ShhN or control conditioned medium for 2 hr . Quantification of N-FRET values described in ( D ) . ( F ) Co-immunoprecipitation analyses of FLAG-Ptch1 and the ‘PPXY’ motif mutants with Myc-tagged Smurf1CA or Smurf2CG ligase-deficient mutants . The immunocomplexes were precipitated using anti-FLAG , and blotted with anti-Myc . DOI: http://dx . doi . org/10 . 7554/eLife . 02555 . 01610 . 7554/eLife . 02555 . 017Figure 7—figure supplement 1 . ShhN treatment and co-expression with Ptch1 caused Smurf2 to redistribute from the nucleus to the cytoplasm . Representative fluorescent images showing subcellular localization of Smurf2GFP in MEFs co-transfected with empty vector or Ptch1-RFP and treated without or with ShhN conditioned medium . DOI: http://dx . doi . org/10 . 7554/eLife . 02555 . 01710 . 7554/eLife . 02555 . 018Figure 7—figure supplement 2 . Neither Smurf1 nor Smurf2 interact with Smo . Western analyses of Ptch1-FLAG or SmoA1-FLAG immunoprecipitated with anti-FLAG in HEK293 cells that were also co-transfected with Myc-tagged Smurf1CA or Smurf2CG ligase-deficient mutants . Although Smurf1CA or Smurf2CG was readily detected in the anti-Ptch1-FLAG precipitates , they did not co-precipitate with Smo-FLAG . DOI: http://dx . doi . org/10 . 7554/eLife . 02555 . 018 To delineate the requirement of Smurfs for Shh-induced Ptch1 turnover , we took the advantage of the conditional Smurf1−/−;Smurf2fl/fl MEFs , and quantified the turnover rate of exogenously expressed Ptch1-FLAG following cyclohexamide treatment without or with removal of the Smurf2 alleles following Ad-cre infection . The results indicated that Ptch1-FLAG was indeed rendered stable against ShhN induced degradation by the removal of the Smurf2fl conditional alleles whereas the stability of Δ2PY was resistant to change in response to either ShhN treatment or eradication of Smurf’s function ( Figure 8A–D ) . The induction by Shh is likely a function of ligand-binding , rather than a signaling outcome , as the loop2 mutant Ptch1 that lacks the ability to bind Shh ( Briscoe et al . , 2001 ) completely lost the capacity to respond to ShhN treatment in wildtype MEFs , although it was more stable in Smurf-null MEFs ( Figure 8E , F ) . We further found that Shh-induced endocytic turnover of Ptch1 was not affected in Smo null MEFs ( Figure 8G , H ) , suggesting that it is an upstream signaling event , independent of Smo function . 10 . 7554/eLife . 02555 . 019Figure 8 . Smurfs are required for the Shh-induced endocytic turnover of Ptch1 . Western analysis of Ptch1-FLAG and Δ2PY-FLAG turnover rates ( A ) and quantification thereof ( B ) in WT MEFs . ShhN and CHX were added for duration as indicated . ( C ) Western analysis of Ptch1-FLAG and Δ2PY-FLAG turnover rates ( C ) and quantification thereof ( D ) in Smurf1−/−;Smurf2fl/fl MEFs infected with Ad-cre . ( E ) Western analysis of Ptch1-Δloop2-FLAG turnover rate and quantification thereof ( F ) in WT ( upper ) and Smurf1−/−;Smurf2fl/fl MEFs infected with Ad-cre ( lower ) . ( G ) Western analysis of Ptch1-FLAG turnover rate and quantification thereof ( H ) in WT ( upper ) and Smo−/− MEFs ( lower ) . Each data point denotes mean ± standard deviation from two independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 02555 . 019 To demonstrate the Ubiquitin E3 ligase activity of Smurfs on Ptch1 , we assayed for the ability of Ptch1-FLAG or Δ2PY to be ubiquitinated by HA-tagged Ubiquitin ( HA-Ub ) in Smurf1−/−;Smurf2fl/fl MEFs . In these cells , Ptch1-FLAG was readily ubiquitinated , but the level of ubiquitination of Ptch1Δ2PY-FLAG was diminished ( Figure 9A ) . More importantly , neither of the two forms of Ptch1 was ubiquitinated after the conditional Smurf2fl alleles were removed with Ad-cre , ( Figure 9A ) . We were also able to demonstrate ubiquitination of Ptch1-FLAG that was produced and isolated from HEK293 cells in an in vitro reconstituted system , in which the level of ubiquitinated species was greatly enhanced by His6-Smurf2 , but not the ligase-inactive His6-Smurf2CG purified from the insect expression system ( Figure 9B ) , indicating a direct enzyme and substrate relationship . Although we were not able to detect mono-ubiquitination , the poly-ubiquitin chains on Ptch1 are likely of both K48 and K63 linkage , as re-expression of Smurf2-Myc in Smurf2 null cells enhanced Ptch1 ubiquitination in the presence of wt , KO , K48 , or K63 ubiquitin ( Figure 9C ) . Finally , ShhN treatment enhanced the level of high molecular weight ubiquitinated Ptch1 species in wildtype MEFs ( Figure 9D ) , consistent with the ability of Shh to induce Ptch1 turnover ( Figure 8 ) . 10 . 7554/eLife . 02555 . 020Figure 9 . Smurfs are required for ubiquitin modification of Ptch1 . ( A ) Western analysis of ubiquitinated Ptch1-FLAG and Ptch1Δ2PY-FLAG in Smurf1−/−;Smurf2fl/fl MEFs infected with Ad-GFP or Ad-Cre . These MEFs were first infected with adenoviruses and then transfected with HA-Ub and the Ptch1 plasmids as marked . The exogenously expressed Ptch1 proteins were immunoisolated using anti-FLAG beads prior to analysis . ( B ) Western analysis of Ptch1-FLAG ubiquitination in vitro in a reconstituted system comprising purified recombinant His6-Smurf2 or the ligase-deficient His6-Smurf2CG from the baculovirus , HA-Ub , and an ATP regeneration system . Ptch1-FLAG was immunoisolated from HEK293 cells and the ubiquitination reaction was carried out on beads . The proteins were eluted prior to Western blot analysis . ( C ) Western analysis of ubiquitinated Ptch1-FLAG in Smurf2−/− MEFs that were also transfected with Wt , KO , K48 , or K63 ubiquitin in the absence or presence of Myc-Smurf2 . ( D ) Western analysis of ubiquitinated Ptch1-FLAG in WT MEFs treated with ShhN or control conditioned medium . Ptch1-FLAG in A-C was resolved by 6% SDS-PAGE , but a 4–12% gradient gel was used in D . DOI: http://dx . doi . org/10 . 7554/eLife . 02555 . 020 Mice deficient in both Smurf1 and Smurf2 were reported embryonic lethal due to absence of planar cell polarity among other pleiotropic defects ( Narimatsu et al . , 2009 ) . More than half of the double null embryos that we generated failed to gastrulate and those rare embryos that did escape seldom passed Theiler stage 13 , thus precluding a thorough analysis of the neural tube phenotype where Shh function is well characterized . To address the physiological relevance of Smurf regulation of Ptch1 endocytosis , we examined the role of Smurfs in sustaining the proliferation of cerebellar granule cell precursors ( GCPs ) , which has an absolute requirement for Shh . For this purpose , we cut cerebellar slices from P7 Smurf1−/−;Smurf2fl/fl pups and cultured them for 12 days in vitro as described ( Kapfhammer , 2010 ) . Anti-NeuN immunofluorescence staining revealed that the number of post-mitotic granule cells were severely reduced in slices infected with Ad-cre viruses ( Figure 10A ) , suggesting that Shh signaling was compromised there . We also isolated GCPs from cerebella of normal P7 pups of the C57/B6 strain , and cultured them in vitro . In the presence of ShhN , GCPs grew healthily for at least 5 days , but siRNA knockdown of Smurf1 and Smurf2 simultaneously blocked GCP proliferation ( Figures 10B , Figure 10—figure supplement 1A , C ) . To ascertain that the effect of Smurf knockdown was Shh-pathway specific , we repeated the above experiment using IGF1 , which is capable of sustaining the proliferation of GCPs in lieu of Shh ( Rao et al . , 2004; Fernandez et al . , 2010 ) , and found that knockdown of Smurfs had no effect on IGF-1-induced GCP growth ( Figure 10C , Figure 10—figure supplement 1B ) . Thus , these data unequivocally demonstrated that Smurf1 and Smurf2 share a critical role in supporting Shh signaling during cerebellar organogenesis . 10 . 7554/eLife . 02555 . 021Figure 10 . Requirements of Smurfs for Shh-induced organogenesis . ( A ) Immunostaining of P7 cerebellar slices cultured in vitro with anti-calbindin ( red ) and anti-NeuN ( green ) . The slices were first infected with control or cre-expressing adenoviruses for 24 hr and then continuously cultured for 12 days . Quantification of EdU incorporated GCPs in cerebellar slices cultured in the presence of ShhN from Figure 10—figure supplement 1A ( B ) or IGF-1 from Figure 10—figure supplement 1B ( C ) , respectively . The data at each time point were derived from four separate fields , and the bars denote standard deviation . DOI: http://dx . doi . org/10 . 7554/eLife . 02555 . 02110 . 7554/eLife . 02555 . 022Figure 10—figure supplement 1 . Smurfs are required for ShhN but not IGF-1 induced GCP proliferation . EdU incorporation by GCPs growing in medium containing ( A ) ShhN or ( B ) IGF-1 . Freshly isolated GCPs from normal C57/B6 mice were seeded in chamber slides that were coated with poly-D-lysine and Matrigel . The cells were then transfected with non-silencing ( NS ) control or Smurf1- and Smurf2-specific siRNAs . 12 hr later , Shh-N or IGF-1 conditioned medium was added to the culture and began the time point 0 hr . EdU was given to cells for 12 hr . ( C ) RT-PCR detection of Smurf1 and Smurf2 mRNAs for monitoring the siRNA knockdown efficiency . DOI: http://dx . doi . org/10 . 7554/eLife . 02555 . 022
Shh plays a fundamental role in setting up the body plan during embryogenesis , and is also critical in guiding stem cell differentiation for maintaining tissue homeostasis in the adult . Cell surface reception of Shh signaling is a multistep process that entails , but is not limited to , ligand engagement , reciprocal movements of Ptch1 exiting from and Smo translocating into the primary cilium , and activation of the G-protein-coupled Smo by still-controversial mechanisms ( Ogden et al . , 2008 ) . The central task of this process is to sense and convert incremental changes in the Shh gradient into corresponding levels of signaling output , thereby allowing the positional cues to be executed . In this study , we extended our knowledge of the Shh signaling activation process by revealing a ubiquitination switch that regulates Ptch1 endocytosis , which is essential in clearing Ptch1 from its site of action in the primary cilium , and to ligand sequestration , as previously described ( Incardona et al . , 2000 ) . Our data demonstrate that ubiquitination of Ptch1 mediated by the two ‘PPXY’ motifs is controlled by HECT-domain E3 ligases Smurf1 and Smurf2 , which are induced by Shh ( Figure 5E , G ) and redistributed into the cytoplasm under Shh influence ( Figure 7B , Figure 7—figure supplement 1 ) . Shh also promotes the association of Ptch1 and Smurfs in intracellular vesicles ( Figure 7E ) , most likely the Cav-1 positive lipid rafts ( Figure 1A , B ) , as well as ubiquitination ( Figure 9D ) and endosomal entry ( Figure 2A–C ) , leading to lysosomal turnover ( Figures 1F , 8A–D ) . So , an increase in the Shh signal strength would cause a corresponding increase in both the production of Ptch1 and its rate of turnover en route from the primary cilium to the lipid rafts and to the endosomes/lysosomes . This regulatory scheme is reminiscent of an electronic amplification circuit , in which a feedback loop added to an open-loop amplifier has the effect of stabilizing the gain and increasing the linearity of the output signal to a given range , which can be controlled by adjusting the feedback strength . By analogy , Shh induction of Gli1 can be viewed as the open-loop amplifier , with Ptch1 providing the negative feedback . In this wiring logic , the graded Shh morphogenic signal can be stably transformed into stepwise output responses tailored for a predetermined cell fate specification . Without endocytosis , Ptch1 would accumulate in the primary cilium ( Figure 1A , B , Figure 1—figure supplement 2 , Figure 3—figure supplement 1 ) , thus hampering Smo import and function . More importantly , without Ptch1 removal/degradation , the amplitude of Shh signaling would be restricted by the accumulation of newly synthesized inhibitory Ptch1 . Oversupplied Ptch1 could also impact on signaling in neighboring cells through non-cell autonomous inhibition . So , Ptch1 endocytosis plays a crucial role in setting the output range of Shh signaling . The presence of Ptc in membranous vesicles has long been noted in Drosophila and mammalian cells ( Capdevila et al . , 1994; Denef et al . , 2000; Ramirez-Weber et al . , 2000; Zhu et al . , 2003 ) , but its significance was not fully appreciated and regulation unknown . Ptc or Ptch1 is a 12-pass transmembrane protein , whose internal sequence spanning from IV to X transmembrane domains resembles the resistance , nodulation , division ( RND ) family of bacterial proton-driven transporter and the sterol-sensing domain found in SREBP and NPC1 ( Carstea et al . , 1997; Taipale et al . , 2002 ) . Substantial evidence in the literature suggests that Ptch1 inhibition of Smo occurs by way of small molecular intermediates that may be transported by Ptch1 through the membrane ( Di Guglielmo et al . , 2003; Bijlsma et al . , 2006; Yavari et al . , 2010 ) . Perhaps it is not a coincidence that we found Ptch1 exits the primary cilium and enters the endocytic pathway via cholesterol and sphingomyelin-rich lipid rafts , whereas Smo was shown previously to enter the primary cilium via Clathrin-coated pits when induced by Shh ( Chen et al . , 2004; Kovacs et al . , 2008 ) . It is possible that Ptch1 and Smo are required to be sorted into different membranous compartments and to keep a mutually exclusive presence in the primary cilium , so that a cross-membrane imbalance of the small molecular intermediates is attained . The RND/sterol-sensing domain is critical to Ptch1 function as multiple inactivating mutations in this region have been found in Drosophila as well as in Gorlin syndrome patients ( Martin et al . , 2001; Strutt et al . , 2001; Taipale et al . , 2002 ) . However , although certain RND mutants of Drosophila Ptc accumulate in endosomes ( Martin et al . , 2001; Strutt et al . , 2001 ) , this domain may be more important to Ptch1 function than to its endocytic turnover , since we found that combining a RND mutation with the 2-PY deletion did not alter the latter's impact on Ptch1 stability ( data not shown ) . Through cDNA-mediated screens , we have identified Smurf1 and Smurf2 as the E3 ligases responsible for generating the sorting signal for Ptch1 endocytosis . Although subsequent experiments indicated that deletion of one Smurf gene was not sufficient to inactivate Shh signaling , siRNA-mediated knockdown of either Smurf1 or Smurf2 was enough to dampen the 8xGliBS reporter response in transfected MEFs . This apparent discrepancy is likely to be reconciled by the mutual , compensatory upregulation of either of the two Smurf genes upon the loss of the other , resulting in the adaptation of single-Smurf-knockout MEFs for a robust Shh signaling response . On the other hand , such an adaptive response might not have been established in time under the conditions found in transiently transfected MEFs in response to siRNA-mediated knockdown . The observation of Shh induction of Smurf expression ( Figure 5E , G ) and cytoplasmic pivoting ( Figure 7B , Figure 7—figure supplement 1 ) further implicated Smurfs in Shh signaling . Previously , Drosophila Ptc was shown to interact with and regulated by Nedd4 ( Lu et al . , 2006 ) , another HECT-domain E3 ligase . In addition , the mouse Ptch1 was also shown to bind Nedd4 , but this interaction triggers apoptosis through ubiquitination of Caspase 9 ( Fombonne et al . , 2012 ) . It is likely that Ptch1 is regulated by multiple E3 ligases with different functional outcomes . Recently , Drosophila DSmurf was identified as a Ptc-interacting partner in a yeast 2-hybrid screen , and shown subsequently as a specific E3 ligase that regulates Ptc stability ( Huang et al . , 2013 ) . However , DSmurf was shown to promote Ptc turnover in the presence of activated SmoSD , bind Smo , and prefer ligand-unbound Ptc as a substrate ( Huang et al . , 2013 ) . We did not observe interaction between mammalian Smurfs and Smo by Co-IP experiments ( Figure 7—figure supplement 2 ) , and found that Shh induction of Ptch1 turnover proceeded unabatedly even in the absence of Smo ( Figure 8G , H ) . In Huang et al . , when ectopically expressed in the anterior compartment of the wing disc , activated SmoSD induced massive amount of Ptc; these two proteins could form a complex at the high levels , much like their mammalian counterparts do when overexpressed in HEK293 cells ( Stone et al . , 1996; Taipale et al . , 2002 ) . Perhaps , DSmurf could recognize this unnatural complex and triggers a proteasomes-mediated degradation , even specifically . Smurf2 was shown previously to function in lipid rafts ( Di Guglielmo et al . , 2003 ) , and the necessity of removing both Smurf1 and Smurf2 to reveal their requirement in Shh signaling strongly argues that this shared function has a deep root in evolution . In any event , our work presents a rather comprehensive view of the Shh pathway activation process . Considering two neighboring cells in a given Shh influence field ( Figure 11 ) , the cell that receives lower Shh input ( upper cell ) encounters a stronger feedback inhibition due to lower endocytic turnover of Ptch1 , resulting in a lower level of Shh signaling output represented by Gli1 . In the cell that receives higher Shh input ( lower cell ) , although the synthesis of Ptch1 is induced , upregulation of Smurfs and the induction of colocalization in lipid rafts ensure a faster Ptch1 turnover such that the level of Ptch1 feedback inhibition is actually low , resulting in higher pathway activity . The endocytic turnover also has impact on the ligand sequestration role of Ptch1 through controlling the availability of the ligand ‘sink’ on cell surface . In this regard , the Smurf-mediated endocytosis of Ptch1 is an essential signaling event , and it is theoretically possible to block Shh function both cell and non-cell autonomously using Smurf inhibitors , thus opening a new route for Shh-targeted cancer treatment . 10 . 7554/eLife . 02555 . 023Figure 11 . A model for the role of Smurf-mediated Ptch endocytosis in Shh signaling . DOI: http://dx . doi . org/10 . 7554/eLife . 02555 . 023
All mice were maintained and handled according to protocols approved by the Animal Care and Use Committee of the National Cancer Institute , NIH . The conditional Smurf2 knockout allele , Smurf2fl was generated by insertion of two loxP sites into introns flanking Exon 9 and 10 through homologous recombination . Further details of the construction will be described elsewhere . Smurf1−/− , Smurf2−/− , and Smurf1−/−;Smurf2fl/fl MEFs were isolated from E14 . 5 embryos and cell immortalization was carried out according to the 3T3 protocol . NIH3T3:Gli-Luc-3T3 and Ptch1−/− MEFs were described previously ( Chen et al . , 2011 ) . Full-length mouse Ptch1 cDNA was obtained from ATCC , and the FLAG , GFP , or RFP-tagged variants of which were generated by PCR and subcloned into the pRK5 vector . The ΔPY mutants of Ptch1 were generated using a PCR-based strategy . All PCR-amplified fragments were sequence verified . Plasmids for Myc-tagged Smurf1 , Smurf1CA , Smurf2 , Smurf2CG , GFP-tagged Smurf2 , HA-tagged Ub , UbKO , UbK63 and UbK48 were described previously ( Zhang et al . , 2001; Yamashita et al . , 2005 , 2008; Tang et al . , 2011; Blank et al . , 2012 ) . RFP-tagged Rab5 , Rab7 , and Lamp1 were acquired from Addgene . siRNAs specific for the mouse HECT family of E3 ligases and cDNAs encoding human HECT E3 ligases were purchased from QIAGEN ( Germantown , MD ) . Approximately 0 . 6 × 105 cells per well were seeded in Lab-Tek chambered slides and cultured for 24 hr . The cells were transfected , allowed to recover for 24 to 36 hr , and then treated with ShhN-CM or other compounds , as indicated . For visualizing ciliary proteins , the transfected cells were starved in DMEM containing 0 . 5% FBS for 24 hr before other treatments . The cells were fixed with 4% paraformaldehyde for 10 min at 4°C , and standard procedures for immunostaining were followed . The primary antibodies used were rabbit anti-Caveolin-1 ( 1:1000; Sigma-Aldrich ( St . Louis , MO ) ) , rabbit anti-Clathrin heavy chain ( 1:200; Cell Signaling Technology ( Danvers , MA ) ) , rabbit anti-Rab5 ( 1:150 , Cell Signaling Technology ) , rabbit anti-Rab7 ( 1:50 , Cell Signaling Technology ) , rabbit anti-Lamp1 ( 1:150; Sigma ) , mouse anti-acetylated Tubulin ( 1:2000; Sigma ) , rabbit anti-Gli3 ( 1:500; R&D ( Minneapolis , MN ) ) , and rabbit anti-Smo ( 1:500; a gift from Dr Rajat Rohatgi ) . Alexa-coupled secondary antibodies were purchased from Life Technologies Corp . Confocal images were acquired on a Carl Zeiss LSM710 microscope . Colocalization coefficient was calculated using Zeiss ZEN2011 program , and quantification of the fluorescence intensity of Ptch1-GFP , Smo , and Gli3 in primary cilia was carried out using Image-Pro as described previously ( Chen et al . , 2011 ) . For FRET analysis , MEFs were transfected with the plasmids encoding Ptch1-CFP or Δ2PY-CFP together with Smurf1-YFP or Smurf2-YFP . Confocal images were acquired with a 40 × objective lens . In track I , cells were excited with a 405-nm laser , and CFP signals were collected in channel II at 470–500 nm . FRET signals were collected in channel III at >530 nm . In track II , cells were excited with a 514-nm laser line , and YFP signals were collected in channel III at >530 nm . FRET efficiency between CFP and YFP , shown as N-FRET , was calculated using Zeiss ZEN2011 program , and the sensitized emission crosstalk coefficients were determined using control cells that expressed only CFP or YFP . Ptch1−/− MEFs were transfected with Ptch1-GFP or Ptch1Δ2PY-GFP along with the Rellina control ( 15:1 ) using Lipofectamine Plus ( Life technologies , Grand Island , NY ) ) . These cells were then re-seeded with NIH3T3:GliBS-luc reporter cells at 5:1 ratio . After 24 hr , the cells were treated with ShhN-CM in different dilutions for additional 36 hr before the luciferase activities were assayed using the luciferase assay system on a GloMax-96 luminometer ( Promega , Madison , WI ) . The firefly luciferase activity from the indicator cells was normalized against the Rellina luciferase activity to correct for transfection efficiency of Ptch1 constructs in the testing Ptch−/− MEFs as the measurement of non-cell autonomous inhibition by Ptch1 . Transfected cells were lysed in modified RIPA buffer ( 50 mM Tris–HCl , pH 7 . 4 , 150 mM NaCl , 1% vol/vol NP-40 , 1% n-Dodecyl β-D-maltoside , 0 . 25% wt/vol sodium deoxycholate , 1 mM DTT , and 1 × Roche cOmplete Protease Inhibitor Cocktail ) for 1 hr at 4°C . The lysate was clarified by centrifugation for 1 hr at 20 , 000×g . The protein concentration was determined using a bicinchoninic acid assay and equal amounts of total protein from each of the samples was supplemented with 6 × SDS loading buffer , incubated at room temperature for 1 hr , subjected to SDS-PAGE , followed by western blot analysis . To assay for interactions between exogenous Ptch1-FLAG and the Myc-Smurfs , transfected Ptch1-FLAG was immunopurified with anti-FLAG M2 agarose beads ( Sigma ) and subjected to SDS-PAGE , followed by western blotting with anti-Myc ( Santa Cruz Biotechnology , Dallas , TX ) . To assay for Ptch1 ubiquitination in vivo , Smurf1−/−/Smurf2flox/flox MEFs were infected with either Ad-GFP or Ad-Cre adenovirus for 24 hr , then transfected with Ptch1-FLAG or Ptch1Δ2PY-FLAG along with HA-Ub using Lipofectamine Plus ( Invitrogen ) . The cells were lysed 24 hr later and Ptch1 and its mutant were isolated with anti-FLAG agarose beads and resolved by SDS-PAGE on 6% or 4–12% gradient gels . The ubiquitinated Ptch1 was then detected with anti-HA ( Roche-Shanghai , China ) . To assay for Ptch ubiquitination in vitro , an ubiquitination assay was modified from a previously described procedure ( Tang et al . , 2011 ) . Ptch1-FLAG was captured from transfected HEK293 cell lysates using anti-FLAG agarose . After a thorough wash , the Ptch1-bound agarose was divided into three aliquots . Empty anti-FLAG agarose was used as a control . The in vitro ubiquitination assay was performed by incubating either Ptch1-bound agarose or control agarose at 37°C for 1 hr with ubiquitin-activating enzyme UBE1 , E2-conjugating enzyme UbcH5c , HA-Ub and ATP ( all from Boston Biochem , Cambridge , MA ) in the presence or absence of purified His6-Smurf2 or His6-Smurf2CG . After the incubation , the supernatant was removed , the agarose thoroughly washed , and the Ptch1-FLAG eluted using the FLAG peptide ( Sigma ) . The eluted fraction was then subjected to Western blot analysis . Sucrose equilibrium density gradient sedimentation experiments were performed as described ( Coulombe et al . , 2004 ) . Briefly , HEK293 cells grown in 10 cm plates were transiently transfected with Ptch1-FLAG or Δ2PY-FLAG along with Myc-Smurf2CG . 48 hr after transfection , the cells were lysed in pre-chilled 2 ml MES buffer , which contains 25 mM MES ( 2-[N-morpholino]ethanesulfonic acid ) , pH 6 . 5 , 150 mM NaCl , 1% Triton X-100 , supplemented with 1 × Roche cOmplete Protease Inhibitor Cocktail and was set on ice for 1 hr . The lysates were mixed with equal volume of 80% ( wt/vol ) sucrose/MES solution and placed at the bottom of an ultracentrifuge tube . Tube was then overlaid in consecutive order with 2 ml each of 30% , 25% , 20% , and 4 ml of a 5% ( wt/vol ) sucrose/MES buffer . After centrifugation at 39 , 000 rpm for 16 hr at 4°C in an SW 41 Ti rotor on Beckman Optima L-100 XP ultracentrifuge , the gradient was separated into twelve 1 ml fractions taken from the top for Western blot analysis . Total RNA was isolated from cultured cells using the RNAiso reagent ( TaKaRa , Shiga , Japan ) , and reverse transcription was carried out using the PrimeScript RT reagent Kit ( TaKaRa ) . Standard RT-PCR was carried out with the following primers: mouse Gli1 ( 5′-TCCAGCTTGGATGAAGGACCTTGT-3′ and 5′-AGCATATCTGGCACGGAGCATGTA-3′ ) , mouse Smurf1 ( 5′-CTACCAGCGTTTGGATCTAT-3′ and 5′-TTCATGATGTGGTGAAGCCG-3′ ) , mouse Smurf2 ( 5′-TAAGTCTTCAGTCCAGAGACC-3′ and 5′-AATCTCTTCCCTAGACACCTC-3′ ) , and mouse HPRT ( 5′-TATGGACAGGACTGAAAGAC-3′ and 5′-TAATCCAGCAGGTCAGCAAA-3′ ) . Real-time PCR was carried out using the FastStart SYBR Green Master mix ( Roche ) on a 7500 Real-Time PCR System ( Applied Biosystems , Grand island , NY ) with primers for mouse Gli1 ( 5′-GCTTGGATGAAGGACCTTGTG-3′ and 5′-GCTGATCCAGCCTAAGGTTCTC-3′ ) and mouse HPRT ( 5′-TATGGACAGGACTGAAAGAC-3′ and 5′-TAATCCAGCAGGTCAGCAAA-3′ ) . Experiments were repeated at least three times , and samples were analyzed in triplicate . Cerebellar slice cultures were prepared as described ( Kapfhammer , 2010 ) . Briefly , sagittal sections ( 350 µm ) were cut from cerebella of P7 Smurf1−/−;Smurf2fl/fl pups using a McIlwain tissue cutter under septic condition . Slices were transferred onto a permeable membrane ( Millicell-CM , Millipore-China , Beijing , China ) in a 6-well plate with 0 . 8 ml of culture medium ( Neurobasal A medium with B27 supplement ) and incubated at 37°C , 5% CO2 . For adenovirus infection , the viral stock ( 3 × 1010 pfu/ml ) was mixed with equal volume of type I collagen gel and applied as a drop on top of each slice , and 5 × 107 pfu of virus was also added in the culture medium . After 24 hr , the infected slices were washed and maintained in culture medium . The medium was changed every 2–3 days for a total of 12 days . Slices were then fixed in 4% paraformaldehyde overnight at 4°C and immunostained with anti-calbindin ( 1:500; Sigma ) and anti-NeuN ( 1:100; Millipore ) . Mouse cerebellar GCPs were isolated from 7-day-old pups according to a published protocol ( Hatten and Shelanski , 1988 ) . Briefly , cerebella were removed aseptically and incubated at 37°C for 5 min in trypsin/DNase buffer . Tissues were then triturated with fine Pasteur pipettes to obtain a single-cell suspension , overlaid on top of a step gradient of 35% and 65% Percoll ( Pharmacia , GE Health-China , Shanghai , China ) and centrifuged at 2 , 000×g for 10 min at 4°C . GCPs harvested from the 35% and 65% Percoll interface were further purified by depleting adherent cells with two consecutive 1-hr incubations in tissue culture dishes , then seeding them in Lab-Tek chambered slides coated with poly-D-lysine and Matrigel , and incubating them at 35°C , 5% CO2 . GCPs were transfected with siRNAs using FugeneHD Transfection Reagent ( Promega ) after 1 hr incubation . Proliferation of transfected GCPs was evaluated using Click-iT EdU cell proliferation assays ( Life Technologies ) at different time points after ShhN-CM or IGF1 ( 100 ng/ml ) treatment . GCPs were incubated with EdU ( 5-ethynyl-2′-deoxyuridine ) for 12 hr before fixation and permeabilization . EdU detection was performed according to the manufacturer's instruction . Images were acquired on a Leica inverted fluorescence microscope ( DMI 300B ) with a 20 × objective lens . Quantification of EdU-positive GCPs was performed using the ImageJ software . | Sonic hedgehog protein fulfils many vital roles in establishing the body plan of multicellular organisms during development . And in adult organisms it regulates the stem cells that maintain organs and tissues . In the embryo , Sonic hedgehog is secreted by certain cells to create a concentration gradient; cells then measure this concentration to work out where they are , which allows them to develop into the right sort of cells . However , many details of this process are not completely understood . At the core of this process are the interactions between the Sonic hedgehog protein , a receptor called Patched1 that is found on plasma membranes , and another membrane protein called Smoothened . The job of Smoothened is to activate proteins that enter the cell nucleus and ‘switch on’ the pathway's target genes , which encode Patched1 and a number of other proteins . The role of Patched1 , on the other hand , is to repress Smoothened . However , when sonic hedgehog binds to Patched1 , the latter is unable to repress Smoothened . Increasing the production of Patched1 is thought to serve two main roles: it prevents activation of the Sonic hedgehog pathway , and it prevents the Sonic hedgehog protein spreading to neighboring cells ( by binding to it ) . But how is the level of Patched1 itself regulated ? Yue et al . now report that two proteins , called Smurf1 and Smurf2 , perform this regulation role in mammalian cells . Smurf1 and Smurf2 are enzymes that attach a molecule called ubiquitin to proteins , setting in train a series of events that leads to the degradation of the protein . Yue et al . now show that Smurf1 and Smurf2 recognize a signal on Patched1 and perform a similar modification , causing the Patched1 to be internalized through an alternate pathway and degraded in lysosomes . This series of events ultimately allow the Sonic hedgehog pathway to be activated . The work of Yue et al . exposes a critical enzymatic step that sorts unbound Patched1 receptors from those that are bound to Sonic hedgehog proteins . Further research is needed to determine if this signaling pathway can be manipulated for therapeutic purposes . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology"
] | 2014 | Requirement of Smurf-mediated endocytosis of Patched1 in sonic hedgehog signal reception |
The roundworm C . elegans is a mainstay of aging research due to its short lifespan and easily manipulable genetics . Current , widely used methods for long-term measurement of C . elegans are limited by low throughput and the difficulty of performing longitudinal monitoring of aging phenotypes . Here we describe the WorMotel , a microfabricated device for long-term cultivation and automated longitudinal imaging of large numbers of C . elegans confined to individual wells . Using the WorMotel , we find that short-lived and long-lived strains exhibit patterns of behavioral decline that do not temporally scale between individuals or populations , but rather resemble the shortest and longest lived individuals in a wild type population . We also find that behavioral trajectories of worms subject to oxidative stress resemble trajectories observed during aging . Our method is a powerful and scalable tool for analysis of C . elegans behavior and aging .
Aging consists of gradual changes in an adult organism that cause a reduction of function and an increase in mortality . Studies of model organisms such as the roundworm C . elegans have identified highly conserved processes and pathways which influence aging , including dietary restriction ( Lakowski and Hekimi , 1998; Greer and Brunet , 2009 ) , insulin/insulin-like signaling ( Kenyon et al . , 1993 ) , and the cytoprotective DAF-16/FOXO pathway ( Greer and Brunet , 2007; Eijkelenboom and Burgering , 2013; Ogg et al . , 1997 ) . Current , widely used methods for long-term measurement of C . elegans aging are based on manual inspection of worm survival on agar plates . These methods are robust and technically simple , but have a number of limitations . They are labor intensive , low in throughput , and are largely focused on lifespan at the population level without access to information about health or behavior during individual life trajectories . Previous efforts have aimed to automate C . elegans survival assays . One method produces high resolution survival curves by monitoring large populations of animals on standard agar plates using flatbed scanners ( Stroustrup et al . , 2013 ) . However , this method monitors only lifespan and is not designed to track individual animals over their entire lifetime . Therefore , this system is limited in its ability to study aging in individual animals . Another method ( Zhang et al . , 2016 ) used small hydrogel compartments between a glass slide and a PDMS membrane to perform long-term longitudinal monitoring of C . elegans aging . However , the device has a number of limitations: it is not easily scalable to large numbers of animals and does not lend itself to screening experiments . The hydrogel device requires complex image analysis software and prevents access to animals during the experiment , limiting the additional phenotypes that can be assayed in tandem . Also , the hydrogel device requires sterile mutations to be crossed into all strains tested , and all experiments must be performed at the restrictive temperature of 25° C , limiting the prospects for genetic screening . Here , we describe a device we call the WorMotel ( WM ) , which is capable of longitudinally tracking behavior of up to 240 uniquely identified animals of any genotype per device for over 60 days . The WM consists of an array of individual wells , each of which is filled with standard agar media , bacterial food , and a single worm , enabling long-term cultivation and imaging of hundreds of uniquely identified animals . By conforming to the ANSI standard microplate format , our method leverages existing scalable automation technology including worm sorters , robotic plate handlers , and chemical library screening tools . We apply our method to quantifying inter-individual and inter-strain differences in behavioral decline during aging and stress , as well as in understanding the relationship between behavior and lifespan .
Conventional 96-well or 384-well microplates are not well suited for worm imaging and cultivation on agar media due to three problems . First , the vertical walls of each well make it difficult to image the worms when they are close to the edge of the wells ( Figure 1—figure supplement 1 ) . Second , worms tend to crawl between the agar and well edges , again making them difficult to image clearly . Third , under humid conditions , animals can climb over the walls of the wells , mixing with other worms . We designed the WorMotel to address these limitations . Each WorMotel consists of a transparent polydimethylsiloxane ( PDMS ) substrate containing a rectangular array of up to 240 wells , produced by molding from an acrylic photopolymer 3D-printed master ( Shepherd et al . , 2011 ) ( Figure 1 , Figure 1—figure supplement 2 ) . The well geometry is optimized for worm cultivation and imaging of a single worm per well . A rounded concave well geometry ( Figure 1a ) provides a clear view of the animal at all positions on the agar surface ( Yu et al . , 2014 ) , and also inhibits worms from burrowing under the agar ( Figure 1—figure supplement 3 ) . A network of narrow moats containing a copper sulfate solution surrounding the wells prevents animals from escaping from their wells ( Figure 1b ) . 10 . 7554/eLife . 26652 . 003Figure 1 . WorMotel design , fabrication , preparation , and experimental setup . ( a ) 3D rendering of WorMotel geometry . ( b ) Schematic cross-section of a single well . ( c ) Fabrication and loading process ( d ) Image of WorMotel filled with agar , bacteria , and adult C . elegans . ( e ) Experimental setup ( f ) Schematic of blue light stimulation system . ( g ) Representative image of 240-well WorMotel ( h ) High resolution image of nine wells , each housing a single young adult N2 worm . DOI: http://dx . doi . org/10 . 7554/eLife . 26652 . 00310 . 7554/eLife . 26652 . 004Figure 1—figure supplement 1 . Comparison of image quality with standard 384-well plate and WorMotel . ( a ) Image of L4 worms on agar in a standard 384-well plate . Scale bar: 2 mm . ( b ) Image of L4 worms on agar in a WorMotel . Individual animals are clearly visible in ( b ) but not in ( a ) . DOI: http://dx . doi . org/10 . 7554/eLife . 26652 . 00410 . 7554/eLife . 26652 . 005Figure 1—figure supplement 2 . 240-Well WorMotel ( a ) 240-well WorMotel PDMS insert . Scale Bar: 1 cm . ( b ) PDMS insert placed inside an OmniTray . ( c ) Computer rendering of the 240-well WorMotel negative master ( d ) Image of a WorMotel filled with agar , bacteria , and worms in an OmniTray . DOI: http://dx . doi . org/10 . 7554/eLife . 26652 . 00510 . 7554/eLife . 26652 . 006Figure 1—figure supplement 3 . WorMotel prevents burrowing . Percent of animals burrowing in the WorMotel ( red squares , n = 96 ) and in a standard 384-well microplate ( blue circles , n = 80 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 26652 . 00610 . 7554/eLife . 26652 . 007Figure 1—figure supplement 4 . Schematic of automated imaging system . An automated plate carousel ( Thermo ) containing eight plate stacks holds up to 240 WorMotel plates , each containing 240 C . elegans . A plate handling robot ( Thermo Orbitor , ThermoFisher Scientific , Philadelphia , PA ) moves plates onto one of 3 imaging/illumination systems , where they are serially imaged . The system is contained in a light-tight enclosure and temperature is controlled to within 0 . 2 C by a temperature controller . All functions are controlled by custom LabView software . DOI: http://dx . doi . org/10 . 7554/eLife . 26652 . 007 Each well is filled with approximately 15 µL of NGM agar , followed by a suspension of bacteria added as food . For lifespan measurements , enough bacteria is added initially to sustain worms throughout their lives , and the animals are left essentially undisturbed for the remainder of the experiment . A single worm is added manually to each well . The WorMotel is sealed inside a polystyrene dish ( Nunc Omnitray , ThermoFisher Scientific ) and imaged under dark field illumination with a CMOS camera ( Imaging Source , Charlotte , NC ) where it remains for the experiment’s duration ( Figure 1e ) ( Churgin and Fang-Yen , 2015 ) . If desired , however , the device may be removed periodically for manual inspection or other longitudinal assays . The pixel resolution for a field of view containing all 240 wells is 36 µm , or roughly one thirtieth the length of an adult C . elegans ( Figure 1g , Videos 1–3 ) . Moving the camera closer to the WM or using a longer focal length lens reduces the field of view but enables higher resolution images to be attained ( Figure 1h , Video 4 ) . 10 . 7554/eLife . 26652 . 008Video 1 . Young ( day 1 ) adults in the 240-well WorMotel . A 10 s long blue light exposure ( bright flash ) occurs at the 15 min mark . FOV: 95 mm x 63 mmDOI: http://dx . doi . org/10 . 7554/eLife . 26652 . 00810 . 7554/eLife . 26652 . 009Video 2 . Detail of 9 wells containing day 1 N2 adults . A 10 s long blue light exposure ( bright flash ) occurs at the 15 min mark . FOV: 14 mm x 14 mmDOI: http://dx . doi . org/10 . 7554/eLife . 26652 . 00910 . 7554/eLife . 26652 . 010Video 3 . Detail of 9 wells containing day 10 N2 adults . A 10 s long blue light exposure ( bright flash ) occurs at the 15 min mark . FOV: 14 mm x 14 mmDOI: http://dx . doi . org/10 . 7554/eLife . 26652 . 01010 . 7554/eLife . 26652 . 011Video 4 . High resolution video of 9 WorMotel wells containing adult animals ( day 1–4 ) . FOV: 14 mm x 14 mmDOI: http://dx . doi . org/10 . 7554/eLife . 26652 . 011 In humans and diverse model organisms , locomotor activity has been used as a measure of health ( Hausdorff et al . , 1997; Grotewiel et al . , 2005; Huang et al . , 2004 ) . It has been shown that spontaneous locomotion on food is a non-ideal measure of health since it assesses food preference in addition to locomotor ability ( Hahm et al . , 2015 ) . As such , a directed behavior is preferable over spontaneous movement . We used the response to blue light illumination as a measure of locomotor ability and therefore health . The WM can be imaged continuously or intermittently . Under intermittent imaging , the WM is automatically imaged at 0 . 2 frames per second for a 30 min interval twice daily . After fifteen minutes , a blue light stimulus , which evokes an escape response in worms ( Edwards et al . , 2008 ) , is applied to the entire plate for 10 s using a set of light emitting diodes ( LEDs ) . In this manner we measure both spontaneous and evoked behavioral responses ( Figures 1f and 2b–e , Videos 1–3 ) . 10 . 7554/eLife . 26652 . 012Figure 2 . Image processing , automated lifespan calculation , and manual validation . ( a ) Activity calculation . Delta image ( right image ) is calculated by pixel-by-pixel subtraction of images taken at time t1 from images taken at time t2 . Examples are shown in which a worm moves ( top ) and does not move ( bottom ) . ( b ) Activity , shown as a heat map , of young adult ( day one ) animals during a 30-min imaging epoch . Arrow indicates 10-s blue light stimulation . Color bar indicates the number of pixels changed over a 60-s interval . ( c ) Activity of adult C . elegans on day 10 . Colorbar uses same scale as panel ( b ) . ( d ) Activity trace from one wild-type animal . Blue light was applied at time zero . ( e ) Average wild-type population behavior on day 1 ( solid curve ) and day 10 ( dashed curve ) ( n = 30 ) . ( f ) Single animal trace of maximum spontaneous ( dashed ) or stimulated ( solid ) activity across entire lifespan . Arrow indicates time of death . ( g ) Lifespan of animals grown on the WorMotel ( solid curves ) or standard plates ( dashed curves ) . DOI: http://dx . doi . org/10 . 7554/eLife . 26652 . 012 Custom MATLAB software ( MathWorks , Natick , MA ) quantifies movement of the animal in each well . Following pixel-by-pixel subtraction of pairs of temporally adjacent frames ( Raizen et al . , 2008 ) , a simple measure of behavior can be defined as the number of pixels whose intensity changes between subsequent image frames ( Figure 2a ) . Highly mobile animals yield high activity values , whereas slowly moving , older , or quiescent animals yield low activity values ( Figure 2b–e ) . Activity analysis generated rich behavioral data capable of characterizing aging and development ( Figure 2f ) ( Nelson et al . , 2013 , Nelson et al . , 2014; Iannacone et al . , 2017 ) . Lifespan is determined as the final time of nonzero movement ( Figure 2f ) . During each imaging epoch we calculate the maximum activity before and after the blue light stimulus , and we term these values the spontaneous and stimulated locomotion , respectively . We recorded the total time spent moving after the stimulus , which we term response duration . Finally , we record the response latency , defined as the delay between the end of the stimulus and the beginning of the animal’s movement . Our WorMotel method is designed to be scalable to large numbers of animals . Using standard automation tools including a plate carousel and plate handling robot ( Figure 1—figure supplement 4 ) we have developed a system capable of intermittently imaging up to 240 plates containing 57 , 600 individually tracked worms . We asked to what extent lifespan results from the WM agreed with those using standard manual methods . We compared the survival curves of animals reared on standard agar plates to those in the WorMotel . We measured the lifespans of wild-type N2 alongside the short-lived strain daf-16 ( mu86 ) ( Ogg et al . , 1997 ) and the long-lived strain daf-2 ( e1370 ) ( Kenyon et al . , 1993 ) at 25° C ( Figure 2g ) . For worms grown on standard plates , lifespan assays were carried out using standard methods ( Kenyon et al . , 1993 ) . As expected , mean lifespan of daf-16 mutants ( WM: 7 . 7 ± 0 . 3 days , n = 61; Manual: 7 . 0 ± 0 . 2 days , n = 117 ) was shorter than that of N2 ( WM: 12 . 3 ± 0 . 3 days , n = 123; Manual: 12 . 2 ± 0 . 5 days , n = 94 ) while daf-2 mutants showed a longer lifespan ( WM: 33 . 5 ± 1 . 9 days , n = 46 , Manual: 30 . 8 ± 2 . 0 days , n = 52 ) . We found no significant difference between survival curves acquired from worms grown on standard plates and those grown on the WorMotel . Moreover , our lifespan results agree with those previously reported for each strain ( Kenyon et al . , 1993 ) . We asked whether the aversive copper sulfate solution in the moat , which helps to retain animals inside their respective wells , had any effect on survival or development . We compared the duration of the L4 stage , a measurement of developmental rate , and lifespan of worms grown on a WorMotel with moats filled with NGMB to a WorMotel with moats filled with 100 mM copper sulfate ( see Materials and methods ) . We found no difference in developmental rate of worms grown from the L3 stage to adulthood in the presence of an NGMB ( L4 Duration = 12 . 2 ± 0 . 2 hr ) and copper sulfate moat ( L4 Duration = 12 . 2 ± 0 . 2 hr , p=0 . 95 ) . Similarly , we found no difference in the mean lifespan of N2 worms grown in the presence of an NGMB ( 19 . 2 ± 0 . 7 days , n = 22 ) or copper sulfate moat ( 18 . 9 ± 1 . 0 days , n = 23 ) . To test the accuracy of our automated assessment of lifespan , we manually scored the lifespan of worms grown on a WorMotel that was also imaged by camera . Manual assessments of lifespan were performed daily within two hours of a thirty-minute imaging epoch . We compared the time of death measured by our software to that measured by a human observer scoring death manually by standard methods ( Kenyon et al . , 1993 ) . For N2 worms , the average difference between manual and automated lifespan measurement was 0 . 66 ± 0 . 6 days ( n = 79 ) , indicating that automated lifespan calculation is accurate within the time resolution of standard lifespan assays . Automated lifespan was always less than or equal to manual lifespan , indicating that dead animals were never incorrectly scored as alive . Furthermore , we found no correlation between the error in automatic lifespan measurement and true lifespan ( p=0 . 18 ) , indicating that the error in automated lifespan score is independent of lifespan itself , i . e . absolute measurement error does not increase for worms with longer lifespan . Together , these results show that with regard to development and lifespan , results from the WorMotel are similar to those using standard methods . Aging in C . elegans is accompanied by a deterioration of many behaviors , including slowing of locomotion ( Hahm et al . , 2015 ) and feeding ( Huang et al . , 2004 ) , and a reduced capacity for learning and memory ( Stein and Murphy , 2012 ) . While many genes have been identified that regulate aging , less is known about the effect of these genes on behavior . To survey the relationship between lifespan and behavior , we used the WM to analyze behavioral trajectories for wild type worms and seven loss-of-function mutants for genes known to influence lifespan and/or behavior: ( 1 ) daf-2 , which encodes the insulin/IGF receptor ( Kenyon et al . , 1993 ) , ( 2 ) age-1 , which encodes phosphoinositide-3-kinase , a component of the insulin/insulin-like signaling ( IIS ) pathway ( Friedman and Johnson , 1988 ) , ( 3 ) daf-16 , which encodes a transcription factor regulating a cytoprotective response ( Ogg et al . , 1997 ) , ( 4 ) tax-4 , which encodes a cyclic nucleotide-gated channel required for some sensory transduction ( Apfeld and Kenyon , 1999 ) , ( 5 ) unc-31 , required for neuropeptide processing ( Ailion et al . , 1999 ) , ( 6 ) lite-1 , a gene encoding a receptor required for normal aversive response to blue light ( Edwards et al . , 2008 ) , and ( 7 ) aak-2 , a gene encoding a subunit of AMP kinase ( Apfeld et al . , 2004 ) . We assayed the activity and survival of individuals within populations of each strain ( Figure 3a–f , Figure 3—figure supplement 1 ) . Mean lifespans of mutant strains relative to wild type agreed with those reported in previous studies ( Table 1 ) . The shapes of behavior curves varied dramatically between strains . 10 . 7554/eLife . 26652 . 013Figure 3 . Automated quantification of behavioral changes during aging . ( a ) Spontaneous behavior heat map for N2 ( n = 445 ) . Color bar indicates the number of pixels changed over a 60-s interval . ( b ) Stimulated behavior heat map for N2 . ( c ) Spontaneous behavior heat map for daf-16 ( n = 234 ) . ( d ) Stimulated behavior heat map for daf-16 . ( e ) Spontaneous behavior heat map for daf-2 ( n = 90 ) . ( f ) Stimulated behavior heat map for daf-2 . ( g ) Survivor population spontaneous activity . ( h ) Survivor population stimulated activity . ( i ) Survivor population response duration . ( j ) Survivor population response latency . DOI: http://dx . doi . org/10 . 7554/eLife . 26652 . 01310 . 7554/eLife . 26652 . 014Figure 3—source data 1 . Includes data for lifespan , spontaneous locomotion , stimulated locomotion , response duration , and response latency for each strain shown . DOI: http://dx . doi . org/10 . 7554/eLife . 26652 . 01410 . 7554/eLife . 26652 . 015Figure 3—figure supplement 1 . Additional behavior heat maps . ( a ) Spontaneous behavior heat map for age-1 mutants . ( b ) Stimulated behavior heat map for age-1 mutants . ( c ) Survivor spontaneous ( dashed curve ) and stimulated ( solid curve ) activity for age-1 mutants . ( d ) Survivor response duration ( dashed curve ) and latency ( solid curve ) for age-1 mutants . ( e–h ) Same data as panels ( a–d ) shown for tax-4 mutants . ( i–l ) Same data as panels ( a–d ) shown for unc-31 mutants . ( m–p ) Same data as panels ( a–d ) shown for lite-1 mutants . ( q–t ) Same data as panels ( a–d ) shown for aak-2 mutants . DOI: http://dx . doi . org/10 . 7554/eLife . 26652 . 01510 . 7554/eLife . 26652 . 016Table 1 . Summary of lifespan dataDOI: http://dx . doi . org/10 . 7554/eLife . 26652 . 016StrainFood sourceLifespan ( Mean ± SD ) ( Days ) NP-value relative to N2 controlN2DA83716 . 8 ± 4 . 4445N/Adaf-16DA83712 . 4 ± 2 . 62343 . 0 x 10-45daf-2DA83728 . 5 ± 9 . 2902 . 1 x 10-16age-1DA83723 . 7 ± 8 . 91194 . 3 x 10-9tax-4DA83722 . 0 ± 5 . 4723 . 6 x 10-6unc-31DA83721 . 2 ± 7 . 21203 . 6 x 10-4lite-1DA83719 . 7 ± 4 . 0900 . 036aak-2DA8379 . 6 ± 1 . 81173 . 9 x 10-30N2HT115 ( EV RNAi ) 20 . 9 ± 10 . 580N/AN2HT115 ( daf-2 RNAi ) 29 . 9 ± 10 . 3381 . 2x 10-6N2HT115 ( odr-10 RNAi ) 21 . 5 ± 6 . 5400 . 53daf-2HT115 ( EV RNAi ) 30 . 0 ± 10 . 5401 . 6 x 10-6daf-2HT115 ( daf-2 RNAi ) 34 . 5 ± 9 . 5397 . 9 x 10-11daf-2HT115 ( odr-10 RNAi ) 36 . 9 ± 7 . 4398 . 2 x 10-15odr-10HT115 ( EV RNAi ) 21 . 5 ± 6 . 6400 . 40odr-10HT115 ( daf-2 RNAi ) 28 . 3 ± 12 . 5384 . 8 x 10-4odr-10HT115 ( odr-10 RNAi ) 21 . 1 ± 7 . 5400 . 49 We included lite-1 worms in order to determine whether these mutants , previously shown to have a reduced response to blue light ( Edwards et al . , 2008; Ward et al . , 2008 ) , can be assayed by our blue light illumination system for measuring stimulated activity . To our surprise , we found little difference in the responses to blue light between N2 and lite-1 mutants , possibly because our light stimulus is higher in irradiance than that previously used , and/or our activity measurement is more sensitive to movement increases than the previously measured body bend frequency ( Edwards et al . , 2008 ) . Regardless , our results for lite-1 worms indicate that even mutants with deficits in blue light response can be assessed by our method . Together , these results show that the WorMotel method is compatible with arbitrary strains . We used the longitudinal data generated by the WorMotel to address questions about the relationship between aging and behavior . While many factors are known to modulate the mean lifespan of a population , less is known about how these factors alter the aging process on an individual level . Zhang et al . recently showed that within a wild-type population , long-lived and short-lived animals differed in two ways ( Zhang et al . , 2016 ) . First , the rate of physiological decline was slower in long-lived individuals , as might be expected . The second , however , was counter-intuitive: the additional lifespan of longer-lived individuals was primarily due to differences toward the end of the lifespan . That is , long-lived animals exhibited longer periods of low physiological function , or ‘extended twilight’ ( Zhang et al . , 2016 ) . A different picture was suggested by a study using automated assays of lifespan in the ‘Lifespan machine’ ( Stroustrup et al . , 2016 ) . In this study it was reported that various genetic and environmental perturbations do not fundamentally change the shape of the survival curve , but rather only compress or dilate it in time . This result was interpreted as suggesting that the aging process in C . elegans is , at least at some point in its pathway , controlled by a single process describable by a single variable corresponding to the rate of aging ( Stroustrup et al . , 2016 ) . We sought to determine to what extent ‘extended twilight’ and/or scaling effects apply at the behavioral level in mutants with altered aging . The concept of a universal scaling parameter in aging would suggest that the short and long-lived individuals within any strain ( whether with normal , short , or long mean lifespan ) would resemble their short and long-lived counterparts in the reference strain , but with a temporal scaling ( Figure 4a ) . If the variations in aging rate among individuals in any isogenic strain are governed by similar factors , we would expect that short and long-lived individuals would display similar late-life characteristics as their wild type counterparts . If , on the other hand , short-lived strains as a whole physiologically more closely resemble short-lived individuals of a wild type population , we might expect them to display late-life characteristics similar to these short-lived individuals ( Figure 4b ) . Similarly , long-lived strains might display a range of late-life decays or alternatively collectively resemble long-lived worms in the reference strain . 10 . 7554/eLife . 26652 . 017Figure 4 . Potential aging models . ( a ) Model 1: Temporal scaling results in identical patterns of behavioral decline when data are normalized by lifespan . Idealized decline curves for wild type ( black dot-dashed ) , short-lived ( red solid ) , and long-lived ( green dashed ) strains . Decline curves are shown as a function of chronological time ( left ) and fraction of life ( right ) . ( b ) Model 2: Long-lived and short-lived strains resemble long-lived and short-lived wild type worms with respect to behavioral decline . Idealized decline curves for a short-lived ( red solid ) and a long-lived ( green dashed ) strain . Decline curves are shown as a function of chronological time ( left ) and fraction of life ( right ) . DOI: http://dx . doi . org/10 . 7554/eLife . 26652 . 017 Wild-type strain N2 worms exhibited an initial decline followed by a ‘plateau’ period of nearly constant spontaneous and stimulated activity and response duration and latency ( Figure 3g–j ) . When we compared the behavior of the shortest-lived and longest-lived quartile of N2 worms , we found that their behavioral declines were qualitatively different . The longest-lived animals exhibited a ‘decline and plateau’ phenotype , in which an initial rapid decline in behavioral capacity is later replaced by a very gradual decline for the remainder of life ( Figure 5a , h ) . By contrast , the shortest-lived animals showed only the rapid decline in behavior before dying ( Figure 5a , g ) . The result that long-lived animals experience a long period of low behavior are consistent with the ‘extended twilight’ reported by Zhang et al . ( Zhang et al . , 2016 ) . 10 . 7554/eLife . 26652 . 018Figure 5 . Mutants with short and long lifespan display patterns of late-life behavioral decline that resemble short and long-lived worms from a wild type population . ( a ) N2 behavior over time for the lowest quartile ( solid curve ) and highest quartile ( dashed curve ) of survivors . ( b ) Data from panel ( a ) plotted as a fraction of each individual’s life . ( c ) daf-16 behavior over time . ( d ) daf-16 behavior over fraction of life . ( e ) daf-2 behavior over time . ( f ) daf-2 behavior over fraction of life . ( g ) Comparison of N2 lowest survivor quartile and daf-16 . ( h ) Comparison of N2 highest survivor quartile with daf-2 . DOI: http://dx . doi . org/10 . 7554/eLife . 26652 . 01810 . 7554/eLife . 26652 . 019Figure 5—figure supplement 1 . Stimulated activity versus chronological time and fraction of life for long-lived individuals and short-Lived individuals of the same strain . ( a ) age-1 mutant behavior as a function of chronological time for the lowest ( solid curve ) and highest ( dashed curve ) survivor quartiles . ( b ) tax-4 mutant behavior as a function of chronological time for the lowest ( solid curve ) and highest ( dashed curve ) survivor quartiles . ( c ) unc-31 mutant behavior as a function of chronological time for the lowest ( solid curve ) and highest ( dashed curve ) survivor quartiles . ( d ) lite-1 mutant behavior as a function of chronological time for the lowest ( solid curve ) and highest ( dashed curve ) survivor quartiles . ( e ) aak-2 mutant behavior as a function of chronological time for the lowest ( solid curve ) and highest ( dashed curve ) survivor quartiles . ( f–j ) Same data as present in panels ( a–e ) presented as a function of life fraction . DOI: http://dx . doi . org/10 . 7554/eLife . 26652 . 019 Short-lived daf-16 mutants declined at a similar rate to N2 , but did not exhibit any plateau phase; instead , daf-16 worms die after their initial behavioral decline ( Figure 3g–j , Figure 5c , d ) . A similar effect was seen in daf-16 response duration and response latency , which do not level off but decrease or increase , respectively , at a similar rate until the time of death . Comparing the activity history of the shortest-lived N2 worms to that of daf-16 as a whole , we found a striking correspondence between the behavioral decline of the two groups ( Figure 5g ) . These results show that the behavioral decline of daf-16 animals is not a scaled version of the wild type distribution of decline , but instead resembles the short-lived individuals in a wild-type population . Long-lived daf-2 ( e1370 ) mutants , in which behavioral quiescence has been previously reported ( Gems et al . , 1998; Gaglia and Kenyon , 2009 ) , exhibited a decline in stimulated activity akin to that observed in N2 and daf-16 followed by a nearly constant low level of stimulated activity and response behaviors for the remainder of life ( Figure 3h ) . Spontaneous activity in daf-2 , on the other hand , declined to near zero within 10 days of adulthood , where it remained until death . Even at very young chronological age ( before day 5 ) , daf-2 mutants perform less well than N2 for each behavior metric scored ( Figure 3g–j ) . The ‘decline and plateau’ phenotype of the longest-lived N2 animals was also evident in both short-lived and long-lived daf-2 animals ( Figure 5e , f ) . Long-lived strains age-1 , tax-4 , and unc-31 also exhibited the ‘decline and plateau’ phenotype ( Figure 3—figure supplement 1 , Figure 5—figure supplement 1 ) . These results show that aging behavior of daf-2 and other long-lived animals , like that of daf-16 animals , does not resemble a scaled version of wild type . Instead , they resemble the longest-lived individuals in a wild-type population , in that they exhibit a long plateau period of low locomotory function during late life . In order to further characterize inter-individual differences in aging , we next sought to quantify the shape of behavioral decline . We analyzed individual behavioral decline as a fraction of life and calculated early and late decline rates ( Figure 5b , d , f , Figure 5—figure supplement 1 , Figure 6a–c ) ( see Materials and methods ) . We then calculated the difference in decline rates to quantify the overall shape of behavioral decline . We found that the change in decline rate negatively correlated with lifespan in all strains tested ( N2: R = −0 . 32 , p=8 . 0×10−12; daf-16: R = −0 . 18 , p=0 . 0054; daf-2: R = −0 . 25 , p=0 . 016; age-1: R = −0 . 43 , p=8 . 0×10−7; tax-4: R = −0 . 41 , p=4 . 2×10−4; unc-31: R = −0 . 49 , p=1 . 6×10−8; lite-1: R = −0 . 54 , p=3 . 9×10−8 , aak-2: R = −0 . 23 , p=0 . 012 ) ( Figure 6a–e ) , indicating that the shape of behavioral decline differed signficantly for individuals with differing lifespans . We observed a smooth transition between the shape of aging behavior between short-lived and long-lived individuals within each strain ( Figure 6a–c , Figure 6—figure supplement 1 ) . Furthermore , our results suggest that there exists a relationship between change in decline rate and lifespan that lies along a continuum across strains in addition to between individuals of the same strain ( Figure 6e , f ) . Therefore , our results suggest that while behavioral decline does not temporally scale with lifespan , the stochastic sources of variability between isogenic individuals modulate the shape of aging along the same axis of variability as between short and long lived strains . For example , variability in the rate of aging may reflect a variability in the nuclear localization of DAF-16 and the activation of its targets . 10 . 7554/eLife . 26652 . 020Figure 6 . Shape of behavioral decline changes continuously with lifespan across individuals and strains . ( a ) Change in decline rate ( pixels/life fraction ) versus lifespan for individual N2 animals . ( b ) Same data as in panel ( a ) presented for daf-16 mutants . ( c ) Same data as in panel ( a ) presented for daf-2 mutants . ( d ) Change in decline rate ( pixels/life fraction ) in lowest ( white ) to highest ( dark gray ) survivor quartiles . * , p<0 . 05; ** , p<0 . 01; *** , p<0 . 001 . ( e ) Change in decline rate ( pixels/life fraction ) versus lifespan for multiple strains . ( f ) Mean decline rate change ( pixels/life fraction ) plotted against mean lifespan for each strain tested . Correlation coefficient r = −0 . 94 , p=0 . 0006 . ( g ) Standard deviation of decline rate change ( pixels/life fraction ) plotted against mean lifespan for each strain tested . Correlation coefficient r = −0 . 70 , p=0 . 055 . DOI: http://dx . doi . org/10 . 7554/eLife . 26652 . 02010 . 7554/eLife . 26652 . 021Figure 6—figure supplement 1 . Change in decline rate versus lifespan in individual animals . ( a ) Change in decline rate versus lifespan for individuals of all strains tested . ( b ) Change in decline rate versus lifespan for individual age-1 mutants . ( c ) Same data as in panel ( b ) presented for tax-4 mutants . ( d ) Same data as in panel ( b ) presented for unc-31 mutants . ( e ) Same data as in panel ( b ) presented for lite-1 mutants . ( f ) Same data as in panel ( b ) presented for aak-2 mutants . DOI: http://dx . doi . org/10 . 7554/eLife . 26652 . 021 Finally , we investigated the level of inter-individual variability in the rate of aging . We found that the standard deviation of decline rate change generally decreased with average lifespan ( Figure 6g ) ( R = − . 70 , p=0 . 055 ) . That is , longer-lived strains exhibited less individual variability than shorter-lived strains . Under a temporal scaling model , both the mean decline rate change and standard deviation of decline rate change would be equal across strains with different lifespans . Therefore , our data argue against a temporal scaling model of aging . In an analysis of scaling in lifespan , Stroustrup et al . showed that the shape of population survival curves during thermal stress was virtually identical to that occurring during aging ( Stroustrup et al . , 2016 ) . That is , the rate of population decline increases with temperature while still obeying the same fundamental kinetics . We have shown that the shape of behavioral decline is not identical for worms with different lifespans within a population . Nevertheless , we asked whether a similar scaling law holds for behavioral decline in populations during acute stress . To test this idea , we added paraquat , which induces oxidative stress via generation of reactive oxygen species ( ROS ) , to the WorMotel agar and monitored the animals’ subsequent behavior and survival ( An et al . , 2003 ) . Oxidative stress , like thermal stress , greatly shortens lifespan . We found that when we added paraquat to a final concentration of 40 mM on day 1 of adulthood , wild type animals survived for 21 . 2 ± 8 . 9 hr ( n = 58 ) , consistent with previous results ( Figure 7a , Table 2 ) ( An et al . , 2003 ) . We compared the shape of decline for animals experiencing stress to animals experiencing normal aging ( Figure 7b , d , f , h ) . We calculated the normalized mean square difference between behavior during aging and stress , and , after correcting for the average activity offset , found that the difference was 4 . 1% for N2 , 9 . 7% for daf-16 , 11 . 0% for daf-2 , and 13 . 2% for age-1 . 10 . 7554/eLife . 26652 . 022Figure 7 . Behavioral decline during acute oxidative stress resembles behavioral decline during aging . ( a , c , e , g ) Behavior and survival heat maps for N2 ( n = 58 ) , daf-16 ( n = 60 ) , daf-2 ( n = 68 ) , and age-1 ( n = 68 ) . ( b , d , f , h ) Comparison of behavioral decline during normal aging ( solid curve ) , stress with paraquat added on day 1 of adulthood ( dashed curve ) , and stress with paraquat added on day 9 of adulthood ( solid curve with circles ) ( i ) Change in decline rate ( pixels/life fraction ) for individuals versus survival on paraquat . DOI: http://dx . doi . org/10 . 7554/eLife . 26652 . 02210 . 7554/eLife . 26652 . 023Figure 7—source data 1 . Includes activity and paraquat survival data for each strain shown . DOI: http://dx . doi . org/10 . 7554/eLife . 26652 . 02310 . 7554/eLife . 26652 . 024Table 2 . Summary of paraquat assay survival data . DOI: http://dx . doi . org/10 . 7554/eLife . 26652 . 024StrainFood sourceLifespan ( Mean ± SD ) ( Days ) NP-value relative to N2 controlN2DA837 + 40 mM paraquat21 . 2 ± 8 . 958N/Adaf-16DA837 + 40 mM paraquat16 . 5 ± 12 . 0600 . 001daf-2DA837 + 40 mM paraquat37 . 4 ± 22 . 0682 . 0 x 10-5age-1DA837 + 40 mM paraquat29 . 7 ± 13 . 8682 . 7 x 10-5 If it is true that behavioral decline during stress beginning at day 1 of adulthood resembles a temporally scaled recapitulation of behavioral decline during normal aging , we reasoned that starting oxidative stress at mid-life should truncate the initial portion of the behavioral decline curve , as the slow decline of early aging should have already occurred naturally . To test this idea , we grew animals on the WorMotel as we did for normal aging experiments . We then added paraquat on day 9 of adulthood , and monitored worms’ subsequent behavior and survival . We observed that the initial decline occurred almost immediately , indicating that , as expected , the initial portion of the aging curve was no longer present in the behavioral stress decline curve ( Figure 7b , d , f , h ) . These results suggest that population-level rate of behavioral decline indeed temporally scales with increased stress . Previously , we observed that during normal aging , the change in decline rate , a measure of the shape of functional decline , negatively correlated with lifespan . Since population behavioral decline seemed to scale between normal aging and stress , we therefore investigated the relationship between individual aging and survival during stress . We once again found a negative correlation between the change in decline rate and survival for animals grown on 40 mM paraquat ( Figure 7i ) . These results show that the shape of behavioral decline during severe oxidative stress was similar to that during and aging , despite the process of aging on oxidative stress occurring about 20 times faster . This result suggests that there exist strong parallels in the worm’s behavioral responses to oxidative stress and to aging . We observed that long-lived daf-2 mutants exhibited greatly reduced locomotion amplitude and movement duration and elevated response latency to aversive blue light ( Figure 3g–j ) . Previous studies have observed similar phenotypes , such as the high degree of dauer-like quiescence in daf-2 adults ( Gems et al . , 1998; Gaglia and Kenyon , 2009 ) . In addition to having increased longevity , daf-2 mutants have been shown to possess greater fat stores ( Ogg et al . , 1997 ) . Reduction of insulin signaling , higher fat stores , and reduced movement are all features of hibernation in mammals , and it has been proposed that the daf-2 mutation confers a constitutive ‘hibernation-like’ phenotype on these animals ( Carey et al . , 2003; Gaglia and Kenyon , 2009 ) . Since animals deprived of fat stores or forced to move during hibernation have reduced survival ( Reeder et al . , 2012 ) , we hypothesized that reduced locomotor behavior might be required for increased lifespan in daf-2 animals . A recent study identified ODR-10 , a G-protein coupled olfactory receptor sensitive to diacetyl , as required for reduced locomotion in daf-2 mutants ( Hahm et al . , 2015 ) . ODR-10 mRNA levels are elevated in daf-2 mutants , and daf-2 mutants on odr-10 RNAi show a greater maximum velocity than controls . In an effort to determine if reduced locomotion was required for increased lifespan , we tested daf-2 mutants with odr-10 RNAi with the WorMotel . We grew N2 , daf-2 ( e1370 ) , and odr-10 ( ky32 ) mutants on Empty Vector ( EV ) , daf-2 , or odr-10 RNAi on the WorMotel to monitor behavior and lifespan . To our surprise , we found no significant difference in spontaneous activity between either N2 or daf-2 worms grown on EV versus odr-10 RNAi ( Figure 8a ) . These results suggest that ODR-10 does not in fact influence the reduced spontaneous movement observed in daf-2 mutants . 10 . 7554/eLife . 26652 . 025Figure 8 . Reduced sensory response is not required for extended longevity in daf-2 mutants . ( a ) Spontaneous activity during days 1–10 and 10–20 of adulthood for N2 grown on Empty Vector RNAi ( n = 80 ) , N2 grown on odr-10 RNAi ( n = 40 ) , daf-2 grown on Empty Vector RNAi ( n = 40 ) , and daf-2 grown on odr-10 RNAi ( n = 39 ) . *p<0 . 05; **p<0 . 01; ***p<0 . 001 . ( b ) Stimulated activity for the same individuals shown in panel ( a ) . ( c ) Response duration for the same individuals shown in panel ( a ) . ( d ) Response latency for the same individuals shown in panel ( a ) . ( e ) Spontaneous activity during days 1–10 and 10–20 of adulthood for N2 grown on Empty Vector RNAi ( n = 80 ) , N2 grown on daf-2 RNAi ( n = 38 ) , odr-10 ( ky32 ) mutants grown on Empty Vector RNAi ( n = 40 ) , and odr-10 ( ky32 ) mutants grown on daf-2 RNAi ( n = 38 ) . *p<0 . 05; **p<0 . 01; ***p<0 . 001 . ( f ) Stimulated activity for the same individuals shown in panel ( e ) . ( g ) Response duration for the same individuals shown in panel ( e ) . ( h ) Response latency for the same individuals shown in panel ( e ) . ( i–l ) Survival curves . DOI: http://dx . doi . org/10 . 7554/eLife . 26652 . 02510 . 7554/eLife . 26652 . 026Figure 8—source data 1 . Includes data for lifespan , spontaneous locomotion , stimulated locomotion , response duration , and response latency for each strain shown . DOI: http://dx . doi . org/10 . 7554/eLife . 26652 . 026 We examined stimulated activity in the same experiments . While we found no difference in early life stimulated activity between N2 or daf-2 mutants grown on EV versus odr-10 RNAi , we did find that late life stimulated activity was elevated in daf-2 mutants grown on odr-10 RNAi , whereas late life stimulated activity was unchanged for N2 grown on odr-10 RNAi ( Figure 8a , b ) . Furthermore , while N2 treated with daf-2 RNAi exhibited significantly decreased stimulated activity compared to animals grown on EV RNAi , odr-10 mutants grown on daf-2 RNAi did not ( Figure 8f ) . Similar results were observed for response duration and latency ( Figure 8c , d , g , h ) . These results suggest that ODR-10 does not directly affect locomotion in daf-2 mutants , but reduces sensitivity of daf-2 mutants to stimuli . Loss of function mutations in a number of genes required for sensory responses , such as such as the cyclic nucleotide channel TAX-4 and the intraflagellar transport particle homologue OSM-6 , have been shown to exhibit extended longevity ( Apfeld and Kenyon , 1999 ) . Because daf-2 mutants exhibit both reduced locomotion in addition to prolonged lifespan , it is possible that the reduced sensory responsiveness of daf-2 mutants is a requirement for their extended longevity . To test this idea , we compared survival curves for worms in which either daf-2 , odr-10 , or both either had loss of function mutations or were knocked down by RNAi . We found that odr-10 mutation and odr-10 RNAi had no effect on wild type lifespan ( Figure 8i , k ) , whereas daf-2 RNAi increased lifespan in both wild type and odr-10 mutants to an equal degree ( Figure 8j , l ) . These results show that while ODR-10 is required for reduced sensory responses in daf-2 mutants , it is not required for extended lifespan .
We have used our WorMotel method to investigate inter-individual and inter-strain variability in behavioral decline and the relationship between behavior and lifespan . Large-scale automated analysis of lifespan and behavior will facilitate screening for genetic and chemical modulators of aging . This system’s ability to longitudinally monitor animals throughput their lifespans may also help identify mechanisms of variability of aging between individuals in a population . In addition to these applications in aging , we have used the WorMotel in studies of other behaviors , such as developmentally-timed quiescence ( lethargus ) ( Nelson et al . , 2013 ) , stress-induced quiescence ( Iannacone et al . , 2017; Nelson et al . , 2014 ) , and adult behavior states ( McCloskey et al . , 2017 ) . Therefore the WorMotel is a flexible tool for assaying C . elegans long-term behavior . We found that short-lived wild type individuals declined in a manner similar to the short-lived strain daf-16 and that long-lived wild type individuals declined in a manner similar to the long-lived strain daf-2 . Furthermore , we found that the shape of decline followed a single curve with respect to lifespan ( Figure 6e ) . These results suggest that the sources of variability in lifespan in individuals also impact functional decline in a corresponding manner . For example , the N2 worm that survives 15 days due to stochastic factors will decline in a similar manner to the daf-16 worm that survives 15 days . Furthemore , individuals with a 30-day lifespan will exhibit a different shape of functional decline , but this shape is dictated by the confluence of genetic and stochastic factors that result in the lifespan of 30 days . One explanation for the extended longevity of insulin signaling mutants such as daf-2 and age-1 is via their shared transcriptional profile with dauer larvae , which can persist in harsh environments by virtue of an upregulation in stress-response and detoxification pathways ( McElwee et al . , 2004 ) . Furthermore , other work has shown that at an advanced age , wild type transcriptional profiles also exhibit similarities with that of dauer larvae ( Lund et al . , 2002 ) . Our finding that behavioral decline adheres to a continuum suggests that long-lived wild type worms may be physiologically and transcriptionally similar to worms with mutations in the insulin signaling pathway . Future experiments comparing gene expression in rapidly or slowly aging worms may elucidate how aging variability is manifest at the molecular level . In addition to this and other studies ( Zhang et al . , 2016 ) , a recent study ( Podshivalova et al . , 2017 ) also observed an extension of late-life behavioral quiescence in N2 and daf-2 mutants . The authors found that intestinal bacterial colonization is a risk factor for death in C . elegans and that daf-2 mutants , which exhibit a greater fraction of late-life decrepitude compared to N2 , were less susceptible to this bacterial colonization . When the authors fed worms killed bacteria , they found a greater lifespan extension in N2 ( 40% ) than daf-2 ( e1368 ) ( 16% ) , suggesting that bacterial colonization is a cause of premature death in N2 worms . Finally , the authors found that feeding worms dead bacteria specifically extended the period of infirmity rather than that of good health , suggesting that bacterial colonization may be a primary cause of lifespan truncation in short-lived individuals . That is , bacterial colonization may cause a reduction in late-life decrepitude in short-lived worms by causing premature death . If bacterial colonization is a fundamental cause of the smaller fraction of late-life decrepitude observed in short-lived worms , the question remains as to why long-lived worms exhibit extended behavioral quiescence in old age . One possibility is that muscle integrity degrades with age faster than other tissues ( Herndon et al . , 2002 ) such that older worms are physically able to move less well than reflected by their probability of dying . Another possibility might be related to the reduction in feeding worms exhibit with age ( Huang et al . , 2004 ) . Feeding and locomotion are linked ( McCloskey et al . , 2017 ) , so it might be that worms that have ceased feeding also tend to cease spontaneous locomotion . It has been hypothesized that normal aging constitutes a low-level stress that results in the slow accumulation of damage leading to senescent decline . While the accelerated aging observed during oxidative stress in our experiments is far greater than what occurs during normal physiological processes , our result that the shape of functional decline is similar during normal aging and acute oxidative stress suggests a potential underlying similarity between these two conditions . Furthermore , we show that the relationship between individual decline and survival is conserved between oxidative stress and normal aging , indicating that the processes governing the unique shape of decline for short-lived and long-lived individuals are preserved . Our results indicate that the process of functional decline can be sped up by at least a factor of twenty while still maintaining a similar average shape . Together , our results suggest that normal aging and acute oxidative stress similarly impact the process of functional decline . Future work will aim to define mechanisms for this similarity . If functional decline were dictated only by lifespan , we expect to observe a single curve relating the shape of decline to lifespan , regardless of environmental conditions . That is , interventions that drastically shorten lifespan , such as the addition of paraquat , should all exhibit an increase in decline rate between early and late life to be continuous with the curve presented in Figure 6e . Instead , however , we observe a translation of the tradeoff we observe during normal aging between the shape of functional decline and lifespan . This indicates that the shape of decline is not dictated by lifespan per se , but instead by the distance of an individual's lifespan relative to some standard in a given set of environmental conditions . For example , during normal aging , a lifespan of about 12 days results in neutral decline , or no change in decline rate between early and late life ( Figure 6a , d , e ) , whereas during oxidative stress , a lifespan of about 12 hr results in neutral decline ( Figure 7i ) . Together , these results suggest that relative to a standard lifespan in a given environment , there exists a defined shape of aging in the individuals whose lifespans differ from that standard as a result of genetic and/or stochastic factors . We found that ODR-10 did not affect baseline locomotion of daf-2 mutants , but did reduce the locomotory response of daf-2 worms to an aversive stimulus . Furthermore , ODR-10 knockdown did not reduce daf-2 lifespan , suggesting that elevated sensory response threshold is not required for increased lifespan of daf-2 animals . We found that ODR-10 suppressed locomotion in daf-2 mutants only in response to stimulation , whereas Hahm et al . ( Hahm et al . , 2015 ) found that ODR-10 suppressed locomotion per se in daf-2 mutants , One simple explanation for this discrepancy is that in Hahm et al . , locomotion assays were conducted soon after worms were manually stimulated due to picking onto the assay plate . The WorMotel allowed us to monitor worm behavior in a long-term unstimulated state in addition to after blue light stimulation . Future experiments may shed further light on whether reduced movement is required for the extended longevity of daf-2 . The WorMotel will be useful in quantifying behaviors that unfold over long periods of time and further exploring the relationships between behavior and lifespan . It has been reported that survival curves scale by a multiplicative constant across a diverse set of genetic and environmental perturbations ( Zhang et al . , 2016 ) . This result was interpreted as being compatible with the sum of biological inputs being filtered through a single state variable that determines the rate of aging . Our results indicate that a single variable is unlikely to be able to account for the inter-individual variability observed in aging . We observe variability in the shape of aging between individuals ( Figures 5a–f and 6a–c ) , indicating that the shape of aging does not scale for individuals with different lifespans . At the same time , we do observe a scaling of the relationship between the shape of aging and lifespan when the mean population rate of aging is accelerated with oxidative stress . Furthermore , we observe a difference in variability in aging decline across genotypes with differences in lifespan ( Figure 6g ) . Therefore , our results indicate the likely existence of at least one additional variable across which the process of aging may vary between individuals within and across populations . Future work may uncover the full space across which the aging process may vary and mechanisms underlying variability in aging .
The following strains were used in this study: N2 , CF1038: daf-16 ( mu86 ) I , CB1370: daf-2 ( e1370 ) III , TJ1052: age-1 ( hx546 ) II , PR678: tax-4 ( p678 ) III , DA509: unc-31 ( e928 ) IV , KG1180: lite-1 ( ce314 ) X , RB754: aak-2 ( ok524 ) X , CX32: odr-10 ( ky32 ) X . All strains were maintained at 15°C under standard conditions ( Stiernagle , 2006 ) . All experiments were carried out at 20°C unless otherwise stated . To fabricate the WorMotel , we developed a 3D-printing based molding method ( Shepherd et al . , 2011 ) . We designed a chip containing a rectangular array of either 48 or 240 rounded wells with 3 mm diameter , 3 mm depth , and center-to-center spacing of 4 . 5 mm ( Figure 1 ) . Each well was surrounded by a 0 . 5 mm wide and 3 mm deep channel , which would serve as the moat . Designs of the WorMotel masters were created using MATLAB . We printed a master corresponding to the negative of this shape with an Objet30 photopolymer 3D printer using the material VeroBlack . To mold the WM devices , we mixed Dow Corning Sylgard 184 PDMS according to the manufacturer’s instructions and poured 35 g or 5 g of PDMS into the 240-well or 48-well masters , respectively . We then degassed the poured PDMS in a vacuum chamber for 1 hr or until no more bubbles were visible . Devices were cured overnight at 40°C and then removed from molds using a spatula . To prepare devices for experiments , the chips were first treated with oxygen plasma for 4 min using a plasma cleaner ( PE-50 , Plasma Etch Inc . , Carson City , NV or Plasmatic Systems Plasma Preen II , Plasmatic Systems . Inc . , North Brunswick , NJ ) . This treatment renders PDMS temporarily hydrophilic , which greatly facilitates the filling of wells and moats . The medium was based on standard NGM media ( Stiernagle , 2006 ) except low-gelling temperature agarose ( gelling temp . 26–30°C , Research Products International , Mount Prospect , IL ) was substituted for agar to minimize solidification of the agar during filling of the wells , and streptomycin ( 200 ng/mL ) was added to the media to minimize bacterial contamination . For lifespan experiments , but not for development experiments , we added 5-fluoro-2’-deoxyuridine ( FUdR ) to prevent growth of progeny . A frozen FUdR stock solution of 10 mg/ml in water was thawed and added to molten agar at 40°C at a concentration of 5 μL per mL just prior to filling . This yielded a final FUdR concentration of 200 μM . The moat solution consisted of 100 mM copper sulfate , which was approximately in osmotic equilibrium with the agar medium via the humidified air inside the chamber . The moat solution was added using a P200 pipette . About 15 µl of molten NGM agarose was added to each well . About 5 µl of a suspension of the Escherichiae coli bacterial strain DA837 ( Davis et al . , 1995 ) , which is a streptomycin-resistant derivative of OP50 ( Brenner , 1974 ) was added to each well after agarose solidification . For aging experiments , late-L4 worms were added to the WorMotel manually with a platinum wire pick . PDMS devices were placed inside either a 90 mm petri plate for 48-well WorMotels or an OmniTray ( Nunc Thermo Scientific ) for 240-well WorMotels . 240-well WorMotels contained alignment tabs to keep devices in alignment with respect to the OmniTray . To maintain humidity inside the dishes , we used water-absorbing polyacrylamide crystals ( AgSAP S , M2 Polymer , West Dundee , IL ) . Sterile water was added to the crystals in a ratio of 150:1 ( water:crystals ) by weight . Approximately 15 g of hydrated crystals were added around the WorMotel . We placed lids on all dishes . To prevent accumulation of water condensation , lids were prepared by coating with a 30% solution of Tween 20 ( Sigma-Aldrich , St . Louis , MO ) in water , which was allowed to dry before use . We wrapped Parafilm ( Bemis , Nennah , WI ) around the sides of the plate to reduce water loss while allowing sufficient gas exchange . Images were captured with an Imaging Source DMK 23GP031 camera ( 2592 × 1944 pixels ) equipped with a Fujinon lens ( HF12 . 5SA-1 , 1:1 . 4/12 . 5 mm , Fujifilm Corp . , Japan ) . We used IC Capture ( Imaging Source ) or Phenocapture imaging software ( http://phenocapture . com/ ) to acquire time lapse images through a gigabit Ethernet connection . For daily imaging we used the time schedule option in Phenocapture to record images every 5 s for a 30 min period twice daily . All experiments were carried out under dark-field illumination using four 4 . 7" red LED strips ( Oznium , Pagosa Springs , CO ) positioned approximately 2" below the WorMotel . Images were saved and processed by a 64-bit computer with a 3 . 40 GHz Intel Core i3 processor and 4 GB of RAM . Images were analyzed using custom-written MATLAB software . Different spatial resolutions can be attained by adjusting the camera’s field of view and thus by modulating the number of wells viewed at once . Imaging six wells at once yields approximately 5 μm resolution , imaging 12 wells yields 7 μm resolution , imaging 48 wells yields 15 μm resolution , and imaging 240 wells yields 36 μm resolution . Temporally adjacent images were subtracted and divided by the average pixel intensity between the two images to generate normalized maps of pixel value intensity change . Depending on the task , time intervals of either 5 or 60 s were used to generate difference images . We refer to such difference images as 5-s or 60-s activity , respectively ( see Aging Behavior Quantification ) . A Gaussian smoothing filter with standard deviation of one pixel was applied to the resulting difference image in order to reduce image noise . A binary threshold of 0 . 25 was used to minimize image noise was then applied to the filtered intensity change image in order to score whether or not movement occurred at each pixel location . All pixels in which movement occurred were summed up and the resulting value was called the ‘activity’ between the two frames . To supply the blue light illumination , we use two high power LEDs ( Luminus PT-121 , Sunnyvale , CA ) secured to an aluminum heat sink and connected in series . We used a relay ( Schneider Electric , France ) controlled by MATLAB through a LabJack ( LabJack Corp . , Lakewood , CO ) or NIDaq ( National Instruments , Austin , TX ) interface to drive the LEDs at a current of 20 A through a power supply . To maximize the blue light irradiance and uniformity at the WorMotel , we constructed a box consisting of four acrylic mirrors with mirrored sides facing inwards and placed it around the WorMotel . We measured irradiance using a silicon power meter ( Coherent , Santa Clara , CA ) . During aging experiments , blue light stimulation was applied once every twelve hours for 10 s . Temperature was continuously monitored with a temperature probe ( LabJack EI1034 ) placed beside the sealed WorMotel . A previous report ( Edwards et al . , 2008 ) found that continuous blue illumination at an irradiance of 2 . 8 mW/mm2 kills N2 worms in 30 min . In our experiments , worms are subject to 20 s of blue light per day ( 10 s per stimulus , two stimuli per day ) . Therefore , it would take 90 days for worms to accrue 30 min of total illumination time with blue light . The irradiance of our blue light stimulus is approximately five times weaker than that used in this Edwards et al . study . Assuming a linear response of blue light dosage toxicity , it would take 450 days for worms to accrue a toxic blue light dose in our experiments . Since the typical worm lifespan is between 15 and 30 days , we believe blue light toxicity in our experiments is not significant . To test whether filling moats with a copper sulfate solution had any effects on worm development or survival , we prepared two WorMotels: in the first , moats were filled with NGM Buffer ( NGMB ) , and in the second , moats were filled with 100 mM copper sulfate . NGMB consists of the same constituents as NGM agar ( Stiernagle , 2006 ) but without peptone , cholesterol , or agar . We manually added L3 larvae to each device and monitored the duration of the L4 stage as previously described ( Nelson et al . , 2013 ) . When these worms reached the first day of adulthood , each animal was manually transferred to a new WorMotel in which the agar contained FUdR in order to assess the effect of moat solution on survival . Worms grown as larvae in the presence of a copper sulfate moat were transferred to a WorMotel in which moats were filled with copper sulfate; likewise for worms grown in the presence of an NGM moat . Lifespan was scored daily by manual methods ( see below ) . We used RNAi clones for EV , daf-2 , and odr-10 supplied by the Ahringer RNAi Libray ( Source Bioscience , Nottingham , UK ) . The bacterium was E . coli strain HT115 . We induced RNAi in liquid culture for two hours using 1 μM IPTG . We added 2 μM IPTG and 25 μg/mL carbenicillin to molten WorMotel agar . Liquid bacteria suspensions were added on top of solidified agar . Worms were manually added to each well with a platinum wire pick . For manual assays , immobile worms were prodded three times with a platinum wire pick . Those that failed to respond were scored as dead . For automated assays , the maximum activity value recorded during the fifteen minutes after each light stimulus was used to determine time of death . Time of death was defined as the time point after the last time point for which the maximum 60-s activity was nonzero . Any worm whose activity was uniformly zero beginning on day 2 of adulthood was assumed to have left its well or was not added due to experimental error , and these wells were censored from further analysis . 15 out of 1230 worms ( 1 . 2% ) were censored in this manner . tax-4 mutants were found to escape their wells as young adults at a much higher rate than all other strains . Therefore , tax-4 mutants found to be absent from their wells at the conclusion of each experiment were censored from analysis . 18 out of 90 worms were censored in this manner . For each 30-min imaging epoch , spontaneous and stimulated locomotion were calculated as the maximum 60-s activity before and after the blue light stimulus , respectively . Spontaneous and stimulated locomotion reflect the maximum movement of an individual before or after blue light stimulus , respectively . The response duration was calculated as the total time during which the 5-s activity was greater than zero after the light stimulus . Response duration reflects the total time spent moving after the stimulus . The response latency was calculated as the time elapsed between the blue light stimulus and the first non-zero 5-s activity . The response latency reflects the time required for an individual to respond to an aversive blue light stimulus . Animals that did not move at all during the 15 min following blue light stimulation were assigned a response latency of 900 s . We first considered the stimulated activity of each individual as a fraction of life rather than chronological time . We defined early-life as 20–60% of an individual’s life . We defined late-life as 60–100% of an individual’s life . Decline rate was calculated to be the negative of the slope of the stimulated activity during either early or late-life . Slope was determined with a linear fit in Matlab . For each individual , the change in decline rate was calculated as the late-life decline rate minus the early-life decline rate . 300 mM paraquat stock solutions were prepared fresh each day . L4 animals were added to a WorMotel as normal . On either day 1 or day 9 of adulthood , 2 μL of paraquat stock solution were added to each well and allowed to dry . The final concentration of paraquat per well was about 40 mM , assuming uniform distribution into the agar . After addition of paraquat to each well , excess liquid was allowed to dry ( about two hours ) , and plates were imaged continuously at 0 . 2 frames per second . Blue light stimulation occurred once per hour for 10 s . Stimulated activity for each animal was determined as the maximum activity between each light stimulus . Lifespan was determined automatically as described above . For a given population , we calculated stimulated activity as a fraction of life during aging and stress . We then corrected the activity during stress by calculating the average difference between the two quantities . By correcting for activity amplitude we could then ask whether the shape of decline was similar during aging and stress regardless of any offset . Finally , we calculated the normalized difference between aging and stress behavior at each fractional time point of life . The average normalized difference was reported as the percentage difference between behavior during aging and stress . Day replicates were defined as WorMotels prepared independently on separate days . For lifespan experiments involving DA837 bacterial food ( Figures 3 , 5 and 6 ) , at least three day replicates were performed for each strain . For lifespan experiments involving RNAi by feeding ( Figure 8 ) , two day replicates were performed . For survival experiments in which Paraquat was added at day 1 of adulthood ( Figure 7 ) , four day replicates were performed . For survival experiments in which Paraquat was added at day 9 of adulthood ( Figure 7b , d , f , h ) , two day replicates were performed . Differences in lifespan and survival distributions ( Tables 1–2 ) were compared using a Wilcoxon rank sum test . Behavioral comparisons ( Figure 6d , Figure 8a–h ) were performed using a two-tailed t-test . | Aging affects almost all living things , yet little is known about the biological changes that occur as we get older . Scientists often study aging in the microscopic roundworm Caenorhabditis elegans because it reproduces quickly and its lifespan is short ( about 2–3 weeks on average ) . To date , investigations have helped to reveal genes that affect overall lifespan . However , it is not known how much these genes also affect the animal’s healthy lifespan or “healthspan” , that is to say , the length of time before advancing age begins to negatively affect health . Until now , studies with worms have often been limited because measuring health and aging required time-consuming and difficult manual experiments . This also meant that worms were studied together as groups , rather than as individuals , providing a simplified picture of what was going on . An automated system in which many single worms can be analyzed and assessed would provide a much more detailed view of the effects of aging on health . Churgin et al . have now developed a device called the WorMotel to allow simultaneous automated examination of 240 worms throughout their entire adult lifespan . The WorMotel is a rectangular slab of clear silicone rubber with small wells in it . A single worm is confined in each well with a source of bacteria for food , and a camera is used to track and monitor each worm’s behavior over time . This device confirmed that worms move more slowly as they get older , which was taken to be a measurement of the worms’ declining health . Worms that lived the longest declined over the first few days and then had a long plateau of very low activity before eventually dying . Short-lived worms became slower and died fairly promptly . Churgin et al . also showed that the worms with mutations that increase lifespan declined in a similar way to the longest-lived normal worms , and that mutants with shorter lifespans declined like the shortest-lived normal worms . Also , normal worms that had been exposed to a chemical called paraquat – which stresses the worm's cells and shortens the worm’s lifespans to a few days – slowed down in a similar manner as aging worms , suggesting that the stress is similar to the aging process . Tools like the WorMotel can improve our understanding of the links between lifespan and healthspan . The tool is designed to be versatile and can be used with standard imaging systems and automated tools , meaning it can be scaled up to deal with tens of thousands of worms at once . Churgin et al . are now using the WorMotel to find other genes that influence healthspan and understand how they contribute to deteriorating health as animals age . Aging affects us all and learning more about healthspan could lead to drugs or interventions to help more people to live healthily for longer . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression",
"tools",
"and",
"resources",
"neuroscience"
] | 2017 | Longitudinal imaging of Caenorhabditis elegans in a microfabricated device reveals variation in behavioral decline during aging |
It has been postulated that a proneural factor , neurogenin 1 ( Ngn1 ) , simultaneously activates the neurogenic program and inhibits the alternative astrogliogenic program when specifying the neuronal fate . While Ngn1 substantially suppresses the activation of the astrogliogenic Jak-Stat pathway , the underlying molecular mechanism was unknown . Here , by employing in vivo and in vitro approaches , we report that Ngn1 binds to the promoter of a brain-enriched microRNA , miR-9 , and activates its expression during neurogenesis . Subsequently , our in vitro study showed that miR-9 directly targets mRNAs of Lifr-beta , Il6st ( gp130 ) , and Jak1 to down-regulate these critical upstream components of the Jak-Stat pathway , achieving inhibition of Stat phosphorylation and consequently , suppression of astrogliogenesis . This study revealed Ngn1 modulated non-coding RNA epigenetic regulation during cell fate specifications .
It is known that leukemia inhibitory factor ( Lif ) -activated Jak-Stat signaling controls the onset of neurogenic-to-astrogliogenic transition ( Bonni et al . , 1997 ) . Lif-activated Jak-Stat signaling pathway in neural stem/progenitor cells ( NPCs ) starts with three critical components at the cell membrane: the heterodimeric Lif receptor Lifr-beta and Il6st , as well as the receptor-associated Janus kinase ( Jak ) , which are involved in activation of the signaling transducer and activator of transcription ( Stat ) ( Sun et al . , 2001 ) . We previously demonstrated that the expression of the proneural basic-Helix-Loop-Helix ( bHLH ) factor neurogenin 1 ( Ngn1 ) inhibits glial differentiation by two distinct mechanisms: ( 1 ) Ngn1 sequesters the transcription co-activators Crebbp ( CBP ) /E1A binding protein p300 ( Ep300 ) away from glial specific promoters; ( 2 ) Ngn1 inhibits Stat1/3 phosphorylation ( Sun et al . , 2001 ) . Moreover , we have previously revealed that during astrogliogenic period , phosphor-Stat1/3 directly induces the expression of Il6st and Jak1 to strengthen Stat signaling and trigger astrogliogenesis ( He et al . , 2005 ) . However , the underlying mechanism of how Ngn1 inhibits Stat1/3 activity to block precocious astrocyte differentiation during the neurogenic period is still elusive . Our previous and current studies showed that protein and mRNA levels of Ngn1 were high during neurogenesis ( E12 to E15 ) but significantly reduced during astrogliogenesis ( P0–P4 ) ( Figure 1A , lower panel ) ( He et al . , 2005 ) . Recently , a group of small non-coding RNA , microRNA ( miRNA ) , has emerged as a novel neuroepigenetic mechanism that can rapidly fine-tune gene expression to regulate developmental timing and cell fate specification ( Lee et al . , 1993; Wightman et al . , 1993; Shibata et al . , 2008 ) . While screening the expression pattern of brain specific miRNAs in developing brains and in adult brains by quantitative PCR ( qPCR ) , we found that expression of miR-9 showed a bell-shaped pattern with their expression reaching maximum level at E16 , right before the onset of astrogliogenesis ( Figure 1A , upper panel ) . To analyze temporal expression of miR-9 in developing mouse brain , we carried out in situ hybridization ( ISH ) analysis . Our ISH data showed that miR-9 was extensively expressed in the ventricular zone ( VZ ) and subventricular zone ( SVZ ) progenitors in prenatal brains . In neonatal brains ( P1–P7 ) , the expression of miR-9 persisted in the SVZ , but at a lower level ( Figure 1B ) . Previous studies by Britz et al . ( 2006 ) and Ge et al . ( 2006 ) showed that the Ngn1 was highly expressed in VZ/SVZ progenitors during neurogenic period . Taken together , these evidences implied that the expression pattern of miR-9 somewhat correlated with the expression of Ngn1 , compatible with a potential regulatory relationship between Ngn1 and miR-9 . 10 . 7554/eLife . 06885 . 003Figure 1 . Neurogenin 1 ( Ngn1 ) directly regulated miR-9 expression . ( A ) Temporal expression of miR-9 and Ngn1 in the developing and adult mouse cortex . Upper panel: Taqman quantitative PCR ( qPCR ) analysis of endogenous miR-9 expression levels . Data were normalized to the loading control U6 non-coding RNA . Lower panel: qRT-PCR of endogenous Ngn1 mRNA levels . Data were normalized to loading control Gapdh . Gapdh , glyceraldehyde-3-phosphate dehydrogenase . ( B ) Expression analysis of miR-9 in the prenatal and neonatal mouse cortex by in situ hybridization ( ISH ) . The mature miR-9 was extensively expressed in the VZ/SVZ progenitors in developing cortex . VZ , ventricular zone . SVZ , subventricular zone . Scale bar: 100 μm . ( C ) Ngn1 up-regulated the activity of the miR-9-2 promoter-driven luciferase reporter but not Ngn1 binding site mutant Ngn1 ( AQ-Ngn1 ) . ( D ) ChIP-qPCR analysis showed Ngn1 bound to miR-9-2 but not miR-99b promoter region in mouse E11 cortical NPCs overexpressing T7-tagged Ngn1 . Data was presented as percentage pull down ( IP using T7 antibody ) comparing to the input . Negative control IgG pull down was also shown ( *p < 0 . 05 , Mann–Whitney test ) . ( E ) ChIP-qPCR of endogenous Ngn1 associated with miR-9-2 promoter in E15 and in P3 mouse cortexes . Data was presented as percentage pull down ( IP using Ngn1 antibody ) comparing to the input . Negative control IgG pull down was also shown ( *p < 0 . 05 , Mann–Whitney test ) . ( F ) Taqman qPCR analysis of the expression level of miR-9 in mouse NPCs in the presence of T7-tagged Ngn1 ( *p < 0 . 02 , Mann–Whitney test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06885 . 003 By searching upstream sequences of the transcription start sites ( TSS ) of all three mouse miR-9 genes , we identified a putative Ngn1 binding site containing the E-box element CATATG located 2 . 5 kb upstream of the miR-9-2 TSS ( Figure 1C ) . Subsequently , we confirmed that Ngn1 , but not its DNA-binding mutant form , AQ-Ngn1 , specifically activated the miR-9-2 promoter-driven luciferase reporter ( Figure 1C ) . As expected , when the E-box element was mutated , Ngn1 activation of the miR-9-2 promoter was abolished ( Figure 1C ) . To further confirm direct binding of Ngn1 to the putative binding site within the miR-9-2 promoter , we performed chromatin-immunoprecipitation combined with qPCR ( ChIP-qPCR ) using cultured mouse E11 cortical NPCs expressing T7-tagged Ngn1 . The ChIP-qPCR analysis showed that Ngn1 directly associated with the miR-9-2 putative promoter region encompassing the E-box element , but not to the promoter of an irrelevant/control brain enriched miRNA , miR-99b ( Figure 1D ) . To determine whether Ngn1 can directly bind to miR-9-2 promoter in vivo , we performed gene-specific ChIP-qPCR experiments using mouse cortex at both the neurogenic stage ( E15 ) when Ngn1 was robustly expressed and the astrogliogenic stage ( P3 ) when Ngn1 expression was diminished . We found that the miR-9-2 promoter was occupied by endogenous Ngn1 in E15 but not in P3 cortex ( Figure 1E ) . Finally , we found that overexpression of Ngn1 significantly up-regulated miR-9 expression ( Figure 1F ) . Collectively , these results demonstrated that Ngn1 could directly control the expression of miR-9 during the neurogenic period . Previous and current studies have showed that ( 1 ) Ngn1 inhibited astrogliogenesis partially by inhibiting the astrogliogenic Jak-Stat pathway ( Sun et al . , 2001 ) ; ( 2 ) miR-9 potentially inhibited Stat activation in mouse embryonic stem cell in vitro ( Krichevsky et al . , 2006 ) ; and ( 3 ) the expression of miR-9 was controlled by Ngn1 during neurogenesis . Based on these evidences , therefore , we predicted that miR-9 should play essential roles in inhibiting astrocyte differentiation . To test this , we built miR-9 overexpression ( miR-9 ) and miR-9 knockdown sponge ( miR-9AS ) constructs ( Figure 2—figure supplement 1A ) . miR-9 , miR-9AS , or control plasmid was delivered into progenitor cells that resided in the ventricular surface by in utero electroporation at E16 ( Figure 2A ) . The electroporated cells ( GFP+ ) migrated to layers 2/3 without obvious migration defect in all experimental conditions . We analyzed the ratios of both astrocyte marker+ cells and neuronal marker+ cells vs total transfected cells ( GFP+ ) at P30 . Our in vivo study showed that overexpression of miR-9 significantly decreased the number of astrocytes by using three astrocyte markers , Aldh1l1 , Slc1a3 ( EAAT1 ) , and Gfap , while knockdown of miR-9 increased it ( Figure 2B–D ) . We found that neither overexpressing nor knocking down miR-9 had significant effect on the number of Rbfox3+ ( NeuN+ ) neurons ( Figure 2B–D ) . By miR-9 ISH combined with Gfap and Rbfox3 immunostaining analysis in adult mouse brains , it became clear that miR-9 was expressed in neurons ( Rbfox3+ ) , but not in astrocytes ( Gfap+ ) ( Figure 2—figure supplement 1B ) , consistent with the notion that miR-9 inhibited the astroglial program . 10 . 7554/eLife . 06885 . 004Figure 2 . miR-9 inhibited astrogliogenesis in vivo . ( A ) Schematic representation of in utero injection of miR-9 constructs followed by electroporation into cortical progenitors resided at the ventricular surface at E16 . The right panels showed that cells electroporated at E16 ( green ) with each condition all migrated to proper cortical layers without overt migration defects . Scale bar: 100 μm . ( B–D ) Left panels showed examples of astrocyte marker+ or neuronal marker+ cells . The arrow pointed Rbfox3+ ( NeuN ) cortical neuron ( green and blue ) . The arrowhead pointed astrocyte marker+ astrocyte ( green and red ) . Scale bar: 10 μm . The arrowhead pointed astrocyte was shown at greater magnification in the last left panel ( confocal montage with orthogonal views taken at the center of the cell , scale bar: 5 μm ) . Right panels showed that overexpression of miR-9 in vivo dramatically attenuated astrogliogenesis , whereas knockdown of miR-9 had the opposite effect ( **p < 0 . 01 , *p < 0 . 5 , Mann–Whitney test ) . Aldh1l1 , aldehyde dehydrogenase 1 family , member L1; Slc1a3 ( EAAT1 ) , solute carrier family 1 ( sodium-dependent glutamate/aspartate transporter 1 ) , member 3; Gfap , glial fibrillary acidic protein . DOI: http://dx . doi . org/10 . 7554/eLife . 06885 . 00410 . 7554/eLife . 06885 . 005Figure 2—figure supplement 1 . ( A ) Upper left panels showed construction of miR-9 overexpression plasmid ( miR-9 ) by inserting pre-miR-9-2 and its flanking sequences to a modified lentiviral plasmid FG12 . Lower left panels showed the construction of miR-9 knockdown lentiviral vector ( miR-9AS ) and knockdown control ( miR-9ASm ) . Four miR-9 bulged antisense sequences were inserted immediately downstream of the H1 promoter of the modified-FG12 vector . H1 , H1 promoter . UbiC , ubiquitin promoter . miR-9AS mutant control , miR-9Sm was constructed by modifying 3-nt in miR-9 seed binding region . Right panels showed the luciferase assay of miR-9 knockdown efficiency . Upper right panel showed the construction of miR-9 luciferase assay plasmid . Three miR-9 bulged antisense repeats were inserted immediately downstream of a firefly luciferase gene . Lower right panel showed relative fold changes of luciferase activity . miR-9 luciferase construct was co-transfected into HEK239T cells with miR-9 and/or miR-9AS . A Renilla luciferase plasmid pRL-TK was used as an internal control . The empty vector FG12 ( con ) and the miR-9 binding site mutant miR-9AS ( miR-9ASm ) were used as negative controls . *p < 0 . 05 , Wilcoxon-Mann-Whitney test . ( B ) Endogenous miR-9 was expressed in neocortical neurons , but not in astrocytes in adult cortex . miR-9 ISH combined with Rbfox3 ( NeuN ) and Gfap double immunostaining in the upper left panels were shown at greater magnification in the right panels . The lower left panels showed miR-9 positive neuron that overlaps with a neuronal maker , Rbfox3 ( green , arrowhead ) , but not Gfap positive astrocyte ( red , arrow ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06885 . 005 To investigate the mechanisms of how miR-9 mediated Ngn1 regulation to prevent precocious astrocyte differentiation , we employed in vitro approaches . miR-9 , miR-9AS , or the control plasmid was delivered into E13 cortical progenitor cells by electroporation ( Figure 3A ) . The ratios of Gfap+ astrocytes and Map2+ neurons in transfected cells ( GFP+ ) were analyzed at 10 DIV ( Days In Vitro ) . Overexpression of miR-9 significantly decreased the number of astrocytes , whereas knockdown of miR-9 had the opposite effect ( Figure 3A ) . Transfection of mouse E11 NPCs with exogenous miR-9 duplexes significantly reduced the number of astrocytes , while having no significant effect on the number of neurons ( Figure 3—figure supplement 1A ) . In agreement with our in vivo and in vitro findings , the luciferase reporter assays showed that miR-9 suppressed the activation of glial-specific Gfap promoter , with little effect on the neurogenic promoter Neurod1 ( Figure 3—figure supplement 1B ) . While Ngn1 could both activate the neurogenic program and simultaneously inhibit the astrogliogenic program ( Figure 3—figure supplement 1C ) , miR-9 appeared to be only involved in the inhibition of glial fate . 10 . 7554/eLife . 06885 . 006Figure 3 . miR-9 inhibited astrogliogenesis via targeting three components of Jak-Stat pathway . ( A ) Upper right panels: schematic representation of in vitro delivery of miR-9 constructs into cortical progenitors by electroporation . Lower left panels showed examples of astrocyte marker Gfap+ or neuronal marker Map2+ cells . Right panel showed overexpression of miR-9 significantly reduced the number of astrocytes 10 days after electroporation , whereas knockdown of miR-9 dramatically promoted astrogliogenesis ( **p < 0 . 01 , *p < 0 . 5 , Mann–Whitney test ) . Overexpression of miR-9 in NPCs did not increase the number of neurons . Knockdown of miR-9 reduced the number of Map2+ neurons . Map2 , Microtubule-Associated Protein 2 . ( B ) Transfection of mouse NPCs with exogenous miR-9 duplex blocked phosphorylation of Stat1/3 without altering their protein levels . ( C ) Luciferase activity of Lifr-beta , Il6st ( gp130 ) , and Jak1 3′ UTR luciferase reporter in the presence of different combinations of control ( con ) , miR-9 , miR-9 inhibitor con , and miR-9 inhibitor in mouse NPCs . ( D ) miR-9 inhibited protein levels of Jak-Stat signaling components in NPCs transfected with control or miR-9 duplex . Right panels: western blotting densitometry analysis of protein level changes . Actb ( β-actin ) serves as the loading control . ( E ) A constitutively active form of Stat3 , Stat3C bypassed the effect of miR-9 inhibition on astrocyte differentiation . ( F ) Schematic representation of Ngn1-regulated miR-9 signaling that modulates Stat1/3 phosphorylation to control cell fate specification . Ngn1 up-regulates miR-9 expression during neurogenesis . miR-9 reduces protein levels of Lifr-beta , Il6st , and Jak1 of Jak-Stat signaling pathway by targeting their 3′ UTRs , which in turn abolish Stat1/3 phosphorylation to suppress astrogliogenesis . Crebbp: CREB binding protein , Smad1: Mothers Against DPP Homolog 1 ( Drosophila ) , Ep300: E1A Binding Protein p300 , E-protein: ubiquitous basic-Helix-Loop-Helix proteins , such as E12 or E47 . The dotted arrows show the inhibition regulations . DOI: http://dx . doi . org/10 . 7554/eLife . 06885 . 00610 . 7554/eLife . 06885 . 007Figure 3—figure supplement 1 . miR-9 and Ngn1 inhibited astrogliogenesis . ( A ) Transfection of mouse E11 cortical NPCs with exogenous miR-9 duplex significantly reduced the number of Gfap+ astrocytes in vitro . ( B ) miR-9 suppressed activation of glial-specific Gfap promoter , with little effect on the neurogenic promoter Neurod1 . ( C ) Ngn1 promoted neurogenesis and suppresses astrogliogenesis . Western blotting analysis showed that overexpression of Ngn1 in mouse E11 cortical NPC promoted Tuj1 expression , while reduced Gfap+ expression ( left panels ) . Tuj1 , beta III tubulin , a neuronal marker . Right panel showed overexpression of Ngn1 in mouse E11 cortical NPC promoted neurogenesis ( Tuj1+ cells ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06885 . 00710 . 7554/eLife . 06885 . 008Figure 3—figure supplement 2 . miR-9 targeted Jak-Stat signaling pathway . ( A ) Predicted duplex formation between mouse Lifr-beta , Il6st , and Jak1 3′ UTR ( top ) and miR-9 ( bottom ) . ( B ) Luciferase activity of wild type Lifr-beta 3′ UTR and miR-9 binding region mutant Lifr-beta 3′ UTR reporter genes with or without miR-9 duplex in mouse E11 cortical NPCs ( *p < 0 . 0001 , Wilcoxon-Mann-Whitney test ) . ( C , D ) Ngn1 suppressed the expression of three major components of Jak-Stat signaling pathway . ( C ) Western blot analysis of Ngn1 suppressed the protein levels of Lifr-beta , Il6st , and Jak1 . Actb served as the loading control . ( D ) Overexpression of Ngn1 decreased the activity of luciferase reporter fused to 3′ UTRs of Il6st , Lifr-beta , and Jak1 mRNAs . ( *p < 0 . 05 , **p < 0 . 005 , Wilcoxon-Mann-Whitney test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06885 . 008 We confirmed that transfection of mouse NPCs with exogenous miR-9 duplex blocked phosphorylation of Stat1/3 without altering their protein levels ( Figure 3B ) , which is in agreement with the work by Krichevsky et al . ( 2006 ) . To explore how miR-9 inhibits Stat phosphorylation , we scanned the 3′ UTRs of three upstream components of the Jak-Stat pathway ( www . targetscan . com ) and found that Lifr-beta , Il6st , and Jak1 contain putative miR-9 binding-sites ( Figure 3—figure supplement 2A ) . Luciferase reporter assays were used to confirm miR-9 targeting of these components . Transfection of exogenous miR-9 duplex led to an average of 40% decrease in luciferase activities of reporter constructs carrying 3′ UTRs of Lifr-beta , Il6st , and Jak1 mRNAs . Such effect was reversed by a miR-9 inhibitor ( duplex ) ( Figure 3C ) . Furthermore , we showed that the effects of miR-9 on the targets were dependent on the miR-9 binding-sites within the 3′ UTRs as the luciferase reporter containing mutant Lifr-beta 3′ UTR ( with mutated miR-9 binding-site ) was resistant to miR-9 inhibition ( Figure 3—figure supplement 2B ) . In line with the luciferase assay , we showed that the transfection of mouse NPCs with exogenous miR-9 duplex reduced protein levels of Lifr-beta , Il6st , and Jak1 ( Figure 3D ) . We confirmed that Ngn1 also down-regulated the expression levels of Lifr-beta , Il6st , and Jak1 ( Figure 3—figure supplement 2C ) ( Sun et al . , 2001; He et al . , 2005 ) . We further verified this result by luciferase assays . Overexpression of Ngn1 suppressed the luciferase activities of all three critical component of Jak-Stat pathway ( Figure 3—figure supplement 2D ) . In supporting the above-mentioned findings , we showed that co-expression of a constitutively active form of Stat3 ( Stat3C ) with miR-9 in NPCs could bypass the effect of miR-9 inhibition on astrocyte differentiation ( Figure 3E ) . In addition , we previously showed that Jak-Stat signaling has a positive auto-regulation loop , where phosphor-Stat1/3 activates Il6st and Jak1 expression ( He et al . , 2005 ) . Therefore , the DNA binding mutant Ngn1 ( AQ-Ngn1 ) , by sequestration of the transcriptional co-activator Crebbp , should also inhibit Stat1/3 mediated activation of Il6st and Jak1 , which in turn inhibits Stat1/3 phosphorylation ( Sun et al . , 2001 ) . It appeared that the Ngn1 sequestration regulation and the Ngn1 induced miR-9 regulation acted jointly to suppress Jak-Stat signaling to warrant the neuronal cell fate specification during the time period of neurogenesis ( Figure 3F ) . Since the first miRNA was discovered 20 years ago , miRNA has emerged as a new mechanism of epigenetic regulation that can rapidly respond to extrinsic cues to pattern the activities of particular target protein-coding genes and generate different types of cells over a short period of time ( Lee et al . , 1993; Wightman et al . , 1993; Sauvageot and Stiles , 2002; Miller and Gauthier , 2007 ) . This study demonstrated a novel molecular mechanism through which miR-9 mediated the action of Ngn1 by suppressing the expression of three major components of Jak-Stat singling , Lifr-beta , Il6st , and Jak1 , which attenuated Stat1/3 phosphorylation to inhibit the astrogliogenic differentiation genes or programs . Our previous and present work revealed that Ngn1 modulated multiple layers of genetic and epigenetic regulations to secure the process of neuronal fate specification .
Small RNA from samples was purified by using the Trizol reagent ( Invitrogen , Thermo Fisher Scientific , Waltham , MA ) . miRNA real-time PCR was performed by using Taqman miRNA assay kit ( Applied Biosystems , Thermo Fisher Scientific , Waltham , MA ) . LNA 5′-DIG-labeled mercury miR-9 probe ( Exiqon , Woburn , MA ) were used in ISH on mouse brains . Briefly , brains of CD-1 embryos were dissected and fixed in freshly prepared 4% paraformaldehyde ( PFA ) ( Sigma-Aldrich , St . Louis , MO ) for 2 hr at room temperature . The fixed brains were perfused in PBS with 30% sucrose overnight at 4°C . The next day , the brain tissues were embedded in Sakura Finetek Tissue-Tek O . C . T . Compound and frozen on dry-ice . The frozen brain was sectioned at 10 μm thickness and mounted on Superfrost PLUS slides . After acetylation , the tissue was permeablized by proteinase K at a final concentration of 5 μg/ml and then pre-hybridized by hybridization buffer ( 50% formamide ( Sigma-Aldrich , St . Louis , MO ) , 5× SSC , 5× Denhardt's solution ( Thermo Fisher Scientific , Waltham , MA ) , 200 μg/ml of yeast RNA ( Thermo Fisher Scientific , Waltham , MA ) , 500 μg/ml of salmon sperm DNA ( Thermo Fisher Scientific , Waltham , MA ) , and 2% Roche blocking reagent ( Roche Applied Sciences , Penzberg , Upper Bavaria , Germany ) ) for 8 hr . For hybridization , 1 pM of the LNA 5′-DIG-labeled mercury probe in 150 μl denaturizing buffer ( 50% formamide , 5× SSC , 5× Denhardt's solution , 200 μg/ml of yeast RNA , 500 μg/ml of salmon sperm DNA , 2% Roche blocking reagent , 0 . 25% CHAPS ( Sigma-Aldrich , St . Louis , MO ) , and 0 . 1% Tween-20 ( Sigma-Aldrich , St . Louis , MO ) ) was added per slide . The hybridization was carried out overnight at 50°C . After stringency washes , hybridization probe was visualized using anti-DIG-alkaline phosphatase conjugated substrate ( Roche Applied Sciences , Penzberg , Upper Bavaria , Germany ) . The immunostaining was carried following the ISH . The primary antibodies used were anti-Gfap polyclonal antibody ( Abcam , Cambridge , MA ) and anti-Rbfox3 ( NeuN ) monoclonal antibody ( Chemicon , EMD Millipore , Billerica , MA ) . Fluorophore conjugated secondary antibody were purchased from Invitrogen ( Molecular Probe , Thermo Fisher Scientific , Waltham , MA ) . Images were taken with a Zeiss LSM510 confocal microscope , processed with software Imaris ( Bitplane , Switzerland ) and composed with adobe Photoshop . To construct miR-9 overexpression lentiviral plasmid , the miR-9-2 pre-miRNA fragment with 80 bp flanking sequences was PCR amplified from CD-1 mouse genomic DNA . The PCR product was first cloned into BamH1 and XhoI sites of pBS-hH1 shuttle vector , and then the pre-miR-9 sequence together with human H1 Pol III promoter of the vector pBS-hH1 was digested by Xbal1 and Xho1 and subcloned into FG12 lentiviral vector . miR-9 overexpression forward oligo: CCGGATCCCTGGAGTTCAGCCAGAGGAA . miR-9 overexpression reverse oligo: CCCTCGAGGGTTTTTACTGTCTCTTGGTTGC . To efficiently suppress miR-9 activity , we inserted four bulged miR-9 antisense sequences immediately downstream of the H1 promoter of FG12 ( Ebert et al . , 2007 ) . The control sequence has 3-nucleotide mutations in the miR-9 ‘seed’-binding region . The bulged miR-9 antisense sequences bind to endogenous mature miR-9 , serving as a miR-9 sponge to inhibit miR-9 cellular activity without altering endogenous miR-9 expression . To construct miR-9 knockdown lentiviral plasmid ( miR-9AS ) , the annealed oligonucleotides for miRNA binding sites with 4-nt spacers for bulged sites were cloned into lentiviral vector FG12 ( CTL ) . miR-9AS forward oligo: GATCTCATACAGCTCTTAACCAAAGAATCTCATACAGCTCTTAACCAAAGAATCTCATACAGCTCTTAACCAAAGAATCTCATACAGCTCTTAACCAAAGATTTTT . miR-9AS reverse oligo: TCGAAAAAATCTTTGGTTAAGAGCTGTATGAGATTCTTTGGTTAAGAGCTGTATGAGATTCTTTGGTTAAGAGCTGTATGAGATTCTTTGGTTAAGAGCTGTATGA . To test the efficiency of miR-9 knockdown plasmid , we cloned miR-9 bulged ( AS ) or miR-9 perfect-matched antisense oligonucleotides duplex into immediate downstream of a firefly luciferase reporter gene of plasmid pIS0 ( a gift kindly provided by David Baltimore ) . miR-9 luciferase assay forward oligo: TCATACAGCTCTTAACCAAAGAATCGTCATACAGCTCTTAACCAAAGAATCGTCATACAGCTAGATAACCAAAGA; miR-9 luciferase assay reverse oligo: CTAGTCTTTGGTTAAGAGCTGTATGACGATTCTTTGGTTAAGAGCTGTATGACGATTCTTTGGTTAAGAGCTGTATGAAGCT; miR-9 Luciferase assay mutant forward oligo: TCATACAGCTCTTAGCTGAAGAATCGTCATACAGCTCTTAGCTGAAGAATCGTCATACAGCTCTTAGCTGAAGA; miR-9 luciferase assay mutant reverse oligo: TAGTCTTCAGCTAAGAGCTGTATGACGATTCTTCAGCTAAGAGCTGTATGACGATTCTTCAGCTAAGAGCTGTATGAAGCT . All the lentiviral shuttle vectors and helper vectors were kind gifts from David Baltimore . The T7-tagged mouse Ngn1 was cloned into an adenoviral shuttle vector pMZL6 containing a GFP expression cassette . Control and myc-tagged mouse Ngn1 adenovirus were previously made ( Sun et al . , 2001 ) . Recombinant adenoviruses were made by co-transfection of the shuttle plasmids with the plasmid pBHG10 into HEK293 cells . Viruses were amplified by infecting HEK293 cells and supernatants were harvested , tittered and frozen at −80°C until infection . For the in utero electroporation , pregnant mice at embryonic day 16 ( E16 ) were deeply anesthetized with isoflurane , uterine horns were carefully exposed through a midline abdominal incision to perform in utero injection and electroporation . Plasmid ( 2 μl , 2 μg/μl ) in saline containing 0 . 01% fast green was injected into the lateral ventricle of the embryos through the uterine wall using a NanoFil Syringe with a 36 gouge needle ( World Procession Instruments , Sarasota , FL ) . After injection , electroporation ( 50 ms square pulses of 40 V with 100 ms , 5 intervals; BTX Electroporator ECM 830 ( Harvard Apparatus , Inc . , Holliston , MA ) was carried out . Then , uterine horns were placed back into the abdominal cavity , and the abdominal wall of the pregnant mouse was sutured . The pups at postnatal day 8 ( P8 ) and day 30 were transcardially fixed with 4% PFA and the brain was continuously fixed in 4% PFA 2 hr at room temperature . Vibratome-sections with GFP positive were subjected to immunohistochemistry . For in utero electroporation experiments , ICR CD-1 mice were used ( Charlies River Laboratories ( San Diego , CA ) . Following the in utero electroporation , the male mice were housed four per cage , maintained on a 12 hr light/dark schedule , and allowed free access to food and water . The protocols are approved by the Institutional Animal Care and Use Committee of the University of California , Los Angeles . Mouse ( CD1 or Balb/c ) cortical neural progenitor cells ( NPCs ) from E11 cortex were dissected , dissociated , and cultured in DMEM/F12 ( Invitrogen ) chemically defined medium supplemented with B27 . HEK293 and HEK293T cells ( ATCC , Manassas , VA ) were cultured in DMEM medium with 10% FBS , penicillin/streptomycin and glutamine . LIF ( 100 ng/ml , R&D Systems , Minneapolis , MN ) was used for astrocyte differentiation ( 1∼3-day long-term treatment ) . Short-term Lif ( 100 ng/ml ) treatment ( 20 min ) was used to detect Stat1/3 phosphorylation . To study the role of miR-9 in regulating astrogliogenesis , CD-1 mouse E13 NPC were isolated and the plasmid ( miR-9 , miR-9AS , con ) was electroporated into the cells using a Nucleofector device ( Lonza , Switzerland ) . To confirm above study , miRNA duplexes and 2′-O-methyl antisense oligonucleotides targeted miR-9 ( miR-9 inhibitor ) and control ( Dharmacon , GE Dharmacon , Lafayette , CO ) were electroporated into mouse NPCs by using the nucleofector and mouse neural stem cell nucleofection kit according to the manufacturer's instructions ( Lonza ) . The cells were fixed with 4% PFA for 10 min at room temperature . The primary antibodies that were used in this study are: anti-Gfap monoclonal antibody ( Sigma ) , anti-Gfap polyclonal antibody ( Abcam ) , anti-Slc1a3 ( EAAT1 ) polyclonal antibody ( Abcam ) , anti-Aldh1l1 polyclonal antibody ( Abcam ) , chicken anti-Map2 antibody ( Abcam ) , anti-Rbfox3 ( NeuN ) monoclonal antibody ( EMD Millipore ) , and anti-beta III Tubulin ( Tuj1 ) monoclonal antibody ( Abcam ) . For mouse antibody ( monoclonal antibody ) on mouse tissues , the endogenous mouse IgG blocking method was used to reduce background straining . Briefly , after normal serum blocking and before primary antibody incubation , the mouse brain sections were incubated with unconjugated affiniPure Fab fragment goat Anti-Mouse IgG ( H + L ) ( Jackson ImmunoResearch Labs , West Grove , PA ) for 1 hr at room temperature . Fluorophore conjugated secondary antibodies were purchased from Invitrogen ( Molecular Probe ) . Images were taken with a Zeiss LSM510 confocal microscope , processed with software NIH ImageJ and Imaris ( Bitplane ) , and composed with adobe Photoshop . The 1 . 9 kb Gfap promoter-luciferase reporter construct ( pGL3-Gfap ) and mouse Neurod1 promoter-luciferase reporter construct ( pGL3-Neurod1 ) were inserted into pGL3 firefly luciferase vector . The mouse miR-9-2 gene promoter sequence and its E-box motif mutant were amplified by PCR and cloned into the pGL3 fly-luciferase construct ( Promega , Madison , WI ) , respectively . The mouse Il6st ( gp130 ) 3′ UTR , Jak1 3′ UTR , Lifr-beta 3′ UTR and miR-9 seed region mutant of Lifr-beta 3′ UTR were amplified by PCR and cloned into a firefly luciferase vector , pIS0 vector after luciferase gene ( a gift from Dr David Bartel ) . The miRNA duplex and 2′-O-methyl oligos and fly-luciferase plasmids were co-transfected into E11 mouse NPCs at P2–P4 using Lipofectamine LTX ( Invitrogen ) . TK-pRL Renilla luciferase construct was used as transfection control ( Promega ) . Approximately 24 hr post-transfection , cells were lysed for dual luciferase assays ( Promega ) . The antibodies used for western were as follows: mouse anti-Gfap ( Sigma ) , rabbit anti-Jak1 ( Santa Cruz Biotechnology , Dallas , TX ) , rabbit anti-Il6st ( Santa Cruz , C-20 ) , rabbit anti-Lifr-beta ( Santa Cruz , C-19 ) , rabbit anti-Stat1 ( BD Biosciences , San Jose , CA ) , mouse anti-Stat3 ( BD Biosciences ) , mouse anti-phosphotyrosine Stat1 ( BD Biosciences ) , rabbit anti-phosphotyrosine Stat3 ( Cell Signaling , Danvers , MA ) , mouse anti-TuJ1 ( Covance , Princeton , NJ ) and mouse anti-β-actin ( Actb ) ( Sigma ) . Secondary goat anti-mouse or anti-rabbit IgG-horseradish antibodies ( Calbiochem , EMD Millipore , Billerica , MA ) were used , and detection was performed using the ECL plus chemiluminescence ( PerkinElmer , Waltham , MA ) on X-Omat Blue films ( Kodak ) . Briefly , ∼108 E11 mouse cortical NPC culture ( usually around passage 2 ) infected with T7-tagged Ngn1 adenoviruses to over 95% infection fate was chemically cross-linked by the additional of 11% formaldehyde solution for 20 min at room temperature . Cells were treated with 1/20 vol of 2 . 5 M glycine to quench the formaldehyde and washed three times with 1× PBS and pellets were harvested and stored at −80°C prior to use . Cells were re-suspended , lysed , and sonicated to solubilize and shear cross-linked DNA . We used a Branson Sonifier 450 and sonicated at power 5 with a microtip for 7–10 cycles of 30 s pulses ( 60-s pause between pulses ) at 4°C while samples were immersed in an ice bath . After saving 50 μl of whole cell extract ( WCE ) from each sample to store at −20°C , the rest of sonicated cell lysate was incubated overnight at 4°C with 100 μl Dynal Protein G magnetic beads ( Invitrogen ) preincubated with 10 μg of the appropriated antibody for at least 8 hr . Beads were then washed five times with RIPA buffer and 1 time with TE containing 50 mM NaCl . Bound complexes were eluted from the beads in elution buffer by heating at 65°C with occasional vortex , and crosslinking was reversed by 6 hr incubation at 65°C . The WCE was also treated for crosslink reversal with additional elution buffer at the same time . Immunoprecipitated DNA and WCE DNA were then purified by treatment with RNaseA , proteinase K and phenol: chloroform: isoamyl alcohol extractions . We used site-specific ChIP-PCR to confirm binding of Ngn1 to miR-9 promoter . Primers were designed to amplify around the E-box location on miR-9-2 promoter . PCR was performed on unamplified DNA samples of several sets of ChIP experiments . The final immunoprecipitated DNA product was dissolved in 70 μl of TE . 2 μl of IP DNA was used in PCR reactions . ∼50 ng of the input WCE DNA samples was used . qPCR was performed on iCycler ( Bio-Rad ) using iQ SYBR Green PCR supermix ( Bio-Rad ) . PCR efficiencies of primers were examined by standard curve of serial-diluted WCEs input and melting curve functionality . The enrichment was calculated as immunoprecipitation signal vs whole cell lysate input ( IP/WCE ) . Antibody used for IP: rabbit anti-Ngn1 ( from Greenberg lab , Harvard ) , rabbit anti-Crebbp ( CBP A-22 ) ( Santa Cruz ) . Normal rabbit anti-IgG antibody ( Santa Cruz ) was used as the negative . The ChIP-PCR primers for miR-9 were GCCACGGTGCTCTTTAATCT ( forward ) and TGGTCACAGCATAAACAACTCA ( reverse ) . The ChIP-PCR primers for miR-99b were GGGTCACCCATTTCCTTCTT ( forward ) and TTCTGAAGGAGGAGGGGATT ( reverse ) . Wilcoxon-Mann-Whitney test was used to evaluate the statistical significance of differences between groups . Data are presented as mean of fold change compared to control group ±SEM . | The brain processes information from all over the body through a complex network of cells called neurons . Other brain cells—including star-shaped cells called astrocytes—support this network . Both neurons and astrocytes originate from the same group of stem cells , which first give rise to neurons in a process called neurogenesis before they switch to producing astrocytes . A protein called neurogenin 1 promotes neurogenesis and suppresses the formation of astrocytes by regulating the activity of particular genes . It does so by binding to a region within the genes called the promoter . A cell communication system ( or ‘signaling pathway’ ) known as the Jak-Stat pathway is required for brain stem cells to make astrocytes . Previous research has shown that neurogenin 1 is present at high levels when stem cells start to make neurons , which leads to the inactivation the Jak-Stat pathway . However , when stem cells start to make astrocytes , the levels of neurogenin 1 decrease and the Jak-Stat pathway is activated . This signaling pathway therefore acts as a switch for the transition from neurogenesis to the formation of astrocytes , but it is not clear exactly how it works . When a gene is active , its DNA sequence is copied to make molecules of ribonucleic acid ( RNA ) . These molecules can be used as templates to assemble proteins—known as messenger RNAs . Alternatively , they may be processed to make another type of RNA called microRNA , which can switch off the activity of particular genes by promoting the destruction of particular messenger RNAs . Zhao et al . studied neurogenesis in the mouse brain and found that neurogenin 1 can directly bind to the promoter of a gene that makes a microRNA called miR-9 . The experiments show that neurogenin 1 increases the activity of this gene so that the amount of miR-9 in brain stem cells increases during neurogenesis . In turn , this microRNA lowers the activity of several critical genes that encode proteins involved in the Jak-Stat pathway . Zhao et al . 's findings reveal that neurogenin 1 promotes neurogenesis and inhibits astrocyte formation by regulating the production of miR-9 . The Jak-Stat pathway plays important roles in nerve injury , neural repair , and the immune system , so drugs that target miR-9 may have the potential to be developed into new therapies to treat diseases that affect the nervous system . | [
"Abstract",
"Main",
"text",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology",
"short",
"report",
"neuroscience"
] | 2015 | Ngn1 inhibits astrogliogenesis through induction of miR-9 during neuronal fate specification |
How long neural information is stored in a local brain area reflects functions of that region and is often estimated by the magnitude of the autocorrelation of intrinsic neural signals in the area . Here , we investigated such intrinsic neural timescales in high-functioning adults with autism and examined whether local brain dynamics reflected their atypical behaviours . By analysing resting-state fMRI data , we identified shorter neural timescales in the sensory/visual cortices and a longer timescale in the right caudate in autism . The shorter intrinsic timescales in the sensory/visual areas were correlated with the severity of autism , whereas the longer timescale in the caudate was associated with cognitive rigidity . These observations were confirmed from neurodevelopmental perspectives and replicated in two independent cross-sectional datasets . Moreover , the intrinsic timescale was correlated with local grey matter volume . This study shows that functional and structural atypicality in local brain areas is linked to higher-order cognitive symptoms in autism .
How long neural information is likely to be stored in a neural area is a fundamental functional property of the local brain region ( Chen et al . , 2015; Hasson et al . , 2015; Himberger et al . , 2018 ) , and has been quantified as temporal receptive window ( Hasson et al . , 2008; Chaudhuri et al . , 2015; Honey et al . , 2012; Lerner et al . , 2011; Stephens et al . , 2013; Yeshurun et al . , 2017 ) , temporal receptive field ( Cavanagh et al . , 2016 ) , or intrinsic neural timescale ( Murray et al . , 2014; Kiebel et al . , 2008; Gollo et al . , 2015; Cocchi et al . , 2016 ) . Computational studies propose that such neural timescales should show a rostrocaudal gradient in the brains ( Kiebel et al . , 2008 ) and that densely interconnected central regions , such as prefrontal and parietal areas , should have slower timescales compared to peripheral sensory areas ( Chaudhuri et al . , 2015; Gollo et al . , 2015 ) . These proposals are supported by empirical observations . Human neuroimaging and macaque electrophysiology studies show that neural timescales tend to be longer in frontal and parietal areas compared to sensory-related regions ( Honey et al . , 2012; Stephens et al . , 2013; Murray et al . , 2014; Ogawa and Komatsu , 2010 ) , and suggested that such a prolonged neural timescale enables these higher-order brain cortices to integrate diverse information for robust sensory perception ( Hasson et al . , 2008; Lerner et al . , 2011; Stephens et al . , 2013; Yeshurun et al . , 2017; Ogawa and Komatsu , 2010; Gauthier et al . , 2012 ) , stable memory processing ( Hasson et al . , 2015; Murray et al . , 2014; Bernacchia et al . , 2011 ) , and accurate decision making ( Cavanagh et al . , 2016; Runyan et al . , 2017 ) . A recent brain stimulation study directly demonstrates that such a hierarchy of the intrinsic timescale is closely related to functional interactions between lower and higher brain regions ( Cocchi et al . , 2016 ) . The heterogeneity of the neural timescale is considered to be a basis of the functional hierarchy in the brain ( Chen et al . , 2015; Hasson et al . , 2015; Himberger et al . , 2018; Chaudhuri et al . , 2015; Gollo et al . , 2015; Cocchi et al . , 2016; Kukushkin and Carew , 2017; Friston and Kiebel , 2009 ) . Given such fundamental roles of local neural dynamics in highly-organised information processing in the brain , we hypothesised that atypical intrinsic neural timescales should be observed in autism . In fact , the core symptoms of this prevalent neurodevelopmental disorder — challenges in socio-communicational skills and repetitive , restricted behaviours ( RRB ) — are often linked to atypical information processing ( Happé and Frith , 2006; Palmer et al . , 2017; Booth and Happé , 2018 ) : weak coherence theory suggests that autism spectral disorder ( ASD ) is associated with impairments of the global integration of diverse information and over-enhancement of individual inputs ( Happé and Frith , 2006; Booth and Happé , 2018 ) ; a recent Bayesian view also attributes autism to overweighing of local sensory information ( Palmer et al . , 2017; Lawson et al . , 2017 ) . These theories suggest that measures of local neural dynamics — such as intrinsic neural timescales — should be linked to the symptomatology of individuals with ASD . Despite such theoretical implications , no study has investigated intrinsic timescales of neural signals in autism . Here , we aimed at exploring this local neural property in the brains of high-functioning individuals with ASD and examining its associations with the core symptoms of this condition .
Based on previous macaque studies ( Chaudhuri et al . , 2015; Cavanagh et al . , 2016; Murray et al . , 2014; Bernacchia et al . , 2011 ) , we calculated an intrinsic neural timescale by assessing the magnitude of autocorrelation of the resting-state brain activity . First , we estimated the sum of autocorrelation function ( ACF ) values in the initial positive period of the ACF ( i . e . , the sum of the area of the green bars in Figure 1a ) . The upper limit of this period was set at the discrete time lag value just before the one where the ACF became non-positive for the first time . To adjust for differences in the temporal resolution of the neural data , we then multiplied the obtained sum of ACF values by the repetition time ( TR ) of the fMRI recording . This product was used as an index for intrinsic neural timescales . This definition was validated by comparing the fMRI-based timescale index to that based on neural data with a higher temporal resolution ( here , simultaneously recorded EEG data ( Deligianni et al . , 2016; Deligianni et al . , 2014 ) ; Figure 1—figure supplement 1 ) . The fMRI-based timescales were strongly correlated with those based on the gamma-band EEG signals ( adjusted R2 = 0 . 71; Figure 1b; see Figure 1—figure supplement 2 for other EEG bands ) . In addition , when the EEG signals were convolved with the hemodynamic response function ( HRF ) , the intrinsic timescales based on the HRF-convolved EEG signals became closer to those based on fMRI signals ( adjusted R2 = 0 . 61; Figure 1c ) . Based on this formulation , we compared the intrinsic neural timescale between 25 high-functioning adults with ASD and 26 age-/sex-/IQ-matched typically developing ( TD ) individuals ( Table 1 ) ( Di Martino et al . , 2014 ) . Both ASD and TD groups showed a similar whole-brain pattern of intrinsic neural timescales: longer timescales in frontal and parietal cortices and shorter timescales in sensorimotor , visual , and auditory areas ( Figure 2a ) . This observation is consistent with previous reports about a hierarchal topography of timescales of local neural activity in brains of mice ( Runyan et al . , 2017 ) , monkeys ( Chaudhuri et al . , 2015; Murray et al . , 2014; Ogawa and Komatsu , 2010 ) , and humans ( Hasson et al . , 2008; Honey et al . , 2012; Lerner et al . , 2011; Stephens et al . , 2013; Yeshurun et al . , 2017 ) . However , we also identified significant differences between the two groups ( Table 2; PFDR <0 . 05 ) . Individuals with ASD had a significantly shorter intrinsic timescale than TD individuals in bilateral postcentral gyri , right inferior parietal lobule ( IPL ) , right middle insula , bilateral middle temporal gyri ( MTG ) , and right inferior occipital gyrus ( IOG ) ( Figure 2b and d ) , whereas the intrinsic timescale in the right caudate was significantly larger in the ASD group ( Figure 2c and e ) . We then tested for any associations between the observed atypical intrinsic neural timescales and the severity of autism , as measured by the Autism Diagnostic Observation Schedule ( ADOS ) ( Lord et al . , 1989 ) . Because previous studies indicate that atypical neural information processing is a common basis for various ASD symptoms ( Happé and Frith , 2006; Belmonte et al . , 2004; Watanabe and Rees , 2017 ) , we first examined associations between the neural timescales and the overall severity of this disorder ( ADOS total scores ) . When no significant link was found in this analysis , we then calculated associations between the neural timescales and specific core symptoms . Of the seven brain regions of interest ( ROIs ) whose intrinsic timescales were shorter in the ASD group ( Figure 2b and d , Table 2 ) , the bilateral postcentral gyri and right IOG showed negative correlations between the intrinsic timescale and overall severity of autism ( rho ≤ –0 . 49 , Puncorrected <0 . 01 , PFDR <0 . 05; Figure 3a ) . The right caudate , a single ROI whose intrinsic timescale was significantly longer in the ASD group ( Figure 2e ) , did not show such an association with the overall ADOS score ( rho = 0 . 19 , p=0 . 34 ) . However , its intrinsic timescale was longer in individuals with more severe repetitive , restricted behaviours ( RRB ) , as measured by ADOS RRB scores ( F3 , 21 = 9 . 9 , p<0 . 001 , main effect of ADOS RRB in a one-way ANOVA; Spearman’s rho = 0 . 57 , p=0 . 002; Figure 3b ) . These brain-symptom associations were preserved even when we conducted this association analysis in a more statistically rigorous manner ( Figure 3—figure supplement 1 ) . That is , we applied the same ROI sets to two independent fMRI datasets ( ETH Zürich and Indiana University datasets; Supplementary Table 1 in Supplementary file 1 ) that were not used in the ROI search , and found negative correlations between the intrinsic timescales and the ADOS total scores in the bilateral postcentral gyri and right IOG ( rho ≤ –0 . 60 ) and a significant association between the intrinsic timescale and the ADOS RRB scores in the right caudate ( F ≥ 6 . 0 , p≤0 . 03 in one-way ANOVAs ) . We then examined whether these observations from adults with ASD could be seen in children with ASD . To this end , we analysed a longitudinal fMRI dataset recorded from adolescent children ( two MRI scans for each participant , interval of the two scans = 2 . 8 ± 0 . 4 years for ASD children , 3 . 0 ± 0 . 4 years for TD children; Supplementary Table 2 in Supplementary file 1 ) ( Di Martino et al . , 2014 ) . We traced the developmental trajectories of the intrinsic neural timescales of the four ROIs whose intrinsic timescales were atypical in the ASD group and associated with the severity of the symptoms . These four ROIs were defined as clusters found in the whole-brain analysis using the adult fMRI data ( Figure 2 , Table 2 ) . In adolescence , we found that the intrinsic neural timescale in bilateral postcentral gyri and right IOG was consistently shorter in individuals with ASD compared to TD individuals ( F1 , 31 > 9 . 0 , Puncorrected <0 . 005 , PFDR <0 . 05 , main effect of diagnosis in repeated-measures two-way ANOVAs with a diagnosis [ASD/TD] × scan order [1 st/2nd] structure; Figure 4a ) . In addition , the decreases in the intrinsic timescale in these areas during this period were predictive of the increases in the overall severity of autism ( rho ≤ –0 . 68 , Puncorrected <0 . 003 , PFDR <0 . 05; Figure 4b ) . In contrast , the intrinsic timescale in the right caudate was consistently longer in the group with ASD during adolescence ( F1 , 31 = 18 . 2 , Puncorrected <0 . 001 , main effect of diagnosis in a repeated-measures two-way ANOVA; Figure 4c ) , and the increase in the intrinsic timescale in this region was associated with progression of RRB symptoms ( F2 , 8 = 10 . 3 , p=0 . 006 , main effect of ADOS RRB changes in a one-way ANOVA; Figure 4d ) . These longitudinal observations are consistent with the cross-sectional findings ( Figures 2 and 3 ) and suggest that autistic atypicality of temporal neural processing in local brain areas may already occur before adolescence . Finally , we explored possible neuroanatomical bases for ( or consequences of ) intrinsic neural timescales by examining relationship with local grey matter volumes ( GMVs ) . We focused on GMV because theoretically , an increase in neuronal density , which is measured by GMV ( Kanai and Rees , 2011 ) , would enhance recurrent neural network activity , and then enlarge the autocorrelation strength in the neural signals . This theoretical assumption was validated by comparisons between intrinsic timescales and GMV across 360 brain areas ( Glasser et al . , 2016 ) : at a group level , these functional and anatomical properties were positively correlated with each other ( TD: r = 0 . 40 , ASD: r = 0 . 38 , p<10–5; Figure 5a ) . Furthermore , the significant correlations were robustly observed at a single-participant level as well ( TD: r ≥ 0 . 29 , ASD: r ≥ 0 . 28 , p<10–5; Figure 5b ) . This association was also seen in the four brain regions whose atypical intrinsic neural timescale was associated with symptoms of autism ( r > 0 . 52 , p≤0 . 005; Figure 5c ) . Given these findings together with atypical GMVs in the four brain regions in individuals with ASD ( t49 >3 . 0 , Puncorrected ≤0 . 004 in two-sample t-tests , PFDR <0 . 05; Figure 5d ) , we can infer that intrinsic neural timescale is a mediator linking atypical GMV and ASD symptoms . This inference was , in fact , consistent with results of mediation analyses ( Figure 5e ) . We replicated our findings in the two independent MRI datasets obtained from adults with ASD , which were collected in ETH Zürich and Indiana University ( Supplementary Table 1 in Supplementary file 1 ) ( Di Martino et al . , 2014 ) . In both datasets , ASD group yielded atypically shorter intrinsic timescales in the bilateral postcentral gyri and right IOG ( t ≥ 4 . 0 , PFDR <0 . 05; Figure 5—figure supplements 1a and 2a ) and a longer intrinsic timescale in the right caudate ( t = 3 . 9 , PFDR <0 . 05; Figure 5—figure supplements 1b and 2b ) . The shorter intrinsic timescale in the bilateral postcentral gyri and right IOG was correlated with the overall severity of ASD ( rho ≤ –0 . 51; Figure 5—figure supplements 1c and 2c ) , whereas the longer intrinsic timescale in the caudate was associated with RRB symptoms ( p<0 . 028 in one-way ANOVAs; Spearman’s rho ≥0 . 74; Figure 5—figure supplements 1d and 2d ) . Moreover , the correlations between the intrinsic timescale and GMV were also replicated ( r ≥ 0 . 49; Figure 5—figure supplements 1e and 2e ) .
We investigated the intrinsic neural timescale , whose length is closely related to the functional hierarchy in the brain ( Hasson et al . , 2015; Murray et al . , 2014; Cocchi et al . , 2016 ) , in high-functioning individuals with autism . By calculating the time-dependent magnitude of autocorrelation function in resting-state fMRI ( rsfMRI ) signals , we found that in adults with ASD , the intrinsic timescale was significantly shorter in the bilateral postcentral gyri and right inferior occipital gyrus , and longer in the right caudate . The shorter intrinsic timescale in these primary sensory/visual areas in autism was correlated with the overall severity of autism . The longer intrinsic timescale in the caudate in autism was associated with the severity of repetitive , restricted behaviours . Moreover , this temporal property in local neural signals was linked to local grey matter volumes ( GMVs ) . These findings indicate the possibility that functional and structural properties in local brain areas could have a critical influence on higher-order cognitive symptoms in autism . Furthermore , we investigated the validity of these observations on longitudinal neuroimaging data recorded from adolescents with ASD . We found atypical development of the intrinsic timescale during adolescence in autism ( Figure 4a and c ) and identified significant associations between such atypical neural development and progression of ASD symptoms ( Figure 4b and d ) . These findings imply that such atypicality in the temporal characteristics of local neural activity may be one of the basic neuro-aetiologies of autism , which needs to be tested using data collected from much younger children with ASD in future studies . The association between the intrinsic timescales and GMVs is theoretically reasonable . Large GMVs are considered to indicate a high density of neurons in local brain regions ( Kanai and Rees , 2011 ) , which is thought to be accompanied with more synapses ( Cullen et al . , 2010 ) and greater synaptic weights ( Perin et al . , 2011 ) . Moreover , computational studies suggest that such a high neuronal and synaptic density should increase reciprocal connections within the areas and enhance local clustering ( Perin et al . , 2013 ) . Given that spontaneous neural activity largely depends on such recurrent neural networks ( Ikegaya et al . , 2004 ) , resting-state neural activities of brain regions with large GMVs would show more repetition patterns and larger autocorrelations . Although future studies have to directly examine this hypothesis , this logic accounts for the significant correlations between the local neural dynamics and local neuroanatomical structures . Some human neuroimaging researches examined autistic local neural dynamics in the sensory-related areas and reported observations that are consistent with the current findings . For example , fMRI studies that investigated intra-participant variability of brain signals across time found atypically large signal variability in the prefrontal region ( Dinstein et al . , 2011 ) , somato-sensory area ( Haigh et al . , 2015; Dinstein et al . , 2012 ) , auditory area ( Haigh et al . , 2015; Dinstein et al . , 2012 ) , and primary visual cortex ( Dinstein et al . , 2012; Dinstein et al . , 2010; Milne , 2011 ) in individuals with ASD . In particular , one study found that the severity of ASD was significantly correlated with such atypical signal variability in the sensory/visual areas ( Dinstein et al . , 2012 ) . If we can assume that such large variability of local brain signals indicates more random brain activity and consequently yields weak autocorrelations , these previous reports can be interpreted as being consistent with the current findings . In contrast , local neural dynamics in the caudate in autism were poorly understood . In fact , the neuroanatomical association between the subcortical region and the RRB symptoms was reported in previous structural MRI studies ( Langen et al . , 2014; Langen et al . , 2009; Langen et al . , 2007; Hollander et al . , 2005; Schuetze et al . , 2016 ) , which is consistent with the current observation about the caudate . However , to the best of our knowledge , no prior research has been conducted on intrinsic neural timescales or signal variability of the caudate in autism . A recent review has suggested that the intrinsic timescale and TRW is not an artefact of neuroimaging signals but closely associated with an ability of local brain areas to pool , normalise , and complete information ( Himberger et al . , 2018 ) . In particular , for sensory information processing , a longer neural timescale is considered to make brain responses more robust against fluctuations in sensory inputs and enable steady and consistent perception ( Himberger et al . , 2018; Honey et al . , 2012; Murray et al . , 2014 ) . Given this , we speculate that the atypically short intrinsic timescale in the primary sensory/visual cortices observed in the ASD group ( Figures 2b and 4a ) might potentially be one of neural bases of perceptual hyper-sensitivity often seen in autism ( American Psychiatric Association , 2013 ) . This rsfMRI study did not adopt a formulation that has been used to measure neural timescales in previous human fMRI studies ( Hasson et al . , 2015; Hasson et al . , 2008; Lerner et al . , 2011; Stephens et al . , 2013; Yeshurun et al . , 2017; Gauthier et al . , 2012 ) , because this definition of neural timescale — so-called temporal receptive window ( TRW ) — was designed for task-related brain activity data and cannot be directly applied to resting-state fMRI data . Instead , based on previous non-human electrophysiology studies ( Cavanagh et al . , 2016; Murray et al . , 2014; Bernacchia et al . , 2011; Runyan et al . , 2017 ) , we defined the intrinsic neural timescale as the magnitude of autocorrelation of brain activity . In addition , to reduce adverse effects of the low sampling rates of fMRI recording , we did not conduct curve fitting to the autocorrelation coefficients as electrophysiological work did ( Cavanagh et al . , 2016; Murray et al . , 2014; Runyan et al . , 2017 ) ; we simply calculated the area under the autocorrelation function ( ACF ) to estimate the autocorrelation strength ( Figure 1a ) . We validated this definition of the intrinsic neural timescales using the simultaneously recorded EEG-fMRI data . Although the fMRI-based neural timescale was different from the EEG-based one by two orders of magnitude , the two measures were significantly correlated with each other ( Figure 1b and Figure 1—figure supplement 2 ) . Moreover , such a difference would be reasonable because an EEG signal is likely to peak ~100 ms after a stimulus onset and an fMRI signal — a product of neurovascular coupling ( Hillman , 2014; Martindale et al . , 2003 ) — tends to take 5 ~ 10 s to peak ( Logothetis et al . , 2001; Yeşilyurt et al . , 2008 ) . In fact , when we convolved the EEG signals with the hemodynamic response function ( HRF ) to take into account such neurovascular coupling , the resultant intrinsic timescales based on HRF-convolved EEG signals were similar in the magnitude to those based on the fMRI data ( Figure 1c ) . These EEG-fMRI comparisons indicate that the fMRI-based neural timescales represent an aspect of local neuronal activity . However , it is beyond the scope of this study to conclude that such an fMRI-based index reflects the same neuronal phenomena as those seen in previous electrophysiology work that calculated neural timescales from spike activity data ( Cavanagh et al . , 2016; Murray et al . , 2014; Ogawa and Komatsu , 2010; Runyan et al . , 2017 ) . To clarify this issue , future studies have to directly compare the fMRI-based neural timescales with those based on neuronal spike activities that are collected simultaneously with fMRI data . Our exploratory study identified significant associations between local brain dynamics and behavioural tendencies in autism; however , biological mechanisms underlying this brain-behaviour link remain unknown . Previous work proposed that such an atypical neural timescale may represent atypical functional hierarchy in information processing in brains ( Himberger et al . , 2018; Chaudhuri et al . , 2015; Gjorgjieva et al . , 2016 ) . However , it is necessary to directly examine this hypothesis by analysing task-related whole-brain activity in ASD populations . Future work also needs to investigate mechanisms linking these local neural dynamics to atypical large-scale brain dynamics seen in autism ( Watanabe and Rees , 2017 ) . Another limitation of this study is the relative homogeneity of the participants with ASD that we have studied . To improve detectability , we limited the ASD group to high-functioning right-handed adult males . Although we confirmed the main findings in an adolescent dataset ( Figure 4 ) , future studies have to examine the current observations in different subsets of ASD cohorts . This resting-state fMRI study investigated how long local brain areas can store information in individuals with autism and identified a shorter intrinsic timescale in the bilateral primary sensory/visual cortices and a longer intrinsic timescale in the right caudate . This atypicality in local neural dynamics was associated with the severity of autism and also correlated with local grey matter volumes . Although these findings should be examined in larger and more diverse cohorts of individuals with ASD , our work highlights the importance of investigating neural dynamics in neuro-psychiatric disorders .
We examined the validity of the current formulation of the intrinsic neural timescales by comparing simultaneously recorded resting-state EEG and fMRI data that were shared in the Open Science Framework ( Deligianni et al . , 2016; Deligianni et al . , 2014 ) . Simultaneous EEG-fMRI data were collected from 17 healthy adults ( 6 females , 32 . 84 ± 8 . 1 years old ) at UCL under the ethical approval from the UCL Research Ethics Committee and informed consent obtained from all the participants ( Deligianni et al . , 2016; Deligianni et al . , 2014 ) . The data were obtained during rest , in which the participants were asked to open their eyes and remain awake with fixating a white cross on a black background . EEG data were recorded by an MRI-compatible EEG system with 64 channels ( BrainCap MR , Germany ) and were preprocessed in MATLAB ( MathWorks , Inc ) and EEGLAB ( Delorme and Makeig , 2004 ) ( sccn . ucsd . edu/eeglab/ ) . First , the EEG data were referenced to the average of all the electrodes , and downsampled to 250 Hz . After conducting band-pass filtering ( 1–80 Hz ) , an optimal basis set ( OBS ) algorithm based on principal component analysis ( Niazy et al . , 2005 ) was used to reduce the gradient artefacts induced by fMRI scanning . Cardio-ballistic artefacts ( CBAs ) were reduced as follows ( Jamison et al . , 2015; Liu et al . , 2012 ) : the alignment of the occipital CBAs was optimized with individual participant’s heartbeat; then , EEG components that were strongly correlated with the occipital CBAs were identified and excluded by conducting independent component analysis ( ICA ) and calculating mutual information . The remaining EEG artefacts induced by eye blinks , eye movements , and muscle activity were removed by ICA . Next , we excluded epochs whose mean global field power was larger than five standard deviations above the mean across the entire recording . Finally , to identify the source location of these preprocessed EEG signals , we conducted source reconstructions of the remaining ICA-components using DIPFIT2 function implemented in EEGLAB , and obtained MNI coordinates for each of the independent components . To reduce ambiguity in the following fMRI analysis , we excluded independent components whose whole-brain activity patterns did not show clear laterality and thus whose sources were calculated to be in both brain hemispheres . We confirmed that the remaining independent components of EEG data were effectively free from fMRI-oriented gradient noise . In fact , the power spectrums of the EEG data showed that the preprocessing procedures significantly reduced fMRI-induced noise in the EEG signals ( Figure 1—figure supplement 1a and b ) . Therefore , we used these EEG data for the intrinsic timescale analysis . We then filtered the preprocessed data to delta ( 1–4 Hz ) , theta ( 4–8 Hz ) , alpha ( 8–13 Hz ) , beta ( 13–30 Hz ) , and gamma ( 30–80 Hz ) bands , and calculated a Hilbert envelope amplitude for each band wave ( Deligianni et al . , 2014 ) . Using the envelope amplitudes , we estimated an intrinsic timescale in the same manner as used for resting-state fMRI signals ( Figure 1a ) . As for the gamma-band EEG signals , we convolved them with the hemodynamic response function ( HRF ) implemented in SPM12 and calculated the intrinsic timescales for the HRF-convolved EEG data ( Figure 1c ) . The MRI data were collected in a 1 . 5T scanner ( Avanto , Siemens ) with a 12-channel head coil . Functional data were recorded using EPI sequence ( TR 2 . 16 s , TE 30 ms , FA 75° , spatial resolution 3 . 3 mm cubic ) , and T1-weighted structural MRI data were also obtained ( Deligianni et al . , 2016; Deligianni et al . , 2014 ) . For each participant , these MRI data were preprocessed in the same manner as mentioned in the main text and calculated the intrinsic timescale for each voxel . We then calculated the average intrinsic timescales for the brain areas corresponding to each independent component of the EEG data . We defined the brain areas as a 4mm-radius sphere whose centres were determined based on the MNI coordinates obtained in the source reconstructions of the EEG data . Through these analyses , we obtained fMRI-based intrinsic neural timescales and compared them with EEG-based ones using linear regression analyses . This study used datasets shared in ABIDE ( Di Martino et al . , 2014 ) . The main analysis was based on a dataset recorded from 25 high-functioning adults with autism spectrum disorder ( ASD ) and 26 typically developing ( TD ) controls in University of Utah ( Table 1 ) . We chose this dataset because of its largest size of high-functioning adults with ASD . We selected participants based on their age ( ≥18 years old ) , sex ( male ) , handedness ( right-handed ) , IQ ( full/verbal/performance IQ ≥ 80 ) , and head motion during scanning ( mean ≤3 mm ) . We focused on high-functioning right-handed male adults to reduce heterogeneity across individuals with ASD ( Jack and A Pelphrey , 2017 ) . The diagnosis of ASD was made based on structured interviews by a clinical expert for ASD in accordance with ADOS and DSM-IV-TR ( Lord et al . , 1989 ) . IQ was evaluated based on Wechsler Abbreviated Scale of Intelligence . Handedness was scored based on Edinburgh Handedness Inventory . The data collection was approved by the local ethics committees ( University of Utah IRB ) , and all participants provided written consent . Resting-state and anatomical MRI data were collected using a 3 . 0T MRI scanner ( Magnetom Trio , Siemens; resting-state MRI , EPI sequence , TR 2 s , TE 28 ms , 40 slices , interleaved , FA 90° , 3 . 4 × 3 . 4 × 3 . 0 mm; anatomical MRI , T1-weighted sequence , TR 2 . 3 s , TE 2 . 91 ms , FA 9° , 1 . 0 × 1 . 0 × 1 . 2 mm ) . The resting-state MRI data were recorded for ~8 min for each participant , during which the participants were asked to relax with their eyes open . The resting-state MRI data were preprocessed with SPM12 ( www . fil . ucl . ac . uk/spm ) . After discarding the first five images , we performed realignment , unwarping , slice-timing correction , and normalisation to the standard template ( ICBM 152 ) . We then removed effects of head motion , white matter signals , and cerebrospinal fluid signals by regression analyses , and finally conducted band-pass temporal filtering ( 0 . 01–0 . 1 Hz ) . Note that we excluded participants whose mean head motions were more than 3 mm . After this exclusion , there was no significant difference in the mean head motion ( p>0 . 1 in a two-sample t-test ) and maximum/mean framewise displacement ( FD ) ( maximum FD , p>0 . 2; mean FD , p>0 . 4 in a two-sample t-test ) between the TD and ASD groups . At a single-participant level , we used these preprocessed fMRI data to evaluate the intrinsic neural timescale for each voxel as follows . First , we estimated an autocorrelation function ( ACF ) of the fMRI signal of each voxel ( time bin = TR ) , and then calculated the sum of ACF values in the initial period where the ACF showed positive values ( i . e . , the sum of the area of the green bars in Figure 1a ) . The upper limit of this period was set at the point where the ACF hits zero for the first time . After repeating this procedure for every voxel , we applied spatial smoothing to the brain map ( Gaussian kernel , full-width at half maximum = 8 mm ) to improve the signal-to-noise ratio . We used this whole-brain map as an intrinsic timescale map in which the value at each voxel is equal to the intrinsic neural timescale of the brain region . After performing this calculation for all participants , we compared the intrinsic neural timescale maps between ASD and TD groups using a random-effects model . We searched for brain areas showing significant differences between the two groups ( PFDR <0 . 05 ) . For the brain regions whose intrinsic neural timescale was significantly different between the individuals with autism and the TD individuals ( Table 2 ) , we explored associations between their intrinsic neural timescales and the ASD symptoms . Because atypical information processing in autism could be a common basis for various ASD symptoms ( Happé and Frith , 2006; Belmonte et al . , 2004; Watanabe and Rees , 2017 ) , we first calculated Spearman’s correlation coefficients between the average intrinsic timescale in the regions and the overall severity of ASD ( ADOS total score ) . The ADOS total scores were defined by the sum of the ADOS social , ADOS communication , and ADOS RRB scores . The regions of interest ( ROIs ) were defined as clusters found in the whole-brain analysis stated above ( Table 2 ) , and an intrinsic timescale for each ROI was given by the average within the corresponding cluster . The multiple comparisons between these brain regions were corrected by FDR . When we found no significant correlations between the neural timescales and ADOS total scores , we calculated associations with the social and RRB symptoms , respectively . The social symptoms were measured as the sum of the ADOS social scores and ADOS communication scores . The associations with the RRB symptoms were evaluated in one-way ANOVA because the ADOS RRB scores were too sparse for an accurate correlation analysis . To minimise any statistical dependence between the brain-symptom association analysis and the ROI search , we repeated the association analysis by applying the same ROIs to two independent MRI datasets that were not used in the ROI search . The data were collected in ETH Zürich and Indiana University and shared through ABIDE ( Supplementary Table 1 in Supplementary file 1 ) ( Di Martino et al . , 2014 ) . The MRI data were collected under the approval of each local ethics committee in the recording site and with the written informed consent of all the participants . Participants were selected based on the same criteria as those in the main analysis ( age ≥18 years old , sex: male , handedness: right-handed , full/verbal/performance IQ ≥ 80 , and mean head motion ≤3 mm ) . This selection excluded three ASD and nine TD individuals from the entire ETH Zürich dataset , and 11 ASD and 10 TD individuals from the Indiana University dataset . As a result , this reproducibility test analysed 10 ASD and 15 TD individuals for the ETH Zürich dataset and 9 ASD and 10 TD individuals for the Indiana University dataset . After conducting the same preprocessing as in the original analysis , we calculated an intrinsic neural timescale for each voxel , and extracted intrinsic timescales for the four ROIs that were defined in the main analysis ( Rt/Lt postcentral gyrus , Rt IOG , and Rt caudate ) . We then tested for the associations between the neural timescale at these ROIs and the severity of ASD . We examined the main findings from the perspective of neurodevelopment . To this end , we analysed longitudinal MRI data that were collected from 11 high-functioning adolescent children with ASD and seven age-/sex-/IQ-matched TD children in University of California Los Angeles ( two scans for each participant; Supplementary Table 2 in Supplementary file 1 ) . The MRI data were preprocessed in the same manner as in the main analysis , and an intrinsic timescale was estimated for each voxel at each time point in each participant . We then extracted intrinsic timescales for the eight brain regions of interest ( ROIs ) . The ROIs were defined as clusters found in the main MRI analysis ( Table 2 ) , and the neural timescales of the regions were given by the average within the corresponding clusters . For each ROI , we compared developmental trajectories of the intrinsic timescale between ASD and TD groups . In addition , we examined whether such developmental changes in the intrinsic timescale are related to changes in clinical severity of autism . We calculated the intrinsic timescale changes by subtracting the intrinsic timescale at the first scan from that at the second scan . The changes in ADOS scores were quantified in the same manner . We investigated the neuroanatomical bases for intrinsic timescale by comparing grey matter volume ( GMV ) to the temporal property of neural signals . GMV was calculated from structural MRI data using SPM12 as follows: the MRI images were segmented into grey matter , white matter , and cerebrospinal fluid using the New Segment Toolbox ( Ashburner and Friston , 2005 ) ; using the DARTEL Toolbox ( Ashburner , 2007 ) , the segmented grey matter images were aligned , warped to a template space , resampled to 1 . 5 mm isotropic voxels , and registered to a participant-specific template . At a whole-brain level , we first segmented these preprocessed grey matter images into 360 areas according to a recently proposed multi-modal brain parcellation system ( Glasser et al . , 2016 ) , and extracted GMV from each area . By applying the same parcellation system to the whole-brain map of the intrinsic timescale , we calculated average intrinsic timescale for each brain segment . We then averaged these anatomical and functional metrics across participants for each brain segment , which yielded a group-average GMV map and a group-average intrinsic timescale map for each group . By comparing these maps with linear regression analyses , we estimated associations between intrinsic timescale and GMV . Next , we examined this function-anatomy correlation in the brain regions whose intrinsic timescale significantly deviated in autism and showed significant associations with the severity of ASD symptoms . Finally , we performed mediation analyses to investigate correlations between intrinsic timescale , GMV , and severity of ASD after normalising these functional , anatomical , and clinical scores . We examined reproducibility of the main findings with the two independent MRI datasets that were used in the confirmatory brain-symptom associations ( see ‘Confirmation of the brain-symptom associations’ in this Materials and methods section; Supplementary Table 1 in Supplementary file 1 ) . We repeated the same voxel-wise comparison of the intrinsic timescale between the ASD and TD groups , calculated associations between the intrinsic timescale and clinical scores in the ASD group , and assessed correlations with GMV . | Autism is a brain disorder that affects how people interact with others . It occupies a spectrum , with severe autism at one end and high-functioning autism at the other . People with severe autism usually have intellectual impairments and little spoken language . Those with high-functioning autism have average or above average IQ , but struggle with more subtle aspects of communication , such as body language . As well as social difficulties , many individuals with autism show repetitive behaviors and have narrow interests . The brains of people with autism process information differently to those of people without autism . The brain as a whole shows less coordinated activity in autism , for example . But whether individual brain regions themselves also work differently in autism is unclear . Watanabe et al . set out to answer this question by using a brain scanner to compare the resting brain activity of high-functioning people with autism to that of people without autism . In both groups , networks of brain regions increased and decreased their activity in predictable patterns . But in individuals with autism , sensory areas of the brain showed more random activity than in individuals without autism . The most random activity occurred in those with the most severe autism . This suggests that the brains of people with autism cannot hold onto and process sensory input for as long as those of neurotypical people . By contrast , a brain region called the caudate showed the opposite pattern , being more predictable in individuals with autism . The most predictable caudate activity occurred in those individuals with the most inflexible , repetitive behaviors . These differences in this neural randomness appear to result from changes in the structure of the individual brain regions . The findings of Watanabe et al . suggest that changes in the structure and activity of small brain regions give rise to complex symptoms in autism . If these differences also exist in young children , they could help doctors diagnose autism earlier . Future studies should investigate whether the differences in brain activity cause the symptoms of autism . If so , it may be possible to treat the symptoms by changing brain activity , for example , by applying magnetic stimulation to the scalp . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2019 | Atypical intrinsic neural timescale in autism |
Despite the high burden of pain experienced by hospitalised neonates , there are few analgesics with proven efficacy . Testing analgesics in neonates is experimentally and ethically challenging and minimising the number of neonates required to demonstrate efficacy is essential . EEG ( electroencephalography ) -derived measures of noxious-evoked brain activity can be used to assess analgesic efficacy; however , as variability exists in neonate’s responses to painful procedures , large sample sizes are often required . Here , we present an experimental paradigm to account for individual differences in noxious-evoked baseline sensitivity which can be used to improve the design of analgesic trials in neonates . The paradigm is developed and tested across four observational studies using clinical , experimental , and simulated data ( 92 neonates ) . We provide evidence of the efficacy of gentle brushing and paracetamol , substantiating the need for randomised controlled trials of these interventions . This work provides an important step towards safe , cost-effective clinical trials of analgesics in neonates .
Considering the short-term stress and long-term neurodevelopmental impact associated with repeated pain exposure in early life ( Brummelte et al . , 2012; Chau et al . , 2019; Morison et al . , 2001; Vinall et al . , 2014 ) , effective pain relief is crucial in neonatal intensive care ( Hall and Anand , 2014; Lim and Godambe , 2017 ) . Nevertheless , as a result of the implicit challenges of measuring pain in neonates , and the ethical and experimental challenges of conducting neonatal clinical trials , few analgesics have proven efficacy in this population ( Allegaert , 2017; Moultrie et al . , 2017; Slater et al . , 2020 ) . Participants in clinical trials risk exposure to potential adverse effects , and therefore every effort should be made to minimise the sample size necessary to demonstrate efficacy ( European Medicines Agency , 2001 ) . However , as factors such as age ( Fabrizi et al . , 2011; Green et al . , 2019; Hartley et al . , 2016 ) , prior pain experience ( Ozawa et al . , 2011; Slater et al . , 2010a ) , stress ( Jones et al . , 2017 ) , sex ( Bartocci et al . , 2006; Verriotis et al . , 2018 ) , illness ( Ranger et al . , 2013 ) , and behavioural state ( Slater et al . , 2006 ) influence noxious-evoked responses , large sample sizes are often required to account for between-subject variability ( Anand et al . , 2004; Ancora et al . , 2013; Hartley et al . , 2018; Kabataş et al . , 2016; Sindhur et al . , 2020; Taddio et al . , 2006 ) . In adult studies , cross-over trial designs are often used to minimise sample sizes by reducing between-subject variability ( Cooper et al . , 2016 ) . However , this approach may not be appropriate when studying pain in neonates as painful medical procedures can only be performed when clinically necessary and within-subject variables that influence pain can change dramatically across sequential test occasions . One approach used to balance demographic characteristics or other prognostic factors across treatment groups in clinical trials is to stratify neonates across treatment arms and to adjust for these factors in the statistical analysis ( McEntegart , 2003 ) . While this can improve comparability across groups for recognised factors , many unknown variables likely influence pain sensitivity , and a more nuanced approach to account for individual differences in noxious-evoked sensitivity could be more effective in reducing sample sizes . In analgesic studies performed in adults , individual pain thresholds can be identified by applying graded increments of experimental stimulus intensity until pain is reported by the participants . This can be used to stratify treatment groups ( Demant et al . , 2014; Smith et al . , 2017; Vollert et al . , 2017 ) or statistically correct for variability in baseline pain thresholds ( Lane et al . , 2010; Sanga et al . , 2013 ) . In neonates , application of graded non-noxious stimuli such as von Frey hairs has previously been used to identify limb reflex withdrawal thresholds ( Andrews and Fitzgerald , 2002; Andrews and Fitzgerald , 1999; Andrews and Fitzgerald , 1994; Kühne et al . , 2012 ) , but these have not been used as a baseline measure in analgesic clinical trials . In the absence of a validated objective biomarker of pain ( Davis et al . , 2020 ) , electroencephalography ( EEG ) -derived measures of brain activity may provide a valuable surrogate marker of pain by measuring the noxious-evoked activation of the cortex . In adults , brain activity during painful procedures is strongly correlated with verbal reports ( Coghill et al . , 2017 ) ; and in neonates , a template of noxious-evoked brain activity that discriminates between noxious and non-noxious procedures has been previously characterised and validated ( Hartley et al . , 2017 ) . Here , we develop and test an experimental paradigm that assesses individual baseline sensitivity in neonates by measuring noxious-evoked brain activity in response to a low-intensity experimental noxious stimulus , and demonstrate that accounting for this measure of noxious-evoked baseline sensitivity substantially reduces the sample sizes required in neonatal studies of analgesic efficacy . The term ‘noxious-evoked baseline sensitivity’ is used to refer to individual neonate’s noxious-evoked baseline brain activity . This will be related to multiple neural and non-neural factors including nociceptive processing , arousal , attention , signal-to-noise ratio of the EEG recording , and differences in head size . In contrast to studies in adults which have shown similar patterns of activity evoked by both painful and non-painful stimuli ( Mouraux et al . , 2011 ) , we have previously shown that the pattern of brain activity that we analyse here is not evoked by visual , auditory , and tactile stimuli which evoke similar levels of physiological arousal ( Hartley et al . , 2017 ) . In Study 1 , we demonstrate that the magnitude of noxious-evoked brain activity in response to an experimental stimulus correlates with the magnitude of brain activity evoked by the clinical procedure and thus reflects baseline sensitivity . In Study 2 , we use simulated data to demonstrate the increased statistical power that can be achieved by including baseline sensitivity as a covariate when analysing the effect of an intervention in small samples . In Study 3 , we test this novel paradigm using a non-pharmacological pain-relieving intervention of known efficacy – gentle touch – prior to heel lancing . Finally , in Study 4 we investigate the analgesic efficacy of oral paracetamol given prior to immunisation in prematurely born neonates . Overall , we demonstrate that measuring and accounting for noxious-evoked baseline sensitivity could improve the design of analgesic efficacy investigations for this patient population .
We hypothesised that a measure of neonatal noxious-evoked baseline sensitivity could be used to account for inter-individual variability in noxious-evoked brain activity in studies of analgesic efficacy ( noxious-evoked sensitivity paradigm , Figure 1 ) and predicted that this would reduce the sample sizes needed in clinical trials . In order to use noxious-evoked brain activity in response to a mild experimental noxious stimulus as a measure of baseline sensitivity , it must be significantly and strongly correlated with the response to the clinically required procedure . In Study 1 , we therefore assessed the feasibility and initial validity of the paradigm by investigating the relationship between individual responses of term neonates to experimental noxious stimuli and a subsequent clinically required heel lance . This was a retrospective study presenting previously unpublished data from term neonates studied between 2014 and 2015 who had received both a heel lance and experimental noxious stimuli on the same test occasion ( n = 9 ) . Whilst this sample size is small , this was a feasibility study and this relationship is retested in Study 3 . The magnitudes of the noxious-evoked brain activity evoked by stimulating a neonate’s foot with a controlled mild experimental noxious stimulus ( force = 64 mN; magnitude range 0 . 15–0 . 62 , Figure 2A ) were strongly correlated with the magnitude of the noxious-evoked brain activity generated by a clinically required heel lance ( range −0 . 07 to 2 . 34 ) in the same neonates ( p=0 . 0025 , R2 = 0 . 77 , Figure 2A ) . Therefore , application of a mild experimental noxious stimulus prior to performing a clinically required painful procedure could provide a novel measure of neonatal baseline sensitivity , which could be used as a covariate in studies of analgesic efficacy to account for inter-individual variability in pain responses ( Figure 1 ) . In the following sections , we simulate and test the impact of applying this novel paradigm in studies investigating the efficacy of pain-relieving interventions . In Study 2 , we used simulated data to investigate whether accounting for individual differences in noxious-evoked baseline sensitivity has the potential to reduce the sample size needed to assess the efficacy of an analgesic intervention . Here , we initially assume that an effective analgesic intervention results in a 40% reduction in noxious-evoked brain activity; this is clinically meaningful as a similar reduction in noxious-evoked brain activity is observed when adults report significantly lower verbal pain scores ( Lorenz et al . , 1997; von Mohr et al . , 2018 ) . We simulated an Intervention Group and Control Group across a range of sample sizes , simulating both baseline sensitivity data and responses to heel lance , and assuming the relationship between these measures observed in Study 1 . The simulated Control Group and the Intervention Group responses were compared using a linear regression with and without baseline sensitivity as a covariate ( see 'Materials and methods' ) . At a significance level of 0 . 05 , the sample size to achieve a given power is substantially reduced when baseline sensitivity is accounted for ( Figure 2B ) . The reduction in sample size that can be achieved by accounting for individual differences in baseline sensitivity is highly dependent on the anticipated effect size of the intervention ( Figure 2C , D ) . For example , at the extremes we considered , with an assumed intervention effect size of 95% ( and power of 95% ) , a sample of 11 neonates per group would be required without accounting for baseline sensitivity compared with eight neonates per group when baseline sensitivity is accounted for , a 27% reduction in sample size . Whereas , by comparison , assuming an intervention effect size of 5% , the sample size required to achieve 95% power is 5458 neonates per group without accounting for individual baseline sensitivity , compared with 660 neonates per group when baseline sensitivity is accounted for – representing an 88% reduction in sample size ( Figure 2C , D ) . Assuming an intervention effect size of 40% , a sample size of 16 neonates per group ( 32 neonates in total ) would be sufficient to observe a significant intervention effect with 95% power if individual differences in baseline sensitivity are accounted for . In contrast , a sample size of 66 neonates per group ( 132 neonates in total ) is required to achieve the same power if neonatal baseline sensitivity is not accounted for ( Figure 2B ) . The percentage reduction in sample size which can be achieved by accounting for individual differences in baseline sensitivity is also dependent on the strength of the correlation between the brain responses evoked by the experimental stimuli and the acute clinical procedure ( Figure 2E ) . When a weak correlation exists between the measures ( calculated using the standard deviation of the correlation residuals ) , the reduction in sample size is low . Conversely , with a strong correlation between measures , a greater reduction in sample size is achieved . For example , with an assumed intervention effect of 40% and the low standard deviation of the residuals observed in Study 1 ( SD of residuals = 0 . 37 ) , accounting for baseline sensitivity results in a sample size reduction of approximately 76% , compared to a sample size reduction of 17% when a high noise level ( SD of residuals = 1 . 7 ) is observed in the correlation ( Figure 2E ) . In a previous study , we reported that a non-pharmacological gentle touch intervention ( brushing a neonate’s leg at a rate of approximately 3 cm/s to optimally stimulate C-tactile fibres ) prior to a clinically required heel lance caused a 40% reduction in noxious-evoked brain activity ( Gursul et al . , 2018 ) . In Study 3 , we used the same non-pharmacological intervention in an independent prospective cohort of healthy neonates that clinically required a heel lance for the purpose of blood sampling and tested the effect of incorporating the noxious-evoked baseline sensitivity paradigm and accounting for inter-individual differences in baseline sensitivity . Based on power calculations from simulated data in Study 2 , assuming a 40% reduction in noxious-evoked brain activity from the intervention and 95% power , a total of 16 neonates were included in the Intervention Group and were gently brushed on the leg ipsilateral to the stimulus site at a rate of approximately 3 cm/s for 10 s prior to heel lancing ( Gursul et al . , 2018 ) . A further 15 neonates were included in the Control Group where the heel lance was performed without gentle brushing . All neonates received mild experimental noxious stimulation prior to heel lancing to assess their individual baseline sensitivity ( see 'Materials and methods' ) . Unlike Study 1 , in which neonates had been stimulated with a force of 64 mN , a force of 128 mN was applied in this prospective cohort to increase the signal-to-noise ratio . The necessary strong correlation between the evoked response to the experimental stimulus and clinical procedure was confirmed in the Control Group ( p=0 . 0013 , R2 = 0 . 65 , Figure 3A ) . Consistent with the previously published study ( Gursul et al . , 2018 ) , the gentle touch intervention resulted in a 39% reduction in the magnitude of the noxious-evoked brain activity , but a significant intervention effect was not observed ( although the result indicated borderline significance ) when baseline sensitivity was not accounted for , likely due to the lack of power with this sample size ( linear regression , t = 1 . 95 , p=0 . 05 , Figure 3B , Study 2 indicates a power of 40% for a sample of this size without accounting for baseline sensitivity , Figure 2B ) . However , when noxious-evoked baseline sensitivity was accounted for as a covariate in the analysis , a significant intervention effect was observed ( linear regression , t = 2 . 29 , p=0 . 026 ) . To further understand these results , we compared the relationship between the responses within the Control Group and Intervention Group . Unlike the significant correlation between the magnitude of noxious-evoked brain activity in response to the experimental noxious stimuli and heel lancing demonstrated in the Control Group ( p=0 . 0013 , R2 = 0 . 65 , Figure 3A ) , this relationship was disrupted in the Intervention Group ( p=0 . 39 , R2 = 0 . 05 , Figure 3C ) . In particular , we observed reduced noxious-evoked brain activity following the gentle brushing intervention in neonates with high baseline sensitivity ( Figure 3D ) , suggesting that the effect of pain-relieving interventions is most prominent in neonates with greater noxious-evoked baseline sensitivity . Noxious stimulation in neonates evokes a range of physiological responses including facial grimacing , reflex withdrawal , and physiological responses ( Cornelissen et al . , 2013; Hartley et al . , 2015; Hatfield and Ely , 2015 ) . There is great value in establishing whether accounting for individual differences in baseline sensitivity can be applied to other pain-related measures . In Study 3 , the magnitude of the reflex withdrawal was also recorded in response to the experimental noxious stimulation and heel lancing . In the Control Group , the magnitude of the reflex withdrawal response to experimental noxious stimulation was significantly correlated with the reflex withdrawal evoked by heel lancing ( p=0 . 009 , R2 = 0 . 36 , Figure 4A ) . However , this correlation in reflex withdrawal activity was weaker than the relationship in the noxious-evoked brain activity , limiting its use ( Figure 2E ) . Assuming an intervention effect of 40% and this level of correlation identified within the same size sample , simulated data reveals that accounting for baseline sensitivity using noxious-evoked reflex activity provides only 17 . 3% power to detect a significant difference between the two groups compared with a power of 11 . 3% without accounting for baseline sensitivity . In this study , the gentle touch intervention did not significantly reduce the magnitude of the reflex withdrawal activity following heel lancing , either when accounting for baseline sensitivity ( linear regression , t = −1 . 43 , p=0 . 17 ) or without accounting for baseline sensitivity ( t = −1 . 73 , p=0 . 10 , Figure 4B ) . While it is possible that reflex withdrawal of the stimulated limb is not modulated by gentle touch , as has previously been suggested ( Gursul et al . , 2018 ) , the intervention clearly disrupted the correlation between baseline reflex sensitivity and the reflex evoked by heel lancing ( p=0 . 25 , R2 = 0 . 1 , Figure 4C ) . The brushing intervention may have caused a change in baseline muscle activity in some individuals resulting in the larger residuals in the Intervention Group ( SD of the residuals – Control Group 15 . 3 µV; Intervention Group 26 . 2 µV ) . In Study 4 , we conducted an opportunistic study to investigate whether the administration of paracetamol prior to immunisation significantly reduces noxious-evoked brain activity . In 2015 , national clinical guidelines recommended the administration of paracetamol at the time of meningitis B immunisation due to its antipyretic effect ( NHS England and Public Health England , 2015 ) . Therefore , our local neonatal unit ( John Radcliffe Hospital ) began administering oral paracetamol to neonates immediately after vaccination . In October 2018 , the local practice guidelines were updated , recommending the administration of oral paracetamol 1 hr pre-vaccination . Prior to the guideline change , we studied 16 neonates who did not receive paracetamol before immunisations ( Control Group ) , recording their noxious-evoked brain activity during immunisations . Following the guideline change , we recorded noxious-evoked brain activity in 16 neonates who received paracetamol 1 hr prior to immunisations ( Intervention Group ) ( see 'Materials and methods' and Figure 5A ) . In the Intervention Group , we explored the relationship between noxious-evoked baseline sensitivity and brain activity evoked by immunisation following paracetamol administration . Noxious-evoked brain activity in response to immunisation was characterised using a fast frame rate video camera to identify the time when the needle first came into contact with the skin ( Hartley et al . , 2014; Verriotis et al . , 2016 ) . For each neonate , up to three immunisations were recorded on the same test occasion . First , we validated the use of the template of noxious-evoked brain activity ( Hartley et al . , 2017 ) to quantify the magnitude of noxious-evoked brain activity from immunisation applied to the thigh ( see Methods to validate the template of noxious-evoked brain activity , Figure 5—figure supplement 1 and Figure 5—figure supplement 2 ) . The magnitude of noxious-evoked brain activity following immunisation was significantly lower in the neonates who received paracetamol prior to vaccination ( linear mixed effects regression model with subject and number of immunisations set as random effects , Control Group mean 0 . 88 [SD 0 . 58] ( n = 15 ) ; Intervention Group mean 0 . 40 [SD 0 . 30] ( n = 14 ) , t = 3 . 61 , p<0 . 001 , Figure 5B ) . In a subset of 11 of the 16 neonates in the Intervention Group , who received paracetamol prior to immunisation , we also recorded responses to experimental noxious stimulation before and approximately 1 hr after paracetamol administration ( Intervention Group subset , Figure 5A ) . As this study was implemented opportunistically following changes in clinical guidelines , responses to experimental noxious stimuli were not recorded in all neonates . Nevertheless , the baseline sensitivity measures that were recorded in response to the experimental noxious stimuli applied prior to paracetamol administration had a range of values that were similar to Studies 1 and 3 ( range: 0 . 09–0 . 77 ) . Likewise , the magnitude of the brain activity evoked by the immunisations was similar to that evoked by heel lance in the previous studies for both the Intervention Group ( Intervention Group [Study 3] range −0 . 06 to 1 . 48; Intervention Group [Study 4] range −0 . 08 to 1 . 22 ) and the Control Group ( Control Group [Study 3] range 0 . 30–2 . 84; Control Group [Study 4] range −0 . 07 to 2 . 14 ) . Although we did not record the baseline sensitivity in neonates in the Control Group , in the absence of a pain-relieving intervention , we would expect the response to be correlated with noxious-evoked brain activity evoked by immunisation . As the correlation between baseline sensitivity and response to immunisation was low in the Intervention Group ( p=0 . 12 , R2 = 0 . 33 , n = 9 , Figure 5C ) , the relationship between these measures was likely disrupted by paracetamol . Similar to the gentle brush intervention , the neonates with high baseline sensitivity , represented by a high magnitude response to experimental noxious stimulation prior to paracetamol administration , had much lower magnitude responses to immunisation than would have been expected without an analgesic intervention ( Figure 5C ) . Similarly , the correlation between the baseline sensitivity ( magnitude of the noxious-evoked baseline sensitivity prior to paracetamol administration ) and the response to the experimental noxious stimuli 1 hr post-paracetamol administration was disrupted ( p=0 . 83 , R2 = 0 . 006 , n = 9 , Figure 5D ) . There was no significant difference in the responses to the experimental noxious stimuli before and after paracetamol administration ( linear regression , before paracetamol mean: 0 . 27 [SD 0 . 38]; after paracetamol mean: 0 . 27 [SD 0 . 35] , t = 0 . 17 , p=0 . 86 , n = 9 ) but we were likely not powered to observe an effect .
We demonstrate that accounting for individual differences in noxious-evoked baseline sensitivity significantly reduces the sample size required to assess the efficacy of analgesics in neonates . Noxious-evoked brain activity in response to a low-intensity experimental noxious stimulus can be used in neonates as a marker of baseline sensitivity and is highly correlated with the magnitude of noxious-evoked brain activity produced by clinically required acute painful procedures . Using both simulated and experimental data , we demonstrate that the sample size required to observe the effects of analgesic interventions ( for a given power and significance level ) can be significantly reduced when noxious-evoked baseline sensitivity is accounted for . Importantly , the percentage reduction in sample size is related to the expected effect size of the intervention and the degree of correlation between the baseline sensitivity measure and the brain activity evoked by the clinical procedure . By testing this novel paradigm in clinical studies , we re-confirm the efficacy of gentle touch as a non-pharmacological intervention that reduces brain activity evoked by heel lancing ( Gursul et al . , 2018 ) and we provide evidence to suggest that oral paracetamol is a candidate analgesic drug for procedural pain in neonates . Although these studies have a number of limitations ( including lack of randomisation ) and only investigate one aspect of the neonatal response to noxious input ( namely an EEG-derived noxious-evoked potential ) , they provide strong evidence to suggest that randomised clinical trials investigating the efficacy of both gentle touch and paracetamol through multi-modal pain assessment measures are warranted . Minimisation of sample sizes is imperative in clinical research , and particularly in neonatal studies given the inherent ethical , recruitment , and experimental challenges associated with studying this patient population . Considering that inter-individual variability drives increases in sample sizes required to demonstrate efficacy , addressing baseline variability is key . The paradigm we present here likely accounts for multiple factors affecting noxious-evoked baseline sensitivity in neonates including potential effects from prior pain exposure during hospitalisation ( Grunau et al . , 2001; Johnston and Stevens , 1996 ) and prematurity ( Slater et al . , 2010a ) . This provides a robust approach to indirectly control for a vast array of known and unknown demographic and environmental factors that influence noxious-evoked brain activity and result in inter-individual variability in responses , as well as potential experimental confounds which differ between individuals ( such as differences in signal-to-noise ratios , head circumference , and skull thickness ) . Responses to other modalities such as visual , auditory , or tactile stimuli could be used to obtain a measure of baseline sensitivity , and background resting state brain activity is also predictive of individual noxious-evoked responses ( Baxter et al . , 2021 ) . However , the aim of this study was to develop an experimental paradigm that accounts for the maximum variability in responses to acute painful procedures , to maximise the power to detect a true effect of an intervention . Applying an experimental noxious stimulus to obtain a measure of baseline sensitivity optimises the model as it optimally matches the main characteristics of a response that would be evoked by a procedure of clinical interest ( e . g . the stimulus can be applied to the same body location , evokes noxious activity as well as other sensory-related brain activity , and it is measured at the same electrode site ) . Given the study aim was to reduce sample size in studies investigating acute pain , it is most appropriate to adjust for inter-individual variability using measures of noxious-evoked baseline sensitivity . In contrast , if alternative studies considered other sensory modalities , for example , visual processing , then a better measure of inter-individual baseline variability would be achieved using a visual stimulus . The experimental noxious stimulus used in these studies provides a practical and ethical paradigm for the assessment of baseline sensitivity in neonates . It is non-tissue damaging in both term and ex-premature neonates , activates A∂ and C fibres ( van den Broeke et al . , 2015 ) , does not evoke changes in facial expression or signs of behavioural distress ( Goksan et al . , 2015; Hartley et al . , 2017; Hartley et al . , 2015 ) , and is acceptable to parents . The application of experimental noxious stimuli provides a reliable measure of baseline sensitivity as the mild stimuli can be repeated and trial averages calculated within individual neonates; an approach that substantially reduces the signal-to-noise ratio as compared with responses recorded in response to a single clinical procedure . Moreover , there is no evidence from our data that the experimental noxious stimuli increase the magnitude of the heel lance response given that the responses to heel lance reported here are similar to previous papers where the experimental noxious stimuli were not applied ( Hartley et al . , 2017 ) , suggesting it is appropriate for use in a clinical setting . Despite the advantage of using this approach , we cannot rule out the potential effects of selection bias ( Bishop , 2020 ) . A relatively high number of trials were rejected due to artefacts , which may be more pronounced when there are stimulus-related movements . If these movements are indicative of a more vigorous response to the noxious input , then it is plausible that we are unavoidably biased towards a subset of the population . The applicability of the noxious-evoked baseline sensitivity paradigm was tested in the context of a pain-relieving intervention that we have previously shown to be effective in reducing noxious-evoked brain activity – gentle touch ( Gursul et al . , 2018 ) . Neonates were gently brushed at a speed of 3 cm/s , which is approximately equivalent to the rate at which parents will naturally stroke their neonates ( Croy et al . , 2016 ) and optimises stimulation of C-tactile fibres ( Löken et al . , 2009 ) . In an independent population of neonates , we re-confirmed that brushing the skin prior to a clinically required heel lance significantly reduces noxious-evoked brain activity . We used our noxious-evoked baseline sensitivity paradigm to indirectly account for many factors that influence the magnitude of noxious-evoked brain activity . In addition , we did not observe a significant difference in reflex withdrawal activity between the control neonates and the neonates who received gentle touch prior to the heel lance , which is consistent with our previous observations ( Gursul et al . , 2018 ) . It is possible that either the magnitude of the reflex withdrawal is genuinely not modulated by the brush intervention or that a modulation in reflex activity would only be observed with a larger sample size . Importantly , a significant but weak correlation was observed between the reflex activity in response to the noxious stimuli and in response to heel lancing in the Control Group , suggesting that the paradigm presented here could be useful in future trials where reflex withdrawal activity is used as an outcome measure . As pain perception is a highly complex sensory and subjective emotional experience generated in the brain ( IASP , 2020 ) , quantifying noxious-evoked brain activity may represent a better proxy pain measure , and a more sensitive marker of analgesic efficacy , compared with reflex signals generated by the spinal cord . In addition to minimising sample sizes , assessing baseline sensitivity may also allow for identification of neonates that would benefit most from analgesic interventions . Neonates with larger noxious-evoked baseline sensitivity had the greatest reduction in response following the intervention . In contrast , neonates with low baseline sensitivity were less likely to demonstrate a benefit of the intervention , as for this clinical procedure the potential reduction in their responses was minimal . This could be due to a floor effect whereby for some neonates noxious-evoked brain responses to heel lance is close to zero and cannot be reduced further . Improving our understanding of inter-individual variability in pain-related responses is pivotal to ensure that for each individual neonate potential adverse effects of analgesics are carefully weighed against potential benefits . In our final study , we demonstrate that paracetamol significantly reduced the magnitude of the noxious-evoked brain activity following immunisation compared with neonates who did not receive paracetamol prior to immunisation . Although this result is consistent with studies in adults , using both EEG ( Bromm et al . , 1992; Pickering et al . , 2013 ) and fMRI ( Pickering et al . , 2015 ) , where paracetamol has been shown to reduce brain activity evoked by noxious procedures , previous studies in neonates have provided insufficient evidence to determine the analgesic efficacy of paracetamol for acute procedural pain ( Ohlsson et al . , 2020 ) . While several studies suggest an opioid-sparing effect of paracetamol ( Ceelie et al . , 2013; Härmä et al . , 2016 ) and reduced need for pain relieveing interventions ( Höck et al . , 2020 ) , the majority of studies do not demonstrate a reduction in behavioural or physiological responses to commonly performed painful procedures , such as heel lancing ( Badiee and Torcan , 2009; Bonetto et al . , 2008; Shah et al . , 1998 ) and invasive eye examinations to screen for retinopathy of prematurity ( Seifi et al . , 2013 ) . The behavioural outcome measures used in these studies may fail to discriminate between pain and distress ( Moultrie et al . , 2017; Slater , 2019 ) , which could limit conclusions related to analgesic efficacy . However , given the small sample size of the present study and that we are only characterising the immediate noxious-evoked brain activity following the needle insertion , rather than the activity associated with the injection of the fluid into the muscle for example , randomised clinical trials that also include other pain-related measures are warranted to assess the benefit of paracetamol administration prior to immunisation . Nonetheless , small studies in adults demonstrate that candidate drugs can modulate pain-related neural activity in the absence of verbally reported analgesia , and these brain-derived measures are recognised as a valuable approach to obtain objective evidence related to potential analgesic efficacy in early proof of concept studies ( Wanigasekera et al . , 2018 ) . The noxious-evoked brain activity measure used here quantifies the evoked potential produced at the central vertex electrode site ( Cz ) , which has been shown to have the greatest and most reproducible response size amplitude compared to other electrodes sites across the brain ( Hartley et al . , 2017; Verriotis et al . , 2016 ) . This measure does not represent all nociceptive activity across the brain and cannot be used to investigate the various aspects of pain perception ( Hartley et al . , 2017 ) ; a multi-modal approach to pain assessment is therefore important in follow-on studies ( van der Vaart et al . , 2019 ) . However , in the absence of verbalisation , neuroimaging methods provide an objective proxy approach which has been used to infer pain perception following noxious events ( Baxter et al . , 2021; Duff et al . , 2020; Gursul et al . , 2019; Hartley et al . , 2017 ) . Paracetamol is administered as an antipyretic for the meningitis B immunisations . An update to our local clinical guidelines was implemented , whereby the paracetamol was administered prior to rather than after immunisation . This meant we were able to opportunistically study whether paracetamol can reduce noxious-evoked brain activity following immunisation . Our study is significantly limited due to the restricted sample sizes , lack of randomisation and blinding , and because in the Control Group , where paracetamol was administered after immunisation , we did not record baseline sensitivity prior to immunisation . Therefore , we do not have data to confirm that the baseline sensitivity is correlated with the magnitude of the evoked brain activity following immunisation; although , given there is no discernible correlation between these measures in the Intervention Group , this strongly suggests that this relationship has been disrupted by paracetamol administration . Furthermore , for neonates with high baseline sensitivity , the brain responses evoked by the immunisation were much lower than would be expected in the absence of the analgesic intervention . To broaden the utility of this paradigm , it will be necessary to characterise the correlation between baseline sensitivity and a range of acute clinical procedures , including immunisation . Although many factors that influence individual variability in responses are accounted for using our noxious-evoked sensitivity paradigm , it does not account for differences in rapidly fluctuating state effects such as differences in attention or sleep state that could vary between the baseline sensitivity testing and the implementation of the clinical procedure . Understanding how state differences influence variability in noxious-evoked responses will facilitate better estimation of the expected responses to clinically required painful procedures . A recent fMRI study demonstrated that noxious-evoked brain activity can be predicted from a neonate’s resting state brain activity as well as the structural integrity of key white matter pathways ( Baxter et al . , 2021 ) . Investigating the role of baseline EEG activity and exploring the neurological differences underlying variability in the noxious-evoked brain activity described here could further improve the utility of the paradigm . In addition , while our paradigm is applicable to many of the most common acute somatic painful procedures which neonates are exposed to including heel lancing , cannulation , and injections , this paradigm may not apply to many types of pain such as visceral pain , post-operative pain , longer procedures , such as retinopathy of prematurity screening , procedures with a slow-rising onset , or chronic pain . In summary , the assessment of pain in non-verbal neonates is challenging ( Slater , 2019 ) and the wide variability in individual responses to painful procedures complicates the assessment of analgesics . Currently there is a paucity of evidence regarding the efficacy of pain-relieving interventions used in neonatal practice ( Baarslag et al . , 2017 ) . Here , we present a paradigm that accounts for individual noxious-evoked baseline sensitivity and we demonstrate its utility in terms of sample size reduction . Using this paradigm in clinical trials could optimise resources , maximise the value of collected data , and ultimately expedite the discovery and validation of urgently needed analgesics for this patient population .
A total of 92 neonates were included in four studies . In Study 1 , the relationship between responses to experimental noxious stimuli and clinically required heel lance was investigated in nine neonates using unpublished data previously collected for other studies . In Study 2 , the potential value of the statistical relationship identified in Study 1 was investigated using computational simulations . In Study 3 , brain activity and reflex withdrawal responses from 38 neonates were recorded to test the paradigm with gentle touch as a pain-relief intervention . In Study 4 , brain activity was recorded from 29 neonates in response to immunisations to test the analgesic efficacy of paracetamol . Additionally , the brain-derived measures to characterise immunisation-evoked activity were validated in a further 16 neonates . The participants were recruited from the Maternity Unit and Newborn Care Unit at the John Radcliffe Hospital , Oxford University Hospitals National Health Service Foundation Trust , Oxford , UK . Medical charts were reviewed , and neonates assessed as clinically stable , not receiving analgesics at the time of study ( except from paracetamol where specified ) , and with no history of neurological problems or maternal substance abuse were eligible for inclusion . Participant demographics are presented in Table 1 . The estimate of cumulative prior pain exposure was quantified from each neonate’s clinical records as the total number of acute skin-breaking procedures ( including heel lances , venepuncture , and intravenous cannulations ) and aspirations ( oropharyngeal or endotracheal ) from time of birth to time of study ( Hartley et al . , 2016 ) . These procedures were chosen based on a prospective epidemiology study describing the most commonly performed clinical procedures neonates are exposed to during hospitalisation ( Carbajal et al . , 2008 ) . Studies were conducted in accordance with the Declaration of Helsinki and Good Clinical Practice guidelines . Ethical approval was obtained from the National Research Ethics Service ( reference 12/SC/0447 ) and informed written parental consent was obtained prior to each study . Electrophysiological activity was recorded from DC to 400 Hz using a SynAmps RT 64-channel headbox and amplifiers ( Compumedics Neuroscan ) . CURRYscan7 neuroimaging suite ( Compumedics Neuroscan ) was used to record activity , with a sampling rate of 2000 Hz . EEG was recorded from eight locations on the scalp ( Cz , CPz , C3 , C4 , Oz , FCz , T3 , T4 ) , with reference at Fz and ground at Fpz ( forehead ) according to the modified international 10–20 system . Preparation gel ( Nuprep gel , D . O . Weaver and Co . ) was used to gently clean the scalp with a cotton bud before disposable Ag/AgCl cup electrodes ( Ambu Neuroline ) were placed with conductive paste ( Elefix EEG paste , Nihon Kohden ) . In Study 3 , surface EMG was recorded from the limb ipsilateral to the site of stimulation . Bipolar EMG electrodes ( Ambu Neuroline 700 solid gel surface electrodes ) were placed on the bicep femoris muscle . EEG signals were filtered from 0 . 5 to 30 Hz with a notch filter at 50 Hz . Epochs were extracted 500 ms before the stimulus and 1000 ms after and were baseline-corrected to the pre-stimulus mean . Epochs were rejected if they contained gross movement artefact . Noxious-evoked brain activity was analysed at the Cz electrode for all trials ( as this is the electrode site at which the maximal evoked response is observed; Hartley et al . , 2017 ) . The previously validated template of noxious-evoked brain activity was projected onto each individual trial in the time window of interest ( 400–700 ms after stimulation when the stimulus was applied to the foot , 300–600 ms after stimulation when the stimulus was applied to the thigh – see Methods to validate the template of noxious-evoked brain activity ) providing a weight indicating the magnitude of the noxious-evoked brain activity ( Hartley et al . , 2017 ) . Each individual trial was first Woody filtered in the time window of interest to achieve maximum correlation with the template , accounting for individual differences in the latency to the response . A maximum jitter of ±50 ms for the experimental noxious stimuli and ±100 ms for the heel lance and immunisations was used for the Woody filtering . In Study 4 , additional variation in the latency of the response occurred from the use of the high-speed video camera . To account for this , responses were first Woody-filtered within a neonate to achieve maximum correlation with the within-subject average . EMG signals were filtered 10–500 Hz , with a notch filter at 50 Hz and harmonics , and rectified . Epochs were extracted from 2 s prior to 4 s after the stimulus . Individual epochs were rejected due to movement artefact in the baseline period . The data was split into 250 ms windows and the root mean square ( RMS ) of the reflex signal was calculated in each window . The average RMS across the first four windows after the stimulus ( first second after stimulation ) was calculated as the magnitude of the reflex withdrawal . Statistical analysis was performed using MATLAB_R2020a ( MathWorks ) . Linear associations were assessed using Pearson correlation tests in Studies 1 , 3 , and 4 . Statistical significance ( alpha<0 . 05 ) was assessed non-parametrically via permutation testing with 10 , 000 permutations using the PALM ( permutation analysis of linear models ) toolbox ( Winkler et al . , 2014 ) . Group differences in Studies 2 , 3 , and 4 were assessed using linear regressions ( unpaired two-sample t-tests , except for the differences in responses to the experimental noxious stimuli before and after paracetamol administration where a paired sample t-test was used ) . When using the paradigm to adjust for baseline sensitivity , we used the following linear regression model: Y^=b0+ b1X1+b2X2 , where Y^ is the magnitude of the response to the clinical procedure , b0 is the intercept , X1 is the intervention , and X2 is the baseline sensitivity . Without accounting for baseline sensitivity , the model used was Y^=b0+ b1X1 . Statistical significance ( alpha<0 . 05 ) in Studies 3 and 4 group comparisons was assessed non-parametrically via permutation testing with 10 , 000 permutations using PALM . The difference between the Intervention and Control Group in Study 4 was assessed using a linear mixed effects model , with subject and number of immunisations set as random effects . Two-sided tests were used for all statistical analyses with a significance level of 0 . 05 . | Hospitalized newborns often undergo medical procedures , like blood tests , without pain relief . This can cause the baby to experience short-term distress that may have negative consequences later in life . However , testing the effects of pain relief in newborns is challenging because , unlike adults , they cannot report how much pain they are experiencing . One way to overcome this is to record the brain activity of newborns during a painful procedure and to see how these signals are modified following pain relief . Randomized controlled trials are the gold standard for these kinds of medical assessments , but require a high number of participants to account for individual differences in how babies respond to pain . Finding ways to reduce the size of pain control studies could lead to faster development of pain relief methods . Here , Cobo , Hartley et al . demonstrate a way to reduce the number of newborns needed to test potential pain-relieving interventions . In the experiments , the brain activity of nine babies was measured after a gentle poke and after a painful clinically required procedure . Cobo , Hartley et al . found that the babies’ response to the gentle poke correlated with their response to pain . Further data analysis revealed that this information can be used to predict the variability in pain experienced by different newborns , reducing the number of participants needed for pain relief trials . Next , Cobo , Hartley et al . used this new approach in two pilot tests . One showed that gently stroking an infant’s leg before blood is drawn from their heel reduced their brains’ response to pain . The second showed that giving a baby the painkiller paracetamol lessened the brain’s response to immunisation . The new approach identified by Cobo , Hartley et al . may enable smaller studies that can more quickly identify ways to reduce pain in babies . Furthermore , this work suggests that gentle brushing and paracetamol could provide pain relief for newborns undergoing hospital acute procedures . However , more formal clinical trials are needed to test the effectiveness of these two strategies . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"medicine",
"neuroscience"
] | 2021 | Quantifying noxious-evoked baseline sensitivity in neonates to optimise analgesic trials |
Seasonal migration is a taxonomically widespread behaviour that integrates across many traits . The European blackcap exhibits enormous variation in migration and is renowned for research on its evolution and genetic basis . We assembled a reference genome for blackcaps and obtained whole genome resequencing data from individuals across its breeding range . Analyses of population structure and demography suggested divergence began ~30 , 000 ya , with evidence for one admixture event between migrant and resident continent birds ~5000 ya . The propensity to migrate , orientation and distance of migration all map to a small number of genomic regions that do not overlap with results from other species , suggesting that there are multiple ways to generate variation in migration . Strongly associated single nucleotide polymorphisms ( SNPs ) were located in regulatory regions of candidate genes that may serve as major regulators of the migratory syndrome . Evidence for selection on shared variation was documented , providing a mechanism by which rapid changes may evolve .
Bird migration is a fascinating and highly variable behaviour that integrates many traits – morphological , physiological and behavioural . Research on a wide range of species has provided important insight into this behaviour , from the incredible distances that birds cover during their journeys ( Alerstam et al . , 2003; Egevang et al . , 2010 ) , to the fine-tuned and precisely controlled timing of migration ( Gwinner and Helm , 2003 ) and the fascinating sensory modalities that allow birds to navigate with amazing precision ( Mouritsen , 2018; Wiltschko and Wiltschko , 1972 ) . The European blackcap Sylvia atricapilla is an iconic migratory species that is well suited to work on the genetics of migration . Blackcaps exhibit dramatic differences in migratory behaviour ( Figure 1a ) , spanning the entire spectrum from exclusively migratory populations in the northern portion of their range to short distance and partially migratory populations in the Mediterranean; non-migratory , or resident , populations occur on both the European continent ( Iberian Peninsula ) and the Atlantic islands . In addition to variation in the propensity to migrate and the distance covered , blackcaps vary in migratory orientation , with a migratory divide ( contact zone between populations that breed adjacent to one another but take different migratory routes ) occurring between populations that migrate southwest ( SW ) and southeast ( SE ) from their breeding grounds in Central Europe in the autumn . A novel migratory route also evolved very recently in this species , with an increasing number of birds migrating northwest ( NW ) from the Central European breeding grounds in the autumn ( Cramp , 1992 ) . Variation in the migratory behaviour of European blackcaps was harnessed in a series of influential papers published in the 1980s and 1990s that detailed the genetic basis of migration . Common garden experiments showed that selectively bred individuals that were reared in isolation from their parents maintain population-specific behaviour , suggesting that there is a genetic component to migration ( Helbig , 1994; Helbig et al . , 1989; Helbig , 1991a; Berthold and Querner , 1981; Pulido and Berthold , 2010 ) . F1 hybrids crossbred between populations that differ in migratory traits exhibited intermediate phenotypes ( orientation , distance and propensity to migrate ) , suggesting that these traits are additively inherited ( Berthold and Querner , 1981; Pulido and Berthold , 2010; Helbig , 1991b ) . Further work with F2 hybrids showed a wider distribution of phenotypes and the recovery of parental phenotypes , indicative of traits that are controlled by only a few major genes ( Helbig , 1996 ) , and selection experiments mating birds according to migratory status showed that the transition between resident and migratory behaviour can occur in just a few generations ( Pulido et al . , 1996; Berthold et al . , 1990 ) . The rapidity with which migratory behaviour can evolve has been supported in natural populations; the NW route taken by some birds was only established in the past 70 years and probably in response to increased food availability during the winter in the United Kingdom ( Berthold et al . , 1992 ) . The blackcap has also been the subject of extensive phylogeographic study . Pérez-Tris et al . ( 2004 ) used mitochondrial ( control ) data from 241 birds and 12 populations across the entire breeding range to show that migratory variation in this species arose recently ( 4 , 000–13 , 000 years ago [ya] ) and has not yet resulted in significant population differentiation . These results could suggest that the genes that control migratory variation have small effect sizes or are restricted to a small portion of the genome . The only populations showing substantial genetic differentiation occurred in the Central European migratory divide ( i . e . , between SW and SE migrants ) , indicating that differences in orientation may help to maintain population differentiation . Resident populations showed evidence of historical bottlenecks followed by sudden expansions , suggesting that blackcaps lost their ability to migrate after secondary colonization of mild areas in southern Europe and on the Atlantic islands . This finding was supported by Voelker and Light ( 2011 ) who used mitochondrial ( ND2 and cytb ) data to reconstruct ancestral states within the genus Sylvia . Limited genetic differentiation between blackcaps was also documented by Dietzen et al . ( 2008 ) using mitochondrial ( cytb ) data; these authors also estimated dates for the colonization of Atlantic islands and for an earlier colonisation of the Canary Islands ( the latter occurring 300 , 000–3 , 000 , 000 ya vs . 4 , 000–40 , 000 ya for colonisation of other islands including the Azores and Cape Verde ) . The phylogeographic studies described above provided important insight into the evolution of migration in blackcaps and other temperate species more generally . When these studies and experimental work are considered together , the thorough set of studies conducted on blackcap migration is arguably unequalled in other species . Surprisingly , these classic experiments and molecular marker-based approaches have been followed by a dearth of genetic work on migration in blackcaps . Here , we leverage our knowledge of this excellent study system by using high-throughput sequencing techniques to provide the first genome-wide characterization of the blackcap . The major objectives of this study were to assemble a high-quality reference genome de novo , and to use whole genome resequencing data from 110 blackcaps ( including birds from each migratory phenotype and encompassing the entire breeding range , Figure 1a ) to ( 1 ) examine population structure and demography in this system , and ( 2 ) study the genetic basis of three migratory traits in unison: the propensity to migrate , migratory distance and orientation . Analyses of population structure and demography revealed novel insights that are important for understanding both the evolutionary history of migration in blackcaps and the underlying genetics of this behaviour . A small number of studies on the genomics of migration have been conducted in songbirds ( Delmore et al . , 2016; Lundberg et al . , 2017 ) . We compare our results to theirs , evaluating a long held hypothesis of a common genetic basis to migratory behaviour ( Liedvogel et al . , 2011 ) . Our results are not only relevant to understanding the genetics of migration in the blackcap . Data on the genetics of complex behaviours is at a premium in the evolutionary literature , which has focused primarily on morphological traits , and migration probably plays an important role in the early stages of speciation in many systems . Our results will speak to the genetic basis of this process .
We used whole genome sequencing ( WGS ) data and an optical map ( Illumina and Bionano Irys technology , Supplementary files 1–3 ) to de novo assemble a hybrid reference genome for the blackcap ( BioProject number PRJNA545868; Guojie Zhang , personal communication ) . The final genome is 1 . 02 Gb in length , comprised of only 96 scaffolds and has a large N50 scaffold length of 22 Mb . Ninety scaffolds mapped to the collared flycatcher Ficedula albicollis genome ( average three scaffolds/chromosome; Supplementary file 3 ) and our annotation strategy , which used both in silico and evidence-based approaches , identified 17 , 982 protein-coding genes . Results from BUSCO and an analysis of UCEs ( ultra-conserved elements ) suggest that our reference is nearly complete , with 92% of single-copy orthologues unique to birds and 97% of UCEs identified in amniotes ( Faircloth et al . , 2012; Supplementary file 2 ) . We aligned WGS data from 110 blackcaps ( including the two birds used in our assembly ) to this reference ( average coverage 17 . 5x , Figure 1a , Supplementary file 4 ) and estimated genotype likelihoods at genome-wide single nucleotide polymorphisms ( SNPs ) . Genomic differentiation was low between migratory populations of different distances and orientations , but unlike earlier work using mitochondrial data ( Pérez-Tris et al . , 2004; Dietzen et al . , 2008 ) , we documented considerable differentiation between migrant and resident populations on both the continent and islands . A PCA separated resident island birds from continental populations on PC1 , and resident continental birds from migrants on PC2 ( Figure 1b ) . Migrants were not clearly distinguished on either PC ( we obtained the same result when we re-ran the PCA excluding islands; Figure 1—figure supplement 1 ) . Results from an ADMIXTURE analysis and estimates of FST confirm this pattern . At a cluster value of two , ADMIXTURE distinguished between island and continental birds ( similar to PC1 ) . At a cluster value of three , populations on the continent were further divided into resident and migratory groups ( similar to PC2 ) , and resident island and continent birds showed some admixture with the migratory group ( Figure 1c ) . No further structure was observed beyond these three clusters ( Figure 1—figure supplement 2 ) . Estimates of FST ranged from 0 . 018 to 0 . 11 , with the highest estimates occurring between migrants and both resident groups ( 0 . 06–0 . 11 for islands , 0 . 042–0 . 05 for continent residents; Figure 1d ) . Evidence for limited population differentiation combined with dramatic differences in the migratory behaviour of blackcaps is ideal for identifying genomic regions that are associated with this focal trait . Specifically , genomic regions associated with migration should standout against this backdrop of limited differentiation , although analyses involving residents will need to account for elevated values of differentiation . A phylogeographic analysis using mitochondrial data suggested that variation in migratory behaviour evolved recently , 4000–13 , 000 ya ( Pérez-Tris et al . , 2004 ) . Our results move this date further back in time , to 30 , 000 ya and the start of the last glacial maximum ( Clark et al . , 2009 ) . Specifically , we used multiple sequentially Markovian coalescent ( MSMC , implemented in MSMC2 ) ( Schiffels and Durbin , 2014; Malaspinas et al . , 2016 ) to characterize the demographic history of blackcaps . The demographic trajectories of migratory , resident continent and resident island birds began to diverge ~30 , 000 ya . The effective population size of migrant and resident island populations expanded and contracted , respectively , while continental residents exhibited a relatively constant effective population size ( Figure 2a; Figure 2—figure supplement 1 ) . Relative cross-coalescence rates ( CCR ) between all three groups exhibited a concomitant drop ~30 , 000 ya ( Figure 2b; Figure 2—figure supplement 2 ) . The drop of relative CCR between migratory and resident island populations was steeper than that between migratory and resident continent populations ( Figure 2b ) , suggesting that genetic separation following the colonization of islands resulted in greater separation than that between continental populations of migrants and residents . Increased differentiation between migrants and resident island birds ( vs . resident continent birds ) was also documented in our PCA ( Figure 1 ) . Results for medium-distance migrants ( NW , SW and SE ) and long-distance migrants are indistinguishable ( Figure 2—figure supplement 4; Figure 2—figure supplement 5; Figure 2—figure supplement 6 ) . One interesting finding from our demographic analyses is that of apparent gene flow between migrant and resident continent birds ~5000 years ago . Specifically , the relative CCR between migrant and resident continent populations started to increase at ~5000 ya ( ~25 , 000 years after initial divergence; Figure 2—figure supplement 3 ) . This admixture event may reflect secondary contact between migrant and resident continent populations and is line with our results from ADMIXUTRE , with admixture documented between these groups at a cluster value of three . The last glacial maximum ended ~19 , 000–11 , 500 ya ( Clark et al . , 2009 ) . After this time , populations would have expanded out of their glacial refugia , and perhaps migrant and resident continent populations came into secondary contact ~5000 years after these expansions began . Similar to our results on population differentiation , island populations exhibit their own evolutionary trajectories following divergence . This result is in line with results from Dietzen et al . ( 2008 ) , who suggested that at least one separate colonization to the Atlantic islands occurred ( earlier than that to the Canaries ) . Here , we transition to study local patterns of genomic differentiation , identifying specific genomic regions that have signatures of selection related to three phenotypes: the propensity to migrate , orientation of migration and distance of migration ( resident continent , short distance SW , medium distance NW , SW , SE and long distance SE populations ) . We excluded resident island birds from these analyses ( because of limited sample size [n = 5] and potential effects from founder events ) and focused on a single resident continent population ( Gibraltar , we obtained similar results using Cazalla de la Sierra , total number of birds included in these analyses = 82 , Supplementary file 4 ) . Positive selection was more common in residents and limited to a few , small genomic regions ( Table 1a ) . For example , hapFLK is a tree-based method that controls for hierarchical population structure . Global and local NJ trees are constructed using haplotype frequencies and regions under selection show longer branch lengths . Only nine regions were found to be under selection ( permutation test , see 'Materials and methods' ) according to this method , and six of these appeared in residents . Figure 3a shows estimates of hapFLK for the entire genome , and Figure 4a exemplifies one region on Super-Scaffold 99 ( syntenic with flycatcher chromosome 3 ) . The average size of these regions was 16 . 7 kb and only six genes occurred within them . We used CAVIAR ( Rochus et al . , 2018 ) to identify variants showing strong associations with selection in these regions . Each region included one to four variants , all of which occurred in non-coding regions ( Supplementary file 5 ) . Previous phylogeographic work suggested that migration is the ancestral state in blackcaps ( Pérez-Tris et al . , 2004; Voelker and Light , 2011 ) . Accordingly , the selection in genomic regions that we identified here is probably involved in the transition from migrant to resident phenotypes . We complemented results from hapFLK with a modified version of the Population Branch Statistics ( PBS ) ( Yi et al . , 2010 ) and the number of segregating sites by length ( nSL ) ( Ferrer-Admetlla et al . , 2014 ) . PBS is an FST-based statistic that estimates allele frequency differences between three or more populations . This parameter can be elevated by linked purifying selection ( or background selection ) within populations that is unrelated to positive selection ( in our case selection related to migration ) . We removed these confounding effects by scaling PBS and subtracting the maximum value of PBS in orthologous windows from that in the non-focal population ( hereafter ‘ΔPBS’ , following Vijay et al . , 2017 ) . nSL is a haplotype-based statistic that focuses on patterns within populations , using segregating sites to measure the length of haplotypes . Linked selection should increase haplotype lengths at genomic regions that are under positive selection . Eight of the nine regions identified by hapFLK also exhibited extreme values of ΔPBS and nSL ( in the top 1% of the distribution ) in the same population as that identified by hapFLK ( ΔPBS Table 1a , Figure 3b for resident birds [estimates for short distance SW , medium distance SE , SW , NW , and long distance SE , Figure 2—figure supplement 4; Figure 2—figure supplement 5; Figure 2—figure supplement 6]; nSL Supplementary file 5 ) . As noted already , population structure and linked selection can elevate differentiation between populations . We controlled for these effects using hapFLK and ΔPBS , respectively , and emphasise that genomic differentiation between populations of blackcaps is low to begin with ( Figure 1d ) . In addition , linked purifying selection would be expected to increase PBS in all populations ( i . e . , not just the focal population ) , but this is not the case . This is exemplified in Figure 5a where estimates of ΔPBS for all populations are shown but these are only elevated in the resident continent population . As a final test of population structure , we re-estimated ΔPBS using resident island birds ( instead of resident continent birds ) . We conservatively excluded these populations from our initial analyses because their sample sizes are small and because genetic drift can affect estimates of differentiation in island populations . Nevertheless , the island populations are also resident and thus these estimates could help to validate the genomic regions that were identified as being under selection in resident populations on the continent . Table 1a summarizes these results , noting which genomic regions exhibited elevated values of PBS on islands . Of particular interest , PBS was elevated in all three island populations at the genomic region on Super-Scaffold 99 ( Figure 5b ) . Combined with findings from hapFLK ( controlling for population structure and relying on haplotypes ) , ΔPBS ( controlling for linked selection and relying on SNP data ) and nLS ( estimated within populations and relying on haplotypes ) , these results provide strong evidence that this specific region contains important variation for the transition to residency , not only on the continent but also on the islands . Note that it is possible that the signatures of positive selection that we document here reflect selection based on different ecological variables involved with the colonization of areas further south on the continent , but at least in the case of Super-Scaffold 99 , we believe that this is rather unlikely as most ecological variables ( biotic and abiotic ) are quite distinct between islands and the continent ( and between the islands themselves ) ( Cropper , 2013; Valente et al . , 2017 ) . The transition to residency is shared , probably representing one of the only shared selection pressures experienced by all of these populations . Note that the lack of consistent results for other regions under selection in the resident continent population does not preclude the potential importance of these regions as , for example , genetic drift on islands would affect which genetic variants were present on islands for selection to act on . Our finding that only a few genomic regions under selection contain genes and that the strongly associated SNPs identified by CAVIAR are in non-coding regions could suggest that cis-regulatory changes are important for the transition from migration to residency . In support of this suggestion , an alignment of predicted mRNAs from several bird species and transcripts from a testis transcriptome of the blackcap placed two of the SNPs from CAVIAR in the 3′ untranslated region ( 3′ UTRs ) of two genes ( GPR83-L on Super-Scaffold 12 and CHST4 on Super-Scaffold 13 , syntenic with flycatcher chromosomes 11 and 4a , respectively ) . Three prime3′ UTRs can act as posttranscriptional regulators; they contain binding sides for microRNAs , which can inhibit translation or target mRNA for degradation ( Mayr , 2017; Barrett et al . , 2012 ) . In fact , previous work with monarch butterflies identified 55 conserved microRNAs that are differentially expressed between summer and migratory butterflies ( Zhan et al . , 2011 ) . Future analyses to validate this suggestion could include the use of qPCR to determine whether GPR83-L and/or CHST4 are in fact differentially regulated between the migratory phenotypes . Future work using techniques aimed at identifying binding sites for transcription factors ( e . g . , ChIP-seq ) could also be useful . We conducted a preliminary analysis here , using HOMER ( Heinz et al . , 2010 ) to detect known transcription factor motifs in the genomic regions that are under selection in residents . Specifically , Ruegg et al . ( 2014 ) used a literature search to identify 25 candidate genes for migration . Four of these genes are transcription factors whose motifs are in the libraries searched by HOMER: three basic helix-loop-helix transcription factors ( bHLH ) ( Clock , Npas2 , and Bmal1 ) and one basic leucine zipper domain ( Nfil3 ) . We found a bHLH motif ( GHCACGTG ) on Super-Scaffolds 12 and 99 ( Figures 3a , b , 4a and 5 ) . The motif on Super-Scaffold 99 is particularly interesting as there is a SNP ( G/T ) at the beginning of the motif that is nearly fixed in continental residents ( the allele frequency for G in Asni , Gibraltar and Cazalla de la Sierra is 1 , 0 . 85 and 0 . 9 , respectively; FST between Gibraltar and medium-distance NW , SW and SE migratory populations is 0 . 15 , 0 . 25 and 0 . 44 , respectively ) . This motif could disrupt or weaken transcription factor binding ( Kasowski et al . , 2010 ) . This is also the genomic region that showed elevated PBS in both resident continent and island populations ( Figure 4a , Figure 5 ) . Clock , Npas2 and Bmal1 are involved in maintaining circadian rhythms . Circadian rhythms synchronize circannual clocks , which are important cues controlling seasonal migratory behaviour ( Gwinner , 1996; Visser et al . , 2010 ) . Concerning the actual identity of genes within regions that are under selection , several have functions that could be related to the transition from migration to residency . For example , LOC100859173 ( located on in the genomic region under selection on Super-Scaffold 12 , the region with a bHLH motif mentioned above; Table 1a ) has been annotated as a probable G-protein coupled receptor that mediates the function of neuropeptide Y ( NPY ) . NPY is localized in the brain of birds and works with Agouti-related peptide ( AGRP ) and proopiomelanocortin ( POMC ) to control energy balance . Specifically , NPY/AGRP neurons stimulate appetite , food intake and fat deposition , while POMC inhibits these processes ( Boswell and Dunn , 2017 ) . It has been hypothesized that the effects of NPY may extend to seasonal changes in energy balance that are important for migration , including hyperphagia and fat deposition ( Boswell and Dunn , 2017 ) . Beyond its role in energy balance , NPY also facilitates learning and memory via the modulation of hippocampal activity and has an effect on circadian rhythms , reproduction , and the contraction of vascular smooth muscles . It has been suggested that a common genetic mechanism or major regulator may control migratory traits ( Liedvogel et al . , 2011; Liedvogel and Lundberg , 2014 ) . A protein such as NPY , or the transcription factors that bind the bHLH motif identified in the prior analysis , could fill this role . Early research on blackcaps was pivotal for demonstrating the existence of a genetic basis of migration and studying its evolution . This is due in large part to the tractability of this species and its variability in migratory behaviour . Here , we have expanded this study system beyond phenotypic and marker-based approaches , launching it into the genomic era and conducting one of the most comprehensive genome-wide analyses of migration to date . Populations of blackcaps began to diverge ~30 , 000 years ago , but differentiation remains low between migratory populations . There is evidence for past gene flow between migratory and resident populations on the European continent but comparison of the contemporary structure of these populations suggests that gene flow may be limited . This is certainly the case for resident island birds . It has been suggested that one single genetic mechanism controls migratory traits and may be shared across broad taxonomic groups . We do not find evidence for one common genetic mechanism across species here , and no protein-coding change is shared across the three focal traits ( propensity , distance and orientation ) that we examined in unison . Future work on gene expression may identify major regulators that control multiple migratory traits , and both NPY and bHLH transcription factors are good candidates . Combined with the additional results that we presented here ( such as the importance of standing genetic variation ) , this information is vital for understanding how predictable the evolution of migration and other complex behavioural traits may be . Blackcaps have not only been relevant to work on the evolution and genetics of migration . Early work in this system suggested that differences in migration might serve as reproductive isolating barriers early in speciation . For example , hybrids were shown to exhibit intermediate orientation behaviour that was predicted to be inferior because it would bring hybrids over large ecological barriers that pure forms avoid ( Helbig , 1991b ) . More recently , it was shown that NW migrants arrive on the breeding grounds earlier than SW migrants , and that these birds mate assortatively on the basis of arrival time , helping to reduce gene flow between phenotypically distinct groups ( Bearhop et al . , 2005 ) . The role of migration in speciation has gained considerable traction in recent years ( Rolshausen et al . , 2009; Bearhop et al . , 2005; Irwin and Irwin , 2005; Rohwer and Irwin , 2011; Turbek et al . , 2018; Delmore and Irwin , 2014; Bensch et al . , 2009 ) and results from our study suggest that selection at a very small number of loci may be sufficient to initiate reductions in gene flow very early in the process of population differentiation and speciation .
Blood samples from two male blackcaps from the Mooswald breeding population at Freiburg im Breisgau , Germany , classified as medium-distance SW migrants ( on the basis of morphometrics and isotope signatures ) were used to assemble the reference genome . Full details on all steps in our genome assembly can be found in Supplementary file 6 ( BioProject number PRJNA545868; Guojie Zhang , personal communication ) . Briefly , genomic DNA from one individual was used to sequence Illumina sequencing libraries ( fragment and mate pair libraries with insert sizes of 2 , 5 and 10 kb ) . 275 . 9 Gb of raw high throughput sequence ( HTS ) data were generated and assembled using ALLPATHS-LG . This assembly was improved several ways ( e . g . , by removing duplicates and closing gaps ) . DNA from the second individual was used to generate two BioNano optical maps ( one using BspQI and the other BssSI ) . These maps were used to super-scaffold HTS scaffolds . Statistics for the final assembly and each stage can be found in Supplementary file 2 . We used SatsumaSynteny ( Grabherr et al . , 2010 ) to determine which avian chromosome each scaffold was found on ( aligning scaffolds to the flycatcher genome , Supplementary file 3 ) . We validated our initial ALLPATHS assembly , the improved ALLPATHS assembly and our final assembly ( including BioNano optical maps ) using BUSCO ( version 3 . 0 . 2 , AUGUSTUS species chicken and aves_odb9 dataset ) and by blasting ultra-conserved elements ( UCEs ) identified by Faircloth ( 2016 ) using whole-genome alignments for the chicken and zebra finch ( Supplementary file 2 ) . We annotated genes with putative functions and protein domains using MAKER . Gene prediction was performed using a de novo testis transcriptome of blackcaps and cDNAs from three avian species ( zebra finch , chicken and flycatchers ) from the ensembl database . Following MAKER , we obtained the predicted protein sequences to annotate genes functionally using blastp and Interproscan . For the final annotation , we only included gene predictions that either had an Annotation edit Distance ( AED ) <0 . 5 and/or a blastp hit ( with the thresholds described above ) and/or a predicted protein domain . We obtained whole genome resequencing ( WGS ) data from 110 male blackcaps ( including WGS data from the two individuals used to generate the reference genomes ) . High molecular weight DNA was extracted from blood withdrawn from the brachial vein , following a standard salt extraction protocol . Individual samples were collected across the European breeding range including three island populations ( Canary Islands , Cape Verde , and Azores ) and covering the entire range of migratory phenotypes . Population phenotype was scored on the basis of morphometry , stable isotope signature and/or ringing-recovery data from selected individuals ( see Supplementary file 4 for a description of how each population was phenotyped ) . Birds were sampled during the breeding season unless indicated otherwise . Specifically , exceptions are a subset of UK overwintering birds ( n = 6 ) sampled during the winter in the British Isles , and a subset of long-distance SE migrants ( n = 5 ) caught during autumn migration and selected on the basis of wing length ( see Supplementary file 4 for details ) . We also obtained WGS data for five garden warblers and three hill babblers , the closest sister taxa to blackcaps , sampled during breeding ( Voelker and Light , 2011 ) . We prepared small insert libraries using DNA from each individual and sequenced five samples per lane on NextSeq 500 with paired-end 150 bp reads . We trimmed reads with trimmomatic ( TRAILING:3 SLIDINGWINDOW:4:10 MINLEN:30 ) ( Bolger et al . , 2014 ) . All analyses made use of data from resequencing reads that were aligned to the reference genome using bwa mem ( Li and Durbin , 2009 ) or stampy in the case of the garden warblers ( divergence time of 0 . 026 based on alignments of UCEs ( Faircloth , 2016; https://github . com/faircloth-lab/phyluce/ ) . GATK ( McKenna et al . , 2010 ) and picardtools ( http://broadinstitute . github . io/picard ) were used to identify and realign reads around indels ( RealignerTargetCreator , IndelRealigner ) as well as remove duplicates ( MarkDuplicates , all default settings ) . We recalibrated the resulting bam files using GATK’s base quality score recalibration ( BQSR ) . Specifically , we called SNPs for each population separately using three different programs and default settings: UnifiedGenotyper from GATK , samtools ( Li et al . , 2009 ) and FreeBayes ( Garrison and Marth , 2012 ) . BQSR requires a set of known variants . We used SNPs identified in all three programs and populations as the set of known variants for the first round of BQSR . We conducted a second round using common SNPs from the three programs that were also of high quality ( QUAL >995 , ~10% of the common SNPs ) . Most of our analyses made use of the BQSR recalibrated bams , calling genotype likelihoods ( GL ) with ANGSD ( version 0 . 910–24-gf84f594 , Korneliussen et al . , 2014 ) and filtering reads that did not map to a unique location , did not have a mapping pair , or had mapping qualities below 20 and flags ≥256 . When it was not possible to use GL as input , we used a vcf that had been run through GATK’s variant quality score recalibration ( VQSR ) . VQSR also requires a set of known SNPs . We used the second set of known SNPs ( common and high-quality ) from BQSR for this analysis and combined variants from all populations into a single vcf file for subsequent analyses . All repetitive regions were excluded from our analyses and those focused on demography did not include the Z chromosome . We estimated FST between all populations using GLs from ANGSD , starting by estimating unfolded site frequency spectrums ( SFS ) for each population ( doSaf 1 , gl 1 ) and using them to obtain joint frequency spectrums ( 2DSFS , realSFS ) for each pair of populations . These 2DSFSs were used as priors for allele frequencies at each site to estimate FST ( realSFS fst index ) . In order to estimate unfolded SFS , we needed an ancestral sequence , or the ancestral state of variants segregating in blackcaps . This sequence was generated using WGS from garden warblers and hill babblers . Specifically , we used samtools to generate fasta files for each garden warbler and hill babbler ( n=5 and n=3 , respectively ) and used rules outlined in Poelstra et al . ( 2014 ) to call ancestral states , with alleles that were homozygous in both outgroup species being considered ancestral and excluding remaining sites ( those that were triallelic or heterozygous in the outgroup species ) . We obtained consensus fasta sequences for each population using ANGSD ( -doFasta 2 –doCounts 1 –minQ 20 –setMinDepth 5 ) and used IQTREE ( Nguyen et al . , 2015 ) to construct maximum likelihood trees for each scaffold in the blackcap genome ( there was no difference in the topology obtained for scaffolds mapping to the Z chromosome so they were included in the consensus , data not shown ) . We summarized the resulting trees using phylip ‘consense’ and constructing an extended majority-rule consensus tree ( in which nodes that were supported by fewer than 50% of the input trees are collapsed ) . We used MSMC2 to infer the demographic history of blackcaps in our dataset . MSMC2 implements the multiple sequentially Markovian coalescent ( MSMC ) model , estimating effective population size by time and relative cross-coalescence rates between any two populations . It allows inference of the expansions and contractions of a population and of the extent and timing of population divergence ( Malaspinas et al . , 2016 ) . Specifically , by running a hidden Markov model ( HMM ) along all possible pairs of haplotypes , MSMC2 estimates the free parameters for a demography model ( a series of effective population sizes as a function of segmented time ) and relative cross-coalescence rates between sequences using a maximum-likelihood approach . After phasing our data using fastphase ( Scheet and Stephens , 2006 ) , we combined individuals into six groups ( medium and long migrants [‘med + long’] , short-distance SW migrants ( ‘short’ ) , resident continent birds , and resident island birds from the Azores , Cape Verde , and Canary Islands ) . We grouped medium ( NW , SW and SE ) and long-distance migrants because they exhibited very little population structure ( Figure 1 ) and indistinguishable demographic histories ( Figure 2—figure supplement 4; Figure 2—figure supplement 5; Figure 2—figure supplement 6 ) . We excluded any birds with less than 15x coverage . This filter left us with all island individuals ( five individuals for each island ) , five short migrants , 19 continental residents , and 44 med + long migrants . To avoid bias associated with the use of unequal numbers of individuals from each group , we randomly down-sampled five individuals from med + long migrants and continental residents to create 10 sample groups . We used the bamCaller . py script provided in the msmc-tools package ( https://github . com/stschiff/msmc-tools; Khvorykh , 2018 ) to create sample-specific callability mask files . We generated a global mappability mask file for the reference genome using GEM ( Derrien et al . , 2012 ) . We inferred effective population size by running MSMC2 separately for each group ( Schiffels and Wang , 2020 ) . We determined the number of clusters for fastPHASE using a cross-validation procedure ( https://github . com/inzilico/kselection/ Khvorykh , 2018 ) . Statistical phasing ( i . e . , phasing without a reference panel ) can be error prone , but fastPHASE is commonly employed in non-model organisms and is well-suited to datasets like ours that include high density SNPs on a physical map ( Scheet and Stephens , 2006; Burri et al . , 2015; Kawakami et al . , 2017 ) . The analysis of cross-coalescence rates requires comparisons between groups and we considered all possible combinations of groups for our analysis ( Schiffels and Wang , 2020 ) . In other words , we ran analyses for all 15 possible combinations ( three between groups on the continent , three between populations on the islands , and nine for comparisons between the three continent groups and three island populations ) . For each pairwise combination , we ran the combineCrossCoal . py script from msmc-tools ( https://github . com/stschiff/msmc-tools ) and computed the relative cross-coalescence rate by dividing the between-populations coalescence rate by the average within-population coalescence rate . We scaled results using a mutation rate of 3 × 10−9/gen/site and a generation time of 2 years ( Nadachowska-Brzyska et al . , 2016; Nadachowska-Brzyska et al . , 2015 ) . hapFLK is a tree-based method that is used to identify genomic regions that are under selection . This program permits the inclusion of two or more populations and accounts for both drift within populations ( different Ne ) and covariance across them ( hierarchical structuring ) . We used the vcf from VQSR as input for this analysis , applying two additional filters for the inclusion of variants: minimum number of individuals/phenotype = 5 and minor allele frequency of 0 . 05 . hapFLK also requires an estimate of the number of clusters into which haplotypes can be grouped . We ran this analysis for the complete dataset including all populations , and for a restricted dataset including only medium-distance migrants . We determined the number of clusters for each dataset separately using fastPHASE ( Scheet and Stephens , 2006 ) and the cross-validation procedure mentioned earlier . Once hapFLK is estimated , it is normalized using rlm in R , and p-values are computed from the chi-squared distribution . We used a permutation analysis to establish a threshold , beyond which genomic regions would be considered to be experiencing positive selection . Specifically , we randomly shuffled population labels 100 times , re-estimated hapFLK and p-values , recorded the lowest p-value for each randomization and set the threshold to the fifth percentile across randomizations . Once these regions were identified , we determined which population was experiencing selection by comparing branch lengths for a tree built using data from the entire genome and one built using data from the region under selection . Note that results from analyses using medium-distance migrants are plotted using the resident phenotype for illustrative purposes , but the analysis was not run using these birds . We include birds from three resident continent populations – Cazalla de la Sierra and Gibraltar in the Iberian Peninsula along with Asni in Morocco ( only three birds were sampled from this African population , precluding its use in the present analysis; Supplementary file 4 ) . The Iberian Peninsula where the other two populations are found is highly heterogeneous as a result of the effects of mountains and plateaus that create variation in seasonality and , consequently , in the intensity of blackcap migratory behaviour ( Pérez-Tris and Tellería , 2002; Tellería et al . , 2001 ) . There is also some evidence in our PCA to show that this heterogeneity has led to some differentiation between populations , as birds from Cazalla de la Sierra exhibit values more similar to migrants on PC2 ( Figure 1c ) . Accordingly , to avoid any confounding effects from population structure , we limited our analysis to birds from Gibraltar . Results using Cazalla de la Sierra instead were very similar . For example , all of the genomic regions identified in Table 1b were also in the top 1% of the ΔPBS distribution when Cazalla de la Sierra was used as the continental reference population instead of Gibraltar . The principle described above for hapFLK focusing on haplotype clusters can also be applied to SNPs ( FLK ) . We used results from an analysis with FLK and limited to genomic regions , which showed evidence of positive selection from hapFLK , to identify independent strongly associated SNPs with CAVIAR ( CAusal Variants Identification in Associated Regions [Hormozdiari et al . , 2014] ) . CAVIAR was originally designed to identify independent causal SNPs in GWAS studies . We followed methods described in Rochus et al . ( 2018 ) to modify this method for FLK , identifying SNPs with p-values <0 . 0001 in hapFLK outlier regions and using a correlation matrix generated by FLK by decomposing signals into loading on orthogonal components ( vs . p-values from a GWAS and LD as is traditionally done with CAVIAR ) . We used a modified version of PBS ( Population Branch Statistic ) to complement results from hapFLK . PBS is similar to FST , but can include more than two populations and identifies regions within each population that exhibit differences in allele frequencies . This statistic was originally designed for three populations , but can be expanded to include more populations ( Zhan et al . , 2014 ) . We used GL from ANGSD to obtain estimates of FST following the procedure described above ( summarized into windows of 2500 kb ) and used the equation below to estimate PBS from these values . This equation is an example that was applied to resident populations ( R ) , where T is log transformed FST between the populations indicated in exponents:TR-NW+ TR-SW+ TR-SE- TNW-SW- TSW-SE3 Recent papers have noted that FST can be elevated by reductions in within-population variation alone and that there are many factors that can reduce variation within populations , including linked selection in areas of reduced recombination that may result from purifying selection ( background selection , [Cruickshank and Hahn , 2014; Noor and Bennett , 2009] ) . It is unlikely that this process affects our results because recombination rate should elevate estimates of PBS in all populations , but this is not the case ( Figure 5a ) . Regardless , we followed methods from Vijay et al . ( 2017 ) to reduce any effects that linked selection may have on our results . Vijay et al . ( 2017 ) used estimates of FST between allopatric populations of crows that did not differ in their trait of interest to control for the effects of linked selection , estimating the difference in estimates of FST in focal populations and maximum FST in non-focal allopatric populations ( ΔFST ) . FST in focal populations would have to extend beyond that in non-focal populations to be considered important in generating the trait of interest . We used the same approach for PBS . For example , ΔPBS for resident continent populations was estimated by finding the difference between PBS in residents and maximum PBS in medium- , short- and long-distance migrants . The former analyses ( hapFLK and PBS ) rely on comparisons between phenotypes . In this last analysis , we focus on the affects that selection can have within a population instead . Specifically , selective sweeps can reduce variation at both the locus under selection and its neighbours ( Smith and Haigh , 1974 ) . Local reductions in variation result in the presence of extended regions of haplotype homozygosity within phenotypes ( long haplotypes at high frequency ) . nSL ( number of segregating sites by length ) ( Ferrer-Admetlla et al . , 2014 ) is similar to the more common iHS , but instead of measuring the decay of haplotype identity as a function of recombination distance , it quantifies this decay of how many mutations remain in other haplotypes present in the dataset . In this way , nSL does not require a genetic map and is more robust to variation in not only recombination rate but also mutation rate . For this analysis , we used selscan ( v . 1 . 20a https://github . com/szpiech/selscan ) and the same vcf used in hapFLK , but split by phenotype ( and scaffold ) . We ran the data through fastPHASE first to phase haplotypes ( using 50 iterations of the EM algorithm , sampling 100 haplotypes from the posterior distribution and using same number of clusters identified for hapFLK ) . We normalized estimates of nSL into the same 2500-kb windows used for PBS . Two sets of preliminary analyses were used to identify regulatory SNPs in the regions identified by hapFLK and PBS as being under selection . First , we focused on 3′ UTRs , downloading predicted mRNAs from Ensembl and NCBI for several bird species , including the Atlantic canary , White-throated sparrow , American crow , Great tit , Collared flycatcher , Zebra finch , Wild turkey , White-rumped munia , Hooded crow , Blue tit and Ground tit . We aligned these sequences with our annotation for the blackcap , and with transcripts assembled from RNAseq data obtained from the testes of a single male blackcap , to determine whether any of the strongly associated SNPs identified by CAVIAR were within 3' UTRs . Alignment files are available upon request . In a second set of analyses , we used HOMER ( Heinz et al . , 2010 ) to identify known transcription factor binding sites ( TFBS ) in genomic regions under selection . Specifically , we used findMotifsGenome . pl with default settings to identify known motifs in each region and scanMotifGenomeWide . pl to identify the specific location in each region where the motif could be found ( permitting no mismatches ) . HOMER includes known motifs for thousands of transcription factors ( mostly for model organisms ) ; we chose to focus on candidate transcription factors identified by previous studies as having an association with migration ( Ruegg et al . , 2014 ) . In our final analysis on migratory distance , we limited our dataset to short- , medium- and long-distance migrants . We coded distance phenotype as an ordinal variable from 1 to 3 and conducted a GWAS analysis using a Bayesian sparse linear mixed model ( BSLMM ) ( Zhou et al . , 2013 ) . We chose BSLMM models here ( instead of hapFLK ) because they allow the inclusion of ordinal variable ( vs . categorical with hapFLK ) . BSLMM models include the phenotype as the response variable and allele frequencies at a set of SNPs as the predictor variable . They also include a term for factors that influence the phenotype and are correlated with genotype ( e . g . , population structure ) . BSLMMs are adaptive models that include linear mixed models ( LMM ) and Bayesian variable selection regression ( BVSR ) as special cases and that learn the genetic architecture from the data . We ran four independent chains for each BSLMM , with a burnin of 5 million steps and a subsequent 20 million MCMC steps ( sampling every 1000 steps ) . We report one hyperparameter from this model ( PVE: the proportion of variance in phenotypes explained by all SNPs , also called chip heritability ) and consider SNPs with inclusion probabilities >0 . 01 following Gompert et al . ( 2013 ) . Note , we chose to run this analysis with GEMMA instead of hapFLK as we did with our other migratory traits ( orientation and propensity ) . This is because our focal variable here ( distance ) is ordinal in nature and this fact would have been lost in hapFLK . We could not code this variable as continuous because the average distance individuals in each population travel on migration is not exactly known . | Every year as the seasons change , thousands of animals migrate huge distances in search of food or better climates . As far as migrations go , there might be none so impressive as the trans-oceanic flights made by small migrating songbirds . These birds can weigh as little as three grams and travel up to 15 , 000 kilometres . Most migrate alone and at night and yet still manage to return to the same location each year . Several strands of research suggest there could be a genetic basis to their migratory behaviour , but exactly which genes control this phenomenon remains poorly understood . One small songbird that has been studied for decades is the European blackcap . This species exhibits a real variety of migration patterns . Some blackcaps travel rather short distances , others much further , and some populations do not migrate at all . Populations that share the same breeding grounds in the summer may migrate in different directions in the autumn . These features make it a good species to study the genetic variation between populations that migrate in different directions and over different distances . However , only in recent years has advancing technology made it possible to comprehensively study an animal’s entire genome , leaving no gene unturned . Now , Delmore et al . have used high-throughput sequencing technologies to trace the evolutionary history of migration in European blackcap and started by assembling a reference genome for the species . Then , the genomes of 110 blackcaps from several populations that take different annual migrations were compared to the reference . This revealed that the populations began to diverge some 30 , 000 years ago and that there was some apparent gene mixing between groups of migrating and resident blackcaps around 5 , 000 years ago . The analysis showed only a small set of genes code for their differences in migration . Additionally , while the candidate genes were shown to be common among blackcaps , the genes identified did not match those reported from studies of other migrating songbirds . Finally , Delmore et al . also noted that the differences between the populations tend to be in the parts of the genome that control whether a given gene is switched on or off , which could explain how new migratory behaviours can rapidly evolve . This study is one of the most comprehensive genomic analysis of migration to date . It is important work as songbirds , like other animals , are responding to increasing pressures of environmental and climate change . In time , the findings could be used to support conservation efforts whereby genetic analyses could determine if certain populations possess enough variation to respond to coming changes in their habitats . | [
"Abstract",
"Introduction",
"Results",
"and",
"discussion",
"Materials",
"and",
"methods"
] | [
"evolutionary",
"biology",
"genetics",
"and",
"genomics"
] | 2020 | The evolutionary history and genomics of European blackcap migration |
An emerging theme in cellular logistics is the close connection between mRNA and membrane trafficking . A prominent example is the microtubule-dependent transport of mRNAs and associated ribosomes on endosomes . This coordinated process is crucial for correct septin filamentation and efficient growth of polarised cells , such as fungal hyphae . Despite detailed knowledge on the key RNA-binding protein and the molecular motors involved , it is unclear how mRNAs are connected to membranes during transport . Here , we identify a novel factor containing a FYVE zinc finger domain for interaction with endosomal lipids and a new PAM2-like domain required for interaction with the MLLE domain of the key RNA-binding protein . Consistently , loss of this FYVE domain protein leads to specific defects in mRNA , ribosome , and septin transport without affecting general functions of endosomes or their movement . Hence , this is the first endosomal component specific for mRNP trafficking uncovering a new mechanism to couple mRNPs to endosomes .
Trafficking of membranes is essential for intracellular logistics . Important membranous carriers are endosomes that transport lipids , proteins , and mRNAs . These large vesicular structures are well-known for their function in endocytosis , transporting plasma membrane proteins to their site of degradation in the lysosome/vacuole system ( Huotari and Helenius , 2011; Rusten et al . , 2012 ) . However , they also carry out other functions , such as receptor recycling or cytoplasmic signalling , and are therefore considered to be multipurpose platforms ( Gould and Lippincott-Schwartz , 2009 ) . Early endosomes are characterised by the presence of Rab5-like small G proteins and their special lipid composition consisting of PI3P lipids ( phosphatidylinositol 3-phosphate; Stenmark et al . , 2002; Kutateladze , 2006 ) . These lipids are recognised by distinct protein domains , such as the FYVE zinc finger ( Stenmark et al . , 1996 ) . Endosomes are actively transported along the microtubule cytoskeleton , which is particularly critical in highly polarised cells , such as neurons and fungal hyphae . In the latter , microtubule-dependent transport supports apical tip growth and secretion of hydrolytic enzymes . This process is streamlined for efficiency and defects in transport result in impaired polar growth and reduced fitness ( Peñalva et al . , 2012; Riquelme and Sánchez-León , 2014 ) . An emerging theme is the intimate linkage of membrane and mRNA trafficking during spatio-temporal control of gene expression ( Kraut-Cohen and Gerst , 2010; Jansen et al . , 2014 ) . Important examples are the actin-dependent co-transport of mRNAs and ER ( endoplasmic reticulum ) during budding in Saccharomyces cerevisiae ( Schmid et al . , 2006 ) or the microtubule-dependent co-transport of mRNAs and endosomes during hyphal growth ( Baumann et al . , 2012; Göhre et al . , 2013 ) . Key factors are RNA-binding proteins that recognise specific localisation sequences within target mRNAs . Together with accessory factors , such as the poly ( A ) -binding protein , they form large macromolecular complexes called mRNPs ( messenger ribonucleoprotein particles , Bullock , 2011; Eliscovich et al . , 2013; Buxbaum et al . , 2015 ) . At present , however , detailed mechanistic insights on the connection of mRNPs to membranes are scarce ( Jansen et al . , 2014 ) . The best fungal model system to study co-trafficking of endosomes and mRNAs is the corn pathogen Ustilago maydis ( Jansen et al . , 2014 ) . Here , the switch from yeast-like to hyphal growth is essential for the infection of its host , and defects in this polar growth correlate with reduced fungal virulence ( Brefort et al . , 2009; Vollmeister et al . , 2012a ) . In hyphae , endosomes shuttle extensively along the microtubule cytoskeleton throughout the entire length of the hyphae ( Steinberg , 2014 ) . Transport is mediated by a cytoplasmic dynein complex ( Straube et al . , 2001 ) transporting Rab5a-positive endosomes towards the microtubule minus-ends and the kinesin-3 type motor Kin3 transports in the opposite direction ( Schuster et al . , 2011 ) . Since endosomes carry the SNARE Yup1 ( soluble N-ethylmaleimide-sensitive-factor attachment receptor; Wedlich-Söldner et al . , 2000 ) and are positive for Rab5a , they were classified as early endosomes , which have initially been proposed to mainly function in endocytosis and signalling ( Steinberg , 2012; Bielska et al . , 2014 ) . Recently , we discovered a novel function for these endosomes , namely mRNA transport throughout the hyphae ( Baumann et al . , 2012 ) , a process that is critical for polar growth and unconventional secretion of the endochitinase Cts1 ( Becht et al . , 2006; Koepke et al . , 2011 ) . The key factor is the RNA-binding protein Rrm4 containing three N-terminal RRMs ( RNA recognition motifs ) for RNA-binding and two C-terminal PABC/MLLE domains ( Figure 1A; Becht et al . , 2005; Zarnack and Feldbrügge , 2010; Baumann et al . , 2012; Vollmeister et al . , 2012b ) . The latter is known from the cytoplasmic poly ( A ) -binding protein and functions as a binding pocket for peptides containing a PAM2 motif ( PABP-interacting motif 2; Albrecht and Lengauer , 2004; Kozlov et al . , 2004; Jinek et al . , 2010; Xie et al . , 2014 ) . 10 . 7554/eLife . 06041 . 003Figure 1 . Loss of Upa1 causes defects in hyphal growth . ( A ) Schematic representation of proteins drawn to scale ( bar , 200 amino acids ) using the following colouring: green , RNA recognition motif ( RRM ) ; dark blue , MLLE domain ( SMART E-values 6 . 7 and 0 . 35 for Rrm4 , Letunic et al . , 2009 ) ; red , PAM2 motif; dark grey , Ankyrin repeats; purple , FYVE domain; light blue , RING domain . ( B ) Comparison of PAM2 sequences found in Upa1 ( accession number UMAG_12183 ) with those of human proteins , such as Paip1 ( accession number NP_006442 . 2 ) , Paip2 ( accession number CAG38520 . 1 ) , eRF3B ( accession number CAB91089 . 1 ) , and Atx2 ( accession number NP_002964 . 3 ) . ( C ) Edge of colonies growing on charcoal-containing medium under hyphae-inducing conditions ( 48 h . p . i . ) . ( D ) Growth of AB33 derivates in the yeast ( left ) and hyphal form ( right; 8 h . p . i . ; size bar , 10 µm ) . Growth direction is marked by arrows . ( E ) Percentage of hyphae ( 8 h . p . i . ) : unipolarity , bipolarity , and septum formation was quantified ( error bars , s . e . m . ; n = 3 independent experiments , >100 hyphae were counted per experiment; note that septum formation is given relative to the values of unipolar or bipolar hyphae set to 100% ) . ( F ) Relative chitinase activity mainly detecting endochitinase Cts1 ( Koepke et al . , 2011 ) in the yeast ( left ) or hyphal form ( right; error bars , s . e . m . ; n = 3 independent experiments ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06041 . 003 Rrm4 specifically associates with shuttling Rab5a-positive endosomes ( Baumann et al . , 2012 ) and binds a specific set of mRNAs encoding , for example , the small G protein Rho3 or the septin Cdc3 ( König et al . , 2009 ) . Studying Cdc3 in more detail revealed that not only its mRNA but also the protein is transported on endosomes in an Rrm4-dependent manner suggesting that endosome-coupled translation is crucial for septin localisation on these membranous carriers and needed for septin filamentation ( Baumann et al . , 2014 ) . This was verified by demonstrating that translationally active ribosomes are transported on endosomes ( Higuchi et al . , 2014 ) . Thus , Rrm4-dependent transport carries out important general functions , such as distributing mRNAs ( König et al . , 2009; Baumann et al . , 2012 ) and associated ribosomes ( Higuchi et al . , 2014 ) , as well as more specific functions such as endosomal septin transport ( Baumann et al . , 2014 ) . Despite the detailed knowledge on microtubule-dependent transport of endosome-coupled mRNA trafficking ( Jansen et al . , 2014 ) , it is still unknown how mRNAs and associated proteins are connected to endosomes . Here , we identified a FYVE protein with specific functions in endosomal mRNP transport by coupling mRNPs to the shuttling vesicles .
To identify factors connecting mRNPs to shuttling endosomes , we had two reasons for initially focusing on proteins containing the MLLE interaction motif PAM2 ( Albrecht and Lengauer , 2004 ) . First , two MLLE domain proteins , namely Rrm4 and Pab1 ( Figure 1A ) , shuttle with Rab5a-positive endosomes along microtubules ( Baumann et al . , 2012 ) , and second , mutations in the C-terminal MLLE domain of Rrm4 interfered with its movement ( Becht et al . , 2006 ) . Closer inspection of Rrm4 revealed that it carries a second region with low similarity to the MLLE domain ( Figure 1A , see below ) . To find potentially interacting PAM2-containing proteins , we performed a HMM motif search for PAM2 screening the genome of U . maydis ( Albrecht and Lengauer , 2004; Kozlov et al . , 2004; Kämper et al . , 2006 ) . Among the 14 obtained candidates , UMAG_12183 was particularly interesting because the encoded protein contained a lipid-binding FYVE domain , and its C-terminal domain architecture resembled the endosomal protein Pib1p from S . cerevisiae or mammalian Rififylin ( Figure 1A; Supplementary file 1; Shin et al . , 2001 ) . In addition to the PAM2 motif ( Figure 1B ) and the FYVE domain , it contained five ankyrin repeats known to be protein–protein interaction interfaces ( Al-Khodor et al . , 2010 ) , and a RING domain involved in ubiquitination ( Figure 1A ) . The protein was designated Upa1 for the U . maydis PAM2 protein . For functional analysis , we deleted upa1 in the laboratory strain AB33 by homologous recombination ( Brachmann et al . , 2004 ) . AB33 expresses an active heterodimeric transcription factor under control of the nitrate regulated nar1 promoter . Since the active heterodimer is sufficient to elicit the morphological transition , hyphae can be induced synchronously and reproducibly by switching the nitrogen source ( Figure 1C–D; Brachmann et al . , 2001; Baumann et al . , 2012 ) . Studying growth of upa1Δ cells revealed no mutant phenotype in the yeast form suggesting that cytokinesis including septa formation was not disturbed ( Figure 1D , see below ) . However , loss of Upa1 caused defects in hyphal growth . At the colony level , shorter hyphae were observed , which were comparable to those of rrm4Δ strains ( Figure 1C ) . At the cellular level , a significant proportion of upa1Δ cells grew bipolar in contrast to unipolar wild-type hyphae and those hyphae that grew unipolar inserted retraction septa with reduced frequency ( Figure 1D–E ) . This again is reminiscent of the mutant phenotype of rrm4Δ strains ( Figure 1C–E; Baumann et al . , 2014 ) . Since rrm4Δ mutants were also disturbed in unconventional secretion of Cts1 , specifically during hyphal growth ( Koepke et al . , 2011; Stock et al . , 2012 ) , we determined extracellular chitinase activity . This revealed that Cts1 secretion was strongly reduced in the upa1Δ strain only in the hyphal form , which was comparable to rrm4Δ strains ( Figure 1F ) . Thus , loss of Upa1 causes defects in hyphal growth and secretion of Cts1 , two cellular processes that are regulated by Rrm4-mediated endosomal mRNA transport . The PAM2 motif is defined as an interaction interface of the MLLE domain of the poly ( A ) -binding protein ( Albrecht and Lengauer , 2004 ) . To address whether the predicted PAM2 motif in Upa1 ( Figure 1B ) interacts with the MLLE domain of Pab1 of U . maydis , we used the yeast two-hybrid assay . To this end , Pab1 ( König et al . , 2009 ) or Upa1 versions were fused at the N-terminus with the DNA-binding domain or activation domain of Gal4p , respectively ( Matchmaker 3 system , Clontech ) . Constructs were transformed into the S . cerevisiae strain AH109 , and control experiments were performed ( Figure 2—figure supplement 1 ) . Interaction was scored by growth on selection plates . Testing Pab1 with full length Upa1 revealed no interaction . However , assaying a version of Upa1 without the FYVE domain ( Upa1ΔF ) did show binding ( Figure 2A ) , suggesting that the FYVE domain might have interfered with the nuclear localisation of the protein . Upa1mPΔF additionally carrying point mutations in the PAM2 motif ( Figure 2A ) or Upa1ΔΝ1ΔF with a deletion of the PAM2 region was no longer able to interact with Pab1 , indicating that the PAM2 motif is necessary for binding ( Figure 2A ) . Analysing only the first 194 amino acids ( Upa1ΔC1 ) showed that this PAM2-containing region of Upa1 is sufficient for interaction ( Figure 2A ) . Using a similar strategy for the interaction partner showed that the MLLE domain of Pab1 is necessary ( Pab1mM ) and sufficient ( MLLEPab1 ) for interaction with Upa1ΔC1 ( Figure 2A ) . To verify these results in independent experiments , we demonstrated that an N-terminal part of Upa1 containing the PAM2 motif interacts with MLLEPab1 in GST pull down assays with the components expressed in Escherichia coli ( see below ) . In summary , Upa1 contains a functional PAM2 motif that interacts with the MLLE of Pab1 . 10 . 7554/eLife . 06041 . 004Figure 2 . The PAM2 motif of Upa1 interacts specifically with the MLLE domain of Pab1 . ( A ) Two-hybrid analysis with schematic representation of variants tested ( left ) and growth plates ( right ) . Yeast cultures were serially diluted 1:5 ( decreasing colony forming units , cfu ) and spotted on respective growth plates assaying for reporter gene expression ( see ‘Materials and methods’ ) . ( B ) Schematic representation of N-terminal truncated Upa1 variants fused at C-terminus with Gfp , drawn to scale ( see Figure 1A; mP , mutation in the PAM2 motif indicated as black bar ) . ( C ) Percentage of hyphae ( 8 h . p . i . ) : unipolarity , bipolarity , and septum formation was quantified ( error bars , s . e . m . ; n = 3 independent experiments , >100 hyphae were counted for each strain per experiment; note that septum formation is given relative to the values of unipolar or bipolar hyphae set to 100% ) . ( D ) Edge of colonies growing on charcoal-containing medium under hyphae-inducing conditions ( 48 hr p . i ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06041 . 00410 . 7554/eLife . 06041 . 005Figure 2—figure supplement 1 . Interaction of Upa1 PAM2/Pab1 MLLE in vivo . ( A ) Schematic representation of the constructs used ( for details see legend in Figure 2A ) . ( B ) Western blot analysis of yeast extracts expressing a Pab1 and Upa1 variants ( given above the lanes ) carrying a Myc tag and a HA epitope tag , respectively; * and ** mark cross reacting proteins . ( C ) Two-hybrid analysis as described in Figure 2 ( 2A marks results already shown in Figure 2A ) . Positive and negative controls ( interaction of p53 with T-Antigen and Lamin C with T-Antigen , respectively ) were recommended by the provider of the Matchmaker 3 system ( Clontech ) . Vector indicates the use of a yeast two-hybrid plasmid without insert . DOI: http://dx . doi . org/10 . 7554/eLife . 06041 . 00510 . 7554/eLife . 06041 . 006Figure 2—figure supplement 2 . The PAM2 motif is dispensable for Cts1 secretion . ( A ) Western blot analysis of strains expressing Upa1-Gfp ( left ) or Rrm4-Gfp ( right ) . Cells were harvested at 0 , 2 and 4 h . p . i . ; detection of α-tubulin Tub1 served as control for equal protein amounts . ( B ) Western blot analysis of strains expressing various N-terminal truncations of Upa1 ( depicted schematically on the left , see Figure 1 ) . α-Gfp antibodies were used for detection of Upa1 and detection of Tub1 served as control for equal protein amounts . Note that the expression of Upa1ΔN3-Gfp to Upa1ΔN5-Gfp was strongly reduced . Therefore , these variants were not analysed further . ( C ) Relative chitinase activity mainly detecting endochitinase Cts1 in the hyphal form ( error bars , s . e . m . ; n = 3 independent experiments; Koepke et al . , 2011 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06041 . 006 In order to test the functional importance of the different Upa1 domains , we first generated a strain expressing Upa1 as functional C-terminal fusion protein with the enhanced version of the green fluorescent protein ( Figure 1C–F , Upa1-Gfp , eGFP ) . This was achieved by homologous recombination at the upa1 locus of strain AB33 resulting in wild-type expression levels to avoid artefacts due to overexpression . Control experiments revealed that the amount of Upa1-Gfp did not change during the switch from yeast to hyphal growth ( Figure 2—figure supplement 2A ) . To test PAM2 functionality , we generated Upa1 variants carrying a mutation in the PAM2 motif of Upa1 ( Upa1mP-Gfp ) or a deletion of an N-terminal part containing that motif ( Upa1ΔN1-Gfp; Figure 2B ) . The protein levels were comparable to that of wild type ( Figure 2—figure supplement 2B ) . Testing for unipolar growth , for hyphal growth of colonies ( Figure 2C–D ) , and for Cts1 secretion ( Figure 2—figure supplement 2C ) showed that the mutant strains did not differ from wild type . Hence , the Pab1 interacting motif PAM2 was dispensable for function . To pinpoint functionally , important regions in the protein additional N-terminal deletions , Upa1ΔN2-6-Gfp ( Figure 2—figure supplement 2B ) , were generated . Only expression of Upa1ΔN2-Gfp and Upa1ΔN6-Gfp was comparable to the wild-type level , and therefore , the function of Upa1ΔN3-5-Gfp could not be assessed ( Figure 2—figure supplement 2B ) . Nevertheless , assaying Upa1ΔN2-Gfp revealed that the strain appeared to be slightly impaired in function ( Figure 2C–D , Figure 2—figure supplement 2C ) , suggesting that there is a functionally important region at the N-terminus ( 143–357 aa ) . The Upa1ΔN6-Gfp expressing strain was indistinguishable from the upa1Δ strain , indicating a complete loss of function ( Figure 2C–D; Figure 2—figure supplement 2C ) . Taken together , the PAM2 motif is dispensable for function of Upa1 . Next , we tested the interaction of Upa1 with the second MLLE domain-containing protein Rrm4 using the yeast two-hybrid assay ( see above ) . Therefore , we fused Upa1-Gfp or Rrm4 versions to the DNA-binding domain or activation domain of Gal4p , respectively . Control experiments were performed as described above ( Figure 3—figure supplement 1A ) . In contrast to Pab1 , Rrm4 interacted with full length Upa1-Gfp ( Figure 3A ) . Testing N-terminal truncations of Rrm4 revealed a minimal interaction domain containing the two predicted MLLE domains , while the C-terminal MLLE domain alone was not sufficient for binding ( Figure 3A ) . Analysing C-terminal deletion of either this or a mutated MLLE domain demonstrated that in contrast to Pab1MLLE , the domain is necessary but not sufficient for interaction with Upa1-Gfp ( Figure 3—figure supplement 1A ) . 10 . 7554/eLife . 06041 . 007Figure 3 . Upa1 contains two PAM2L motives for interaction with Rrm4 . ( A ) Two-hybrid analysis with schematic representation of variants tested ( left ) and growth plates ( right ) . Yeast cultures were serially diluted 1:5 ( decreasing cfu ) and spotted on respective growth plates assaying for reporter gene expression . ( B ) Two-hybrid analysis as in ( A ) . Red rectangle indicates minimal region in Upa1 interacting with Rrm4 . ( C ) Two-hybrid analysis as in ( A ) . Upa1 region identified in ( B ) was analysed by linker scanning mutagenesis ( mutations indicated as black bar , Mut1-12 ) . ( D ) Comparison of PAM2 and PAM2L sequences as in Figure 1B . Note , the second PAM2L motif was only mutated in Mut7 ( E ) PAM2L sequences of Upa1 compared to related sequences from basidiomycetes ( U . m . , Ustilago maydis UMAG_12183 / XP_758247 . 1; S . r . , Sporisorium reilianum , accession number sr13323 / CBQ72642 . 1; U . h . , Ustilago hordei accession number UHOR_03 , 485 / CCF52210 . 1; P . a . Pseudozyma antarctica GAK65366 . 1; C . c . Coprinopsis cinerea CC1G_00 , 427 / XP_001837291 . 2; C . p . Coniophora putanea XP_007767511 . 1; L . b . Laccaria bicolor XP_001876756 . 1; A . d . Auricularia delicate XP_007337909 . 1 ) . ( F ) GST co-purification experiments with components expressed in E . coli N-terminal His6-tagged versions of Upa1 , Upa1mP , and Upa1mPL ( amino acids 1–363 ) were expressed to the same level ( first input lane , I1; see Figure 3—figure supplement 5B ) . MLLE domains of Pab1 or Rrm4 ( MLLEPab1 or Rrm4ΔN5 , respectively ) were expressed as GST fusion proteins ( second input lane , I2 ) . After GST affinity chromatography proteins were eluted ( lanes marked with "E" ) . Interaction studies were performed with whole protein extracts from E . coli to demonstrate specific binding . ( G ) Schematic representation of Upa1 variants carrying mutations ( black boxes ) in the PAM2 and PAM2L regions . ( H ) Percentage of hyphae ( 8 h . p . i . ) : unipolarity , bipolarity , and septum formation was quantified ( error bars , s . e . m . ; n = 3 independent experiments , >100 hyphae were counted per experiment; note that septum formation is given relative to the values of unipolar or bipolar hyphae set to 100% ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06041 . 00710 . 7554/eLife . 06041 . 008Figure 3—figure supplement 1 . Upa1 interacts with Rrm4 in vivo . Control experiments for the detailed two-hybrid analysis given in Figure 3 . ( A ) Two-hybrid analysis mapping the domain in Rrm4 that interacts with Upa1 was carried out as described in Figure 3 . Positive and negative controls ( interaction of p53 with T-Antigen and Lamin C with T-Antigen , respectively ) were recommended by the provider of the Matchmaker 3 system ( Clontech ) . Note that Rrm4ΔN2 failed to interact with Upa1-Gfp for unknown reasons . ( B ) Western blot analysis of yeast extracts expressing Upa1-Gfp and Rrm4 variants ( given above the lanes ) carrying a Myc tag and a HA epitope tag , respectively . * and ** mark cross reacting proteins . Schematic representation of variants is given on the left . ( C ) Western blot analysis of yeast extracts expressing Upa1-Gfp variants ( schematically shown on the left ) and Rrm4 ( given above the lanes ) . Upa1-Gfp was detected with αc-Myc-antibody and Rrm4 with αHA-antibody . DOI: http://dx . doi . org/10 . 7554/eLife . 06041 . 00810 . 7554/eLife . 06041 . 009Figure 3—figure supplement 2 . The evolutionarily conserved core of both PAM2L motifs is essential for interaction with Rrm4 . ( A ) Sequences of the PAM2L motifs from Upa1 ( conserved core in white and alanine mutations are given in red ) . ( B ) Two-hybrid analysis with schematic representation of variants tested ( left ) and growth plates ( right ) . Yeast cultures were diluted 1:5 ( decreasing colony forming units , cfu ) and spotted on respective growth plates assaying for reporter gene expression . ( C ) Western blot analysis of yeast extracts expressing Upa1-Gfp and Rrm4 variants ( given above the lanes ) carrying a Myc tag and a HA epitope tag , respectively . * and ** mark cross reacting proteins . Schematic representation of variants is given in B . DOI: http://dx . doi . org/10 . 7554/eLife . 06041 . 00910 . 7554/eLife . 06041 . 010Figure 3—figure supplement 3 . Conserved PAM2L motif in the Upa1 N-terminal region . Sequence comparison of Upa1 homologues of various basidiomycetes ( names and accession numbers are given in Figure 3 ) . Remarkably , the FVYP sequence of PAM2L ( red box ) is the only conserved region at the N termini . DOI: http://dx . doi . org/10 . 7554/eLife . 06041 . 01010 . 7554/eLife . 06041 . 011Figure 3—figure supplement 4 . Conserved PAM2L motif in the central region of Upa1 . Sequence comparison of Upa1 homologues from various basidiomycetes ( names and accessions numbers given in Figure 3 ) . The FxYP sequence of PAM2L ( red box ) is conserved in the central part . The computationally predicted FYVE zinc finger domain is given in blue ( SMART; Letunic et al . , 2009 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06041 . 01110 . 7554/eLife . 06041 . 012Figure 3—figure supplement 5 . Sequence specific recognition of the PAM2 and PAM2L sequence with the MLLE domains of Pab1 and Rrm4 , respectively . ( A ) Results of Figure 3F are given with the important region enlarged showing the co-purifying Upa1 variants ( red boxes ) . ( B ) Coomassie stained SDS-PAGE gels of protein fractions analysed by GST-pulldown assays shown in Figure 3F . On the left side , protein extracts of E . coli expressing all variants of His6-Upa1 are shown . The corresponding band is labelled . Please note , that these proteins—like all variants of Upa1—exhibit a band at a higher kDa size than predicted . On the right side , experimental steps of the pulldown experiment are shown ( I = Input; FT1 = flow through 1; FT2 = flow through 2; E = elute of bound proteins ) . The band height of each protein is indicated on the right . ( B ) Two-hybrid analysis with schematic representation of variants tested and growth plates as described in Figure 2 . Variants of a N-terminal region of Upa1 ( Upa1N2; amino acid 1 to 408 ) carrying no mutation ( wt ) , a mutation in the PAM2 motif ( mPAM2 ) or a mutation in the PAM2L motif ( mPAM2L ) , were tested against full length as well as MLLE-containing versions of Rrm4 and Pab1 . DOI: http://dx . doi . org/10 . 7554/eLife . 06041 . 01210 . 7554/eLife . 06041 . 013Figure 3—figure supplement 6 . Sequence specific recognition of the PAM2 and PAM2L sequence with the MLLE domains using purified components . Purified protein fractions ( His6-tagged Upa1 versions and GST-tagged MLLE domains of Pab1 and Rrm4 ) analysed by GST-pulldown assays as shown in Figure 3F . Coomassie-stained gels are shown in ( A ) and results of Western blot analysis in ( B ) using αHis and αGST antibodies . DOI: http://dx . doi . org/10 . 7554/eLife . 06041 . 01310 . 7554/eLife . 06041 . 014Figure 3—figure supplement 7 . The PAM2L motifs are functionally important for efficient secretion of Cts1 . ( A ) Schematic representation of Upa1 variants carrying mutations ( black boxes ) in the PAM2 and PAM2L regions . ( B ) Relative chitinase activity detecting endochitinase Cts1 ( Koepke et al . , 2011 ) in the hyphal form ( error bars , s . e . m . ; n = 3 independent experiments ) . ( C ) Western blot analysis of strains expressing Upa1-Gfp versions ( depicted in A; α-Gfp antibodies were used for detection of Upa1 and detection of Tub1 served as control for equal protein amounts ) . ( D ) Kymographs of hyphae expressing Upa1-Gfp versions showing bidirectional movement of signals as diagonal lines and indicating that the mutations in the PAM2L motives did not cause drastic differences in endosomal movement of Upa1-Gfp versions . DOI: http://dx . doi . org/10 . 7554/eLife . 06041 . 014 Screening N-terminal truncations of Upa1 , we observed that , surprisingly , the PAM2 sequence was not needed for interaction with Rrm4 ( Upa1ΔN1-Gfp , Figure 3B ) . Instead , the Rrm4-interacting region was mapped to the centre of Upa1 ( Figure 3B , Figure 3—figure supplement 1B , C ) . Moreover and also unexpectedly , an Upa1 variant without the central Rrm4-interacting region was still able to interact with Rrm4 ( Upa1ΔF-Gfp , Figure 3B ) , suggesting the presence of two yet unknown interacting regions . Therefore , we performed a more detailed analysis of the central interaction region of Upa1 ( position 886 to 1030 ) , applying linker scanning mutagenesis with 10 amino acid block mutations . Only a mutation in region 948 to 958 caused loss of binding ( mutation 7 in Figure 3C ) . Interestingly , this region exhibited sequence similarity to the PAM2 motif ( designated PAM2L , PAM2-like; Figure 3D ) , and furthermore , a second PAM2L sequence was found in the N-terminus of Upa1 ( Figure 3D ) supporting the initial two-hybrid data ( Figure 3B ) . Mutating the conserved five amino acid core EFxxP or the highly conserved phenylalanine residue alone confirmed that these amino acids in both PAM2L motifs are crucial for the interaction with Rrm4 ( Figure 3—figure supplement 2 ) . A phylogenetic analysis of Upa1 revealed that both PAM2L sequences are conserved in related fungal proteins ( Figure 3E , Figure 3—figure supplement 3 and Figure 3—figure supplement 4 ) . Thus , Upa1 interacts with Pab1 and Rrm4 via similar but not identical motifs , and a single PAM2L motif is sufficient for interaction . To compare the binding specificities , we analysed the interaction of the Upa1 N-terminus containing PAM2 and the N-terminal PAM2L against full length Pab1 and Rrm4 as well as minimal regions of both proteins . Furthermore , we verified the observations performing GST co-purification experiments using protein variants expressed in E . coli ( Figure 3F , Figure 3—figure supplement 5 and Figure 3—figure supplement 6 ) . The results were consistent with the mapping analysis ( Figure 3B–C; Figure 3—figure supplement 2 ) showing that the MLLE-containing regions of Rrm4 and Pab1 exclusively recognise the corresponding PAM2L and PAM2 motifs , respectively . Importantly , we addressed whether the PAM2L motifs are also needed for Upa1 function during hyphal growth . To this end , we generated strains expressing Upa1-Gfp versions carrying mutations in the PAM2L motifs ( Figure 3G ) . Scoring unipolar growth and secretion of Cts1 revealed consistent results leading to the conclusion that one PAM2L motif is sufficient for function , but if both are mutated functionality of Upa1 is lost . Additional mutations in the PAM2 motif made no difference indicating that it is the PAM2L motifs that are indeed crucial for activity ( Figure 3G–H; Figure 3—figure supplement 7 ) . In essence , Upa1 interacts directly with Rrm4 via two novel , functionally important PAM2L motifs . For further support of these interaction studies , we investigated the subcellular localisation of Upa1-Gfp in hyphae of U . maydis . The protein localised exclusively in the cytoplasm and was mainly present on distinct units that shuttled bidirectionally throughout the hyphae ( Figure 4A–B; Video 1 ) . No staining of other specific compartments , such as vacuoles was visible ( Figure 4A ) . The observed motility was comparable to the bidirectional movement of Rrm4-Gfp and Pab1-Gfp ( Figure 4B; Figure 4—figure supplement 1A–C; Videos 2 , 3 ) that are known to shuttle on Rab5a-positive endosomes ( see below; Baumann et al . , 2012; Baumann et al . 2014 ) . Note that in contrast to Pab1-Gfp , the cytoplasmic signal of Upa1-Gfp is weak suggesting that the Upa1/Pab1 interaction is restricted to shuttling units ( Figure 4—figure supplement 1A–C ) . 10 . 7554/eLife . 06041 . 015Figure 4 . Endosomal targeting of Upa1 is functionally important . ( A ) Micrograph ( size bar , 10 μm ) and corresponding kymograph of hyphae expressing Upa1-Gfp showing bidirectional movement of signals as diagonal lines ( arrowheads , Video 1 ) . ( B ) Bar diagrams depicting amount of processive Upa1-Gfp signals ( left , processive units per 10 μm hyphal length to accommodate for size differences between individual hyphae; error bars , s . d . ; more than 30 hyphae per strain ) and their velocity ( right; velocity of tracks showing >5 μm processive movement; error bars , s . d . ; 10 to 12 hyphae and more than 900 tracks per strain ) . ( C ) Hyphae treated with microtubule inhibitor benomyl . Micrograph ( size bar , 10 μm ) and corresponding kymograph showing static signals as vertical lines ( arrowheads; Video 4 ) . ( D ) Hyphae expressing Upa1-Gfp and carrying deletion in kin3 . Micrograph ( size bar , 10 μm ) and corresponding kymograph showing static signals as vertical lines ( arrowheads; Video 5 ) . Arrow points towards residual movement . ( E ) Schematic representation of Upa1 fused at C-terminus with Gfp drawn to scale ( see Figure 1A ) . ( F ) Percentage of hyphae ( 8 h . p . i . ) : unipolarity , bipolarity , and septum formation was quantified ( error bars , s . e . m . ; n = 3 independent experiments , >100 hyphae were counted per experiment; note that septum formation is given relative to the values of unipolar or bipolar hyphae set to 100% ) . ( G ) Relative chitinase activity mainly detecting endochitinase Cts1 in the hyphal form ( Koepke et al . , 2011; error bars , s . e . m . , n = 3 independent experiments ) . ( H , I ) Micrographs ( size bar , 10 μm ) and corresponding kymographs of hyphae expressing Upa1ΔR-Gfp ( H ) or Upa1ΔFR-Gfp ( I ) ( Videos 10 , 11 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06041 . 01510 . 7554/eLife . 06041 . 016Figure 4—figure supplement 1 . The FYVE domain is crucial for function . ( A–C ) Micrographs ( size bar , 10 μm ) of hyphae expressing Upa1-Gfp ( A ) , Rrm4-Gfp ( B ) , or Pab1-Gfp ( C ) and corresponding kymographs showing bidirectional movement of signals as diagonal lines ( arrowheads; Videos 1–3 ) . Note that the cytoplasmic background signals of Upa1-Gfp resemble Rrm4-Gfp rather than Pab1-Gfp suggesting that Upa1 does only interact with Pab1-Gfp on the cytoplasmic surface of endosomes . ( D ) Schematic representation of C-terminal truncated Upa1 variants fused at its C-terminus with Gfp drawn to scale ( see Figure 1A ) . ( E ) Edge of colonies growing on charcoal-containing medium under hyphae-inducing conditions ( 48 h . p . i . ) . ( F ) Western blot analysis of strains expressing various C-terminal truncations of Upa1 ( depicted schematically in D ) . α-Gfp antibodies were used for detection of Upa1-Gfp variants and detection of Tub1 served as control for equal protein amounts . ( G ) Micrographs ( size bar , 10 μm ) and corresponding kymographs of hyphae expressing Upa1-Gfp variants shown in Figure 4 ( Videos 6–9 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06041 . 01610 . 7554/eLife . 06041 . 017Video 1 . Upa1-Gfp moves bidirectionally in a hypha of AB33upa1-Gfp . Video corresponds to Figure 4A ( size bar = 10 μm , timescale in seconds , 200 ms exposure time , 150 frames , 5 frames/s display rate; QuickTime format , 6275 kB ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06041 . 01710 . 7554/eLife . 06041 . 039Video 2 . Rrm4-Gfp moving bidirectionally in a hypha of AB33rrm4-Gfp . Video corresponds to Figure 4—supplement figure 1B ( size bar = 10 μm , timescale in seconds , 150 ms exposure time , 150 frames , 6 frames/s display rate; QuickTime format , 1536 kB ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06041 . 03910 . 7554/eLife . 06041 . 040Video 3 . Pab1-Gfp moving bidirectionally in a hypha of AB33pab1-Gfp . Note , that in contrast to Upa1-Gfp and Rrm4-Gfp ( Videos 1 and 2 ) a higher background signal can be attributed to non-transported poly-adenylated mRNAs . Video corresponds to Figure 4—supplement figure 1C ( size bar = 10 μm , timescale in seconds , 150 ms exposure time , 150 frames , 6 frames/s display rate; QuickTime format , 473 kB ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06041 . 040 Upa1-Gfp movement was inhibited by treatment with the microtubule inhibitor benomyl ( Figure 4C; Video 4 ) and deletion of kin3 , which encodes the plus-end directed motor for endosomal movement , resulted in the accumulation of Upa1-Gfp signals in the centre of the cells where minus-ends of microtubules are located ( Figure 4D; Video 5; Baumann et al . , 2012 ) . Hence , Upa1 appears to localise on endosomes that shuttle along microtubules . 10 . 7554/eLife . 06041 . 018Video 4 . Upa1-Gfp movement is dependent on the microtubule cytoskeleton . Treating AB33upa1-Gfp for one hour with 50 μM microtubule destabilising drug benomyl , inhibits the bidirectional movement of Upa1-Gfp . Video corresponds to Figure 4C ( size bar = 10 μm , timescale in seconds , 200 ms exposure time , 150 frames , 5 frames/s display rate; QuickTime format , 2862 kB ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06041 . 01810 . 7554/eLife . 06041 . 019Video 5 . Loss of the plus-end directed kinesin-3 Kin3 disturbs Upa1-Gfp movement in AB33upa1-Gfp/kin3Δ . Immobile Upa1-Gfp accumulations can be seen in the middle of the cell . Residual movement might be due to dynein activity . Video corresponds to Figure 4D ( size bar = 10 μm , timescale in seconds , 200 ms exposure time , 150 frames , 5 frames/s display rate; QuickTime format , 2382 kB ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06041 . 019 To map the endosome binding domain , we analysed the subcellular localisation of the respective mutated versions Upa1mP-Gfp , Upa1ΔN1-Gfp , -ΔN2 and -ΔN6 ( Figure 3A ) . This revealed that all versions containing the FYVE domain shuttled on endosomes ( Figure 4—figure supplement 1G; Videos 6–9 ) consistent with the assumption that the FYVE domain is sufficient for endosome interaction . 10 . 7554/eLife . 06041 . 041Video 6 . Upa1mP-Gfp moving bidirectionally in a hypha of AB33upa1mP-Gfp . Mutating the PAM2 motif by amino acid exchanges does not inhibit movement of Upa1mP-Gfp . Video corresponds to Figure 4—supplement figure 1G top left ( size bar = 10 μm , timescale in seconds , 200 ms exposure time , 150 frames , 5 frames/s display rate; QuickTime format , 2901 kB ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06041 . 04110 . 7554/eLife . 06041 . 042Video 7 . Upa1ΔN1-Gfp moving bidirectionally in a hypha of AB33upa1ΔN1-Gfp . Loss of the first 143 amino acids including the PAM2 motif does not inhibit movement of Upa1ΔN1-Gfp . Video corresponds to Figure 4—supplement figure 1G top right ( size bar = 10 μm , timescale in seconds , 200 ms exposure time , 150 frames , 5 frames/s display rate; QuickTime format , 2695 kB ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06041 . 04210 . 7554/eLife . 06041 . 043Video 8 . Upa1ΔN2-Gfp moving bidirectionally in a hypha of AB33upa1ΔN2-Gfp . Loss of the first 337 amino acids including the PAM2 motif does not inhibit movement of Upa1ΔN2-Gfp although an increase in stationary background signals can be seen . Video corresponds to Figure 4—supplement figure 1G bottom left ( size bar = 10 μm , timescale in seconds , 200 ms exposure time , 150 frames , 5 frames/s display rate; QuickTime format , 5614 kB ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06041 . 04310 . 7554/eLife . 06041 . 044Video 9 . Upa1ΔN6-Gfp moving bidirectionally in a hypha of AB33upa1ΔN6-Gfp . The C-terminal part of Upa1 containing the FYVE and RING domains is sufficient for movement although an increase in stationary background signals can be seen . Video corresponds to Figure 4—supplement figure 1G bottom right ( size bar = 10 μm , timescale in seconds , 200 ms exposure time , 150 frames , 5 frames/s display rate; QuickTime format , 4388 kB ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06041 . 044 For functional analysis of the Upa1 C-terminus containing the RING and FYVE domain , we expressed corresponding C-terminal truncations ( Figure 4E , Figure 4—figure supplement 1D-F ) . Phenotypic analysis revealed that the RING domain ( Upa1ΔR-Gfp ) was dispensable for function under the tested conditions , whereas deletion of a C-terminal region containing the RING and FYVE domain ( Upa1ΔFRGfp ) results in loss of function ( Figure 4F–G , Figure 4—figure supplement 1E ) . Upa1ΔR-Gfp still shuttled on endosomes , but the additional deletion of the FYVE domain in Upa1ΔFRGfp abolished movement ( Figure 4H–I; Videos 10 , 11 ) . In summary , the FYVE domain of Upa1 mediates endosomal localisation and , importantly , endosomal targeting of Upa1 is crucial for its function during polar growth and Cts1 secretion . 10 . 7554/eLife . 06041 . 020Video 10 . Upa1ΔR-Gfp moving bidirectionally in a hypha of AB33upa1ΔR-Gfp . The C-terminal part comprising aa 1241–1287 including the RING domain is not necessary for shuttling of Upa1 . Video corresponds to Figure 4H ( size bar = 10 μm , timescale in seconds , 200 ms exposure time , 150 frames , 5 frames/s display rate; QuickTime format , 4069 kB ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06041 . 02010 . 7554/eLife . 06041 . 021Video 11 . Upa1ΔFR-Gfp shows an even distribution in a bipolar growing hypha of AB33upa1ΔFR-Gfp . Thus , the C-terminal part comprising of aa 1048–1287 including the FYVE and RING domains is necessary for localising Upa1 to endosomes . Video corresponds to Figure 4I ( size bar = 10 μm , timescale in seconds , 200 ms exposure time , 150 frames , 5 frames/s display rate; QuickTime format , 5201 kB ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06041 . 021 To study whether Upa1 is indeed part of the endosomal compartment that is positive for Rab5a and Rrm4 , Upa1-Gfp was expressed first with Rab5a-Cherry , an N-terminal fusion of Rab5a with the red fluorescent protein mCherry ( Baumann et al . , 2012 ) . Rab5a-Cherry localises to shuttling endosomes and exhibits additional staining in the cytoplasm ( Figure 5A; Video 12 ) , which most likely marks the endomembrane system proposed to be late endosomes involved in endocytosis ( Higuchi et al . , 2014 ) . Dynamic co-localisation experiments using dual view technology and msALEX microscopy ( millisecond alternating laser excitation , Baumann et al . , 2014 ) revealed that Upa1-Gfp co-localises extensively with motile Rab5a-positive endosomes ( Figure 5A , Figure 5—figure supplement 1A; about 90% percent in both directions ) . A second marker for this motile endosomal compartment is Rrm4 , which in contrast to Rab5a does not stain other membrane compartments . Consistently , the vast majority of Upa1-Gfp signals co-localises with Rrm4-Rfp ( Figure 5B , Figure 5—figure supplement 1B; Video 13 , about 90% percent in both directions ) . Thus , Upa1-Gfp is present on almost all Rab5a- and Rrm4-positive endosomes . 10 . 7554/eLife . 06041 . 022Figure 5 . Upa1 is crucial for Rrm4 movement on Rab5a-positive endosomes . ( A ) Dynamic co-localisation studies of Upa1-Gfp ( left ) and Rab5a-Cherry ( right ) using dual view and msALEX microscopy ( see ‘Materials and methods’; arrowheads indicate co-localising signals ) . Micrographs ( size bar , 10 μm ) and corresponding kymographs of hyphal tip ( Video 12 ) . ( B ) Dynamic co-localisation studies of Upa1-Gfp ( left ) and Rrm4-Rfp ( right ) as in ( A ) ( Video 13 ) . ( C–F ) Micrographs ( size bar , 10 μm ) and corresponding kymographs of hyphae expressing Rab5a-Gfp ( C ) , Rab5a-Gfp/upa1Δ ( D ) , Rrm4-Gfp ( E ) , or Rrm4-Gfp/upa1Δ ( F ) ( Videos 14–17 ) . ( G ) Bar diagrams depicting amount of Rab5a-Gfp signals per 10 μm hyphae in wt and upa1Δ cells ( left , error bars , s . d . ; >22 hyphae ) , as well as amount of Rrm4-Gfp signals per 10 μm hyphae in wt and upa1Δ cells ( right , error bars , s . d . ; >15 hyphae ) . ( H ) Number of Rab5a-Gfp ( left ) and Rrm4-Gfp ( right ) —signals passing zones in the middle of the hyphae and 10 μm from the apical pole in wt and upa1Δ cells , respectively ( passing events of signals/s , error bars , s . d . ; more than 15 hyphae ) . ( I ) Dynamic co-localisation studies of Rrm4-Gfp ( left ) and Rab5a-Cherry ( right ) using dual view and msALEX microscopy ( arrowheads indicate co-localising signals ) . ( J ) Same analysis as in ( I ) using a strain carrying a deletion in upa1 . Corralled movement of Rrm4 signals not found associated with Rab5a is highlighted by red arrows . DOI: http://dx . doi . org/10 . 7554/eLife . 06041 . 02210 . 7554/eLife . 06041 . 023Figure 5—figure supplement 1 . Upa1 co-localises with Rab5a and Rrm4 , but loss of Upa1 does not affect long-distance transport of endosomes . ( A ) Bar diagram showing the percentage of mobile Upa1-Gfp signals that co-localise with Rab5a-Cherry in the retrograde and anterograde direction ( error bars s . e . m . ; n = 11 hyphae ) . ( B ) Bar diagram showing the percentage of mobile Upa1-Gfp signals that co-localise with Rab5a-Cherry in the retrograde and anterograde direction ( error bars s . e . m . ; n = 10 hyphae ) . ( C ) DIC images of loss of function mutants in kin3 , upa1 , and rrm4 showing that cytokinesis is not disturbed . ( D ) Quantification of sporidia appearance in single cell form or cell aggregates ( error bars s . e . m . ; n = 3 experiments with >100 cells each ) . ( E ) FM4-64 uptake assays showing wild-type and upa1Δ mutants . FM4-64 uptake is not disturbed . ( F ) Bar diagram showing the velocity of processive units ( >5 μm/30 s ) of Rab5a-Gfp ( left ) , Rrm4-Gfp ( middle ) and Pab1-Gfp ( right ) in wt and upa1Δ background strains ( error bars , s . d . ; 10 hyphae each strain with >1000 tracks [Upa1-Gfp] , >550 tracks [Rrm4-Gfp] and >260 tracks [Pab1-Gfp] analyzed ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06041 . 02310 . 7554/eLife . 06041 . 024Figure 5—figure supplement 2 . Rrm4 and Pab1 movement is altered in the absence of Upa1 . ( A–C ) Representative kymographs of the movement of Rab5a-Gfp ( A ) , Rrm4-Gfp ( B ) and Pab1-Gfp ( C ) in wt and in upa1Δ background strains . Each movie was taken with an exposure time of 200 ms and consisted of 150 frames . DOI: http://dx . doi . org/10 . 7554/eLife . 06041 . 02410 . 7554/eLife . 06041 . 025Figure 5—figure supplement 3 . Rrm4 does not influence endosomal localisation of Upa1 . ( A ) Kymograph of Rrm4-Gfp in upa1Δ background strain treated with benomyl showing that residual movement is microtubule-dependent . ( B ) . Western blots of Rrm4-Gfp and Pab1-Gfp in wt and upa1Δ background strains 6 hr after induction of hyphal growth . α-Gfp antibodies were used for detection of Upa1-Gfp variants and detection of Tub1 served as control for equal protein amounts . ( C ) Kymograph of Upa1-Gfp in rrm4Δ background showing that Rrm4 is not needed for Upa1 movement . ( D ) Bar diagrams comparing the amount of Upa1-Gfp signals ( left; error bars , s . d . ; n = 37/23 cells ) , the velocity of processive Upa1-Gfp units ( middle; >5 μm/30 s; error bars , s . d . ; 10 hyphae each strain with >800 tracks ) and the number of Upa1-Gfp signals ( passing a line 10 μm from the apical pole ) in wt and rrm4Δ filaments ( right; error bars , s . d . ; more than 20 filaments ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06041 . 02510 . 7554/eLife . 06041 . 026Figure 5—figure supplement 4 . Residual processive movment of Rrm4-Gfp takes place on endosomes . Dynamic co-localisation studies of Rrm4-Gfp ( left ) and ( A ) Yup1-Cherry ( right ) or ( B ) FM4-64-stained signals ( right ) using dual view and msALEX microscopy . Arrowheads indicate co-localising processive signals and corralled movement of Rrm4-Gfp signals is highlighted by red arrows . DOI: http://dx . doi . org/10 . 7554/eLife . 06041 . 02610 . 7554/eLife . 06041 . 027Video 12 . Upa1-Gfp and Rab5a-Cherry ( upper and lower part , respectively ) co-localise in shuttling units in hyphae of AB33upa1-Gfp/rab5a-Cherry . Videos were recorded simultaneously using dual-colour detection and correspond to Figure 5A ( size bar = 5 μm , timescale in seconds , 70 ms alternating exposure time , 200 frames , 15 frames/s display rate; QuickTime format , 561 kB ) DOI: http://dx . doi . org/10 . 7554/eLife . 06041 . 02710 . 7554/eLife . 06041 . 028Video 13 . Upa1-Gfp and Rrm4-Rfp ( upper and lower part , respectively ) co-localise in shuttling units in hyphae of AB33upa1-Gfp/rrm4-Rfp . Videos were recorded simultaneously using dual-colour detection and correspond to Figure 5B ( size bar = 10 μm , timescale in seconds , 70 ms alternating exposure time , 100 frames , 15 frames/s display rate; QuickTime format , 132 kB ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06041 . 028 Next , we analysed the influence of the loss of Upa1 on the multiple functions of Rab5a-positive endosomes . Previously , it was shown that these endosomes function in cytokinesis and cell separation . For example , loss of the FYVE domain-containing guanine nucleotide exchange factor ( GEF ) Don1 or the plus-end directed Kinesin-3 type motor Kin3 resulted in the formation of cell aggregates ( Schink and Bölker , 2009 ) . Closer inspection revealed no cell separation defect in upa1Δ or rrm4Δ strains ( Figure 2B; Figure 5—figure supplement 1C , D ) indicating that Upa1 , like Rrm4 ( Becht et al . , 2005; Baumann et al . , 2014 ) , is not involved in endosomal functions during cytokinesis . Another function for these Rab5a-positive endosomes is their role in endocytosis . This is mainly based on the observation that the styryl dye FM4-64 follows the endocytotic pathway by initially staining the plasma membrane followed by staining Rab5a-positive shuttling endosomes and lastly vacuoles ( Higuchi et al . , 2014 ) . Testing endocytotic uptake of FM4-64 revealed no differences in the uptake and labelling of shuttling endosomes when comparing wild-type and upa1Δ strains ( Figure 5—figure supplement 1E ) , suggesting that Upa1 is not involved in endocytosis . Next , we tested the shuttling of Rab5a-Gfp in hyphae comparing wild-type and upa1Δ strains . This showed neither a difference in the velocity of Rab5a-Gfp-positive endosomes nor in the bidirectional movement of Rab5a ( Figure 5C–D , Figure 5H , Figure 5—figure supplement 1F; Videos 14 , 15 ) . Hence , the endosomal protein Upa1 is not essential for general endosome functions . 10 . 7554/eLife . 06041 . 029Video 14 . Rab5a-Gfp moving bidirectionally in a hypha of AB33rab5a-Gfp . Video corresponds to Figure 5C ( size bar = 10 μm , timescale in seconds , 150 ms exposure time , 150 frames , 6 frames/s display rate; QuickTime format , 248 kB ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06041 . 02910 . 7554/eLife . 06041 . 030Video 15 . Rab5a-Gfp moving bidirectionally in a hypha of AB33rab5a-Gfp/upa1Δ . Loss of Upa1 results in bipolar growing cell , but does not affect Rab5a-Gfp shuttling . Video corresponds to Figure 5D ( size bar = 10 μm , timescale in seconds , 150 ms exposure time , 150 frames , 6 frames/s display rate; QuickTime format , 219 kB ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06041 . 030 However , testing the impact on Rrm4-Gfp revealed that its movement was drastically impaired in upa1Δ strains . Although the velocity of processive Rrm4-Gfp signals is the same as in wild type ( Figure 5—figure supplement 1F ) , in the absence of Upa1 , we observed fewer processive signals and a significant increase in signals exhibiting corralled movement ( Figure 5E–G , Figure 5—figure supplement 2B; Videos 16 , 17 ) . This altered movement was particularly eminent when quantifying processive signals reaching the apical region of hyphal tips ( Figure 5H ) . This suggests that endosome association of Rrm4 was impaired . Analysing the altered movement of Rrm4 revealed that it was still microtubule-dependent ( Figure 5—figure supplement 3A ) . The decline in processive Rrm4-Gfp signals was not due to a reduced protein amount , since expression is comparable in wild-type and upa1Δ strains ( Figure 5—figure supplement 3B ) . Importantly , the processive Rrm4-Gfp signals co-localised with Rab5a and the endosomal marker protein Yup1 , as well as with the lipophilic dye FM4-64 . These co-localisation studies confirm that Rrm4 movement was specifically altered in the absence of Upa1 and that residual processive movement took place on endosomes ( Figure 5I–J; Figure 5—figure supplement 4 ) . Rrm4-Gfp signals exhibiting corralled movement did not co-localise with the membrane markers used , suggesting that these large accumulations constitute aberrant forms that fail to associate with the machinery for long-distance transport . Furthermore , addressing the endosomal shuttling of Upa1-Gfp in the absence of Rrm4 revealed that Upa1-Gfp movement is indistinguishable from wild type , suggesting that Upa1 attaches to endosomes independently of Rrm4 ( Figure 5—figure supplement 3C , D ) . In summary , Upa1 is dispensable for general endosomal functions but is crucial for endosomal recruitment of Rrm4 . 10 . 7554/eLife . 06041 . 031Video 16 . Rrm4-Gfp moving bidirectionally in a hypha of AB33rrm4-Gfp . Video corresponds to Figure 5E ( size bar = 10 μm , timescale in seconds , 150 ms exposure time , 150 frames , 6 frames/s display rate; QuickTime format , 2324 kB ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06041 . 03110 . 7554/eLife . 06041 . 032Video 17 . Rrm4-Gfp moving bidirectionally in a hypha of AB33rrm4-Gfp/upa1Δ . Loss of Upa1 disturbs shuttling of Rrm4-Gfp , as seen by increased corraled movement . Video corresponds to Figure 5F ( size bar = 10 μm , timescale in seconds , 150 ms exposure time , 150 frames , 6 frames/s display rate; QuickTime format , 741 kB ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06041 . 032 Previously , it was shown that Rrm4 functions in endosomal transport of mRNAs and associated ribosomes ( Baumann et al . , 2012 , 2014; Higuchi et al . , 2014 ) . To address whether these functions are altered in the absence of Upa1 , we first studied the movement of Pab1-Gfp , which is an established marker for mRNA ( Baumann et al . , 2012 , 2014 ) . In contrast to wild type , the number of Pab1-Gfp-positive , processive signals reaching the apical pole was strongly reduced in the upa1Δ strain , even though the protein level in both strains are comparable ( Figure 6A–C , Figure 5—figure supplement 2C; Videos 18 , 19 ) . The processive signals exhibited the same velocity in wild type and upa1Δ cells ( Figure 5—figure supplement 1F ) indicating that the transport machinery itself is not affected whereas the loading of the mRNA cargo to endosomes is ( Figure 6B–C ) . Consistently , the number of signals exhibiting corralled movement increased , and the number of signals reaching the apical pole decreased as described for Rrm4 ( Figure 6C , Figure 5—figure supplement 2 ) . These results suggest that the disturbed Rrm4 localisation in the absence of Upa1 also affects mRNA loading onto endosomes . 10 . 7554/eLife . 06041 . 035Figure 6 . Upa1 functions specifically in mRNP function of endosomes . ( A , B ) Micrographs ( size bar , 10 μm ) and corresponding kymographs ( Videos 18 , 19 ) of hyphae expressing Pab1-Gfp ( A ) or Pab1-Gfp/upa1Δ ( B ) . ( C ) Bar diagrams depicting amount of Pab1-Gfp signals per 10 μm hyphae in wt and upa1Δ cells ( left; error bars , s . d . ; more than 13 hyphae ) and number of Pab1-Gfp signals passing two zones in the middle of the hyphae and 10 μm from the apical pole in wt and upa1Δ cells ( right; passages of signals/s , error bars , s . d . ; >13 hyphae ) . ( D ) Analysing subcellular localisation of ribosomal protein Rps2-Gfp as an example . Micrographs ( size bar , 10 μm ) of hyphae expressing Rps2-Gfp ( nucleus indicated by asterisk ) . 20-μm area was bleached by laser irradiation about 10 μm from the hyphal tip . Arrowhead indicates processive signal entering the bleached area . ( E , F , G ) Kymographs of hyphal areas bleached with laser irradiation , as shown in ( D ) . Arrowheads indicate processive signals . DOI: http://dx . doi . org/10 . 7554/eLife . 06041 . 03510 . 7554/eLife . 06041 . 036Video 18 . Pab1-Gfp moving bidirectionally in a hypha of AB33pab1-Gfp . Video corresponds to Figure 6A ( size bar = 10 μm , timescale in seconds , 150 ms exposure time , 150 frames , 6 frames/s display rate; QuickTime format , 729 kB ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06041 . 03610 . 7554/eLife . 06041 . 037Video 19 . Pab1-Gfp moving bidirectionally in a hypha of AB33pab1-Gfp/upa1Δ . Loss of Upa1 disturbs shuttling of Pab1-Gfp , as seen by the drastically decreased number of processive Pab1-Gfp signals . Video corresponds to Figure 6B ( size bar = 10 μm , timescale in seconds , 150 ms exposure time , 150 frames , 6 frames/s display rate; QuickTime format , 353 kB ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06041 . 037 Next , we studied endosomal transport of Gfp-labelled ribosomal proteins of the small and large subunit in dependence of Upa1 . Endosomal movement of ribosomal proteins was visualised by bleaching an area about 10 μm from the hyphal tip followed by video microscopy ( Baumann et al . , 2014 , 2015; Figure 6D–G ) . Rps2-Gfp , Rpl25-Gfp , and Rps19-Gfp exhibited bidirectional movement in shuttling units ( Figure 6E–G ) , which were previously shown to be Rrm4-positive endosomes ( Baumann et al . , 2014; Higuchi et al . , 2014 ) . Note , that whereas Rps2-Gfp and Rpl25-Gfp were expressed ectopically , Rps19-Gfp was expressed at the homologous locus under control of the endogenous promoter resulting in stronger signals . Loss of Upa1 caused a severe reduction in shuttling signals ( analysing five hyphae each revealed 42 , 47 , and 78 processive signals in Rps2-Gfp , Rpl25-Gfp , and Rps19-Gfp in wild type but 0 , 1 , 5 in upa1Δ strains , respectively ) . Thus , in the absence of Upa1 , Rrm4 functions such as transport of mRNAs and associated ribosomes are disturbed . To test whether Rrm4-dependent septin mRNA transport was also affected ( Baumann et al . , 2014 ) , we used λN-based RNA live imaging by expressing a cdc3 mRNA with 16 BoxB in its 3′ UTR , and the λN RNA-binding peptide fused with a nuclear localisation signal ( NLS ) and triple Gfp ( λN*NLS-Gfp3 , Figure 7A; Baumann et al . , 2014 , 2015 ) . In this experimental set-up , unbound λN*NLS-Gfp3 was redirected to the nucleus , improving the cytoplasmic background signal ( Figure 7—figure supplement 1A; Video 20 ) . Consistent with earlier results , λN*NLS-Gfp3-labelled cdc3B16 mRNA co-localised with Rrm4-Cherry-positive endosomes ( Figure 7B; Baumann et al . , 2014 ) . However , deletion of upa1 resulted in impaired cdc3B16 mRNA transport ( Figure 7C–D ) . We observed fewer processive particles and those that were detected showed a lower range of movement ( Figure 7C–D ) . For the few examples that exhibited processive movement over a certain distance , the velocity was comparable to wild type ( Figure 7D ) indicating that the mRNAs can move with the speed of endosomes , but the attachment appeared to be less stable . 10 . 7554/eLife . 06041 . 045Figure 7 . Loss of Upa1 disturbs Rrm4-dependent transport of cdc3 septin mRNA and protein . ( A ) Schematic representation of components of the modified λN RNA reporter system ( Potef , constitutively active promoter; Tnos , heterologous transcriptional terminator; cdc3B16 carries 16 copies of boxB hairpin in its 3′ UTR; λN*NLS-Gfp3 , modified λN peptide fused to triple Gfp; and NLS; Baumann et al . , 2014 ) . ( B ) Dynamic co-localisation of strain expressing Rrm4-Cherry , λN*NLS-Gfp3 , and cdc3B16 . Kymograph with directed particles ( arrowheads ) . ( C ) Kymograph of upa1Δ strain expressing λN*NLS-Gfp3 protein and cdc3B16 mRNA . Occasionally directed particles are observed that exhibit altered processive movement . ( D ) Diagram showing range of movement of λN*NLS-Gfp3-labelled mRNAs ( vertical bar = mean , 151 mRNA particles for wt and 33 particles for upa1Δ ) on the left and velocity λN*NLS-Gfp3-labelled mRNAs on the right ( error bars , SD; 54 and 96 hyphae for wt and upa1Δ , respectively ) . ( E ) Micrographs of Cdc3G or Cdc3G/upa1Δ expressing hyphae ( maximum projection of z-stacks with 0 . 27 μm steps; size bar , 10 μm ) . Arrowhead marks gradient of septin filaments emanating from the hyphal tip . ( F ) FRAP analysis of Cdc3-Gfp or Cdc3-Gfp/upa1Δ expressing hyphae 7–10 h . p . i . ( about 10 μm from the hyphal tip; data were fitted to uniphasic exponential equation , dashed lines indicate half time of recovery; n = 3 independent experiments with 4–6 hyphae per experiment; error bars represent s . e . m . ) . Fluorescence is normalised to plateau ( Baumann et al . , 2014 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06041 . 04510 . 7554/eLife . 06041 . 046Figure 7—figure supplement 1 . Endosome-dependent movement of cdc3 mRNA and protein . ( A ) Hyphal tip of a strain expressing the λN*NLS-Gfp3 protein and cdc3B16 mRNA . Micrograph ( size bar , 10 μm ) and corresponding kymograph show directed particles ( arrowheads; Video 20 ) . ( B ) Micrographs ( size bar , 10 μm; asterisk marks retraction septum ) and corresponding kymographs of hyphae expressing Cdc3-Gfp ( top ) or Cdc3-Gfp/upa1Δ ( bottom ) . Arrowheads indicate Cdc3-Gfp-positive endosomes . DOI: http://dx . doi . org/10 . 7554/eLife . 06041 . 04610 . 7554/eLife . 06041 . 038Video 20 . Video of cdc3B16 particle visualised by λN*NLS-Gfp3 . Arrowhead at the beginning of the video marks the starting point of the moving particle ( size bar = 5 μm , timescale in seconds , 150 ms exposure time , 150 frames , 6 frames/s display rate , QuickTime format , 5495 kB ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06041 . 038 Rrm4 is needed for the correct localisation of Cdc3 protein on endosomes and in septin filaments forming a gradient that emanates from the hyphal tip ( Baumann et al . , 2014 ) . Analysing the subcellular localisation of functional Cdc3-Gfp revealed that its localisation on endosomes was severely disturbed in upa1Δ strains ( Figure 7—figure supplement 1B ) . Thus , without Upa1 , hardly any Cdc3 protein could be detected on endosomes . Also the subcellular localisation of Cdc3-Gfp in septin filaments was altered in upa1Δ strains . Similar to rrm4Δ strains , septin filaments were still formed , but the gradient at the hyphal tip was lost ( Figure 7E ) . To verify that the disturbed formation of septin filaments correlated with altered endosomal delivery of septin protein in the absence of Upa1 , we performed fluorescence recovery after photobleaching ( FRAP ) experiments to analyse hyphal tips ( Baumann et al . , 2014 ) . Due to the long maturation time of Gfp in the order of several minutes , local translation of newly synthesised protein at the hyphal tip can be excluded Baumann et al . , 2014 ) . Using the identical set-up as described before ( Baumann et al . , 2014 ) , we determined a half time of recovery ( t1/2 ) of 4 . 2 min for wild-type hyphae ( Figure 7F ) . In upa1Δ hyphae , t1/2 was substantially increased to 14 min ( Figure 7F , Baumann et al . , 2014 ) confirming that Upa1-dependent septin transport is crucial for efficient assembly into filaments at the hyphal tip . In essence , these results demonstrate consistently , that Upa1 is of specific importance for Rrm4-dependent endosome functions .
Aiming at the identification of endosomal components involved in mRNP transport , the PAM2 protein Upa1 caught our attention because of its FYVE and RING domains . This domain organisation is similar to Pib1p in S . cerevisiae and mammalian Rififylin , two proteins which appear to function in endosomal protein sorting . Although their precise roles are still unclear ( Shin et al . , 2001; Coumailleau et al . , 2004 ) , they might function in ubiquitination during protein sorting due to the presence of the RING domain found in RNF-type E3 ubiquitin ligases ( Nikko and Pelham , 2009 ) . The FYVE domain interacts with PI3P lipids and thereby targets proteins to endosomes and endocytotic vesicles ( Stenmark et al . , 2002; Lee et al . , 2005; Kutateladze , 2006 ) . Consistently , it was already demonstrated in U . maydis that the guanine GEF Don1 is targeted to Rab5a-positive endosomes in the yeast form via its FYVE domain . Don1 specifically regulates the small GTPase Cdc42 and its efficient endosomal delivery to the site of septation that is crucial for cytokinesis ( Schink and Bölker , 2009 ) . Here , we demonstrate that the FYVE domain of Upa1 is necessary to target the protein to the identical endosomal compartment , and that this localisation is essential for Upa1 activity specifically during filamentous growth ( Figure 8 ) . 10 . 7554/eLife . 06041 . 033Figure 8 . Upa1 functions specifically in endosomal mRNA transport . Model proposing Upa1 function during endosomal mRNA transport . Microtubules are given in blue , kinesin-3 type motor Kin3 transports endosomes in the plus-end direction . The small GTPase Rab5a ( magenta ) marks this specific endosomal compartment , described as early endosomes ( Higuchi et al . , 2014 ) . Upa1 binds endosomes via FYVE domain and the MLLE domain ( M ) of Pab1 via PAM2 motif ( P ) , as well as the MLLE domain of Rrm4 via PAM2L motif ( PL; for simplicity only one motif is shown ) . Note that the interaction of Upa1 with Pab1 is dispensable , whereas the interaction of Upa1 with Rrm4 is crucial for the endosomal localisation of septin mRNA ( blue line with poly[A]-tail ) , septin protein ( blue ) and ribosomes ( orange ) . A currently unknown adaptor protein is highlighted with a question mark . DOI: http://dx . doi . org/10 . 7554/eLife . 06041 . 03310 . 7554/eLife . 06041 . 034Figure 8—figure supplement 1 . Rrm4 and Upa1 homologues in fungi . ( A ) Domain architecture of proteins was predicted by analysing amino acid sequence of the proteins using SMART ( http://smart . embl-heidelberg . de/smart/set_mode . cgi ? NORMAL=1 ) and NCBI conserved domain search ( http://www . ncbi . nlm . nih . gov/Structure/cdd/wrpsb . cgi ) . Rrm4-homologues: U . maydis ( UMAG_10836 ) ; Sporisorium reilianum ( sr14312 ) ; Ustilago hordei ( UHOR_05154 ) ; Pseudomonas antarctica ( GAK64672 . 1 ) ; Coprinopsis cinerea ( XP_001832566 . 2 ) ; Coniophora putanea ( XP_007771597 . 1 ) , Punctularia strigosozonata ( XP_007384300 . 1 ) ; Tinea versicolor ( XP_008042363 . 1 ) , Acaulospora delicata ( XP_007341926 . 1 ) ; L . bicolor ( XP_001881076 . 1 ) ; Piriformospora indica ( CCA67340 . 1 ) ; Upa1 homologues: U . maydis ( UMAG_10836 ) ; S . reilianum ( sr13323 ) ; U . hordei ( UHOR_03485 ) ; P . antarctica ( GAC72163 ) ; C . cinerea ( XP_001837291 . 2 ) ; C . putanea ( XP_007767511 . 1 ) , P . strigosozonata ( XP_007382070 . 1 ) ; T . versicolor ( XP_008035292 . 1 ) , A . delicata ( XP_007337909 . 1 ) ; L . bicolor ( XP_001876756 . 1 ) ; P . indica ( CCA71703-CCA71704 ) , Malassezia globosa ( XP_001732453 . 1 ) ; S . cerevisiae ( YDR313C ) ; Schizosaccharomyces pombe ( NP_595987 . 1 ) ; Candida albicans ( XP_719309 . 1 ) ; Aspergillus nidulans ( ANID_11 , 932 . 1 ) ; Neurospora crassa ( CU08360 ) ; Homo sapiens ( NP_001017368 . 1 ) . ( B ) Alignment of the protein sequence of MLLE domains of the Poly ( A ) -binding proteins from U . maydis ( UmPab1; UMAG_03 , 449 ) and humans ( HsPab1; AAH23520 ) , as well as the N-terminal situated MLLE domains of Rrm4 ( UmRrm4; UMAG_10836 ) and homologues . DOI: http://dx . doi . org/10 . 7554/eLife . 06041 . 034 In contrast to Pib1p and Rififylin , Upa1 contains the aforementioned PAM2 , two PAM2Ls and five ankyrin repeats . The latter is a wide-spread protein–protein interaction motif of about 30 amino acids in length found in the human cytoskeletal protein Ankyrin ( Mosavi et al . , 2004; Li et al . , 2006 ) . Unfortunately , the ankyrin repeats in Upa1 escaped our analysis because of the instability of protein variants lacking the repeats ( Figure 2—figure supplement 2B ) . PAM2 motifs are found in a number of interaction partners of the mammalian protein PABPC1 , such as translational initiation factor eRF3 , nuclease subunit Pan3 and miRNA regulator GW182 ( Hoshino et al . , 1999; Uchida et al . , 2004; Tritschler et al . , 2010; Zekri et al . , 2013 ) . Structural and functional analysis revealed a conserved core motif ( consensus xxLNxxAxEFxP; Kozlov et al . , 2010 ) , which is inserted in the peptide-binding pocket of the MLLE domain ( Albrecht and Lengauer , 2004; Kozlov et al . , 2004; Jinek et al . , 2010 ) . In Upa1 , this motif is necessary and sufficient for interaction with the MLLE domain of Pab1 , confirming its bioinformatic identification . However , to our surprise the PAM2 motif is dispensable for Upa1 function , which could be explained with a redundant Pab1 interaction domain . Such a scenario was already described during miRNA regulation in Drosophila melanogaster . The key factor GW182 interacts with PABPC1 via two regions , the PAM2 sequence and a second region that provides indirect interaction with PABPC1 . Due to these redundant binding modes , the PAM2 motif of GW182 is functionally dispensable for PABP binding and silencing in D . melanogaster ( Eulalio et al . , 2009; Huntzinger et al . , 2010 ) . Relatedly , we found that Upa1 interacts with MLLE domains of Rrm4 via two PAM2-like sequences . This motif contains a conserved core D/E D/E D/EFVYP showing a similarity to the core of the PAM2 sequence ( EFxP ) including the essential phenylalanine that is inserted into a hydrophobic binding pocket of the MLLE domain ( Kozlov et al . , 2004 ) . Note , that mutations in the phenylalanine of both PAM2-like sequences in Upa1 resulted in loss of Rrm4 interaction ( Figure 3—figure supplement 2 ) . A phylogenetic sequence comparison revealed that closely related basidiomycetes , which contain an Rrm4 homologue with MLLE domain , also possess Upa1 homologues with at least one of the two PAM2L motifs ( Figure 8—figure supplement 1A ) . Experimentally , we found that although both RNA-binding proteins , Pab1 and Rrm4 , contain similar MLLE domains ( Becht et al . , 2005 , Figure 8—figure supplement 1B ) , they apparently differ in their sequence specificity ( Figure 3F; Figure 3—figure supplement 5 and Figure 3—figure supplement 6 ) . Thus , we hypothesize that in principle both PAM2 and PAM2L sequences are able to interact with MLLE domains but based on their specific interaction with Rrm4 , the novel PAM2L motifs are crucial for endosomal mRNP recruitment ( Figure 8 ) . In accordance , mutations in the PAM2L motifs lead to loss of Upa1 functionality ( Figure 3H ) . Based on the fact that Upa1 needs to be present on endosomes and interacts with the RNA-binding proteins Pab1 and Rrm4 , we hypothesize that Upa1 specifically functions during endosomal mRNP targeting ( Figure 8 ) . Importantly , we can exclude that Upa1 is necessary for other known biological functions of this Rab5a-positive endosomal compartment , since septum formation , cytokinesis , FM4-64 uptake , or Rab5a shuttling are not altered in the absence of Upa1 ( Wedlich-Söldner et al . , 2002; Schink and Bölker , 2009; Baumann et al . , 2012 ) . Consistently , also Rrm4 is dispensable for basic functions of endosomes , such as their movement or the association of Rab5a to endosomes ( Baumann et al . , 2012 ) . Importantly , we observed specific defects in Rrm4-dependent endosomal mRNA transport in the absence of Upa1: ( i ) the amount of aberrant bipolar growing hyphae is increased; ( ii ) cts1 secretion is specifically disturbed in the hyphal form; ( iii ) processive movement of Rrm4 is disturbed; ( iv ) endosomal transport of the mRNA indicator Pab1 , as well as of mRNA-associated ribosomes is strongly reduced; ( v ) septin mRNA and protein are strongly affected in endosomal transport; ( vi ) efficient delivery of septin protein to the hyphal tip is disturbed . These observations all point toward the conclusion that Upa1 has a specific function in endosomal targeting of Rrm4 ( Figure 8 ) . In the absence of Upa1 , endosomal Rrm4 functions are disturbed , and consequently , mRNA and associated ribosomes are transported less efficiently . In line with this , specific functions of Rrm4 in septin mRNA and protein transport are also affected , overall leading to characteristic defects of hyphal functions , such as unipolar growth and Cts1 secretion . Thus , Upa1 is a key factor in endosomal recruitment of Rrm4 . However , since endosomal movement of Rrm4 was not completely abolished in the absence of Upa1 , we envision the presence of additional factors involved ( Figure 8 ) . Noteworthy , the observed defects in Rrm4 function are fully consistent with our earlier model that Rrm4 mediates local translation of transported mRNAs for the endosomal transport of the translation products ( Figure 8; Baumann et al . , 2014; Jansen et al . , 2014 ) . Recent data suggest that there is a close connection between mRNA and membrane trafficking ( Kraut-Cohen and Gerst , 2010; Jansen et al . , 2014; Berkovits and Mayr 2015 ) . Prominent examples are actin-dependent co-transport of mRNAs and ER during budding in S . cerevisiae ( Schmid et al . , 2006 ) , endosomal transport of viral RNA ( Ghoujal et al . , 2012 ) or endosomal miRNA-dependent processes ( Kim et al . , 2014 ) . In this respect , one of the key questions is how mRNPs are connected to membranes . Mammalian p180 , for example , contains a lysine-rich RNA-binding domain in concert with a membrane spanning domain and thereby attaches mRNAs to the surface of ER membranes to support their local translation ( Cui et al . , 2012 ) . She2p from yeast harbours a lipid-binding domain and is able to recognise membrane curvature supporting specific ER association of mRNAs during their transport to daughter cells ( Genz et al . , 2013 ) . Furthermore , neuronal PICK1 contains a banana-shaped BAR ( Bin-Amphiphysin-Rvs ) domain for interaction with curved membranes . Recently , it was shown that this endosome-associated factor specifically interacts with Argonaute 2 , a core component of the miRNA machinery . This provides a mechanism of how miRNAs can be attached to endosomes to carry out specific functions , such as miRNA assembly or translational regulation in neurons ( Antoniou et al . , 2014 ) . Here , we demonstrate that a FYVE domain protein directly couples the key RNA-binding protein of mRNA transport to endosomes by novel PAM2L motifs . Thereby , mRNPs and associated ribosomes are attached to endosomes during microtubule-dependent trafficking . This transport process is important in distributing mRNAs and ribosomes throughout the highly polarised cells ( König et al . , 2009; Baumann et al . , 2012; Higuchi et al . , 2014 ) , as well as in the delivery of translation products such as septins to the growing tip ( Baumann et al . , 2014; Jansen et al . , 2014 ) . In essence , we provide first mechanistic insights into how mRNPs and associated ribosomes are attached to endosomes during long-distance transport .
E . coli K-12 derivates DH5α ( Bethesda Research Laboratories ) and Top10 ( Life Technologies , Carlsbad , CA , USA ) were used for cloning purposes . S . cerevisiae strain AH109 ( Clontech Laboratories Inc . , Mountain View , CA , USA ) was used for yeast two-hybrid analyses . Transformation and cultivation were performed using standard techniques . Growth conditions for U . maydis strains and source of antibiotics were described elsewhere ( Brachmann et al . , 2004 ) . Strains were constructed by the transformation of progenitor strains with linearized plasmids . All homologous integration events were verified by Southern blot analysis ( Brachmann et al . , 2004 ) . For ectopic integration , plasmids were linearized with SspI and targeted to the ipS locus ( Loubradou et al . , 2001 ) . Genomic DNA of wild-type strain UM521 ( a1b1 ) was used as a template for PCR amplifications unless otherwise noted . Detailed information is given in Supplementary files 2–7 . Accession numbers of U . maydis genes used in this study: upa1 ( UMAG_12183 ) , rrm4 ( UMAG_10836 ) , pab1 ( UMAG_03494 ) , kin3 ( UMAG_06251 ) , rab5a ( UMAG_10615 ) , yup1 ( UMAG_05406 ) , rps2 ( UMAG_05139 ) , rpl25 ( UMAG_05998 ) , rps19 ( UMAG_11551 ) , and cdc3 ( UMAG_10503 ) . Cell suspensions were grown for about 12 hr in 3 ml CM supplemented with 1% glucose ( glc ) at 28°C . 4 μl of the densely grown cells were spotted on NM-glc plates containing 1% ( wt/vol ) activated charcoal , sealed with parafilm and incubated at 28°C for 48hr . Pictures were taken using a Stemi 2000C stereomicroscope ( Zeiss , Oberkochen , Germany ) with a mounted Canon PowerShot A650 IS camera ( Canon Germany GmbH , Krefeld , Germany ) . U . maydis cell suspensions were grown at 28°C for about 12 hr in 20 ml CM supplemented with 1% glucose ( glc ) and set to an OD600 of 0 . 5 . Either sporidial cells were directly measured or filamentous growth was induced by shifting to NM ( 1% glc ) and subsequent incubation at 28°C for 6–9 hr . 30 μl of the culture was mixed with 70 μl 0 . 25 μM 4-Methylumbelliferyl β-D-N , N′ , N″-triacetylchitotrioside ( Sigma–Aldrich , Taufkirchen , Germany ) , a specific substrate for endochitinolytic activity . After incubation for 1 hr ( protected from light ) , the reaction was stopped by adding 200 μl 1M Na2CO3 . Enzymatic activity was measured by detecting the fluorescent product with a fluorescence spectrometer Infinite M200 ( Tecan Group Ltd . , Männedorf , Switzerland ) using an excitation and emission wavelength of 360 nm and 450 nm , respectively . At least three independent biological experiments were performed with three technical replicates per strain ( Koepke et al . , 2011 ) . The two-hybrid system Matchmaker 3 from Clontech was used . Yeast two-hybrid strains were co-transformed with derivates of pGBKT7-DS and pGADT7-Sfi ( Supplementary file 5 ) and were grown on SD plates without leucine and tryptophan at 28°C for 4 days . Transformants were patched on SD plates without leucine and tryptophan ( control ) or on SD plates without leucine , tryptophan , histidine , and adenine ( selection ) . Plates were incubated at 28°C for 3 days to test for growth under selection condition . For qualitative plate assays , cells ( SD -leu , -trp , OD600 of 0 . 5 ) were diluted with sterile water in 1:5 steps , and 4 μl drops were spotted on control and selection plates and incubated at 28°C for 3 days . Colony growth was documented with a LAS 4000 imaging system ( GE Healthcare Life Sciences , Little Chalfont , United Kingdom ) . Expression of hybrid proteins was analysed by Western blot ( see below ) . Preparation of protein extracts of U . maydis and S . cerevisiae cells was carried out according to published protocols ( Baumann et al . , 2012; Clontech ) . For the latter , cells were grown over night at 28°C in SD -leu-trp medium to an OD600 of about 0 . 75 . The exact OD600 was recorded and together with the culture volume used to calculate the OD-units ( 50 ml × OD600 of 0 . 75 = 37 . 5 OD units ) . Cells were harvested by centrifugation ( 2000×g , 5 min ) and resuspended in 100 μl yeast cracking buffer ( 40 mM Tris–HCl [pH 6 . 8] , 8 M Urea , 5% [wt/vol] SDS , 0 . 1 mM Na2-EDTA , 0 . 4 mg/ml bromophenol blue , 0 . 1% [vol/vol] β-mercaptoethanol , 7% [vol/vol] benzamidine , and 5% [vol/vol] PMSF ) per 7 . 5 OD units ( e . g . , 37 . 5 OD-units/7 . 5 = 500 μl YCB ) . The suspension was transferred to 2 ml reagent tubes , 100 μL of glass beads added , and the sample boiled at 99°C , while shaking ( 1000 rpm ) . Immediately , the cells were cooled on ice and subsequently analysed by Western Blot or stored at −70°C . For Western blotting protein samples were resolved by 8% SDS-PAGE and transferred to a PVDF membrane ( GE Healthcare ) by semi-dry blotting . Western blot analysis was conferred with anti-GFP ( clones 7 . 1 and 13 . 1 ) , anti-c-Myc ( clone 9E10; Roche ) , anti-alpha-Tubulin ( clone DM1A ) , anti-HA ( clone 12CA5 ) and anti-GST ( Sigma ) antibodies . A mouse IgG HRP conjugate ( H+L; Promega , Madison , WI ) was used as a secondary antibody . Activity was detected using the AceGlow blotting detection system ( Peqlab , Erlangen , Germany ) . In order to identify crucial amino acids for the interaction of Upa1 with Rrm4 , we performed a linker scanning mutagenesis of plasmid pGBKT7-Upa1ΔN7/ΔFR-Gfp ( aa 883 to 1947 ) resulting in 10 amino acid substitutions of the sequence AASAAATAAS . Serines and threonine were introduced by the Golden Gate cloning system ( Terfrüchte et al . , 2014 ) to prevent potential translational problems resulting from 10 consecutive alanines . pGBKT7-Upa1ΔN7/ΔFR-Gfp was used as a template for two PCR reactions amplifying two products , which lie directly upstream and downstream of the targeted sequence of 30 nucleotides . Oligonucleotide combinations u2 and p2 were used for the upstream situated sequence and combinations d2 and p1 for the downstream sequences ( Supplementary file 7 ) . The resulting products were subcloned in pDest ( pUMa1467 , Terfrüchte et al . , 2014 ) using BsaI . From these storage plasmids , the mutagenized alleles were introduced into pGBKT7-Upa1ΔN7ΔFRGfp as a SfiI/SfiI-fragment . Mutagenesis was verified by an introduced SacII-restriction site as well as by sequencing . Derivates of plasmids pGEX and pET15B ( Supplementary file 6 ) were transformed into E . coli Rosetta . Overnight cultures were diluted 1:50 in a final volume of 100 ml . Protein expression was induced with IPTG for 4 hr . Cells were pelleted , resuspended in 10 ml lysis buffer ( 20 mM Tris-Cl , pH 7 . 5; 200 mM NaCl; 1 mM EDTA; pH 8 . 0; 0 . 5% Nonidet P-40; 1 tablet protease inhibitor per 50 ml; Roche , Mannheim , Germany ) and lysed by sonication . 50 μl glutathione beads ( GE Healthcare ) were washed 3 times with lysis buffer . For each pulldown , 500 μl cell lysate with GST-tagged protein was added to the washed beads , incubated for 2 hr at 4°C and subsequently washed 5× with lysis buffer . 1 ml cell lysate containing different Upa1 variants was added directly to loaded GST columns , incubated for 1 hr at 4°C and subsequently washed 5 times with lysis buffer . Beads were boiled 6 min at 99°C . 10 μl of each fraction was loaded on SDS-PAGE for analysis and stained with Coomassie blue . For protein purification , GST-purification was performed as described above . For GST-tagged proteins , 1 . 5 ml glutathione beads ( GE Healthcare ) were equilibrated with lysis buffer . Cell lysate was loaded onto the columns , incubated , and washed . The GST-tagged proteins were eluted in elution buffer ( 50 mM Tris-HCl; pH 7 . 5; 200 mM NaCl; 20 mM Glutathione ) and glutathione was removed via PD-10 Desalting columns ( GE Healthcare ) . For the His6-tagged Upa1 variants , cells were pelleted , resuspended in His-lysis buffer ( 50 mM NaH2PO4; 300 mM NaCl; 10 mM imidazole; pH 8 . 0 ) and lysed by sonication . Cell lysates were loaded onto 1 . 5 ml Ni-NTA agarose columns ( Qiagen , Hilden , Germany ) and incubated for 1 hr at 4°C . The Ni-NTA agarose was washed 3× with His-lysis buffer ( increasing Imidazole concentration from 10 to 20 mM ) . His-tagged proteins were eluted in His-lysis buffer ( containing 250 mM imidazole ) . The pulldown experiments with purified proteins were performed similar as described above . Glutathione beads were loaded with100 μl purified GST-tagged proteins and 400 μl lysis buffer . Loaded GST-columns were incubated with 200 μl purified His6-tagged Upa1 variants and 300 μl lysis buffer . Proteins were eluted by boiling , and 10 μl of each fraction was analysed after SDS-PAGE by colloidal Coomassie staining and Western blotting . For the Western blots , membranes were probed with αHis antibody ( 1/10 , 000; Sigma ) and αGST antibody ( 1/5 , 000; Molecular Probes ) . Standard microscopy was carried out with our set-up as described before ( Baumann et al . , 2014 ) . 20 ml cultures of cells were grown to an OD600 of 0 . 5 in CM supplemented with 1% glucose ( glc ) and shifted to NM ( 1% glc ) to induce filamentous growth for 6–9 hr . For quantification of bipolarity , hyphae were observed with a 63× Plan-Apochromat objective in combination with a Spot Pursuit CCD camera . Pictures of more than 100 cells were taken and scored for unipolar or bipolar growth , as well as for septum formation . At least three independent experiments were performed . Staining of hyphae with FM4-64 was done as described elsewhere ( Baumann et al . , 2012 ) . Briefly , 500 μl of filament suspension were labelled in 0 . 8 μM FM4-64 ( Life Technologies ) . After 30–60 s of incubation at room temperature , samples were subjected to microscopic analysis . For analysis of signal number , velocity and passages through a defined zone of Gfp fusion protein hyphae were observed with a 63× Plan-Apochromat ( NA 1 . 4 ) in combination with a Spot Pursuit CCD camera . Videos were recorded with an exposure time of 200 ms and 150 frames taken . Kymographs were generated from these videos and analysed using Metamorph ( Version 7 . 7 . 0 . 0; Molecular Devices , Seattle , IL , USA ) . Signals were counted manually discriminating between processive movement , corralled movement ( covered distance of hypha per 22 . 5 s < 5 μm; rapid changes of direction ) , or static signals ( no movement , straight line ) . Velocity was determined by quantifying processive signals ( movement >5 μm ) . Note that one signal could exhibit different speeds ( i . e . , upon reversal of direction ) . Those velocities were handled as individual data points and not averaged . Passing events were quantified at two defined regions: 10 μm from the apical tip or in the middle of the filament . The number of passing events reflects the overall crossing of signals through this zone . All parts of the microscope systems were controlled by the software package MetaMorph ( version 7; Molecular Devices ) , which was also used for image processing including the adjustment of brightness and contrast , as well as measurements , quantifications , kymographs , and maximum projections of z-stacks . Fluorescence micrographs are displayed inverted unless otherwise stated . For RNA live imaging , we improved our λN-based system ( König et al . , 2009; Baumann et al . , 2014 , 2015 ) and fused protein λN*Gfp3 to a NLS resulting in λNNLSGfp3 . Free λNNLS-Gfp3 , which does not bind cytoplasmic cdc3B16 mRNA , is targeted to the nucleus , thereby improving the signal to noise ratio in the cytoplasm substantially . 20 ml culture was grown to an OD600 of 0 . 5 in CM supplemented with 1% glucose ( glc ) and shifted to NM ( 1% glc ) to induce filamentous growth for 9 hr . For excitation of Gfp , the 488 nm laser line was set to 60% . Hyphae were observed with a 63× Plan-Apochromat ( NA 1 . 4 ) in combination with a CoolSNAP HQ2 camera . Each video was recorded with 150 ms/frame and contained 150 frames . For quantification of directed movement , kymographs were generated to study the number , velocity , range , and direction of particle movement . For analysis of directionality , particles that reversed direction were counted twice . To determine the average number of particles per 100 μm of hyphae , the total length of hyphae was measured and divided by the number of particles . For AB33λNNLS-Gfp3/Potefcdc3B16 , we counted 151 signals in 5024 μm corresponding to about 54 hyphae with an average length of 94 μm . In AB33λNNLS-Gfp3/Potefcdc3B16/upa1Δ 33 , particles were detected in 6610 μm corresponding to 96 hyphae with an average length of 69 μm . Photobleaching was adapted from our previous publication ( Baumann et al . , 2015 ) . In order to visualise moving ribosomal proteins , 15 μm of the respective hyphae were photobleached prior to detection of Gfp fluorescence . The 405 nm laser was set to 29% output power . Total bleach time was 77 ms . The 488 nm laser was set to 50% output ( exposure time 150 ms , binning 2 ) . Design and analysis of FRAP experiments was previously described ( Baumann et al . , 2014 , 2015 ) . An area of 16 μm from hyphal tips was bleached with 8 . 3% laser power . The beam diameter was set to13 pixels , and the bleach time was 7 ms per pixel . Bleaching was carried out in 11 z-planes through fungal hyphae with a z-distance of 0 . 5 μm . Fluorescence recovery was acquired with an exposure time of 500 ms in a z-stack of 11 planes with a z-distance of 0 . 5 μm ( open camera shutter ) . Every minute , a z-stack was collected for a period of 65 min . | DNA contains the instructions to build proteins . These instructions are first copied to make a molecule of messenger RNA ( or mRNA for short ) . A large machine called the ribosome then reads the mRNA molecule and translates it to build a protein . Many proteins must get to particular locations in a cell to carry out their roles . For some proteins , this is achieved by transporting the mRNAs to the right location before they get translated , via a process called ‘mRNA trafficking’ . However , mRNAs do not move by themselves; instead they bind to a host of mRNA-binding proteins , and the ribosomes that are required for translation to take place . Cells also move proteins between different locations using small bubble-like structures called vesicles . These vesicles are surrounded by a membrane , and so this process is known as ‘membrane trafficking’ . Previous work has shown that these two processes are often linked , as vesicles can also carry mRNA molecules . But it is not fully understood how mRNA molecules are connected to vesicles . Now , Pohlmann et al . have used a fungus called Ustilago maydis as a model system to investigate how mRNAs and vesicles can move together in cells that grow to form filament-like structures called hyphae . This fungus uses these filaments to penetrate into plant tissues and causes a disease called corn smut . The experiments revealed a vesicle protein called Upa1 that contains a new type of binding site that allows Upa1 to bring an important RNA-binding protein to the surface of vesicles . Since the RNA-binding protein binds mRNA and the translating ribosomes , this can explain how mRNAs can associate with membranes to move together along hyphae . When Pohlmann et al . engineered fungi that lacked the gene for Upa1 , these mutants had problems transporting their mRNAs and associated ribosomes . These findings reveal a direct connection between mRNA trafficking and membrane trafficking . Future studies could now investigate whether similar processes take place in other cells that grow as long filaments , such as plant pollen tubes or nerve cells . These studies might provide new insights into plant reproduction or brain activity . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology",
"microbiology",
"and",
"infectious",
"disease"
] | 2015 | A FYVE zinc finger domain protein specifically links mRNA transport to endosome trafficking |
We have discovered a new mechanism of monoallelic gene expression that links antigenic variation , cell cycle , and development in the model parasite Trypanosoma brucei . African trypanosomes possess hundreds of variant surface glycoprotein ( VSG ) genes , but only one is expressed from a telomeric expression site ( ES ) at any given time . We found that the expression of a second VSG alone is sufficient to silence the active VSG gene and directionally attenuate the ES by disruptor of telomeric silencing-1B ( DOT1B ) -mediated histone methylation . Three conserved expression-site-associated genes ( ESAGs ) appear to serve as signal for ES attenuation . Their depletion causes G1-phase dormancy and reversible initiation of the slender-to-stumpy differentiation pathway . ES-attenuated slender bloodstream trypanosomes gain full developmental competence for transformation to the tsetse fly stage . This surprising connection between antigenic variation and developmental progression provides an unexpected point of attack against the deadly sleeping sickness .
Functional variation between cells in a population is often achieved by selective expression of one protein from a pool of possibilities . To achieve this goal , a variety of mechanisms have evolved that ensure allelic exclusion , namely the silencing of expression of all but one member of a gene family , either temporarily or for the remainder of the life of the cell . In the mammalian central nervous system , each olfactory sensory neuron expresses only one olfactory receptor ( OR ) from a family of ∼1200 genes ( Buck and Axel , 1991 ) . Before an OR gene is expressed , all alleles are silenced and converted to heterochromatin ( Magklara et al . , 2011 ) . A limiting enzymatic activity then stochastically removes the heterochromatin marks from one allele to activate it . The expressed OR protein mediates a feedback loop that inhibits removal of heterochromatin marks from all other alleles , preventing their transcription ( Serizawa et al . , 2003; Lyons et al . , 2013 ) . Allelic exclusion commonly occurs in pathogens that exploit antigenic variation of their cell surface proteins to keep ahead of the host immune response , usually as part of a population survival strategy . The malaria parasite Plasmodium falciparum , for example , expresses only one out of 60 members of the var gene family , each coding for different versions of the surface virulence factor PfEMP1 ( Guizetti and Scherf , 2013 ) . Monoallelic expression of the variant surface glycoprotein ( VSG ) in Trypanosoma brucei is a particularly striking example of allelic exclusion in a pathogen . In the mammalian host , the cell surface is covered with millions of copies of VSG that form a dense layer that is virtually impervious to host-derived antibodies ( Cross , 1975; Engstler et al . , 2007; Schwede et al . , 2011 ) . VSGs are highly immunogenic and provoke a rapid and efficient immune response that diminishes the parasite population . Only trypanosomes that have successfully switched to expression of another , structurally similar but immunologically distinct VSG survive . Trypanosomes possess several hundred VSG genes ( Berriman et al . , 2005 ) . Their potentially unlimited capacity for antigenic variation forms the basis of trypanosome persistence and virulence . VSG genes are expressed from one of ∼15 telomeric expression sites ( ES ) but only one of these is transcribed at any given time ( Hertz-Fowler et al . , 2008 ) . Chromatin remodeling appears to play an important role in maintaining the monoallelic expression of the active ES ( Horn and McCulloch , 2010 ) . Antigenic variation can result from either a change in the active ES , usually gene conversion at the VSG locus , or by an epigenetic change that results in silencing of the active ES and transcription of a previously silent ES ( in situ switch ) . No mechanism or factor has been identified that is involved in the initiation of the latter . The complex life cycle of T . brucei represents a succession of proliferative and quiescent developmental forms , which vary widely in cell architecture and function ( MacGregor et al . , 2012 ) . Throughout the parasite's life cycle , the plasma membrane is covered with a series of different surface coats . Antigenic variation only occurs in the mammalian host , and not in the transmitting tsetse fly . Consequently , the trypanosome bloodstream form ES is exclusively functional in the proliferating slender bloodstream stage of the parasite . In response to quorum sensing , the trypanosomes differentiate to the quiescent stumpy form . Stumpy forms are competent for the next developmental transition to procyclic forms , which occurs after ingestion by a tsetse fly ( Reuner et al . , 1997 ) . This developmental transition is accompanied by silencing of the active ES so that allelic exclusion is lost . How this silencing is achieved is unknown . In this study , we show that high-level expression of an ectopic VSG transgene is sufficient to initiate the directional silencing of the active ES from the telomere inwards . The silencing of the active ES is dependent on the histone methyltransferase DOT1B . Thus , the VSG itself mediates a feedback loop that controls the state of monoallelic gene expression . Furthermore , we show that transcriptional activity of the ES directly determines cell proliferation and developmental competence via a subset of expression site associated genes ( ESAGs ) . Thus , the trypanosome ES can be regarded as a fine-tuned regulator at the crossroads of antigenic variation , proliferation , and development .
To simulate a transcriptional switching event ( in situ switch ) in the most straightforward manner , we adopted a strategy that allows inducible high-level expression of VSG121 in trypanosomes that have VSG221 in their active expression site ( 221ES ) . This was achieved by introducing a VSG121 gene under the control of a tetracycline-inducible T7 RNA polymerase promoter ( 121tet ) into an rDNA spacer locus ( Figure 1A ) . We decided to use the T7 RNA polymerase-based expression system instead of a polymerase I-based system to minimize interference with the endogenous transcription machinery . Quantitative Northern blot analysis revealed an almost instantaneous increase of VSG121 mRNA , reaching 82 ( ±1 ) % of wild type levels within 2 hr ( Figure 1B ) . This was followed by a reduction of endogenous VSG221 transcripts , which declined with kinetics compatible with the mRNA half-life ( Ehlers et al . , 1987 ) . The total VSG mRNA initially peaked at 180% , but within 8 hr leveled to approximately wild-type quantities . Thus , the VSG mRNA population was effectively exchanged in this time period . 10 . 7554/eLife . 02324 . 003Figure 1 . Inducible VSG121 expression leads to endogenous VSG silencing . ( A ) Illustration of the ectopic VSG expression strategy . VSG121 integration into the transcriptionally silent rDNA spacer is achieved in the absence of tetracycline by using a constitutive 10% T7 promoter driving the BLE resistance cassette . High-level expression of ectopic VSG121 occurs upon tetracycline induction of a full T7 promoter . ( B ) Quantification of VSG mRNA and protein levels during the course of tetracycline-induced VSG121 expression . The values are percentages ± SD for two independent clones normalized to the parental 221ES or VSG121 wild-type cells . Total RNA samples were prepared at the time points indicated and analyzed by dot blotting . The blots were hybridized with VSG-specific fluorescent DNA probes and quantified by normalization to beta-tubulin mRNA using a Licor Odyssey near infrared scanner . Protein equivalents of 6 × 105 cells were dot blotted and incubated with anti-VSG121 or anti-VSG221 antibodies . Quantification was done by normalization to the paraflagellar rod ( PFR ) protein using the Licor Odyssey system . DOI: http://dx . doi . org/10 . 7554/eLife . 02324 . 00310 . 7554/eLife . 02324 . 004Figure 1—figure supplement 1 . VSG immunofluorescence after tetracycline-induced VSG121 expression . Double indirect immunofluorescence with mouse anti-VSG121 ( green ) , rabbit anti-VSG221 ( red ) antibodies and DAPI staining ( blue ) was performed 0 , 12 , and 24 hr after tet induction . 121wt and parental 221ES cells served as controls for antibody specificity . Scale , 20 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 02324 . 004 The VSG121 protein was detectable within 4–6 hr after induction ( Figure 1B ) . After 24 hr , the surface coat of all cells was dominated by VSG121 ( Figure 1—figure supplement 1 ) and the amount of VSG221 had declined to 21 ( ±2 ) % ( Figure 1B ) . Thus , expression of an ectopic VSG at wild-type quantities caused a reduction in the abundance of the ES-resident VSG , which decreased with a half-life of 12 hr that corresponds to the population doubling time within the first 24 hr of induction . The reduction in steady state levels of VSG221 mRNA was caused by an active process , which specifically affects the ES-resident VSG , either through reduction of transcription or by VSG mRNA degradation . To distinguish between these possibilities , we quantified other mRNAs derived from the ES . As transcription of the ES is polycistronic , any transcriptional silencing would also affect genes upstream of the VSG . Initially , the levels of three ESAG mRNAs were measured over a period of 8 hr after VSG121 induction ( Figure 2A , B ) . It appears that mRNAs from the telomere proximal ESAG1 and 2 genes decreased faster than mRNA from the telomere distal ESAG12 gene . To examine whether this effect was the result of different mRNA half-lives , we inserted the same GFP transgene either upstream of the VSG gene ( promoter distal ) or next to the ES promoter ( promoter proximal ) ( Figure 2A ) . Upon induction of VSG121 expression , the promoter proximal GFP mRNA decreased with significantly slower kinetics ( p<0 . 018; two-way ANOVA test ) than that of the promoter distal GFP ( Figure 2C , left panel ) , revealing that the kinetics of GFP mRNA reduction was determined solely by the position of the GFP gene within the active ES . Thus , VSG121 overexpression caused a gradual loss of ES activity that started at the telomere and propagated towards the VSG promoter . This process was completed within one cell division cycle and was maintained for 120 hr ( Figure 1B ) . Then the ES was gradually reactivated in the reverse direction ( Figure 2C ) . The ES was not silenced completely , so the effect did not constitute a bona fide ES silencing , but a fine-tuned ES attenuation . We postulate that this VSG-induced attenuation represents the initiation step of an antigenic in situ switch . 10 . 7554/eLife . 02324 . 005Figure 2 . Ectopic VSG121 expression induces gradual ES attenuation . ( A ) Scheme illustrating the anatomy of the VSG221 expression site in a set of reporter lines . Cell line 221ES . 121tet was used to quantify three different ESAG transcripts . In GFPESpro-221ES . 121tet , a GFP reporter gene was inserted just downstream of the ES promoter and in GFPEStel-221ES . 121tet just upstream of the endogenous VSG221 . NEO , neomycin resistance; BLAS , blasticidin resistance; numbered boxes , ESAGs; arrow , ES promoter . ( B ) ESAG1 and 2 transcripts decrease faster than ESAG12 mRNA . After 0 , 4 , and 8 hr of ectopic VSG121 induction , mRNA levels of three ESAGs were quantified using 32P-labeled probes on Northern blots and normalized to 18S rRNA using a fluorescent probe . Values are relative to the parental 221ES cells ( P ) . ( C ) Expression site attenuation starts at the telomere and is released in the reverse direction . The GFP reporter lines were induced for ectopic VSG121 expression and mRNA levels of GFP , VSG221 and VSG121 were quantified using fluorescently labeled probes , normalized to beta-tubulin mRNA ( upper panel ) . Values are relative to the non-induced levels ± SD for two independent clones ( lower panel ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02324 . 005 The immediate early effect of VSG121 overexpression was the attenuation of the active ES ( Figures 1 and 2 ) . Parasite growth was not affected for the first 24 hr of tetracycline induction , but then cell proliferation slowed down significantly ( Figure 3A ) . The cell cycle was analyzed by determining the nucleus ( N ) and kinetoplast ( K ) configuration and showed an accumulation of 1K1N cells with the kinetoplasts having a pre-replication morphology and a reduction of cells containing a dividing kinetoplast ( Figure 3B , C ) . These observations are consistent with the slowed growth resulting from a G1 phase prolongation . To test this model , the percentage of cells in S-phase was determined at different times after induction of VSG121 expression . Cells were incubated with the nucleoside-analogue EdU for either 4 or 18 hr ( Figure 3D ) . In the control ( 0 hr ) , 80% of trypanosomes had entered nuclear S-phase within the 4-hr labeling window . In contrast , after 24 hr of VSG121 expression only 17% incorporated EdU into nuclear DNA within an equivalent 4-hr window . However , when the parasites were incubated with EdU for 18 hr , 86% had entered S-phase . Thus , the cells were clearly not arrested in the cell cycle , but showed a specific , at least fivefold expansion of the G1-phase from 3 to 15 hr . Remarkably , all trypanosomes remained fully motile and no morphological abnormality or cell death was observed . The marked G1-phase prolongation obviously required a reduction in metabolic flux and anabolic pathways , including protein synthesis , which in turn might be reflected in decreased overall RNA synthesis . Therefore , we measured total RNA synthesis by labeling nascent transcripts with BrUTP and detection with an anti-BrdU antibody . After 8 hr of tetracycline induction , when VSG221 transcripts had already decreased to 20% , the general transcription rate was not impaired at all . Only after 24 hr was the global transcription reduced to 30% ( p<0 . 0001; unpaired t test ) ( Figure 3E ) . This formally proves that ES attenuation precedes growth retardation and general transcriptional diminution . To test if RNA polymerase I and II transcription was both affected , we inserted luciferase reporter transgenes either downstream of an rDNA promoter or into the tubulin-locus ( Figure 3—figure supplement 1 ) . In both the cases , luciferase activity was down to 30–50% within 2 days . Thus , all tested regions of the genome were slowly silenced . This also explains the gradual decrease in the amount of ectopically expressed VSG121 mRNA , which started after 24 hr of tetracycline induction ( Figure 1B ) that is at a time when the active VSG221 ES was already attenuated . 10 . 7554/eLife . 02324 . 006Figure 3 . Expression site attenuation causes G1 retardation and transcriptional shut-down . ( A ) Cumulative growth curve after induction of ectopic VSG121 expression . Two independent clones were analyzed for 9 days in the absence ( −tet ) or presence ( +tet ) of 1 µg/ml tetracycline . ( B ) Cell cycle analysis . 221ES . 121tet cells were induced for VSG121 expression and fixed at the time points indicated . The configurations of the mitochondrial genome ( kinetoplast , K ) and the nucleus ( N ) were visualized by light microscopy following staining with DAPI . The configuration of K and N was subdivided into 1K1N for cells having single copies of each organelle , 1Kv1N for those with an early dividing kinetoplast and 1Kb1N with a late dividing kinetoplast . Cells in G2/M phase reveal 2K1N configuration and post-mitotic trypanosomes are marked by 2K2N . Abnormal configurations are scored as ‘other’ . For each time point , n >300 cells were analyzed . Significance was determined using Fisher's exact test . ( C ) Cell cycle retardation has no effect on cell morphology . Trypanosomes before ( −tet ) and 3 days after VSG121 induction ( +tet ) were stained with DAPI and the cell surface labeled using AMCA-sulfo-NHS . The arrowhead points to a 1Kv1N cell . Scale bar: 5 µm . ( D ) Expression site attenuation leads to reduced growth , but not to cell cycle arrest . Incorporation of EdU into 221ES . 121tet cells induced for VSG121 expression . The trypanosomes were induced for ectopic VSG121 expression and at 0 , 1 , 2 , and 3 days , they were incubated with 50 µM EdU for both 4 and 18 hr . After chemical fixation the cells were co-stained with DAPI and analyzed by light microscopy ( n >100 each ) . ( E ) Quantification of RNA synthesis rate after ectopic VSG121 expression . After 0 , 8 , 24 , and 48 hr of tetracycline induction , cells were incubated in the presence of 0 . 5 mM BrUTP for 15 min at 33°C , subsequently chemically fixed and incubated with a monoclonal mouse anti-BrdU and an Alexa488-coupled goat anti-mouse antibody . The intensities of 3D images ( 100 images; z-step 100 nm ) were measured in summed slice projections using the ImageJ software and are expressed as arbitrary units . To exclude variations in signal intensities due to different positions within the cell cycle , only G1 cells were analyzed . After 8 hr of tet induction , no altered fluorescence intensity was observed . After both 24 and 48 hr , fluorescence intensity was significantly decreased to 30% ( p<0 . 0001; unpaired t test ) . For each time point more than 60 cells were analyzed . Red lines and error bars show mean ± SD . DOI: http://dx . doi . org/10 . 7554/eLife . 02324 . 00610 . 7554/eLife . 02324 . 007Figure 3—figure supplement 1 . Growth retardation is accompanied by a reduced transcriptional rate of Pol I and Pol II . Ectopic expression of VSG121 causes reduction in gene expression . ( A ) Luciferase reporters were inserted downstream of an rDNA polymerase I promoter and ( B ) into the polymerase II-transcribed tubulin locus . Values are given as mean RLU ± SD of 1 × 105 cells for three ( Luctub . 221ES . 121tet ) and two ( LucrPro . 221ES . 121tet ) independent clones . DOI: http://dx . doi . org/10 . 7554/eLife . 02324 . 007 The histone methyltransferase DOT1B is involved in the silencing of the active ES during in situ switching ( Figueiredo et al . , 2008 ) . To test if DOT1B was also required for the attenuation of the active ES upon ectopic VSG expression , we inserted the tetracycline-inducible VSG121 gene into DOT1B-deleted trypanosomes ( Δdot1b . 221ES . 121tet ) ( Janzen et al . , 2006 ) . Following induction of VSG121 expression , the decrease of the ES-resident VSG221 protein occurred with the same kinetics as in the parental cells ( Figure 4A ) . However , unlike in the parental cell line the protein levels did not increase again . Furthermore , the cells did not display a growth phenotype ( Figure 4B ) . Therefore , we measured ES activity just upstream of the VSG221 gene by inserting a luciferase reporter transgene into both Δdot1b and parental cell lines ( Figure 4C ) . As expected , the luciferase activity decreased rapidly in the parental cells ( Figure 4C , right panel ) . However , in the DOT1B-knockout trypanosomes luciferase activity was only marginally and transiently reduced at day 2 ( Figure 4C , left panel ) . Thus , in DOT1B-deleted cells , the ES was not attenuated , except for the most telomere-proximal part , where the VSG221 gene resides . This suggests that silencing of a VSG gene can be uncoupled from ES silencing . While ES attenuation seems to require DOT1B-dependent histone tri-methylation , VSG silencing does not . This is remarkable as it suggests a mechanism that specifically represses the telomeric VSG gene but not any other part of the ES ( Figure 4A ) . Furthermore , these results formally prove that the ES-resident VSG can be complemented in trans by a heterologous VSG gene . We conclude that the activation of a new VSG ES initiates an in situ switch , and that high-level expression of a second VSG is sufficient to attenuate and eventually silence the formerly active ES . This feedback loop assures monoallelic expression of the VSG . 10 . 7554/eLife . 02324 . 008Figure 4 . Expression site attenuation requires the histone methyltransferase DOT1B . ( A ) Quantification of VSG protein levels during the course of tetracycline-induced VSG121 expression in DOT1B-depleted cells . Quantification was done by normalization to the paraflagellar rod protein ( PFR ) , using the Licor Odyssey system . The values are percentages ± SD for two independent clones normalized to the parental Δdot1b or VSG121 wild-type cells . ( B ) Growth curve after induction of ectopic VSG121 expression in a Δdot1b cell line . Two independent clones were analyzed for 25 days in the absence ( −tet ) or presence ( +tet ) of 1 µg/ml tetracycline . ( C ) A luciferase reporter gene ( LUC ) was inserted into the active VSG221 ES ( upper panel ) of Δdot1b . 121tet and 221ES . 121tet cells . VSG121 expression was induced in the resulting LucESΔdot1b . 121tet and LucES . 221ES . 121tet cells and luciferase activity of three ( LucESΔdot1b . 121tet ) or two ( LucES . 221ES . 121tet ) independent clones was measured at the time points indicated and expressed as relative light units ( RLU ) ± SD . DOI: http://dx . doi . org/10 . 7554/eLife . 02324 . 008 There is a period in the life cycle of trypanosomes , when allelic exclusion of VSG is lifted , namely during differentiation to the quiescent stumpy stage . This developmental transition occurs in the mammalian bloodstream and involves ES-attenuation and growth arrest . The VSG-induced ES attenuation described here does not cause a bona fide cell cycle arrest and the parasites do not display the typical morphology of stumpy trypanosomes . Nevertheless , we hypothesized that the ES-attenuated parasites could have adopted stumpy-like characteristics and actually may represent the enigmatic intermediate stage , which is not yet committed to stumpy differentiation . Therefore , we tested for expression of the cell surface transporter protein associated with differentiation 1 ( PAD1 ) , which is a molecular marker for stumpy cells and is not expressed in the dividing slender stage ( Dean et al . , 2009 ) . In addition to a PAD1-specific antibody , we used a transgene , GFP:PADutr , encoding a GFP with a nuclear localization signal and a PAD1 3′UTR ( J Sunter , A Schwede and M Carrington , unpublished ) , which confers the stage specificity of PAD1 ( MacGregor and Matthews , 2012 ) . This reporter allowed for live cell analysis of PAD1-expression by flow cytometry ( Figure 5—figure supplement 1 ) . Prior to induction of the ectopic VSG , the cells showed no PAD1-expression and no GFP signal . After 48 hr of VSG121 overexpression , however , the PAD1 protein was clearly detectable on the surface of trypanosomes in the G1-phase of the cell cycle ( Figure 5A , B ) . Likewise , the GFP:PADutr reporter was visible in most G1 cells , but was also expressed in a minor portion of trypanosomes in other cell cycle phases ( Figure 5C ) . This is in agreement with the observation that the PAD1 mRNA is detectable prior to the protein ( MacGregor et al . , 2012 ) . We conclude that in the absence of the quorum-sensing factor SIF and at 37°C , the VSG overexpression-induced ES attenuation triggers G1-phase prolongation and concomitant expression of a bona fide stumpy marker in monomorphic slender stage trypanosomes . Notably , PAD1 expression was induced despite the global transcriptional attenuation that started at the same time . This again is reminiscent of the situation in stumpy parasites , where PAD1 expression is induced while general transcription decreases ( Amiguet-Vercher et al . , 2004 ) . 10 . 7554/eLife . 02324 . 009Figure 5 . Expression site attenuation leads to PAD1 surface expression . ( A ) GFP:PADutr and endogenous PAD1 are both expressed following ES attenuation . Maximum intensity projections of three-channel , 3D images ( 100 images , z-step: 200 nm ) of GFP:PADutr . 221ES . 121tet cells induced for VSG121 . After 48 hr , the cells were chemically fixed and incubated with an anti-PAD1 antibody ( red ) and counterstained with DAPI ( grey ) . The GFP:PADutr reporter is expressed in the nucleus ( green ) . The upper panel shows an example of G1 cells co-expressing the native PAD1 ( red ) and the reporter GFP:PADutr ( green ) . S- and G2-phase cells did not show PAD1 expression ( lower panel ) . Scale bar: 10 µm . ( B and C ) Cell cycle distribution of trypanosomes expressing the native PAD1 protein ( B ) and the GFP:PADutr reporter ( C ) . Both the GFP-reporter and the PAD1 protein are expressed specifically in G1 cells . PAD1 expression is limited to G1 . GFP:PAD1utr is visible in more G1 cells and also in other cell cycle stages . Values show percentages of positive cells of the indicated cell cycle stages and p-values were calculated using the Fisher's exact test . DOI: http://dx . doi . org/10 . 7554/eLife . 02324 . 00910 . 7554/eLife . 02324 . 010Figure 5—figure supplement 1 . Flow cytometry analysis of GFP:PADutr expression in live cells . ( A ) GFP expression of the parental 221ES cell line with the GFP:PADutr reporter inserted into the tubulin locus . Cells were analyzed for GFP expression before ( day 0 ) and 1 and 2 days after the addition of 100 µM pCPT-cAMP , which is a membrane permeable analogue of cAMP and induces stumpy gene expression . At day 2 , 42 . 3 ( ±3 . 7 ) % of the cells were GFP positive ( mean ± SD for two clones ) . ( B ) Flow cytometry analysis of 221ES . 121tet cells with the GFP:PADutr reporter . Cells were induced for VSG121 expression and analyzed for GFP expression at the indicated times . At day 2 , 31 . 1 ( ±0 . 1 ) % of the cells showed GFP expression ( mean ± SD for two clones ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02324 . 010 Next , we asked if PAD1 was fully functional as a transporter on the cell surface of slender trypanosomes . If so , the parasites should have become sensitive to the differentiation trigger cis-aconitate , the chemical cue that triggers progression from the stumpy to the procyclic insect stage . This developmental step is accompanied by loss of the VSG coat and expression of the surface protein EP . We induced VSG121 expression for 48 hr and then incubated the trypanosomes in the absence of tetracycline with 6 mM cis-aconitate . Immunofluorescence analysis revealed EP1 surface expression as early as 6 hr after addition of cis-aconitate ( Figure 6—figure supplement 1 ) . Thus , the mechanism of cell surface access block that prevents EP1 from being routed to the plasma membrane of proliferating bloodstream parasites had been deactivated , just as in stumpy cells ( Engstler and Boshart , 2004 ) . To quantify the EP1 surface expression , samples were collected 0 , 6 , and 24 hr after addition of cis-aconitate and analyzed by flow cytometry . In the control cells , no fluorescent EP1 signal was detectable ( Figure 6A , B ) . In contrast , VSG121-expressors revealed strong EP1 cell surface expression after only 6 hr of cis-aconitate treatment . After 24 hr , 70 ( ±6 ) % of cells were EP1-positive ( Figure 6A , B ) . Furthermore , the trypanosomes also altered their cell architecture , adopting a stumpy-like morphology after 6 hr of cis-aconitate treatment . After 24 hr , the posterior part of the cells had elongated and the kinetoplast had repositioned and segregated ( Figure 6C , Figure 6—figure supplement 1 ) . These morphological changes are hallmarks of the differentiation to the procyclic insect stage . 10 . 7554/eLife . 02324 . 011Figure 6 . G1-retardation renders trypanosomes sensitive to the differentiation trigger cis-aconitate and provides full developmental competence at 37°C . The response of trypanosomes to cis-aconitate was determined 48 hr after induction of ectopic VSG121 expression . ( A ) Flow cytometry of EP1 surface expression at 37°C . The induced cell line 221ES . 121tet was treated with cis-aconitate at 37°C in bloodstream form medium for 0 , 6 , and 24 hr ( +tet ) . Non-induced cells ( −tet ) were used as a control . ( B ) Percentage of cells expressing EP1 at the surface in ( A ) . Values represent mean ± SD of two independent clones . ( C ) Cell cycle analysis after treatment with cis-aconitate for various lengths of time . At least 100 cells were analyzed for each time point . For details see legend to Figure 2 . ( D ) Flow cytometry analysis of EP1 surface expression at 27°C . The experiments were conducted as described in ( A ) except that cells were grown in differentiation medium DTM at 27°C . ( E ) Percentage of cells expressing EP1 at the surface in ( D ) . Values represent mean ± SD of two independent clones . As expected for differentiation at 27°C , after 24 hr of cis-aconitate treatment , both populations were expressing EP1 . This is due to cold-shock induction of EP1 in the presence of the dominant differentiation trigger . Values are mean ± SD of two independent clones . ( F ) Cis-aconitate treatment releases growth attenuation . After induction for ectopic VSG121 expression for 48 hr , 221ES . 121tet cells were transferred to differentiation medium at 27°C in the presence or absence of cis-aconitate and kept at a density of 2 × 106 cells/ml . Cumulative cell numbers per ml are shown as mean ± SD of two independent clones . DOI: http://dx . doi . org/10 . 7554/eLife . 02324 . 01110 . 7554/eLife . 02324 . 012Figure 6—figure supplement 1 . Immunofluorescence detection of the insect stage surface protein EP1 . The trypanosomes were incubated at 37°C with 6 mM cis-aconitate in bloodstream form culture medium HMI-9 for 0 , 6 , and 24 hr . Red indicates EP1 fluorescence and blue is DAPI staining of the nucleus ( N ) and kinetoplast ( K ) . Note the characteristic posterior elongation of the cell marked by an arrowhead . Scale bar , 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 02324 . 012 It is important to note that the above experiments were conducted at 37°C to exclude the possibility that EP1 expression was induced by cold-shock ( Engstler and Boshart , 2004 ) . As growth of procyclic trypanosomes in fact requires temperatures of 27°C or below and a specific culture medium , we repeated the experiment at 27°C in the differentiation medium DTM ( Overath et al . , 1986 ) . Following tetracycline removal , ES-attenuated trypanosomes were incubated with cis-aconitate for 6 and 24 hr , and then analyzed for EP1 expression using flow cytometry . After 6 hr , 58 ( ±8 ) % of cells revealed strong EP1 fluorescence on the cell surface , while only 5 ( ±2 ) % of the control population were EP1-positive ( Figure 6D , E ) . Interestingly , the ES-attenuated trypanosomes accelerated the cell cycle after 24 hr and started growing as procyclic trypanosomes ( Figure 6F ) . This delayed growth response is a characteristic of the differentiation of stumpy trypanosomes to the procyclic insect stage ( van Deursen et al . , 2001 ) and probably marks the terminal shutdown of the attenuated ES . We conclude that the VSG-induced ES attenuation triggers the cell biological reprogramming of the parasite and causes the concomitant gain of full developmental competence . We initiated the shutdown of the active ES by providing wild-type levels of a second VSG . While the overall amount of VSG did not become limiting at any time during our experiments , transcription of ESAGs was obviously affected early on ( Figure 2B ) . Thus , we surmised that the loss of ESAG homeostasis caused growth retardation and developmental reprogramming of the parasite . As shown above , the reduction of ES activity is gradual , affecting the telomere-proximal ESAG1 and 2 first . Therefore , we initially examined if loss of these two ESAGs resulted in G1-prolongation and/or PAD1-induction . The GFP:PADutr cell line was first transfected with an tetracycline-inducible RNAi-construct targeting ESAG1 . Growth was significantly but transiently impaired at days 2 and 3 of RNAi ( Figure 7A ) . The GFP:PADutr expression was monitored by flow cytometry ( Figure 7B ) . Whereas no GFP was detected in the first 24 hr , there was an accumulation of GFP-positive cells after 48 hr and a peak of GFP expression at day 3 . Then the GFP-signal decreased again and was close to pre-induction at day 5 , when normal growth had resumed . 10 . 7554/eLife . 02324 . 013Figure 7 . Depletion of ESAG 1 or 2 by RNAi causes distinct growth defects and induction of PAD1 expression . ( A and C ) Growth of GFP:PADutr cells induced for RNAi against either ( A ) ESAG1 or ( C ) ESAG2 is presented as the cumulative mean cell number ± SD for two ( ESAG1 ) and three ( ESAG2 ) independent clones . Upper panels in ( A ) and ( C ) show Northern blots specific for ESAG1 and 2 , respectively , normalized to 18S rRNA . ( B and D ) Flow cytometric analysis of GFP:PADutr expression in a time course of ( B ) ESAG1-RNAi and ( D ) ESAG2-RNAi . DOI: http://dx . doi . org/10 . 7554/eLife . 02324 . 01310 . 7554/eLife . 02324 . 014Figure 7—figure supplement 1 . Down-regulation of a subset of three ESAGs induces GFP:PADutr . Growth curves ( upper panels ) and flow cytometry analysis ( lower panels ) of GFP:PADutr cells induced for RNAi against several ESAGs . Depletion of ESAGs 3 ( A ) and 6/7 ( B ) causes severe growth phenotypes without induction of GFP:PADutr expression . In contrast , RNAi against ESAGs 8 ( C ) and 12 ( D ) leads to transient growth retardation after day 2 , but only ESAG8 depletion leads to the appearance of a fraction of GFP-positive cells at day 3 . All growth curves show the cumulative cell numbers/ml of uninduced ( −tet ) and induced ( +tet ) cells as mean ± SD for 2 ( ESAGs 3 , 8 and 12 ) and 3 ( ESAG 6/7 ) independent clones . DOI: http://dx . doi . org/10 . 7554/eLife . 02324 . 014 Next , we knocked down ESAG2 in GFP:PADutr cells . No growth defect was observed within the first 2 days of RNAi , but the fraction of GFP-positive cells increased twofold in the first 24 hr . After 3 days , GFP was detected in 75% of all cells . At this time , cell growth stopped and the parasites started to die ( Figure 7C , D ) . It should be mentioned that ESAG1-RNAi reduced transcript levels to only 30% , while ESAG2-RNAi was more effective and reduced the mRNA to less than 5% , which may explain the lethal phenotype . To examine the possibility that down-regulation of any ESAG induces PAD1 expression , we targeted other ESAGs of the VSG221 ES . In addition to ESAG1 and 2 , only ablation of ESAG8 caused PAD1 induction , albeit in a less pronounced manner . The distinctive growth phenotypes observed after down-regulation of ESAGs 3 , 6 , 7 , and 12 were not accompanied by PAD1 expression at any time ( Figure 7—figure supplement 1 ) . We also depleted ESAG5 by RNAi and no growth phenotype was observed . These results strongly suggest that G1-retardation and PAD1 expression is caused by the depletion of three ESAGs . Thus , loss of ESAG homeostasis is an immediate early signal for developmental stage transition . Although ES attenuation provides the cells with full differentiation competence , it does not lead to an irreversible developmental commitment , as is the case in stumpy cells . Starting at day 5 of VSG121 induction , the attenuation of the VSG221 ES was gradually lifted . Interestingly , the kinetics of ES re-activation was also directional , starting at the promoter and proceeding towards the telomere . Thus , the linearity of ES attenuation was reversed ( Figure 2C , left panel ) . VSG221 mRNA slowly increased to pre-induction levels between days 5 and 8 ( Figure 1B ) . At day 8 , the VSG221 surface coat had returned and VSG121 was back to pre-induction levels ( Figure 1B ) . With onset of VSG ES reactivation , the G1-retardation was lifted , cell cycle progression was accelerated , and at day 10 , the population doubling times were back to normal ( 6 hr ) ( Figure 3A ) . This exact timing was reproducible in many independent experiments with subcloned populations . We have analyzed 20 clonal populations and all clones re-activated the VSG221 ES; no indication of ES switching was found . We cultivated the trypanosomes for an additional 10 days in the absence of tetracycline and then re-induced expression of VSG121 . More than 20% of the population was responsive to re-induction and showed strong VSG121 expression , pointing against destructive mutations in the inducible expression system . As an additional control , we have expressed the VSG121 gene using a constitutive T7 promoter . All resulting clonal populations revealed reduced levels of ectopic VSG121 expression , comparable to inducible VSG121-expressors after ES-re-activation ( after 5 days ) . Importantly , in those cell lines the tetracycline system was still fully functional , as shown by tetracycline induction of an unrelated transgene ( data not shown ) . These observations strongly argue against selection of revertants that prevent VSG121 expression by interfering with the T7-system . Thus , ES attenuation opens a time-window in which the trypanosomes linger in an intermediate state and can progress in two directions , either by differentiating to the insect stage or by returning to proliferation in the mammalian host . This questions the common view that trypanosome development is inevitably unidirectional as in the case of the SIF-induced pathway ( MacGregor et al . , 2011 ) .
The unique feature of the mammalian infective form of African trypanosomes is the plasma membrane with a dense VSG coat . The parasite has evolved an efficient system of antigenic variation that relies on monoallelic expression of VSG genes ( Borst , 2002 ) . Though the system is optimized for a monotypic coat , two VSGs have to be handled on the same cell surface during antigen switching . Although nothing is known about the length of this period , the different underlying molecular mechanisms suggest variable time scales . Telomere exchange or gene conversion are probably rather fast processes as loss of VSG mRNA by RNAi results in rapid cell cycle arrest ( Sheader et al . , 2005 ) . A similar scenario applies if the active ES would be silenced before a new one is activated . In contrast , time would not necessarily be limiting if an in situ switch would involve activation of a new ES before silencing of the old one . In this study , we simulated the activation of a new ES by high-level , inducible expression of a second heterologous VSG . The induction of VSG121 expression resulted in the rapid decrease of VSG221 mRNA and protein . The ectopic VSG121 dominated the surface coat for several days . The ESAGs were also down regulated , showing that induction of a second VSG gene attenuates the entire active ES . The insertion of reporter genes showed that this ES attenuation starts at the telomere and progresses towards the promoter region . Our data further indicate that the initial spike in the production of ectopic VSG acts as an immediate early signal , which is relayed to an epigenetic ES silencing machinery that requires the histone methyltransferase DOT1B . This model further explains the expression levels in stable VSG double-expressors that have a second VSG inserted into the active ES , just upstream of the endogenous copy ( Muñoz-Jordán et al . , 1996 ) . In these double-expressors , both VSGs are expressed in a 50/50 ratio , not exceeding a total amount of 100% compared to wild-type single expressors ( data not shown ) . The generation of such double-expressors requires antibiotic selection for 7 days after transfection . In that time , the trypanosomes react to the initial overexpression of VSG , which occurs immediately after recombination of the transgene into the active ES , and attenuate the ES . When the VSG expression is adapted , attenuation is hold , not affecting the upstream located ESAGs . Thus , high-level expression of a second VSG alone is sufficient to initiate attenuation of the active ES . Importantly and unlike the endogenous VSG , this second VSG does not have to be expressed from an ES and it does not need to be transcribed by RNA polymerase I . Furthermore , VSG silencing and shutdown of the active ES are mechanistically distinct events , as in DOT1B-knockout trypanosomes , the ES-resident VSG is silenced upon VSG overexpression , but the ESAGs are not . No phenotype is observed after VSG overexpression in the DOT1B-knockout cells , proving that neither the initial surplus of VSG production nor the non-trypanosomal T7 RNA polymerase system is toxic . In the mammalian olfactory receptor ( OR ) system , the expression of an ectopic OR transgene is sufficient to mediate a feedback inhibition that prevents the expression of the active endogenous OR gene , independent of the receptor signal transduction pathway ( Nguyen et al . , 2007 ) . This is strongly reminiscent of the VSG-mediated attenuation of the trypanosome ES reported here . Despite the enormous evolutionary distance between kinetoplastids and mammals the mechanistic principle appears to be conserved . The trypanosome model obviously offers some opportunities that are not readily available in the OR system , such as following the kinetics of initiation and reversal of gene silencing . Furthermore , in trypanosomes an array of genes is attenuated , whereas in the olfactory system only the OR is silenced . Importantly , the OR system fulfills a single function , namely detection of an almost infinite number of odorant molecules . Our work suggests that the control of monoallelic expression of VSG links at least three key-processes , namely antigenic variation , host range determination , and parasite development . Our findings clearly support a model in which the VSG ES orchestrates events at the crossroads of antigenic variation , cell cycle , and development . The trypanosome cell can operate for significant periods in a dormant state . This ability may become crucial in the course of an in situ switch to a non-functional or impaired VSG or upon activation of a defective ES . Any trypanosome with an attenuated ES will experience a time window of full developmental competence . Thus , the stumpy life cycle stage with its genuine cell cycle arrest can be bypassed . While our experimental system clearly initiates VSG switching , it is not suited for completion of the process , most likely because it does not provide functional ESAGs . Interestingly , the cells react by not fully silencing the ES , but rather attenuating its activity to about 20% , probably by chromatin remodeling after histone methylation . In response to ES attenuation trypanosomes enter a semi-quiescent state with a specifically extended G1-phase . The parasites remain in this slow growth state for about 5 days and then resume normal growth without any sign of reduced fitness . While ES attenuation is triggered by the brief period of true VSG overexpression , the entry into the dormant phase is the consequence of ESAG shortage . It is important to note that ES attenuation does not lead to complete ESAG exhaustion , but rather to gradual reduction in expression . Thus , individual ESAGs could reach growth-limiting levels at different concentrations and hence , could induce G1-retardation at different times . In fact , depletion of ESAG1 mRNA by RNAi leads to transient growth retardation after 48 hr . The cells recover from this phenotype 2 days later , which is in agreement with the existence of ESAG1 knockout lines ( Carruthers et al . , 1996 ) . Importantly , even depletion of the non-essential and ES-specific ESAG8 ( Hoek and Cross , 2001 ) becomes growth limiting for a transient period . In contrast , ESAG2-RNAi is irreversibly lethal and therefore , a complete ES shutdown without functional rescue from non-ES copies or a new ES would be detrimental . Although essential , the down-regulation of ESAG2 to the level prevailing in ES-attenuated cells ( 20% ) is tolerated . However , the function of these ESAGs remains elusive . ESAG1 and 2 are N-glycosylated , membrane-associated proteins localized in the flagellar pocket . It is suggested that both could function in the endosome pathway ( Nolan et al . , 1999; Pays et al . , 2001 ) . Thus , ESAG1 and 2 could act in concert , which would explain the in parts redundant RNAi phenotypes . In contrast , ESAG8 is found in the nucleolus and the cytoplasm , and interacts with a Puf family member ( Hoek et al . , 2000; Hoek et al . , 2002 ) . This suggests that ESAG8 plays a role in mRNA regulation and might be linked to the differentiation pathway . Thus , the concept of transient ES attenuation rather than immediate and complete ES silencing may be of some physiological relevance; it could allow probing for ES performance , for example ESAG functionality , VSG switching orders , quality of novel mosaic VSGs , or newly assembled expression sites . If both VSGs are compatible , and produce a mixed and dense surface coat and the ESAGs of the new ES are fully functional , the old expression site is rapidly and completely silenced with no effects on cell cycle progression . However , if any component of the new ES fails or is not sufficiently functional , the cells can adjust growth , in the extreme , to very basal levels . This way the parasites may either gain time to adapt to the new conditions or reverse the attenuation and re-activate the old ES . We suggest that the latter case resembles an unsuccessful switching event , which would escape experimental detection . In fact , we can only observe successful VSG switching events , thereby neglecting the surprising capacity of trypanosomes to reverse switching directions . This flexibility could become particularly productive in varying host environments . Provided a sufficiently high ES activation frequency exists , the parasites could effectively probe for host serum-compatible ESAGs ( Bitter et al . , 1998; Chaves et al . , 1999; Pays et al . , 2001 ) . Hence , ES switching plasticity could determine the host range . One extreme example for this is Trypanosoma b . rhodesiense , which is human-infective and thought to be a host-range variant of T . b . brucei . The parasites are specifically adapted to human serum by expression of the serum-resistance antigen ( SRA ) from one particular VSG ES in most isolates . The structure of the SRA-containing ES is unusual as it is missing several ESAGs present in all other ESs . During an infection of a human , an in situ switch to a VSG in another ES would cause loss of SRA expression , with lethal consequences . Thus , T . b . rhodesiense must not switch off the SRA-containing ES , thus making the ESAG-dependent differentiation pathway obsolete . This may be the reason why the T . b . rhodesiense ES has lost ESAG1 , 2 and 8 . The fact that ES attenuation causes growth retardation could also become mechanistically relevant during another phase of the trypanosome life cycle , namely upon differentiation to the G1/G0-arrested stumpy stage . The developmental transition from the proliferating slender to the quiescent stumpy bloodstream stage is initiated by quorum sensing of the trypanosome-derived compound SIF ( Reuner et al . , 1997 ) . The monomorphic cells used in this study are unresponsive to SIF and hence , should have lost the capacity to differentiate to the tsetse-infective stage ( Vassella et al . , 1997 ) . In our experiments , however , VSG-induced ES attenuation renders monomorphic trypanosomes fully competent for development . After 2 days of VSG overexpression the stumpy stage marker PAD1 was strongly up-regulated , specifically in G1-phase cells , and despite global transcriptional silencing being initiated at the same time . This indicates a highly specific event , which , in the absence of SIF , is orchestrated solely by the state of the VSG expression site . We are currently investigating this alternative pathway in a pleomorphic strain . The induced expression of a second VSG indeed leads to the formation of stumpy forms ( data not shown ) . While VSG itself is not involved , as it is not limiting in our system , three ESAGs signal ES attenuation to the differentiation pathway . Depletion of ESAGs 1 , 2 and 8 mRNA by RNAi resulted in induction of the PAD1 reporter GFP:PADutr . Down-regulation of any of the other ESAGs did not cause GFP:PADutr expression . This underlines the specificity of the pathway and suggests that ES activity controls the induction of transient developmental competence . Remarkably , by embedding the endogenous PAD1 protein into the VSG coat , the ES attenuated trypanosomes became fully responsive to the developmental trigger cis-aconitate . They initiated transformation to the procyclic insect stage with the same kinetics as stumpy parasites . There is , however , one principal feature that distinguishes SIF-mediated stumpy formation from ES-regulated cell dormancy . While stumpy parasites are terminally differentiated and can only survive if taken up by an insect , the onset of differentiation is fully reversible in the ES-attenuated trypanosomes . We postulate that these parasites linger in an intermediate state with open developmental choices ( Figure 8 ) . They have the capacity to complete ES switching by adapting to the new ES or , alternatively , they can re-activate the old , attenuated ES . In both of these cases the parasites resume growth as mammalian bloodstream form cells . If taken up by the tsetse fly , however , they would immediately transform to the procyclic insect stage . 10 . 7554/eLife . 02324 . 015Figure 8 . Model of the dual roles of expression site attenuation for antigenic variation and development of T . brucei . Activation of a new ES leads to attenuation of the previously active one in a DOT1B-dependent manner . The attenuation and subsequent growth retardation allows ES quality control via ESAG1 , 2 and 8 . If the newly activated ES is functional , the old one is rapidly silenced , resulting in a successful VSG switch . If , however , a non-functional or non-compatible ES is activated , the cells react with a specific prolongation of the G1-cell cycle phase . This dormancy can last for up to 5 days and is fully reversible . Within this period the trypanosomes become competent for developmental stage transition , which involves expression of the stumpy marker PAD1 . The stumpy stage , however , is irreversibly arrested in G0 and therefore committed to a life in the tsetse fly . In contrast , ES attenuation causes a reversible G1 dormancy , which allows either re-activation of the attenuated ES ( unsuccessful switch ) or survival by development in the insect vector . Thus , ES attenuation operates at the crossroads of antigenic variation and development . DOI: http://dx . doi . org/10 . 7554/eLife . 02324 . 015 This high flexibility might become biologically relevant especially during early infections , when certain ESAGs or VSGs are not compatible with the new host environment: cells either adapt or regain fly-infectivity . Thus , the state of the expression site must not be regarded as a simple on-off switch , but rather represents a fine-tuned module at the unexpected crossroads of antigenic variation and development .
All cell lines generated in this study are based on T . brucei 427 MITat 1 . 2 13-90 bloodstream forms ( 221ES ) ( Wirtz et al . , 1999 ) and were cultivated in HMI-9 medium with 10% fetal bovine serum ( Sigma-Aldrich , St . Louis , MO ) at 37°C and 5% CO2 . For T7-polymerase and tetracycline repressor maintenance , cells were selected with 5 µg/ml hygromycin and 2 . 5 µg/ml G418 . For transfections , 3 × 107 parasites were electroporated with 10 µg of linearized plasmid DNA using the AMAXA Nucleofector II ( Lonza , Switzerland ) . Details of expression constructs and cell lines are provided in Supplementary file 1 . The VSG121 open reading frame with the full 3′UTR was amplified from MITat1 . 6 wild-type genomic DNA with the primers VSG121_fw and VSG121_rev , subcloned into a pBluescriptSK , mobilized with HindIII and cloned into the pLew82v4 ( 24 , 009; Addgene plasmid ) . The resulting plasmid pRS . 121 was NotI-linearized and transfected into the bloodstream form 221ES , giving rise to the 221ES . 121tet transformants . pkD . GFP was generated by cloning a 1 . 6 kb SpeI/EcoRI fragment containing a GFP ORF flanked by VSG221 5′ and 3′UTRs from p3845 ( kindly provided by A Schwede and M Carrington , Cambridge , UK ) into the VSG221 ES targeting vector pkD ( Muñoz-Jordán et al . , 1996 ) . Cell lines GFPESpro-221ES . 121tet and GFPEStel-221ES . 121tet were generated by inserting p3845 and pkD . GFP , respectively , into 221ES . 121tet . For the LucES cell lines , the luciferase gene was mobilized with HindIII from pLew82v4 , overhangs were filled in and the resulting product was ligated to an EcoRI digested and blunted pkD vector . The resulting plasmid pkD . Luc was linearized with AvrII and KpnI and transfected into 221ES . 121tet or Δdot1b . 121tet . pRib . Luc was generated by cloning the luciferase gene between the HindIII and BamHI sites of pTSA-Rib ( Xong et al . , 1998 ) , in which the tubulin 3′UTR had been replaced by the PARP 3′ region and the hygromycin resistance by puromycin resistance . The resulting plasmid was SphI linearized and transfected into 221ES . 121tet , creating LucrPro . 221ES . 121tet . pTub . Luc was created by replacing a BsaBI/SphI fragment from pHD328 ( Wirtz et al . , 1994 ) with a BsaBI/SphI fragment from pLew20 ( Wirtz et al . , 1999 ) and the ble gene was replaced by a puromycin resistance gene . The resulting plasmid was linearized with NotI and transfected into 221ES . 121tet , resulting in Luctub . 221ES . 121tet . For generating the GFP:PADutr reporter cell lines , the plasmid p4231 ( J Sunter , A Schwede and M Carrington , unpublished ) was transfected into the parental 221ES or the 221ES . 121tet cell line . For RNAi , the target sequences were amplified from genomic DNA of 221ES cells using the following primers: ESAG1_fw ( 5′-TTGTGTTGATGCATG-3′ ) and ESAG1_rev ( 5′-TCGGTCTTG GTTTAG-3′ ) ; ESAG2_fw ( 5′-GAAATAGTGATTGCCG-3′ ) and ESAG2_rev ( 5′-CAAACTCAGCTAATGC-3′ ) ; ESAG3_fw ( 5′AAAAAAAAGCTTTCCTTC AAGATGAAG AAGC-3′ ) and ESAG3_rev ( 5-AAAAAAGGATCCAAACAAGTCATTCTCCTTGACC-3′ ) ; ESAG6/7_fw ( 5′-AAAAAAAAGCTTGTTTTGGTTTGTGCTGTTGG-3′ ) and ESAG6/7_rev ( 5′-AAAAAAGGATCCATACTTTCCGCACCCAAGC-3′ ) ; ESAG8_fw ( 5′-GCACTACGTGATCTGGAAGC-3′ ) and ESAG8_rev ( 5′-CATAGAGCACCCTC AAGTGG-3′ ) ; ESAG12_fw ( 5′-AGCGGTGTCAATATTC-3′ ) and ESAG12_rev ( 5′-AGGAGGAAGGAGTTTG-3′ ) . The PCR products were subcloned into a pBluescriptSK+ and mobilized with HindIII and SpeI ( ESAG1 and 2 ) , HindIII and BamHI ( ESAG3 , ESAG6/7 ) , or XbaI and XhoI ( ESAG8 ) to ligate them into the RNAi plasmid p2T7 ( Shi et al . , 2000 ) . The resulting plasmids were linearized with NotI and transfected into the GFP:PADutr . 221ES cell line . Newly synthesized DNA was labeled with 5-ethynyl-2′-deoxyuridine ( EdU ) using the Click-iT EdU Alexa Fluor 488 Imaging Kit ( Invitrogen , Carlsbad , CA ) , essentially following the manufacturer's instructions . The cells were incubated with 50 µM EdU in HMI-9 medium , washed in buffer and chemically fixed . After permeabilization with 0 . 1% Triton X-100 , the reaction cocktail was added and the cells were washed twice . BrUTP labeling of nascent transcripts was achieved as described in Smith et al . ( 2009 ) . Total RNA was isolated from 1 × 108 trypanosomes using the Qiagen RNeasy Mini Kit ( Qiagen , Netherlands ) essentially following the manufacturer's instructions . For RNA quantification with fluorescently labeled probes ( GFP-probe: GCCGTTCTTCTGCTTGTCGGCCATGATATAGA; VSG121-probe: GGCTGCGGTTACGTAGGTGTCGATGTCGAGATTAAG; VSG221-probe: CAGCGTAAACAACGCACCCTTCGGTTGGTCGTCTAG; Tubulin-probe: ATCAAAGTACACATTGATGCGCTCCAGCTGCAGGTC; 18SrRNA-probe: CAACCAAACAAATCACTCCACCGACCAAAA ) , 3 µg of total RNA was denatured with glyoxal and blotted with a Minifold Dotblotter ( Schleicher & Schuell , Germany ) onto nitrocellulose and hybridized overnight at 42°C . ESAG transcripts were detected with 32P-labeled probes and quantified using a Phosphorimager . To precipitate protein , the cell lysates were mixed with ice-cold acetone and centrifuged at 20 , 000×g for 10 min . The sediment was air-dried and resuspended in sample buffer ( 2% SDS , 10% glycerol , 60 mM Tris–HCl , pH 6 . 8 , 1% β-mercaptoethanol ) to yield equivalents of 2 × 105 cells/µl . For protein quantification , 3 µl was spotted onto nitrocellulose , air-dried , and incubated with specific polyclonal anti-VSG antibodies ( rabbit anti-VSG221/C-term , 1:5000; rabbit anti-VSG121 , 1:2000 ) and mouse monoclonal anti-PFR antibody ( L13D6 , 1:20 ) . For fluorescence detection , an IRDye680-conjugated goat-anti-mouse antibody and an IRDye800-conjugated goat-anti-rabbit antibody ( 1:10 , 000; Licor , Lincoln , NE ) were used . Protein and RNA were analyzed and quantified using the Licor Odyssey System . 1 × 106 to 1 × 107 trypanosomes were washed in ice-cold PBS and resuspended in 100 µl of cell culture lysis buffer ( Promega , Madison , WI ) . 5 µl of lysate were added to 45 µl of luciferase substrate and the luminescence was immediately measured with a Tecan Infinite M200 plate reader ( Tecan , Switzerland ) . IF for surface detection of procyclin EP1 was done as in Engstler and Boshart ( 2004 ) . For VSG IF , chemically fixed cells were incubated with a mouse anti-VSG121 ( 1:100 ) and a rabbit anti-VSG221 ( 1:100 ) antibody , followed by incubation with an Alexa488- and Alexa594-conjugated anti-mouse and anti-rabbit antibody , respectively ( 1:1000 ) . For PAD1 detection , chemically fixed cells were permeabilized with 0 . 05% Triton X-100 for 10 min and incubated with a rabbit anti-PAD1 antibody ( 1:100 ) , followed by incubation with an Alexa488-conjugated anti-rabbit antibody ( 1:1000 ) . Flow cytometry was performed with a BD Bioscience FACSCalibur Flow Cytometer . Data were analyzed with the BD CellQuest Pro Software . For each sample , 20 , 000 cells were counted . | African sleeping sickness is a potentially lethal disease that is caused by a parasite called T . brucei and spread by tsetse flies . Like many of the parasites that cause tropical diseases , T . brucei employs genetic trickery to evade the immune systems of humans and other mammals . This involves changing the variant surface glycoprotein ( VSG ) coat that surrounds the parasite on a regular basis in order to remain one step ahead of the immune system of its host: while the immune system looks for invaders wearing a particular coat , the parasites are spreading through the host in a completely different coat . To infect other hosts , the parasite must undergo changes that allow it to re-infect the tsetse fly . Therefore , besides the ‘antigenic variation’ that allows it to change its surface coat when it is in the blood of its host , T . brucei must undergo a more fundamental metamorphosis before it is capable of colonizing the tsetse fly . However , many details of the changes that allow the parasites to re-infect flies are not understood . T . brucei has several hundred VSG genes clustered in about 15 regions known as expression sites , but only a single expression site is active at any given time . Each expression site also contains a number of other genes known as expression site-associated genes ( ESAGs ) . Antigenic variation can occur as a result of different VSG genes within the same expression site being expressed as proteins , or when the active expression site is silenced and another expression site is activated . This is another process that is not fully understood . Batram et al . now reveal that the expression of VSG genes , antigenic variation and the changes that allow the parasites to re-infect flies are all related to each other . This suggests that the expression site could provide a new point of attack in the fight against African sleeping sickness . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression",
"cell",
"biology"
] | 2014 | Expression site attenuation mechanistically links antigenic variation and development in Trypanosoma brucei |
The prion protein ( PrPC ) is highly expressed in the nervous system and critically involved in prion diseases where it misfolds into pathogenic PrPSc . Moreover , it has been suggested as a receptor mediating neurotoxicity in common neurodegenerative proteinopathies such as Alzheimer's disease . PrPC is shed at the plasma membrane by the metalloprotease ADAM10 , yet the impact of this on prion disease remains enigmatic . Employing conditional knockout mice , we show that depletion of ADAM10 in forebrain neurons leads to posttranslational increase of PrPC levels . Upon prion infection of these mice , clinical , biochemical , and morphological data reveal that lack of ADAM10 significantly reduces incubation times and increases PrPSc formation . In contrast , spatiotemporal analysis indicates that absence of shedding impairs spread of prion pathology . Our data support a dual role for ADAM10-mediated shedding and highlight the role of proteolytic processing in prion disease .
The cellular prion protein ( PrPC ) is a glycosylphosphatidylinositol ( GPI ) -anchored lipid raft constituent highly expressed in neurons of the central nervous system ( CNS ) . Misfolding of PrPC into its pathogenic isoform , PrPSc , occurs by a self-perpetuating process of templated conformational conversion and causes transmissible and invariably fatal prion diseases ( Prusiner , 1982 ) . In these diseases , PrPC not only represents the substrate for conversion but also a prerequisite of prion-associated neurotoxicity ( Büeler et al . , 1993; Brandner et al . , 1996; Mallucci et al . , 2003 ) . In contrast to PrPC , PrPSc is prone to aggregation and is characterized by a high β-sheet content and partial resistance to digestion with proteinase K ( PK ) . Moreover , PrPSc constitutes an essential component of infectious prion particles or ‘prions’ ( Prusiner , 1982; Silveira et al . , 2005 ) . Recently , a central role in more common proteinopathies , such as Alzheimer's disease ( AD ) , has been attributed to PrPC where it was shown to act as a neuronal receptor for neurotoxic amyloid-β ( Aβ ) oligomers ( Lauren et al . , 2009; Dohler et al . , 2014 ) and a variety of other disease-associated β-sheet rich protein assemblies including PrPSc ( Resenberger et al . , 2011 ) . Despite some degree of controversy regarding this receptor function and its relevance in AD ( Balducci et al . , 2010; Benilova and De Strooper , 2010; Calella et al . , 2010; Kessels et al . , 2010 ) , binding of pathogenic oligomers to the flexible N-terminus of PrPC is thought to initiate a cascade of events that mediates their neurotoxicity and results in synaptic degeneration and neuronal loss ( Resenberger et al . , 2011; Larson et al . , 2012; Um et al . , 2012 ) . PrPC is subject to two conserved proteolytic cleavage events under physiological conditions , termed α-cleavage and shedding ( Borchelt et al . , 1993; Harris et al . , 1993; Chen et al . , 1995; Mange et al . , 2004 ) . These cleavages likely influence physiological functions of PrPC ( Aguzzi et al . , 2008; Linden et al . , 2008 ) and , importantly , its role in neurodegenerative diseases ( reviewed in Altmeppen et al . , 2013 ) . While α-cleavage , occurring in the middle of the PrPC sequence and producing a soluble N1 and a membrane-attached C1 fragment , confers neuroprotection with regard to prion diseases ( Lewis et al . , 2009; Westergard et al . , 2011; Turnbaugh et al . , 2012; Campbell et al . , 2013 ) and Aβ-associated neurotoxicity ( Guillot-Sestier et al . , 2009; Resenberger et al . , 2011; Beland et al . , 2012; Guillot-Sestier et al . , 2012; Fluharty et al . , 2013 ) , the role of an extreme C-terminal cleavage in close proximity to the plasma membrane , termed shedding , remains enigmatic . On the one hand , by reducing the substrate for prion conversion and the receptor for neurotoxicity , this proteolytic release of almost full length PrPC from the plasma membrane could be protective with regard to prion diseases ( Marella et al . , 2002; Heiseke et al . , 2008; Altmeppen et al . , 2012 ) . Moreover , recombinant or transgenically expressed anchorless PrPC mimicking shed PrPC has been shown to antagonize both PrPSc propagation and Aβ neurotoxicity , respectively , thus indicating a protective activity by blocking toxic conformers and preventing their access to the cell ( Meier et al . , 2003; Calella et al . , 2010; Nieznanski et al . , 2012; Fluharty et al . , 2013; Yuan et al . , 2013 ) . On the other hand , shedding could accelerate an ongoing prion disease by increasing production and subsequent spread of anchorless prions within the CNS . In fact , PrPSc can be shed from the cellular surface in vitro ( Taylor et al . , 2009 ) and anchorless PrPC can , in principle , be converted to PrPSc , resulting in an altered type of prion disease in transgenic mice ( Chesebro et al . , 2005 , 2010 ) . High expression of anchorless PrPC leads to formation of prions and a late onset neurological disease ( Stöhr et al . , 2011 ) . While the in vivo identity of the protease ( s ) responsible for the α-cleavage of PrPC remains a matter of controversy ( Vincent et al . , 2001; Tveit et al . , 2005; Taylor et al . , 2009; Oliveira-Martins et al . , 2010; Altmeppen et al . , 2011 , 2012; Beland et al . , 2012; Liang et al . , 2012; Wik et al . , 2012; Mays et al . , 2014; McDonald et al . , 2014 ) , we recently showed that a disintegrin and metalloproteinase ADAM10 is the relevant sheddase of PrPC in vivo ( Altmeppen et al . , 2011 ) , as previously suggested ( Parkin et al . , 2004; Taylor et al . , 2009 ) and subsequently confirmed ( Wik et al . , 2012; Ostapchenko et al . , 2013; McDonald et al . , 2014 ) by in vitro experiments of others . Using conditional NestinADAM10 knockout mice ( Jorissen et al . , 2010 ) , we recently showed that depletion of the protease in neural precursors abolishes shedding and leads to posttranslational accumulation of PrPC in the early secretory pathway indicative of a regulatory role of ADAM10 in PrPC membrane homeostasis ( Altmeppen et al . , 2011 ) . However , due to the perinatal lethality of these mice , the influence of ADAM10 on the course of prion disease remained unsolved . In this study we used novel viable ADAM10 conditional knockout ( ADAM10 cKO or A10 cKO ) mice with specific postnatal deletion of the protease in forebrain neurons ( Prox et al . , 2013 ) . We show that lack of ADAM10 leads to ( i ) elevated ( membrane ) levels of PrPC , ( ii ) drastically shortened incubation times of prion disease , ( iii ) increased prion conversion , and ( iv ) upregulation of calpain levels . In contrast , spread of prion-associated pathology within the brain was reduced . It is intriguing that the absence of a protease involved in the constitutive processing of PrPC significantly impacts the course of prion disease , thus enforcing the relevance of proteolytic processing events in neurodegeneration .
Maintenance of plasma membrane levels of PrPC by ADAM10-mediated shedding may represent a conserved mechanism in different cellular lineages . To further investigate this we generated neural stem cells ( NSCs ) from embryonic day ( E ) 14 wild-type and A10 cKO mice with inactivation of the Adam10 gene in neuroectodermal progenitor cells ( NestinA10 KO mice ) ( Jorissen et al . , 2010 ) . As shown in Figure 1A , surface biotinylation experiments on neuronally differentiated NSCs revealed that membrane levels of PrPC were increased 1 . 56-fold ( ±0 . 12; SEM ) in the absence of ADAM10 ( n = 9 independent samples ) compared with wild-type controls ( set to 1 ± 0 . 13; n = 9 ) . Moreover , genetic reintroduction of Adam10 into NSC cultures of NestinA10 KO mice was sufficient to reduce membrane levels of PrPC ( 0 . 95 ± 0 . 11; n = 8 ) and thus to restore physiological wild-type conditions . Nucleofection of NestinA10 KO cells with a vector lacking the Adam10 cDNA did not show any effect on PrPC membrane levels ( 1 . 55 ± 0 . 18; n = 5 ) . Indirect immunofluorescence analyses of non-permeabilized neuronally differentiated NSCs confirmed the biochemical results by showing increased intensity of PrPC surface immunostaining in NestinA10 KO cells and NestinA10 KO cells nucleofected with a control vector compared with wild-type control cells , and a similar intensity of PrPC surface immunostaining in A10-nucleofected NestinA10 KO and control cells ( Figure 1B ) . 10 . 7554/eLife . 04260 . 003Figure 1 . Characterization of PrPC levels in different cellular models of ADAM10 deficiency . ( A ) Representative Western blots showing membrane levels of PrPC as revealed by surface biotinylation ( I; on the left ) and total PrPC levels in lysates ( I; on the right ) as well as ADAM10 surface expression ( II ) of neuronally differentiated neural stem cells ( NSCs ) from NestinA10 KO and littermate control mice and after genetic reintroduction of Adam10 ( A10 KO + ADAM10 ) or nucleofection with control vector ( A10 KO + Vector ) into NSCs of NestinA10 KO mice . Flotillin served as loading control . ( III ) Quantification of densitometric analysis of PrPC membrane levels of experimental groups mentioned above ( n = 9 independent samples for controls [set to 1]; n = 9 for NestinA10 KO; n = 8 for NestinA10 KO + ADAM10; n = 5 for NestinA10 KO + Vector; significance: **p = 0 . 0054; ##p = 0 . 0014; *p = 0 . 0336 ; #p = 0 . 0212 ) . Error bars indicate SEM . ( B ) Representative immunofluorescent PrPC ( green ) surface staining of neuronally differentiated NSCs derived from NestinA10 KO ( without [second row] or with genetic reintroduction of ADAM10 [third row] or vector only [fourth row] ) and littermate control mice ( first row ) , respectively . Tubulin ( red ) was stained after permeabilization of cells to confirm neuronal differentiation of NSCs . DAPI ( blue ) marks nuclei . ( C ) Representative immunostaining of PrPC ( green ) and ADAM10 ( red ) in permeabilized ( upper two rows ) and non-permeabilized ( lower three rows ) murine embryonic fibroblasts ( MEFs ) derived from mice with a complete knockout of ADAM10 ( ADAM10 KO ) or wild-type mice ( control ) . Higher resolution of white boxes is shown in the bottom row and reveals colocalization of PrPC and ADAM10 at the plasma membrane of wild-type control MEFs . Scale bars in B and C represent 10 µm . ( D ) Western blot analysis of cell-associated PrPC levels in ADAM10 knockout ( A10 KO ) and wild-type ( wt ) MEF lysates ( left part: actin served as loading control ) . Levels of shed PrPC were assessed in cell culture media supernatants of ADAM10 knockout and wild-type MEFs by filter column concentration ( conc . media ) and immunoprecipitation ( IP ) with a PrPC-specific antibody respectively ( right part ) . ( E ) Levels of cell-associated ( neuronal lysates ) and shed PrPC ( IP of media supernatants ) in primary neuronal cultures of prion protein knockout ( Prnp0/0 ) , wild-type ( wt; C57BL/6 ) , prion protein overexpressing ( tga20 ) , and NestinA10 KO mice at embryonic day 14 . IgG-HC and IgG-LC mark signals for heavy and light chain of the capturing antibody POM2 . DOI: http://dx . doi . org/10 . 7554/eLife . 04260 . 003 In addition , we analyzed murine embryonic fibroblasts ( MEFs ) derived from mice with a complete knockout of ADAM10 ( Hartmann et al . , 2002 ) with regard to PrPC levels ( Figure 1C and D ) . As expected , we found increased total PrPC levels in ADAM10 knockout MEFs by ( i ) immunofluorescence analysis of permeabilized cells ( Figure 1C , upper part ) and ( ii ) Western blot assessment of MEF lysates ( Figure 1D , left part ) . In non-permeabilized wild-type MEFs , colocalization could be observed between the protease ADAM10 and its substrate PrPC at the plasma membrane ( Figure 1C , bottom row ) . Next , we directly investigated the shedding of PrPC in ADAM10 knockout and wild-type MEFs by biochemical analysis of culture supernatants . Results obtained with concentrated media and with immunoprecipitation of shed PrPC from media showed that shedding is impaired in ADAM10 knockout MEFs compared with wild-type MEFs ( Figure 1D , right part ) . Finally , we assessed PrPC levels and the shedding of PrPC in primary neurons of NestinA10 KO and wild-type control mice as well as in mice deficient for PrPC ( Prnp0/0 ) and in PrPC overexpressing Tg ( Prnp ) a20 mice ( known and hereafter referred to as tga20 ) ( Figure 1E ) . Confirming our previous study ( Altmeppen et al . , 2011 ) , shedding of PrPC was absent in NestinA10 KO neurons while tga20 neurons showed increased levels of shed PrPC compared with wild-type controls . Taken together , data obtained from different murine cellular models having a deletion of Adam10 confirmed the role of this protease as the functionally relevant sheddase of PrPC and , thus , as a regulator of PrPC membrane homeostasis . Using conditional NestinA10 knockout mice , we previously showed that lack of ADAM10-mediated shedding leads to increased neuronal levels of PrPC ( Altmeppen et al . , 2011 ) . However , due to the perinatal lethality of these mice we were unable to investigate the impact of ADAM10 deficiency on the course of prion disease . Therefore , new conditional Camk2aADAM10 knockout mice ( ADAM10 cKO or A10 cKO ) lacking Adam10 in neurons of the forebrain were produced and characterized ( Prox et al . , 2013 ) . These mice were viable and used for prion inoculations performed in this study . First , we analyzed PrPC levels in A10 cKO and littermate controls at postnatal day ( P ) 19 . Reduction of ADAM10 expression was accompanied by increased PrPC amounts as revealed by Western blot analysis of cortical homogenates ( Figure 2A ) . Residual ADAM10 most likely resulted from glial cells not depleted of ADAM10 . In contrast to the cortex , differences in ADAM10 expression and PrPC levels between A10 cKO and littermate controls were not seen in the cerebellum , a brain region not affected by our knockout strategy ( Casanova et al . , 2001; Prox et al . , 2013 ) ( Figure 2A ) . Immunohistochemical analysis in P19 mouse brains confirmed the Western blot data by showing increased immunostaining for PrPC in A10 cKO mice in several regions of the forebrain , as exemplified by the hippocampus and cortex , whereas no such increase was observed in the cerebellum ( Figure 2B ) . In addition , a coronal brain section showing costaining of PrPC with the neuronal marker NeuN correlates with the Camk2a driven ADAM10 knockout strategy ( Casanova et al . , 2001 ) by showing increased PrPC expression in hippocampal and cortical areas as well as in the striatum ( Figure 2—figure supplement 1 ) . Increased PrPC levels in A10 cKO mice did not result from transcriptional upregulation as there were no significant differences in PrPC mRNA levels in the forebrain at P19 as assessed by quantitative RT-PCR ( mean ± SD: controls: 5 . 96 ± 0 . 86; A10 cKO: 6 . 32 ± 1 . 04; p = 0 . 23; n = 5 ) ( Figure 2C ) . We also analyzed PrPC levels in 35-week-old A10 cKO and littermate control mice . In contrast to newborn A10 cKO mice ( Prox et al . , 2013 ) , adult A10 cKO mice did not show differences regarding weight and body size ( Figure 2D and Video 1 ) . Again , as shown at P19 , adult A10 cKO mice also showed increased levels of PrPC in the cortex and hippocampus compared with their littermate controls , as assessed by immunohistochemistry at 35 weeks of age ( Figure 2E ) . Taking these observations together , A10 cKO mice are viable and show a temporally unrestricted posttranslational increase of PrPC in neurons . 10 . 7554/eLife . 04260 . 004Figure 2 . Characterization of PrPC expression in juvenile and adult Camk2a-Cre A10 cKO mice and littermate controls . ( A ) Western blot analysis of mature ADAM10 ( mADAM10 ) and PrPC expression in different brain homogenates from the cortex ( left ) and cerebellum ( right ) of A10 cKO and control mice at postnatal day ( P ) 19 . Actin served as a loading control . ( B ) Immunohistochemical detection of PrPC in forebrain of P19 mice of both genotypes . Overview ( top ) and magnifications showing cortex ( Cx ) and hippocampal CA2 and CA3 regions . Overview and details of cerebellum ( Cb ) are shown below . Scale bars represent 100 µm ( insets for Cb: 50 µm ) . ( C ) Quantitative RT-PCR analysis of PrPC mRNA levels in A10 cKO mice and controls at P19 ( n = 5 for each genotype ) . GAPDH served as a control for normalization . Error bars indicate SD . ( D ) Adult age-matched A10 cKO and wild-type littermates had a comparable body size at 35 weeks of age . ( E ) Representative immunohistochemical staining of PrPC in cortex ( Cx ) and hippocampal CA1 region of adult ( 35 weeks ) A10 cO and control mice . Prion protein knockout ( Prnp0/0 ) and overexpressing mice ( tga20 ) served as negative and positive controls , respectively ( scale bar: 100 µm ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04260 . 00410 . 7554/eLife . 04260 . 005Figure 2—figure supplement 1 . Regional distribution of increased PrPC levels in a coronal brain section of an ADAM10 cKO mouse compared with a wild-type littermate control . Brain sections derived from postnatal day 19 mice of both genotypes co-stained for PrPC ( brownish signal ) and neuronal marker NeuN ( red signal ) . Scale bar represents 200 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 04260 . 00510 . 7554/eLife . 04260 . 006Video 1 . Clinical presentation of a prion-infected ADAM10 cKO and littermate control mouse at 95 days post inoculation ( dpi ) . Terminally diseased ADAM10 cKO mouse ( cage on the left ) directly prior to termination of the experiment showing lack of nest-building behaviour and late hypoactivity whereas a wild-type littermate ( cage on the right ) matched for dpi presents with a regular nest and normal activity . DOI: http://dx . doi . org/10 . 7554/eLife . 04260 . 006 Based on previous reports ( Casanova et al . , 2001; Prox et al . , 2013 ) and on our findings presented here ( Figure 2 ) , a qualitative representation showing the regional distribution of the neuronal Camk2a-Cre expression ( and thus of the ADAM10 knockout and increased PrPC expression ) is provided in Figure 3A . In addition , this scheme depicts information on our experimental design described in detail below , including the site of prion inoculation and brain regions chosen for immunohistochemical or biochemical analyses . 10 . 7554/eLife . 04260 . 007Figure 3 . Overview of the experimental design and summary of most important findings . ( A ) Scheme of a mouse brain combining a qualitative representation of the Camk2a driven ADAM10 knockout strategy and information on the sampling of specimen . The site of intracerebral inoculation of mice with RML prions is indicated by the red encircled dot . Samples of frontal brain ( dotted box ) were taken for biochemical analysis and determination of infectivity titers ( bioassay ) . The rest of the brain was formalin-fixed and embedded in paraffin . Coronal sections were prepared from different layers ( dashed lines ) with varying distance to the site of prion inoculation ( as indicated by blue arrows ) and assessed by immunohistochemical analysis . ( B ) Qualitative comparison of mouse genotypes ( A10 cKO , controls , tga20 ) with regard to PrPC or PrPSc levels , prion-associated neuropathology ( including spongiosis , astrocytosis , and microglia activation ) and prion infectivity titers according to brain region and time point . Reference to corresponding figures showing original data is provided . Cb = Cerebellum; Cx = Cortex; Hc = Hippocampus; Stri = Striatum; Tha = Thalamus; n . a . = not assessed; Qualities: +++ = high/strong; ++ = medium/moderate; + = low/basal; ( + ) = very low/weak; o = none . DOI: http://dx . doi . org/10 . 7554/eLife . 04260 . 007 Next , we studied how the absence of the PrPC sheddase impacts on the course of prion disease . For better orientation , Figure 3B summarizes the most important findings presented in the following paragraphs including comparisons of PrPSc levels , prion-associated neuropathology , and infectivity titers for the different genotypes , brain regions , and time points investigated in this study . Moreover , this overview provides reference to the corresponding figures showing the original data described below . We inoculated A10 cKO ( n = 7 ) and littermate control mice ( n = 8 ) at 6 weeks of age intracerebrally with the Rocky Mountain laboratory ( RML ) strain of prions . Mock ( CD1 ) -inoculated A10 cKO mice ( n = 5 ) or wild-type littermates ( n = 11 ) and prion inoculated , prion hypersensitive , PrPC overexpressing tga20 mice ( n = 8 ) served as negative and positive controls , respectively . Interestingly , we found significantly reduced incubation times until onset of terminal prion disease in the A10 cKO mice ( 103 ± 7 days post inoculation ( dpi ) ; SD ) compared with littermate controls ( 146 ± 2 dpi ) ( Figure 4 and Table 1 ) . Clinical manifestations ( starting with a neglect of nest formation and fur cleaning , followed by presentation of a stiff tail and gait abnormalities , and terminating with kyphosis , weight loss , and hypoactivity ) and disease duration ( 35 ± 4 days for A10 cKO mice and 40 ± 2 . 4 days for controls ) were comparable between A10 cKO and littermate control mice ( Figure 4 , Table 1 , Video 1 ) . In line with other studies ( Fischer et al . , 1996; Sandberg et al . , 2011 ) , tga20 mice showed fastest disease development and had to be sacrificed between 59 and 65 dpi ( Figure 3 ) whereas mock-infected A10 cKO and control mice were taken out of the experiment at 200 dpi without any signs of prion disease ( Figure 4 and Table 1 ) . 10 . 7554/eLife . 04260 . 008Figure 4 . Survival curves of mice upon intracerebral inoculation with RML prions . Age-matched juvenile ( 6-week-old ) A10 cKO mice ( n = 7 ) , littermate control mice ( n = 8 ) , and tga20 mice ( n = 8 ) were intracerebrally inoculated with RML prions ( +RML ) and time until development of terminal prion disease was measured as days post inoculation ( dpi; ***p = 7 . 6 × 10−7 ) . As a negative control ( mock ) , 5 A10 cKO mice and 11 littermate controls were intracerebrally inoculated with brain homogenate of CD1 mice and sacrificed without clinical signs at 200 dpi . DOI: http://dx . doi . org/10 . 7554/eLife . 04260 . 00810 . 7554/eLife . 04260 . 009Table 1 . Clinical presentation of A10 cKO , littermate control and tga20 mice inoculated with RML prionsDOI: http://dx . doi . org/10 . 7554/eLife . 04260 . 009RML prionsCD1 mockClinical manifestationsA10 cKO ( n = 7 ) Controls ( n = 8 ) tga20 ( n = 8 ) A10 cKO ( n = 5 ) Controls ( n = 11 ) First clinical signs ( dpi ) 68 ( ±3 . 5 ) 106 ( ±2 . 5 ) n . a . NoneNoneDuration of clinical signs ( d ) 35 ( ±3 . 8 ) 40 ( ±2 . 4 ) n . a . --Time to terminal disease ( dpi ) 103 ( ±6 . 6 ) 146 ( ±2 . 1 ) 62 ( ±2 . 8 ) **Progression of signsSteadySteadyRapid––Quitting of nest-buildingAllAllAll––Ungroomed coatMost ( 6/7 ) Most ( 6/8 ) Rare ( 2/8 ) ––Stiff tailAllMost ( 7/8 ) Most ( 5/8 ) ––Gait disturbanceAllAllAll––Hind leg paresisRare ( 2/7 ) Rare ( 3/8 ) Most ( 7/8 ) ––KyphosisMost ( 6/7 ) Most ( 6/8 ) Rare ( 3/8 ) ––Weight lossAllAllMost ( 6/8 ) ––Late hypoactivityAllAllMost ( 5/8 ) ––Negative controls included A10 cKO and littermate controls inoculated with CD1 brain homogenates ( mock ) . Asterisks indicate that CD1 mock-inoculated animals were sacrificed at 200 days post inoculation ( dpi ) without any clinical signs . In summary , depletion of the PrPC sheddase ADAM10 in neurons of the forebrain led to a significant reduction in incubation times of more than 40 days until onset of terminal prion disease without affecting overall clinical presentation . We performed biochemical and neuropathological analyses of terminally prion-diseased A10 cKO and control mice in order to elucidate the reasons for reduced incubation times in the A10 cKO mice . In terminally diseased mice , comparable levels of PrPSc were found in frontal brain homogenates of both genotypes as assessed by Western blot analysis after digestion of samples with PK ( Figure 5A ) . As expected , no PrPSc was found in mock-inoculated controls of each genotype ( Figure 5A on the right ) . Since alterations in incubation times may indicate generation of different prion strains ( Barron et al . , 2003 ) with altered PrPSc glycosylation pattern and altered sizes of core fragments of PrPSc , we investigated whether lack of ADAM10-mediated shedding resulted in an altered PrPSc banding pattern . To this end , we measured the relative proportion of di- , mono- , and unglycosylated forms of PrPSc and sizes of PrPSc core fragments . No differences in the PrPSc glycopattern ( Figure 5B ) or sizes of PrPSc core fragments ( Figure 5A ) between terminally diseased A10 cKO and control mice were detected , which argues against the generation of a modified prion strain . 10 . 7554/eLife . 04260 . 010Figure 5 . Analysis of terminally prion-diseased A10 cKO mice and littermate controls . ( A ) Detection of PrPSc by Western blot analysis of proteinase K ( PK ) -digested ( PK+ ) forebrain homogenates of prion-infected ( RML+; blot on the left ) and mock-inoculated ( RML−; blot on the right ) A10 cKO mice and controls at terminal stage of disease . Samples from three ( RML+ ) or two ( RML− ) representative animals per genotype are shown . Digestion controls included prion-diseased/undigested ( RML+/PK− ) , prion-negative/PK-digested ( RML−/PK+ ) , or prion-diseased/digested ( RML+/PK+ ) brain homogenates . ( B ) Comparison of proportions of di- , mono , and unglycosylated PrPSc forms between A10 cKO ( n = 4 ) and control mice ( n = 4; error bars indicate SEM ) . ( C ) Representative histological analysis in the forebrain of terminally prion-diseased ( RML terminal ) and healthy mock-inoculated ( CD1 mock ) A10 cKO and control mice including H&E staining and immunostaining of PrPSc , GFAP ( for detection of astrocytes ) , and Iba-1 ( for detection of microglia ) . Scale bar in overviews: 200 µm; scale bar for inlays showing magnifications of representative areas: 100 µm . ( D ) Electron microscopy photographs showing intraneuronal vacuoles ( asterisk in ( a ) and ( b ) ) containing membranous structures ( small arrowheads ) , enlarged and densely packed multivesicular bodies ( arrows in ( b ) ) as well as autophagic membranes ( arrow in ( c ) ) and autolysosomes that were found in terminally prion-infected forebrains of both genotypes ( exemplified here for an A10 cKO brain ) . Scale bars represent 500 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 04260 . 01010 . 7554/eLife . 04260 . 011Figure 5—figure supplement 1 . Neuropathological features in the thalamus of ADAM10 cKO and control mice . Representative histological analysis in the thalamic brain region of terminally prion-diseased ADAM10 cKO and control mice including H&E staining and immunohistochemical detection of PrPSc , GFAP ( for detection of astrocytes ) , and Iba-1 ( for detection of microglia ) . Scale bar: 100 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 04260 . 01110 . 7554/eLife . 04260 . 012Figure 5—figure supplement 2 . Electron microscopic analysis . Overviews showing vacuolization ( asterisks ) in forebrain samples of ADAM10 cKO ( A ) and control mice ( B ) at a terminal stage of prion disease . High abundance of clusters containing tubulovesicular structures ( TVS; arrows in ( A ) and ( C ) ) was only found in terminal ADAM10 cKO brain ( 5 out of 13 square grids showed clusters of TVS whereas only 1 out of 15 square grids presented with TVS in terminal wild-type control mice ) . Scale bars represent 500 nm ( A , B ) or 200 nm ( C ) . Inset in ( C ) shows magnification of TVS . DOI: http://dx . doi . org/10 . 7554/eLife . 04260 . 012 Prion infection caused spongiform lesions , massive astrocytosis , and microgliosis in both genotypes ( Figure 5C on the left ) . Similar levels of PrPSc immunopositivity were found at this terminal stage of prion disease ( Figure 5C ) , thus confirming our biochemical data ( Figure 5A ) . Comparable morphological findings were observed in the thalamic brain region ( Figure 5—figure supplement 1 ) . Electron microscopic analysis revealed the presence of typical spongiform changes , dystrophic neurites , and abundant autophagy including enlarged multivesicular bodies and autolysosomes in both genotypes ( Figure 5D ) . Remarkably , prion-diseased A10 cKO mice presented with an extraordinarily high abundance of tubulovesicular structures ( TVS ) ( Figure 5—figure supplement 2 ) ( Jeffrey and Fraser , 2000; Liberski et al . , 2010 ) . These are spherical structures of 25–37 nm diameter that are specific for prion diseases though devoid of PrP ( Jeffrey and Fraser , 2000; Liberski et al . , 2008 , 2010 ) . Thus , our model may allow purification of these structures and could contribute to unraveling the nature and relevance of TVS in prion diseases . Taken together , these data indicate that , at terminal prion disease , the lack of PrPC shedding does not influence local PrPSc distribution while it does affect the appearance of TVS . To elucidate the order of events underlying changed disease kinetics , we measured PrPSc production in A10 cKO mice and littermate controls . Since in A10 cKO mice levels of PrPC are posttranslationally increased , we also investigated this issue in tga20 mice with genetically increased Prnp expression . To this aim , we performed Western blot analysis of total PrP ( i . e . , PrPC plus PrPSc ) and PK-resistant PrPSc in A10 cKO and littermate control mice matched for days post prion inoculation ( 95 dpi ) as well as in terminally prion-diseased tga20 mice ( 65 dpi ) . Relative levels of total PrP were highest in tga20 mice ( 5 . 0 ± 0 . 3; SEM ) , with increased levels in A10 cKO mice ( 2 . 8 ± 0 . 4 ) compared with controls ( set to 1 ± 0 . 1 ) ( A10 cKO vs control: **p = 0 . 0022; tga20 vs control: ***p = 0 . 0005; A10 cKO vs tga20: **p = 0 . 0015 ) ( Figure 6A ) . Surprisingly , relative levels of PrPSc were significantly elevated in A10 cKO mice ( 3 . 9 ± 0 . 8 ) , with moderate levels in littermate controls ( set to 1 ± 0 . 2 ) and reduced levels in tga20 mice ( 0 . 6 ± 0 . 3 ) ( A10 cKO vs control: *p = 0 . 049; A10 cKO vs tga20: *p = 0 . 032 ) . There were no atypical PrP patterns as assessed by PK digestion at 4°C ( ‘cold PK’ ) or with lower dilutions of PK ( Figure 6—figure supplement 1 ) . In summary , A10 cKO mice showed significantly increased PrPSc formation compared with littermate controls and tga20 mice , which cannot solely be explained by increased neuronal PrPC levels . 10 . 7554/eLife . 04260 . 013Figure 6 . PrPSc formation , neuropathology , toxic signaling , and calpain levels at a matched time point . ( A ) Assessment of total PrP ( no proteinase K ( PK ) ) and PrPSc amounts ( +PK; blot is shown with short and longer exposition ) by parallel replica Western blot analysis in forebrain homogenates of age-matched A10 cKO mice and littermate controls ( both at 95 days post inoculation ( dpi ) ; n = 3 for each genotype ) as well as terminally diseased tga20 mice ( at 65 dpi; n = 3 ) . Actin was detected in the undigested homogenates ( no PK ) and served as loading control . Densitometric quantification of relative protein amounts from two technical replicates is shown on the right . ( B ) Morphological analysis of neuropathological lesions in forebrains ( showing hippocampal and cortical brain regions ) of A10 cKO , littermate controls , and tga20 mice at the aforementioned time points ( scale bars: 200 µm in overviews and 100 µm in insets and for PrPSc ) . ( C ) Biochemical assessment of candidate toxic signaling pathways showing protein levels of total Fyn ( t-Fyn ) , phosphorylated ( Tyr416 ) Src ( p-Src ) as well as total ( t-Erk1/2 ) and phosporylated ( Thr202/Tyr204 ) Erk1/2 ( p-Erk1/2 ) . Actin served as a loading control and for normalization ( # and § indicate use of the same actins as corresponding signaling proteins were detected on the same Western blot ) . Quantitative densitometric analysis of relative p-Src/t-Fyn ( left ) and p-Erk/t-Erk ratio ( right ) ( n . s . = not significant ) . ( D ) Representative Western blot analysis ( left ) and quantification of three technical replicates ( right ) of calpain levels in aforementioned samples . Levels of ADAM10 are shown in ( C ) and ( D ) to confirm the ADAM10 status . Error bars indicate SEM; *p <0 . 05; **p <0 . 01 , ***p <0 . 001 ( p values of Student's t-test are given in the main text ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04260 . 01310 . 7554/eLife . 04260 . 014Figure 6—figure supplement 1 . ‘Cold proteinase K ( PK ) ’ and partial PK digestion of forebrain samples from ADAM10 cKO , control and tga20 mice at 95 days post inoculation ( dpi ) . ( A ) Western blot analysis showing ‘cold PK’ digestions of forebrain homogenates from two animals per genotype with 200 µg/ml PK at 4°C for 1 hr compared with classical digestion with 20 µg/ml PK at 37°C . No atypical digestion pattern is detected . Antibody POM1 was used for detection . ( B ) Representative Western blot analysis of frontal brain homogenates incubated with decreasing amounts of PK or without PK . Blot was first incubated with 1E4 antibody and then re-probed with POM1 . Apart from differences in total PrP and PrPSc amounts between the genotypes described in the main Figure 6 , neither atypical bands nor low abundance or small PrP digestion fragments ( ∼10 kDa ) are detected . DOI: http://dx . doi . org/10 . 7554/eLife . 04260 . 01410 . 7554/eLife . 04260 . 015Figure 6—figure supplement 2 . Calpain levels and calpain substrates in forebrain homogenates of ADAM10 cKO and control mice . ( A ) Representative Western blot analysis of calpain expression in non-prion infected ADAM10 cKO and control mice . In addition , premature ( pADAM10 ) , mature ADAM10 ( mADAM10 ) and PrPC were detected . Quantification of calpain levels by densitometric analysis of ADAM10 cKO ( n = 4 ) and control mice ( n = 5 ) is shown below . Actin served as loading control . ( B ) Western blot analysis of spectrin ( FL = full length ) and spectrin breakdown products ( SBDP ) in prion-infected ADAM10 cKO and control mice at 95 days post inoculation ( dpi; n = 3 per genotype ) . FL spectrin was reduced in ADAM10 cKO mice whereas calpain-dependent SBDP ( marked by asterisk at 150/145 kDa ) were not increased . ( C ) Western blot analysis of p35 and p25 levels in prion-infected samples mentioned in ( B ) . A reduction in p35 levels was found in two out of three ADAM10 cKO mice . A band corresponding to p25 was not detectable . # indicates that blots used for calpain detection ( Figure 6D ) were re-probed with an antibody against p35/p25 and thus the same actin signals were used as loading controls . DOI: http://dx . doi . org/10 . 7554/eLife . 04260 . 015 Neuropathological changes in the forebrain , including spongiosis , astrocytosis and microgliosis , were more prominent in A10 cKO mice ( Figure 6B ) . Remarkably , severe differences could be observed for PrPSc immunopositivity , with A10 cKO mice showing prominent PrPSc deposits in cortical and hippocampal regions while controls and tga20 mice showed only moderate levels , thus confirming the biochemical data ( Figure 6A ) . Next , we addressed the question whether elevated ( membrane ) levels of PrPC in A10 cKO and tga20 mice caused an increase in PrPC-mediated neurotoxic signaling events . Therefore , we biochemically assessed two candidate signaling pathways , via the Src kinase Fyn and the MAP kinases Erk1/2 , that have previously been reported to be activated upon binding of toxic oligomers to PrPC ( Mouillet-Richard et al . , 2000; Schneider et al . , 2003; Larson et al . , 2012; Um et al . , 2012; Pradines et al . , 2013 ) . However , we did not observe any significant differences in the activation state of these kinases in forebrain homogenates between tga20 , A10 cKO , and littermate control mice at the chosen time point ( Figure 6C ) . A10 cKO mice showed significantly increased PrPSc production ( Figure 6A , B ) compared with controls ( at 95 dpi ) and tga20 mice ( at 65 dpi ) . Since PrPSc-associated neurotoxicity and cell death have been linked to formation of membrane pores by binding and clustering of PrPSc to PrPC at the plasma membrane resulting in calpain upregulation ( Falsig et al . , 2012; Sonati et al . , 2013 ) , we investigated this in our mice . Specifically , we biochemically analyzed calpain expression in the frontal brain of all experimental groups at the aforementioned time points . Interestingly , we found significant upregulation of calpain expression in A10 cKO mice ( 1 . 78 ± 0 . 24; SEM ) compared with littermate controls ( set to 1 ± 0 . 07; *p = 0 . 034 ) and tga20 mice ( 0 . 93 ± 0 . 14; *p = 0 . 022 ) ( Figure 6D ) , thus correlating with PrPSc levels ( Figure 6A , B ) . This seemed to be specific for prion disease as no differences were detected in non-prion infected A10 cKO mice ( mean: 0 . 90 ± 0 . 26 SEM; n = 4 ) when compared with age-matched wild-type controls ( set to 1 ± 0 . 19; n = 5; p = 0 . 7 ) ( Figure 6—figure supplement 2A ) . In prion disease we observed a reduction of the described calpain substrates spectrin ( Falsig et al . , 2012 ) and neuron-specific activator p35 ( Lee et al . , 2000 ) , suggesting that increased expression in A10 cKO mice correlated with the activity of calpain ( Figure 6—figure supplement 2B , C ) . However , it should be noted that we failed to detect differences in specific cleavage products of spectrin and p35 . Our data suggest that lack of ADAM10-mediated PrPC shedding leads to increased production of PrPSc and upregulation of calpain expression , whereas activation of proposed toxic PrPC-dependent signaling pathways was not observed . We next investigated whether reduced incubation times and increased PrPSc conversion rates in A10 cKO mice were accompanied by elevated prion infectivity titers and therefore performed bioassays ( Figure 7 ) . We intracerebrally inoculated forebrain homogenates of A10 cKO and littermate control mice at terminal ( ∼103 dpi for A10 cKO and ∼146 dpi for controls; see Table 1 ) and preclinical ( 60 dpi , 35 dpi ) prion disease into tga20 ( n = 4 for each sample ) reporter mice . In addition , we assessed the 95 dpi time point ( n = 6 tga20 mice per sample ) at which A10 cKO mice were clinically affected while controls were still asymptomatic and where we observed significant differences in PrPSc loads ( Figure 6A , B ) . Titers of infectious prions increased with progression of disease , showed only some inter-individual variation , and reached a plateau phase towards the terminal stage of disease . However , they were independent of ADAM10 expression since we did not find differences between the two genotypes . At early time points ( 35 dpi ) prion titers were low ( mean: A10 cKO: 3 . 9 ± 0 . 2 logLD50; control: 4 . 0 ± 0 . 2 ) , intermediate titers were seen at 60 dpi ( A10 cKO: 5 . 0 ± 0 . 5; control: 4 . 9 ± 0 . 8 ) , and highest titers were observed in terminally ill mice ( A10 cKO: 5 . 8 ± 0 . 2; control: 6 . 0 ± 0 . 2 ) ( Figure 7 ) . Interestingly , significantly increased PrPSc amounts in A10 cKO mice at 95 dpi ( Figure 6A , B ) did not result in elevated prion infectivity titers ( A10 cKO: 5 . 74 ± 0 . 05; control: 5 . 66 ± 0 . 08; p = 0 . 2 ) . 10 . 7554/eLife . 04260 . 016Figure 7 . Titers of prion infectivity in terminally diseased and preclinical A10 cKO mice and controls . Prion titers in forebrain homogenates of A10 cKO mice and littermate controls at terminal or matched ( preclinical ) time points ( 95 days post inoculation [dpi] , 60 dpi , and 35 dpi ) after prion inoculation as assessed by bioassays in tga20 reporter mice . Each dot indicates the titer ( shown as logLD50 ) assessed in a single reporter mice . Bars indicate mean values from six ( for the 95 dpi time point ) or four ( all other time points ) tga20 mice . Error bars indicate SD . DOI: http://dx . doi . org/10 . 7554/eLife . 04260 . 016 Since PrPSc is shed from the plasma membrane in vitro ( Taylor et al . , 2009 ) and anchorless PrPC can convert into PrPSc ( Chesebro et al . , 2005 , 2010; Stöhr et al . , 2011 ) , we were interested to investigate if lack of PrP shedding affects the spread of prion pathology within the brain . Therefore , we performed a morphology-based spatiotemporal analysis in prion-infected A10 cKO and littermate control mice . We assessed a brain region in direct proximity to the site of prion administration ( striatum ) and brain regions ( cerebellum and brain stem ) distant from the inoculation site and not directly affected by the depletion of ADAM10 ( Figure 2; see Figure 3 for overview ) ( Casanova et al . , 2001; Prox et al . , 2013 ) . Whereas for striatum we observed neuropathological changes typically found in terminal prion disease irrespective of the ADAM10 status , we found an apparently reduced degree of spongiosis in the cerebellum and brain stem of A10 cKO mice compared with terminally diseased littermate controls ( Figure 8A and Figure 8—figure supplement 1 ) . In line with this , other prion-associated pathological hallmarks , including PrPSc immunopositivity , astrocytosis involving Bergmann glia , and microglia activation , were observed in the cerebellum of controls , but were virtually lacking in this brain region in A10 cKO mice ( Figure 8A ) . 10 . 7554/eLife . 04260 . 017Figure 8 . Spatiotemporal analysis of prion-associated pathology . ( A ) Comparison of neuropathological features including PrPSc deposition , spongiotic changes ( presented in H&E stainings ) , astrocyte ( GFAP ) and microglia ( Iba-1 ) activation in striatum ( as a site in close proximity to prion inoculation ) and cerebellum ( resembling a brain area distant to prion inoculation ) in A10 cKO and littermate control mice at a terminal stage of prion disease ( i . e . , ∼103 days post inoculation [dpi] for A10 cKO and ∼146 dpi for controls ) . While similar pathological alterations were observed in the striatum of both genotypes , prion-associated lesions in the cerebellum were almost absent in A10 cKO mice . ( B ) At a matched time point ( 95 dpi for A10 cKO and littermate controls; 65 dpi [i . e . , terminal disease] for tga20 ) , prion-related pathology was found in the striatum of all genotypes analyzed . As described earlier for the cortex and hippocampus ( Figure 6 ) , differences in PrPSc amounts could also be observed in the striatum . In the cerebellum of both time-matched A10 cKO mice and littermate controls , prion-associated lesions were largely absent whereas tga20 mice showed all relevant neuropathological features already at 65 dpi . Representative pictures for at least three animals per genotype and time point are shown . Scale bars represent 200 µm ( overviews ) or 100 µm ( insets ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04260 . 01710 . 7554/eLife . 04260 . 018Figure 8—figure supplement 1 . Spongiosis in brain stem and cerebellum of ADAM10 cKO and control mice . Representative histological analysis of spongiotic vacuolization as revealed by H&E staining of the cerebellum and brain stem of terminally prion-diseased control and ADAM10 cKO mice . Upper row shows overview and lower two rows show higher magnifications of cerebellum and brain stem ( scale bars represent 100 µm ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04260 . 018 Reduced disease duration in A10 cKO mice leading to diminished dissemination of prion-associated pathology may represent a possible explanation for these findings . In order to control for this , we assessed neuropathological alterations at a matched time point ( 95 dpi ) in both A10 cKO mice and littermate controls . Additionally , we included terminally prion-diseased tga20 mice ( 65 dpi ) . In the striatum , typical neuropathological changes were seen in all groups of mice . Similar to hippocampus and cortex ( Figure 6B ) , increased PrPSc immunopositivity and astrogliosis were seen in A10 cKO mice ( Figure 8B ) . In line with our biochemical findings ( Figure 6A ) , tga20 mice showed lowest PrPSc amounts in the forebrain . Interestingly , in the cerebellum and brain stem , relevant prion-typical neuropathological changes were observed in tga20 mice at 65 dpi but were almost absent in both A10 cKO and littermate controls at 95 dpi ( Figure 8B ) . This argues against a temporal cause for the observed differences and rather supports an accelerating role of shedding in the spread of prion disease within the CNS . Since tga20 mice show enhanced rates of PrPC shedding ( Figure 1E ) ( Altmeppen et al . , 2011 ) , our spatiotemporal data may indicate a correlation between PrPC shedding and the efficiency of prion spread within the brain , with tga20 mice having the highest efficiency whereas A10 cKO mice show the least efficient dissemination of prion pathology ( see model in Figure 9 ) . 10 . 7554/eLife . 04260 . 019Figure 9 . Model of the dual role of ADAM10-mediated shedding in prion disease . ADAM10 regulates PrPC levels at the plasma membrane and releases almost full length PrP into the extracellular space . Thereby it affects ( i ) neurotoxicity , ( ii ) PrPSc formation , and ( iii ) spreading of prion pathology . ( i ) Lack of ADAM10 ( as assessed here by use of A10 cKO mice ) results in elevated PrPC membrane levels . Membrane levels of PrPC ( as a receptor ) likely determine PrPSc-associated neurotoxicity ( as indicated by sizes of thunderbolts and skulls ) and thereby incubation times with shortest survival in tga20 mice and reduced incubation times in A10 cKO mice compared with wild-type littermates with longest survival ( order reflected by grey triangles on the left ) . ( ii ) Shed PrP , which is most efficiently produced in tga20 and absent in A10 cKO mice , might block formation of PrPSc . This is reflected by the different PrPSc amounts found in our different experimental groups ( A10 cKO > wild-type littermates > tga20 ) . The combination of increased PrPC membrane levels and PrPSc formation in A10 cKO mice might favor increased production of membrane pores ( as indicated in the middle row on the left ) and neurotoxic Ca2+ influx with possible ( ‘ ? ’ ) involvement of calpain . ( iii ) Finally , spread of prion-associated pathology within the brain also seems to be affected by the levels of ADAM10 expression since tga20 mice showed enhanced whereas A10 cKO mice showed reduced dissemination of neuropathological features ( as indicated by size of arrowheads on the right ) . Key references supporting this model are given in the text . Based on this model , stimulation of ADAM10 might therefore offer a treatment option . With regard to incubation times , protective effects by reducing local membrane bound PrPC amounts and by producing a protective soluble fragment able to block PrPSc formation seem to predominate the disadvantage of increased spread by production of anchorless prions . DOI: http://dx . doi . org/10 . 7554/eLife . 04260 . 019
Proteolytic processing of the prion protein is critically involved in controlling its physiological functions . For α-cleavage , a protective role for both prion and AD is postulated and attributed to destruction of the neurotoxic domain and production of the neuroprotective N1 fragment respectively ( Guillot-Sestier et al . , 2009; Resenberger et al . , 2011; Westergard et al . , 2011; Beland et al . , 2012; Guillot-Sestier et al . , 2012; Fluharty et al . , 2013 ) . Moreover , the C1 fragment of PrPC is involved in myelination ( Bremer et al . , 2010 ) . The protective role of α-cleavage is also supported by the fact that α-cleavage-resistant PrPC deletion mutants are intrinsically neurotoxic upon expression in mice ( Shmerling et al . , 1998; Baumann et al . , 2007 ) ( reviewed in Yusa et al . , 2012 ) . In contrast , our knowledge of the consequences of PrPC shedding , though being evolutionary conserved and constitutively occurring under physiological conditions , is limited . In a previous study we have shown that ADAM10 represents the principal PrPC sheddase in vivo with lack of ADAM10 resulting in a posttranslational increase of PrPC levels ( Altmeppen et al . , 2011 ) . This pointed to a key role for this protease in the regulation of PrPC levels at the neuronal surface , yet its impact on prion diseases remained unknown . We recently generated A10 cKO mice , a viable model with selective postnatal ablation of ADAM10 in neurons of the forebrain ( Prox et al . , 2013 ) . In the present study we demonstrate that prion-inoculated A10 cKO mice show considerably reduced incubation times and increased conversion of PrPC to PrPSc . In contrast , spread of prion pathology within the brain is reduced , thus indicating a dual role for ADAM10 in the pathophysiology of prion disease ( Figure 9 ) . ADAM10 acts as a sheddase for a number of neuronally expressed proteins with important functions in the development and homeostasis of the brain ( Yang et al . , 2006; Weber and Saftig , 2012 ) . For PrPC it is now widely accepted that ADAM10 is the PrPC sheddase ( Parkin et al . , 2004; Taylor et al . , 2009; Altmeppen et al . , 2011; Wik et al . , 2012; Ostapchenko et al . , 2013; McDonald et al . , 2014 ) . Here we extend our previous findings on increased levels of PrPC in the absence of ADAM10 ( Altmeppen et al . , 2011 ) in different cellular models and observe colocalization between PrPC and ADAM10 at the plasma membrane , which is regarded as the primary site of ADAM10-mediated shedding events ( Lammich et al . , 1999; Chen et al . , 2014 ) . In addition to accumulation of PrPC in the early secretory pathway ( Altmeppen et al . , 2011 ) , we here show that lack of ADAM10 also results in elevated membrane levels of PrPC . This supports a model where ADAM10-mediated shedding controls PrPC membrane homeostasis ( Altmeppen et al . , 2013 ) . However , cell culture models as well as mice with pan-neuronal depletion of ADAM10 did not allow investigation of the role of PrP shedding in prion diseases ( Altmeppen et al . , 2011 ) . Therefore , we used A10 cKO mice ( Prox et al . , 2013 ) in which we observed increased PrPC levels that were not caused by elevated PrPC mRNA levels . Onset of terminal prion disease in mice upon intracerebral inoculation with defined prion strains occurs after defined incubation times . Once clinical symptoms appear , individual mice reliably succumb to disease within days . In contrast to peripheral prion inoculation where the integrity of the lymphoreticular or autonomous nervous system plays a decisive role in neuroinvasion of prions , incubation time after intracerebral inoculation is mainly influenced by PrPC levels ( Mabbott et al . , 2000; Glatzel et al . , 2001; Prinz et al . , 2003 ) . This is nicely exemplified by tga20 mice constitutively overexpressing PrPC ( Fischer et al . , 1996; Sandberg et al . , 2011 ) . Our intracerebrally inoculated A10 cKO mice developed prion disease >40 days earlier than controls , corresponding to a shortening of incubation times of ∼30% . Such dramatic reductions upon intracerebral inoculation with high-dose RML prions have so far only been achieved when PrPC is massively overexpressed ( Fischer et al . , 1996 ) . Less dramatic reductions were , for instance , described in mice lacking superoxide dismutase 1 ( Akhtar et al . , 2013 ) or overexpressing the heat shock protein Hspa13 ( Grizenkova et al . , 2012 ) . The only other study investigating the role of ADAM10 in prion disease showed that transgenic overexpression of bovine ADAM10 in mice led to prolongation of prion disease ( Endres et al . , 2009 ) , a finding which is complementary to our data on the effect of ADAM10 deficiency . However , these authors linked this effect to transcriptional downregulation of PrPC and not to proteolytic processing events . In contrast , data from the present and previous studies show that transcriptional regulation of PrPC is not affected by ADAM10 whereas proteolytic processing clearly is ( Parkin et al . , 2004; Taylor et al . , 2009; Altmeppen et al . , 2011; Wik et al . , 2012; Ostapchenko et al . , 2013; McDonald et al . , 2014 ) . We considered the possibility that non-PrP dependent effects resulting from ADAM10 deficiency may have contributed to disease acceleration . Indeed , ADAM10 is involved in a number of developmental and immunological processes and A10 cKO mice show learning deficits ( Weber and Saftig , 2012; Prox et al . , 2013 ) . It is virtually impossible to dissect out the individual contribution of the multitude of ADAM10 targets to the pathophysiology of prion disease . For instance , microglia contribute to prion pathophysiology and express CD40L , a known ADAM10 substrate . Prion-inoculated CD40L-deficient mice show decreased incubation times ( Burwinkel et al . , 2004 ) . As CD40L is not expressed on neurons and its receptor ( CD40 ) is not processed by ADAM10 , we do not expect that this dyad is altered by our neuron-specific knockout strategy ( Schönbeck and Libby , 2001; Burwinkel et al . , 2004 ) . Another microglia-related mechanism relevant to prion disease and influenced by ADAM10 is the Cx3cl1/Cx3cr1 signaling complex . The receptor ( Cx3cr1 ) is expressed on microglia whereas its ligand Cx3cl1 ( Fractalkine ) is an ADAM10 substrate expressed on neurons . Cx3cl1-deficient mice inoculated with prions show slightly decreased incubation times to disease ( Grizenkova et al . , 2014 ) . Therefore , we do not assume that these events explain the strong effect observed in our study . Most other immune system-mediated effects on prion pathogenesis are only relevant for peripheral prion inoculation and neuroinvasion and do not play major roles in intracerebral prion administration ( Aguzzi et al . , 2003 ) . Furthermore , A10 cKO mice do not show any gross morphological abnormalities in the CNS , have a normal life span once they have survived a critical weaning period , and mock inoculated mice behave normally ( Prox et al . , 2013 ) . Finally , we consider that increased PrPSc levels found in A10 cKO mice strongly argue in favor of PrP-dependent effects of ADAM10 depletion . Analysis of A10 cKO mice and littermate controls at 95 dpi as well as tga20 mice at 65 dpi ( which represents terminal disease for these mice ) revealed a significant difference in PrPSc formation between A10 cKO and the other two experimental groups . This cannot be explained by the increased presence of PrPC , since PrPC overexpressing tga20 mice showed low levels of PrPSc even at the terminal stage of the disease . How can this be explained ? The molecular basis underlying neurodegeneration in prion diseases is under debate and different mechanisms of neurotoxicity have been proposed . In the receptor model , PrPC binds oligomeric β-sheet rich protein species , such as PrPSc , and mediates neurotoxic signaling ( Mouillet-Richard et al . , 2000; Lauren et al . , 2009; Resenberger et al . , 2011; Larson et al . , 2012; Um et al . , 2012 ) . In the pore formation model , PrPSc oligomers , either directly or upon binding to PrPC , form a membrane channel leading to exaggerated Ca2+ influx and calpain activation ( Falsig et al . , 2012; Sonati et al . , 2013 ) . In our study we found evidence for the link between excessive generation of PrPSc and increased calpain levels in A10 cKO mice , whereas basal ( in littermate controls ) and low ( in tga20 mice ) PrPSc levels were not associated with calpain upregulation . Tga20 mice are hypersensitive to prion disease yet do not show significant levels of PrPSc even though PrPC is strongly overexpressed . In a previous study we have shown and re-evaluated here that ADAM10-mediated shedding of PrPC in neurons of tga20 mice is increased approximately threefold ( Altmeppen et al . , 2011 ) . Of note , soluble GPI-anchorless forms of PrPC , which may be regarded as correlates of shed PrPC , and experimentally-induced release of PrPC in vitro impair PrPSc formation ( Marella et al . , 2002; Meier et al . , 2003; Kim et al . , 2009; Yuan et al . , 2013 ) . Thus , high levels of shed PrPC in tga20 mice may impair PrPSc formation whereas lack of shed PrPC in A10 cKO mice may favor PrPSc formation ( Figure 9 ) . In line with this concept , PrPSc production was decreased in mice moderately overexpressing ADAM10 ( Endres et al . , 2009 ) . In both tga20 and A10 cKO mice , increased levels of PrPC at the plasma membrane should facilitate PrPC-mediated neurotoxicity and thereby determine incubation times . Interestingly , these data provide a potential explanation for the tga20 paradox of low PrPSc formation despite high PrPC levels and short incubation times ( Fischer et al . , 1996 ) . A10 cKO mice and controls showed identical prion titers . On the other hand levels of PrPSc were elevated and no atypical protease-sensitive forms of PrPSc were found . It is becoming increasingly evident that titers of prion infectivity , PrPSc , and the presence of potentially neurotoxic PrP conformers are not congruent ( Lasmezas et al . , 1997; Barron et al . , 2007; Piccardo et al . , 2007; Krasemann et al . , 2013 ) . In fact , the exact composition of infectious ‘prions’ ( i . e . , the entity determined by bioassay ) is not fully understood and it may be that ADAM10-mediated shedding affects formation of PrPSc while it does not influence production of ‘prions’ . We did not observe induction of proposed neurotoxic signaling pathways ( Mouillet-Richard et al . , 2000; Schneider et al . , 2003; Larson et al . , 2012; Um et al . , 2012; Pradines et al . , 2013 ) . It may be that these pathways occur spatially and temporally restricted and future studies will have to address this question using refined methods of analysis . Alternatively , other yet to be discovered pathways may be involved . In A10 cKO mice , the combination of elevated PrPC membrane levels and increased PrPSc amounts may favor increased pore formation and Ca2+ influx as indicated by elevated calpain levels . However , since we did not detect unequivocal evidence for calpain activation , further studies are required to substantiate this model . In view of the fact that prion-infected mice expressing anchorless PrPC show PrPSc formation and deposition in ectopic sites ( Lee et al . , 2011 ) and prion propagation in the CNS ( Chesebro et al . , 2005 ) , we investigated whether shedding contributes to the spread of prion pathology . In fact , A10 cKO mice show prion-related pathology and PrPSc deposition at the site of ( striatum ) and adjacent to prion administration ( hippocampus , frontal cortex , thalamus ) , whereas brain regions distant from the inoculation site ( cerebellum , brain stem ) were unaffected , even at terminal stages of disease . In contrast , in tga20 mice , dissemination of prion-related pathology occurred efficiently at a much earlier time point . Non-PrP-dependent effects resulting from ADAM10 deficiency , such as altered microglial activity or reduced processing of proinflammatory molecules , may have contributed to impaired spreading ( Weber et al . , 2013; Prox et al . , 2013 ) . However , the influence of the immune system , including microglial activation on intracerebral spread of prions , is limited , therefore a correlation between PrP shedding and the efficiency of prion spread is possible ( Glatzel et al . , 2004; Grizenkova et al . , 2014 ) . In line with our findings , it has been shown that , upon prion infection of mice , heterozygous expression of normal and anchorless PrP results in earlier death compared with wild-type mice ( Chesebro et al . , 2005 ) . Thus , shedding of either PrPC ( followed by cell-free conversion [Kocisko et al . , 1994] ) or PrPSc , both of which have been shown to occur ( Taylor et al . , 2009; Altmeppen et al . , 2011 ) , should be considered as an alternative mechanism of prion spread ( Figure 9 ) aside from direct transsynaptic cell-to-cell transfer ( Glatzel et al . , 2001; Shearin and Bessen , 2014 ) , exosomal ( Alais et al . , 2008 ) and viral transfer ( Leblanc et al . , 2006 ) , or tunneling nanotubes ( Gousset et al . , 2009 ) . It remains to be investigated whether ADAM10-mediated shedding is also involved in the release of bona fide prions into body fluids such as cerebrospinal fluid , blood , or nasal secretions , which may potentially increase the risk of transmission ( Tagliavini et al . , 1992; Perini et al . , 1996; Parizek et al . , 2001; Bessen et al . , 2010 ) . In conclusion , our data indicate that proteolytic processing , as described here for the shedding of the prion protein , represents a master switch in the pathophysiology of prion diseases by drastically affecting PrPSc formation and incubation times . Interestingly , protective effects of ADAM10 might be hijacked in prion diseases since a reduction of the protease in prion disease has recently been reported in vitro and in vivo ( Chen et al . , 2014 ) . By contrast , shedding also seems to affect the spread of prion pathology within the CNS . Understanding this dual role of ADAM10 in prion disease brings together current concepts of prion biology and might reveal a mechanistic insight into important pathophysiological processes . Apart from prion disease , in other neurodegenerative proteinopathies , where PrPC-PrPSc conversion does not play a role , ADAM10-mediated shedding of PrPC might solely be protective due to reduction of the receptor as well as the production of a soluble blocker of toxic oligomers . In view of a potential role of PrPC in AD , further research on proteolysis as a regulatory event in prion biology will likely impact on more common neurodegenerative conditions .
In this study we took advantage of the recently generated Camk2a-Cre Adam10 conditional knockout ( A10 cKO ) mouse model with a depletion of Adam10 in neurons of the forebrain . Generation as well as a detailed description of these mice can be found elsewhere ( Prox et al . , 2013 ) . Assessment of floxed alleles ( Jorissen et al . , 2010 ) and the Camk2a-Cre status ( Casanova et al . , 2001; Prox et al . , 2013 ) in transgenic mice was performed by PCR . Our breeding strategy of crossing Adam10F/F with Camk2a-Cre:Adam10F/+ yielded approximately 15% viable Cre-positive Adam10F/F offsprings ( A10 cKO ) ( Prox et al . , 2013 ) . Cre-negative Adam10F/F or Adam10F/+ littermates were used as wild-type controls . In addition , we used prion protein overexpressing Tg ( Prnp ) a20 mice ( herein referred to as tga20 ) ( Fischer et al . , 1996 ) as controls for our biochemical and immunohistochemical analysis and as reporter mice in our bioassays for determination of prion infectivity titers as described below . Finally , brain sections of prion protein knockout ( Prnp0/0 ) mice ( Büeler et al . , 1992 ) were used as an internal negative control for our immunohistochemical staining procedure . Our study was carried out in accordance with the principles of laboratory animal care ( NIH publication No . 86-23 , revised 1985 ) as well as the recommendations in the Guide for the Care and Use of Laboratory Animals of the German Animal Welfare Act on protection of animals . The protocol was approved by the Committee on Ethics of the Freie und Hansestadt Hamburg—Amt für Gesundheit und Verbraucherschutz ( permit number 48/09 , 81/07 and 84/13 ) . Inoculations of mice were performed under deep ketamine and xylazine hydrochloride anesthesia . All efforts were made to minimize suffering of the animals including careful observation and special treatment ( i . e . , administration of wet food and incubation on a warming plate ) of mice immediately after inoculations until recovery . In brief , 6−9-week-old A10 cKO mice ( n = 12 ) , littermate controls ( n = 17 ) , and tga20 mice ( n = 11 ) were inoculated with 30 µl of 1% homogenate of Rocky Mountain Laboratory ( RML ) prions ( RML 5 . 0 inoculum , corresponding to 3 × 105 LD50 ) into the forebrain . Mice were checked on a weekly basis and observation was intensified to a two-day schedule following the appearance of first clinical signs . To assess disease characteristics in preclinical mice , A10 cKO , wild-type littermates , and tga20 mice ( n = 3 for each genotype ) were sacrificed at 35 dpi . Moreover , two A10 cKO mice and two wild-type control mice were taken at 60 days for determination of preclinical prion titers . For matched comparison with terminal A10 cKO mice , preclinical littermate controls ( n = 4 ) were sacrificed at day 95 after inoculation . All other mice ( A10 cKO: n = 7; controls: n = 8; tga20: n = 8 ) were sacrificed and analyzed when they reached terminal prion disease ( Table 1 ) . Additionally , we performed mock inoculations with 30 µl of 1% brain homogenate from uninfected CD-1 mice into age-matched A10 cKO ( n = 5 ) and littermate controls ( n = 11 ) . These animals were sacrificed at 200 dpi lacking any signs of prion disease ( see Table 1 ) . For the assessment of titers of prion infectivity in our preclinical and terminally diseased A10 cKO and control mice , we performed bioassays in tga20 reporter mice . To this end , 20 µl of a 1% homogenate of frontal brain derived from an animal to be investigated were intracerebrally inoculated into four or six tga20 mice . Prion titers ( y; shown as logLD50 ) in the original sample were then calculated according to the equation y = 11 . 45 − 0 . 088 × x , with x being the time to terminal disease of tga20 reporter mice ( in dpi ) . Generation of primary neurons of E14 embryos of NestinA10 KO , wild-type C57BL/6 , Prnp0/0 and tga20 mice as well as immunoprecipitation of shed PrPC from conditioned media was described ( Altmeppen et al . , 2011 ) . MEFs of A10 KO and control mice were initially described in ( Hartmann et al . , 2002 ) and maintained under standard cell culture conditions . Conditioned media were collected after 18–24 hr and prepared for immunoprecipitation of PrPC ( Altmeppen et al . , 2011 ) . Alternatively , media supernatants were concentrated 10× using ultrafiltration spin columns ( Vivaspin 500 , Sartorius Stedim Biotech , Goettingen , Germany ) . Samples of frontal brain were processed as 10% ( wt/vol ) homogenates in RIPA buffer ( 50 mM Tris–HCl pH 8 , 150 mM NaCl , 1% NP40 , 0 . 5% Na-Deoxycholat , 0 . 1% SDS ) supplemented with Complete Mini protease inhibitor cocktail ( PI; Roche Diagnostics , Mannheim , Germany ) . Samples were smashed 30× on ice using a Dounce homogenizer and subsequently incubated on ice for 15 min before resuspending by 15× pipetting up and down . Homogenates were centrifuged at 12 , 000×g for 6 min at 4°C and total protein content in supernatants was determined by colorimetric analysis ( QuickStart Bradford 1× Dye , Biorad , Hercules , CA ) following the manufacturer's instructions . Samples were then normalized to yield equal protein amounts and boiled in 4× loading buffer ( 250 mM Tris–HCl , 8% SDS , 40% glycerol , 20% β-mercaptoethanol , 0 . 008% Bromophenol Blue , pH 6 . 8 ) for 6 min at 96°C . Gel electrophoresis was performed using 20–50 µg of total protein per lane and 8% , 10% , or 12% SDS-PAGE gels . Proteins were subsequently wet-blotted onto nitrocellulose membranes ( Biorad ) and membranes were blocked for 1 hr using 5% non-fat dry milk in TBS-T . Immunoblot analysis was performed using the following primary antibodies: rabbit anti-ADAM10 ( 1:1000; polyclonal antiserum , B42 . 1 ) , mouse monoclonal antibodies against PrPC ( POM1 [for most of the experiments] or POM2 [for detection of shed PrP ( Figure 1E ) ] ( Polymenidou et al . , 2008 ) , 1:2500; A Aguzzi , Zurich , Switzerland ) , mouse monoclonal 1E4 ( 1:500; Sanquin , Amsterdam , The Netherlands ) , rabbit against p35/p25 ( 1:1000; mAB #2680; Cell Signaling ) , mouse Anti-Spectrin ( 1:500; MAB1622; Millipore , Bedford , MA ) , rabbit polyclonal antibody against calpain ( H-240 , 1:200; Santa Cruz Biotechnology , Santa Cruz , CA ) , and mouse monoclonal antibody against β-actin ( 1:2000; Millipore ) or rabbit anti-Actin ( 1:2000; A5060; Sigma ) . Incubation with primary antibodies ( diluted in 5% non-fat dry milk ) was carried out overnight at 4°C on a shaking platform . Membranes were then washed 3× for 5 min with TBS-T , incubated for 45 min at room temperature with HRP-conjugated anti-mouse or anti-rabbit secondary antibodies ( Promega , Madison , WI ) diluted in 5% non-fat dry milk in TBS-T , and washed 6× for 5 min with TBS-T . The signal was detected after incubation of membranes with Pierce ECL or SuperSignal West Femto substrate ( Thermo Scientific , Waltham , MA ) using a CD camera imaging system ( Biorad ) . If necessary , quantification of signal strengths was performed using QuantityOne software ( Biorad ) to measure the ratio of protein bands ( glycotyping of PrPC ) or to determine relative protein expression against β-actin . For assessment of PrPSc levels and for use in inoculation experiments ( Bioassay ) , 20% ( wt/vol ) homogenates of frontal brain were prepared in sterile phosphate buffer saline ( PBS ) without protease inhibitors . Again , samples were smashed 30× on ice using a Dounce homogenizer and subsequently spun down at 1000 rpm for 2 min . The resulting supernatant was either further diluted in PBS to yield a 1% homogenate used for inoculation of tga20 reporter mice ( see above ) or 2–5 µl were digested with 20 µg/ml PK ( Roche ) in a total volume of 22 µl RIPA buffer for 1 hr at 37°C to assess the PrPSc content . For detection of atypical prion fragments , selected samples were digested with 1 or 10 µg/ml or without PK for 1 hr at 37°C . ‘Cold PK’ treatment was performed with 200 µg/ml PK for 1 hr at 4°C . Digestion was stopped by adding 10× loading buffer and boiling for 6 min at 96°C . Subsequent SDS-PAGE and Western blot analysis was performed as described above with the exception that commercial Mini-PROTEAN TGX Any kD gels ( Biorad ) were used . Morphological analysis was performed as described previously ( Altmeppen et al . , 2011; Prox et al . , 2013 ) . In brief , brains were dissected and fixed by immersion in 4% buffered formalin overnight . In the case of prion- or mock-inoculated animals , samples were inactivated for 1 hr in 98–100% formic acid before export from the facility . These samples were then incubated overnight with an excess of 4% buffered formalin . Samples were dehydrated and embedded in low melting point paraffin following standard laboratory procedures . Sections of 4 μm were prepared and either stained with hematoxylin and eosin ( HE ) following standard laboratory procedures or submitted to immunostaining following standard immunohistochemistry procedures using the Ventana Benchmark XT machine ( Ventana , Tucson , AZ ) . Briefly , deparaffinated sections were boiled for 30–60 min in 10 mM citrate buffer ( pH 6 . 0 ) for antigen retrieval . All solutions were from Ventana . Sections were incubated with primary antibody in 5% goat serum ( Dianova , Hamburg , Germany ) , 45% Tris buffered saline ( TBS ) pH 7 . 6 , 0 . 1% Triton X-100 in antibody diluent solution ( Zytomed , Berlin , Germany ) for 1 hr . The following primary antibodies were used: monoclonal POM1 ( Polymenidou et al . , 2008 ) ( 1:100; Prof Dr Aguzzi , Zürich , Switzerland ) , SAF84 antibody against PrPC and PrPSc ( 1:100; Cayman Chemicals , Ann Arbor , MI ) , anti-GFAP ( 1:200; M0761 , DAKO , Hamburg , Germany ) , anti-Iba-1 ( 1:500; 019-19741 , Wako Chemicals , Neuss , Germany ) , anti-NeuN ( 1:50; MAB377 , Millipore ) . Detection was with anti-rabbit or anti-goat histofine Simple Stain MAX PO Universal immunoperoxidase polymer or Mouse Stain Kit ( for detection of mouse antibodies on mouse sections ) . All secondary antibody polymers were purchased from Nichirei Biosciences ( Tokyo , Japan ) . Detection of antibodies was performed with Ultra View Universal DAB Detection Kit or Ultra View Universal Alkaline Phosphatase Red Detection Kit from Ventana according to standard settings of the machine . Experimental groups were stained in one run , thereby providing identical conditions . The counterstaining was also performed by the machine according to common protocols . Additional negative controls included sections treated with secondary antibody only . For PrPSc detection , mounted tissue sections ( 4 µm ) were pretreated with 98% formic acid for 5 min . Further processing was performed on an automated staining machine ( Benchmark XT , Ventana ) . Briefly , sections were pretreated with 1 . 1 mM sodium citrate buffer ( 2 . 1 mM Tris–HCl , 1 . 3 mM EDTA , pH 7 . 8 ) at 95°C for 30 min , digested with low concentration of PK for 16 min , incubated in Superblock for 10 min and then incubated with the PrP-specific antibody SAF 84 ( see above ) , followed by secondary antibody treatment and detection . Small pieces of forebrain ( 2–3 mm3 ) of terminally prion-diseased A10 cKO mice and littermate controls were fixed in glutaraldehyde , postfixed in osmium tetroxide for 1−2 hr , dehydrated through a series of graded ethanols and propylene oxide , and embedded in Epon . Semi-thin sections were stained with toluidine blue . Ultrathin sections were stained with lead citrate and uranyl acetate , and specimens were examined using a JEM 100 C transmission electron microscope ( JEOL , Tokyo , Japan ) . For the assessment of TVS , sample grids were divided into grid squares of equal size and the number of grid squares presenting TVS clusters was determined as published previously ( Falsig et al . , 2012 ) . Adherently growing NSC cultures were established from the ganglionic eminence of 14-day-old wild-type and NestinA10 KO mice as described elsewhere ( Jorissen et al . , 2010; Altmeppen et al . , 2011 ) . In brief , we first established neurosphere cultures according to standard protocols ( Ader et al . , 2001; Pressmar et al . , 2001 ) . After two passages , neurospheres were enzymatically dissociated using Accutase ( PAA Laboratories , Pasching , Austria ) and cells were further cultivated in tissue culture flasks coated with 0 . 1% Matrigel ( BD Biosciences , Franklin Lakes , NJ ) in DMEM/F12 ( Life Technologies , Carlsbad , CA ) supplemented with 2 mM glutamine , 5 mM HEPES , 3 mM sodium bicarbonate , 0 . 3% glucose ( all from Sigma–Aldrich , St Louis , MO; in the following termed NS medium ) and 10 ng/ml epidermal growth factor ( EGF ) , 10 ng/ml fibroblast growth factor-2 ( FGF-2 ) ( both from Tebu-Bio , Le-Perray-en-Yvelines , France ) , 1% N2 and 1% B27 ( both from Life Technologies ) to establish adherently growing NSC cultures ( Conti et al . , 2005; Jung et al . , 2013 ) . To express ADAM10 in NestinA10 cKO NSCs , the mouse Adam10 cDNA was cloned into pcDNA3 . 1/Zeo ( − ) ( Life Technologies ) , and the linearized plasmid was used to transfect NestinA10 KO NSC using the Nucleofector technology ( Lonza , Basel , Switzerland ) as described previously ( Richard et al . , 2005; Altmeppen et al . , 2011 ) . As control , NestinA10 KO NSCs were nucleofected with pcDNA3 . 1/Zeo ( − ) lacking the Adam10 cDNA . To select for positive cells , cells were further cultivated in NS medium supplemented with 10 ng/ml EGF , 10 ng/ml FGF-2 , 1% N2 , 1% B27 , and 200 µg/ml zeocin . To induce neuronal differentiation of wild-type , NestinA10 KO and A10-nucleofected NestinA10 KO NSCs , cells were plated onto coverslips coated with 1% Matrigel and maintained for 4 days in NS medium supplemented with 5 ng/ml FGF-2 , 1% N2 , and 2% B27 . Subsequently , cells were cultivated for another 4 days in a 1:1 mixture of NS medium and Neurobasal medium ( Life Technologies ) containing 0 . 25% N2 and 2% B27 . After neuronal differentiation of NSCs in six-well plates , cells were washed 2× with cold PBS and incubated for 30 min with 0 . 5 mg EZ-Link Sulfo-NHS-SS-Biotin ( Thermo Scientific ) in PBS at 4°C under gentle agitation on a rocking platform . Cells were then washed 3× for 5 min at 4°C with 0 . 1% BSA in PBS and lysed with 500 µl RIPA buffer as described above . After centrifugation , supernatants were diluted 1:1 with Triton dilution buffer ( 100 mM TEA , 100 mM NaCl , 5 mM EDTA , 0 . 02% NaN3 , 2 . 5% Triton X-100 , pH 8 . 6 , +PI ) and incubated for 1 hr with 200 µl pre-washed NeutrAvidin agarose beads ( Thermo Scientific ) at room temperature on a rotary wheel . Centrifugation was performed at 1000×g for 1 min and supernatant was taken as ‘lysate control’ . Beads were washed 3× with wash buffer ( 20 mM TEA , 150 mM NaCl , 5 mM EDTA , 1% Triton X-100 , 0 . 2% SDS , 0 . 02% NaN3 , pH 8 . 6 , +PI ) and spun down . Two more washing steps were performed with final wash buffer ( 20 mM TEA , 150 mM NaCl , 5 mM EDTA , pH 8 . 6 , +PI ) before adding 50 µl of 4× loading buffer including DTT and boiling for 6 min at 96°C to release biotinylated proteins from the beads . Supernatants were taken and loaded onto gels for SDS-PAGE as described above . ‘Lysate controls’ and biotinylated samples were detected via immunblotting using POM1 antibody . As reference and specificity control for surface biotinylated samples , membrane marker Flotillin-1 ( murine purified anti-Flot-1; 1:1 . 000; BD Bioscience ) was used . For surface staining of PrPC in neuronally differentiated NSCs , live cells were incubated with primary antibody against PrPC ( POM1 ) for 1 hr at 4°C . Cells were then fixed in 4% paraformaldehyde in PBS ( pH 7 . 4 ) , blocked in PBS containing 0 . 1% bovine serum albumin and 0 . 3% Triton X-100 ( both from Sigma ) , and incubated with polyclonal rabbit anti-β-tubulin III antibodies ( Sigma ) overnight at room temperature to identify nerve cells . Primary anti-PrPC and anti-β-tubulin III antibodies were detected with anti-mouse Cy2- and anti-rabbit Cy3-conjugated secondary antibodies ( Jackson Immunoresearch Laboratories , West Grove , PA ) , respectively , and cell nuclei were stained with 4′ , 6-diamidino-2-phenylindole ( DAPI; Sigma ) . For MEF cells , surface staining of PrPC and ADAM10 was achieved by incubating live cells for 1 hr at 4°C with primary antibodies POM1 and monoclonal rat anti-mouse ADAM10 ectodomain antibody ( 1:100 , R&D Systems , Minneapolis , MN ) , respectively . These antibodies were also applied after permeabilization of cells with 0 . 2% Triton X-100 in PBS . Goat anti-mouse IgG Alexa Fluor 488 and goat anti-rat IgG Alexa Fluor 568 ( both from Life Technologies ) were used as secondary antibodies prior to mounting the coverslips with DAPI Fluoromount-G ( SouthernBiotech , Birmingham , AL ) onto object slides . Cortical regions from A10 cKO mice ( n = 5 ) and littermate controls ( n = 5 ) were prepared and total RNA was isolated using the NucleoSpin RNAII kit ( Macherey Nagel , Dueren , Germany ) according to the manufacturer's instructions . 2 µg of DNAse-treated RNA was used for cDNA synthesis using the RevertAid cDNA Synthesis Kit ( Thermo Fisher Scientific ) . Gene expression analysis for mouse PrPC and GAPDH were performed using the mouse Universal ProbeLibrary System ( Roche Applied Science ) for qRT-PCR experiments ( PrPC: 5′-GCCGACATCAGTCCACATAG , 5′-GGAGAGCCAAGCAGACTATCA , Probe: #71 ) on a LightCycler480 ( Roche Diagnostics ) . PrPC gene expression levels were depicted as percentage of GAPDH expression using the ΔCT method for calculation . PCR efficiency of each assay was determined by serial dilutions of standards and these values were used for calculation . Statistical comparison of incubation times , Western blot quantifications , and qRT-PCR results between experimental groups was performed using Student's t-test for two-group comparisons with consideration of statistical significance at p values <0 . 05 ( * ) , <0 . 01 ( ** ) , and <0 . 001 ( *** ) . | Prion proteins are anchored to the surface of brain cells called neurons . Normally , prion proteins are folded into a specific three-dimensional shape that enables them to carry out their normal roles in the brain . However , they can be misfolded into a different shape known as PrPSc , which can cause Creutzfeldt-Jakob disease and other serious conditions that affect brain function and ultimately lead to death . The PrPSc proteins can force normal prion proteins to change into the PrPSc form , so that over time this form accumulates in the brain . They are essential components of infectious particles termed ‘prions’ and this is why prion diseases are infectious: if prions from one individual enter the brain of another individual they can cause disease in the recipient . The UK outbreak of variant Creutzfeldt-Jakob disease in humans in the 1990s is thought to be due to the consumption of meat from cattle with a prion disease known as mad cow disease . An enzyme called ADAM10 can cut normal prion proteins from the surface of neurons . However , it is not clear whether ADAM10 can also target the PrPSc proteins and what impact this may have on the development of prion diseases . Here , Altmeppen et al . studied mutant mice that were missing ADAM10 in neurons in the front portion of their brain . These mice had a higher number of normal prion proteins on the surface of their neurons than normal mice did . When mice missing ADAM10 were infected with prions , more PrPSc accumulated in their brain and disease symptoms developed sooner than when normal mice were infected . This supports the view that mice with higher numbers of prion proteins are more vulnerable to prion disease . However , disease symptoms did not spread as quickly to other parts of the brain in the mice missing ADAM10 . This suggests that by releasing prion proteins from the surface of neurons , ADAM10 helps PrPSc proteins to spread around the brain . Recently , it has been suggested that prion proteins may also play a role in Alzheimer's disease and other neurodegenerative conditions . Therefore , Altmeppen et al . 's findings may help to develop new therapies for other forms of dementia . The next challenge is to understand the precise details of how ADAM10 works . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience",
"microbiology",
"and",
"infectious",
"disease"
] | 2015 | The sheddase ADAM10 is a potent modulator of prion disease |
Synaptotagmin-1 ( Syt1 ) acts as a Ca2+ sensor for neurotransmitter release through its C2 domains . It has been proposed that Syt1 promotes SNARE-dependent fusion mainly through its C2B domain , but the underlying mechanism is poorly understood . In this study , we show that the C2B domain interacts simultaneously with acidic membranes and SNARE complexes via the top Ca2+-binding loops , the side polybasic patch , and the bottom face in response to Ca2+ . Disruption of the simultaneous interactions completely abrogates the triggering activity of the C2B domain in liposome fusion . We hypothesize that the simultaneous interactions endow the C2B domain with an ability to deform local membranes , and this membrane-deformation activity might underlie the functional significance of the Syt1 C2B domain in vivo .
Ca2+-triggered neurotransmitter release by synaptic exocytosis is an exquisitely regulated process for interneuronal communication . The core machinery governing the process includes the SNAREs synaptobrevin , syntaxin-1 and SNAP-25 , which form tight SNARE complexes to bridge synaptic vesicles to the plasma membrane and catalyze membrane fusion ( Jahn and Scheller , 2006; Sutton et al . , 1998; Weber et al . , 1998 ) . Syt1 , the Ca2+ sensor for the fast component of Ca2+-triggered release ( Chapman , 2008; Fernandez-Chacon et al . , 2001; Geppert et al . , 1994 ) , confers Ca2+ sensitivity to SNARE-dependent synaptic vesicle fusion . The triggering function of Syt1 depends on its interplay with membranes , SNAREs , complexins and other key proteins of the release machinery ( Rizo and Xu , 2015 ) . Syt1 consists of two C2 domains , known as C2A and C2B . The C2A and C2B domains adopt similar structures and bind three and two Ca2+ ions , respectively , through their Ca2+-binding loops located at the top of the structures ( Fernandez et al . , 2001; Shao et al . , 1998; Sutton et al . , 1995 ) . These loops mediate penetration of Syt1 C2 domains to acidic membranes containing phosphatidylserine ( PS ) in response to Ca2+ , and this activity is required for the triggering function of Syt1 ( Chapman and Davis , 1998; Fernandez-Chacon et al . , 2001; Rhee et al . , 2005 ) . In addition , Syt1 readily binds to phosphatidylinositol-4 , 5-bisphosphate ( PI ( 4 , 5 ) P2 ) via its C2B domain in a Ca2+-independent manner , which helps increase the apparent Ca2+ affinity of Syt1 and thereby enhances release probability ( Bai et al . , 2004; Li et al . , 2006; Radhakrishnan et al . , 2009; van den Bogaart et al . , 2011a ) . Furthermore , Syt1 binds to SNAP-25 and syntaxin-1 , and to SNARE complexes mainly through its C2B domain , which is believed to position Syt1 on the pre-fusion SNARE complexes to trigger release in response to Ca2+ ( Brewer et al . , 2015; Zhou et al . , 2015; de Wit et al . , 2009; Mohrmann et al . , 2013 ) . Although the biochemical properties of Syt1 have been studied in detail , the functional importance of individual properties has remained unclear . It has been found that disrupting Ca2+ binding to C2B impairs release much more strongly than disruption of C2A Ca2+ binding sites in vivo ( Mackler et al . , 2002; Nishiki and Augustine , 2004; Robinson et al . , 2002 ) . Moreover , previous work showed that isolated C2B , instead of C2A , can promote SNARE-dependent membrane fusion in response to Ca2+ in vitro ( Gaffaney et al . , 2008; Xue et al . , 2008 ) . These findings suggested that C2B plays a more preponderant role than C2A in neurotransmitter release . A number of studies have revisited this issue and suggested that the functional importance of C2B arises partly from its ability to deform membranes . For instance , C2B can induce vesicle clustering and/or membrane curvature in response to Ca2+ ( Arac et al . , 2006; Martens et al . , 2007; Hui et al . , 2009; Xue et al . , 2008 ) . However , it is unclear why C2B has such membrane-deformation activity while C2A does not , given the fact that both C2 domains of Syt1 exhibit similar Ca2+-dependent membrane-insertion properties in vitro . Another potential reason for the striking functional asymmetry of the Syt1 C2 domains was provided by recent studies showing that interactions between C2B and SNARE complexes are crucial for the function of Syt1 in neurotransmitter release ( Zhou et al . , 2015; Brewer et al . , 2015 ) . However , these studies have yielded conflicting results . For instance , a recently solved Syt1–SNARE complex crystal structure showed a relatively large binding interface between Syt1 and the SNARE complex that involves two basic residues ( Arg398 and Arg399 , referred to as the R398 R399 region , see Figure 1A ) on the bottom of C2B ( Zhou et al . , 2015 ) , whereas another dynamic Syt1–SNARE complex structure model obtained by nuclear magnetic resonance ( NMR ) indicated a SNARE binding interface that is located on the polybasic patch ( Lys326 and Lys327 , referred to as the K326 K327 region , see Figure 1A ) at the side of C2B ( Brewer et al . , 2015 ) . Moreover , both of these basic regions have been previously implicated in binding to acidic membrane lipids , such as PS and PI ( 4 , 5 ) P2 ( Xue et al . , 2008; van den Bogaart et al . , 2011a ) . Some other reports argued against the interaction between Syt1 and the SNARE complex , and suggested that the triggering function of Syt1 requires specific binding of C2B to acidic membranes rather than binding to SNARE complexes ( Honigmann et al . , 2013; Park et al . , 2015 ) . Taken together , although it seems clear that C2B can interact with either acidic SNARE complexes or acidic membranes due to the abundance of highly positive charges around its surface ( Figure 1B ) , the binding mode of C2B with SNARE complexes and membranes underlying the actual mechanism of Syt1 in release remains elusive . 10 . 7554/eLife . 14211 . 003Figure 1 . Overview of the structure features of Syt1 C2 domains and the core SNARE complex . ( A and B ) Structural diagrams ( A ) and electrostatic surface potential ( B ) of Syt1 C2A ( PDB entry 1BYN ) and C2B ( PDB entry 1TJX ) domain . Residues K326 and K327 on the side , R398 and R399 on the bottom , and Ca2+ ions on the top of C2B are shown as blue and yellow spheres , respectively . Black boxes display the basic patches that include the residues shown in A . ( C and D ) Structural diagram ( C ) and electrostatic density map ( D ) of the core SNARE complex ( PDB entry 1N7S ) . Residues D51 , E52 and E55 are displayed as red spheres . Black box displays the acidic patch , which includes the residues shown in C . Syx , syntaxin-1; SN25 , SNAP-25; Syb , synaptobrevin-2 . The electrostatic surface potential was calculated by generating local protein contact potential ( pymol software ) and scaled from -5kT/e to 5kT/e , with red and blue denoting negative and positive potential , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 14211 . 003 To address these conundrums , we systematically investigated the binding properties of the C2B domain with SNARE complexes and membranes using diverse biophysical techniques . Our results showed that , prior to the Ca2+ signal , the C2B domain interacts with PI ( 4 , 5 ) P2 and membrane-anchored SNARE complexes through the K326 K327 region and the R398 R399 region , respectively . We also found that the two interactions persist during insertion of the Ca2+-binding loops into the membrane upon Ca2+ influx . Consistent with a Syt1 triggering model proposed recently by Brunger and colleagues ( Zhou et al . , 2015 ) , our data suggest that the local membrane deformation driven by the C2B domain constitutes a key step for triggering fusion .
As mentioned above , Syt1 C2B contains two highly basic regions at the side and the bottom of the structure ( i . e . , the K326 K327 and R398 R399 regions; see Figure 1A , B ) ; both regions have been implicated in interactions with the SNARE complex ( Brewer et al . , 2015; Zhou et al . , 2015 ) . Conversely , an acidic patch located in the middle portion of the SNARE complex ( e . g . , residues D51 , E52 , and E55 on SNAP-25 , likely with adjacent residues on syntaxin-1 , see Figure 1C , D ) has been suggested to mediate binding to C2B ( Brewer et al . , 2015; Mohrmann et al . , 2013; Zhou et al . , 2015 ) . Using the GST pull-down assay , we first re-examined interactions between the two basic regions on C2B and the acidic patch on the assembled SNARE complex in the absence of Ca2+ . Consistent with previous results ( Brewer et al . , 2015 ) , we found that both C2AB ( soluble fragment of Syt1 harboring C2A and C2B ) and C2B bound to assembled SNARE complexes , whereas C2A did not ( Figure 2A , B , note that the C2A binding might be too weak to be detected on the gel ) . The SNARE complex containing the SNAP-25 D51A/E52A/E55A mutation ( referred to as SN25 3M ) showed impaired binding ability to C2B ( Figure 2A , B ) , in agreement with previous findings ( Brewer et al . , 2015; Mohrmann et al . , 2013; Zhou et al . , 2015 ) . Note that the SNARE complex bearing SN25 3M was less resistant to SDS ( Figure 2A and Figure 2—figure supplement 1A ) , yet its assembly was normal ( Figure 2—figure supplement 1B , C ) . As expected , disruption of the C2B Ca2+-binding sites ( D363N/D365N , referred to as C2b ) caused no effect on the SNARE complex binding ( Figure 2A , B ) . However , disruption of either the K326 K327 region or the R398 R399 region ( K326A/K327A , K326E/K327E or R398Q/R399Q , referred to as C2B2KA , C2B2KE or C2B2RQ , respectively ) impaired binding of C2B to the SNARE complex ( Figure 2A , B ) , indicating that both basic regions of C2B contribute to the SNARE complex binding . This data reproduced previously contradictory findings ( Brewer et al . , 2015; Zhou et al . , 2015 ) , suggesting heterogeneous interactions between C2B and the SNARE complex in solution ( Park et al . , 2015 ) . 10 . 7554/eLife . 14211 . 004Figure 2 . Different Ca2+-independent interactions of Syt1 with membranes and SNARE complexes . ( A and B ) Binding of Syt1 soluble fragments and their mutants to the core SNARE complex measured by GST pull-down assay ( A ) and quantification of the C2B binding ( B ) . Asterisks in A show the bands of bound protein . 3M , GST-tagged SNARE complex bearing the SNAP-25 D51A/E52A/E55A mutation; H3 , the SNARE motif of syntaxin-1; neg . ctrl . , negative control , which represents C2B bound to GST-H3 . Representative gel from one of three independent experiments is shown . Data are processed by Image J ( NIH ) and presented as the mean ± SD ( n = 3 ) , technical replicates . ( C ) Schematic diagram of the liposome co-flotation assay . After centrifuging , liposomes ( orange ) and bound proteins ( blue ) were co-floated on the top of the density gradients , remaining unbounded proteins left in the bottom of the gradients . ( D and E ) Co-flotation of C2AB , C2B and their mutants with liposomes bearing 1% PI ( 4 , 5 ) P2 in the absence of Ca2+ ( D ) and quantification of the binding ( E ) . WT/2RQ/2KA/2KE , Syt1 C2AB or C2B , and the mutants bearing R398Q/R399Q , K326A/K327A or K326E/K327E mutations , respectively; S , supernatant; P , pellet formed by centrifuging . Representative gel from one of three independent experiments is shown . Data are processed by Image J ( NIH ) . ( F and G ) FRET between NBD labeled Syt1 C2B-H315C and rhodamine labeled liposomes with or without 1% PI ( 4 , 5 ) P2 ( F ) and quantification of the emission fluorescence of rhodamine at 587 nm ( G ) . Liposome compositions are presented below the diagram; all reactions were performed in the absence of Ca2+ . ( H ) C2B-induced liposome clustering measured in the presence of SNARE complexes . The change in particle size as a function of the C2B concentration was measured by dynamic light scattering ( DLS ) . All data plots are presented as the mean ± SD ( n = 3 ) , technical replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 14211 . 00410 . 7554/eLife . 14211 . 005Figure 2—figure supplement 1 . The SNAP-25 3M mutation displayed less resistance to SDS and no influence on SNARE complex assembly . ( A ) Assembled GST-tagged SNARE complexes were analyzed by SDS-PAGE before pull-down assays . ( B ) The assembly of pre-assembled syntaxin-1–SNAP-25 complex and synaptobrevin monitored by fluorescence anisotropy . ( C ) The assembly of syntaxin-1 , SNAP-25 and synaptobrevin monitored by fluorescence anisotropy . Fluorescent probe ( BODIPY FL ) was labeled on synaptobrevin S61C . The small difference between WT and 3M in B is insignificant given the noise in the traces . WT , wild type; 3M , SN25 D51A/E52A/E55A mutations; neg . ctrl . , negative control , which represents the addition of excess unlabeled synaptobrevin . DOI: http://dx . doi . org/10 . 7554/eLife . 14211 . 00510 . 7554/eLife . 14211 . 006Figure 2—figure supplement 2 . PS-containing liposome clustering induced by C2B and its mutants in the presence of Ca2+ . DOI: http://dx . doi . org/10 . 7554/eLife . 14211 . 006 This heterogeneous binding might be due to the absence of membrane lipids . PI ( 4 , 5 ) P2 selectively localizes to the plasma membrane , binds to Syt1 and plays crucial functions in neurotransmitter release ( van den Bogaart et al . , 2011b ) . Using a liposome co-flotation assay ( Figure 2C ) , we found that , in the absence of Ca2+ , both C2AB and C2B ( wild type , WT ) bound efficiently to liposomes containing 1% PI ( 4 , 5 ) P2 ( Figure 2D , E ) . This PI ( 4 , 5 ) P2-binding ability of Syt1 relies on the K326 K327 region but not the R398 R399 region , as C2AB2KE/C2AB2KA or C2B2KE/C2B2KA completely abolished binding to PI ( 4 , 5 ) P2-containing liposomes whereas C2AB2RQ or C2B2RQ did not ( Figure 2D , E ) . To confirm this binding preference , we measured the fluorescence resonance energy transfer ( FRET ) between 7-nitrobenz-2-oxa-1 , 3-diazole ( NBD ) labeled C2B and rhodamine-labeled liposomes ( Hui et al . , 2011 ) containing 1% PI ( 4 , 5 ) P2 in the absence of Ca2+ . When NBD was placed close to the K326 K327 region ( H315C-NBD ) , we detected a robust energy transfer ( Figure 2F , G ) . Comparably , Ca2+-independent FRET signals were considerably weaker when PI ( 4 , 5 ) P2 was removed from liposomes ( Figure 2F , G ) . The addition of excess assembled SNARE complexes in the reaction caused no obvious effect on the FRET between C2B and PI ( 4 , 5 ) P2-containing liposomes ( Figure 2F , G ) . These results , together with a recent finding that Syt1 binds to PI ( 4 , 5 ) P2-containing liposomes in physiological ionic conditions that contain ATP and Mg2+ ( Park et al . , 2015 ) , suggest that the K326 K327 region binds specifically to PI ( 4 , 5 ) P2-containing membranes in the absence of Ca2+ . Given that the side K326 K327 region of C2B binds specifically to PI ( 4 , 5 ) P2 on membranes , we explored whether the bottom R398 R399 region tends to bind the SNARE complex . Previous studies suggested that the R398 R399 region binds acidic membranes ( i . e . , PS ) and participates in liposome clustering ( Arac et al . , 2006; Xue et al . , 2008 ) . In our study using PS-containing liposomes , we indeed found that C2B clustered liposomes in response to Ca2+ ( Figure 2—figure supplement 2 ) . Both C2B2RQ and C2b failed to cluster liposomes , while C2B2KE was able to cluster liposomes ( Figure 2—figure supplement 2 ) , supporting the idea that the bottom R398 R399 region and the top Ca2+-binding loops associate two opposite acidic membranes in response to Ca2+ ( Xue et al . , 2008 ) . However , it is noteworthy that the PS-containing liposomes used in these experiments lacked the SNAREs ( Arac et al . , 2006; Xue et al . , 2008 ) . We therefore re-analyzed the liposome-clustering activity of C2B in the presence of soluble SNARE complexes , as a function of the concentrations of C2B . Robust clustering was observed as the C2B concentration increased in the presence of 1 mM Ca2+ ( Figure 2H ) . To our surprise , the addition of 10 μM soluble SNARE complexes in the reaction strongly impaired liposome clustering but the clustering was capable of gradually recovery as the concentration of C2B exceeded that of SNARE complexes ( when C2B was above 10 μM , Figure 2H ) . Thus , the likely explanation is that the weaker R398 R399–PS interaction can be replaced by the stronger R398 R399–SNARE complex binding . These results suggest that the R398 R399 region binds preferentially to SNARE complexes rather than acidic membranes . We further directly measured the C2B ( R398 R399 ) –SNARE complex interaction using a bimane-tryptophan quenching assay in the absence of membranes . The bimane-tryptophan quenching assay has been previously used to study the structure and movements of proteins and has shown its sensitivity in short-distance electron transfer measurements ( <10 Å ) ( Islas and Zagotta , 2006; Mansoor et al . , 2002; Taraska and Zagotta , 2010 ) . In this case , a single tryptophan mutation ( T285W ) at the bottom of C2B that is adjacent to the R398 R399 region was introduced ( note that C2B contains two native tryptophans at residues 390 and 404 that are both far from the R398 R399 region , see Figure 3A ) . In addition , SNAP-25 was labeled with bimane via a single cysteine mutation ( R59C ) close to its acidic patch ( D51 E52 E55 ) , and was then assembled into the SNARE complex ( Figure 3A ) . In contrast to a donor only condition ( no addition of C2B ) and the addtion of C2B ( WT ) , the addition of C2B T285W induced robust quenching of bimane fluorescence on the SNARE complex ( Figure 3B , C ) . Comparably , C2B T285W containing the R398Q/R399Q mutation ( referred to as T285W2RQ ) showed strongly impaired binding to the SNARE complex , whereas C2B T285W containing the K326E/K327E mutation ( referred to as T285W2KE ) did not have such an effect ( Figure 3B , C ) . Consistent with previous results ( Zhou et al . , 2015 ) , these data support the specific R398 R399–SNARE complex interaction . 10 . 7554/eLife . 14211 . 007Figure 3 . Persistence of the R398 R399–SNARE complex interaction in the presence of ATP and Mg2+ . ( A ) Schematic diagrams of bimane-labeled SNARE complex and Syt1 C2B . Tryptophan was introduced at the bottom of C2B ( T285W , orange stick ) , which is close to residues R398 and R399; two native tryptophans ( W390 and W404 ) are indicated as red sticks; residues K326 K327 and R398 R399 are shown as blue sticks; Ca2+ ions are displayed as yellow spheres . ( B and C ) Quenching of bimane fluorescence on the SNARE complex with the addition of C2B T285W and the mutants in the absence of ATP and Mg2+ ( B ) and quantification of the results ( C ) . ( D and E ) Quenching of bimane fluorescence on the SNARE complex with the addition of C2B T285W in the presence of ATP and Mg2+ ( D ) and quantification of the results ( E ) . Donor only , no addition of Syt1 C2B; T285W , C2B bearing the T285W mutation; T285W2RQ and T285W2KE , C2B T285W bearing the R398Q/R399Q or K326E/K327E mutations , respectively . Data are presented as the mean ± SD , technical replicates . n . s . , not significant ( p > 0 . 05 ) ; *p<0 . 05; ***p < 0 . 001; one-way ANOVA , n = 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 14211 . 007 We also investigated the R398 R399–SNARE complex interaction in the absence of membranes and in the presence of ATP and Mg2+ . Robust quenching of bimane fluorescence was still observed with the addition of C2B T285W ( Figure 3D , E ) , suggesting that the specific R398 R399–SNARE complex interaction persists in a physiological ionic environment . Besides , the K326E/K327E mutation or ATP , but not Mg2+ , slightly influenced the interaction between the R398 R399 region and the SNARE complex ( Figure 3B–E ) , likely owing to the electrostatic shielding . This data is inconsistent with a recent study showing that the Syt1–SNARE complex interaction is abolished at conditions containing ATP and Mg2+ ( Park et al . , 2015 ) . The reason for this contradiction is explained in the Discussion . Since C2B binds to PI ( 4 , 5 ) P2 and the SNARE complex in the absence of Ca2+ , we measured the binding affinity between C2B and the SNARE complex in the presence of membranes and in the absence of Ca2+ by using the bimane-tryptophan quenching assay . Titration of C2B T285W to liposomes containing bimane-labeled cis-SNARE complexes yielded a Kd of 1 . 53 ± 0 . 04 μM in the absence of PI ( 4 , 5 ) P2 , and a Kd of 0 . 86 ± 0 . 04 μM in the presence of PI ( 4 , 5 ) P2 ( Figure 4 ) , indicating a rather strong interaction between C2B ( R398 R399 ) and the membrane-anchored cis-SNARE complex when PI ( 4 , 5 ) P2 is present . This higher binding affinity arises likely because PI ( 4 , 5 ) P2 helps recruit C2B to the membrane through the K326 K327 region and thereby increases the encounter between the R398 R399 region of C2B and the SNARE complex on membranes . 10 . 7554/eLife . 14211 . 008Figure 4 . Binding Kd between C2B and the membrane-anchored SNARE complex . Cis-SNARE complexes were reconstituted on liposome via the syntaxin-1 transmembrane domain . PI ( 4 , 5 ) P2 increased the binding affinity between Syt1 C2B and the membrane-anchored SNARE complex in the absence of Ca2+ . Plots show the quenched efficiency of the bimane-labeled cis-SNARE complex reconstituted on liposomes ( 65% PC + 20% PE + 15% PS ) with the titration of Syt1 C2B T285W in the presence ( black ) and absence ( red ) of 1% PI ( 4 , 5 ) P2 . Diagram in the solid box is the close-up view of the data in the dashed box . Plots are presented as the mean ± SD , technical replicates . ***p<0 . 001; multiple t-test using Holm-Sidak method , n = 5 . Non-linear curve fit were achieved by the Michaelis-Menten equation where Vmax was constrained to 100 ( % Quenched efficiency ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14211 . 008 Taken together , all above binding results obtained in the absence of Ca2+ support the idea that Syt1 'docks' two membranes via its transmembrane domain anchored on vesicles and its C2B domain binding to the plasma membrane ( Honigmann et al . , 2013; van den Bogaart et al . , 2011a ) . It is conceivable that C2B pre-adsorbs to the plasma membrane prior to Ca2+ influx , through its side K326 K327 region interacting with PI ( 4 , 5 ) P2 , and its bottom R398 R399 region binding to surrounding SNAREs ( e . g . , SNAP-25 ) or assembled SNARE complexes . We next sought to explore interactions of C2B with SNARE complexes and membranes in the presence of Ca2+ using the liposome co-flotation assay . Previous studies have observed a strong binding of C2B to membranes owing to the insertion of the C2B Ca2+-binding loops into PS-containing liposomes in the presence of 1 mM Ca2+ ( Arac et al . , 2006; Hui et al . , 2011 ) . However , such strong binding would hinder the detection of interactions of C2B with PI ( 4 , 5 ) P2 or membrane-anchored SNARE complexes at the same time in our co-flotation experiments . To alleviate such an effect , we applied higher ion strength ( 250 mM KCl , instead of 150 mM KCl used in other experiments throughout the work ) and weakened the Ca2+-binding loops–Ca2+–PS interaction by lowering the concentration of Ca2+ from 1 mM to 0 . 1 mM ( Figure 5—figure supplement 1 ) . In the condition containing 250 mM KCl and 0 . 1 mM Ca2+ , we could not detect the interaction between C2B and PS-containing liposomes when the SNAREs were absent ( Figure 5A , B and Figure 5—figure supplement 1 ) . As such , we could not detect binding of C2B to membrane-anchored cis-SNARE complexes in the absence PS or Ca2+ ( Figure 5A , B ) . Efficient binding was only observed when cis-SNARE complexes and PS were both present on liposomes in the presence of 0 . 1 mM Ca2+ ( Figure 5A , B ) . Intriguingly , inclusion of 1% PI ( 4 , 5 ) P2 on liposomes that already contain PS and cis-SNARE complexes dramatically enhanced C2B binding in the presence of 0 . 1 mM Ca2+ , leaving very little C2B left in the pellet ( Figure 5C , D ) . In contrast , selective removal of PS , PI ( 4 , 5 ) P2 or cis-SNARE complexes on liposomes led to strong impairment of C2B binding ( Figure 5C , D ) , suggesting a synergy among the C2B ( Ca2+–binding loops ) –Ca2+–PS , C2B ( K326 K327 ) –PI ( 4 , 5 ) P2 and C2B ( R398 R399 ) –SNARE complex interactions in the presence of 0 . 1 mM Ca2+ . 10 . 7554/eLife . 14211 . 009Figure 5 . Synergistic interactions of C2B with membrane-anchored SNARE complexes , PI ( 4 , 5 ) P2 and PS in the presence of 0 . 1 mM Ca2+ . ( A and B ) Co-flotation of Syt1 C2B with liposomes in the absence of PI ( 4 , 5 ) P2 ( A ) and quantification of the results ( B ) . ( C and D ) Co-flotation of Syt1 C2B with liposomes in the presence of 1% PI ( 4 , 5 ) P2 ( C ) and quantification of the results ( D ) . Cis-SNARE complexes were reconstituted on liposome via the syntaxin-1 transmembrane domain . Liposomes compositions in A and C contain 65% PC , 20% PE , 15% PS with and without 1% PI ( 4 , 5 ) P2 . S , supernatant; P , pellet . Representative gel from one of three independent experiments is shown . Data are processed by Image J ( NIH ) and presented as the mean ± SD ( n = 3 ) , technical replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 14211 . 00910 . 7554/eLife . 14211 . 010Figure 5—figure supplement 1 . Binding of C2B to PS-containing liposomes in different Ca2+ concentrations . ( A ) Co-flotation of C2B with liposomes ( 65% PC + 20% PE + 15% PS ) in the presence of 250 mM KCl and different Ca2+ concentrations as indicated . ( B ) Quantification of the results in A . Representative gel from one of three independent experiments is shown . Data are processed by Image J ( NIH ) and presented as the mean ± SD ( n = 3 ) , technical replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 14211 . 010 Next , we characterized the above synergistic interactions in the presence of 1 mM Ca2+ and 150 mM KCl in a real-time manner . We first monitored insertion of the Ca2+-binding loops into membranes in response to Ca2+ . For this purpose , we introduced cysteine mutations at the C2B T285W Ca2+-binding loop 2 or 3 ( T285W N333C or T285W I367C , respectively , see Figure 6A ) and then labeled them separately with NBD and monitored their membrane-insertion abilities . Note that as an environment-sensitive probe , NBD exhibits a large increase in fluorescence intensity when it is transferred from an aqueous to a hydrophobic environment ( Crowley et al . , 1993 ) . In the presence of 1 mM Ca2+ , addition of either T285W N333C-NBD or T285W I367C-NBD to liposomes containing PS and cis-SNARE complexes caused a marked increase in intensity ( Figure 6B , C ) , confirming that the loops insert into membranes . In contrast , when additional 1% PI ( 4 , 5 ) P2 was applied , both T285W N333C-NBD and T285W I367C-NBD showed even higher intensities ( Figure 6D , E ) , implying that the membrane-insertion ability of the Ca2+-binding loops is enhanced with the assist of the C2B–PI ( 4 , 5 ) P2 interaction . Obviously , compared to T285W N333C-NBD , T285W I367C-NBD exhibited a remarkable enhancement in intensity in response to Ca2+ ( Figure 6D , E ) , suggesting that the loop 3 located on the same side as the K326 K327 region ( see Figure 6A ) on C2B is more accessible to membranes . These results imply that the K326 K327 region of C2B persistently sticks to PI ( 4 , 5 ) P2 on the plasma membrane with the insertion of the Ca2+-binding loops into membranes . This supports the idea that Ca2+-independent pre-adsorption of the K326 K327 region to PI ( 4 , 5 ) P2-harboring membranes helps 'steer' the Ca2+-triggered membrane insertion of Syt1 toward the plasma membrane ( Bai et al . , 2004 ) . 10 . 7554/eLife . 14211 . 011Figure 6 . Persistence of C2B–SNARE complex and C2B–PI ( 4 , 5 ) P2 interactions upon insertion of the Ca2+-binding loops into membranes . ( A ) Schematic diagrams of C2B and membrane-embedded cis-SNARE complex . NBD was labeled on N333C or I367C ( green sticks ) on C2B separately; bimane was labeled on SN25 R59C ( orange sphere ) ; tryptophan was introduced at the bottom of C2B ( T285W , orange stick ) , which is close to residues R398 and R399; two native tryptophans ( W390 and W404 ) are indicated as red sticks; residues K326 K327 and R398 R399 are shown as blue sticks; Ca2+ ions are displayed as yellow spheres . ( B–E ) Detecting Ca2+-triggered membrane insertion of the Ca2+-binding loops using NBD fluorescence reporters in the absence ( B and C ) and presence ( D and E ) of PI ( 4 , 5 ) P2 . Emission spectra were collected from 500 nm to 620 nm . ( F–I ) Detecting FRET between tryptophan ( T285W ) on C2B and bimane-labeled SNARE complexes reconstituted on liposomes in the absence ( F and G ) and presence ( H and I ) of PI ( 4 , 5 ) P2 . Emission spectra were collected from 400 nm to 600 nm . Donor only , no addition of Syt1 C2B . DOI: http://dx . doi . org/10 . 7554/eLife . 14211 . 011 Meanwhile , we simultaneously monitored the interaction between C2B and membrane-anchored SNARE complexes using the bimane-tryptophan quenching assay when monitoring the C2B Ca2+-binding loops inserting into membranes . The addition of T285W N333C-NBD or T285W I367C-NBD induced robust quenching of bimane fluorescence on membrane-anchored cis-SNARE complexes under all conditions with or without Ca2+ and PI ( 4 , 5 ) P2 ( Figure 6F–I ) , suggesting that the R398 R399 region binds consistently to the membrane-anchored SNARE complex before and after Ca2+ influx . Indeed , crystal structures obtained in the absence and presence of Ca2+ produced the same binding interface that involves residues R398 and R399 between Syt1 and the SNARE complex ( Zhou et al . , 2015 ) . Together , these results suggest that the C2B ( R398 R399 ) –SNARE complex interaction is Ca2+ independent and this interaction persists during insertion of the Ca2+-binding loops into the membrane . In all , based on the findings that both the R398 R399–SNARE complex interaction and the K326 K327–PI ( 4 , 5 ) P2 interaction persist during entry of the Ca2+-binding loops into membranes with PS , it is conceivable that the three simultaneous interactions of C2B induce membrane deformation in response to Ca2+ ( see the Discussion ) , as suggested by Brunger and colleagues ( Zhou et al . , 2015 ) . The above results show that the C2B ( Ca2+–binding loops ) –Ca2+–PS , C2B ( K326 K327 ) –PI ( 4 , 5 ) P2 and C2B ( R398 R399 ) –SNARE complex interactions occur simultaneously in the presence of Ca2+ . We next sought to investigate whether all three interactions are required for the triggering activity of C2B in membrane fusion . It has been widely reported that Syt1 C2AB or C2B can promote SNARE-dependent liposome fusion in response to Ca2+ in vitro ( Chapman , 2008; Xue et al . , 2008; Martens et al . , 2007 ) . However , to what extent the promotion activity of C2AB ( or C2B ) in vitro reflects the triggering activity of Syt1 in vivo must be interpreted with caution because the liposome-clustering ability of C2AB ( or C2B ) correlated strongly with its activity in promoting SNARE-dependent liposome fusion ( Arac et al . , 2006; Hui et al . , 2011; Xue et al . , 2008 ) . Accelerated liposome fusion might arise from the enhanced membrane docking and SNARE pairing caused by C2AB ( or C2B ) ( Hui et al . , 2011 ) . To detect the actual activity of C2AB ( or C2B ) in triggering fusion , in our liposome fusion system ( Figure 7A , B ) we utilized poly-D-lysine as a Ca2+-independent factor to mimic liposome docking and increase SNARE pairing ( Hui et al . , 2011 ) ; added complexin-1 to arrest liposomes in a 'ready-for-fusion' state ( Kyoung et al . , 2011; Diao et al . , 2012; Lai et al . , 2014 ) ; and finally applied 1 mM Ca2+ to trigger fusion . A lower concentration of C2AB or its mutants ( 0 . 5 µM ) was applied in the system to avoid liposome clustering . 10 . 7554/eLife . 14211 . 012Figure 7 . Ca2+-dependent simultaneous C2B–SNARE complex–membrane interactions underlie the function of Syt1 in triggering fusion . ( A and B ) Schematic diagrams of the lipid mixing ( A ) and content mixing ( B ) assay . Liposome compositions are presented below the diagram . Cpx , complexin-1 . ( C and I ) Poly-D-lysine promoted SNARE-dependent lipid mixing ( C ) and content mixing ( I ) in the absence of Ca2+ . ( D and J ) Cpx inhibited SNARE-dependent lipid mixing ( D ) and content mixing ( J ) in the absence of Ca2+ . ( E and K ) C2AB and C2B triggered fast lipid mixing ( E ) and content mixing ( K ) whereas C2A did not in response to Ca2+ . ( F and L ) Disruption of the C2B–SNARE complex–membrane interactions abolished fast lipid mixing ( F ) and content mixing ( L ) . ( G and M ) The functional analysis of the Ca2+-binding loops mutations on C2AB in triggering lipid mixing ( G ) and content mixing ( M ) . ( H and N ) Quantification of the lipid-mixing ( H ) and content-mixing results ( N ) in E–G and K–M , respectively . Data are presented as the mean ± SD ( n = 3 ) , technical replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 14211 . 01210 . 7554/eLife . 14211 . 013Figure 7—figure supplement 1 . Liposome clustering induced by Poly-D-lysine in a concentration-dependent manner . 100 μM liposomes ( 65% PC + 20% PE + 15% PS ) mixed with different concentrations of poly-D-lysine were incubated for 40 min and particle size was monitored by DLS . Data plots are presented as the mean ± SEM , technical replicates , degrees of freedom ( n ) are indicated on the top of the plots . DOI: http://dx . doi . org/10 . 7554/eLife . 14211 . 01310 . 7554/eLife . 14211 . 014Figure 7—figure supplement 2 . Liposome clustering and SNARE pairing monitored during liposome fusion . ( A ) C2B did not cluster t-liposomes ( bearing syntaxin-1–SNAP-25 complex ) in the presence of Ca2+ . Plain liposomes or t-liposomes ( 59% PC + 20% PE + 20% PS + 1% PI ( 4 , 5 ) P2 ) bearing 0 . 5 μM syntaxin-1–SNAP-25 complex or were mixed with 0 . 5 μM C2B , 1 mM Ca2+ , after 40 min incubation , particle size of liposomes was monitored using DLS . Data are presented as the mean ± SD , technical replicates . n . s . , not significant ( p > 0 . 05 ) ; ***p<0 . 001; two-way ANOVA , n = 5 . ( B ) SNARE complexes were largely assembled before Ca2+ triggering and not promoted upon Ca2+ triggering ( 1480 s–1500 s ) . Liposome compositions are indicated below the diagram . 0 . 2 mM EDTA , complexin-1 ( cpx ) , Syt1 C2B and poly-D-lysine were present all the time unless indicated , Ca2+ was added to trigger fusion after 1480 s incubation . BODIPY FL ( donor ) and TMR ( tetramethylrhodamine , acceptor ) were labeled on Syb and Syx ( D44C , green sphere and S200C , magenta sphere as shown in the schematic diagram , respectively ) separately . DOI: http://dx . doi . org/10 . 7554/eLife . 14211 . 01410 . 7554/eLife . 14211 . 015Figure 7—figure supplement 3 . No leakiness of liposomes detected in the content-mixing experiments . ( A ) Schematic diagram of the normal content mixing and the leakiness control assays . In the leakiness control , both v-liposomes and t-liposomes were loaded with 40 mM sulforhodamine . Liposome compositions are indicated below the diagram . ( B ) Leakiness was not detected in SNARE-dependent content mixing promoted by poly-D-lysine . ( C ) Leakiness was not detected in Ca2+-triggered content mixing in the presence of poly-D-lysine , complexin and Syt1 C2AB . DOI: http://dx . doi . org/10 . 7554/eLife . 14211 . 015 The fusion between liposomes ( t-liposomes ) reconstituted with syntaxin-1–SNAP-25 complex and synaptobrevin-containing liposomes ( v-liposomes ) was monitored ( Figure 7A , B ) by using a FRET based lipid-mixing and content-mixing assay ( Ma et al . , 2013; Yang et al . , 2015 ) . Consistent with previous observations ( Hui et al . , 2011 ) , addition of 2 µg/ml poly-D-lysine in the mixtures efficiently induced liposome clustering ( Figure 7—figure supplement 1 ) , thereby enhancing SNARE-dependent lipid mixing and content mixing in the absence of Ca2+ ( in the presence of 0 . 2 mM EDTA , Figure 7C , I ) . This data is consistent with the idea that liposome clustering can promote fusion through facilitating SNARE pairing . Complexin-1 inhibited both lipid mixing and content mixing in the presence of poly-D-lysine and in the absence of Ca2+ ( Figure 7D , J ) , whereas further addition of 1 mM Ca2+ at 1500 s triggered both lipid mixing and content mixing ( Figure 7E , K ) . Consistent with previous studies ( Gaffaney et al . , 2008; Xue et al . , 2008 ) , the triggering activity strictly required C2B instead of C2A ( Figure 7E , K ) . We emphasize that C2B was unable to cluster SNARE-bearing liposomes at a concentration of 0 . 5 μM in this fusion system ( Figure 7—figure supplement 2A ) , and trans-SNARE complexes were mostly assembled at the ready-for-fusion stage arrested by complexin-1 ( Figure 7—figure supplement 2B ) . In addition , SNARE pairing was not obviously promoted during the triggering ( Figure 7—figure supplement 2B ) . These results pinpoint the triggering activity of C2B in membrane fusion . Furthermore , to exclude the possibility that content-mixing signals represented by de-quenching of sulforhodamine fluorescence arise from liposome leakiness , we performed control experiments where both v- and t-liposome were loaded with sulforhodamine ( Figure 7—figure supplement 3A ) . Leakiness was not detected in SNARE-dependent content mixing in the presence of poly-D-lysine with and without C2AB–Ca2+ ( Figure 7—figure supplement 3B , C ) , indicating that the real membrane fusion events were observed in our experiments . Using this system , we investigated whether the simultaneous interactions of C2B with PS , PI ( 4 , 5 ) P2 , and the SNARE complex are critical for the triggering activity of C2B . We found that the Ca2+-binding sites mutation ( C2b ) strongly impaired the ability of C2B to trigger both lipid mixing and content mixing in response to Ca2+ ( Figure 7F , L ) . Similar results were obtained when using C2B2KE or C2B2RQ mutations ( Figure 7F , L ) . These results are consistent with those physiological data in previous studies ( Li et al . , 2006; Xue et al . , 2008; Zhou et al . , 2015 ) , suggesting that the simultaneous interactions of C2B with the SNARE complex and membranes are crucial for the triggering function of Syt1 . In addition , we assessed the functional importance of the Ca2+-binding loops of C2A versus C2B in triggering liposome fusion with this system . We found that disrupting the C2A Ca2+-binding sites ( D230N/D232N , C2aB ) caused a moderate impairment in the lipid mixing and content mixing , whereas disrupting the C2B Ca2+-binding sites ( D363N/D365N , C2Ab ) totally abolished the ability of C2AB to trigger fusion ( Figure 7G , M ) . Disruption of the Ca2+-binding sites on both C2 domains ( D230N/D232N/D363N/D365N , C2ab ) completely abolished the triggering effect ( Figure 7G , M ) . Quantification of the lipid-mixing and content-mixing activities of Syt1 and its mutants are shown in Figure 7H , N . Thus , this fusion system successfully reconstituted the triggering role of C2AB and C2B in vitro and reproduced the relative importance of the C2 domain Ca2+-binding sites observed in vivo ( Mackler et al . , 2002; Nishiki and Augustine , 2004; Robinson et al . , 2002; Shin et al . , 2009 ) . Actually , previous in vitro studies ( Kyoung et al . , 2011; Diao et al . , 2012; Zhou et al . , 2015 ) have successfully reconstituted the full-length Syt1–SNARE complex machinery at physiological Syt1 copy number and also included complexin-1 . To complement these studies , we used the fusion system described in Figure 7A , B with C2AB replaced by full-length Syt1 ( reconstituted on v-liposomes ) ( Figure 8A , B ) . Besides , poly-D-lysine was excluded because full-length Syt1 can 'dock' two membranes ( van den Bogaart et al . , 2011a ) . We found that Syt1AB ( WT ) and its mutations on the Ca2+-binding sites ( Syt1aB , Syt1Ab and Syt1ab , respectively ) all stimulated lipid mixing and content mixing in the absence of Ca2+ and complexin-1 ( Figure 8C , F ) , consistent with the docking role of Syt1 . As expected , complexin-1 arrested liposomes bearing Syt1 at the ready-for-fusion stage , and further addition of 1 mM Ca2+ at 1500 s triggered fusion ( Figure 8D , G ) . Consistent with the results observed in our C2AB-based fusion experiments ( Figure 7 ) , Syt1AB and Syt1aB efficiently triggered lipid mixing and content mixing while Syt1Ab and Syt1ab did not ( Figure 8D , E , G , H ) , verifying the functional importance of the C2B Ca2+-binding sites in Ca2+-triggered liposome fusion . Note that a physiological ratio of full-length Syt1 and the SNAREs were reconstituted in our experiments ( Figure 8I ) as previously reported ( Lai et al . , 2014; Zhou et al . , 2015 ) . 10 . 7554/eLife . 14211 . 016Figure 8 . Functional analysis of the Ca2+-binding loops on full-length Syt1 in triggering liposome fusion . ( A and B ) Schematic diagrams of the lipid mixing ( A ) and content mixing ( B ) . ( C and F ) Syt1 stimulates lipid mixing ( C ) and content mixing ( F ) in the absence of Ca2+ and complexin-1 ( cpx ) . ( D and G ) The functional analysis of the Ca2+-binding loops on Syt1 full-length in triggering lipid mixing ( D ) and content mixing ( G ) . ( E and H ) Quantification of the lipid-mixing ( E ) and content-mixing results ( H ) in D and G , respectively . Data are presented as the mean ± SD ( n = 3 ) , technical replicates . ( I ) Analysis of reconstituted proteins on liposomes by SDS-PAGE . Mock injection represents the addition of the buffer ( no Ca2+ ) instead of CaCl2 during triggering . DOI: http://dx . doi . org/10 . 7554/eLife . 14211 . 016
Syt1 acts as a Ca2+ sensor and plays key functions in neurotransmitter release through its C2 domains . The C2 domains exhibit similar overall structures and Ca2+-induced membrane-insertion properties but differ strikingly in their function during release . Despite the fact that the functional significance of the C2B domain in vivo has been successfully reproduced in SNARE-dependent membrane fusion assays in vitro the mechanism by which C2B acts with the SNAREs and membranes to promote fusion is unclear . In the present study , we suggest a membrane-bending property of the C2B domain that arises from its simultaneous interactions with SNARE complexes and membranes . The relevance of our proposed C2B–SNARE complex–membrane interactions is supported by the present study and many previously reported data ( Radhakrishnan et al . , 2009; van den Bogaart et al . , 2011a; Zhou et al . , 2015 ) , concordantly with the increasing realization that Syt1 cooperates with the SNARE complex and membranes in neurotransmitter release . The functional importance of C2B arises most likely from its unique structure that contains abundant basic residues on its surface , which endows C2B with the ability to bind acidic SNARE complexes and/or membranes . Contiguously electrostatic potentials created by the bottom R398 R399 residues and the side K326 K327 residues of C2B contribute to the SNARE binding , as observed in Figure 2A , B . However , as demonstrated with our co-flotation and FRET experiments ( Figure 2C–G ) , the presence of PI ( 4 , 5 ) P2-containing membranes shifts the equilibrium towards an energetically favorable binding where the K326 K327–PI ( 4 , 5 ) P2 interaction dominates . The much higher efficiency of the K326 K327 region , compared to the R398 R399 region , in PI ( 4 , 5 ) P2 binding arises most likely because the K326 K327 region contains a much higher positive-charge density , which enables it to bind tightly to lipid head groups with a highly negative-charge density [i . e . , PI ( 4 , 5 ) P2-microdomains at active zones] ( Honigmann et al . , 2013; Joung et al . , 2012; Park et al . , 2012 ) . Consistent with this , further investigations on the C2B–SNARE complex interaction using a more sensitive bimane-tryptophan quenching assay ( Figure 3 and 4 ) indicate that the R398 R399 region of C2B binds to the SNARE complex or the membrane-anchored SNARE complex in a Ca2+-independent manner . Thus , the existence of the K326 K327–PI ( 4 , 5 ) P2 interaction and the R398 R399–SNARE complex interaction prior to Ca2+ influx likely recruits Syt1 to the fusion sites , which underlies the docking function of Syt1 , as suggested previously ( de Wit et al . , 2009; Honigmann et al . , 2013 ) . A recent report ( Park et al . , 2015 ) argued against the Syt1–SNARE complex interaction because this interaction measured in the study appeared to be completely abolished in the presence of ATP and Mg2+ . However , Syt1 used in the study was labeled at residue 342 , which is close to the polybasic patch , suggesting that this study actually measured the FRET between the K326 K327 region and the SNARE complex , and it is unlikely that it reflects the real Syt1–SNARE complex interaction . Instead , our study found that the R398 R399–SNARE complex interaction persists in the absence and presence of ATP and Mg2+ ( Figure 3 ) . It is also noteworthy that C2B T285W and SNAP-25 R59C mutations used in our bimane-tryptophan quenching assay seem unlikely to affect the particular interaction between Syt1 and the SNARE complex , because these residues are outside the 'primary' interface ( interface area: 720 Å2; including residues Arg398 and Arg399 ) between Syt1 and the SNARE complex ( Zhou et al . , 2015 ) . In addition , the Cα–Cα distance between the two labeling sites is measured at ~12 Å based on the Syt1–SNARE complex structures ( Zhou et al . , 2015 ) , consistent with the relatively large effect on the FRET observed in our experiments . Moreover , we measured a reasonably strong binding Kd [0 . 86 ± 0 . 04 μM and 1 . 53 ± 0 . 04 μM in the presence and absence of PI ( 4 , 5 ) P2 , respectively] between C2B and the SNARE complex in the presence of membranes ( Figure 4 ) . Thus , our binding results , together with the observations that both spontaneous and Ca2+-evoked release are not affected by the presence of 3 mM ATP ( Zhou et al . , 2015 ) , strongly suggest that the interaction between Syt1 and the SNARE complex is not affected by ionic shielding and is physiologically relevant . The finding that the R398 R399 region binds preferentially to the SNARE complex in this study seems to be incompatible with our previous studies ( Arac et al . , 2006; Xue et al . , 2008 ) . This discrepancy may arise from the different experimental conditions used between the present work and our previous studies . Re-examination of liposome clustering in a more stringent condition containing abundant SNARE complexes showed that the liposome-clustering ability of C2B is totally abrogated ( Figure 2H ) . This data suggests that the R398 R399–PS binding might be displaced by the R398 R399–SNARE complex interaction . However , our results could not completely rule out the possibility that a small population of Syt1 molcules act to shorten the distance between membranes via the direct interaction of the R398 R399 region with acidic phospholipids in response to Ca2+ . By detecting penetration of C2B into membranes ( with PS ) as well as binding of C2B to the SNARE complex and PI ( 4 , 5 ) P2 at the same time ( Figures 5 and 6 ) , our results provides acceptable proof for the persistence of the K326 K327–PI ( 4 , 5 ) P2 interaction and the R398 R399–SNARE complex interaction during Ca2+ influx . The presence of the simultaneous SNARE-containing membrane binding of the top Ca2+-binding loops , of the side K326 K327 region , and of the bottom R398 R399 region , leads to a possible membrane-deformation mechanism of Syt1 ( Figure 9 ) : before Ca2+ influx , both the K326 K327–PI ( 4 , 5 ) P2 interaction and the R398 R399–SNARE complex interaction fasten C2B in an orientation parallel with the plasma membrane , leaving the Ca2+-binding loops held back from membrane insertion ( Figure 9A ) ; Ca2+ influx strongly induces fast insertion of the Ca2+-binding loops into membranes , rearranging C2B to an orientation vertical to the plasma membrane . The persistent binding of C2B to PI ( 4 , 5 ) P2 and the membrane-anchored SNARE complex would thus exert a membrane bending force to buck the local membrane outward in response to Ca2+ ( Figure 9B ) . Furthermore , it is likely that binding of C2B to PI ( 4 , 5 ) P2 and SNARE complexes before Ca2+ influx would lead to a ring-like arrangement of C2B molecules around the fusion pore ( Wang et al . , 2014 ) , which would then facilitate bucking local membranes upon collective bending forces in response to Ca2+ . This possible membrane-deformation mechanism is in good agreement with the Syt1 working model proposed by Brunger and colleagues ( Zhou et al . , 2015 ) , and is supported by a recent observation that local membrane protrusions ( 5 nm in height , similar to the size of one C2B molecule ) bucked on the surface of GUV ( giant unilamellar vesicles ) require the presence of Syt1 and assembled SNARE complexes ( Bharat et al . , 2014 ) . 10 . 7554/eLife . 14211 . 017Figure 9 . A working model of Syt1 in triggering membrane fusion . ( A ) Binding of Syt1 C2B to PI ( 4 , 5 ) P2 and primed trans-SNARE complex on the plasma membrane before Ca2+ influx . ( B ) The simultaneous interactions of C2B with primed trans-SNARE complex and PI ( 4 , 5 ) P2-PS-containing membranes in response to Ca2+ cause bucking toward the synaptic vesicle of the plasma membrane . Note that a similar model has been recently proposed ( Zhou et al . , 2015 ) . ( C ) Membrane bucking might cooperate with the action of Syt1 C2B in displacing inhibitory complexin-1 to facilitate the continuous helical SNARE complex assembly , thus triggering membrane fusion and neurotransmitter release . SV , synaptic vesicle; PM , pre-synaptic membrane . DOI: http://dx . doi . org/10 . 7554/eLife . 14211 . 017 The proposed membrane-deformation mechanism is supported by the finding that all three interactions of C2B are required for the triggering function of the Syt1 in liposome fusion ( Figure 7 ) . This membrane-deformation mechanism explains very well the recent observation that isolated C2B readily bends membranes ( Hui et al . , 2009 ) , and suggests that membrane insertion of the Ca2+-binding loops alone is not sufficient to drive membrane deformation without the assist of the R398 R399 and the K326 K327 regions . Thus , the functional importance of the C2B Ca2+-binding sites observed in previous in vivo studies and our present liposome fusion experiments ( Figures 7 and 8 ) can be explained: disruption of the C2B Ca2+-binding sites abrogates the simultaneous interaction of C2B with acidic membrane lipids [i . e . , PI ( 4 , 5 ) P2 and PS] and the membrane-anchored SNARE complex , so that the ability of C2B to bend membranes is absolutely abolished . Our data reinforce the idea that local membrane deformation by Ca2+–Syt1 is key for the triggering function of Syt1 in release ( Hui et al . , 2009; Zhou et al . , 2015 ) . Our data also reinforce the notion that the coordinated efforts of two or more interactions from one protein can induce membrane deformation ( McMahon and Boucrot , 2015 ) . The liposome fusion results support the notion that SNARE complexes are already partially assembled before Ca2+ influx ( Figure 7 and Figure 7—figure supplement 2B ) , which enables complexin-1 binding and thereby strains such a complex in a stage ready for fusion ( Rizo and Xu , 2015 ) . In response to Ca2+ , C2B-induced membrane bucking would reduce the distance between two apposed membranes , which might cooperate with the action of C2B in displacing inhibitory complexin-1 , to facilitate the continuous helical SNARE complex assembly that propagates through the linker region into the transmembrane domains ( Figure 9C ) . Thus , high curvature stresses induced by C2B and the energy released from the C-terminal SNARE complex assembly might be coupled together in response to Ca2+ to overcome the energy barrier for membrane fusion . It is noteworthy that C2A would have an ancillary role by binding to one membrane and helping to dictate the apparent Ca2+ affinity of Syt1 ( Robinson et al . , 2002; Zhou et al . , 2015 ) . Altogether , our results add increasing evidence for the triggering mechanism by which Syt1 acts in concert with the SNARE complex and membranes to promote membrane fusion .
The cytoplasmic domain of rat Syt1 ( known as C2AB ) used in this study comprises residues 140–421 , and the C2A and C2B domains comprise residues 140–266 and 270–421 , respectively . All Syt1 fragments or their mutants , full length rat synaptobrevin-2 and its cytoplasmic domain ( residues 29–93 ) , the H3 domain of rat syntaxin-1a ( residues 191–253 ) and full-length rat complexin-1 were constructed into pGEX-6p-1 vector ( GE Healthcare; Piscataway , NJ ) ; full-length rat Syt1 ( all cysteines were mutated to alanine except the cysteine residue at position 277 ) , rat C-terminal syntaxin-1a ( residues 183–288 , without Habc domain ) , full-length human SNAP-25a ( with its four cysteins mutated to serines ) and SNAP-25a 3M ( D51A/E52A/E55A ) were constructed into pET28a vector ( Novagen; Australia ) ; rat syntaxin-1a C-terminal ( residues 183–288 , without Habc domain ) and human SNAP-25a ( with its four cysteins mutated to serines ) or SNAP-25a 3M were constructed into pETDuet-1 vector ( Novagen ) . All the recombinant proteins above were expressed in E . coli BL21 DE3 cells and purified as previously described ( Lai et al . , 2014; Ma et al . , 2013; van den Bogaart et al . , 2011a ) . Point mutations were prepared by using the QuickChange Site-Directed Mutagenesis Kit ( Agilent Technologies; Santa Clara , CA ) . Purified GST-H3 ( residues 191–253 of syntaxin-1a ) was incubated with SNAP-25 or SNAP-25 3M ( SN25 3M ) and synaptobrevin ( residues 29–93 ) overnight and analyzed by SDS-PAGE to confirm the SNARE complex formation . 20 μM GST-SNARE complex or GST-H3 was mixed with 10 μM Syt1 fragments and 20 μl 50% ( v/v ) Glutathione Sepharose 4B affinity media ( GE Healthcare ) to a final volume of 50 μl . After 2 hr gentle shaking at 4°C , beads were washed 3 times using 25 mM HEPES pH 7 . 4 , 150 mM KCl , and 10% glycerol ( buffer A ) . Samples were analyzed by SDS-PAGE . All experiments were performed in the absence of Ca2+ . Lipid powder ( all from Avanti Polar Lipids; Alabaster , AL ) was dissolved in chloroform at a concentration of 10 mg/ml for storage at -20°C , except for brain PI ( 4 , 5 ) P2 ( from porcine's brain ) in chloroform:methanol:water 20:9:1 at 1 mg/ml . Lipids were mixed at the proper ratio as indicated in the figures or legends to a final concentration of 5 mM and dried under nitrogen followed by vacuum for at least 3 hr . Lipid films were dissolved in buffer A containing 0 . 2 mM Tris ( 2-carboxyethyl ) phosphine ( TCEP , Sigma Aldrich; St . Louis , MO ) and 1% CHAPS ( w/v , Amresco; Solon , OH ) and vortexed for 5 min . For preparing proteoliposomes , purified proteins dissolved in 1% CHAPS ( w/v ) were added into the dissolved lipid films to a final protein-to-lipid ratio of 1:200 ( for SNAREs ) and/or 1:1000 ( for Syt1 full-length ) , respectively; for plain liposomes , equivalent buffer A containing 1% CHAPS ( w/v ) was added; after 30 min incubation at room temperature , the mixtures were dialyzed against buffer A containing 0 . 1 mM TCEP and 1 . 0 g/L Bio-beads ( Bio-Rad; Hercules , CA ) at 4°C 3 times . The prepared proteoliposomes were checked using Dynamic Light Scattering ( DLS ) on a DynaPro Nanostar ( Waytt Technology , Santa Barbara , CA ) before using . Liposome ( 2 mM total lipids ) compositions are indicated in the figures or legends . Liposomes were incubated with 10 μM proteins ( ±Ca2+ ) in buffer A ( unless stated otherwise ) for 40 min at room temperature . The liposomes and bound proteins were isolated by flotation on a Histodenz ( Sigma Aldrich ) density gradients ( 40%:30% ) using a SW 55 Ti rotor ( Beckman Coulter; Boulevard Brea , CA ) at 163 , 000 ×g for 40 min . Samples from the top and the bottom of the gradient ( 20 μl ) were taken and analyzed by SDS-PAGE and Coomassie Brilliant Blue ( CBB ) staining . For liposome-protein FRET experiments ( Figure 2F , G ) , 100 μM liposomes were mixed with 5 μM Syt1 C2B H315C-NBD and 20 μM soluble SNARE complex . Fluorescence was monitored in a physiological ion condition ( buffer A ) on a PTI QM-40 fluorescence spectrophotometer ( PTI; Edison , NJ ) with an excitation wavelength of 460 nm and an emission spectra from 500 to 650 nm . For bimane-tryptophan electron transfer in the absence of membranes ( Figure 3 ) , 1 μM assembled SNARE complex which harbors monobromobimane ( mBBr , Molecular Probes; Eugene , OR ) labeled SNAP-25a R59C was mixed with 2 μM Syt1 C2B or its mutant ( as indicated in the figures ) , and additional 1 mM magnesium chloride ( analytical grade ) and 3 mM ATP ( Bio Basic Inc . ; Canada ) were incorporated as indicated . Fluorescence was monitored on a PTI QM-40 fluorescence spectrophotometer with an excitation wavelength of 380 nm and emission spectra from 400 to 600 nm . For the Kd measurement between Syt1 C2B and membrane-anchored SNARE complex in Figure 4 , Syt1 C2B T285W was mixed with 200 μM liposome [64% POPC , 20% POPE , 15% DOPS and/or 1% PI ( 4 , 5 ) P2 , removed PI ( 4 , 5 ) P2 was supplied with POPC] bearing bimane-labeled SNARE complex ( with a protein-to-lipid ratio of 1:200 ) with indicated concentration . Bimane fluorescence was monitored on a PTI QM-40 fluorescence spectrophotometer with an excitation wavelength of 380 nm and an emission wavelength of 470 nm . Data plots were fitted using the Michaelis-Menten equation , where Vmax was constrained to 100 ( % Quenched efficiency ) . For bimane-tryptophan electron transfer and NBD membrane-insertion assay ( Figure 6 ) , 100 μM liposomes ( bearing bimane-labeled SNARE complex with syntaxin-1 transmembrane domain anchored on liposomes and with a protein-to-lipid ratio of 1:200 ) were mixed with 5 μM NBD-labeled Syt1 C2B . A dual excitation of 380 nm ( for bimane ) and 460 nm ( for NBD ) and a dual emission spectrum of 400–600 nm and 500–620 nm was used to collect the fluorescence of bimane and NBD , respectively . Fluorescence anisotropy assay in Figure 2—figure supplement 1 was carried out as previously described ( Wiederhold and Fasshauer , 2009 ) . 200 nM BODIPY FL ( Molecular Probes ) labeled synaptobrevin ( residues 29–93 , S61C ) was mixed with 1 μM syntaxin-1 ( residues 191–253 ) and SNAP-25 or 1 μM pre-incubated syntaxin-1–SNAP-25 complex . For the SNARE-pairing assay shown in Figure 7—figure supplement 2B , 0 . 2 mM EDTA , 2 μg/ml poly-D-lysine , 20 μM complexin and 0 . 5 μM Syt1 C2B were incorporated into a mixture of t-liposome ( 100 μM lipids and 0 . 5 μM syntaxin-1 S200C-tetramethylrhodamine [TMR , Molecular Probes]-SNAP-25 ) and v-liposome ( 50 μM lipids and 0 . 25 μM synaptobrevin D44C-BODIPY FL ) unless otherwise indicated . After incubation for 1480 s , 1 mM Ca2+ was added . Donor fluorescence was monitored with an excitation wavelength of 485 nm and an emission wavelength of 513 nm . All lipid compositions are indicated in the figures or legends . All experiments were performed at 25°C in a 1-cm quartz cuvette in buffer A . Liposome clustering assay was carried out as previously described ( Arac et al . , 2006; Xue et al . , 2008 ) . Briefly , 100 μM liposomes were mixed with 1 mM Ca2+ , with/without 10 μM of a soluble SNARE complex or the equivalent volume of buffer and the indicated concentration of Syt1 C2B was incorporated for 40 min incubation at room temperature . Liposome compositions are indicated in the figures or legends . Particle sizes were analyzed by DLS using a DynaPro Nanostar ( Waytt Technology ) at 25°C . General procedures are indicated in Figure 7A and Figure 8A . For lipid mixing using the soluble Syt1 fragments , 0 . 2 mM EDTA , 2 ug/ml poly-D-lysine , 20 μM complexin and 0 . 5 μM Syt1 fragments were added to a mixture of 100 μM t-liposomes ( bearing 0 . 5 μM syntaxin-1–SNAP-25 ) and 50 μM v-liposomes ( bearing 0 . 25 μM synaptobrevin ) . For lipid mixing using reconstituted full-length Syt1 , poly-D-lysine was excluded , and v-liposomes ( 50 μM ) were reconstituted with synaptobrevin ( bearing 0 . 25 μM synaptobrevin ) and 0 . 05 μM full-length Syt1 ( with a protein-to-lipid ratio of 1:1000 ) . After incubation , 1 mM Ca2+ was added to trigger lipid mixing at 1500 s . Donor ( NBD ) fluorescence were monitored on a PTI QM-40 fluorescence spectrophotometer with an excitation wavelength of 460 nm and an emission wavelength of 538 nm . Fluorescence in Figure 7 was normalized to the initial fluorescence intensity . Fluorescence in Figure 8 were normalized to the fluorescence intensity at 0 . 1% Triton X-100 . All experiments were carried out at 25°C in buffer A . Lipid compositions are indicated in the figures or legends . General procedures are indicated in Figure 7B and Figure 8B . 40 mM sulforhodamine B ( Sigma ) was loaded into v-liposome ( harboring synaptobrevin with or without full-length Syt1 ) without lipid probes . Other details are the same as with lipid mixing assays . Leakiness control was performed with 40 mM sulforhodamine B both loaded into t-liposomes and v-liposomes . Fluorescence was monitored on a PTI QM-40 fluorescence spectrophotometer with an excitation wavelength of 565 nm and an emission wavelength of 580 nm . Fluorescence normalization is the same as that used in the lipid-mixing assay . All experiments were carried out at 25°C in buffer A . Lipid compositions are indicated in the figures or legends . Additional 10% Cholesterol ( Chol , from ovine wool , Avanti Polar Lipids ) was incorporated into v-liposomes to prevent leakiness . Prism 6 . 01 ( Graphpad ) and Image J ( NIH ) were used for graphing and statistical tests , all of which are described in figure legends . | Information travels around the nervous system along cells called neurons , which communicate with each other via connections called synapses . When a signal travelling along one neuron reaches a synapse , it triggers the release of molecules known as neurotransmitters . These molecules are then taken up by the next neuron to pass the signal on . Neurotransmitters are stored in compartments called synaptic vesicles and their release from the first neuron depends on the synaptic vesicles fusing with the membrane that surrounds the cell . This “membrane fusion” process is driven by a group of proteins called the SNARE complex . Membrane fusion is triggered by a sudden increase in the amount of calcium ions in the cell , which leads to an increase in the activity of a protein called synaptotagmin-1 . A region of this protein known as the C2B domain is able to detect calcium ions , and it can also bind to the cell membrane and SNARE complex proteins . However , it is not clear what roles these interactions play in driving the release of neurotransmitters . Wang , Li et al . have used a variety of biophysical techniques to study these interactions in more detail using purified proteins and other cell components . The experiments show that all three interactions occur at the same time and are all required for synaptotagmin-1 to trigger membrane fusion . Wang , Li et al . propose that these interactions allow synaptotagmin-1 to bend a section of the cell membrane in response to calcium ions . The experiments also show that the C2B domain interacts more strongly with the SNARE complex than previously thought . A future challenge is to observe whether synaptotagmin-1 works in the same way in living cells . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"structural",
"biology",
"and",
"molecular",
"biophysics",
"neuroscience"
] | 2016 | Synaptotagmin-1 C2B domain interacts simultaneously with SNAREs and membranes to promote membrane fusion |
IFIT ( interferon-induced with tetratricopeptide repeats ) proteins are critical mediators of mammalian innate antiviral immunity . Mouse IFIT1 selectively inhibits viruses that lack 2'O-methylation of their mRNA 5' caps . Surprisingly , human IFIT1 does not share this antiviral specificity . Here , we resolve this discrepancy by demonstrating that human and mouse IFIT1 have evolved distinct functions using a combination of evolutionary , genetic and virological analyses . First , we show that human IFIT1 and mouse IFIT1 ( renamed IFIT1B ) are not orthologs , but are paralogs that diverged >100 mya . Second , using a yeast genetic assay , we show that IFIT1 and IFIT1B proteins differ in their ability to be suppressed by a cap 2'O-methyltransferase . Finally , we demonstrate that IFIT1 and IFIT1B have divergent antiviral specificities , including the discovery that only IFIT1 proteins inhibit a virus encoding a cap 2'O-methyltransferase . These functional data , combined with widespread turnover of mammalian IFIT genes , reveal dramatic species-specific differences in IFIT-mediated antiviral repertoires .
Mammalian antiviral defenses rely on the combined functions of a vast collection of immunity genes . One important arm of the immune system is innate , or cell intrinsic , antiviral immunity , which establishes an antiviral state in cells by signaling through the cytokine interferon ( IFN ) and subsequent upregulation of hundreds of genes known collectively as IFN-stimulated genes ( ISGs ) . ISGs serve as the first line of defense against viruses and typically function to sense pathogen-associated molecular patterns ( PAMPs ) or directly restrict virus replication ( Schneider et al . , 2014; Schoggins , 2014 ) . Among the most highly upregulated ISGs are members of the IFIT ( interferon-induced with tetratricopeptide repeats ) genes . Upon IFN-stimulation or viral infection , the mRNA levels of IFITs increase 100- to 1000-fold , and IFIT proteins have been implicated in inhibition of a broad range of viruses ( Diamond and Farzan , 2013; Fensterl and Sen , 2015; Vladimer et al . , 2014 ) . However , the number and identity of IFIT genes can vary substantially between species . For instance , while humans have five intact IFIT genes ( IFIT1 , 1B , 2 , 3 and 5 ) , rats have four ( IFIT1 , 1b , 2 and 3 ) and mice have six ( IFIT1 , 1b , 1c , 2 , 3 and 3b ) ( Fensterl and Sen , 2011; 2015; Liu et al . , 2013 ) . The functional consequences of IFIT family evolution are unknown , in part because the antiviral functions and specificities of IFITs are incompletely characterized . Initial studies with IFIT1 and IFIT2 from humans and mice indicated that these proteins might mediate their antiviral activity by inhibiting mRNA translation through interactions with the translation initiation factor eIF3 ( Guo et al . , 2000; Hui et al . , 2003; Terenzi et al . , 2005 ) . In this way , IFITs appeared to function similarly to another critical mediator of the innate immune system , Protein Kinase R ( PKR ) , by globally inhibiting mRNA translation . In the case of PKR , the recognition of cytoplasmic double-stranded RNA , a by-product of viral replication , triggers its activity and the global repression of protein synthesis ( Dever et al . , 2007 ) . Such a 'self versus non-self' molecular pattern has been more enigmatic for IFIT proteins , and it has been challenging to determine how IFITs discriminate viral from host RNAs to repress viral replication specifically . An elegant means by which one IFIT protein distinguishes 'self versus non-self' mRNAs was revealed by recent studies on mouse IFIT1 . During mammalian mRNA processing , the 5' cap region is 2'O-methylated from a cap0-structure ( 7mGpppN , where 7mG is the 7-methyl guanosine , ppp is the triphosphate linkage , and N is any nucleotide ) to a cap1-structure ( 7mGpppNm ) ( Banerjee , 1980 ) . This reaction is catalyzed in the host nucleus by a dedicated 2'O-methyltransferase , known as a cap1-methyltransferase ( hCMTR1 in humans ) ( Belanger et al . , 2010 ) . Interestingly , many viruses have evolved ways to mimic host cap1-structures ( Banerjee , 1980; Decroly et al . , 2012 ) . For several viruses that replicate in the cytoplasm , such as poxviruses , flaviviruses , coronaviruses , and rhabdoviruses , 2'O-methylation of the cap is catalyzed by a virally-encoded cap1-methyltransferase . For other viruses , such as orthomyxoviruses , arenaviruses , and bunyaviruses , the effect is achieved by ‘cap-snatching’ , in which a segment of cap1-modified host mRNA is appended to viral mRNAs . Either strategy results in methylated ( cap1- ) mRNAs , suggesting that unmethylated ( cap0- ) mRNAs could be recognized as a 'non-self' pattern that elicits host immunity . Indeed , mouse IFIT1 was discovered to inhibit replication of numerous viruses naturally lacking or mutated to lack 2'O-methylation by directly binding and inhibiting translation of cap0-mRNAs ( Daffis et al . , 2010; Hyde et al . , 2014; Ma et al . , 2014; Menachery et al . , 2014; Szretter et al . , 2012; Zust et al . , 2011; Habjan et al . , 2013; Kimura et al . , 2013; Kumar et al . , 2014 ) . In this way , mouse IFIT1 selectively inhibits viruses that translate proteins from 'non-self' cap0-mRNAs , while the host protects itself via 'self' cap1-structures on its mRNAs ( Diamond , 2014; Hyde and Diamond , 2015 ) . Given the importance of the cap0-mRNA versus cap1-mRNA specificity in directing mouse IFIT1’s repressive effects against viruses , we might expect that other mammalian IFIT1 genes would preserve such discrimination . Surprisingly , studies on human IFIT1 have belied this expectation . For instance , human IFIT1 was shown to inhibit mRNA translation and replication of parainfluenza virus 5 ( PIV5 ) , despite the fact that PIV5 encodes a cap1-methyltransferase and PIV5 mRNAs are 2'O-methylated on their caps ( Andrejeva et al . , 2013 ) . Other studies have implicated human IFIT1 in inhibition of hepatitis C virus ( HCV ) ( Raychoudhuri et al . , 2011; Wang et al . , 2003 ) , human papillomavirus ( HPV ) ( Terenzi et al . , 2008 ) , influenza A virus ( IAV ) and vesicular stomatitis virus ( VSV ) ( Pichlmair et al . , 2011 ) , none of which are predicted to translate proteins from cap0-mRNAs . These seemingly contradictory results regarding the antiviral specificities of mouse IFIT1 and human IFIT1 have led to a conundrum in the field regarding the molecular functions and antiviral specificity of IFIT proteins in general . However , one implicit assumption underlying the expectation that human and mouse IFIT1 should function similarly is that mouse and human IFIT1 represent orthologous genes . Here , we show that this is not the case . Using detailed phylogenetic analyses of IFIT genes in vertebrates , made possible by deconvolving the confounding effects of recurrent gene conversion , we show that human IFIT1 and mouse IFIT1 are two distinct paralogous genes that diverged early in mammalian evolution . Mouse genomes only have representatives of one of these paralogous genes ( which we rename IFIT1B ) , whereas several primates have lost IFITB but retained IFIT1 . We further show that these two divergent IFIT1 paralogs have distinct specificities . Using a genetic assay we developed in budding yeast ( which exploits yeast's inherent lack of a cap1-methyltransferase ) , we show that mouse and other IFIT1B proteins , but not IFIT1 proteins , discriminate cap0- from cap1-mRNA methylation by selectively inhibiting growth only when a cap1-methyltransferase is missing . Consistent with this activity , we show that only IFIT1B proteins can inhibit replication of a virus that lacks functional cap1-methylation . In contrast , we find that human and other IFIT1 proteins can inhibit the growth of yeast or of a virus that encodes a cap1-methyltransferase . Our findings thus resolve the apparent mystery of IFIT1’s altered antiviral discrimination mechanism in mouse and human and delineate an important role for human IFIT1 in restriction of viruses including those that produce cap1-mRNAs . Our analyses also reveal a high degree of dynamism and alternate retention of IFIT1 or IFIT1B paralogs in mammalian genomes . This IFIT gene turnover might profoundly impact the spectrum of viruses that different mammalian species are capable of restricting .
The expectation that mouse and human IFIT1 should have a similar antiviral function and specificity implicitly assumes that these two genes are orthologous , i . e . , they diverged when the common ancestors of humans and mice diverged as species . However , there is evidence suggesting that this assumption may not be valid . First , there have been extensive changes to the number and identity of IFIT genes between humans and mice ( [Fensterl and Sen , 2011;2015; Liu et al . , 2013] and Figure 1A ) . Furthermore , mouse IFIT1 is more similar at the sequence level to another human IFIT , the poorly characterized human IFIT1B , than to human IFIT1 ( 57% versus 53% pairwise amino acid identity ) . Based on these data , and the contradictory functional data for mouse IFIT1 and human IFIT1 antiviral activities ( Andrejeva et al . , 2013; Daffis et al . , 2010; Pichlmair et al . , 2011; Pinto et al . , 2015 ) , we formally tested for IFIT1 orthology between mouse and human . To do so , we first undertook detailed phylogenetic analyses of the IFIT genes found in several well-assembled mammalian genomes ( Figure 1A ) . 10 . 7554/eLife . 14228 . 003Figure 1 . Discordant signatures of synteny and phylogeny for mammalian IFIT genes . ( A ) Alignment of the IFIT gene locus from several mammalian genomes . Colored arrows indicate intact IFIT genes and grey arrows indicate neighboring syntenic genes . At left is a schematic phylogenetic tree showing the relatedness of the indicated species . In carnivores such as cats and ferrets , the IFIT locus is split with IFIT5 in one chromosomal location and the remainder of the IFITs in another . In the marmoset genome , the IFIT1B gene has been pseudogenized by frameshift and nonsense mutations , as indicated by the dashed white arrow . Note that the nomenclature for the IFIT genes marked with asterisks is as previously proposed; we suggest a revised nomenclature scheme for those genes in this report from Figure 2 onwards . A version of this figure , with revised gene names and coloring , is found in Figure 2—figure supplement 4 . ( B ) Maximum likelihood phylogenetic tree generated using an alignment of the entire gene sequence of the indicated IFITs . Bootstrap values greater than 80% are shown along the supported branch . Gene names with asterisks next to them have been revised from Figure 2 onward . Accession numbers for all sequences are found in Supplementary file 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 14228 . 003 Our phylogenetic analyses were facilitated by the fact that IFIT genes are located in a single locus in most mammalian species ( Liu et al . , 2013 ) . For example , human chromosome 10 contains a single locus that encodes all five intact IFIT genes: IFIT1 , IFIT1B , IFIT2 , IFIT3 and IFIT5 ( Figure 1A ) . We found the IFIT locus organization is similar in many other mammalian genomes including African green monkeys , rabbits , cats , ferrets , and armadillos ( Figure 1A ) . In this shared syntenic arrangement , two IFIT1 paralogs are found next to each other in the same sense orientation as IFIT2 and IFIT3 , suggesting that a single duplication of IFIT1 may have occurred early in mammalian evolution , i . e . , before the radiation of placental mammals over 100 million years ago . In primates , rabbits and carnivores , the paralogs immediately adjacent to IFIT3 are named IFIT1B in sequence databases , whereas the distal paralogs are named IFIT1 . Based on this previous nomenclature , we will henceforth refer to these as the IFIT1 and IFIT1B gene families . In contrast , the IFIT locus in mice and rats is arranged quite differently than other mammalian genomes . Instead of two paralogs , mice have three IFIT1 paralogs , two of which are in the opposite orientation to IFIT2 and IFIT3 ( Figure 1A ) . Thus , shared synteny analyses are not suitable for assigning orthology/ paralogy relationships for the mouse IFIT1 genes . We , therefore , constructed maximum likelihood phylogenetic trees of full-length IFIT gene sequences across these divergent mammalian species to determine how mouse IFIT1 genes are related to those found in other species ( Figure 1B ) . We found unambiguous phylogenetic signatures that delineate IFIT2 , IFIT3 and IFIT5 genes into distinct , monophyletic clades that diverged as expected based on known mammalian species evolution ( O'Leary et al . , 2013 ) . In contrast to these three gene families , we were unable to resolve IFIT1 and IFIT1B genes into distinct clades in our phylogenetic analyses . Instead , similar to previous studies ( Liu , 2013 ) , we found several instances of IFIT1 and IFIT1B from the same species ( e . g . , cat ) appearing more phylogenetically related to each other than to their similarly named genes from a sister species ( e . g . , ferret ) ( Figure 1B ) . The discordant phylogenetic signatures of IFIT1 and IFIT1B evolution could have two alternative explanations . The first is that species-specific duplication may have independently given rise to IFIT1 and IFIT1B genes in different genomes , as has been previously proposed ( Liu et al . , 2013 ) . Such a duplication pattern might explain why cat IFIT1B is more phylogenetically related to cat IFIT1 than to IFIT1B genes from other genomes . However , such an explanation would require recurrent duplication of IFIT genes in many lineages into the same genetic location and orientation , and is thus unlikely . We , therefore , considered an alternate explanation . In this alternative , IFIT1 and IFIT1B genes duplicated early in mammalian evolution but have recurrently recombined with each other . Resulting gene conversion would scramble the phylogenetic relatedness of the two genes without altering their genomic locations . Such gene conversion has been shown to contribute to the evolution of several multigene families including mammalian interferon alpha genes ( Benovoy and Drouin , 2009; Hurles , 2004; Petronella and Drouin , 2011; Santoyo and Romero , 2005; Song et al . , 2011; Woelk et al . , 2007; Yasukochi and Satta , 2015 ) . To address this possibility , we compared the sequences of IFIT1 and IFIT1B genes within species and between species to look for evidence of gene conversion . We found strong characteristic signatures of gene conversion frequently occurring between the 5' portions of IFIT1 and IFIT1B genes . For example , the sequences of IFIT1 and IFIT1B genes in cats are 86% identical at the whole gene level . However , these two genes are 97% identical in the 5' ends of the genes ( nucleotides 1–948 in cat IFIT1 ) but only 63% identical in the 3' ends of the genes ( Figure 2A ) . This dichotomy is apparent even if we restrict the analysis to just synonymous substitutions , where we observe a much higher rate of synonymous substitutions in the 3' region of the alignment ( Figure 2—figure supplement 1 ) . Such an uneven distribution of sequence differences strongly suggests that the 5' ends of IFIT1 and IFIT1B underwent gene conversion recently in the evolution of the cat IFIT locus , leading to sequence homogenization of the first two-thirds of the gene . In contrast , comparing orthologous genes between species ( e . g . IFIT1B between cat and ferret ) reveals an even distribution of differences across the gene sequences ( Figure 2B ) indicating that both the 5' and 3' regions have diverged to the expected degree between species . Moreover , this pattern of gene conversion is seen in pairwise alignments of many IFIT1 and IFIT1B genes . For instance , between ferret IFIT1 and IFIT1B , the 5' ends are 98% identical , but 3' ends are only 62% identical ( Figure 2B ) . A similar dichotomy is evident in a comparison between rabbit IFIT1 and IFIT1B and between armadillo IFIT1 and IFIT1B ( Figure 2—figure supplement 1 ) . Interestingly , all of these gene conversion events have a similar recurrent breakpoint approximately two-thirds of the way through the gene sequence ( Figure 2—figure supplement 1 ) that maps to a 'pivot' point between subdomains in the homologous IFIT5 structure ( Abbas et al . , 2013 ) ( Figure 2—figure supplement 2 ) . In all cases , the 5' ends of the IFIT1 genes are largely homogenized , whereas the 3' ends appear to be consistently diverging . Thus , our findings suggest that recurrent gene conversion has confounded the phylogenetic relationships of the 5’ segment of IFIT1 and IFIT1B , which likely do not accurately represent the actual ancestry of these genes . 10 . 7554/eLife . 14228 . 004Figure 2 . An ancient gene duplication gave rise to paralogous IFIT1 and IFIT1B gene families . ( A ) Comparison of the nucleotide sequence of cat IFIT1 and IFIT1B indicating gene conversion between the two paralogs . The middle rectangle represents a pairwise sequence alignment , with black vertical lines indicating differences between the two gene sequences . Percent nucleotide identity for the whole gene comparison is shown above the alignment schematic , with the 5' end and 3' end percent identities shown below . ( B ) Additional pairwise sequence alignments are shown as in part A . Additional examples supporting a common recombination breakpoint are shown in Figure 2—figure supplement 1 . Mapping of the recombination breakpoint onto IFIT structural models ( Abbas et al . , 2013; Yang et al . , 2012 ) is shown in Figure 2—figure supplement 2 . ( C ) Maximum likelihood phylogenetic tree of IFIT genes using only the 3' end of the gene alignment ( corresponding to bases 907–1437 of human IFIT1 ) . Bootstrap values greater than 80% are shown along the supported branch . For clarity , IFIT2 , IFIT3 and IFIT5 branches are collapsed and represented as triangles . A maximum likelihood phylogenetic tree of the 5' end of the IFIT gene alignment is shown in Figure 2—figure supplement 3 . A figure showing revised gene names on the gene locus alignment is shown in Figure 2—figure supplement 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 14228 . 00410 . 7554/eLife . 14228 . 005Figure 2—figure supplement 1 . Recurrent gene conversion across several diverse mammalian species has utilized a similar recombination breakpoint . Pairwise alignments between IFIT1 and IFIT1B genes in the indicated species are represented as in Figure 2A , with black vertical lines indicating differences between the two gene sequences . Percent identity for the whole gene comparison is shown directly above the alignment schematic , with the 5' end and 3' end percent identities shown directly below . Below each pairwise alignment schematic is a sliding window calculation using K-Estimator ( Comeron , 1999 ) of the rate of synonymous changes ( dS ) for each alignment . Each 200 nt window is represented by a red horizontal line at the calculated dS value . The recombination breakpoint ( red vertical arrow ) between regions of high sequence identity and low dS ( 5' end ) and low sequence identity and high dS ( 3' end ) is similar across these distantly related species . DOI: http://dx . doi . org/10 . 7554/eLife . 14228 . 00510 . 7554/eLife . 14228 . 006Figure 2—figure supplement 2 . Gene conversion of IFIT1 and IFIT1B homogenizes some IFIT structural domains while leaving others to independently evolve . Structures of human IFIT5 complexed with RNA ( [Abbas et al . , 2013] , PDB code 4HOT ) and the human IFIT2 dimer ( [Yang et al . , 2012] , PDB code 4G1T ) are shown . In both cases , residues in the N-terminus of the protein that align to the recurrently recombining region of IFIT1 and IFIT1B ( see Figure 2—figure supplement 1 ) are shown in green , while the non-recombining region is shown in blue . For IFIT5 , the major structural domains as described in ( Abbas et al . , 2013 ) are indicated , with the recombination breakpoint ( Figure 2—figure supplement 1 ) falling in the 'pivot' region between the second and third subdomains . For IFIT2 , one monomer is shown in grey , revealing that the non-recombining region lies outside of the swapped interface that underpins IFIT2 dimerization ( Yang et al . , 2012 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14228 . 00610 . 7554/eLife . 14228 . 007Figure 2—figure supplement 3 . Phylogenetic tree of the recombining region of mammalian IFIT genes . Maximum likelihood phylogenetic tree of IFIT genes using only the region 5' to the recombination breakpoint ( see Figure 2C and Figure 2—figure supplement 1 ) . Bootstrap values greater than 80% are shown along the supported branch . The red and blue coloring is based on the phylogeny shown in Figure 2C . DOI: http://dx . doi . org/10 . 7554/eLife . 14228 . 00710 . 7554/eLife . 14228 . 008Figure 2—figure supplement 4 . Synteny of mammalian IFITs with updated IFIT1 and IFIT1B gene names . Alignment of the IFIT gene locus from several mammalian genomes as in Figure 1 . Gene colors and names have been revised to correspond to the phylogenetic relatedness of genes shown in Figure 2C . DOI: http://dx . doi . org/10 . 7554/eLife . 14228 . 008 To determine the phylogenetic relationship of IFIT1 and IFIT1B genes in the absence of gene conversion , we created a maximum likelihood phylogenetic tree of the same IFIT sequences shown in Figure 1B , but with just the nucleotide sequences 3' of the observed recombination breakpoint ( s ) ( Figure 2C ) . In contrast to the analyses using the full-length IFIT genes ( Figure 1B ) or the 5' end of the IFIT genes ( Figure 2—figure supplement 3 ) , we now discern clear separation of IFIT1 and IFIT1B genes . Both IFIT1 and IFIT1B genes form distinct monophyletic clades that diverge according to mammalian species evolution , similar to other IFIT genes . Based on this phylogenetic concordance , as well as the agreement with synteny data , we infer that this phylogeny reflects the actual ancestry of IFIT1 and IFIT1B genes in mammals . Importantly , these data indicate that IFIT1 and IFIT1B duplicated early in mammalian evolution . Following this duplication , the 3' ends of the two genes have been diverging according to speciation events , whereas the 5' ends have recurrently recombined . These data also reveal that all three mouse IFIT1 paralogs ( previously named mouse IFIT1 , IFIT1b and IFIT1c ) and both rat paralogs ( previously named rat IFIT1 and IFIT1b ) unambiguously group within the IFIT1B gene family . We , therefore , infer that mouse and rat have lost all copies of IFIT1 , likely through a deletion of both IFIT1 and the neighboring IFIT5 . Instead , mouse and rat genomes now bear recently duplicated copies of IFIT1B . As a result of our phylogenetic analyses , we henceforth refer to these genes as mouse/rat IFIT1B ( previously mouse/rat IFIT1 ) , IFIT1B2 ( previously mouse/rat IFIT1b ) and IFIT1B3 ( previously mouse IFIT1c ) ( Figure 2C and Figure 2—figure supplement 4 ) . This recombination-aware phylogenetic approach thus reveals substantial changes in the IFIT gene repertoire even within this limited number of mammalian species . To determine the broader impact of IFIT gene duplication , loss and conversion beyond our initial sampling of mammalian species , we assembled over 200 IFIT genes from 51 vertebrate species and conducted maximum likelihood phylogenetic analyses ( Figure 3—figure supplement 1 ) . We found that most placental mammals contain members of the IFIT1 , 1B , 2 , 3 and 5 gene families ( Figure 3 ) . In contrast , marsupials encode a more limited set of IFIT genes including a single gene that predates the IFIT2/IFIT3 duplication and a single gene that predates IFIT1/IFIT1B divergence ( Figure 3 and Figure 3—figure supplement 1 ) . Consistent with other analyses ( Liu et al . , 2013; Varela et al . , 2014 ) , we also observed independent duplication of IFIT genes in fish , as well as in birds and monotremes ( Figure 3 and Figure 3—figure supplement 1 ) . Based on these analyses , we infer that the human-like IFIT repertoire ( IFIT1 , 1B , 2 , 3 and 5 ) was established when placental mammals diverged from non-placental mammals approximately 100–200 mya while other vertebrates have undergone lineage-specific expansions of IFITs . 10 . 7554/eLife . 14228 . 009Figure 3 . Widespread gene birth , gene loss and gene conversion in vertebrate IFIT genes . Summary table of the IFIT gene repertoire of 51 vertebrate species ( see complete phylogeny in Figure 3—figure supplement 1 as well as expanded regions of trees built from the 5' and 3' end of the IFIT gene alignment in Figure 3—figure supplements 2 and 3 ) . Colored boxes indicate the presence of a given IFIT gene sequence , with multiple copies indicated as a number to the right of the colored box . Lightened boxes indicate that only a partial or pseudogenized copy of the gene can be found in the genome . On the phylogenetic tree to the left , X's indicate deletion or pseudogenization of either IFIT1 or IFIT1B genes from that lineage ( see Figure 3—figure supplement 4 for primate examples ) . Also shown are predicted occurrences of gene conversion ( red-around-blue circles ) between IFIT1 and IFIT1B sequences based on discordance between phylogenetic trees generated from 3' or 5' regions of the genes ( for carnivore examples see Figure 3—figure supplement 2 ) . Accession numbers for all sequences are found in Supplementary file 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 14228 . 00910 . 7554/eLife . 14228 . 010Figure 3—figure supplement 1 . Phylogeny of vertebrate IFIT genes . Maximum likelihood phylogenetic tree of over 200 IFIT genes from 51 vertebrate species . Bootstrap values greater than 80% are shown along the supported branch . An ancestral IFIT was duplicated multiple times early in mammalian evolution , resulting in many species having IFIT1 , IFIT1B , IFIT2 , IFIT3 and IFIT5 . As in Figure 1B , IFIT1 and IFIT1B genes cannot be resolved into distinct phylogenetic clades based on the full-length gene alignment . DOI: http://dx . doi . org/10 . 7554/eLife . 14228 . 01010 . 7554/eLife . 14228 . 011Figure 3—figure supplement 2 . Recurrent gene conversion is restricted to the 5' ends of IFIT genes . GARD and SBP analyses were run on the complete IFIT gene phylogeny shown in Figure 3—figure supplement 1 . These analyses indicate recurrent gene conversion between IFIT1 and IFIT1B paralogs 5' to the breakpoint indicated in red , while there is no evidence for gene conversion in the 3' end of the genes . Below are portions of the maximum likelihood phylogenetic tree of the recombining 5' region ( left ) and the non-recombining 3' region ( right ) of vertebrate IFITs highlighting only the region of the tree showing carnivore IFIT1 and IFIT1B gene sequences . Bootstrap values greater than 80% are shown along the supported branch . DOI: http://dx . doi . org/10 . 7554/eLife . 14228 . 01110 . 7554/eLife . 14228 . 012Figure 3—figure supplement 3 . Primate IFIT1 and IFIT1B genes . A portion of the maximum likelihood phylogenetic tree of the non-recombining 3' region of vertebrate IFITs is shown as in Figure 3 supplement 2 . Bootstrap values greater than 80% are shown along the supported branch . Primate IFIT1 and IFIT1B gene phylogenies are expanded , with pseudogenization of chimpanzee and New World monkey ( marmoset and squirrel monkey ) IFIT1B genes by frameshift and nonsense mutations ( see Figure 3—figure supplement 4 ) indicated by red-strikethroughs . Subsequent analyses suggest that human IFIT1B also lacks activity , although it encodes an intact open reading frame . DOI: http://dx . doi . org/10 . 7554/eLife . 14228 . 01210 . 7554/eLife . 14228 . 013Figure 3—figure supplement 4 . Pseudogenization of IFIT1B genes from several primate lineages . Location of frameshift and nonsense mutations in pseudogenized primate IFIT1B genes are shown . Gene sequences were aligned to intact human IFIT1B and frameshift-inducing insertions or deletions ( red triangles ) or nonsense mutations ( red asterisks ) are shown in the corresponding position in the sequence . DOI: http://dx . doi . org/10 . 7554/eLife . 14228 . 013 Consistent with our smaller sample , we found that IFIT1 and IFIT1B could not be resolved using the whole gene sequence , whereas IFIT2 , IFIT3 and IFIT5 could be easily separated into distinct monophyletic lineages . To formally test for recombination breakpoints between IFIT1 and IFIT1B , we analyzed the complete set of full-length IFIT genes with GARD and SBP ( Kosakovsky Pond et al . , 2006 ) , two tools that determine whether there is discordance between phylogenetic trees built from different regions of an alignment . Both analyses found strong statistical support ( GARD p-value <0 . 001 , SBP 100% model averaged support ) for a recombination breakpoint approximately two-thirds of the way through the IFIT gene sequence ( Figure 3—figure supplement 2 ) . Importantly , performing the same analyses for the 3' region of the IFIT alignment found no evidence for additional recombination breakpoints ( GARD p-value >0 . 1 , SBP 0% model averaged support ) , supporting the assertion that these regions have not recombined during mammalian IFIT evolution . Indeed , construction of a phylogenetic tree of this non-recombining region of vertebrate IFITs shows that IFIT1 and IFIT1B genes cluster into distinct monophyletic clades ( Figure 3—figure supplement 2 and supplement 3 ) . In contrast , a phylogenetic tree created from just the 5' end of vertebrate IFITs reveals numerous instances of gene conversion , including at least five separate cases in the carnivores alone ( Figure 3—figure supplement 2 ) . With this enhanced ability to assign IFIT genes to distinct IFIT1 or IFIT1B families after separating the confounding effects of gene conversion , we were able to detect lineage-specific duplication and loss events . For instance , in primates , we found that IFIT1 and IFIT1B genes have diverged as expected based on primate phylogeny and that all primates encode an intact IFIT1 . In contrast , we found that IFIT1B has been pseudogenized at least twice in the primate lineage as the result of the introduction of nonsense codons and frameshift mutations , once in chimpanzees and again in both sequenced New World monkeys ( Figure 3—figure supplement 3 and supplement 4 ) . Extending our surveys across all 51 sampled vertebrate species , we found numerous instances of IFIT gene birth , gene loss and gene recombination ( Figure 3 ) . In total , we found evidence for at least seven independent instances of IFIT1B loss , two separate instances of IFIT1 loss , and at least 13 independent instances of IFIT1/1B gene conversion . Interestingly , it appears that most changes in the gene repertoire of IFITs across mammalian species are focused on IFIT1 and IFIT1B , although there are also instances of lineage-specific gene births and loss in other IFIT genes as well . For example , IFIT3 has duplicated in mice , whereas IFIT5 has been lost in several rodents ( which may have occurred coincident with IFIT1 loss ) . These changes have led to a significant evolutionary turnover in the IFIT gene repertoire of different mammalian lineages , prompting us to investigate next the functional consequences of this evolutionary turnover of IFIT1/ IFIT1B paralogs in mammals . The ancient duplication and retention of IFIT1 and IFIT1B in most mammalian lineages suggested that they have non-redundant functions . Based on our phylogenetic findings , in addition to previous data suggesting human IFIT1 and mouse IFIT1B inhibit different viruses ( Andrejeva et al . , 2013; Daffis et al . , 2010; Pichlmair et al . , 2011; Pinto et al . , 2015 ) , we hypothesized that IFIT1 and IFIT1B might have evolved an alternative means of differentiating 'self versus non-self' and distinct antiviral specificities . We , therefore , investigated whether IFIT1 and IFIT1B proteins have distinct molecular discrimination properties . Previous characterizations of mouse IFIT1B clearly established its role in recognizing the methylation status of 5' cap structure of mRNAs as a way to distinguish 'self' , or host , mRNAs from 'non-self' , or viral , mRNAs ( Daffis et al . , 2010; Diamond , 2014; Fensterl and Sen , 2015; Hyde and Diamond , 2015 ) ( Figure 4A ) . However , most previous studies investigating mRNA cap requirements for IFIT1B-mediated growth inhibition have been carried out in the context of a virus infection under interferon-induced conditions . These approaches to understanding the function of individual IFIT proteins are potentially complicated by the induction of hundreds of other ISGs , as well as the fact several IFIT paralogs have been reported to bind to each other and function in a heterooligomeric complex ( Habjan et al . , 2013; Pichlmair et al . , 2011 ) . 10 . 7554/eLife . 14228 . 014Figure 4 . A yeast genetic assay recapitulates IFIT1B molecular specificity . ( A ) Schematic of mRNA cap structures and the proposed role of mouse IFIT1B in recognition and inhibition of mRNAs containing cap0-structures . Methylation of the first transcribed nucleotide of mRNAs , represented by an 'N' , is performed by cap1-methyltransferases , such as human CMTR1 ( hCMTR1 ) . However , cap1-methyltransferase activity is absent in budding yeast . Mouse IFIT1B binds and inhibits translation of cap0-mRNAs , but its activity is blocked when mRNAs contain a cap1-structure . ( B ) A budding yeast growth assay for IFIT1B function . Since budding yeast lack a cap1-methyltransferase , their mRNAs are predicted to be targeted by mouse IFIT1B . Shown are 10-fold serial dilutions of yeast spotted on media in which the indicated IFIT is either not expressed ( Glucose ) or expressed ( Galactose ) . Rows marked with '-' indicate an empty vector control . ( C ) Predicted role for hCMTR1 in blocking IFIT1B-mediated growth inhibition . ( D–E ) Yeast growth inhibition assays as in part B , but with galactose-induced co-expression of the indicated methyltransferase ( cap1-methyltransferase ( hCMTR1 ) wildtype or catalytic mutant ( K239A ) ) . In these panels , both IFIT and hCMTR1 expression are only induced upon grown on galactose-containing media . Protein expression data corresponding to these yeast strains grown in galactose are shown in Figure 4—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 14228 . 01410 . 7554/eLife . 14228 . 015Figure 4—figure supplement 1 . Expression of IFIT1B and cap1-methyltransferase proteins in yeast . Western blots of yeast described in Figure 4 . Lanes marked with '-' indicate an empty vector control . Shown are blots probed with anti-Flag antibody ( for IFIT1B expression ) , anti-hCMTR1 ( for cap1-methyltransferase expression ) and anti-PGK1 ( as a loading control ) . The location of molecular weight markers ( in kiloDaltons ) are indicated on the right of the Western blots . DOI: http://dx . doi . org/10 . 7554/eLife . 14228 . 015 We , therefore , wished to assay the specificity of IFIT1 and IFIT1B molecular discrimination in an orthogonal system . To this end , we turned to the budding yeast Saccharomyces cerevisiae , whose mRNAs only have cap0-structures ( Banerjee , 1980 ) due to a lack of a cap1-methyltransferase . Based on this absence of a cap1-methyltransferase , we hypothesized that expression of mouse IFIT1B in budding yeast might inhibit yeast growth just as IFIT1B inhibits replication of viruses with cap0-mRNAs . Indeed , using a galactose-induced IFIT1B expression system , we found that expression of mouse IFIT1B potently inhibits yeast growth ( Figure 4B ) . Importantly , this growth inhibition occurs in the absence of any other IFIT or component of the innate immune system , indicating that mouse IFIT1B does not require any other mammalian-specific cofactor to perform this function . We next hypothesized that introducing the human cap1-methyltransferase hCRMT1 to yeast would rescue yeast growth by blocking the activity of IFIT1B ( Figure 4C ) . Strikingly , we observed nearly complete rescue of yeast growth when we co-expressed the human cap1-methyltransferase ( hCMTR1 ) , but not a catalytic mutant of hCMTR1 ( K239A ) ( Belanger et al . , 2010 ) ( Figure 4D ) . We , therefore , conclude that the mouse IFIT1B-mediated growth inhibition of budding yeast is specifically due to the lack of a functional cap1-methyltransferase in yeast , accurately recapitulating the known antiviral molecular specificity of IFIT1B ( Figure 4C ) . Based on our phylogenetic predictions , we next investigated whether other members of the IFIT1B gene family have similar properties as the well-characterized mouse IFIT1B . If so , we predicted that they would also cause growth inhibition of yeast , but be rescued by overexpression of the human cap1-methyltransferase . To test this hypothesis , we expressed IFIT1B from several primate species with or without co-expression of hCMTR1 ( Figure 4E ) . Consistent with our phylogenetic prediction , we observed robust growth inhibition as a result of expression of gibbon and African green monkey ( AGM ) IFIT1B that could be rescued by expression of hCMTR1 . Thus , despite the long divergence separating mouse and primate IFIT1B genes , they share molecular discrimination properties . Surprisingly , we observed no growth inhibition by human IFIT1B . This lack of activity even upon equivalent expression in yeast ( Figure 4—figure supplement 1 ) is unusual among intact primate IFIT1Bs , suggesting that human IFIT1B lacks function . This result is consistent with previous suggestions ( Fensterl and Sen , 2011 ) that human IFIT1B may be non-functional even though it appears to be encoded by an intact open reading frame . We next tested our hypothesis that IFIT1 has evolved a distinct molecular function from that of IFIT1B . Similar to mouse IFIT1B , we found that expression of human IFIT1 resulted in a potent inhibition of yeast growth ( Figure 5A ) . However , in contrast to mouse IFIT1B , we observed no rescue of human IFIT1-mediated growth inhibition upon overexpression of hCMTR1 ( Figure 5B ) . IFIT1 proteins from gibbon and African Green Monkey ( AGM ) recapitulate the findings from human IFIT1 ( Figure 5C ) . These data indicate that IFIT1 proteins do not distinguish between absence or presence of a cap1-methyltransferase ( Figure 5D ) but instead , likely recognize another as-yet-undefined 'non-self' molecular pattern , other than Cap0-mRNA , which is present in yeast . 10 . 7554/eLife . 14228 . 016Figure 5 . IFIT1-mediated growth inhibition is not relieved by a cap1-methyltransferase . ( A–C ) Yeast growth inhibition assays as in Figure 4 with the indicated IFIT1 . ( D ) In contrast to IFIT1B , IFIT1 inhibits yeast growth regardless of whether cap1-methyltransferase is present . ( E ) Pairwise comparison of the amino acid sequence of cat IFIT1 and IFIT1B , with black vertical lines indicating differences between the two protein sequences . Shown below are the pairwise amino acid identities for the N- and C-terminal regions of the alignment . ( F ) Yeast growth inhibition assays as in Figure 4 with cat IFIT1 or IFIT1B . In these panels , both IFIT and hCMTR1 expression are only induced upon grown on galactose-containing media . Protein expression data corresponding to these yeast strains grown in galactose are shown in Figure 5—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 14228 . 01610 . 7554/eLife . 14228 . 017Figure 5—figure supplement 1 . Expression of IFIT1 proteins in yeast . Western blots of yeast described in Figure 5 . Lanes marked with '-' indicate an empty vector control . Shown are blots probed with anti-Flag antibody ( for IFIT1 or IFIT1B expression ) , anti-hCMTR1 ( for cap1-methyltransferase expression ) and anti-PGK1 ( as a loading control ) . The location of molecular weight markers ( in kiloDaltons ) are indicated on the right of the Western blots . DOI: http://dx . doi . org/10 . 7554/eLife . 14228 . 017 These yeast genetic data indicate that human IFIT1 and mouse IFIT1B have evolved distinct molecular functions . As described above , resolution of IFIT1 and IFIT1B genes into distinct monophyletic clades required considering only the non-recombining 3' end of the genes ( Figure 2 ) . We , therefore , asked whether differences in the C-terminal end of the IFIT1 and IFIT1B were responsible for changes in their molecular specificity , as suggested by our phylogenetic analyses . Our initial chimeras between mouse IFIT1B and human IFIT1 , or gibbon IFIT1 and IFIT1B , were unable to recapitulate the yeast growth inhibition . However , this result may be expected , as each of these proteins has diverged without gene conversion for >30 million years and may have acquired epistatic interactions between the N-terminus and the C-terminus . As an alternate strategy to test whether the C-terminus is responsible for differences between IFIT1 and IFIT1B function , we turned to a natural pair of IFIT1/IFIT1B proteins in which the N-termini had recently undergone gene conversion . Cat IFIT1 and IFIT1B are 96% identical at the amino acid level in the N-terminus of the protein whereas the C-terminus is only 46% identical ( Figure 5E ) . Upon overexpression in yeast , both cat IFIT1 and IFIT1B were capable of causing growth inhibition ( Figure 5F ) . However , only inhibition by cat IFIT1B was rescued by hCMTR1 , suggesting that the C-terminus of the IFIT1 paralogs is critical for the differences in the molecular specificity of these two proteins . We , therefore , conclude that IFIT1 and IFIT1B gene families perform distinct molecular functions in the species in which they are found and that recurrent gene conversion of the N-termini of IFIT1/IFIT1B genes in mammals has not led to a loss of their separate functions . We showed that IFIT1 and IFIT1B proteins are phylogenetically distinct and are functionally distinct in their dependence on cap1-methyltransferase in our yeast assay . Next , we hypothesized that they have non-overlapping antiviral activities . To test this hypothesis , we created cell lines that constitutively overexpress different IFIT1 or IFIT1B genes ( Figure 6A ) , circumventing the need to stimulate IFIT expression with interferon and allowing us to evaluate different IFIT1 or IFIT1B gene functions in an isogenic background . We then tested whether either human IFIT1 or mouse IFIT1B can specifically inhibit replication of a virus lacking a functional cap1-methyltransferase . For these experiments , we compared replication of a wildtype ( cap1-mRNA expressing ) vaccinia virus to replication of a vaccinia virus that has a catalytic mutation in its cap1-methyltransferase ( Latner et al . , 2002 ) . This assay allowed us to test whether cap1-methyltransferase could enable a virus to block IFIT-mediated antiviral activity ( Figure 6B ) . Consistent with previous studies with these same viruses ( Daffis et al . , 2010 ) , we found that expression of mouse IFIT1B can potently inhibit replication of a mutant vaccinia virus that lacks cap1-methyltransferase activity , but was unable to inhibit replication of wildtype vaccinia virus ( Figure 6C ) . While the previous mouse knockout experiment had demonstrated that mouse IFIT1B is necessary for inhibition of viruses encoding cap0-mRNAs ( Daffis et al . , 2010 ) , our findings show that singular overexpression of mouse IFIT1B is also sufficient for such inhibition . We next extended our analyses to other members of the IFIT1 and IFIT1B gene families to determine if cap0-mRNA dependence was a conserved feature of their antiviral activity . Consistent with our yeast assay data , we found that gibbon and AGM IFIT1B proteins function similarly to mouse IFIT1B , and can specifically inhibit the replication of a virus expressing cap0-mRNAs ( Figure 6D ) . Also consistent with our analyses of IFIT1B in yeast ( Figure 4E ) , human IFIT1B lacks antiviral activity despite being expressed in these cell lines . These results suggest that most IFIT1B proteins ( except for the putatively non-functional human IFIT1B ) share cap0-mRNA recognition as a discriminant of their antiviral function . 10 . 7554/eLife . 14228 . 018Figure 6 . IFIT1B , but not IFIT1 , inhibits replication of a cap0-containing virus . ( A ) Cell lines were generated that express human IFIT1 or mouse IFIT1B , as assessed by western blot . ( B ) Predicted role for vaccinia virus cap1-methyltransferase ( J3 ) in blocking IFIT1B-mediated viral inhibition . ( C ) Cell lines expressing the indicated IFIT protein were infected with either wildtype ( left ) or cap1-methlytransferase mutant ( J3 K175A , right ) vaccinia virus at an MOI of 0 . 01 . Virus was harvested at the indicated time point and titered . Error bars represent standard deviation of three biological replicates . ( D ) As in panel C , but only showing data 24 hr post-infection with wildtype ( left ) or the cap1-methyltransferase mutant ( right ) vaccinia virus . Shown below are protein expression levels from uninfected cell lines . DOI: http://dx . doi . org/10 . 7554/eLife . 14228 . 018 In contrast to the activity of most IFIT1B proteins , human IFIT1 did not inhibit replication of the vaccinia virus methyltransferase mutant ( Figure 6C ) . Consistent with our yeast data , these results demonstrated that human IFIT1 does not discriminate 'self versus non-self' based on cap0-mRNA versus cap1-mRNA specificity . Likewise , neither gibbon nor AGM IFIT1 was able to inhibit the cap0-mRNA expressing vaccinia virus ( Figure 6D ) . We , therefore , conclude that in keeping with their phylogenetic divergence , IFIT1B and IFIT1 lineages evolved to rely on distinct cues for their function . Such non-redundancy could explain the retention of both IFIT1 and IFIT1B in most mammalian genomes . These results also suggest that there likely exist viruses that are specifically inhibited by IFIT1 but not IFIT1B . To identify such viruses , we took advantage of previous studies , which showed that knockdown of human IFIT1 can augment replication of several viruses expressing a functional cap1-methyltransferase ( Andrejeva et al . , 2013; Pichlmair et al . , 2011 ) . Based on our findings and others ( Daffis et al . , 2010; Hyde et al . , 2014; Ma et al . , 2014; Menachery et al . , 2014; Szretter et al . , 2012; Zust et al . , 2011; Habjan et al . , 2013; Kimura et al . , 2013; Kumar et al . , 2014 ) we expect that all cap1-mRNA expressing viruses would be resistant to IFIT1B restriction . We , therefore , infected our cell lines expressing either human IFIT1 or mouse IFIT1B with vesicular stomatitis virus ( VSV ) , a virus known to express mRNAs with a cap1-structure ( Ma et al . , 2014; Banerjee , 1980 ) . As expected , we found that mouse IFIT1B had no effect on either virally-encoded GFP expression ( Figure 7A ) or viral titers ( Figure 7B ) . VSV viral titers were also unaffected by other IFIT1B genes ( Figure 7C ) . In contrast , we observed a substantial decrease in both viral-GFP expression and viral titer in cells expressing human IFIT1 ( Figure 7A , B ) . This inhibitory activity was also conserved in other primate IFIT1 proteins ( Figure 7C ) , suggesting that multiple IFIT1 proteins are insensitive to cap1-methyltransferase and share the property of inhibiting VSV ( Figure 7D ) . Together with our vaccinia virus data ( Figure 6 ) , these data support the conclusion that IFIT1B and IFIT1 encode non-redundant antiviral specificities that together expand the antiviral range of the IFIT gene repertoire in mammals . 10 . 7554/eLife . 14228 . 019Figure 7 . IFIT1 , but not IFIT1B , inhibits replication of VSV , a cap1-containing virus . ( A ) Cell lines were infected with vesicular stomatitis virus ( VSV ) encoding a GFP protein at an MOI of 0 . 01 . After 12 hr , cells were visualized for expression of the virally-encoded GFP . ( B ) Infections were performed as in part A , but at the indicated timepoint , virus was harvested and titered . Error bars represent standard deviation of three biological replicates . ( C ) Infections were performed as in part A , harvested after 12 hr , and titered . Error bars represent standard deviation of three biological replicates . ( D ) In contrast to IFIT1B , IFIT1 inhibits VSV growth arrest regardless of the fact that cap1-methyltransferase is present . DOI: http://dx . doi . org/10 . 7554/eLife . 14228 . 019
Rapid gene evolution is a hallmark of antiviral proteins . Immunity proteins are driven to innovate continually to maintain effective defenses against an evolving barrage of pathogens ( Daugherty and Malik , 2012 ) . One common genetic mechanism of innovation employed by host genomes is gene duplication , which enables hosts to either escape viral antagonism or evolve new inhibitory mechanisms through subfunctionalization . Indeed , many antiviral genes expressed as part of the interferon response represent members of multigene families , whose lineage-specific expansions and contractions have shaped contemporary repertoires of antivirals in mammalian genomes ( Brunette et al . , 2012; Daugherty and Malik , 2012; Munk et al . , 2012; Tareen et al . , 2009; Zhang et al . , 2012 ) . In several of these multigene families , gene conversion has further diversified the innate immune repertoire of different species ( Buchmann , 2014; Mitchell et al . , 2015; Woelk et al . , 2007 ) . Despite being among the most highly expressed interferon-stimulated genes ( ISGs ) , the function of the IFIT antiviral genes is still incompletely understood . Previously , several IFITs , including mouse IFIT1B ( previously IFIT1 ) and human IFIT1 , were shown to inhibit protein synthesis ( Guo et al . , 2000; Hui et al . , 2003; Terenzi et al . , 2005 ) . These findings are consistent with mRNA translation control being a major axis of regulation of viral replication ( Mohr and Sonenberg , 2012; Li et al . , 2015 ) . Furthermore , the finding that mouse IFIT1B represses the translation of mRNAs lacking Cap1-structures appeared to resolve the conundrum of how at least this IFIT protein discriminates between 'self' and 'non-self' ( Daffis et al . , 2010 ) . However , the broad application of this result for IFIT1 function was complicated by apparently contradictory findings that mouse and human representatives of IFIT1 have different antiviral specificities ( Andrejeva et al . , 2013; Daffis et al . , 2010; Pichlmair et al . , 2011; Pinto et al . , 2015 ) . Thus , it remained unclear how different IFIT proteins recognize and inhibit replication of different viruses , and whether and how the changes in IFIT gene composition between species ( e . g . humans and mice ) have altered their antiviral repertoire . To provide an evolutionary framework for understanding how the IFIT antiviral repertoire evolved in mammals , we performed in-depth phylogenetic analyses . We find unambiguous evidence that , although they were initially assumed to be orthologs , mouse IFIT1B ( previously mouse IFIT1 ) and human IFIT1 represent paralogous genes that diverged close to the origin of placental mammals . We further showed that the IFIT gene family , especially IFIT1 and IFIT1B genes , have undergone recurrent bouts of gene duplication , gene loss , and gene conversion . These processes have resulted in a wide diversity of IFIT genes across mammalian species . Such changes in IFIT gene composition might be expected to affect the range of viruses that can be inhibited by IFITs in different mammalian genomes . Based on their long divergence , we predicted that mouse ( and other ) IFT1B and human ( and other ) IFIT1 protein might possess different antiviral specificities . Consistent with this prediction , our data indicate that only IFIT1B proteins distinguish 'self from non-self' mRNAs by recognizing an unmethylated ( cap0- ) mRNA structure , a molecular pattern that is absent on 2'O-methylated ( cap1- ) host mRNAs . However , the general antiviral effectiveness of IFIT1B proteins is limited due to the fact that every mammalian virus family , except alphaviruses , has evolved a way to produce cap1-mRNAs . Viral cap1-mRNAs are produced either by nuclear transcription and use of the host nuclear capping machinery ( e . g . , retroviruses and many DNA viruses ) , or cap-snatching ( e . g . , orthomyxoviruses ) , or virally encoding cap1-methyltransferases ( e . g . , poxviruses and rhabdoviruses ) ( Banerjee , 1980; Decroly et al . , 2012; Hyde and Diamond , 2015 ) . Even in the case of alphaviruses , which produce cap0-mRNAs ( Banerjee , 1980 ) and are therefore predicted to be susceptible to IFIT1B-mediated inhibition , secondary structure changes at the 5' end of alphaviral mRNAs can blunt IFIT1B antiviral action ( Hyde et al . , 2014 ) . Given all of these viral counter-strategies , it might appear that IFIT1B is on the losing side of the arms race between host and most viruses . These counter-strategies may partially explain why we find that IFIT1B has been deleted or pseudogenized at least seven separate times in mammalian evolution . Even in humans , which encode a full-length IFIT1B ORF , IFIT1B activity appears to have been lost both due to lack of interferon-inducibility ( Fensterl and Sen , 2011 ) as well as loss of protein function ( Figure 4E and 6D ) . Humans and most other mammalian species ( excluding rodents ) do , however , have an intact member of the IFIT1 gene family . IFIT1 mechanism and antiviral specificity have been enigmatic in part due to incorrect comparisons with mouse IFIT1B . By experimentally separating the functions of IFIT1 and IFIT1B from the rest of the innate immune system , we now show that IFIT1 has evolved a molecular specificity that differs from IFIT1B and is not blocked by cap1-methyltransferase . For example , using a yeast genetic assay , we show that IFIT1 , unlike IFIT1B , potently inhibits yeast growth independent of whether human cap1-methyltransferase is co-expressed . Moreover , IFIT1 , but not IFIT1B , inhibits replication of VSV , a virus encoding a cap1-methyltransferase . These results establish that these two paralogous immunity factors have distinct molecular function and highlight a significant role for IFIT1 in specific antiviral defense that is distinct from IFIT1B . Our study does not elucidate the molecular means by which human IFIT1 inhibits viral replication . While several possibilities exist for how IFIT1-mediated translational repression might be controlled , we favor the possibility that IFIT1 , like its paralog IFIT1B , distinguishes a 'self versus non-self' pattern on mRNA to selectively inhibit viral replication . Supporting this model is the fact that in addition to IFIT1B , IFIT2 and IFIT5 have been reported to have sequence- or structure-specific binding to RNA ( Habjan et al . , 2013; Katibah et al . , 2013; 2014; Yang et al . , 2012 ) , suggesting that numerous IFITs may restrict viral replication through recognition of distinct viral RNA patterns . Although the presumed molecular pattern that IFIT1 recognizes remains unknown , our data suggest that mammalian hosts , as well as vaccinia virus , possess a 'self' molecular pattern to prevent IFIT1-mediated inhibition whereas yeast and VSV display a 'non-self' molecular pattern that cannot block IFIT1 action . One previous proposal suggested that human IFIT1 antiviral activity resulted from direct binding and sequestration of 5' triphosphate ends of viral replication intermediates , thus interfering with replication rather than directly inhibiting translation ( Pichlmair et al . , 2011 ) . We instead favor a model in which IFIT1 recognizes a distinct 'self versus non-self' pattern to inhibit mRNA translation directly , similar to IFIT1B . Biochemical data suggests that IFIT1 , like IFIT1B , binds to cap-proximal RNA and prevents binding of mRNAs by the translation initiation factor eIF4F ( Kumar et al . , 2014 ) . Moreover , human IFIT1 was shown to not just inhibit replication of PIV5 , but to inhibit translation of PIV5 2'O-methylated mRNAs ( Andrejeva et al . , 2013 ) . Finally , our observation that IFIT1 expression also inhibits yeast growth suggests that it is not a viral replication intermediate , but rather a 'non-self' pattern , which is recognized by IFIT1 , similar to the lack of mRNA 2'O-methylation for IFIT1B . While the methylation state of the first transcribed nucleotide ( cap0-structure versus cap1-structure ) is not the determinant of IFIT1 specificity , other chemical modifications near the cap , or mRNA sequence determinants , may allow IFIT1 to distinguish 'self' from 'non-self' to selectivity inhibit viral replication . Although gene conversion and gene turnover initially confounded IFIT phylogenetic analysis , it may now present a significant opportunity to understand the biochemical basis of the different antiviral specificities of IFIT1 and IFIT1B . For instance , two cat IFIT proteins , which are practically identical in their N-terminal two-thirds , but divergent in their C-terminus , fully recapitulate the expected properties of IFIT1 and IFIT1B in our yeast assay . Use of other such ‘natural’ chimeras can help guide the biochemical dissection of what provides IFIT1B with its unique cap0-mRNA recognition properties and provide insight into how IFIT1 has evolved its distinct specificity . The naturally occurring recombination breakpoint in the IFIT1/1B gene conversion tracts also suggests features of IFITs that are critical for a common aspect of IFIT1/1B activity versus those that promote their distinct molecular specificity . In the structures of IFIT5 bound to the 5' triphosphate containing end of RNA , three 'subdomains' surround the RNA ( Abbas et al . , 2013 ) . The recombination breakpoint we observed falls within the 'pivot' region separating the second and third subdomain ( Figure 2—figure supplement 2 ) , suggesting the first two subdomains may be important for a common conserved function of IFIT1 and IFIT1B whereas the third subdomain might confer changes to the molecular specificity . We also note that the breakpoint occurs just outside of the homooligomerization domain in the structure of IFIT2 ( Yang et al . , 2012 ) . As some IFITs are thought to oligomerize for function ( Habjan et al . , 2013; Pichlmair et al . , 2011 ) , the homogenization of the region corresponding to the oligomerization domain in IFIT2 suggests that IFIT1 and IFIT1B might also need to hetero-oligomerize in species in which both are present . Based on our new understanding of the divergent functions of IFIT1 and IFIT1B , we hypothesize that duplication or loss of different IFIT genes might directly influence the ability of a particular host species to defend against specific viral pathogens . For instance , the duplication of IFIT1B genes in rodents might have been driven by pressure from alphaviruses or other undiscovered cap0-mRNA expressing viruses , and may augment the defenses of rodent hosts against such viruses . In contrast , the loss of IFIT1 in many rodents might leave them more susceptible to infection by viruses inhibited by IFIT1 , such as VSV . Similarly , loss of IFIT1B function in a broad range of species , including humans , may result in greater ability of cap0-mRNA expressing viruses to replicate in those species . In that regard , rodents may be better prepared than humans , chimpanzees , New World monkeys and several other species to defend against cap0-mRNA expressing viruses such as alphaviruses ( Figure 3 ) . Thus , IFIT gene duplication may be the host's response to the evolutionary arms race where viruses are continually masking their RNAs with 'self' features . On the other hand , when the selective pressure from viruses displaying a given 'non-self' feature ( e . g . Cap0-mRNAs ) has been relaxed , the specific IFIT gene that restricts those viruses may be pseudogenized or lost permanently . Although compensatory mechanisms may have evolved that mitigate the consequences of loss of a single IFIT , our results indicate that lineage-specific loss of IFIT genes eliminates an elegant means to discriminate ‘self versus non-self’ RNAs from the host's armament . We predict that the dynamic evolution of IFIT genes across mammals will have important consequences for species-specific antiviral immunity .
Mammalian IFIT genes were identified in genomes of the indicated species using human IFIT protein sequences to query the non-redundant ( nr ) database using tBLASTn ( Altschul et al . , 1997 ) ( see Supplementary file 1 for accession numbers ) . In human IFIT genes , the entire protein-encoding sequence except for the N-terminal methionine is encoded on a single exon . For the species indicated in Figure 1 , we used this information to include only IFIT sequences that were found on a single uninterrupted region of DNA in either NCBI or UCSC genome databases . For the broader panel of species shown in Figures 2 and 3 , several of the genomes are not as well assembled and we therefore did not eliminate genes based on these criteria but insisted that >90% of the gene sequence be available in NCBI . IFIT sequences were aligned based on their translated sequence using MUSCLE ( Edgar , 2004 ) implemented in Geneious ( Kearse et al . , 2012 ) . All alignments were manually curated using Geneious ( Kearse et al . , 2012 ) . Maximum likelihood phylogenetic trees of IFIT nucleotide sequences were generated using the HKY85 substitution model in PhyML ( Guindon et al . , 2010 ) using 1000 bootstrap replicates for statistical support . K-estimator ( Comeron , 1999 ) was to calculate the rate of synonymous change ( dS ) for each 200 nt window of pairwise comparisons of IFIT1 and IFIT1B genes . To examine the alignments for evidence of recombination breakpoints , we used the SBP and GARD algorithms implemented at DataMonkey . org ( Kosakovsky Pond et al . , 2006 ) . Pairwise percent identity calculations and graphical representations were made using Geneious ( Kearse et al . , 2012 ) . Phylogenetic trees were visualized using FigTree ( http://tree . bio . ed . ac . uk/software/figtree/ ) . Protein structural figures were generated using PyMol ( https://www . pymol . org ) . Previous studies have utilized budding yeast as a genetic assay system for PKR-mediated mRNA translation inhibition by showing that PKR expression in yeast arrests growth ( Dever et al . , 1993 ) . We therefore asked whether expression of IFIT genes in yeast would also arrest growth . IFIT genes with an N-terminal 3xFlag tag were cloned downstream of the Gal1-10 promoter in the Cen-based plasmid , p413 , using primers described in Supplementary file 2 . Yeast ( strain BY4741 ) were transformed and selected on synthetic complete media lacking histidine and containing 2% glucose ( SC -his GLU ) . For expression of methyltransferases and IFITs , human methyltransferases were integrated into strain BY4741 to replace the His3 gene and then transformed with p413 plasmids expressing IFITs . First , the human genes for hCMTR1 or hCMTR1 K239A methyltransferases were cloned downstream of the Gal1-10 promoter of pRS305-Gal1 using primers described in Supplementary file 2 . The resulting plasmids were used as templates to amplify the entire region spanning the Gal promoter to the Leu2 gene using primers described in Supplementary file 2 with 70 bp homology to the genomic regions flanking the His3 gene . The resulting PCR products were transformed into strain BY4741 and yeast were selected on media lacking leucine ( SC -leu GLU ) . Yeast with integrated galactose-inducible methyltransferases were subsequently transformed with IFIT genes as described above and selected on media lacking both leucine and histidine ( SC -leu/-his GLU ) . All yeast plating assays were performed on selective media with either 2% glucose ( for uninduced controls ) or 2% galactose ( for induction of IFIT and methyltransferase genes ) . A single yeast colony was picked from a freshly streaked plate of transformed yeast and grown overnight at 30 degrees in liquid media ( SC -his GLU or SC -his/-leu GLU ) . Saturated overnight cultures were serially diluted 10-fold from a starting OD600 of 1 and spotted onto selective plates containing glucose or galactose . Plates were incubated at 30 degrees for 48–72 hr . The retroviral packaging vector pQCXIP ( Clontech , Mountain View , CA ) was used as a backbone for generation of all stable cell lines . First , an insert containing mCherry followed by a T2A site followed by a 3xFlag tag was cloned downstream of the CMV promoter of pQCXIP , resulting in plasmid pMD143 ( see Supplementary file 2 ) . All IFIT genes were inserted in frame with the 3xFlag tag of pMD143 using primers and templates described in Supplementary file 2 . The resulting plasmids were used to generate VSV-g pseudotyped retroviruses that were then used to transduce BSC40 cells , an African green monkey derived kidney epithelial cell line . The BSC40 cells have been routinely screened for mycoplasma infections and validated as being of African Green monkey origin using RT-PCR analyses . We also note that BSC40 cells are not on the list of commonly misidentified cell lines maintained by the International Cell Line Authentication Committee . Stably transduced BSC40 cells were selected and maintained in culture media ( DMEM with 10% FBS ) containing 10 ug/ml puromycin . All cell lines were assayed within the first 20 passages following transduction . Vaccinia virus strain WR wildtype or cap1-methyltransferase mutant ( J3 K175R ) were a generous gift of R . Condit ( Latner et al . , 2002 ) . VSV-GFP was a generous gift of J . Rose ( Boritz et al . , 1999 ) . For infectivity assays , BSC40 cells stably transduced with pMD143 containing various IFIT gene inserts were seeded to 24-well plates ( ~50000 cells/well ) and allowed to grow overnight . Cells were incubated with vaccinia virus or VSV-GFP at a multiplicity of infection ( MOI ) of 0 . 01 for two hours , followed by a media change . Vaccinia virus was harvested at the indicated time post-infection by freeze/thaw disruption of cells . VSV-GFP was harvested from cell supernatant at the indicated time post-infection . Viruses were titered on BSC40 cells either with agar overlays ( for VSV-GFP ) or without ( for vaccinia virus ) . All infectivity data is reported as average of three biological replicates ( independent infections followed by titering ) plus and minus the standard deviation . The following commercially available primary antibodies were used: mouse M2 anti-Flag ( Sigma-Aldrich , St . Louis , MO; F1804 ) , rabbit anti-hCMTR1 ( Sigma-Aldrich; HPA029979 ) , mouse anti-PGK1 ( Invitrogen , Carlsbad , CA; 459250 ) , and rat anti-tubulin ( Millipore , Temecula , CA; CBL270 ) . Goat anti-mouse , anti-rat and anti-rabbit HRP-conjugated secondary antibodies were from Santa Cruz Biotechnology ( Dallas , TX ) . For analysis of protein expression from stably transduced BSC40 cells , ~200000 cells were harvested and lysed by boiling in 2x SDS sample buffer . For analysis of protein expression from yeast , a single yeast colony was picked from a freshly streaked plate of transformed yeast and grown overnight in liquid selective media containing 2% raffinose ( SC -his/-leu RAF ) . Cultures were diluted in SC -his/-leu RAF to 0 . 5 OD600 and grown 3 more hours at which point galactose was added to a final concentration of 2% . After 90 min , cultures were spun down and frozen . Cell pellets were resuspended in 2x SDS sample buffer containing protease inhibitors and bead beaten for 30 s . All cell lysates ( yeast and mammalian ) were run on 4–12% Bis-Tris gels ( Invitrogen ) and transferred onto nitrocellulose membrane . Blocking buffer and antibody dilution buffer for hCMTR1 blots was 5% bovine serum albumin in phosphate buffered saline ( PBS ) with 0 . 1% Tween-20 . Blocking buffer and antibody dilution buffer for all other blots was 5% nonfat dried milk in PBS with 0 . 1% Tween-20 . | When a virus is detected in the body , hundreds of different proteins in the immune system are rapidly produced as a first line of defense to limit the ability of the virus to multiply and spread . Many of these ‘innate’ immunity proteins have rapidly evolved in mammals in escalating molecular 'arms races' with the viruses they target . This makes it more difficult to work out exactly what job each protein performs . Even when the role of a specific protein has been determined in mice , for example , it does not always follow that the human protein with the same name will perform the same role . The IFIT1 proteins are some of the most highly produced innate immunity proteins in mammals during viral infections . In the infected cell , host and viral proteins are both made from templates made of molecules of ribonucleic acid ( RNA ) . Previous work showed that the IFIT1 protein in mice is able to exploit a critical chemical difference between host and virus RNA to selectively block the production of virus proteins . However , other research suggests that the human IFIT1 protein does not use the same chemical difference to distinguish between host and virus RNA . Here , Daugherty et al . unravel the complicated evolutionary history of IFIT1 proteins to show that the mouse and human proteins are not as closely related to each other as first thought . Instead , they belong to two different protein families with distinct roles in fighting viruses . Further experiments show that the human and mouse IFIT proteins likely discriminate between host and viral RNA using different cues , leading to their action against different sets of viruses . Daugherty et al . ’s findings highlight that there are additional undiscovered chemical differences between host and viral RNA that the immune system can exploit to selectively target and stop viruses from multiplying . Furthermore , these findings re-emphasize the often-overlooked differences between the immune systems of mice and humans . The finding that mammals have such a diverse set of IFIT1 immunity proteins may directly explain why different species are susceptible to different viruses . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"microbiology",
"and",
"infectious",
"disease"
] | 2016 | Evolution-guided functional analyses reveal diverse antiviral specificities encoded by IFIT1 genes in mammals |
Replication checkpoint is essential for maintaining genome integrity in response to various replication stresses as well as during the normal growth . The evolutionally conserved ATR-Claspin-Chk1 pathway is induced during replication checkpoint activation . Cdc7 kinase , required for initiation of DNA replication at replication origins , has been implicated in checkpoint activation but how it is involved in this pathway has not been known . Here , we show that Cdc7 is required for Claspin-Chk1 interaction in human cancer cells by phosphorylating CKBD ( Chk1-binding-domain ) of Claspin . The residual Chk1 activation in Cdc7-depleted cells is lost upon further depletion of casein kinase1 ( CK1γ1 ) , previously reported to phosphorylate CKBD . Thus , Cdc7 , in conjunction with CK1γ1 , facilitates the interaction between Claspin and Chk1 through phosphorylating CKBD . We also show that , whereas Cdc7 is predominantly responsible for CKBD phosphorylation in cancer cells , CK1γ1 plays a major role in non-cancer cells , providing rationale for targeting Cdc7 for cancer cell-specific cell killing .
Eukaryotic DNA replication depends on the formation of pre-Replicative Complex on chromatin during the G1 phase of cell cycle , which is mediated by assembly of Orc , Cdc6 , Cdt1 and Mcm proteins . Cdc7 kinase plays a crucial role in initiation of DNA replication by phosphorylating Mcm proteins , essential as a part of the replicative helicase ( Masai and Arai , 2002; Masai et al . , 2010; Labib , 2010 ) . Inhibition of DNA replication by hydroxyl urea ( HU ) or UV triggers cellular responses known as replication stress checkpoint ( Branzei and Foiani , 2009 ) . Conserved checkpoint kinases ( Mec1-Rad3-ATR ) are activated in response to replication stress , which ultimately activates Rad53-Cds1-Chk1 effector kinases , that inhibits progression of S phase as well as entry into M phase to reduce the genomic instability potentially caused by replication fork arrest . The stalled replication fork caused by replication stress results in activation of the ATR-ATRIP complex and its association with TopBP1 , another activator of ATR . Activated ATR phosphorylates Chk1 , the crucial step for activation of the key effector kinase . ATR-mediated phosphorylation of Chk1 requires Clapsin as a mediator/adaptor for the signal transfer . Indeed , ATR phosphorylates Chk1 in the Clapsin-Chk1 complex more efficiently than Chk1 alone in the absence of Claspin in virto ( Lindsey-Boltz et al . , 2009 ) . Cdc7 kinase has been implicated in replication checkpoint responses . In budding yeast , it is known that Rad53 checkpoint effector kinase directly phosphorylates Dbf4 upon fork stalling to block late origin firing most likely through inhibition of the Cdc7 kinase function ( Zegerman and Diffley , 2010; Lopez-Mosqueda et al . , 2010; Duch et al . , 2011 ) , although the mechanism of this regulation is not clear ( Weinreich and Stillman , 1999 ) . In mammalian cells , it was implicated in repair of stalled replication forks through Rad18 ( Day et al . , 2010 ) . In addition to its role in the effector phase of checkpoint , the role of Cdc7 kinase in checkpoint activation has been suggested . In fission yeast , activation of checkpoint kinase was impaired in cdc7 mutant cells ( Shimmoto et al . , 2009; Matsumoto et al . , 2010 ) . However , a possibility that the reduced number of active replication forks in these mutants is responsible for compromised checkpoint activation could not be ruled out ( Shimada et al . , 2002 ) . However , the impaired checkpoint activation in cdc7 bypass mutants ( ∆mcm4N ∆cdc7 in budding yeast and rif1∆ hsk1∆ in fission yeast; Sheu and Stillman , 2010; Hayano et al . , 2011; Ogi et al . , 2008; our unpublished data ) provided strong evidence that Cdc7 is required for checkpoint activation in yeasts . In mammalian cells , induced knockout of Cdc7 gene in mouse ES cells as well as siRNA-mediated inhibition of Cdc7 expression in cancer cells resulted in almost complete loss of Chk1 activation in response to HU or UV irradiation ( Kim et al . , 2008 ) . Later , requirement of Cdc7 for timely checkpoint activation in cancer cells was confirmed by using a compound that inhibits Cdc7 kinase ( Rainey et al . , 2013 ) . Thus , it is now well established that Cdc7 is required for replication checkpoint activation . Furthermore , replication stress-induced hyperphosphorylation of Claspin/Mrc1 exemplified by mobility-shift on PAGE is largely gone in cells where Cdc7 activity is compromised ( Kim et al . , 2008; Shimmoto et al . , 2009 ) . This suggests a possibility that Cdc7 may regulate checkpoint through Claspin/Mrc1 . However , the precise mechanism by which Cdc7 activates replication checkpoint has not been known . Claspin , originally discovered as a factor that binds to Chk1 in Xenopus egg extract ( Kumagai and Dunphy , 2000 ) , and its yeast homologue , Mrc1 , are essential for activation of downstream effector kinases ( Chk1 and Cds1/Rad53 , respectively ) , and are required for replication checkpoint control as a mediator ( Chini and Chen , 2003; Yoo et al . , 2006; Lindsey-Boltz et al . , 2009; Alcasabas et al . , 2001; Osborn and Elledge , 2003; Tanaka and Russell , 2001 ) . Claspin/Mrc1 is required also for efficient fork progression ( Lin et al . , 2004; Petermann et al . , 2008; Scorah and McGowan , 2009; Szyjka et al . , 2005 ) . Claspin interacts with various replication factors and other factors including ATR , Chk1 , Cdc7 kinase , Cdc45 , Tim , MCM4 , MCM10 , PCNA , DNA polymerases α , δ , ε , And-1 , and Rad9 ( Gambus et al . , 2006; Izawa et al . , 2011; Lee et al . , 2005; Brondello et al . , 2007; Serçin and Kemp , 2011; Gold and Dunphy , 2010; Uno and Masai , 2011; Liu et al . , 2012; Hao et al . , 2015 ) , as well as with DNA ( Sar et al . , 2004; Zhao and Russell , 2004 ) suggesting its role at the replication forks and potentially in initiation . Yeast Mrc1 was shown to move along with replication fork , linking the helicase components to the replicative polymerases ( Katou et al . , 2003 ) . More recently , Mrc1 , in conjunction with Tof1/Csm3 , was shown to stimulate DNA replication fork progression in an in vitro reconstitution assay system ( Yeeles et al . , 2017 ) . We recently reported a novel role of Claspin as a recruiter of Cdc7 kinase for efficient phosphorylation of Mcm proteins required for initiation ( Yang et al . , 2016 ) . Cdc7-recruiting function and its potential role in origin firing regulation was reported also for fission yeast Mrc1 ( Matsumoto et al . , 2017; Masai et al . , 2017 ) . The role of Claspin/Mrc1 as a replication checkpoint mediator is well established from yeasts to human . In metazoan Claspin , phosphopeptide motifs ( CKBD [Chk1-binding domain] or CKAD [Chk1-activating domain] ) were identified that are required for regulated binding of Chk1 ( Kumagai and Dunphy , 2003 ) . In vitro reconstituted system was also reported in which Chk1 activation could be monitored in the presence of ATR ( Lindsey-Boltz et al . , 2009 ) . In Xenopus egg extracts , conserved serine-864 and serine-895 are phosphorylated upon replication stress and this phosphorylation is required for checkpoint activation . CKBD is required for checkpoint activation , and a phosphopeptide containing CKBD is sufficient for binding to Chk1 . In human cells , Thr-916 residue in CKBD was reported to be phosphorylated in response to HU and the 3A mutant ( T916A S945A S983A ) is deficient in checkpoint activation ( Chini and Chen , 2006 ) . The nature of kinase ( s ) responsible for CKBD phosphorylation has been controversial . Chk1 kinase was reported to phosphorylate Thr-916 in vitro ( Chini and Chen , 2006 ) , although it was later reported that Chk1 is not responsible for the phosphorylation of the CKAD in vivo ( Bennett et al . , 2008 ) . On the other hand , Dunphy’s group reported that casein kinase is a potential kinase responsible for phosphorylating CKBD ( Meng et al . , 2011 ) . They showed that casein kinases can phosphorylate in vitro the critical threonine residue in CKBD , and that casein kinase γ1 , among different types of casein kinases , promotes phosphorylation of CKBD in vivo and its depletion reduced phosphorylation of Thr-916 of the CKBD polypeptide ( 899-953 ) ectopically expressed in human cells , and diminished checkpoint activation . Here , through CRISPR/Cas9 system , we have established a derivative of HCT116 ( human colon cancer cell line ) in which the promoter of Cdc7 gene is mutated . This cell line ( HCT116-323 ) , expressing Cdc7 at a low level , replicates its DNA at a normal rate , but replication checkpoint activation was significantly reduced . The AP ( acidic patch ) motif near the C-terminus of Claspin interacts with Cdc7 and is required for Cdc7-mediated phosphorylation of Claspin and also for its interaction with various replication proteins . We have generated a DE/A mutant ( APDE/A ) of Claspin , in which all the acidic residues in aa988-1086 were replaced by alanine , and have shown that this mutant does not interact with Cdc7 and is not phosphorylated by Cdc7 in spite of the presence of all the serine/threonine residues . We speculated this is due to the inability of the APDE/A mutant to recruit Cdc7 ( Yang et al . , 2016 ) . We found that APDE/A is not able to activate Chk1 in response to HU , and cannot bind to Chk1 in spite of the presence of intact CKBD sequence . We surmised this is probably due to the absence of required phosphorylation of CKBD in APDE/A mutant . Through mass spectrometry analyses , a number of phosphorylation sites were identified near CKBD , including S945 in CKBD . Cdc7 depletion resulted in loss of most of these phosphorylations including S945 . These results provide strong evidence for the proposal that Cdc7 plays a crucial role in phosphorylation of CKBD . We noted that complete suppression of Chk1 activation requires depletion of both Cdc7 and CK1γ1 in all the cell lines tested . The dependency of Chk1 kinase activation on each kinase differs between cancer cells and non-cancer cells; cancer cells expressing a high level of Cdc7 exhibit higher dependency on Cdc7 , whereas non-cancer cells with a lower level of Cdc7 depend more on Chk1γ1 . We propose that either Cdc7 or CK1γ1 can phosphorylate CKBD for checkpoint activation , while the cellular context affects the pathway choice in different cells .
Previously , we reported that conditional knockout of Cdc7 in mouse ES cells resulted in loss of Chk1 activation ( measured by Chk1 S317 phosphorylation ) in response to HU or UV irradiation . We also showed siRNA-mediated inhibition of Cdc7 expression resulted in significant reduction of Chk1 activation in cancer cells ( Kim et al . , 2008 ) . It was later reported that Cdc7 inhibition using an inhibitor delayed the checkpoint activation in cancer cells ( Rainey et al . , 2013 ) . We reexamined this in human cancer cell lines by knocking down Cdc7 expression with siRNA ( Figure 1—figure supplement 1 ) . In the human colon cancer cell line , HCT116 , Chk1 activation was detected at 15 min after addition of 2 mM HU , and increased up to 50 min . At 48 hr after transfection of Cdc7 siRNA , Cdc7 expression was reduced by over 90% . Under this condition , Chk1 activation was reduced by ~90% at 30 min compared to the parental cells . In order to more precisely examine the effect of the Cdc7 level on the replication checkpoint activation , we generated a derivative of HCT116 . By using CRISPR-Cas9 , we generated a mutant cell line , HCT116-323 , in which Cdc7 expression level was reduced due to a deletion in its promoter region . The deletion covers 12 bp near the transcription initiation site ( Figure 1—figure supplement 2 ) . The growth of HCT116-323 was slightly slower than the parent ( data not shown ) , but the DNA synthesis or S phase progression was not noticeably affected . There was no significant difference in the level of BrdU incorporation ( Figure 1A ) , and S phase proceeded in a similar timing for up to 8 hr after release from the double thymidine block . However , at 10 hr after release , the wild-type HCT116 cells proceeded to G1 , while HCT116-323 cells have not divided yet ( Figure 1B ) . This suggests that HCT116-323 cells have some difficulty going through mitosis , although eventually they enter mitosis . The Cdc7 protein level was reduced by 90% in HCT116-323 compared to the parent HCT116 . After HU treatment , Chk1 activation was reduced by about 80% at 30 min in HCT116-323 compared to the parent ( Figure 1C ) , confirming that a certain level of Cdc7 activity is required for full activation of Chk1 in HCT116 . Claspin is a phosphoprotein , and at least some of these phosphorylations may be mediated by Cdc7 kinase ( Kim et al . , 2008; Rainey et al . , 2013 ) . We expressed various mutant forms of Claspin tagged with Flag at the C-terminus in 293T ( human embryonic kidney cells 293 expressing SV40 large T antigen ) cells and pulled down with anti-Flag antibody . We generated ST27A , ST5A and ST19A , in which all the serines/threonines in aa903 ~1120 , aa1121 ~1218 , and aa1219 ~1337 , respectively , were replaced by alanine . Whereas Chk1 was co-pulled down with the wild-type and ST5A and ST19A , it was not pulled down with ST27A ( Figure 2A , lanes 8 , 10–12 ) . This is likely due to the alanine substitution of serine/threonine of CKBD in this mutant , as was shown before ( Kumagai and Dunphy , 2003; Meng et al . , 2011; Chini and Chen , 2006 ) . We previously reported that the acidic patch ( AP; aa988-1086 ) of Claspin is required for its interaction with Cdc7 kinase . The DE/A mutant of AP ( APDE/A ) in which all the acidic residues in AP were replaced with alanine is deficient in interaction with Cdc7 and fails to support phosphorylation of Mcm in non-cancer cells ( Yang et al . , 2016 ) . Intriguingly , APDE/A mutant did not cause defect in phosphorylation of Mcm in cancer cells . In the same pull down assays in 293 T cells , Chk1 was not pulled down with the APDE/A mutant as well , although CKBD is intact in this mutant ( Figure 2A , lane 9 ) . Using the Claspin conditional knockout MEF cells , we examined the ability of the APDE/A mutant to support Chk1 activation . The wild-type or APDE/A mutant Flag-tagged Claspin was stably expressed in the mutant MEF cells and endogenous Claspin was knocked out by transfection of Ad-Cre . The Claspin mutant cells without transgene did not show Chk1 activation ( Figure 2B , lane 2 , pChk1 ) , as expected from the loss of Claspin , whereas expression of the wild-type Claspin restored it ( Figure 2B , lane 4 ) . In contrast , the APDE/A mutant did not restore the Chk1 activation , consistent with its reduced binding to Chk1 ( Figure 2B , lane 6 ) . We examined the function of the Claspin APDE/A mutant for checkpoint activation in U2OS cells ( human bone osteosarcoma epithelial cells ) as well . The wild-type or APDE/A mutant Claspin was expressed on a retroviral vector . The endogenous Claspin was knocked down by siRNA specific to its 3’-noncoding segment . The wild-type Claspin could restore the Chk1 phosphorylation in Claspin-depleted cells , whereas the APDE/A mutant did not fully restore it ( Figure 2C , lanes 10 and 12 ) . We previously showed that APDE/A is not phosphorylated by Cdc7 in vivo as well as in vitro , presumably due to impaired recruitment of Cdc7 to Claspin ( Yang et al . , 2016 ) . Thus , the above results are consistent with the notion that Cdc7-mediated phosphorylation of Claspin is required for checkpoint activation . CKBD was discovered as a motif to which Chk1 specifically binds ( Kumagai and Dunphy , 2000 ) . This binding was reported to be dependent on the phosphorylation of the conserved threonine or serine present in the CKBD . We have confirmed phospho-dependent interaction by pull down assays . Biotinylated CKBD-containing phosphopeptides ( NC [aa901-955 with phospho-Thr-916 and phospho-Ser-945] , N [aa905-925 with phospho-Thr-916] and C [aa934-954 with phospho-Ser-945] ) were mixed with the extract from 293 T cells and the pulled-down materials were examined by western analyses . Chk1 was efficiently recovered with NC but not with N or C ( Figure 2—figure supplement 1A; compare lanes 4 , 6 and 8 ) . When the biotinylated NC oligopeptide was pretreated with λ-phosphatase , binding was completely lost ( Figure 2—figure supplement 1A , lane 5 ) . Similar results were obtained with the purified Chk1 protein ( Figure 2—figure supplement 1B and C ) . This is consistent with previous findings , and indicates that at least two phosphorylated CKBD boxes are required for efficient Chk1 binding ( Kumagai and Dunphy , 2003; Chini and Chen , 2006 ) . If the dephosphorylated oligopeptide was incubated with Cdc7 prior to pull down , the level of Chk1 pull down increased ( Figure 2—figure supplement 1C , compare lanes 8 and 9 ) . The recovery is not complete , because the efficiency of Cdc7-mediated phosphorylation is inefficient on the NC oligopeptide due to the absence of AP , the recruiter of Cdc7 kinase . This suggests a possibility that Cdc7 phosphorylates CKBD and promotes the interaction of Claspin with Chk1 . We therefore examined the effect of Cdc7 on the interaction of Claspin with Chk1 in cells . 293 T cells were either treated with Cdc7 siRNA or non-treated and stimulated by 2 mM HU for 4 hr . Cell extracts were prepared and immunoprecipitation was conducted by using anti-Chk1 antibody . Claspin was coimmunoprecipitated with Chk1 , and the amount of the precipitated Claspin increased after HU ( Figure 3A , lanes 5 and 6 ) . In Cdc7-depleted cells , the amount of the precipitated Claspin significantly decreased ( Figure 3A , lanes 7 and 8 ) , indicating that the Claspin-Chk1 interaction depends on Cdc7 in 293 T cells . We have conducted similar experiments by using Cdc7 promoter mutant cell line ( HCT116-323 ) . Immunoprecipitation with anti-Chk1 antibody pulled down the endogenous Claspin protein in the wild-type HCT116 treated with HU ( Figure 3B , lane 3 ) . However , the amount of coimmunoprecipitated Claspin was significantly reduced in HCT116-323 in which the Cdc7 protein level was reduced ( Figure 3B , lane 4 ) . These results establish that binding of Chk1 to Claspin requires Cdc7 function in HCT116 as well . Above in vitro and in vivo ( cellular ) results strongly suggest that Cdc7 phosphorylates CKBD for recruitment of Chk1 . To evaluate this possibility in more detail , we conducted mass spectrometry analyses of Claspin after treatment with HU in 293 T cells . 293 T cells , either transfected with Cdc7 siRNA or control siRNA , were treated with 2 mM HU for 24 hr or non-treated . Cell lysates prepared with RIPA buffer were subjected to immunoprecipitation with anti-Claspin antibody . The precipitated Claspin bands were detected on SDS-PAGE ( Figure 4—figure supplement 1 ) and were subjected to mass spectrometry analyses . In the control siRNA transfected cells with HU , we detected 31 phosphorylated amino acids , 25 of which are present in the C-terminal half of the protein . In the siCdc7 transfected cells with HU , 35 phosphorylated amino acids were detected , 21 of which are present in the C-terminal half of the protein . Furthermore , 20 or 14 phosphorylated amino acids were detected on the 266-amino acid segment ( 720 ~ 985 ) overlapping with a part of the CKBD segment in the control siRNA or siCdc7 transfected cells with HU , respectively . 11 phosphorylated serines and threonines in the 266 amino acid segment disappeared after transfection of siCdc7 in comparison with the control siRNA ( with HU ) . Thr-916 and Ser-945 were among those that disappeared after Cdc7 depletion ( Figure 4 and Figure 4—figure supplement 1B ) . These results support the conclusion that Cdc7 plays a major role in phosphorylating CKBD for checkpoint activation in 293 T cells . Dunphy’s group previously reported that casein kinase 1 γ1 ( CK1γ1 ) can promote interaction between Claspin and Chk1 through phosphorylating CKBD in cancer cells ( Meng et al . , 2011 ) . We evaluated the potential role of CK1γ1 in replication checkpoint in conjunction with Cdc7 . When CK1γ1 was reduced in U2OS by siRNA , Chk1 activation was reduced by about half , consistent with the previous report ( Figure 5A , lanes 9–12 ) . Cdc7 knockdown reduced the Chk1 activation by about 80% , but still weak Chk1 S317 signal remained ( Figure 5A , lanes 5–8 ) . In Cdc7-CK1γ1 double knockdown , Chk1 activation was almost completely gone ( Figure 5A , lanes 13–16 ) . Codepletion of Cdc7 and CK1γ1 led to additive effect on decrease of Chk1 activation in HeLa ( human cervix epitheloid carcinoma ) cells as well ( Figure 5—figure supplement 1 ) . CK1γ1 depletion resulted only in partial suppression of Chk1 activation also in HCT116 ( Figure 5B , lanes 5–8 ) . However , that in HCT116-323 , where Chk1 activation is reduced due to decreased Cdc7 expression , almost completely suppressed it , suggesting the residual Chk1 activation in HCT116-323 is mediated by CK1γ1 ( Figure 5B , lanes 13–16 ) . BrdU incorporation of HCT116-323 was not significantly affected by CK1γ1 depletion ( Figure 5—figure supplement 2 ) , suggesting that suppression of Chk1 activation in HCT116-323 by siCK1γ1 is not caused by reduced DNA replication . All these results strongly suggest that both Cdc7 and CK1γ1 play roles in Chk1 activation in response to replication stress and complete suppression of Chk1 activation requires inactivation of both kinases . In cancer cells , we consistently observed that Cdc7 depletion reduced Chk1 activation more than CK1γ1 depletion did . Indeed , cell death measured by TUNEL assays increased after siRNA transfection targeted either to Cdc7 or CK1γ1 . HU treatment increased TUNEL positive cells in Cdc7 siRNA treated cells but not in CK1γ1 siRNA treated cells ( Figure 5—figure supplement 3A and B , compare row 2 and 10 or 3 and 11 in B ) . Unexpectedly , in NHDF ( normal human dermal cells ) , CK1γ1 depletion almost completely impaired Chk1 activation ( by 90% ) , whereas Cdc7 depletion only partially ( by 50% ) inhibited it ( Figure 5C , lanes 6 and 4 ) . In MEF cells , Chk1 activation was inhibited by more than 80% by CK1γ1 depletion ( Figure 2B , compare lanes 1 and 7 ) . Similarly , in other non-cancer cells RPE-1 ( human epithelial cells immortalized with hTERT ) and TIG-3 ( a normal diploid fibroblast cell line ) , depletion of CK1γ1 significantly reduced Chk1 activation ( by 90% for both ) , whereas that of Cdc7 reduced it by 60% in both cells ( Figure 5—figure supplement 4 , lanes 6 and 7 ) , supporting our conclusion that CK1γ1 plays a major role in checkpoint activation in non-cancer cells . How do Cdc7 and CK1γ1 collaborate for checkpoint activation ? One possibility is that CK1γ1 acts as a priming kinase for subsequent phosphorylation by Cdc7 . Phosphorylation of a substrate by Cdc7 kinase has been shown to be facilitated by the ‘priming phosphorylation’ by other kinases in a number of instances ( Masai et al . , 2000; Sasanuma et al . , 2008; Wan et al . , 2008; Francis et al . , 2009; Murakami and Keeney , 2014 ) . These possibilities were examined by conducting the in vitro phosphorylation assays using CKBD containing polypeptides . We first conducted phosphorylation assays using a Claspin-derived polypeptide ( aa897-1100 ) as a substrate . This substrate contains both AP and CKBD . Both Cdc7 and CK1γ1 phosphorylated this polypeptide in a dose-dependent manner . Addition of Cdc7 in the presence of a low amount of CK1γ1 increased the phosphorylation level of the polypeptide . At a low level of Cdc7 , the effect of CK1γ1 was additive on the level of the phosphorylation , whereas , at the highest concentration of Cdc7 , similar levels of phosphorylation were observed regardless the presence or absence of CK1γ1 ( Figure 6—figure supplement 1 ) . CK1γ1 alone also could achieve the similar level of phosphorylation when added in a sufficient amount ( data not shown ) . These results are consistent with the notion that Cdc7 and CK1γ1 can phosphorylate this polypeptide in an independent fashion . In order to more directly examine whether Cdc7 phosphorylates CKBD , we have prepared the 50 amino acid polypeptide ( 906 ~ 955 ) containing two CKBD ( wild-type ) and their derivatives containing amino acid substitutions at selected serine/threonine residues , and used them as substrates for in vitro phosphorylation assays with Cdc7 and/or CK1γ1 . Cdc7 was able to phosphorylate the 50 aa polypeptide ( Figure 6 , lanes 2 , 13–15 ) , and this phosphorylation was significantly reduced by alanine substitutions of threonine and serine in CKBD ( CKBD-A; Figure 6 , compare lanes 15 and 16 ) , indicating that CKBD is the major target of Cdc7 kinase on this polypeptide . To assess the roles of other serine/threonine residues on the polypeptide , we replaced all other serine/threonine residues with either alanine ( Others-A ) or glutamic acid ( Others-E ) . Cdc7 was able to phosphorylate Others-A , albeit at a reduced level compared to Wild-type ( Figure 6 , lanes 15 and 19 ) , whereas Others-E was phosphorylated by Cdc7 at about three times more efficiently than was Wild-type ( Figure 6 , lanes 15 and 22 ) . This is probably due to the effect of acidic residues that facilitate the substrate recognition by Cdc7 . In contrast , CK1γ1 did not very efficiently phosphorylate Wild-type ( Figure 6 , lanes 6 , 10–12 ) . Consistently , addition of CK1γ1 did not stimulate Cdc7-mediated phosphorylation of Wild-type was ( Figure 6 , lanes 2–5 ) . CK1γ1 alone did not phosphorylate Others-E very efficiently , either ( Figure 6 , lanes 23 ) . These results are consistent with the conclusions that Cdc7 phosphorylates preferentially the serine/threonine residues in CKBD in vitro and that CK1γ1 does not serve as a priming kinase for phosphorylation of CKBD by Cdc7 . They also suggest that Cdc7 is more active than CK1γ1 as a CKBD-phosphorylating kinase in vitro . These results indicate that Cdc7 and CK1γ1 independently phosphorylate CKBD , most notably at the Thr-916/Ser-945 . Then , how is the kinase selected for phosphorylation of CKBD in cells ? We examined other cell lines for dependency of checkpoint activation on Cdc7 ( Figure 7—figure supplement 1 ) . We used XL413 , a Cdc7-specific inhibitor , to inhibit Cdc7 kinase . In HL60 ( human promyelocytic leukemia cell line ) , KM12-Luc ( human colon cancer cell line ) , SK-BR-3-Luc ( human breast cancer cell line ) and HCT15 ( human colorectal adenocarcinoma cell line ) , pretreatment with XL413 reduced HU-induced Chk1 S317 phosphorylation by 32 , 25 , 24% and 61% , respectively . In contrast , in NCI-H1975-Luc ( human lung adenocarcinoma cell line ) and NUGC-3 ( human gastric cancer cell line ) , the effect of XL413 was weaker ( 15% and 4% inhibition , respectively ) . We noted that the levels of Cdc7 were significantly lower in these latter cell lines than in other cancer cell lines ( Cheng et al . , 2013 ) . Non-cancer cells including NHDF , RPE-1 , and TIG-3 exhibited more dependency on CK1γ1 than on Cdc7 for checkpoint activation ( Figure 5C and Figure 5—figure supplement 4 ) . We therefore examined the levels of Cdc7 and CK1γ1 proteins in different cell lines . By comparing the western signals of the whole cell extracts from a fixed number of various cells with those of the control protein samples of known quantity , we estimated the numbers of Cdc7 and CK1γ1 molecules per cell . The results indicate that the numbers of Cdc7 molecules in cancer cells are 10–50 fold more than those in non-cancer cells . On the other hand , the numbers of the CK1γ1 molecules in non-cancer and cancer cells are roughly the same ( 1 . 4 ~ 2 . 4×105; except for HeLa cells which have four-fold more than NHDF; Figure 7—figure supplement 2 ) . The results suggest a possibility that Cdc7 predominantly serves as a CKBD kinase in the majority of cancer cells , since it is overexpressed , whereas in NHDF , CK1γ1 is more abundant than Cdc7 , and thus serves as a major kinase for CKBD phosphorylation . In order to test the above possibility , we examined the effects of manipulation of protein expression levels on the checkpoint pathway choice . In HCT116 , CK1γ1 depletion does not significantly affect the Chk1 activation , whereas Cdc7 depletion reduced it by over 50% ( Figure 7A , lanes 2 and 3 ) . In contrast , in HCT-323 that expressed Cdc7 at a reduced level , CK1γ1 depletion reduced the Chk1 activation almost completely ( Figure 7A and B , compare lanes 5 and 7 ) . We also examined whether overexpression of Cdc7 modulates the effect of CK1γ1 depletion . Cdc7 was expressed in NHDF by a lentivirus expression vector and the cells were adjusted for 3 days . Then , CK1γ1 or Cdc7 was depleted by siRNA for two days . HU-induced Chk1 S317 phosphorylation was reduced by more than 70% by CK1γ1 depletion in mock-transfected cells ( Figure 7C and D , lanes 1 and 3 ) , but the inhibition was mitigated in Cdc7-expressing cells ( Figure 7C and D , lanes 5 and 7 ) , whereas stronger inhibition by Cdc7 siRNA ( by ~60% ) was observed in Cdc7-overexpressing NHDF cells ( Figure 7C and D , lanes 5 and 6 ) . This effect appears to depend on the ‘adaptation’ period . Incubation of Cdc7-overexpressing cells for three weeks before addition of siRNA resulted in even better dependence on Cdc7 for checkpoint activation ( Figure 7—figure supplement 3 ) . These results are consistent with the idea that the relative levels of Cdc7 and CK1γ1 are one of the factors that determine the pathway for replication stress-induced Claspin phosphorylation .
The following results presented in this study show that Cdc7 is predominantly responsible for phosphorylation of CKBD of Claspin in response to replication stress in most of the cancer cells . Mass spectrometry analyses revealed the presence of the clusters of phosphorylation sites in and near CKBD , and notably at Thr-916/Ser-945 , conserved serines/threonines in the two CKBD . The phosphorylation of Thr-916/Ser-945 is not detected in Cdc7-depleted cells consistently in three independent experiments ( Figure 4 and Figure 4—figure supplement 1 ) . These two CKBD appear to be important for binding to Chk1 , since the peptide containing both CKBD bound to Chk1 but not the peptide containing only one ( Figure 2—figure supplement 1B ) . Using human in vitro cell-free system , mutation of Thr-916 and Ser-945 was shown to inhibit its association with Chk1 ( Clarke and Clarke , 2005 ) . Our analyses indicate that Cdc7 recruited to AP ( aa988 ~ 1086 ) phosphorylates multiple serine/threonine residues in the adjacent segment ( aa720 ~968 ) . These sites do not match very well with previously reported phosphorylation sites of Claspin phosphorylated by Cdc7 in vitro ( Rainey et al . , 2013; the phosphorylation sites do not include serine/threonine in CKBD; only S721 and S744 in aa720 ~ 968 ) . Claspin may adopt a conformation that is more favorable for recruitment of Cdc7 kinase in vivo , that may permit more localized phosphorylation in aa720 ~ 968 . Cdc7 likes acidic environment , and thus initial phosphorylation may trigger further recruitment of Cdc7 , causing autoactivating loop of phosphorylation . Indeed , the CKBD-derived peptide , in which serine-threonine residues other than Thr-916 and Ser-945 were replaced by glutamic acid , was phosphorylated by Cdc7 with much better efficiency in vitro ( Figure 6 ) . On the other hand , some Cdc7-dependent phosphorylation sites outside CKBD were detected only in one or two experiments out of the three independent experiments ( Figure 4 ) . This may suggest promiscuous or stochastic nature of phosphorylation of Claspin by Cdc7 recruited to AP . CK1γ1 were previously reported to phosphorylate CKBD ( Meng et al . , 2011 ) . Casein kinase is similar to Cdc7 in its acidophilic preference for substrate selection . Thus , it is predictable that casein kinase also recognizes CKBD which is embedded in acidic environment . In fact , CK1γ1 can phosphorylate CKBD+AP containing polypeptides with efficiency similar to that of Cdc7 ( Figure 6—figure supplement 1 ) . We examined the possibility that CK1γ1 may initially act as a priming kinase , which phosphorylates serine/threonine residues near the Cdc7 target site . This priming phosphorylation would create acidic environment which may facilitate the recognition by Cdc7 . However , we were unable to detect priming phosphorylation events on the aa897-1100 polypeptide containing CKBD used in this study . Both kinases appear to independently and additively phosphorylate this polypeptide in vitro ( Figure 6 and Figure 6—figure supplement 1 ) . This is consistent with the fact that , in cancer cells , Cdc7 depletion inhibits checkpoint activation more vigorously than CK1γ1 depletion does , and that combination of both depletions results in almost complete inhibition of checkpoint activation . Although Cdc7 can phosphorylate Thr-916/Ser-945 on the 50aa oligopeptide ( 906 ~ 955 ) in vitro , albeit to a lower extent , CK1γ1 was not able to efficiently phosphorylate it ( Figure 6 ) . CK1γ1 may need the adjacent polypeptide segment to efficiently recognize Claspin . Furthermore , Cdc7 only poorly phosphorylates ST27A Claspin , in which all the serines/threonines in aa903 ~1120 containing CKBD were replaced by alanine ( Yang et al . , 2016 ) . CK1γ1 efficiently phosphorylates the same polypeptide , suggesting that CK1γ1 can phosphorylate Claspin outside of aa903-1120 ( Figure 6—figure supplement 2 ) . The full-length Claspin protein binds to purified Chk1 protein in vitro . This in vitro binding occurs with both APDE/A and or ST27A ( data not shown ) , whereas binding of the 55 aa CKBD polypeptide to Chk1 completely depends on phosphorylation of Thr-916/Ser-945 . On the other hand , either APDE/A or ST27A do not bind to Chk1 in vivo by immunoprecipitation ( Figure 2A ) , suggesting better specificity for the phosphorylated CKBD in the cells . And-1 was reported to promote the interaction between Claspin and Chk1 , thereby stimulating efficient Chk1 activation by ATR ( Hao et al . , 2015 ) . ATR-ATRIP was also shown to be required for Claspin-mediated Chk1 activation in Xenopus egg extracts ( Kumagai et al . , 2004 ) . Thus , other factors are likely to improve the specificity and efficacy of the Claspin-Chk1 interaction in cells . In cancer cells examined , Cdc7 kinase plays a major role in mediating checkpoint activation ( Figure 5A and B , Figure 5—figure supplement 1 and Figure 5—figure supplement 5 ) . In contrast , in non-cancer cultured cells , such as NHDF , PRE-1 or TIG-3 , CK1γ1 plays a predominant role for checkpoint activation ( Figure 5C and Figure 5—figure supplement 4 ) . The Claspin APDE/A mutant cannot support checkpoint activation even in MEF cells ( Figure 2 ) , which depends more on CK1γ1 for checkpoint activation ( Figure 2B ) . This is due to the inability of the APDE/A mutant to interact with CK1γ1 as well ( data not shown ) . The selection of a kinase for checkpoint activation appears to be correlated with the relative levels of the two kinases in given cells , since reduction of the Cdc7 level in HCT116 increased dependency on CK1γ1 ( Figure 7 ) , while overexpression of Cdc7 in normal cells decreased dependency on CK1γ1 ( Figure 7 and Figure 7—figure supplement 3 ) . Cdc7 inhibition or depletion reduces DNA synthesis both in cancer and non-cancer cells . However , it induces cell death in the former cells but not in the latter ( Montagnoli et al . , 2006; Montagnoli et al . , 2008; Sawa and Masai , 2008; Ito et al . , 2008; Ito et al . , 2012 ) . Our results indicate higher dependency on Cdc7 for checkpoint activation in cancer cells than in non-cancer cells . The checkpoint defect by the loss of Cdc7 specifically observed in cancer cells may explain cancer cell-specific cell death induction by Cdc7 inhibition ( Ito et al . , 2008; Ito et al . , 2012 ) , since failure to properly respond to replication stress would be directly linked to genome instability . In summary , our findings in this report show that Cdc7 is a major kinase that mediates phosphorylation of CKBD of Claspin in response to replication stress , especially in Cdc7-overexpressing cancer cells . This finding also clarifies the role of Cdc7 in replication checkpoint activation . Another important finding of this report is that complete shutdown of Chk1 activation requires loss of both Cdc7 and CK1γ1 , and that contribution of each kinase on Chk1 activation may vary from one cell-type to another and may be at least partly determined by the abundance of each protein and cellular context . The crucial role of Cdc7 in checkpoint activation in cancer cells but not in non-cancer cells may provide molecular explanation for cancer cell-specific cell death induced by Cdc7 inhibition .
293T , HeLa , NHDF , HCT116 and U2OS were obtained from ATCC . Claspin flox /- Mouse Embryonic Fibroblasts ( MEFs ) were established from E12 . 5 embryos ( Yang et al . , 2016 ) . Claspin flox /- MEFs stably expressing the wild-type or APDE/A mutant Claspin were established by infecting recombinant retroviruses expressing these cDNAs ( Yang et al . , 2016 ) . Cells were grown in Dulbecco’s modified Eagle’s medium ( high glucose ) supplemented with 15% fetal bovine serum ( NICHIREI ) , 2 mM L-glutamine , 1% sodium pyruvate , 100 U/ml penicillin and 100 μg/ml streptomycin in a humidified atmosphere of 5% CO2 , 95% air at 37°C . HCT-15-Luc#1 , NCI-H1975-Luc , NUGC-3 were grown in RPMI supplemented with 10% fetal bovine serum ( Gibco ) ; HL-60 were grown in RPMI supplemented with 55 μM β- 2-mercaptoethanol and 10% fetal bovine serum ( Gibco ) ; 5K-BR-3-Luc were grown in McCoy’s 5A supplemented with 10% fetal bovine serum ( Gibco ) ; KM12-Luc , RPE-1 and TIG-3 were grown in Dulbecco’s modified Eagle’s medium supplemented with 10% fetal bovine serum ( Gibco ) . Anti-Chk1 ( sc-8408 or sc-7898 , 1:200 ) and anti-MCM2 ( sc-9839 , 1:200 ) antibodies were from Santa Cruz Biotechnology . Anti-MCM2 S53 ( A300-756A , 1:1000 ) antibody was from Bethyl . Anti-Cdc7 ( K0070-3 , 1:1000 ) , anti-Chk1 ( K0085-3 , 1:1000 ) and anti-Flag ( M185-3L , 1:1000 ) antibodies were from MBL . Anti-CK1γ1 was from Biorbyt ( orb37898 , 1:2000 ) . Anti-Chk1 S317 ( #2341 , 1:1000 ) and anti-tubulin ( T5168 , 1:1000 ) antibodies were from Cell Signaling and Sigma-Aldrich , respectively; anti-human Claspin , anti-mouse Claspin anti-MCM4 , and anti-MCM4 S6T7 antibodies were previously described ( Masai et al . , 2006; Yang et al . , 2016 ) . Purified Cdc7-ASK ( 05–109 ) , Chk1 ( 02–117 ) and CK1γ1 ( 03–105 ) kinase were from Carna Bioscience , Inc Cdc7-ASK was also purified from Sf9 cells as previously described ( Masai et al . , 2000 ) . The Claspin-coding segment ( XhoI/XbaI fragment ) of CSII-EF MCS-mAG ( monomeric Azami Green; Karasawa et al . , 2003 ) -TEV-6His-Claspin-3Flag , CSII-EF MCS-6His-Claspin-3Flag or CSII-EF MCS-6His-Claspin-HA plasmid DNA was replaced by DNA fragments encoding portions of Claspin or its mutant forms , obtained by PCR amplification , to express truncated or mutant forms of Claspin . The EcoRI-HpaI fragment containing the wild-type or APDE/A mutant Claspin DNA from mAG-TEV-6His-Claspin-3xFlag was inserted at the EcoRI/SnaBI site of pMX-IP ( Addgene ) to construct retroviral expression vectors . A short deletion was introduced into the promoter segment of the Cdc7 gene basically according to the method previously described ( Shalem et al . , 2014 ) . Briefly , target sequence ( CR1 , CR2 , CR3 , or CR4; see Figure 1—figure supplement 2 for sequences ) was inserted into the lentiCRISPR v2 vector , and the lentivirus carrying the target sequence was prepared . A colorectal cancer cell line HCT116 , which is pseudodiploid and has two copies of the Cdc7 gene ( https://cansar . icr . ac . uk/cansar/cell-lines/HCT-116/copy_number_variation/chromosome_1/ ) , was used for targeting . Forty-eight hours after infection , 4 µg/ml of puromycin was added for 4 days and resistant cells were plated in 96 well plates to isolate single cell-derived clones . Cdc7 expression in each clone was examined by western blot analysis and genomic DNA sequences were determined with PCR products of the targeted segment and also with the plasmids after cloning of the PCR fragments . 1 . 6 μg of expression plasmid DNA in 100 μL of 150 mM NaCl were mixed with 100 μl of 150 mM NaCl supplemented with 7 μl 1 mg /mL PEI ( polyethylenimine ‘MAX’ [MW25 , 000; Cat . 24765; Polyscience , Inc . ] ) . After 30 min incubation at a room temperature , the solution was added to 293 T cells cultured in six well plates with 2 ml of fresh D-MEM in each well ( Uno et al . , 2012 ) . HeLa , U2OS and HCT116 cells were transfected siRNA by oligofectamine ( Thermo Fisher Scientific ) for 48 hr . NHDF cells were transfected siRNA by lipofectamine 2000 ( Thermo Fisher Scientific ) for 48 hr . The siRNA sequences are provided in Supplementary file 1 or Supplementary file 2 . Proteins in SDS-sample buffer were incubated at 96°C for 1 min , were run on 4 ~ 20% gradient SDS-PAGE ( ATTO ) and then transferred to Hybond ECL membranes ( GE Healthcare ) , followed by western blot analysis with the indicated antibodies . Blots were then incubated for 1 hr with the secondary antibody conjugated to horseradish peroxidase . Detection was conducted with Lumi-Light PLUS Western Blotting Substrate ( Roche ) and images were obtained with LAS3000 ( Fujifilm ) . 1 × 105 cells of HCT116 and its Cdc7 promoter mutant derivative were plated in 6-well plates . Cells were incubated and cell numbers were counted at day 1 , 2 , 3 , 4 , and five by staining with Trypan blue . BrdU was added to HCT116 and Cdc7 promoter mutant cells in 6-well plates at 20 µM for 20 min . Cells were harvested and were fixed at −20°C by 75% ethanol . After wash with the wash buffer ( 0 . 5% bovine serum albumin in phosphate-buffered saline ) , cells were treated with 2 N HCl for 20 min and then with 0 . 1 M sodium borate ( pH8 . 5 ) for 2 min at room temperature . Cells were then treated with FITC-conjugated anti-BrdU antibody ( BD biosciences , 51–23 , 614 l ) for 20 min at room temperature in the dark , and further incubated with propidium iodide ( 25 μg/ ml ) and RNaseA ( 100 μg/ ml ) for 30 min at room temperature , followed by analyses with FACS ( fluorescence-activated cell sorter; Yoshizawa-Sugata and Masai , 2007 ) . Two pmole of biotinylated phosphopeptides , treated with λPPase or not , were mixed with cell lysates or with purified Chk1 protein , pulled down by streptavidin beads . After washing , samples were analyzed by western blotting using anti-Chk1 antibody . Purified Cdc7 protein and ATP were also added together with purified Chk1 protein when the template peptides needed to be phosphorylated . 293 T cells transiently expressing wild-type or mutant Claspin proteins ( C-terminally tagged with 3x-Flag ) were lysed in CSK buffer ( 10 mM PIPES-KOH [pH6 . 8] , 100 mM potassium glutamate , 300 mM sucrose , 1 mM MgCl2 , 1 mM EGTA , 1 mM DTT , 1 mM Na3VO4 , 50 mM NaF , 0 . 1 mM ATP , protease inhibitor-PI tablet [Roche] and 0 . 5 mM PMSF ) , containing 0 . 1% TritonX-100 and 10 units⁄ ml Benzonase ( Amersham plc . ) . After rotating for 60 min in cold room , the supernatants were incubated with anti-Flag M2 affinity beads ( SIGMA , A2220 ) for 60 min at 4°C . The beads were washed with CSK buffer three times and proteins bound to the beads were analyzed by western blotting . For immunoprecipitation of Claspin , 293T or HCT116 cells were transfected with indicated siRNA or control siRNA for 48 hr , and were further treated with 2 mM HU for indicated time or non-treated . Cells were lysed by CSK buffer as above and supernatants were mixed with anti-Claspin antibody conjugated to Dynabead protein G ( Invitrogen ) at 4°C for 1 hr . After wash with PBS with 0 . 01% Tween-20 , the proteins on beads were separated on 4–20% gradient gel and analyzed by western blotting with anti-Claspin and anti-Chk1 antibodies . 293 T cells ( two 10 cm dishes ) transfected with a plasmid expressing wild-type or mutant/truncated versions of Claspin for 40 hr were lysed with CSK buffer as above . The proteins of the supernatants were mixed anti-Flag M2 affinity beads ( Sigma-Aldrich ) and washed by Flag wash buffer ( 50 mM NaPi [pH7 . 5] , 10% glycerol , 300 mM NaCl , 0 . 2 mM PMSF and PI tablet ) . The bound proteins were eluted by Flag elution buffer ( 50 mM NaPi [pH7 . 5] , 10% glycerol , 30 mM NaCl , 200 µg ⁄ml 3xFlag peptide [SIGMA] , 0 . 1 mM PMSF and PI tablet ) ( Uno et al . , 2012 ) . To examine the interaction between the purified Claspin ( C-terminally Flag-tagged ) and Chk1 , the recombinant wild-type or mutant Claspin proteins were incubated with purified Chk1 ( Carna Bioscience ) in a reaction buffer ( 40 mM HEPES-KOH [pH 7 . 6] , 20 mM K-glutamate , 2 . 5 mM MgCl2 , 1 mM EGTA , 0 . 01% TritonX-100 , 1 mM Na3VO4 , 50 mM NaF , 1 mM ATP , 0 . 5 mM PMSF and PI tablet ) at 4°C for 1 hr . Anti-Flag M2 affinity beads were added and beads were recovered by centrifugation , washed twice with the reaction buffer . Proteins attached to the beads were analyzed by SDS-PAGE and detected by western blotting . To examine the effect of Cdc7 on interaction between Claspin and Chk1 , purified Claspin protein was first incubated for 30 min at 30°C with λPPase to remove any preexisting phosphates , and Claspin was recovered with anti-Flag M2 affinity beads . After thoroughly washing the beads with washing buffer ( 25 mM HEPES-KOH [pH 7 . 6] , 0 . 01% TritonX-100 , 0 . 5 mM PMSF and PI tablet ) , the recovered Claspin attached to anti-Flag M2 affinity beads was mixed with Chk1 alone or Chk1 and Cdc7 in the same reaction buffer at 30°C for 30 min and at 4°C for 1 hr . After washing three times with the washing buffer , proteins attached to the beads were analyzed by SDS-PAGE and detected by western blotting . 1 × 105 of Claspin flox /- MEFs or Claspin flox /- MEFs stably expressing the wild type or APDE/A mutant Claspin were infected with Ad-Cre for 48 hr or not treated . Cells , treated with 2 mM HU for 3 hr , were harvested , were resuspended in sample buffer and the whole cell extracts were analyzed by SDS-PAGE , followed by western blotting analyses . 293 T cells ( 15 cm dish , five plates ) were transfected with Cdc7 siRNA or control siRNA for 24 hr , and 2 mM HU was added or non-treated for 24 hr . Cells were lysed by RIPA buffer ( 50 mM Tris-HCl [pH8 . 0] , 150 mM NaCl , 0 . 5 % w/v sodium deoxycholate , 1 . 0 % NP-40 and 0 . 1% SDS ) containing 1 mM DTT , 1 mM Na3VO4 , 50 mM NaF , 0 . 1 mM ATP , 0 . 5 mM PMSF and PI tablet . The supernatants were mixed with anti-Claspin antibody conjugated to Dynabead protein G ( Invitrogen ) at 4°C for 1 hr . After washing with PBS containing 0 . 01% Tween-20 , the proteins on beads were separated on 4–20% gradient gel and stained by CBB . Claspin protein was extracted from the gel , digested by Trypsin and phospho-threonines or -serines were analyzed by mass spectroscopy . The results are presented in Supplementary spreadsheet . Claspin-derived polypeptides ( 100 ng ) or CKBD peptides ( 200 ng ) were incubated with the Cdc7-ASK complex and/or CK1γ1 in the kinase reaction buffer ( 40 mM Hepes-KOH [pH7 . 6]; 2 . 5 mM spermine; 5 mM MgCl2; 0 . 5 mM EGTA; 0 . 5 mM EDTA; 1 mM Na3VO4; 1 mM NaF; 2 mM DTT; 10 μM ATP; 1 μCi [γ-ATP] ) for 1 hr at 30°C . One-fourth volume of 5x sample buffer was added , heated at 75°C for 1 min and analyzed by SDS-PAGE , followed by silver staining and autoradiogram . For cell death analyses , DNA fragmentation was analyzed by TUNEL assay using In Situ Direct DNA Fragmentation Assay Kit ( Abcam ) according to the manufacturer’s protocol . In brief , 1 × 105 U2OS cells were transfected with siRNA for 72 hr , and treated with 2 mM HU for 0 , 24 or 48 hr . Harvested cells were fixed by 1% paraformaldehyde and washed by PBS , followed by addition of 70% ice-cold ethanol for 30 min . After washing by wash buffer , cells were mixed with staining solution for 1 hr at 37°C and rinsed by the rinse buffer . After removing the rinse buffer by centrifugation , cells were re-suspended in Propidium Iodide/RNase A solution and incubated in the dark for 30 min at room temperature . The FITC labeled cells were analyzed by FACS . Whole cell extracts prepared from 5 × 104 cells of 293T , HeLa , HCT116 , U2OS and NHDF were run on SDS-PAGE along with known amounts of purified Cdc7 and CK1γ1 proteins and were analyzed by western blotting using the antibodies against these proteins . 293 T cells were transfected with the lentivirus vector expressing Cdc7 ( on CSII-CMV-puro ) , pMDLg/pRRE , pRSV-Rev and pMD2 . G . Platinum-A ( Plat-A ) cells were transfected by the retrovirus vector expressing mAG-WT Claspin or mAG- APDE/A Claspin ( on pMX-IP ) . The fusion of a fluorescent protein at the N-terminus of Claspin does not affect its functions . In both cases , virus-containing medium were harvested at 2 days ( for lentivirus ) or 3 days ( for retrovirus ) after transfection and the viruses were concentrated by centrifugation . The concentrated Cdc7-expressing lentiviruses were used for infection of NHDF . The Claspin-expressing retroviruses were for infection of U2OS cells . In each case , at 2 days after infection , puromycin ( 4 µg/ml ) was added and resistant cells were selected for 2 days . Cells used were authenticated cell lines obtained from ATCC or JCRB Cell Bank . Mycoplasma contamination tests gave negative results on all the cells used . Error bars represent the mean ± s . d . values calculated from three independent replicate experiments . | It takes a human cell between six and eight hours to copy all three billion letters of its genome . During this time , any interruption to the process can lead to genetic errors , putting the cell in danger of developing disease . To guard against this , cells use a checkpoint system , testing their own health before , during and after DNA replication to make sure that they are ready for the next step . If a cell detects a problem while copying its DNA , it responds by activating proteins called checkpoint kinases . These stop the cell from continuing until the problem is resolved . One of these checkpoint kinases is a protein called Chk1 , which switches on if the cell gets stuck part way through copying its DNA . To switch Chk1 on , the cell first needs to activate a protein called Claspin . Activating Claspin involves adding a chemical phosphate group to part of the Claspin protein . A third protein takes on this role , but its identity is controversial . Recent research points to a protein called casein kinase 1 , but it was also possible that another protein , Cdc7 kinase , might be involved . To find out , Yang et al . used gene editing to lower the levels of Cdc7 in human cancer cells . The cells were able to copy their DNA under normal conditions , but they struggled to activate Chk1 when DNA replication stopped . Biochemical tests revealed that this was because , without Cdc7 , Claspin was not receiving the phosphate group it needed . Even so , the cancer cells still had some Chk1 activation , which meant that they must be able to activate some of their Claspin . So , Yang et al . tried getting rid of both Cdc7 and the other candidate protein , casein kinase 1 . This stopped Chk1 activation completely , revealing that although the cancer cells mainly used Cdc7 to activate Claspin , they also used casein kinase 1 . In tests on non-cancerous cells , the results were the other way around; healthy cells mainly used casein kinase 1 and relied less heavily on Cdc7 . These differences could prove useful for drug design . One of the challenges in cancer treatment is producing drugs that target cancer cells while leaving healthy cells unharmed . Future research could explore whether blocking Cdc7 could stop Chk1 activation in cancer cells only . This could stop the diseased cells fixing problems with their DNA replication , making it harder for them to survive . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression"
] | 2019 | Cdc7 activates replication checkpoint by phosphorylating the Chk1-binding domain of Claspin in human cells |
In diverse organisms , nanostructures that coherently scatter light create structural color , but how such structures are built remains mysterious . We investigate the evolution and genetic regulation of butterfly scale laminae , which are simple photonic nanostructures . In a lineage of buckeye butterflies artificially selected for blue wing color , we found that thickened laminae caused a color shift from brown to blue . Deletion of the optix patterning gene also altered color via lamina thickening , revealing shared regulation of pigments and lamina thickness . Finally , we show how lamina thickness variation contributes to the color diversity that distinguishes sexes and species throughout the genus Junonia . Thus , quantitatively tuning one dimension of scale architecture facilitates both the microevolution and macroevolution of a broad spectrum of hues . Because the lamina is an intrinsic component of typical butterfly scales , our findings suggest that tuning lamina thickness is an available mechanism to create structural color across the Lepidoptera .
Structural colors are both visually delightful and abundant in nature . Organisms deploy structural colors to display hues for which they lack pigments ( frequently blues and greens ) , to create specific optical effects such as iridescence or light polarization , and to mediate ecological interactions , including intraspecific signaling and camouflage . Unlike pigmentary color , which is caused by molecules that selectively absorb certain wavelengths of light , structural colors result from the constructive and destructive interference of light as it interacts with nanoscale , precisely-shaped physical structures that are made of a high refractive index material ( e . g . keratin , chitin , or cellulose ) . Despite the clear importance of structural color for living systems , the biological production of structural colors has long eluded characterization ( Cuthill et al . , 2017 ) . Many experimental techniques depend on harnessing variation to dissect biological processes , but photonic structures are so small that quantitatively measuring variation in their dimensions is technically demanding , especially for high-throughput sampling , detecting subtle variation that may segregate within populations , or analyzing over developmental time in vivo . The color itself is easier to quantify , but has limited utility as a proxy for nanostructural dimensions , since structural colors and pigments often co-occur and covary . While recent studies ( Parnell et al . , 2018; Matsuoka and Monteiro , 2018; Brien et al . , 2018 ) have made early headway toward describing genetic regulation of structural colors , much work remains to decipher the evolutionary , developmental , and genetic bases of structural coloration , and lab-tractable systems with intraspecific variation in structural coloration are needed . We present a promising system , the butterfly genus Junonia , with extensive variation in a simple structural color , and show how structural simplicity is a tactical advantage when seeking to unravel mechanisms for the biological production of nanostructures . In butterflies , photonic nanostructures occur within the architecture of scales . Scales are the fundamental coloration unit on butterfly wings and have a Bauplan consisting of a grid of ridges and crossribs , supported by a lower lamina that is a simple plane ( Figure 1A ) . Scales are composed of chitin and may also have embedded pigments . Intricate architecture and a high refractive index make scales a pliable substrate for photonic innovations , and indeed scales have been evolutionarily elaborated in many ways for impressive optical effects ( Ghiradella , 1985 ) . Even the simplest butterfly scales can produce structural color , via the lower lamina acting as a thin film reflector . Thin films are the simplest photonic structure and consist of a layer of high refractive index material , on the order of hundreds of nanometers thick , surrounded by a material with a contrasting refractive index , such as air ( Figure 1B ) . Light is reflected from each surface of the film , and these two reflections interfere with each other . If the two reflections remain in phase , which depends on the extra distance traveled through the film and the wavelength , then they interfere constructively to produce observable color ( Mason , 1927; Yeh et al . , 1978 ) . Conversely , wavelengths ( colors ) that undergo destructive interference have decreased brightness . While it is known that the thickness of the lower lamina is one parameter that controls structural color wavelength ( Stavenga et al . , 2014 ) and that thickness can respond to artificial selection in the laboratory ( Wasik et al . , 2014 ) , it is not known how general this mechanism is in natural evolution . It is also unknown how lamina structural colors are genetically regulated and whether any recognized butterfly wing patterning genes regulate lamina thickness . Here , we use mutants with deletion of the optix wing patterning gene , artificial selection on wing color , and genus-wide wing color variation to test the role of lamina thickness in generating butterfly color . We show that butterflies in the genus Junonia thoroughly exploit the relationship between film thickness and color , using the thin films necessarily present in their scales to produce a broad spectrum of hues by tuning lamina thickness . These lamina colors work in tandem with pigments to define the wing pattern elements that distinguish populations , sexes , and species , indicating that the ability to vary lamina thickness has been an important microevolutionary and macroevolutionary tool in this group , and likely in butterflies more broadly .
Here we describe a novel instance of rapid , artificially selected color shift from brown to blue wing color in J . coenia buckeye butterflies ( Figure 1D–E ) and identify the structural changes that enabled the color shift . Edith Smith , a private butterfly breeder , began selectively mating buckeyes with a few blue scales on the costal margin of the dorsal forewing ( E . Smith , personal communication , Sep . 2014 ) . After five months of selective breeding , blue spread to the dorsal hindwing of some individuals . By eight months , there was a noticeable increase in blue surface area , and within roughly 12 months ( on the order of 12 generations ) , most butterflies in the breeding colony were visibly blue over the majority of their dorsal wing surface . On the forewing , areas proximal to M1 were visibly blue , except the discal bars ( Figure 1—figure supplement 1 ) . On the hindwing , blue shift did not include the distal-most wing pattern elements , namely EI-EIII and eyespots . At its strongest , the phenotype may include blue scales cupping the posterior forewing eyespot and/or a blue sheen in all distal elements of the forewing . Smith maintained the blue colony for several years , introgressing a few progeny from crosses to wild-caught buckeyes about once per year to maintain genetic diversity . Over time , she noted the emergence of a variety of short-wavelength colors , ranging from purple to green . Two years after focused selection , she estimated that the population was 85% blue , 8% green , 2% purple , and 5% brown . Like many familiar examples of human selection ( e . g . domesticated animals , crop plants ) , outcomes are informative even without complete experimental documentation of the selective process ( Akey et al . , 2010; Wright et al . , 2005 ) . These selected blue buckeyes provide a previously unexploited opportunity to study structural color . They demonstrate rapid and extensive evolutionary color change , and make a stark contrast to wild-type brown populations with which they are still interfertile . Conveniently , the artificially selected taxon , J . coenia , is a recognized model species for butterfly developmental genetics ( Carroll et al . , 1994; Nijhout , 1980b ) . The selected blue individuals resemble naturally evolved color variants in the sister species , J . evarete ( Figure 1F ) , and offer a useful comparison to a previously reported artificial selection experiment in butterflies ( Wasik et al . , 2014 ) . To pinpoint the cause of blueness in artificially selected butterfly scales , we characterized cover scales from the dorsal hindwing ( Figure 2A–D ) . Butterfly wings have two classes of scales arranged in alternating rows that form two layers: superficial cover scales and underlying ground scales . Cover and ground scales frequently have contrasting size , shape , and color , and their juxtaposition can be important for wing color ( Stavenga et al . , 2014 ) . When isolated and laid in the abwing orientation they occupy on the wing ( i . e . ridges facing up ) , cover scales were blue ( Figure 2B ) . However , when flipped over and viewed in adwing orientation , which exposes only the lower lamina , scales appeared more brightly blue and iridescence was more apparent ( Figure 2B’ and D ) . We tested whether the blue was structural rather than pigment-based by immersing the full scale in oil with a refractive index matched to that of chitin ( Figure 2B’’’ ) . Index-matching eliminates the possibility of reflection and structural color , leaving only pigment-based coloration . We measured the scale’s absorption spectrum under these conditions ( Figure 3A ) , which revealed that blue scales did have some pigment , presumably a brown ommochrome ( Nijhout and Koch , 1991 ) , but this pigment cannot account for blueness . The pigment was located in the scale ridges ( Figure 2—figure supplement 1B ) . Lepidopteran structural colors may occur in the lamina , lumen , ridges , or crossribs . To isolate which of these features had the nanostructure responsible for blue structural color , we dissected the scales ( Figure 2B” , Figure 2—figure supplement 1A ) . After removing all other scale components , we found that the bare lower lamina was sufficient for blue structural color ( Figure 2B” ) . We also examined regions with all scale components except the lamina and found that these pieces of lamina-less scale were not blue ( Figure 2—figure supplement 1C ) . We thus focused on investigating nanostructure in the lamina . To discern between a single or multilayer lamina and take precise measurements , we cross-sectioned the lamina and viewed it with Helium Ion Microscopy ( HIM ) ( Figure 2C ) . HIM imaging indicated the lower lamina was a simple monolayer of chitin with a thickness of 187 ± 18 nm ( SD , Figure 1C ) , which is a reasonable thickness to reflect blue as a dielectric thin film ( Stavenga et al . , 2014 ) . We next investigated whether ground scales also contributed to blueness after artificial selection . The ground scales generally had similar architecture to the cover scales , but with less uniform lamina color: ground scales exhibited a color gradient from the stalk outward ( Figure 4A–B’ ) . Correspondingly , ground scales had a similar mean thickness but more variability than cover scales ( 197 ± 31 nm ) . Ground scales were much more heavily pigmented than cover scales ( Figure 3B , Figure 4B” ) , such that the abwing surface was black ( Figure 4B ) . The extra pigmentation in ground scales enhances spectral purity by absorbing light transmitted through the cover scales , thus reducing backscatter and making the observed blue color more saturated , ( similar to Siddique et al . , 2016 ) . We conclude that cover scale laminae are the major source of blueness in artificially selected buckeye butterfly scales , while melanic ground scales secondarily enhance spectral purity . For comparison , we tested the source of color in wild-type brown scales and found that they also had structural color ( Figure 2E–H ) . Brown cover scales had the same general architecture and no significant difference in the amount of brown pigment compared to blue cover scales ( Figure 3A , Type III nested Analysis of Variance ( nested ANOVA ) , Figure 3—source data 1 ) . The salient difference was lamina thickness: brown scales were markedly thinner , measuring only 107 ± 14 nm ( nested ANOVA , p=0 . 00157 , Figure 1C , Figure 2G ) . A 107 nm chitin thin film reflects a desaturated golden color due to reflectance of many long wavelengths . This golden structural color was confirmed by the adwing scale color , the color of the bare lamina in dissected scales , and the adwing reflectance spectra of brown scales ( Figure 2F’–F” , H ) . Therefore , though brown coloration is often attributed to pigmentation , wild-type brown cover scales also had a structural color , one simply tuned to enhance different wavelengths . Artificial selection also altered the ground scales relative to wild type ( Figure 4A–D ) . Selected ground scales were significantly more absorbing than wild-type ( brown ) ground scales ( Figure 3B , nested ANOVA , Figure 3—source data 1 ) , which is consistent with increased pigmentation that decreases backscatter in blue wing regions . The wild-type ground scales were thinner on average than the blue ground scales , although the difference was not statistically significant ( wild type 156 ± 33 nm; selected 197 ± 31 nm , nested ANOVA , p=0 . 37 , Figure 1C ) . This comparison was complicated by the fact that although rainbow lamina colors in optical images suggest that each ground scale likely had highly variable thickness ( Figure 4B’ , D’ ) , we were only able to take measurements from a single cross-section position along any given scale’s proximodistal axis . Consequently , our dataset seriously underestimates the within-scale variation and overestimates the between-scale variation , making this model poorly equipped to detect any differences between groups due to potential masking of the effect . A more conclusive test would require a way to thoroughly sample thickness variation within each ground scale and larger sample sizes . Our results qualitatively suggest that if artificial selection increased lamina thickness in ground scales , the effect was likely less extreme than in cover scales: blue cover scales were on average 80 nm thicker than wild-type , while the observed mean thicknesses of selected and wild-type ground scale laminae differed by only 41 nm . We conclude that the artificially selected buckeye butterflies rapidly evolved blue wing color via a 74% mean increase in lamina thickness in cover scales and , possibly , a modest increase in ground scales . The effect was further amplified by increased pigmentation in ground scales , but without removing brown pigment from cover scales . Our results show that structural color can evolve quickly by modifying one dimension of an existing structure , and the process is facilitated by the initial presence of previously unrecognized structural color in wild-type brown J . coenia . Since the artificially selected J . coenia wing pattern resembles natural iridescent variants in the sister species , J . evarete ( Figure 1F ) , we obtained hindwings from brown and blue J . evarete individuals from different geographic locations and tested whether blue cover scales in this species were also associated with increased lamina thickness ( Figure 2I–P ) . We found that the same mechanism explained color differences between geographic color variants: lamina thickness distributions from brown and blue scales were non-overlapping , with blue scales having 78% thicker scale laminae on average ( blue 199 ± 14 nm; brown 112 ± 13 nm; Figure 1C ) . Brown and blue scales had generally similar pigmentation , although brown scales were somewhat more absorbing at short wavelengths ( Figure 3C , Figure 3—source data 1 ) . Furthermore , in blue J . evarete , the ground scales were darkly pigmented . Thus , the artificially selected blue buckeyes recapitulate natural variation at the level of scale coloration between sister species . Recently , Zhang et al . used CRISPR/Cas9 to generate mosaic knockout mutants of optix ( Zhang et al . , 2017 ) , a gene previously associated with pigment variation in butterfly wings ( Reed et al . , 2011 ) . Surprisingly , in addition to pigmentation phenotypes , optix mutants in J . coenia gained blue iridescence in wing scales . We tested phenotypically mutant blue scales from mosaic butterflies generated by Zhang et al . to determine what structural or pigmentary changes created the color change ( Figure 2Q–T ) . Where blue scales occured in the background region of the dorsal wing , blueness was due to similar factors as identified in artificially selected buckeye scales . Blue scales had markedly thicker laminae than wild-type brown scales ( 212 ± 11 nm , Figure 1C ) . The concentration of brown pigment in the cover scales was significantly reduced relative to wild-type scales within the same mosaic wing ( Figure 5A , Mann-Whitney U , Figure 5—source data 1 ) but comparable to pigmentation in artificially selected butterfly scales ( Figure 3A , Figure 3—source data 1 ) . Ground scales ( Figure 4E–F” ) were likewise similar to selected blue ground scales , having thick and variable laminae ( 199 ± 31 nm , Figure 1C ) and significantly increased pigmentation ( Figure 5B , Mann-Whitney U , Figure 5—source data 1 ) . Overall , blue scale identity in optix mutants was caused by similar mechanisms as artificially selected blue . optix mutant phenotypes also affected structural colors and pigments differently across wing pattern elements . As originally postulated ( Zhang et al . , 2017 ) , excess melanin was produced in some ventral wing regions ( Figure 6A–D , Figure 5C ) . We also observed regions where both pigment and structure were dramatically changed . For example , discal bars on the dorsal forewing , which are normally orange , gained blue scales through both converting lamina structural color to blue and replacing orange with brown pigment ( Figure 6E–H , Figure 5D ) . The kinds of pigmentation effects were diverse: optix mutation increased the quantity ( Figure 5B , C ) , decreased the quantity ( Figure 5A ) , or switched the identity ( Figure 5D ) of the pigment in different scales ( Mann-Whitney U , Figure 5—source data 1 ) . Because the butterflies were mosaic mutants , some of this phenotypic variability could be due to genotypic differences between clones ( i . e . mono- versus biallelic gene deletion , as well as the exact size of the deletion ) ( Zhang et al . , 2017 ) . However , much of the variation in outcome could also be observed within single clones that spanned multiple wing pattern elements ( defined by the Nymphalid ground plan [Nijhout , 1991] , Figure 1—figure supplement 1 ) , suggesting that the patterning roles of optix are quite context specific . In summary , optix knockout can have varied effects in a single scale by altering pigmentation , nanostructures , or both . These findings are consistent with optix’s described role as a developmental patterning gene that determines gross switches between discrete scale fates , and which , directly or indirectly , can regulate diverse downstream factors ( Martin et al . , 2014 ) . Since appropriate coloration critically depends on the proper combination of pigment and structural colors in both cover and ground scales ( e . g . Wilts et al . , 2011; Wilts et al . , 2017 ) , it is of particular interest that optix can regulate all of these components simultaneously . optix mosaic knockout mutants demonstrate that lamina thickness can be experimentally perturbed and highlight a multifunctional candidate genetic pathway for coordinated color evolution . Relatives of J . coenia exhibit extensive color and pattern diversity , and blue structural colors in particular show patterns of variation that hint at ecological relevance ( e . g . sexual dichromatism , seasonal polyphenism ) ( Figure 7A ) . To assess the importance of lamina thickness variation in macroevolutionary color diversity , we sampled cover scales from nine species in the genus Junonia and a tenth species , Precis octavia , which belongs to the tribe Junoniini and exhibits seasonally polyphenic wing coloration . We prioritized large pattern elements that distinguish color forms within species . We compared scales using optical imaging , immersion index-matching , spectrophotometry , and Helium Ion Microscopy . All scales sampled had typical Nymphalid scale structure with a single plane of chitin forming the lower lamina . We tested whether the relationship between lamina thickness and color that we observed in experimental contexts applies more broadly . We sought to address two questions: First , does lamina thickness reliably predict lamina color , as measured from the adwing surface ? While it is known that the thickness of a dielectric film controls the film’s reflectance , other variables such as refractive index , surface roughness , and pigmentation within the film also factor into reflectance , and these could plausibly vary among taxa . Second , how variable is lamina thickness ? What range of thicknesses occur , and is there evidence for either quantized or continuous thickness variation ? To address these questions , we measured reflectance spectra from the adwing surface of disarticulated cover scales from the 23 wing regions indicated in Figure 7A . We then cross-sectioned scales , imaged with HIM , and measured thickness . We found that lamina thickness varied continuously between 90–260 nm , indicating that all thicknesses over a more than 2 . 5-fold range are accessible ( Figure 8A ) . To better visualize the relationship between thickness and lamina color , we clustered similar samples into five color groups ( Materials and methods ) . Lamina colors in these groups could be described as gold , indigo , blue , and green , with a fifth variable group that included magenta , copper , and reddish colored scales ( labeled as ‘red’ in Figure 8 ) . Thickness differed significantly between all color group pairwise comparisons ( Figure 8A , nested ANOVA: p<2×10−16 , with post hoc Tukey’s Honestly Significant Difference test: p<3×10−8 for all pairwise comparisons ) . The color groups were also associated with different reflectance profiles ( Figure 8B ) . In some cases , we obtained variable measures within individual specimens , which reflects biological color variation between adjacent scales , as well as varying color within individual scale laminae along their proximal-distal and lateral axes . A particularly striking example of the latter came from J . atlites . While the wing appeared light gray , at higher magnification individual scales could be seen to be multicolored ( Figure 7G’ ) , and thickness measures from J . atlites overlapped the ranges of all color groups ( Figure 8A , see further analysis below ) . Lamina thickness had a consistent relationship with adwing scale reflectance for the taxa and color range we sampled . The order of color shift as lamina thickness increased followed Newton’s series , which is the characteristic color sequence for thin films ( Mason , 1927; Shevtsova et al . , 2011 ) . This sequence can be understood in terms of an oscillating thin film reflectance function , which shifts toward longer wavelengths as film thickness increases ( Figure 8C–G ) . The thinnest films appeared gold due to reflectance of all the longer wavelengths ( Figure 8C ) . In mid-thickness laminae , a mix of two oscillations determined color: reflectance of the first oscillation was shifted toward far red wavelengths , while a second reflectance peak rose in the ultraviolet ( Figure 8D ) . Visible reflectance of thicker laminae was dominated by the peak of the second oscillation as it moved from indigo to green ( Figure 8E–G ) . That the trend between thickness and reflectance holds broadly suggests that color changes in Junonia butterfly scales have recurrently evolved via lamina thickness adjustments . Moreover , the consistency of the relationship between thickness and reflectance is useful . For example , structural variation could be rapidly surveyed by extracting fitted thickness estimates from reflectance measurements , a much less laborious process than sectioning for electron microscopy . We next tested whether the extensive variation in lamina structural color among Junonia butterflies , explained by lamina thickness , also drives variation in overall wing color . An alternative hypothesis would be that composite wing color is usually dominated by pigmentation , particularly by pigments distributed on the outward-facing abwing surfaces of cover scales , above the lamina thin film . We measured pigmentation in cover scales from the same regions ( Figure 7A ) to test the relative importance of pigments and lamina structural colors for wing color . ( Structural colors and pigments are listed per each specimen in Supplementary file 1 and representative examples are shown in Figure 7B–M” . ) Pigmentation was highly variable among Junonia species ( Figure 7B–M” , Figure 9 , Supplementary file 1 ) . This included marked differences in pigmentation between regions of a single wing ( e . g . yellow and blue regions in J . hierta , Figure 7B–E” and 9A ) and also variation between color forms and species throughout the genus ( e . g . between sexes in J . orithya , Figure 9C , and seasonal forms in P . octavia Figure 7H–M” and 9B ) . Absorbance spectra varied in both shape and magnitude . Variation in magnitude , such as between the red band and the wet season morph of P . octavia ( Figure 9B ) , represents differences in pigment abundance . We also observed distinct absorbance spectral shapes , which can indicate the identity of the pigment ( for example , contrast the spectral shape of the yellow pigment in J . hierta , Figure 9A , versus red pigment in P . octavia , Figure 9B , versus brown pigment in J . orithya , Figure 9C ) . Notwithstanding the clear importance of pigmentation among Junonia butterflies , pigment variation was insufficient to explain the breadth of wing color diversity , and lamina structural colors made up the shortfall . The importance of lamina structural color was most obvious in scales that entirely lacked pigments . For example , the blue basal aura regions of male J . westermanni , J . hierta , and J . oenone wings had unpigmented cover scales with structurally blue laminae ( Figure 7B–C” , Figure 9A ) . Most of the pigmentless scales we sampled were blue , with the notable exception of J . atlites scales ( Figure 7F–G” , Figure 9D ) . These scales had rainbow gradient laminae , which presumably create the overall light gray by additive color mixing ( Vukusic et al . , 2009 ) . J . atlites demonstrates that lamina structural color can fundamentally drive wing color even in neutrally colored wing regions that are not obviously iridescent , and also that thickness can be patterned at fine spatial resolution within a single lamina . In most wing regions , color was determined by the interaction of both lamina structure and pigments . For example , in the cover scales of J . hierta ( Figure 7D–E” , Figure 9A ) , the yellow lamina structural color and yellow pigment were mutually reinforcing , with the lamina sensibly reflecting wavelengths that the pigment does not absorb . Other examples help delineate how much pigment is required to overpower the lamina color . In blue J . evarete , pigments in the cover scale ridges absorbed approximately 0 . 2 AU ( Absorbance Units , that is 37% of light not transmitted , Figure 3C ) of the blue wavelengths that the lamina reflected most brightly ( Figure 2L ) . With this ratio , wing hue was still driven by the lamina structural color . The cover scales of J . orithya were similar ( Figure 9C ) , having a neutral dark pigment ( i . e . a pigment that absorbs all visible wavelengths ) in the scale ridges . Perhaps dark pigment in the ridges functions like a Venetian blind to limit iridescence , so that at high viewing angles , where iridescence would be most pronounced , light from the lamina is quenched . Because of their range of pigment concentrations , P . octavia specimens were also useful to test the tradeoff between pigment abundance and lamina color influence . When viewed at high resolution , scales from the wet season morph of P . octavia contained red pigment in the ridges and crossribs ( max absorbance 0 . 12 ± 0 . 02 , Figure 7K , K’ , Figure 9B ) , while reflected light from the blue lamina spilled through the windows between ridges . Viewed macroscopically , this combination made a lightly saturated red . To display a richly saturated red , much more pigment was required , as seen in the red band of the dry season morph ( max absorbance 0 . 38 ± 0 . 04 , Figure 9B , Figure 7L–M” ) . These reddest scales also had thinner , structurally magenta and copper colored laminae that may further reinforce redness ( Figure 8A , Figure 7M’ ) . The concentration of red pigment was the most important driver of the color difference between P . octavia seasonal morphs . The blue and red morphs had only a subtle difference in lamina thickness ( Figure 8A ) , and the laminae of both were blue ( Figure 7I’ , K’ ) , but the blue morph lacked any red pigment ( Figure 9B ) . Overall , Junonia wing color was determined by complex mix-and-matching of different lamina thicknesses and pigments . A thin film lower lamina was present in all scales , but its influence on wing color was adjusted by the amount and placement of pigment , especially in the upper surface of the scale . Pigments can mask lamina structural color at high enough density , depending on the placement and color of the pigment as well as the color of the lamina . In our tests , when pigmentation absorbed ≤0 . 2 AU of the relevant wavelengths , it did not cancel out lamina structural color . We compared our empirical data to Fresnel’s classical thin film equations , which model the reflectance of an idealized dielectric thin film ( Yeh et al . , 1978; Fresnel , 1834 ) . This model has previously been used to estimate the thickness of butterfly scale laminae based on their adwing reflectance spectra ( Stavenga et al . , 2014; Wilts et al . , 2017 ) . For each sample , we modeled the expected reflectance using our thickness measurements , and then compared to the measured reflectance spectra . We used 1 . 56 for the refractive index of chitin ( Vukusic et al . , 1999 ) and a maximal angle of illumination of 30° following Stavenga ( 2014 ) ( because spectra were measured through an objective lens with a numerical aperture of 0 . 5 ) . To account for measurement error , we modeled films over all thicknesses within one standard deviation of the measured mean per sample ( red envelopes , Figure 8—figure supplement 1A ) . We also modeled films with Gaussian thickness distributions for each sample , following Siddique et al . ( 2016 ) . This model is analogous to a single uneven film with mean thickness and surface roughness defined by the measured thickness and sample standard deviation ( solid red lines , Figure 8—figure supplement 1A ) . We found that qualitatively the model describes the main behaviors of our data: reflectance oscillates with a given frequency and brightness , and the function shifts toward longer wavelengths as thickness increases , causing perceived color to cycle through Newton’s series . Quantitatively , mean maxima and minima in the reflectance function were offset laterally for every specimen , by about 40–80 nm , with the modeled curves blue-shifted relative to the observed . A similar blue shift has been reported in butterfly scale laminae before ( Wasik et al . , 2014 ) . The comparison improves if we assume a higher refractive index or thickness . However , to align modeled and measured spectra would require either an impossibly high refractive index ( around 1 . 75 ) or increased thickness outside the error range of our measures ( 20–25 nm thicker than mean measurements ) . Possibly the lateral offset is due to a combination of the former . Alternatively , these results could indicate that scales have additional properties not fully described by the model . There are a number of differences between the idealized film and real scales , including curvature of the film and possible birefringence of the ridges . The lamina itself may not necessarily have a uniform material composition or refractive index . For example , contrasting sublayers within the lamina ( as in Trzeciak et al . , 2012 ) could create extra reflective interfaces . Thus , our data are compatible with the expected behaviors of thin films , but modeling the specific case of butterfly scale laminae with quantitative precision may require additional parameters or calibration to an empirical dataset . Modeled spectra also revealed that lamina thin films are likely UV reflective . Our spectral measurements were restricted to 400–700 nm , which roughly corresponds to the human-visible range and is adequate for demonstrating the relationship between morphology and reflectance . However , the modeled spectra show sub-400 nm reflectance of varying brightness for many lamina thicknesses that occur in Junonia ( Figure 8—figure supplement 1A ) . Ultraviolet spectral bands are biologically relevant , since butterflies are known to have UV-sensitive photoreceptors ( Briscoe , 2008 ) , and this UV reflectance should be kept in mind when considering the possible ecological implications of lamina structural colors .
This study leverages the simplest photonic nanostructures , thin films , to interrogate the evolution and genetic regulation of structural color in Junonia butterfly scales . While there is a large body of literature attributing optical properties to various biological nanostructures , such claims commonly rest on correlation between mathematical models and spectral measurements . Here , we use three different experimental manipulations of the structure ( dissection , artificial selection on wing color , and knockout of the optix gene ) in addition to broad interspecies comparisons to establish that lower lamina thickness quantitatively controls structural color wavelength in Junonia butterfly scales . The relationship between lamina thickness and wavelength holds over a wide range of thicknesses ( 90–260 nm ) that generate Newton’s color series for dielectric thin films . Moreover , lamina structural color is one important determinant of overall wing color , including in wing regions that also contain pigments . Lamina structural colors contribute to the color differences that distinguish sexes , species , seasonal variants , and selectively-bred lineages of Junonia butterflies , highlighting that quantitatively tuning lamina thickness is a vehicle for color evolution in both micro and macroevolutionary contexts . Because the lower lamina is part of the typical architecture of butterfly scales , our findings have broad implications for future research on adult color in numerous butterfly taxa . Foundational literature drew a distinction between highly derived scales with vivid structural colors and ‘standard , undifferentiated scales , ’ which conform to the butterfly scale Bauplan , have a simple monolayer lower lamina , and ‘are not truly iridescent , that is they do not produce brilliant structural colors’ ( Ghiradella , 1991 ) . However , within the past ten years , individual examples of thin film interference from the lower lamina have emerged in diverse Lepidoptera , including in simple scales ( Stavenga et al . , 2014; Wasik et al . , 2014; Siddique et al . , 2016; Wilts et al . , 2017; Trzeciak et al . , 2012; Stavenga et al . , 2018; Giraldo and Stavenga , 2016 ) . These newer descriptions and our thorough examination of many scales indicate two points: first , although thin films are indeed less brilliant than some other classes of Lepidopteran photonic structures ( thin films only reflect around 20% of incident light ) , they are a consequential source of structural color . Second , thin films occur in many butterfly and moth lineages and likely arose early in Lepidopteran evolution . The lower lamina has a thin film morphology in all scales that resemble the scale Bauplan , meaning that reflectance from the lamina is the shared condition except where it is masked by either heavy pigmentation or a derived structure with higher optical contrast . Because butterflies commonly produce multiple lamina colors across wing pattern elements and scale types , it is probable that the developmental genetic networks for quantitatively varying lamina thickness are deeply conserved as well . Hence , it will be useful to report which lamina colors are present , in addition to identifying pigments , when describing butterfly colors . Physical constraints inherent to thin film colors may help explain the division of color space between pigments and photonic structures . It is not well understood why certain hues seem to be more often produced by pigments while others are more often produced by structural colors ( e . g . the abundance of blue structural colors but lack of blue pigments in birds [Stoddard and Prum , 2011] and the rarity of one class of red structural color in birds and beetles [Magkiriadou et al . , 2014] ) . In Junonia , we show that by tuning thickness , thin film laminae can produce nearly all the spectral colors ( i . e . yellow , green , blue , indigo , UV ) , and even light achromatic colors ( e . g . light gray in J . atlites ) via color mixing across a gradient . Yet thin films are fundamentally incapable of producing certain colors , notably dark brown , black , and pure red . The medium thickness films that most nearly approach red have inherently poor color properties due to the oscillating nature of the thin film reflectance function . Since the colors of mid-thickness films are a mix of two reflectance peaks ( Figure 8C ) , they are reddish but not pure or well-saturated , and are better described as copper , magenta , and purple . Further , mid-thickness films are not bright: they reflect less total visible light than other thicknesses we observed ( compare Figure 8D–8C , E–G ) . By contrast , red , black , and brown are prevalent pigment colors in Junonia , making pigments and thin film structural colors complementary color palettes with little overlap . The optical limitations of thin films may have partially determined how pigment families and scale architecture evolved in early butterfly lineages , which in turn initialized whether pigments or structures provide the most accessible route to evolve specific hues during subsequent diversification . Our findings uncover a link between artificially selectable responses in lamina thickness and natural butterfly color variation , and expand on a previous artificial selection study on butterfly wing color ( Wasik et al . , 2014 ) which selected for violet structural color in Bicyclus anynana . In both J . coenia and B . anynana , color shift was accomplished by modifying the dimension of an existing structure , the lower lamina , with pigmentation being less important . Since the selected taxa diverged 78 million years ago ( Wahlberg et al . , 2009 ) this similarity may be informative about evolvability in nymphalid butterflies generally . However , artificial selection in B . anynana primarily increased thickness in the obscured layer of ground scales , which can only weakly influence color , whereas Bicyclus species with naturally evolved violet wing color have violet thin films in their cover scales . In our study , artificial selection continued longer ( 12 vs . 6 generations ) and elicited a more extreme response ( 74% vs . 46% increase in lamina thickness ) . Moreover , in J . coenia , we show that lamina thickness increased in the cover scales and fully recapitulated the naturally evolved mechanism of structural color in the sister species J . evarete . The thickness increases caused a stark wing color change plainly visible by eye , with appropriate wing patterning that also resembled J . evarete ( thickened blue scales filled the background dorsal wing , while eyespots , distal pattern elements , and the ventral wing were unaffected ) . Our results robustly connect a rapid microevolutionary process to macroevolutionary diversity . By using butterflies with CRISPR/Cas9-generated knockout of the optix gene , we are able to provide insight into the genetic regulation of lamina thin films . It was previously known that the optix wing patterning gene can regulate a switch between wild-type brown and blue iridescent wing color in J . coenia ( Zhang et al . , 2017 ) , but the mechanistic basis for the color switch remained unknown . Specifically , it was unclear whether optix regulated scale structure itself , or whether optix deletion merely caused the loss of brown pigment , thus unveiling a pre-existing iridescent structure . Here , we show explicitly that in certain wing regions and scale types , optix deletion substantially increases lamina thickness . Our findings also amend the earlier conclusion that optix represses structural coloration in J . coenia ( Zhang et al . , 2017 ) . Rather , by regulating lamina thickness , optix regulates the wavelength of a photonic structure that exists in both wild types and mutants . This distinction has implications for the likely identities and behavior of downstream genetic factors , as well as the developmental basis of mutant blue coloration . For example , rather than preventing a cascade of downstream genes from acting to erect a photonic structure de novo , optix may subtly regulate the expression of a gene or genes that directly regulate lamina thickness , such as chitin synthase . Additionally , we uncover disparate effects of optix deletion on pigmentation , including promoting , suppressing , and switching the identity of pigments in different scale types . In aggregate , these results show that optix’s functions in J . coenia are highly context specific , depending on both wing region and scale type ( i . e . ground or cover scale ) . Moreover , because optix can regulate both pigmentary and structural color , the optix pathway is an especially interesting candidate for coordinated color evolution , and further work on the detailed regulation of optix and its downstream targets is called for . The possibility that optix plays a role in patterning blue wing regions in the artificially selected J . coenia is especially intriguing , and motivates future investigation into whether the optix locus , or other loci in the optix pathway , were the targets of artificial selection in that population . In summary , thin film reflectors , a morphologically simple class of photonic structures , are experimentally manipulable and broadly employed in the lower lamina of Junonia butterfly wing scales . Lamina thickness explains variation in structural color wavelength , responds to selection on wing color , and is regulated by the optix wing patterning gene . Tuning lamina thickness facilitates both microevolutionary and macroevolutionary shifts in wing color patterning throughout the genus Junonia , making the buckeye butterflies a promising study system with which to decipher the genetic and developmental origins of structural color .
Reared J . coenia were fed fresh Plantago lanceolata or artificial diet ( Southland Products , Lake Village , AK ) as larvae and kept at 27–30°C on a 16/8 hour day/night cycle . Artificially selected blue J . coenia were purchased as larvae from Shady Oak Butterfly Farm in 2014 ( Brooker , FL ) . Wild-type J . coenia were from an established laboratory colony , originally derived from females collected in Durham , North Carolina ( Nijhout , 1980a ) or were collected in California . We acquired preserved specimens from various vendors and collaborators ( Supplementary file 1 ) , including optix mutant butterflies from Zhang et al . ( 2017 ) . Species-level identification was generally unambiguous . However , relationships among Neotropical Junonia are not well-resolved and the limited molecular data available do not cleanly support current designations ( Neild and D'Abrera , 2008; Pfeiler et al . , 2012; Gemmell et al . , 2014 ) . Two recognized species , J . evarete and J . genoveva , have large ranges with extensive overlap and many variable color forms , including both brown and blue . We therefore described three Neotropical specimens as belonging to the J . evarete species complex to avoid accidental misidentification . Available diagnostic details , including ventral antenna club color and full collection details , are in Supplementary file 1 . Scales were laid on glass slides . Optical images of scales were taken with a Keyence VHX-5000 digital microscope ( 500-5000x lens ) . For refractive index matching , we used immersion oil ( nD = 1 . 56 ) from Cargille Laboratories ( Cedar Grove , New Jersey ) , and imaged with transmitted light . Scales were dissected by hand using a capillary microinjection needle . Whole wings were also imaged on the Keyence VHX-5000 , using the 20-200x lens . For reflectance spectra , individual scales were laid flat on a glass slide , with the adwing surface facing up . We collected spectra of the adwing surface with an Ocean Optics Flame-S-UV-Vis-Es spectrophotometer mounted on a Zeiss AxioPhot reflected light microscope with a 20x/0 . 5 objective and a halogen light source . We took two technical replicates of each scale , with a minimum sample size of 3 scales per specimen . Measurements were normalized to the reflectance of a diffuse white reference ( BaSO4 ) . Data were recorded with SpectraSuite 1 . 0 software with three scans to average and a boxcar width of 7 pixels . The software wizard determined optimal integration time from the reference sample; time was generally about 0 . 007 seconds . Spot size was roughly circular , 310 μm in diameter , and centered on the scale . We processed spectra in RStudio 1 . 0 . 153 with the package ‘pavo , ’ version 0 . 5–4 ( Maia et al . , 2013 ) . We first smoothed the data using the procspec function with fixneg set to zero and span set to 0 . 3 . We then normalized the data using the ‘minimum’ option of the procspec function , which subtracts the minimum from each sample . Because we use a diffuse standard and scales are specular , raw spectra overestimate reflectance . We therefore followed Stavenga ( 2014 ) in dividing spectra by a correction factor . We used a smaller correction factor of only 2 . 5 , because in our setup the scale does not fill the full field of view . Absorption spectra from scales submerged in index-matched oil were collected and processed similarly , but under transmitted light with an integration time of 0 . 01 seconds , and without the ‘minimum’ option . Surface imaging by HIM provides increased depth of field and enhanced topographic contrast compared to Scanning Electron Microscopy for a range of biological and other materials ( Joens et al . , 2013 ) , including butterfly wing scales ( Boden et al . , 2012 ) . Samples were prepared for HIM by laying the wing on a glass slide with the region of interest facing down , wetting with ethanol , and freezing with liquid nitrogen . We then promptly cross-sectioned the wing through the region of interest with a new razor blade . After the sample warmed and dried , we used a capillary microinjection needle to transfer individual cut scales onto carbon tape . Scales were placed overhanging the edge of a strip of carbon tape , with one end pressed into the tape . We optically imaged the tape strip as a color reference and then transferred the tape to the vertical edge of a 90° stepped pin stub ( Ted Pella #16177 ) . While non-conductive samples can be imaged by HIM using low energy electrons for charge neutralization , we found that the unsupported overhanging edges of our scales tended to bend due to local charging ( Allen et al . , 2019 ) . We thus sputter coated with 4 . 5–13 nm of Au-Pd using a Cressington 108auto or Pelco SC5 . Images ( secondary electron ) of the sectioned scales were acquired with a Zeiss ORION NanoFab Helium Ion Microscope using a beam energy of 25 keV and beam current of 0 . 8–1 . 8 pA ( 10 μm aperture , spot size 4 ) . We then used the line measurement tool in ImageJ software to measure lamina thickness from the micrographs . We corrected measurements for slight variations in working distance not accounted for by the software scale bar , using Tcorrect = ( Traw ) /9058 μm x d μm , where d is the measured working distance and 9058 μm is the reference working distance . Because these are point measures , and thickness and color vary extensively along the proximal-distal and lateral axes of individual laminae , we took measurements from multiple different positions along each cut scale . All thickness data were based on a minimum of 12 measures drawn from a minimum of 3 scales per specimen/treatment . Thickness of female J . westermanni scales was not measured because specimens were unavailable . Even with vertical mounting , the sectioned surface of the scale was not always perfectly perpendicular to the direction of the imaging beam , largely due to the scales’ tendency to curve . Viewing angle is critical , since measurements taken from a projected image viewed under erroneous tilt could cause systematic underestimation of thickness . We therefore tilted the microscope stage until the scale lamina was perpendicular at the measurement site , as diagnosed by observing an inflection point in lamina curvature ( i . e . a switch between the upper and lower surfaces being visible ) . Thickness was only measured at visible inflection points ( Figure 8—figure supplement 1B–D ) . We performed a tilt calibration to test the precision of our inflection point criterion and determined that an inflection point was only visible if the sample was within 4–5° of perpendicular . Since erroneous tilt is limited to 5° , thickness underestimation is limited to 1 nm . Slight overestimations are likely , due to the sputter coating . However , since sputtering was done from a direction perpendicular to the direction of measurement , sputtering primarily increased the length dimension of the scales , rather than their thicknesses . We did not adjust thickness measurements to attempt to remove slight increases due to sputtering . Note that if we had subtracted a few nanometers from each thickness measurement , then modeled and measured spectra would be offset from each other to a greater degree . The sectioned scale shown in Figure 1A was milled using the gallium ion beam of the Zeiss ORION NanoFab ( beam energy 30 keV , beam current 300 pA ) . Statistical analyses were conducted in R 3 . 2 . 2 . For Figure 8A–B , specimens were grouped following the largest natural breaks in the data for two metrics , mean thickness and weighted average reflected wavelength , which were in good agreement . For statistical tests involving lamina thickness , we used a Type III ANOVA with either treatment ( Figure 1C ) or color group ( Figure 8A ) as the fixed effect , and with the individual and scale identity corresponding to each measurement as nested random effects . We used this nested ANOVA model to account for the nonindependence among multiple thickness measurements per scale and multiple scales per individual butterfly . ANOVA was implemented with the default settings of the R package ‘lme4’ , including its application of Satterthwaite’s approximation for uneven sample sizes . For statistical tests on absorbance in Figure 3A–B , we used the same method except with only a single nested random effect for individual identity , since there were not multiple absorbance measurements per scale . We modeled the reflectance from chitin thin films as previously described by Stavenga ( 2014 ) , including integrating reflectance for values of θ from zero to the maximal angle of illumination ( i . e . averaging reflectances to simulate the inverted cone of light collected by the objective lens used in microspectrophotometry , given its numerical aperture ) . Specifically , since our objective had NA = 0 . 5 , we calculated reflectance over values of θ from 0 to 30° , multiplied by 2πθ , and then averaged over the cumulative circular surface area . For the model with Gaussian thickness distributions , we followed ( Siddique et al . , 2016 ) using n = 400 observations from the simulated thickness distribution . | From iridescent blues to vibrant purples , many butterflies display dazzling ‘structural colors’ created not by pigments but by microscopic structures that interfere with light . For instance , the scales that coat their wings can contain thin films of chitin , the substance that normally makes the external skeleton of insects . In slim layers , however , chitin can also scatter light to produce color , the way that oil can create iridescence at the surface of water . The thickness of the film , which is encoded by the genes of the butterfly , determines what color will be produced . Yet , little is known about how common thin films are in butterflies , exactly how genetic information codes for them , and how their thickness and the colors they produce can evolve . To investigate , Thayer et al . used a technique called Helium Ion Microscopy and examined the wings of ten related species of butterflies , showing that thin film structures were present across this sample . However , the different species have evolved many different structural colors over the past millions of years by changing the thickness of the films . Next , Thayer et al . showed that this evolution could be reproduced at a faster pace in the laboratory using common buckeye butterflies . These insects mostly have brown wings , but they can have specks of blue created by thin film structures . Individuals with more blue on their wings were mated and over the course of a year , the thickness of the film structures increased by 74% , leading to shiny blue butterflies . Deleting a gene called optix from the insects also led to blue wings . Optix was already known to control the patterns of pigments in butterflies , but it now appears that it controls structural colors as well . From solar panels to new fabrics , microscopic structures that can scatter light are useful in a variety of industries . Understanding how these elements exist and evolve in organisms may help to better design them for human purposes . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"evolutionary",
"biology"
] | 2020 | Structural color in Junonia butterflies evolves by tuning scale lamina thickness |
Odor attraction in walking Drosophila melanogaster is commonly used to relate neural function to behavior , but the algorithms underlying attraction are unclear . Here , we develop a high-throughput assay to measure olfactory behavior in response to well-controlled sensory stimuli . We show that odor evokes two behaviors: an upwind run during odor ( ON response ) , and a local search at odor offset ( OFF response ) . Wind orientation requires antennal mechanoreceptors , but search is driven solely by odor . Using dynamic odor stimuli , we measure the dependence of these two behaviors on odor intensity and history . Based on these data , we develop a navigation model that recapitulates the behavior of flies in our apparatus , and generates realistic trajectories when run in a turbulent boundary layer plume . The ability to parse olfactory navigation into quantifiable elementary sensori-motor transformations provides a foundation for dissecting neural circuits that govern olfactory behavior .
Fruit-flies , like many animals , are adept at using olfactory cues to navigate toward a source of food . Because of the genetic tools available in this organism , Drosophila melanogaster has emerged as a leading model for understanding how neural circuits generate behavior . Olfactory behaviors in walking flies lie at the heart of many studies of sensory processing ( Root et al . , 2008; Su et al . , 2012 ) , learning and memory ( Aso et al . , 2014; Owald et al . , 2015 ) , and the neural basis of hunger ( Root et al . , 2011; Tsao et al . , 2018 ) . However , the precise algorithms by which walking flies locate an odor source are not clear . Algorithms for olfactory navigation have been studied in a number of species , and can be broadly divided into two classes , depending on whether the organisms typically search in a laminar environment or in a turbulent environment . In laminar environments , odor concentration provides a smooth directional cue that can be used to locate the odor source . Laminar navigators include bacteria ( Brown and Berg , 1974 ) , nematodes ( Pierce-Shimomura et al . , 1999 ) , and Drosophila larvae ( Gomez-Marin et al . , 2011; Gershow et al . , 2012 ) . In each of these organisms , a key computation is detection of temporal changes in odor concentration , which drives changes in the probability of re-orientation behaviors . In turbulent environments , odors are transported by the instantaneous structure of air or water currents , forming plumes with complex spatial and temporal structure ( Crimaldi and Koseff , 2001; Crimaldi et al . , 2002; Webster and Weissburg , 2001 ) . Within a turbulent plume , odor fluctuates continuously , meaning that instantaneous concentration gradients do not provide simple information about the direction of the source . Navigation in turbulent environments has been studied most extensively in moths ( Kennedy and Marsh , 1974; David et al . , 1983; Baker , 1990; Kuenen and Carde , 1994; Rutkowski et al . , 2009 ) but has also been investigated in flying adult Drosophila ( van Breugel and Dickinson , 2014 ) and marine plankton ( Page et al . , 2011 ) . In these organisms , the onset or presence of odor drives upwind or upstream orientation , while loss of odor drives casting orthogonal to the direction of flow . An important distinction between laminar and turbulent navigation algorithms is that the former depend only on the dynamics of odor concentration , while the latter rely also on measurements of flow direction derived from mechanosensation or optic flow ( Cardé and Willis , 2008 ) . Also unclear is the role of temporal cues in turbulent navigation . Several studies have suggested that precise timing information about plume fluctuations might be important for navigation ( Baker , 1990; Mafra-Neto and Cardé , 1994 ) , or that algorithms keeping track of the detailed history of odor encounters may promote chemotaxis ( Vergassola et al . , 2007 ) , but the relationship between odor dynamics and olfactory behaviors has been challenging to measure experimentally ( Pang et al . , 2018 ) . In comparison to these studies , olfactory navigation in walking flies has not been studied as quantitatively . A walking fly in nature will encounter an odor plume that is developing close to a solid boundary . Such plumes are broader , exhibit slower fluctuations , and allow odor to persist further downwind from the source , compared to the airborne plumes encountered by flying organisms ( Crimaldi and Koseff , 2001; Crimaldi et al . , 2002; Webster and Weissburg , 2001 ) . Navigational strategies in these two environments might therefore be different . In laboratory studies , walking flies have been shown to turn upwind when encountering an attractive odor ( Flügge , 1934; Steck et al . , 2012 ) , and downwind when odor is lost ( Bell and Wilson , 2016 ) . However , flies can also stay within an odorized region when wind cues provide no direction information , by modulating multiple parameters of their locomotion ( Jung et al . , 2015 ) . Finally , walking flies have been shown to turn towards the antenna that receives a higher odor concentration ( Borst and Heisenberg , 1982; Gaudry et al . , 2013 ) . It is not clear how these diverse motor programs work together to promote navigation toward an attractive odor source in complex natural environments . Here , we set out to define elementary sensory-motor transformations that underlie olfactory navigation in walking fruit flies . To this end , we designed a miniature wind-tunnel paradigm that allows us to precisely control the wind and odor stimuli delivered to freely walking flies . Using this paradigm , we show that flies , like other organisms , navigate through distinct behavioral responses to the presence and loss of odor . During odor , flies increase their ground speed and orient upwind . Following odor loss , they reduce their ground speed and increase their rate of turning . By blocking antennal wind sensation , we show that mechanosensation is required for the directional components of these behaviors , while olfaction is sufficient to induce changes in ground speed and turning . This implies that olfactory navigation is driven by both multi-modal and unimodal sensori-motor transformations . We next used an array of well-controlled dynamic stimuli to define the temporal features of odor stimuli that drive upwind orientation and turn probability . We find that behavioral responses to odor are significantly slower than peripheral sensory encoding , and are driven by an integration of odor information over several hundred milliseconds ( for upwind orientation ) and several seconds ( for turn probability ) . To understand how these elementary responses might promote navigation in a complex environment , we developed a simple computational model of how odor dynamics and wind direction influence changes in forward and angular velocity . We show that this model can recapitulate the mean behavior of flies responding to a pulse stimulus , as well as the variability in response types observed across flies . Finally , we examine the behavior of our model in a turbulent odor plume measured experimentally in air , finding that its performance is comparable to that of real flies in the same environment . These simulations suggest that integration over time may be a useful computational strategy for navigating in a boundary layer plume , allowing flies to head upwind more continuously in the face of odor fluctuations , and to generate re-orientations clustered at the plume edges . Moreover , they suggest that multiple independent forms of sensing —flow sensing , temporal sensing , and spatial sensing— can work cooperatively to promote attraction to an odor source . Our description of olfactory navigation algorithms in walking flies , and the resulting computational model , provide a quantitative framework for analyzing how specific sensory-motor transformations contribute to odor attraction in a complex environment , and will facilitate the dissection of neural circuits contributing to olfactory behavior .
To investigate the specific responses underlying olfactory navigation , we developed a miniature wind-tunnel apparatus in which we could present well-controlled wind and odor stimuli to walking flies ( Figure 1A and B and Materials and methods ) . Flies were placed in rectangular arenas , where they were exposed to a constant flow of filtered , humidified air , defining the wind direction . Into this airflow we injected pulses of odor with rapid onset and offset kinetics , producing a front of odor that was transported down the arena at 11 . 9 cm/s . The time courses of odor concentration and air speed inside the behavioral arena were measured using a photo-ionization detector ( PID ) and an anemometer ( Figure 1E ) . Because flies were free to move about the chamber , and because the odor from takes about 1 s to advect down the arena , flies encountered and lost the odor at slightly different times . We therefore used PID measurements made a several locations in the arena to warp our behavior data to the exact times of odor onset and offset ( see Materials and methods , Figure 1—figure supplement 1 ) . We used genetically blind flies ( norpA36 mutants ) in order to remove any possible contribution of visual responses . Flies were starved 5 hr prior to the experiment , and were tested for approximately 2 hr ( from ZT 2–4 ) , in a series of 70 second-long trials with blank ( wind only ) and odor trials randomly interleaved . We observed that in the presence of 10% apple cider vinegar ( ACV ) , flies oriented upwind , and moved faster and straighter ( Figure 1C , magenta traces ) . This ‘ON’ response peaked 4 . 4±2 . 5 s after odor onset , but remained as long as odor was present . Following odor offset , flies exhibited more tortuous and localized trajectories ( Figure 1C , cyan traces ) . This ‘OFF’ response resembles local search behavior observed in other insects ( Willis et al . , 2008 ) , and persisted for tens of seconds after odor offset . These two responses are usually readily perceptible and distinguishable by observing the movements of flies during an odor pulse ( Figure 1C , Video 1 ) . On trials without odor , flies tended to aggregate at the downwind end of the arena ( Figure 1D ) . To analyze these responses quantitatively , we first noted that flies alternated between periods of movement and periods of immobility ( Figure 3—figure supplement 1A–B ) . To focus on the active responses of flies , we considered in our analyses only those periods in which flies were moving , and we established a threshold of 1 mm/s below which flies were considered to be stationary ( see Materials and methods ) . Then , we analyzed how flies’ movements changed in response to an odor pulse by extracting a series of motor parameters ( Figure 1F , see Materials and methods ) . We computed each measure both as a function of time ( Figure 1F ) and on a fly-by-fly basis for specific time intervals before , during , and after the odor presentation ( Figure 1G ) . During odor presentation , upwind velocity ( i . e . speed of flies along the longitudinal axis of the arenas ) and ground speed both increased significantly , while angular velocity and curvature ( i . e . ratio between angular velocity and ground speed ) decreased after an initial peak . This resulted in the straighter trajectories observed during odor; the initial peak observed in angular velocity and curvature corresponds to big turns performed by flies to orient upwind after odor onset . Following odor offset , angular velocity increased , while ground speed decreased , resulting in the increased curvature characteristic of local search ( Figure 1F , G ) . Since an increase in probability of reorientation has been traditionally identified as a hallmark of localized search ( Brown and Berg , 1974; Pierce-Shimomura et al . , 1999; Gomez-Marin et al . , 2011; Gershow et al . , 2012 ) , we calculated the turn probability of flies in our arena as a binarized version of curvature around a threshold of 20 deg/mm . Indeed , turn probability increased as well after odor offset ( Figure 1F , G ) . Upwind velocity also became negative after odor offset , although this response was weaker than the upwind orientation during odor , and peaked later than the changes in ground speed and curvature . Although most of the flies we tested showed ON and OFF responses as described above , we observed considerable variability between individuals ( Figure 1—figure supplement 2 ) . Individuals varied in the strength of their odor responses , with some flies exhibiting strong upwind orientation and search , while others showed little odor-evoked modulation of behavior ( Figure 1—figure supplement 2A–C ) . Motor parameters from the same individual in different trials were correlated , whereas parameters randomly selected from different individuals were not ( Figure 1—figure supplement 2D ) . Thus , the movement parameters of the ‘average fly’ depicted in Figure 1 underestimate the range of search behaviors shown by individuals , with particular flies exhibiting both much stronger and much weaker ON and OFF responses . There was a slight tendency for responses to be weaker during the first few trials; afterwards , this behavior was stable ( on average ) across the entire experimental session ( Figure 1—figure supplement 2F ) . Sighted flies of the same genetic background also showed ON and OFF responses ( Figure 1—figure supplement 3 ) , with increases in upwind velocity and ground speed during odor , and increases in angular velocity and decreased ground speed after odor offset . However , the increase in angular velocity appeared to be weaker , on average , in these flies . Together , these data indicate that apple cider vinegar drives two distinct behavioral responses: an ON response consisting of upwind orientation coupled with faster and straighter trajectories , and an OFF response consisting of slower and more curved trajectories . We next asked whether any change in behavior could be produced by odor in the absence of wind information . Previous studies have found that optogenetic activation of orco+ neurons did not elicit attraction ( Suh et al . , 2007 ) , unless wind was present ( Bell and Wilson , 2016 ) . However , modulation of gait parameters by odor has also been observed when the wind is directed perpendicular to the plane of the arena ( Jung et al . , 2015 ) . To ask whether walking flies could respond to odor in the absence of wind , we stabilized the third segment of the antennae using a small drop of UV glue . Fruit flies sense wind direction using stretch receptors that detect rotations of the third antennal segment ( Yorozu et al . , 2009 ) . This manipulation therefore renders flies ‘wind-blind’ ( Budick et al . , 2007; Bhandawat et al . , 2010 ) . We found that wind-blind flies showed severely impaired directional responses to odor and wind . Upwind velocity was not significantly modulated either during the odor or after ( Figure 2A–B , top ) . Indeed , odor-induced runs in different directions ( either up- or downwind or sideways ) could be observed in individual trajectories ( Figure 2C ) . In addition , the downwind positional bias seen in the absence of odor was reduced ( Figure 2D ) . The average arena position of wind-blind flies on no-odor trials was no different from that of intact flies in the absence of wind ( Figure 2D ) . Thus , antennal wind sensors are critical for the oriented components of olfactory search behavior . However , wind-blind flies still responded to odor by modulating their ground speed and angular velocity . Wind-blind flies increased their curvature after odor offset and also increased their ground speed during odor ( Figure 2B ) . These changes can be seen in the examples shown in Figure 2C , where flies adopt somewhat straighter trajectories during odor , and exhibit local search behavior following odor offset . These results imply that odor can directly modulate gait parameters to influence navigation in the absence of wind . Together these experiments show that olfactory navigation depends both on multimodal processing ( odor-gated upwind orientation ) , and on direct transformation of odor signals into changes in ground speed and curvature . Because natural odor stimuli are highly dynamic , we next asked what features of the odor signal drive ON and OFF responses . To address this question , we presented flies with a variety of dynamically modulated stimuli . We focused our analysis on upwind velocity and turn probability , as measures of the ON and OFF response , respectively , as these parameters provided the highest signal-to-noise ratio . We first looked at how ON and OFF behaviors depended on the concentration of the odor stimulus . In these experiments , different groups of flies were exposed to square pulses of apple cider vinegar at dilutions of 0 . 01% , 0 . 1% , 1% and 10% ( Figure 3A–B ) . We found that both upwind velocity during odor and turn probability after offset grew with increasing odor concentration between 0 . 01% and 1% , but saturated or even decreased at 10% ( Figure 3A–B ) . These responses were well fit by a Hill function with a dissociation constant κd of 0 . 072% ( for ON ) and and 0 . 127% ( for OFF; Figure 3A and B , left and right insets ) . The fitted Hill coefficient was very close to 1 ( 1 . 03 for ON and 1 . 06 for OFF ) . A saturating Hill function nonlinearity is to be expected from odor transduction kinetics , and has been found to describe encoding of odor stimuli by peripheral olfactory receptor neurons ( Kaissling et al . , 1987; Nagel and Wilson , 2011; Gorur-Shandilya et al . , 2017; Schulze et al . , 2015 ) , and central olfactory projection neurons ( Olsen et al . , 2010 ) . A decrease in response at the highest intensities could arise from inhibitory glomeruli that are recruited at higher odor intensity , as has been described in Semmelhack and Wang , 2009 ) . We next wondered whether OFF behaviors could be elicited by gradual decreases in odor concentration , as turning behavior in gradient navigators is sensitive to the slope of odor concentration ( Brown and Berg , 1974; Pierce-Shimomura et al . , 1999 ) . To perform this experiment , we used proportional valves to deliver a pulse of saturating concentration ( 10% ACV ) , that then decreased linearly over a period of 2 . 5 , 5 or 10 s ( Figure 3C–D , Materials and methods ) . We observed that turn probability began to grow gradually as soon as the odor concentration started to decrease ( Figure 3D , white arrow ) , but peaked close to the point where the linear off ramp returned to baseline ( black arrow ) . This result suggests some form of sensitivity adaptation , that allows the fly to respond to a small decrease from a saturating concentration of odor . We also noted that upwind velocity remained positive during these ramps ( Figure 3C , white arrow ) , suggesting that ON and OFF responses can be driven —at least partially— at the same time . Finally , we wished to gauge the ability of flies to follow rapid fluctuations in odor concentration , as occurs in real odor plumes . Indeed , olfactory receptor neurons can follow odor fluctuations up to 10–20 Hz ( Nagel and Wilson , 2011; Kim et al . , 2011 ) , and these rapid responses have been hypothesized to be critical for navigation in odor plumes ( Nagel and Wilson , 2011; Gorur-Shandilya et al . , 2017 ) . To test the behavioral response of flies to rapid odor fluctuations , we used proportional valves to create ascending and descending frequency sweeps of 10% ACV between approximately 0 . 1 and 1 Hz ( Figure 3E–H ) . The peak frequency we could present was limited to 1 Hz , as we found that frequencies higher than this became attenuated at the downwind end of the arena , presumably because odor diffuses as it is transported downwind , blurring the differences between peaks and troughs in the stimulus ( see Materials and methods ) . In addition , we presented a ‘plume walk’: an odor waveform created by taking an upwind trajectory at fly pace through a boundary layer plume measured using planar laser imaging fluorescence ( PLIF; Figure 3I–J , see Materials and methods , Connor et al . , 2018 ) . As in previous experiments , we warped all behavioral data to account for the fact that flies encounter the odor fluctuations at different times depending on their position in the arena ( Figure 1—figure supplement 1 and Materials and methods ) . In addition , we excluded behavioral data points within 3 mm of the side walls , where boundary layer effects would cause slower propagation of the stimulus waveform . We also excluded responses occurring after each fly reached the upwind end of the arena , where arena geometry would constrain their direction of movement . The resulting traces represent our best estimate of the time courses of behavioral parameters ( Figure 3—figure supplement 1 ) , although we cannot completely rule out some contribution of odor diffusion or arena geometry . We found that upwind velocity tracked odor fluctuations at the lowest frequencies , but that modulation became attenuated at higher frequencies ( end of the ascending frequency sweep and start of the descending frequency sweep; Figure 3E and G ) , suggesting low-pass filtering of the odor signal . Similarly , upwind velocity peaked in response to nearly every fluctuation in the ‘plume walk’ , but remained elevated during clusters of odor fluctuations ( Figure 3I ) . The frequency-dependent attenuation was seen in both ascending and descending frequency sweeps , arguing against it being an effect of position in the arena , or duration of exposure to odor . Attenuation was not due to the filter imposed on trajectories during processing , as it was visible also when this filtering step was omitted ( Figure 3—figure supplement 1C–D ) . We think it is also unlikely to be due to a limit on our ability to measure fast behavior reactions . We observed rapid decreases in ground speed in response to click stimuli that did not attenuate at higher frequencies ( Figure 3—figure supplement 1C , F ) , arguing that the attenuation seen with odor does not reflect a limit on detecting rapid behavioral responses . Turn probability at offset showed even stronger evidence of low-pass filtering . Fluctuations in turn probability were attenuated during the higher frequencies of both frequency sweeps , and the strongest responses occurred at the end of the stimulus to the absence of odor ( Figure 3F , H , J ) . The initial peaks in turn probability most likely represent the initial upwind turn , rather than an OFF response . Together these experiments provide detailed measurements of the way that ON and OFF behaviors depend on the history of odor encounters . Moreover , they suggest that the two responses depend on odor history in different ways , with rapid fluctuations leading to elevated ON responses and suppressed OFF responses . We next sought to develop computational models that could account for the behavioral dynamics described above . A challenge was that behavioral responses saturated at concentrations above 1% ACV , and they were also modulated by small decreases and fluctuations from a higher concentration ( 10% ) . This suggests some form of adaptation , in which the sensitivity of behavior to odorant shifts over time , allowing responses to occur near what was previously a saturating concentration . Sensitivity adaptation has been described at the level of olfactory receptor neuron transduction and can be implemented as a slow rightward shift in the Hill function that describes intensity encoding ( Kaissling et al . , 1987; Nagel and Wilson , 2011; Gorur-Shandilya et al . , 2017 ) . We therefore modeled adaptation by filtering the odor waveform with a long time constant τA and using the resulting signal to dynamically shift the midpoint of the Hill function to the right ( see Materials and methods ) . The baseline κd of the Hill function was taken from the fits in Figure 3A and B . We call this process ‘adaptive compression’ ( Figure 4A ) as it both compresses the dynamic range of the odor signal ( from orders of magnitude to a linear scale ) , and adaptively moves the linear part of this function to the mean of the stimulus . We then tested four models for the ON response: one with adaptive compression followed by a low-pass filter ( 'ACF' ) , one with filtering followed by adaptive compression ( 'FAC' ) , and the same models without adaptation ( 'CF' and 'FC' respectively ) . We note that the FC model , with filtering followed by a fixed nonlinearity , is most similar to traditional linear-nonlinear models . For simplicity , we parameterized the low-pass filter by a single time constant τON , that describes the amount of smoothing seen in the response ( Materials and methods ) . We first fit models of the ON response to all upwind velocities shown in Figure 3 , omitting and reserving the 'plume walk' stimulus to use as a test . We found that both models with adaptation performed better than models without , and that the model with adaptive compression first ( 'ACF' , Figure 4A ) outperformed the adaptive model with filtering first ( 'FAC' , Figure 4B ) . As shown in Figure 4C , model ACF correctly predicted saturation with increasing odor concentration , and also the fact that responses to high odor concentrations exhibit adaptation while those to low odor concentrations do not . This model also correctly predicted the attenuation seen during frequency sweeps ( Figure 4D and E ) , although some details of response timing early in the stimulus were not matched . We note that behavioral responses used for fitting were recorded in three different experiments with different sets of flies , and we used a single set of parameters to fit all responses; some differences between real and predicted response ( for example the timing of response onset in Figure 3D and E vs C ) may reflect differences in responses across experiments . The time constant of filtering was 0 . 72 s ( see Table 1 ) , significantly slower than encoding in peripheral ORNs ( Kim et al . , 2011; Nagel and Wilson , 2011 ) . The time constant of adaptation was very slow ( 9 . 8 s ) . Models without adaptation ( pink trace in Figure 4D–E ) exhibited strong saturation during the frequency sweep , which was not observed experimentally . We next fit the OFF response using four related models . In this case , the adaptive compression step was the same , but we used a differentiating filter instead of a low-pass filter , to generate responses when the odor concentration decreases from a previously high level . This filter was parameterized by two time constants , τOFF1 and τOFF2 , that describe the time intervals over which the current and past odor concentrations are measured ( Figure 4F , Materials and methods ) . Again we found that models with adaptation outperformed those without , and that the adaptive model with compression first very slightly outperformed the adaptive model with filtering first ( Figure 4G ) . This model reproduced reasonably well the responses of flies to odor ramps ( Figure 4H ) . The slow time constant of filtering was 4 . 84 s , accounting for the selectivity of the OFF response to low frequencies during frequency sweeps ( Figure 4I and J ) . The time constant of adaptation was of similar magnitude to that derived from fitting the ON response ( 10 . 62 s ) . To further assess the best-performing ON and OFF models ( those with adaptive compression followed by filtering ) , we tested the performance of these models on the 'plume walk' stimulus . We found that the ON model reproduced most major contours in the 'plume walk' response ( Figure 4K ) , although there was some discrepancy in the timing of peaks early in the response as for the frequency sweeps ( Figure 4D ) . The OFF model also captured many of the major peaks in the behavioral response ( Figure 4L ) , as well as the time course of the slow offset response after the end of the stimulus . Overall , the RMSE errors between predictions and data for the plume walks were comparable to those for the stimuli we used for fitting . We conclude that models featuring adaptive compression followed by linear filtering provide a good fit to behavioral dynamics over a wide range of stimuli . To understand how the ON and OFF functions defined above might contribute to odor attraction , we incorporated our ON and OFF models into a simple model of navigation . In our model ( Figure 5A–C ) , we propose that odor dynamics directly influence ground speed and turn probability through the ON and OFF functions developed and fit above . Specifically , ON ( t ) drives an increase in ground speed and a decrease in turn rate , leading to straight trajectories , while OFF ( t ) drives a decrease in ground speed and an increase in turn rate , leading to local search ( Figure 5B ) . Ground speed ( v ) and turn probability ( P ( t ) ) of our model flies are then defined by ( 1 ) v ( t ) =v0+κ1ON ( t ) -κ2OFF ( t ) ( 2 ) P ( t ) =P0-κ3ON ( t ) +κ4OFF ( t ) where v0 and P0 are baseline values extracted from behaving flies ( Figure 1F ) . Second , we propose that turning has both a probabilistic component , driven by odor , and a deterministic component , driven by wind . In the absence of any additional information about how these turn signals might be combined , we propose that they are simply summed . To model deterministic wind-guided turns , we constructed a sinusoidal desirability function or ‘D-function’ which drives right or leftward turning based on the current angle of the wind with respect to the fly . Such functions were originally proposed to explain orientation to visual stripes ( Reichardt and Poggio , 1976 ) . In an upwind D-function , wind on the left ( denoted by negative ψ values ) drives turns to the left ( denoted by negative θ˙ values ) , and vice-versa ( Figure 5C , magenta trace ) . Conversely , in a downwind D-function , wind on the left drives turns to the right , and vice-versa ( black trace ) . Supporting the notion of a wind direction-based D-function , we found that the average angular velocity as a function of wind direction in the period immediately after odor onset had a strong ‘upwind’ shape ( Figure 5D , magenta trace ) , while the angular velocity after odor offset had a weaker ‘downwind’ shape ( Figure 5D , black trace ) . In our navigation model , the angular velocity of the fly is then given by ( 3 ) θ˙ ( t ) =ρ ( t ) G+κ5ON ( t ) Du ( φ ) +κ6Dd ( φ ) where ρ ( t ) is a binary Poisson variable with rate P ( t ) and G is the distribution of angular velocities drawn from when ρ is 1 ( see Materials and methods ) . This first term generates probabilistic turns whose rate depends on recent odor dynamics . The second term is an upwind D-function , gated by the ON function , that produces strong upwind orientation in the presence of odor . The final term is a constant weak downwind D-function that produces a downwind bias in the absence of odor . This navigation model is parameterized by six coefficients ( κ1-κ6 ) that determine the strength with which the ON and OFF functions modulate ground speed , turn probability , and the drive to turn up- or downwind . For example , κ1 determines how much the forward velocity increases when the ON function increases by a specific amount . We first adjusted these parameters so that average motor parameters calculated from simulations of our model in response to a 10 s odor pulse would match the ground speed , upwind velocity , and turn probability of the ‘mean fly’ seen in Figure 1 ( Figure 5E , see Materials and methods and Table 2 ) . Similar to real flies , this model produced upwind runs during the odor pulse and searching after odor offset ( Figure 5F ) . Average upwind velocity during the odor and turn probability after the odor were comparable to measurements from real flies ( compare Figure 5G and Figure 1G ) . As a second test , we set the coefficients controlling wind orientation ( κ5 and κ6 ) to zero , making the model fly indifferent to wind direction and mimicking a wind-blind real fly . In this case , the model produced undirected runs during odor and search behavior at odor offset , as in our data ( compare Figure 5H–I and Figure 2A–B ) . We also asked whether our model could account for variability in behavior seen across flies ( Figure 1—figure supplement 2 ) . To address this question , we asked whether differences in behavior could be accounted for by applying fly-specific scale factors to the ON and OFF functions of the model . To define these scale factors , we returned to our main data set ( Figure 1 ) and computed an ON scale value for each fly equal to its mean upwind velocity , divided by the mean upwind velocity across flies . An OFF scale value was computed similarly by taking the mean turn probability for a fly divided by the mean across flies . This procedure allowed us to express the behavior of each fly as a scaled version of the group average response . Next , keeping all other parameters in our navigation model fixed as previously fitted , we scaled the ON and OFF functions to match the value of individual flies . The trajectories produced by these scaled models resembled the behavior of individual flies both qualitatively and quantitatively . For example , scaling down the ON and OFF functions produced similar behavior to a weak searching fly ( Figure 5J , compare directly to green-highlighted examples in Figure 1—figure supplement 2A ) , while scaling up the ON and OFF function produced behavior similar to a strongly-searching fly ( Figure 5K , compare directly to blue-highlighted examples in Figure 1—figure supplement 2A ) . Together , these results support the idea that our model captures essential features of how flies respond to odor and wind in miniature wind-tunnels , including the responses of intact and wind-blind flies , and variations in behavior across individuals . Thus , this model provides a basis for examining the predicted behavior of flies in more complex environments . Finally , we sought to test whether our model could provide insight into the behavior of real flies in more complex odor environments . To that end we constructed two equivalent wind tunnels capable of delivering a turbulent odor plume ( Figure 6A; see Materials and methods ) . In one tunnel ( New York ) , we incorporated IR lighting below the bed and cameras above it to image fly behavior in response to a turbulent odor plume . In the second tunnel ( Colorado ) , we used a UV laser light sheet and acetone vapor to obtain to high-resolution movies of the plume for use in modeling ( Figure 6B , Connor et al . , 2018 ) . These two apparatuses had similar dimensions , and matched odor delivery systems and wind speeds . We used photo-ionization detector measurements to corroborate that the shape and dynamics of the plume in the New York tunnel was similar to the one measured in Colorado ( Figure 6B ) . We next examined the behavior of walking flies in this wind tunnel . Flies were of the same genotype and were prepared for experiments in the same way as those used previously . They were constrained to walk by gluing their wings to their backs with a small drop of UV glue and by placing a 1cm-wide water-filled moat at the edge of the arena . We first tested flies with wind only ( no odor ) at 10 cm/s . As in our miniature wind tunnels , we found that flies uniformly preferred the downwind end of the arena ( Figure 6C ) . In the absence of wind , this preference was reduced ( Figure 6D ) . We observed no preference for the upwind end of the tunnel ( which received greater ambient light from the room ) or for the odor tube , confirming that these norpA36 flies lacked phototaxis and visual object attraction . Finally , we examined behavior in the presence of a plume of ACV 10% , and we observed diverse responses ( Figure 6E ) . Of 66 flies , 37 ( 56% ) successfully located the odor source , walking upwind and lingering in a small region close to the odor tube ( Figure 6E , left trace ) . Other flies searched in the middle of the arena without getting close to the source ( Figure 6E , middle trace , 18% ) , while others headed downwind and remained at the downwind end of the arena ( Figure 6E , right trace , 15% ) . The rest of the flies ( seven flies ) either moved very little or moved mostly along the sides of the tunnel . To compare the performance of our model to the behavior of the flies , we ran simulations with our model using the plume movie measured in the Colorado wind tunnel as a virtual environment ( Video 2 ) . At each time step , we took the odor concentration at the location of the simulated fly and used this to iteratively compute ON and OFF functions and update the fly’s position accordingly ( Figure 6F–H ) . We observed that model flies produced trajectories similar to those of real flies in the wind tunnel . For example , some flies responded to odor with general movement upwind interrupted by occasional excursions out of the plume ( Figure 6F ) ; overall , 66% successfully came within 2 cm of the odor source . Other model flies searched but failed to locate the source ( 17% of trials; Figure 6H , left trace ) , while others ‘missed’ the plume and moved downwind ( 17% of trials; Figure 6H , right trace ) . Using a single set of model parameters fit to the mean behavioral responses in Figure 1F , we found that our model yielded a similar —although somewhat higher— success rate than real flies ( Figure 6I , 66% versus 56% success rate ) . Given the large degree of variability in behavior across individuals , we wondered if this variability could account for the difference in success rates between real and model flies . We therefore ran simulations incorporating variability in fly behavior . In each trial of this simulation , we randomly drew a pair of ON and OFF scale values ( as described previously ) and used it to scale the ON and OFF responses of the model for that trial . Introducing variability in the model decreased the success rate to 45% ( Figure 6I ) , and made it slightly worse than that of real flies in the wind tunnel . This simulation produced 27% ‘failed’ searches and 28% trials in which flies ‘missed’ the plume and went downwind . The simulations described above indicate that the trajectories produced by our model in a turbulent environment are qualitatively similar to those produced by real flies . To gain insight into the roles that ON and OFF behaviors play in this environment , we color-coded model trajectories according to the magnitude of the ON and OFF functions underlying them ( Figure 6F–G ) . We observed that the ON function was dominant throughout most of the odorized region , while excursions from the plume elicited strong OFF responses that frequently resulted in the model fly re-entering the plume . OFF responses were also prominent near the odor source , where they contributed to the model fly lingering as observed in real flies . ON and OFF magnitudes varied over a much smaller range than the range of odor concentrations , suggesting that the adaptive compression we incorporated into the model helps flies to respond behaviorally over a greater distance downwind of the source . Plotting the strengths of both responses as a function of position in an odor plume supported this analysis of individual trajectories ( Figure 6J–K ) . This analysis showed ON being active in the area within the plume , and more active the closer to the center of the plume ( Figure 6J ) , where the concentration of odor is higher and intermittency is lower . This suggests that ON responses are responsible for making flies progress within the odor area , allowing them to eventually reach the odor source . The OFF function was most active in the area surrounding the odor plume ( Figure 6K ) , suggesting it plays a role in relocating the plume after flies walk outside of it and the odor signal is lost . OFF values were also high just upwind of the source . Notably , OFF values were generally low within the plume , even though large fluctuations do occur within this region . This suggests that the slow integration time of the OFF response may help it to detect the edges of the time-averaged plume , allowing flies to slow down and search only when the plume has genuinely been exited . To assess the relative role of ON and OFF functions in promoting source localization , we ran a series of simulations in an odor plume ( 500 trials each ) , systematically changing the scaling factors of the ON and OFF functions ( Figure 6L ) . We observed that performance increased with both functions , but that ON was more critical for success in the plume , producing large improvements in performance as it increased . This is consistent with the idea that wind direction is a highly reliable cue in this environment ( indeed , it is likely more reliable in our model than in reality , as we did not incorporate local variations in flow induced by turbulence into our model ) . To test the idea that ON and OFF might have different importance in a windless environment , we repeated the analysis just described in a simulated Gaussian odor gradient with no wind ( Figure 6M ) . In this environment , success rates were lower , but the contributions of ON and OFF were more similar , with higher success rates when the OFF function was the strongest for any given strength of the ON function . These results suggest that ON and OFF responses have different impact on success depending on the features of the environment . In addition to the ON and OFF functions described here , walking Drosophila have also been shown to perform spatial comparisons across their antennae , and to turn toward the antenna that receives a higher odor concentration ( Borst and Heisenberg , 1982; Gaudry et al . , 2013 ) . Such turns can be produced using optogenetic activation of olfactory receptor neurons in one antenna , arguing that they are independent of wind sensing ( Gaudry et al . , 2013 ) . Because the fly’s antennae are located so close to one another , and because it has been unclear what kind of spatial information a plume provides , the role of these spatial comparisons in plume navigation has been questioned ( Borst and Heisenberg , 1982 ) . To ask whether such comparisons could contribute to source finding in the boundary layer plume that we measured , we incorporated a fourth term into the total angular velocity in our model: ( 4 ) θ˙ ( t ) =ρ ( t ) G+κ5ON ( t ) Du ( φ ) +κ6Dd ( φ ) +κ7 ( Cl−Cr ) Here , Cl and Cr represent the odor concentrations at the left and right antennae , processed by the same adaptive compression function used previously ( see Materials and methods ) . The left antenna was taken to be at the position of the fly , and the right antenna was taken to be one pixel ( 740 μm ) to the right . The results of these simulations depended heavily on the choice of gain κ7 . Based on the results of ( Borst and Heisenberg , 1982 ) and ( Gaudry et al . , 2013 ) , we estimated a gain of approximately 40 deg/s when the concentration difference between the two antennae is maximal . In this case , spatial comparisons did not contribute significantly to the probability of successfully finding the source ( Figure 7A–C ) . However , if we increased the gain to 300 deg/s , we found that performance of the model improved significantly , from 67% to 76% . Under these conditions , trajectories remained closer to the center of the plume and were less dispersed around the source ( Figure 7A–B , third column ) . We observed a contrary phenomenon when we switched the position of the antennae in the model , so that information from the right side was interpreted as left , and vice-versa . This made model flies more prone to leave the area of the plume and wander off , decreasing their success rate to 54% ( Figure 7A–C , fourth column ) . In the absence of wind sensation , flies performing a correct bilateral comparison were unable to locate the odor source ( Figure 7A–C , fifth column ) . These results argue that nearby locations in the plume contain information that can be used to aid navigation ( if the gain is high enough ) , but that this information is insufficient to find the odor source in the absence of wind . To explore how performance depended on the interaction of wind sensation and spatial sensing , we varied the strength of these two behavioral components ( Figure 7D ) . This analysis showed that some wind sensing is absolutely required to find the odor source , as almost no flies find the source when the wind coefficients are set to zero . However , in the presence of wind , bilateral sensing , controlled by κ7 , improves performance , with the greatest improvements coming at the highest gain . Thus , although the contributions of wind sensing and bilateral sensing sum linearly to control angular velocity in our model , their effects on finding the source are nonlinear , presumably because of the structure of the plume itself . In addition , we asked whether both temporal sensing and spatial sensing contribute to performance in the plume . To do this , we varied the magnitude of the OFF response and the gain of bilateral sensing ( Figure 7E ) , while keeping the strength of wind sensation constant . In this case , we observed that both components contributed to increased performance . This is consistent with our observations of model trajectories , which suggest that the OFF response and bilateral sensing work together to help reorient model flies into the plume when they wander out of it . Together these results suggest that three different forms of sensation—flow sensing ( wind ) , temporal sensing ( OFF response ) , and spatial sensing ( bilateral comparisons ) —can all contribute to finding an odor source , but that the precise contribution of each mechanism depends both on the environment and on the gain or sensitivity of the animal to each measurement . These data support the idea that olfactory navigation in complex environments can be decomposed into several largely independent sensori-motor transformations and provide a foundation for investigating the neural basis of these components .
The ability to navigate toward attractive odors is widespread throughout the animal kingdom and is critical for locating both food and mates ( Bell and Tobin , 1982 ) . Taxis toward attractive odors is found even in organisms without brains , such as E . coli , and is achieved by using activation of a receptor complex to control the rate of random re-orientation events , called tumbles or twiddles ( Falke et al . , 1997 ) . Precise quantification of the behavior elicited by controlled chemical stimuli has been critical to the dissection of neural circuits underlying navigation in gradient navigators such as C . elegans ( Gray et al . , 2005 ) and Drosophila larvae ( Tastekin et al . , 2015 ) . Larger organisms that navigate in air or water face fundamentally different problems in locating odor sources ( Cardé and Willis , 2008; Murlis et al . , 1992 ) . Odors in open air are turbulent . Within a plume , odor concentration at a single location fluctuates over time , and local concentration gradients often do not point toward the odor source ( Crimaldi and Koseff , 2001; Webster and Weissburg , 2001 ) . To solve the problem of navigating in turbulence , many organisms have evolved strategies of combining odor information with flow information . For example , flying moths and flies orient upwind using optic flow cues during odor ( Kennedy and Marsh , 1974; David et al . , 1983; van Breugel and Dickinson , 2014 ) . Marine invertebrates travel upstream when encountering an attractive odor ( Page et al . , 2011 ) . Although neurons that carry signals appropriate for guiding these behaviors have been identified ( Olberg , 1983; Namiki et al . , 2014 ) , a circuit-level understanding of these behaviors has been lacking . Obtaining such an understanding will require quantitative measurements of behavior coupled with techniques to precisely activate and inactivate populations of neurons . In recent years , the fruit-fly D . melanogaster has emerged as a leading model for neural circuit dissection ( Simpson , 2016 ) . The widespread availability of neuron-specific driver lines , the ease of expressing optogenetic reagents , and the ability to perform experiments in a high-throughput manner have established the fruit-fly as a compelling experimental model . Here , we have developed a high-throughput behavioral paradigm for adult flies that allows for precise quantification of fly movement parameters as a function of well-controlled dynamic odor and wind stimuli . An important distinction between our paradigm , and others previously developed for flies ( Jung et al . , 2015; van Breugel and Dickinson , 2014; Bell and Wilson , 2016 ) , is that it allows us to control the odor and wind stimuli experienced by the flies regardless of their movement . This ‘open loop’ stimulus presentation allowed us to measure the dependence of specific behaviors on odor dynamics and history . In addition , our paradigm allows for movement in two dimensions ( in contrast to Steck et al . , 2012; Bell and Wilson , 2016 ) , which allowed us to observe and quantify search behavior elicited by odor offset . By combining this paradigm with techniques to activate and silence particular groups of neurons , it should be possible to dissect the circuits underlying these complex multi-modal forms of olfactory navigation . In our behavioral paradigm , we observed two distinct behavioral responses to a pulse of apple cider vinegar: an upwind run during odor , and a local search at odor offset . Previous studies have suggested that flies cannot navigate toward odor in the absence of wind ( Bell and Wilson , 2016 ) , while others have suggested that odor modulates multiple parameters of locomotion , resulting in an emergent attraction to odorized regions ( Jung et al . , 2015 ) . Our findings suggest a synthesis of these two views . We find that upwind orientation requires wind cues transduced by antennal mechanoreceptors . In contrast , offset searching is driven purely by changes in odor concentration . In computational model simulations , we found that when wind provided a reliable cue about source direction , wind orientation was the major factor in the success of a model fly in finding the source . However , when wind cues were absent , ON and OFF behaviors both played equal roles . In real environments , wind direction is rarely completely reliable ( Murlis et al . , 2000 ) , so both behaviors are likely to contribute to successful attraction . The ON and OFF responses that we describe here have clear correlates in behaviors described in other organisms . The upwind run during odor has been described previously ( Flügge , 1934; Steck et al . , 2012 ) and seems to play a similar role to the upwind surge seen in flying insects ( Vickers and Baker , 1994 ) . Upwind orientation in walking flies appears to depend entirely on mechanical cues while upwind orientation during flight has been shown to be sensitive to visual cues ( Kennedy , 1940; Kennedy et al . , 1981; van Breugel and Dickinson , 2014 ) . Searching responses after odor offset have been observed in walking cockroaches ( Willis et al . , 2008 ) , and have been observed in adult flies following removal from food ( Dethier , 1976; Kim and Dickinson , 2017 ) but have until recently not been reported in flies in response to odor ( Sayin et al . , 2018 ) . The OFF response seems to play a role related to casting in flying insects , allowing the fly to relocate an odor plume once it has been lost , although the response we observed did not have any component of orientation orthogonal to the wind direction , as has been described in flight ( Kennedy and Marsh , 1974; van Breugel and Dickinson , 2014 ) . OFF responses were weaker in flies lacking the norpA36 allele , suggesting that vision may be able to substitute to some degree for search behavior , or that the norpA36 allele itself promotes more vigorous searching . A common feature of chemotaxis strategies across organisms is the use of temporal cues to guide behavior . In gradient navigators , the dependence of behavior on temporal features of odor is well established . Bacteria respond to decreases in attractants over an interval of about 2 s ( Block et al . , 1982 ) . Pirouettes in C . elegans are driven by decreases in odor concentration over a window of 4–10 s ( Pierce-Shimomura et al . , 1999 ) . The temporal features of odor that drive behavioral reactions in plume navigators are less clear . Studies of moth flight trajectories in a wind tunnel have suggested that moths respond to each filament of odor with a surge and cast ( Baker , 1990; Vickers and Baker , 1994 ) , and cease upwind flight in a continuous miasma of odor ( Kennedy et al . , 1981 ) . These findings have led to the idea that the rapid fluctuations found in plume are critical for promoting upwind progress ( Baker , 1990; Mafra-Neto and Cardé , 1994 ) . In contrast , Drosophila have been observed to fly upwind in a continuous odor stream ( Budick and Dickinson , 2006 ) , suggesting that a fluctuating stimulus is not required to drive behavior in this species . Flight responses to odor have been described as fixed reflexes ( van Breugel and Dickinson , 2014 ) , although they have also been shown to depend on odor intensity and history ( Pang et al . , 2018 ) . Measurement of these dependencies has been hampered by the inability to precisely control the stimulus encountered by behaving animals . Here , we have used an open loop stimulus and a very large number of behavioral trials , to directly measure the dependence of odor-evoked behaviors on odor dynamics and history . We find that in walking Drosophila , ON behavior ( upwind orientation ) is continuously produced in the presence of odor . ON behavior exhibited a filter time constant of 0 . 72 s , significantly slower than encoding of odor by peripheral olfactory receptor neurons ( Kim et al . , 2011; Nagel and Wilson , 2011 ) . We think it is unlikely that this represents a limit on our ability to measure behavioral reactions with high temporal fidelity , as we observed very rapid , short-latency freezing in response to valve clicks that were faster and more reliable than olfactory responses . One possible explanation for this difference is that olfactory information may be propagated through multiple synapses before driving changes in motor behavior , while the observed freezing may be a reflex , executed through a more direct coupling of mechanoreceptors and motor neurons . OFF responses ( increases in turn probability ) were driven by differences between the current odor concentration , and an integrated odor history with a time constant of 4 . 8 s . This long integration time was evident in responses to frequency sweeps and to the ‘plume walk’ , where increases in turn probability were only observed in response to relatively slow odor fluctuations , or to long pauses between clusters of odor peaks . This filtering mechanism may allow the fly to ignore turbulent fluctuations occurring within the plume , and to respond with search behavior only when the overall envelope of the plume is lost . The neural locus of this offset computation is unclear . Olfactory receptor neurons that are inhibited in the presence of odor can produce offset responses when odor is removed ( Nagel and Wilson , 2011 ) ; such inhibitory responses are generally odorant specific ( Hallem and Carlson , 2006 ) . In addition , inhibition after odor offset is observed in many olfactory receptor neurons , and the dynamics of this inhibition have been shown to predict offset turning in Drosophila larvae ( Schulze et al . , 2015 ) . Alternatively , the OFF response could be computed centrally in the brain . For example , many local interneurons of the antennal lobe are broadly inhibited by odors ( Chou et al . , 2010 ) and exhibit offset responses driven by post-inhibitory rebound ( Nagel and Wilson , 2016 ) . Rebound responses grow with the duration of inhibitory current ( Nagel and Wilson , 2016 ) , providing a potential mechanism for slow integration . Experiments testing the odor and glomerulus specificity of the OFF response could be used to distinguish between these possibilities , as ORN temporal responses are specific to particular odorants ( Hallem and Carlson , 2006 ) , while LN temporal responses are similar across odorants ( Chou et al . , 2010 ) . In addition to low-pass filtering , we found that behavioral responses to odor were best fit by models that included a compressive nonlinearity—in the form of a Hill function—whose sensitivity was slowly adjusted by adaptation . This type of adaptive compression has been observed in the transduction responses of Drosophila olfactory receptor neurons ( Kaissling et al . , 1987; Nagel and Wilson , 2011; Gorur-Shandilya et al . , 2017 ) . Additional adaptation has been observed at synapses between first and second order olfactory neurons ( Nagel et al . , 2015; Cafaro , 2016 ) . Adaptation at multiple sites in the brain may contribute to the relatively slow adaptation time constants we measured for behavior ( 9 . 8 and 10 s for ON and OFF respectively . ) Our adaptive compression model has some similarity to the quasi-steady state model of ( Schulze et al . , 2015 ) , in which sensitivity to odor is dynamically adjusted to a running average of recent changes in odor history . Similar to that study in larvae , our study also suggests that events early in olfactory transduction can shape the time course of subsequent motor responses . Why might olfactory behavior in walking flies reflect integration of olfactory information over time while upwind flight in moths appears to require a rapidly fluctuating stimulus ? Several possibilities are worth considering . One is that the temporal demands of walking differ from those of flight . A flying moth travels at much faster speeds and over longer distances than a walking fly and will therefore traverse a plume in less time . Second , plumes developing near a boundary are broad and relatively continuous , while those in open air , particularly at the long distances covered by moths , are much more intermittent ( Crimaldi and Koseff , 2001; Celani et al . , 2014; Yee et al . , 1993 ) , again making detection of the plume edge potentially more important than responding rapidly to each plume encounter . Finally , receptor-odorant interactions can have different kinetics ( Nagel and Wilson , 2011 ) and may induce differing amounts of adaptation ( Cao et al . , 2016 ) . Differences in temporal processing of odors across species could also therefore reflect differences in the kinetics of individual odor-receptor interactions . Experiments expressing moth receptors in fly neurons , or comparing the history-dependence of flight vs walking reactions in the same species , may help resolve these differences . Rapid odor fluctuations have also been observed to impair upwind progress in some moth species ( Riffell et al . , 2014 ) . To relate elementary sensory-motor transformations to behavior in complex odor environments , we developed a simple model of olfactory navigation . In our model , different forms of sensation , such as flow sensing ( wind ) , temporal sensing ( offset response ) and spatial sensing ( comparisons across the antennae ) each produce distinct changes in forward and in angular velocity . The contributions of each form of sensing are summed to generate total turning behavior . Our model differs from previous models of turbulent navigation ( Pyk et al . , 2006; Balkovsky and Shraiman , 2002; van Breugel and Dickinson , 2014 ) in that it does not specify any distinct behavioral states such as ‘upwind orientation’ or ‘casting . ’ This is consistent with the observation that intermediate behavior , in which a positive upwind velocity overlaps with an increase in angular velocity , can be observed during decreasing odor ramps . Our model also differs from those requiring the animal to derive and maintain an estimate of the source position ( Vergassola et al . , 2007; Masson , 2013 ) . The only ‘memory’ required by our model is a slow adaptation and an offset response with a long integration time . Slow adaptation has been observed in the responses of olfactory receptor neurons and projection neurons ( Kaissling et al . , 1987; Nagel and Wilson , 2011; Nagel et al . , 2015; Cafaro , 2016; Gorur-Shandilya et al . , 2017 ) , while offset responses with long integration times have been observed in antennal lobe interneurons ( Nagel and Wilson , 2016 ) . Thus , both these types of history-dependence have been experimentally demonstrated . To validate our model , we showed that it can reproduce several features of experimentally observed fly behavior . First , the model can produce the upwind run during odor and the local search at offset that we observe in response to odor pulses in our miniature wind-tunnels . Second , it can produce straighter trajectories during odor and local search in the absence of wind information . Third , variation in the scale of the ON and OFF functions can generate the type of variability we observe in behavior across flies . Finally , the model produces a distribution of behaviors ( source finding , intermediate search , and downwind orientation ) similar to that of real flies when tested in a turbulent odor plume . Despite these similarities , there are aspects of fly behavior that our model does not capture . For example , we were unable to precisely match the distribution of angular velocities observed in our data and still produce realistic trajectories . This suggests that there is additional temporal structure in real fly behavior that our model lacks . There are also discrepancies between our model predictions and the timing of responses near odor onset ( particularly in the frequency sweep responses ) that might reflect the simplicity of the filter model used , or might reflect real variability in the latency of flies to respond to odor . Nevertheless , our model provides a relatively straightforward way to understand the relationship between temporal filtering of odors , sensory-motor coupling , and behavior in various odor environments . It should thus facilitate studies relating changes in neural processing to olfactory behavior . A question left open by our model is the role of spatial sensing ( bilateral comparisons ) in guiding navigation . We found that if the gain was set high enough , this form of sampling could significantly improve the model’s performance ( unrealistic gain values , of 1500 deg/s , could produce performance rates of over 95% success ) . This result is surprising , as previous studies have concluded that nearby samples taken in turbulent plume do not contain usable information ( Borst and Heisenberg , 1982 ) . However , recent studies have suggested that plumes may contain more usable spatial information than previously thought ( Boie et al . , 2018 ) , particularly when the plume forms near a solid boundary ( Gire et al . , 2016 ) . Using average gain values estimated from studies in tethered flies on a trackball ( Borst and Heisenberg , 1982; Gaudry et al . , 2013 ) we found that bilateral sampling contributed fairly little to performance , because the concentration differences across the antennae were typically quite small . In previous studies , bilateral sampling has been investigated largely using long-lasting odor stimuli of fixed concentration . It would be interesting in the future to ask whether flies can respond more strongly to small concentration differences when they are embedded in a fluctuating environment like the one measured here .
We used genetically blind norpA36 mutants , ( Ostroy and Pak , 1974; Pearn et al . , 1996 ) to avoid visual contributions to behavior . The norpA36 allele was backcrossed for seven generations to an isogenic w1118 stock ( Bloomington 5905 , also known as iso31 as described in ( Ryder et al . , 2004 ) that exhibits robust walking behavior ( Stavropoulos and Young , 2011 ) , using PCR to follow the allele through backcrossing . norpA36 males were crossed to w1118 virgins and virgin female norpA36/+ progeny were backcrossed to w1118 males . In each subsequent generation , 15 to 20 virgin females were backcrossed singly to w1118 males and genomic DNA was extracted from each female after several days of mating . PCR amplification was performed with primers flanking the norpA36 deletion ( oNS659 AAACCGGATTTCATGCGTCG and oNS660 TGTCCGAGGGCAATCCAAAC; 95C 2 min , 30x ( 95C 20 s , 60C 10 s , 72C 15 s , 72C 10 min ) to identify heterozygous norpA36/+ mothers giving rise to wild-type ( 172 bp ) and mutant ( 144 bp ) products . After seven generations of backcrossing , single males were crossed to an isogenic FM7 stock to generate homozygous stocks , and those bearing norpA36 were identified with PCR . Both w1118 norpA36 and w+ norpA36 stocks were generated during backcrossing . We used only w1118 norpA36 flies for behavior . For this reason , we used w1118 flies as ‘sighted’ controls , although the w1118 allele does affect vision as well . All flies were collected at least 1 day post-eclosion . After collection , flies were housed in custom-made cardboard boxes at room temperature ( 21 . 5-23 . 5C ) , with a light cycle of 12 hr , for at least 3 days prior to experiments to allow habituation . Different boxes were shifted by two hours relative to the others to allow us to perform several experiments with the same conditions in the same day . At the time of the experiments , flies were 5 to 14 days old ( average age was 7 . 1±1 . 8 days ) . Prior to the experiments , flies were starved for 5 hr in an empty transparent polystyrene vial with a small piece of paper soaked in distilled water to humidify the air . Experiments were performed between 2–4 hr after lights on ( ZT 2-ZT 4 ) . Our behavioral apparatus ( Alvarez-Salvado and Nagel , 2018 ) was modified from the design of Bell and Wilson ( 2016 ) and was designed to allow us to monitor the position and orientation of flies walking freely in two dimensions while tightly controlling the odor and wind stimuli they experienced . The behavioral arena was composed of several layers of laser-cut plastic , all 30 by 30 cm in size with varying thicknesses ( detailed below ) , in which different shapes were cut to create an internal air circuit and four individual behavioral chambers that measured 14 by 4 by 0 . 17 cm each . The arena was designed using Adobe Illustrator ( design: Adobe Systems , San Jose , CA; plastics: Pololu Corp , Las Vegas , NV and McMaster , Robbinsville , NJ; laser cutting: Pololu ) . The internal layers —in which the individual chambers were cut— were made of 0 . 5 mm-thick PETG ( McMaster reference: 9513K123 ) , 0 . 8 mm delrin ( McMaster: 8575K131 ) , and 0 . 4 mm fluorosilicone rubber ( McMaster: 2183T11 ) . Additionally , the arenas had a floor and ceiling layers made of 4 . 5 mm clear acrylic ( Pololu ) . The ceiling was held in place with seven set screws; combined with the fluorosilicone rubber layer this ensured that air did not escape from the chambers and produced more uniform odor concentrations throughout the arena . Each behavioral chamber had a separate air inlet through which charcoal-filtered air was supplied , and an outlet at the opposite end . A series of baffles in the PETG layer , as well as the short vertical extent of the chambers ( 1 . 7 mm ) ensured laminar flow of air through our chambers ( calculated Reynolds number 11 . 5 ) . Total airflow through the arena , as measured by anemometer , was 11 . 9cm/s . The arena was placed in an imaging chamber constructed from a breadboard ( Thorlabs ) and 80/20 posts ( McMaster: 47065T101 ) held in place with brackets ( McMaster: 47065T236 ) . Illumination was provided by an LED panel composed of an aluminum sheet ( McMaster: 88835K15 ) covered with infrared ( IR ) LED strips ( Environmental lights , irrf850-5050-60-reel ) . A diffuser ( Acrylite: WD008 ) was placed between the LED panel and the arena to provide uniform lighting . Flies were imaged from below the arena using a monochrome USB 3 . 0 camera ( Basler: acA1920-155um ) and a 12 mm 2/3’’ lens ( Computar: M1214-MP2 ) . An IR filter ( Eplastics: ACRY31430 . 125PM ) was placed between the camera and the arena . LEDs were controlled using buckblock drivers ( Digikey ) . An Arduino microprocessor ( teensy 2 . 0 , PJRC ) was used to strobe the IR LEDs at 50 Hz and to synchronize them with each camera frame . Imaging and stimulus delivery were controlled by custom software written in Labview ( National Instruments , Austin , TX ) . Timing of odor was controlled by a National Instruments board ( PCIe-6321 ) . Flies were tracked by comparing the camera image at each time point to a background image taken prior to the experiment . Background-subtracted images were thresholded and binarized; a region of interest per chamber was then taken for further processing . Particle filtering functions were applied to each region of interest to remove particles less than 3 pixels ( 0 . 4 mm ) long or greater than 50 pixels ( 6 . 8 mm ) long . A particle analysis function was used to identify the fly in each chamber and to compute its center of mass and orientation . Since the particle analysis function could only determine the fly’s orientation up to 180 ( i . e . it cannot distinguish the front and back of the fly ) , we used a second algorithm to uniquely determine the animal’s orientation . Each background-subtracted image was passed through a second thresholding operation with a lower threshold intended to include the translucent wings . The center of mass of this larger particle was compared to the center of mass of the smaller wingless particle to determine the orientation of the fly in 360 . Orientation measurements were strongly correlated with movement direction , but provided a smoother readout of heading direction when its velocity was low . Position ( X and Y coordinates ) and orientation were computed in real time during data collection and saved to disk . Wind and odor stimuli were delivered through inlets at the upwind end of the arena . Each arena was supplied with a main air line that provided charcoal-filtered wind . Wind flow rate was set to 1 L/min by a flowmeter ( Cole-Parmer , Vernon Hills , IL ) . This line could be shut off by a three-way solenoid valve ( Cole Parmer , SK-01540-11 ) in order to measure behavior in the absence of wind ( Figure 2 ) . To measure air flow , we used an anemometer ( miniCTA 54T30 , Dantec Dynamics , Skovlunde , Denmark ) , inserting the probe into the chambers through holes on a special ceiling made for this purpose . The anemometer was calibrated by measuring the outlet of a single 25 mm diameter tube ( filled with straws to laminarize flow ) connected directly to a flow meter . The measured air velocity was 11 . 9 cm/s . Odor was delivered via rapidly switching three-way solenoid valves ( LHDA1233115H , The Lee Company , Westbrook , CT ) located just below the arena , that directed odorized air either to the chambers or to a vacuum . Each chamber had its own valve , and odor was injected just downstream of the main air inlet , 1 . 7 cm upstream of the baffle region of the chamber . Charcoal-filtered air was odorized by passing it through a scintillation vial filled with 20 ml of odorant solution ( apple cider vinegar or ethanol ) , then directed through a manifold ( McMaster: 4839K721 ) to each of the four valves . Importantly , the vials containing the odor solution were almost full , creating a relatively small head space where odor could readily accumulate . Odorized air flow rate was set to 0 . 4 l/min using flowmeters . During non-odor periods , odorized air was directed into a vacuum manifold and away from the apparatus . Flow rates in the arena and vacuum manifold were matched to eliminate transients in odor concentration during switching . An equal volume of 'balancing' air was injected into the arena during these periods to maintain a constant air flow rate throughout the experiment . Balancing air was humidified by passing over an identical scintillation vial filled with water and was delivered by an identical three-way valve . Odor and balancing valves fed into a small t-connector , that was suspended from the arena using ≈1 cm of tygon tubing ( 0 . 8 mm inner diameter , E-3603 ) . This design , in which odor flowed continuously and was switched close to the arena , produced rapid odor dynamics with few concentration artifacts , but also a small mechanical stimulus when the valve was switched . This odor delivery system was using for experiments in Figures 1 and 2 , and for intensity experiments in Figure 3A–B . To produce analog odor stimuli including ramps , frequency sweeps , and the plume walk stimulus , we added two-way proportional valves ( EVP series , EV-P-05-0905; Clippard Instrument Laboratory , Inc . , Cincinnati , Ohio ) 20 cm upstream of the odor and balancing scintillation vials . Proportional valves were driven indepentendly by valve drivers ( EVPD-2; Clippard ) and were calibrated so that their maximal opening would produce the same flow rate as in experiments using three-way valves . ( three-way valves were held open during experiments with analog stimuli . ) Proportional valves produce increasing airflow with applied current; however , they exhibit both nonlinearity and histeresis , in which the effect of a driving current depends on the past and current state of the valve . To generate our desired stimulus waveforms , we first provided an ascending and descending ramp stimulus to the valves and measured the subsequent odor waveform in the behavioral chambers using a PID ( see below ) . We used the results of that measurement as a lookup table to create a driving current command that produced the desired odor waveforms . Lookup tables for odor and balancing valves were measured separately . We used PID measurements at several locations in the arena to verify that the delivered odor waveform matched our desired odor waveform . We used an anemometer ( see below ) to verify that the total flow rate during the stimulus ( in which odor and balancer valves were run together ) did not vary by more than 1% . To measure odor concentration in our arenas we used a photo-ionization detector ( miniPID , Aurora Systems , Aurora , Canada ) inserted into the arena , again using a special ceiling . All calibration measurements were made using 10% ethanol , which provided higher signal to noise than ACV . Measurements at the top of the arenas revealed an average rise time of ≈180 ms and a fall time of ≈220 ms for square pulses delivered using three-way valves . The latency of the measured odor onset from nominal odor onset increased linearly with distance from the odor source ( up to 900-1000 ms at the downwind end of the arena ) , consistent with our measurement of air velocity ( Figure 1—figure supplement 1 ) . For frequency sweep stimuli , we observed some widening of peaks with distance down the arena , consistent with the effects of diffusion ( Figure 1—figure supplement 1 ) . Diffusion thus set the upper limit on the frequency of stimuli that we could reliably deliver within our arena ( about 1Hz ) . Presenting higher frequency stimuli would require higher wind speeds , but we found that higher wind speeds caused flies to stop moving , as previously observed ( Yorozu et al . , 2009 ) . Each experiment lasted approximately 2 hr , during which flies performed an average of 86 . 7±7 . 7 trials . ( Some trials were discarded due to tracking problems , as described below , and not all experiments lasted exactly the same amount of time ) . Each trial lasted 70 s , and was followed by a gap of ≈6 seconds while the computer switched to the next trial . There were three to four types of trials that were randomly interleaved during the experiment . One of those types was always a blank trial , in which flies only experienced clean air flow . The other types corresponded to different types of odor stimuli , that were dependent on the experiment: namely , square odor pulses for experiments in Figures 1 , 2 and 3A–B; odor ramps in Figure 3C–D; frequency sweeps and plume data in Figure 3E–J . To ensure repeatability , data for all experiments was collected over several different days ( 5 to 9 , often non-consecutive ) . For Figure 1 , we used data from experiments performed over a period of 7 months . For experiments in Figure 2 , we rendered flies ‘wind-blind’ by anesthetizing them on a cold plate and cutting their aristae and stabilizing their antennae . We cut the aristae by clipping them with fine forceps at the lowest possible level without touching the antennae . Then , we put a very small drop of ultra-violet ( UV ) glue on the anterior side of the antennae , falling between the second and third segments , as well as touching the rest of the clipped aristae . We then used a pen-sized ultra-violet light to cure the glue , and made sure it was solid before putting the flies back to their home vials to recover for 24 hr . The whole procedure took approximately 5 min , and never longer than 10 . We did this procedure in a pair of flies at a time , stabilizing the antenna of one and using the other as sham-treated ( it was placed on the cold plate and under the UV light exactly like the treated fly was ) . For experiments in Figure 6 , approximately 48 hr before the experiment , we applied a drop of UV glue connecting both wings of the fly or to each wing hinge . This prevented flies from flying while allowing us to still use the wings to detect heading . All analyses were performed in Matlab ( Mathworks , Natick , MA ) ( Alvarez-Salvado and Nagel , 2018; copy archived at https://github . com/elifesciences-publications/AlvarezSalvado_ElementaryTransformations ) . X and Y coordinates and orientation information were extracted from the data files , and any trials with tracking errors ( i . e . flies’ position was missed at some point ) were discarded ( this occurred rarely ) . In some trials , we observed orientation errors in the form of sudden changes of approximately 180 . In these cases , orientation was corrected by calculating the heading of the flies using X and Y coordinates , and filling in the gaps in orientation using the orientation that best correlated with that information , producing coherent and continuous orientation vectors . Coordinates and orientations were low-pass filtered at 2 . 5 Hz using a two-pole Butterworth filter to remove tracking noise that was produced especially when flies were not moving . X and Y coordinates were then converted to mm , and trials in which flies moved less than a total of 25 mm were discarded . Distance moved was calculated as the length of the hypotenuse between two subsequent pairs of coordinates . We next calculated a series of gait parameters from each trial’s data . Ground speed was obtained by dividing the distance moved by the time interval of each frame ( 20 ms ) . Upwind velocity was calculated using the derivative of the filtered Y coordinates divided by the time interval of 20 ms . Angular velocity was calculated as the absolute value of the derivative of the filtered unwrapped orientation ( i . e . orientation with phases corrected to be continuous beyond 0° or 360° ) divided by the time interval of 20 ms . For all gait parameters shown ( ground speed , upwind velocity , angular velocity ) , we excluded data points in which ground speed was less than 1 mm . This was necessary because flies spend a large amount of time standing still . Distributions of gait parameters are therefore highly non-Gaussian , with large peaks at 0 ( Figure 3—figure supplement 1A ) , and parameter means are highly influenced by the number of zeros . In addition , the probability of moving ( obtained by binarizing the ground speed with a threshold of 1 mm/s ) changes dramatically in response to odor , and remains high for tens of second after odor offset ( Figure 3—figure supplement 1B ) . Exclusion of the large number of zeros from average gait parameters produced more reliable estimates of these parameters . Curvature was calculated by dividing angular velocity by ground speed ( excluding any points where ground speed was less than 1 mm/s ) . Turn probability was calculated binarizing curvature with a threshold of 20 deg/mm . Because it takes a little over a second for the odor waveform to advect down the arena , the exact time of odor encounter and loss depends on the position of the fly within the arena . This advection delay has a strong effect on our estimates of gait parameter dynamics , particularly for fluctuating sinusoidal stimuli . We therefore developed a warping procedure to align behavioral responses to the actual time at which each fly encountered the odor on each trial . To implement this procedure , we first recorded the PID response to each stimulus at three different points along the arena ( Figure 1—figure supplement 1 ) . We then calculated the delay for the odor to reach the position of the fly for each time frame during the odor stimulus , and shifted all the data points back by this amount . The periods before and after the odor stimulation are also shifted according to the initial position of flies in the odor period . This method can skip a data point when the fly moves upwind or can repeat a data point when the fly moves downwind , but the majority of the data are conserved and the resulting waveforms resemble very much the initial ones . After warping , all trials from all flies can be equally compared to a standard PID measurement done at the top of the arenas ( i . e . the odor source ) . Warping was applied to all data shown in Figures 1–3 . Note that in experiments using three-way valves ( Figure 1 ) , the click of the valve produced a brief freezing responses that was visible as a dip in ground speed . However , because of the warping , the time of the valve click is distributed across flies , as their ground speeds have been aligned to the time of odor encounter rather than the time of valve opening . This results in a smeared dip in the ground speed trace near the beginning and end of the odor stimulus . For experiments using frequency sweeps and plume walks , we additionally excluded data obtained after the fly reached the top end of the chamber , as well as data from within 3 mm of the side walls . These exclusions were made to minimize the effect of arena geometry on gait parameter estimates , and to exclude regions where boundary layer effects would cause the odor waveform to advect more slowly . To calculate the data shown in the insets of Figure 3E–H , and in Figure 3—figure supplement 1F , we used a jackknife procedure to resample the responses of flies to frequency sweep stimuli . We made 10 estimates of the mean , excluding 34 trials from each estimate . To estimate the modulation of upwind velocity and ground speed in response to each cycle of the stimuli , we took the times between minima of the stimulus waveform as the limits for each cycle of the ascending frequency sweep; for the descending frequency sweep we used the intervals between maxima of the odor waveform . Within those limits , we calculated the minimum-to-maximum amplitude for each of the 10 different mean responses . The results shown in the figures are the mean of these estimates as a function of frequency of the corresponding stimulus cycles . The frequency of the cycles was estimated as 1 over the duration of the cycle . Error bars in the figure insets represent the standard error ( SE ) across estimates , calculated as ( 5 ) SE=n-1n∑i=1n ( x¯i-x¯ ) 2nwhere x¯i is each of the peak-to-peak estimates excluding one fly , x¯ the estimate including all flies , and n the number of data subsets used ( 10 ) . In Figure 1G , Figure 2B and Figure 5G , we compared the mean values of different motor parameters from the same fly in three different periods of time in the trials , namely: ‘before odor’ from −30 to 0 s before the odor , 'during odor' from 2 to 3 seconds during the odor , and ‘after odor’ from 1 to 3 s after odor offset . We performed a Wilcoxon signed rank paired test for each of those comparisons and corrected the threshold for statistical significance alpha using the Bonferroni method . All significant comparisons were marked with asterisks in the figures , and the exact p values obtained are presented in the following tables . ComparisonUpwind velocityGround speedAngular velocityCurvatureTurn probabilityBefore–during odor2 . 0⋅10-123 . 9⋅10-91 . 7⋅10-34 . 9⋅10-52 . 3⋅10-3Before–after odor6 . 3⋅10-27 . 7⋅10-61 . 2⋅10-115 . 5⋅10-107 . 3⋅10-14During–after odor1 . 4⋅10-121 . 5⋅10-109 . 5⋅10-117 . 1⋅10-104 . 8⋅10-12 p values for comparisons made in Figure 1G . The alpha value after correcting for multiple comparisons was 0 . 0167 . ComparisonUpwind velocityGround speedCurvatureBefore–during odor0 . 270 . 0160 . 34Before–after odor0 . 840 . 850 . 003During–after odor0 . 410 . 0080 . 002 p values for comparisons made in Figure 2B . The alpha value after correcting for multiple comparisons was 0 . 0167 . ComparisonUpwind velocityTurn probabilityBefore–during odor1 . 3⋅10-831 . 2⋅10-55Before–after odor9 . 0⋅10-461 . 3⋅10-83During–after odor1 . 3⋅10-831 . 3⋅10-83 p values for comparisons made in Figure 5G . The alpha value after correcting for multiple comparisons was 0 . 0001 . ComparisonUpwind velocityGround speedAngular velocityCurvatureTurn probabilityBefore–during odor9 . 4⋅10-114 . 6⋅10-75 . 6⋅10-11 . 0⋅10-13 . 2⋅10-6Before–after odor3 . 7⋅10-91 . 7⋅10-55 . 2⋅10-52 . 1⋅10-68 . 1⋅10-10During–after odor3 . 2⋅10-95 . 2⋅10-93 . 1⋅10-34 . 9⋅10-69 . 3⋅10-5 p values for comparisons made in Figure 1—figure supplement 3B . The alpha value after correcting for multiple comparisons was 0 . 0167 . To estimate the Standard Error of the proportion of successful trials shown in Figure 6I and in Figure 7C , we used the formula ( 6 ) SE=p ( 1-p ) nwhere p was the proportion of successful trials and n the number of trials . To test for statistical differences in Figure 7C , we calculated a z statistic by normal approximation of the corresponding binomial distributions according to: ( 7 ) z=p1−p2p ( 1−p ) ( 1n1+1n2 ) where p1 and p2 are the probabilities of success in the two distributions being compared , p is the probability of both distributions combined , and n1 and n2 are the number of trials in the two distributions . We then estimated the p values by evaluating a normal cumulative distribution function of a standard normal distribution for the resulting z values . This analysis yielded the following results: Comparisonz statisticp valueκ7=0 VS κ7=40deg/s1 . 030 . 30062κ7=0 VS κ7=300deg/s3 . 500 . 00046κ7=0 VS κ7=300deg/sSWAP4 . 120 . 00004 z statistics and p values for comparisons made in Figure 7C . The alpha level used was 0 . 05 . Our computational model was composed of two parts ( Alvarez-Salvado and Nagel , 2018 ) . In the first , we asked whether simple phenomenological models , comprised of a linear filtering step , and a nonlinear adaptive compression function , were capable of capturing the dynamics of upwind velocity and turn probability in response to a wide array of odor waveforms . We compared fits of four model versions to our behavioral data , and tested the resulting best fit model by predicting responses to the plume walk stimulus . These fits comprise the two temporal functions which we call ON and OFF . In the second part , we asked whether a simple navigational model , based on the ON and OFF functions fit to the data and described in Figure 5 , was capable of reproducing the types of trajectories we observed experimentally and of locating the source of a real odor plume . In addition , this model allowed us to test the contribution of each of its components to successful odor localization . In the model , we first compute two temporal functions of the odor stimulus , ON and OFF . These two signals are then used to modulate ongoing behavioral components ( angular velocity and ground speed ) which iteratively update the fly’s position . The model can be run in open loop , as in our behavioral expeirments , by providing an odor input as a function of time , or in closed loop , where the odor concentration at any point in time depends on the fly’s position in a real or virtual space . All computational modelings were performed in Matlab . Differential equations were simulated using the Euler method with a time step of 20 ms . We generated a turbulent odor plume in a low-speed bench-top wind tunnel with a flow-through design . Two wind tunnels were built , one in Colorado ( for plume measurements ) and one in New York ( for behavioral measurements ) . In the Colorado wind tunnel , air entered the tunnel through a bell-shaped contraction ( 4:1 ratio ) and passed through a turbulence grid ( 6 . 4 mm diameter rods with a 25 . 5 mm mesh spacing ) prior to the test section . The test section was 30 cm wide , 30 cm tall , and extended 100 cm in the direction of the flow . Air exited the test section through a 15 cm long honeycomb section used to isolate the test section from a fan located in the downstream contraction . The fan generated a mean flow of air through the tunnel at 10 cm/s . Acetone was released isokinetically into the center of the test section through a 0 . 9 cm diameter tube aligned with the flow . The tube opening was located 10 cm downstream of the turbulence grid and 6 mm above a false floor spanning the length and width of the test section . The New York tunnel was designed similarly , except that test section measured 38 cm by 38 cm by 92 cm , the honeycomb was 5 cm long , and odor was released from a 1cm diameter tube at floor level . Air flow was 10cm/s and odor release was isokinetic as in the Colorado wind tunnel . The New York tunnel was fitted with an aluminum IR light panel ( Environmental lights , irrf850-5050-60-reel ) 2 . 5 cm below a diffuser ( Acrylite: WD008 ) and a 1 cm thick acrylic layer that acted as the arena floor . A channel 1 cm wide and 0 . 4 cm deep was milled into this arena and filled with water to constrain flies to walk within the imaging area , 31 cm wide and 87 cm long . Two cameras ( Point Grey: 2 . 3MP Mono Grasshopper3 USB 3 . 0 ) with 12 mm 2/3' lenses ( Computar: M1214-MP2 ) were suspended approximately 45 cm above the arena to image fly movement . Tracking code was written in Labview and used the same algorithms as described above to extract position and heading at 50Hz . To measure plume structure and dynamics in air , we used a planar laser-induced fluorescence ( PLIF ) system ( Lozano et al . , 1992 ) to image a plume of acetone vapor ( Connor et al . , 2018 ) . A UV laser light sheet entered the test section of the tunnel through a slit along the length of the test section to excite acetone vapor . A camera imaged the resulting acetone fluorescence in the test section through a glass window . The imaging area covered up to 30 cm downwind from the odor source and up to 8 cm to both sides . The plume was imaged in the 1-mm-thick laser sheet centered on the tube 6 mm above the bed . A total of 4 min were recorded . Images were then post-processed into calibrated matrices of normalized concentrations . We produced acetone vapor by bubbling an air and helium gas mixture through flasks partially filled with liquid acetone . To reduce fluctuations in concentration , a water bath maintained flask temperature at 19 deg C which was approximately 2 degrees cooler than ambient air temperature to prevent condensation . To account for the density of acetone , we blended air ( 59% v/v ) and helium ( 41% v/v ) for the carrier gas . Assuming 95% saturation after contact with the liquid acetone , the gas mixture was approximately 25% acetone by volume and neutrally buoyant . An Nd:YAG pulsed laser emitted light at a wavelength of 266 nm and a frequency of 15 Hz to illuminate the acetone plume . After excitation at that wavelength , acetone vapor fluoresces with an intensity proportional to its concentration . A high-quantum efficiency sCMOS camera imaged the acetone plume fluorescence at 15 Hz . To enhance signal and minimize noise , we collected data in a dark environment , used a lens with high light-gathering capabilities ( f/0 . 95 ) , and binned the pixels from 2048x2048 native resolution to 512x512 resolution ( 0 . 74 mm/pixel ) . Images were post-processed using an algorithm adapted from ( Crimaldi , 2008 ) to correct for variations in laser sheet intensity , lens vignette , and pixel-to-pixel gain variation . The correction used a spatial map of the image system response by imaging the test section while it was filled with a constant and uniform distribution of acetone . Signal intensities were normalized by the intensity at the tube exit such that concentrations have average values between 0 and 1 . The 'plume walk' stimulus was generated by taking the time course of odor concentration along a linear trajectory going upwind through a plume movie at 6 mm/s ( the average ground speed of our flies ) , starting 8 . 9 cm laterally from the midline and 30 cm downwind from the source . | All kinds of animals use their sense of smell to find things . Doing this is difficult because odors in air travel as plumes , which meander downwind and break apart . Scientists are interested in learning the rules that animals use to decipher these odor signals and trace them back to their source . For example , do animals use patterns of timing in the odor , differences between smell at the two nostrils , or the direction of the wind ? Scientists would also like to know how animal’s brain circuits decipher this information . Tiny fruit flies make a good model for studying the way animals detect odors because scientists have already learned a great deal about how their brains work . There are also many tools available to help scientists study the brain circuits of fruit flies . Now , Álvarez-Salvado et al . show that fruit flies use multiple senses to track odors to their source . In the experiments , fruit flies that were blind and could not fly were placed in tiny wind tunnels and their behavior in response to a smell or no smell in the tunnel was carefully documented . When the flies detected an odor , they turned to face the wind using their antennae to detect wind direction and run toward it . When flies lost track of an odor they began to search for it at the spot where they last smelled it . Next , Álvarez-Salvado et al . created a computer model that recreated the flies’ behavior and was able to find the odor source as well as real flies . The model added together these basic behaviors to successfully recreate the flies’ odor-search strategy . Other animals are often better than humans at finding odor sources . As a result , people use pigs to find truffles and dogs to find lost hikers . The computer model Álvarez-Salvado et al . developed might help design robots that can search for truffles , hikers , or landmines , without risking the lives of animals . It might also be useful for designing autonomous vehicles that must respond to many types of information in changing environments to make decisions . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2018 | Elementary sensory-motor transformations underlying olfactory navigation in walking fruit-flies |
Bile is an important component of the human gastrointestinal tract with an essential role in food absorption and antimicrobial activities . Enteric bacterial pathogens have developed strategies to sense bile as an environmental cue to regulate virulence genes during infection . We discovered that Vibrio parahaemolyticus VtrC , along with VtrA and VtrB , are required for activating the virulence type III secretion system 2 in response to bile salts . The VtrA/VtrC complex activates VtrB in the presence of bile salts . The crystal structure of the periplasmic domains of the VtrA/VtrC heterodimer reveals a β-barrel with a hydrophobic inner chamber . A co-crystal structure of VtrA/VtrC with bile salt , along with biophysical and mutational analysis , demonstrates that the hydrophobic chamber binds bile salts and activates the virulence network . As part of a family of conserved signaling receptors , VtrA/VtrC provides structural and functional insights into the evolutionarily conserved mechanism used by bacteria to sense their environment .
Bile is an important component of the human gastrointestinal tract that plays a major role in the emulsification and solubilization of lipids ( Begley et al . , 2005 ) . Its main components are bile salts , cholesterol , and phospholipids that together possess antimicrobial activities , challenging the survival and colonization of both pathogenic and commensal bacteria ( Begley et al . , 2005 ) . There is increasing evidence that enteric bacterial pathogens can utilize bile as a signaling cue to regulate gene expression profiles during host infection . For example , Salmonella typhimurium senses bile to repress the genes involved in invasion when growing in the intestinal lumen and induce their expression to promote invasion upon the penetration of the mucosal layer wherein the concentration of bile is decreased ( Prouty and Gunn , 2000; Prouty et al . , 2004 ) . In Shigella spp . , bile salts increase the pathogen's adherence and invasion of epithelial cells ( Pope et al . , 1995 ) . Bile also has important effects on the pathogenicity of Vibrio species , as observed in multiple studies with pathogenic strains of V . cholerae and V . parahaemolyticus ( Gupta and Chowdhury , 1997; Schuhmacher and Klose , 1999; Krukonis and DiRita , 2003; Faruque et al . , 1998 ) . V . cholerae produces two major virulence factors during infection , cholera toxin ( CT ) and toxin-coregulated pilus ( TCP ) , and these factors are repressed by ToxT in the presence of bile ( Gupta and Chowdhury , 1997; Schuhmacher and Klose , 1999; Krukonis and DiRita , 2003; Faruque et al . , 1998 ) . However , it was also reported that bile can activate the production of CT independent of ToxT ( Hung and Mekalanos , 2005 ) . Non-O1/non-O139 V . cholerae strains that do not encode CT and TCP can cause gastroenteritis and utilize bile to activate a pathogenic type III secretion system ( Dziejman et al . , 2005; Chaand et al . , 2015; Alam et al . , 2010 ) . Similarly , V . parahaemolyticus type III secretion system 2 ( T3SS2 ) is induced specifically by bile salts during infection , resulting in acute gastroenteritis ( Gotoh et al . , 2010 ) . Despite these important discoveries , the mechanism of sensing bile salts by pathogenic bacteria remains unknown . To investigate this mystery , we used V . parahaemolyticus as a model to elucidate how Vibrio spp . sense bile salts as a signal to regulate the expression of virulence genes . V . parahaemolyticus is a globally-spread , Gram-negative , halophilic bacterial pathogen that inhabits marine and estuarine environments and is the world’s leading cause of acute gastroenteritis due to the consumption of raw or undercooked seafood ( Broberg et al . , 2011; Zhang and Orth , 2013 ) . During recent years , rising temperatures in the ocean has contributed to this pathogen’s worldwide dissemination ( Nair et al . , 2007; Velazquez-Roman et al . , 2013; McLaughlin et al . , 2005; O'Boyle and Boyd , 2014; Daniels et al . , 2000 ) . V . parahaemolyticus is also the causative agent of the devastating shrimp disease Acute Hepatopancreatic Necrosis Disease ( AHPND ) ( Tran et al . , 2013; Lee et al . , 2015 ) . Over the last decade , enormous progress has been made to elucidate virulence factors used by this pathogen and tools are available to study virulence mechanisms at the genetic and biochemical levels ( de Souza Santos et al . , 2015 ) . As part of its virulence repertoire , V . parahaemolyticus encodes two T3SSs: the cytotoxic T3SS1 and the enterotoxic T3SS2 . T3SSs encode needle-like secretion apparatuses used to deliver bacterial effector proteins , called Vops , which allow V . parahaemolyticus to invade and kill infected host cells ( Broberg et al . , 2011; Makino et al . , 2003; Galán and Wolf-Watz , 2006 ) . Each T3SS delivers a specific set of Vops into a host cell ( Broberg et al . , 2011 ) . For example , the T3SS1 exclusively secretes VopQ , VopR , VPA0450 and VopS and uses the effectors to orchestrate a multifaceted host cell death . This system is found in all strains of V . parahaemolyticus , and therefore is proposed to be important for bacterial survival in the environment ( Broberg et al . , 2011 ) . T3SS2 is present in clinical isolates of V . parahaemolyticus and is a key bile salt-induced virulence system that causes enterotoxicity and symptoms associated with gastroenteritis during infection ( Broberg et al . , 2011; Park et al . , 2004; Hiyoshi et al . , 2010; Ritchie et al . , 2012 ) . Bile salts not only induce Vops , such as VopA and VopC , but also the needle-like secretion apparatus , including components such as T3SS2 translocon VopD2 ( Broberg et al . , 2011; Zhang and Orth , 2013; Trosky et al . , 2004 ) . In V . parahaemolyticus , the activation of T3SS2 by bile salts is regulated by two transmembrane ToxR-like transcription factors , VtrA ( VPA1332 ) and VtrB ( VPA1348 ) ( Gotoh et al . , 2010; Kodama et al . , 2010 ) . Homologues of VtrA and VtrB , named VttRA and VttRB respectively , have been identified in T3SS-containing non-O1/non-O139 V . cholerae strains and function in a similar way ( Alam et al . , 2010 ) . Despite the identification of these transcription factors , the molecular mechanism underlying bile salts sensing and T3SS2 signal propagation by V . parahaemolyticus and V . cholerae is unknown . Here , we identify VtrC ( VPA1333 ) from V . parahaemolyticus as a previously unrecognized component necessary for bile salt sensing and T3SS2 activation . vtrC is constitutively co-transcribed with vtrA as an overlapping 3’ open reading frame . Furthermore , we demonstrate that the periplasmic domains of the transmembrane proteins , VtrA and VtrC , form a functional complex that binds bile salts to activate VtrA’s cytoplasmic DNA binding domain , which in turn induces T3SS2 via the downstream transcription factor VtrB . The structure of VtrA/VtrC periplasmic domains reveals an obligate heterodimer where VtrC recruits structural elements from VtrA to complete a β-barrel with a hydrophobic inner chamber that binds bile salts . A co-crystal structure of the VtrA/VtrC heterodimer with a bile salt reveals that ligand binding occurs in the hydrophobic inner chamber of VtrC , similar to that found for the family of monomeric calycins . Mutations of residues within the hydrophobic chamber that are important for bile salt-binding disrupt bile salt activation of T3SS2 . Collectively , we uncover a mechanism by which bacteria can sense bile salts , and reveal how an evolutionarily conserved receptor senses an environmental cue to induce the production of virulence factors .
vtrC ( vpa1333 ) is a previously uncharacterized gene in V . parahaemolyticus that is located directly downstream of vtrA ( Figure 1A ) . The open reading frames for vtrA and vtrC overlap by 17 nucleotides , suggesting that they are likely in the same operon and function in the same biological process . We observed that the gene organization of vtrA , vtrC and vtrB in the genome is highly conserved in other Vibrio and related species ( Figure 1A ) . Both V . cholerae non-O1/O139 strains and Grimontia hollisae possess a T3SS similar to T3SS2 in V . parahaemolyticus and cause gastroenteritis during human infection ( Dziejman et al . , 2005 ) . RT-PCR of the region spanning vtrA and vtrC showed that these genes are indeed in the same operon and co-transcribed before and after bile salt-mediated T3SS2 induction ( Figure 1B , lanes 3 , 5 ) . PSI-BLAST analysis with the predicted VtrC protein sequence revealed that it is conserved in various Vibrio and related species , as well as in Moritella . VtrC has a highly conserved N-terminal transmembrane signal anchor that is predicted to direct and retain proteins in the bacterial membrane ( Figure 1C ) . PSI-BLAST analysis using VtrA's sequence without the cytoplasmic DNA binding domain identified homologues in the same group of bacteria where VtrC is found ( Figure 1—figure supplement 1 ) . Furthermore , ten other strains of bacteria contain homologous genes of vtrA and vtrC but not vtrB ( Figure 1—figure supplement 2 ) . These results support the hypothesis that VtrA and VtrC evolved independently of VtrB . 10 . 7554/eLife . 15718 . 003Figure 1 . VtrC is conserved in various Vibrio and other species with VtrA-like sequences . ( A ) The gene organization of vtrA , vtrC , and vtrB is conserved in the T3SS2-like pathogenicity island of Vibrionaceae family species , with variable numbers of inserted genes ( indicated by numbers in parentheses ) between vtrC ( or its homologous gene ) and vtrB ( or its homologous gene ) . ( B ) RT-PCR to amplify the target region shown by the bar that spans vtrA and vtrC . - Bile , Vibrio parahaemolyticus POR1 strain growing in LB without bile salts; + Bile , POR1 growing in LB supplemented with 0 . 05% bile salts . - RT , without reverse transcriptase; + RT , with reverse transcriptase; lane 1: DNA marker . RT-PCR is representative of three independent experiments . ( C ) Multiple protein sequence alignment of VtrC and its homologues . GI number of each protein is listed before the species names . Residues are highlighted according to group-wise conservations: mainly hydrophobic ( yellow ) and small ( gray ) . Signal anchor: the predicted N-terminal transmembrane domain . An empty line is inserted between the species that contain ( top ) or lack ( bottom ) vtrB ( or vtrB homologous gene ) . *marks protein sequences that were translated from nucleotide to include the entire N-terminus . DOI: http://dx . doi . org/10 . 7554/eLife . 15718 . 00310 . 7554/eLife . 15718 . 004Figure 1—figure supplement 1 . Multiple protein sequence alignment of VtrA and its homologues . GI number of each protein is listed before the species names . Residues are highlighted according to group-wise conservations: mainly hydrophobic ( yellow ) , small ( gray ) , conserved positive charge ( blue ) , conserved negative charge ( red ) , conserved S/T ( orange ) , and invariant polar position ( black ) . An empty line is inserted between the two groups of species that either contain ( top ) or lack ( bottom ) vtrB ( or vtrB homologous gene ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15718 . 00410 . 7554/eLife . 15718 . 005Figure 1—figure supplement 2 . The gene organization of vtrA and vtrC in species that lack vtrB . The gene organization of vtrA and vtrC homologous genes is conserved in Vibrionaceae and Moritellaceae family species . DOI: http://dx . doi . org/10 . 7554/eLife . 15718 . 005 VtrA and VtrB are involved in the activation of the T3SS by bile salts . Therefore , we set out to determine whether VtrC also plays a role in this pathway . Because vtrA and vtrC overlap by 17 nucleotides , we generated a deletion of vtrC from the V . parahaemolyticus POR1 strain by retaining vtrC’s first 33 nucleotides thereby leaving the vtrA open reading frame intact ( see Materials and methods ) . Deletion of vtrC completely abolished the activation of T3SS2 by bile salts , as shown by the loss of expression and secretion of the T3SS2 effectors VopA and VopC , and of the T3SS2 translocon VopD2 ( Figure 2A , B ) . Complementation of vtrC deletion by a vector expressing VtrC fully restored the activity of T3SS2 ( Figure 2A , B ) , indicating that the phenotypes observed in the vtrC deletion were not caused by a polar effect on neighboring genes . The effect of VtrC is specific to T3SS2 because its deletion had no impact on the expression and secretion of the T3SS1 effector VopS ( Figure 2C ) . Taken together , these results indicate that VtrC is required for T3SS2 activation by bile salts . 10 . 7554/eLife . 15718 . 006Figure 2 . VtrC is essential for the activity of V . parahaemolyticus T3SS2 in the presence of bile salts . Expression ( Cell ) and secretion ( Sup ) of V . parahaemolyticus T3SS components were analyzed by Western blot . Loading control ( LC ) is shown for total protein lysate . ( A–B ) Expression and secretion of T3SS2 effectors VopA and VopC , and translocon VopD2 by V . parahaemolyticus POR1 derivative strains with the empty pBAD vector ( Empty ) , vtrC deletion ( ΔvtrC Empty ) or vtrC complementation ( ΔvtrC p-vtrC ) containing a pBAD vector expressing VtrC under the putative promoter of its operon ( 1kb upstream of vtrA ) . Protein-specific antibodies were used to detect VopA and VopC . Anti-FLAG antibody was used to detect C-terminal endogenously FLAG-tagged VopD2 . -/+ Bile , V . parahaemolyticus grown in LB without bile salts ( - ) or supplemented with 0 . 05% bile salts ( + ) . Non-specific band is indicated with an asterisk . ( C ) Expression and secretion of T3SS1 effector VopS by V . parahaemolyticus POR1 derivative strains with the empty pBAD vector ( Empty ) , vtrC deletion ( ΔvtrC Empty ) or vtrC complementation ( ΔvtrC p-vtrC ) . Anti-FLAG antibody was used to detect endogenously C-terminal FLAG-tagged VopS . -/+ DMEM , V . parahaemolyticus grown in LB ( - ) or DMEM ( + ) . Data is representative of three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 15718 . 006 Analysis of the VtrC protein sequence using membrane topology prediction programs , TMpred , TMHMM 2 . 0 , HMMtop 2 . 0 , and Phobius indicates VtrC is an inner membrane protein with the N-terminal 1–9 amino acids facing the cytoplasm , a single transmembrane helix ( 10–24 amino acids ) and the C-terminal domain ( 25–161 amino acids ) in the periplasm ( Figure 3A ) . To verify the cellular localization and orientation of VtrC , we used two reporter proteins from Escherichia coli , alkaline phosphatase PhoA and β-galactosidase LacZ , whose activities depend on their subcellular localization . PhoA is active only in the periplasm after disulfide bond formation and dimerization , whereas LacZ only exhibits enzymatic activity in the cytoplasm where it can fold properly ( van Geest and Lolkema , 2000 ) . PhoA or LacZ were fused to either the N-terminus or the C-terminus of VtrC and their activities were then measured to determine the localization of VtrC ( Liu , 2003 ) . As expected , PhoA was active when fused to the C-terminus of VtrC , but not N-terminus ( Figure 3B ) . In addition , LacZ was more active when fused to the N-terminus of VtrC ( Figure 3C ) . These observations demonstrate that VtrC is localized to the inner membrane with the N-terminal 9 amino acids in the cytoplasm and the C-terminal domain in the periplasm . 10 . 7554/eLife . 15718 . 007Figure 3 . VtrC is an inner membrane protein with the N-terminus in the cytoplasm and the C-terminus in the periplasm . ( A ) Predicted cellular localization and orientation of VtrC . Active form of PhoA or LacZ fused to VtrC based on prediction . ( B ) Alkaline phosphatase PhoA activity of POR1 with the empty pBAD vector ( Empty ) , expressing N terminal PhoA-VtrC ( p-phoA-vtrC ) or C terminal VtrC-PhoA ( p-vtrC-phoA ) fusion protein . ( C ) β-galactosidase LacZ activity of POR1 with the empty pBAD vector ( Empty ) , expressing N terminal LacZ-VtrC ( p-lacZ-vtrC ) or C terminal VtrC-LacZ ( p-vtrC-lacZ ) fusion protein . ****p<0 . 0001 , n = 3 , +/- S . E . M . Data is representative of three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 15718 . 007 Previous studies have shown that VtrA directly controls the expression of VtrB and that both proteins are inner membrane proteins with their N-terminal DNA binding domain oriented in the cytoplasm and their C-terminal region in the periplasm ( Figure 4A ) ( Kodama et al . , 2010 ) . Given that VtrC is constitutively expressed and its C-terminal domain is oriented in the periplasm , we hypothesized that VtrC , like VtrA , functions upstream of VtrB . The deletion of vtrC did not change the mRNA level of VtrA ( Figure 4B ) . Interestingly , deletion of vtrC resulted in decreased protein levels of VtrA independent of bile salts ( Figure 4C , lanes 3 and 7 ) , suggesting that the presence of VtrC might stabilize VtrA . Ectopic expression of VtrC from a plasmid rescued VtrA levels ( Figure 4C , lanes 4 and 8 ) . As was previously observed for VtrA , the absence of VtrC prevented the induction of vtrB upon T3SS2 activation with bile salts ( Figure 4D ) . Thus , VtrC appears to function upstream of VtrB via a signaling cascade including VtrA . The observed degradation of VtrA in the absence of VtrC suggested that there might be a direct physical interaction between VtrA and VtrC . 10 . 7554/eLife . 15718 . 008Figure 4 . VtrC is necessary for maintaining VtrA protein level and the induction of VtrB . The effects of VtrC on VtrA and VtrB were characterized using V . parahaemolyticus POR1 derivative strains with the empty pBAD vector ( Empty ) , vtrA deletion ( ΔvtrA Empty ) , vtrC deletion ( ΔvtrC Empty ) or vtrC complementation ( ΔvtrC p-vtrC ) containing a pBAD vector expressing VtrC under the putative promoter of its operon ( 1kb upstream of vtrA ) . ( A ) Cellular localization and orientation of VtrA , VtrB and VtrC . ( B ) qRT-PCR analysis of VtrA mRNA level relative to POR1 with the empty pBAD vector ( Empty ) grown in LB without bile salts . ( C ) Western blot analysis of VtrA protein level . Protein specific antibody was used to detect VtrA . Non-specific band is indicated with an asterisk . ( D ) qRT-PCR analysis of VtrB mRNA level relative to POR1 with the empty pBAD vector ( Empty ) grown in LB without bile salts . -/+ Bile , V . parahaemolyticus grown in LB without bile salts ( - ) or supplemented with 0 . 05% bile salts ( + ) . For qRT-PCR analysis , expression of vtrA and vtrB was normalized to the expression of the control gene fliA . Data is representative of three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 15718 . 008 To test whether VtrA and VtrC form a complex in vivo , we performed co-immunoprecipitation using endogenous VtrA and ectopically expressed N-terminal FLAG-tagged VtrC . Analysis of FLAG immunoprecipitates revealed that endogenous VtrA co-precipitated with VtrC , indicating that these two proteins interact in vivo ( Figure 5A , lanes 3 and 6 ) . VtrA and VtrC interact in the presence and absence of bile salts , suggesting that these proteins interact independently of bile salts before activation of T3SS2 by bile salts , and remain in a complex after treatment with bile salts ( Figure 5A ) . Notably , ectopic expression of FLAG-VtrC , but not FLAG-VtrC that is deleted for its N-terminal transmembrane domain ( FLAG-VtrCΔN30 ) , fully complemented the deletion of vtrC and restored the activity of T3SS2 , confirming that the wild type FLAG-tagged is functional ( Figure 5—figure supplement 1 ) . However , we noticed that the apparent molecular weight of full-length FLAG-VtrC ( 12 kDa ) is smaller than the expected size of this protein ( 20 kDa ) suggesting that VtrC may be processed in the periplasm . In any event , our results support that VtrC and VtrA interact in vivo . 10 . 7554/eLife . 15718 . 009Figure 5 . VtrA and VtrC form a complex . ( A ) Co-immunoprecipitation ( Co-IP ) of endogenous VtrA and vector induced FLAG-VtrC . pBAD vector induced N-terminal FLAG-tagged VtrC was immunoprecipitated with anti-FLAG affinity gel from V . parahaemolyticus POR1 derivative strains that express only VtrA ( Empty ) , only N-terminal FLAG-tagged VtrC ( ΔvtrA p-FLAG-vtrC ) , or both ( p-FLAG-vtrC ) . Protein-specific antibody was used to detect VtrA . Anti-FLAG antibody was used to detect FLAG-VtrC . -/+ Bile , V . parahaemolyticus grown in LB without bile salts ( - ) or supplemented with 0 . 05% bile salts ( + ) . Non-specific band is indicated with an asterisk . Data is representative of three independent experiments . ( B ) Top , gel filtration analysis of the VtrA/VtrC periplasmic domain complex , void volume of the column is indicated as V0 . Bottom , SDS-PAGE analysis of samples from the fractions corresponding to the elution peak of the complex . DOI: http://dx . doi . org/10 . 7554/eLife . 15718 . 00910 . 7554/eLife . 15718 . 010Figure 5—figure supplement 1 . N-terminal FLAG-tagged VtrC is functional . Expression ( Cell ) and secretion ( Sup ) of V . parahaemolyticus T3SS2 effector VopA by POR1 derivative strains with the empty pBAD vector ( Empty ) , vtrC deletion ( ΔvtrC Empty ) , vtrC complementation by N-terminal FLAG-tagged wild type ( ΔvtrC p-FLAG-vtrC ) or mutant vtrC ( ΔvtrC p-FLAG-vtrC ΔN30 ) that expresses pBAD vector induced protein under the arabinose inducible promoter . Protein-specific antibody was used to detect VopA . Anti-FLAG antibody was used to detect N-terminal FLAG-tagged VtrC ( both wild type and ΔN30 mutant ) . -/+ Bile , V . parahaemolyticus grown in LB without bile salts ( - ) or supplemented with 0 . 05% bile salts ( + ) . Loading control ( LC ) is shown for total protein lysate . Data is representative of three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 15718 . 010 Based on the orientation of VtrA and VtrC , we hypothesized these proteins would interact through their periplasmic domains . We found that the VtrA periplasmic domain ( aa 161–253 ) is soluble when expressed in E . coli; however , attempts to express and purify the periplasmic domain of VtrC were unsuccessful . We predicted that the VtrA/VtrC interaction might stabilize the VtrC periplasmic domain . Therefore , we co-expressed the periplasmic domains of these two proteins in E . coli: amino acids 161–253 of VtrA and N-terminal His-tagged 31–161 of VtrC . Both proteins co-purified by Ni-NTA-affinity chromatography and remained as a stable and soluble complex during size exclusion chromatography ( SEC ) . The VtrA/VtrC complex eluted as a 29 . 5 kDa species by SEC , indicating a 1:1 heterodimer of VtrA ( 11 . 0 kDa ) and VtrC ( 16 . 9 kDa ) periplasmic domains ( Figure 5B ) . VtrA alone appeared to elute as a monomer of approximately 11 kDa ( Figure 5B ) . To further understand the nature of the interaction between the periplasmic domains of VtrA and VtrC , we crystallized the complex and obtained its X-ray structure . The structure was solved by multiple-wavelength anomalous dispersion phasing using anomalous signals from selenomethionine and refined to a resolution of 2 . 70 Å using native data from an isomorphous crystal . Two of the five selenomethionines were used for phasing . The two N-terminal methionines and Se-Met 115 in the loop that covers the beta barrel of VtrC were not observed in the final electron density map and were not included in the final model . The asymmetric unit contains one complex with one copy each of the VtrC and VtrA periplasmic domains ( Figure 6A ) . The VtrA and VtrC subunits in the heterodimer make extensive interactions with each other ( Figure 6B ) , with an interface area of 1 , 149 Å2 ( Krissinel and Henrick , 2007 ) . An analysis of the macromolecular interfaces in this crystalline lattice by the web server PDBePISA ( http://www . ebi . ac . uk/pdbe/pisa/pistart . html ) ( Krissinel and Henrick , 2007 ) indicate that the VtrA/VtrC heterodimer is the only stable quaternary structure in solution . 10 . 7554/eLife . 15718 . 011Figure 6 . Structure of the VtrA/VtrC heterodimer . ( A ) Cartoon representation of the periplasmic domain complex formed by VtrA ( green ) and VtrC ( blue , light to dark gradient from N-terminus to C-terminus ) . Side chains of Hβ-Iβ loop residues are shown as sticks . ( B ) Detailed view of the VtrA/VtrC interface . Selected residues that form polar contacts ( black dashed lines ) , as well as potential bile salt binding residues are shown as sticks . ( C ) Overlay of surface and ribbon models of VtrC showing interior cavity . Side chains of residues lining the cavity are shown as sticks in yellow for hydrophobic residues ( Ala , Val , Ile , Leu , Met , Phe , Tyr , Trp ) and green for all other . DOI: http://dx . doi . org/10 . 7554/eLife . 15718 . 01110 . 7554/eLife . 15718 . 012Figure 6—figure supplement 1 . Activation of T3SS2 by individual bile acid . Secretion ( Sup ) of V . parahaemolyticus T3SS2 effector VopA by POR1 . Protein-specific antibody was used to detect VopA . V . parahaemolyticus grown in LB without bile salts ( - ) , supplemented with 0 . 05% crude bile ( Bile ) or 0 . 5 mM individual bile acids . CDC: chenodeoxycholate , CA: cholate , TDC: taurodeoxycholate , GDC: glycodeoxycholate . Loading control ( LC ) is shown for total protein lysate . Data is representative of three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 15718 . 012 The VtrC subunit consists of eight β-strands and a short α-helix . The eight β-strands of VtrC form a β-sheet meander that folds into itself to form a β-barrel with a small space between the first and last β-strands ( Aβ and Iβ ) ( Video 1 ) . The β-barrel is lined on the inside by hydrophobic side chains ( Figure 6C ) and is covered on one side by the short α-helix ( Fα ) and an adjacent disordered loop , which could not be modeled in its entirety due to the poor electron density of this region ( Figure 6A ) . The Hβ-Iβ loop is particularly rich in aromatic side-chains , promoting us to hypothesize that this loop could be involved in binding the steroid ring structure of bile acids ( Figure 6A , B ) . Despite the lack of sequence similarity to proteins of known function , VtrC shares structural similarity with lipocalins , fatty acid binding proteins ( FABPs ) , avidins , metalloprotease inhibitors ( MPIs ) , and other members of the calycin superfamily of proteins ( Figure 7 ) ( Flower et al . , 2000 ) . The inside of the calycin β-barrel often acts as a binding site for fatty acids and other hydrophobic molecules , such as retinol and biotin ( Flower et al . , 2000 ) . Comparison of the overall VtrC structure with other calycins suggests that it adopts a new fold within the superfamily . VtrC falls in between the lipocalins , which include a C-terminal helix that packs against the barrel on the corresponding VtrA interacting surface; and the FABPs , which include a β-hairpin insertion into the typical 8-stranded β-barrel in the position of the VtrC disordered loop . Interestingly , VtrA/VtrC would be the first member of this family that is an obligate dimer and not only binds a hydrophobic ligand but also transmits a signal upon binding . 10 . 7554/eLife . 15718 . 013Video 1 . Structure of the VtrA/VtrC heterodimer . Rotation of VtrA/VtrC heterodimer demonstrates the binding cavity of the beta-barrel . Further rotation shows how VtrA contributes a strand to the ‘open’ beta barrel formed by VtrC . Extensive hydrogen bonding is observed between VtrA and VtrC . DOI: http://dx . doi . org/10 . 7554/eLife . 15718 . 01310 . 7554/eLife . 15718 . 014Figure 7 . Structure based distance tree of VtrA/VtrC heterodimer and members of the calycin superfamily . Structures representatives of calycin superfamilies: MPI , avidin , FABPs , lipocalins , and triabin with bound ligands in the centers of the barrels ( magenta spheres ) were chosen . The conserved 8-stranded barrel core found in all structures is colored from dark to light grey for all representatives and dark to light blue for VtrC . The Hβ-Iβ loop containing the presumed VtrC helical lid ( red ) is unique to the subunit , as compared to the corresponding loops from representative structures ( light pink ) . This loop also corresponds to the position of the inserted FABP β-hairpin ( light pink ) . The functionally analogous helical lid in FABPs ( red ) is located after the first β-strand of the barrel . The triabin structure closes the binding pocket with hydrophobic residues from both the FABP lid loop ( red ) and the VtrC loop ( pink ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15718 . 014 The VtrA subunit adopts an alpha/beta fold with a five-stranded β-sheet and two α-helices on one side of the sheet ( Figure 6A ) . The bulk of the contacts with VtrC is through VtrA's five-stranded β-sheet on the side opposite to the α-helices and involves both hydrophobic and polar contacts . VtrA Fβ intercalates between VtrC Aβ and Iβ , making numerous polar contacts with VtrC Aβ and closing the gap in the VtrC β-barrel ( Figure 6B ) . The incorporation of structural elements from VtrA into the β-barrel fold suggests that VtrC may not fold properly in isolation and is consistent with our inability to purify the VtrC periplasmic domain in the absence of VtrA . Interestingly , the N-terminus of the VtrA periplasmic domain , leading to the transmembrane helix , contacts VtrC Gβ and is in close proximity to the aromatic-rich Hβ-Iβ loop ( Figure 6B ) , leading us to hypothesize that these residues might be involved in signal transmission across the membrane in response to ligand binding . Based on the structural similarities between the monomeric calycins and the VtrA/VtrC heterodimer , we predicted that VtrA/VtrC complex could bind bile salts . Previously , Gotoh et al . ( Gotoh et al . , 2010 ) established that bile salts are the component of bile that activates the T3SS2 . We have recapitulated this data and confirmed which purified bile salts could activate T3SS2 . We validated that taurodeoxycholate ( TDC ) and glycodeoxycholate ( GDC ) , but not chenodeoxycholate ( CDC ) or cholate ( CA ) , could activate the V . parahaemolyticus T3SS2 ( Figure 6—figure supplement 1 ) . Based on this information we moved forward with biophysical experiments to test whether the VtrA/VtrC heterodimer is the bile salt receptor using a relevant bile salt , TDC . We first tested whether the VtrA/VtrC periplasmic domain heterodimer could bind bile salts using isothermal titration calorimetry ( ITC ) . Negative power deflections were observed throughout the titration of the bile salt TDC into the VtrA/VtrC solution . These results indicated that TDC binds to the VtrA/VtrC heterodimer in an exothermic manner ( Figure 8 ) , with a dissociation constant ( KD ) of 315 . 4 nM ( Figure 8A ) . The stoichiometry of TDC binding to VtrA/VtrC is approximately 1:1 ( n = 0 . 94 ) . Taken together our results suggest that VtrA and VtrC form a functional complex that can bind bile salts . 10 . 7554/eLife . 15718 . 015Figure 8 . VtrA/VtrC periplasmic domain complex binds the bile salt taurodeoxycholate ( TDC ) . ( A ) ITC-derived binding curves of the VtrA/VtrC complex with TDC . Thermodynamic parameters were determined by global fitting of triplicate isotherms ( presented in black , red , and cyan ) . The dissociation constant ( KD ) and enthalpy ( ΔH ) values are reported followed by the 1σ error intervals in parenthesis . Data is representative of two independent experiments . ( B ) Structure of the VtrA/VtrC periplasmic domain complex binding TDC ( green ) . VtrA and VtrC follow same coloring scheme as in Figure 6 . ( C ) Detailed view of the TDC binding site . Hydrogen bonds between protein and TDC are represented as dashed lines . DOI: http://dx . doi . org/10 . 7554/eLife . 15718 . 01510 . 7554/eLife . 15718 . 016Figure 8—figure supplement 1 . Electron density around TDC molecules . Kicked Fo-Fc omit maps of the three TDC molecules bound to the A/B , C/D and E/F heterodimers ( A ) and the TDC mediating lattice contacts ( B ) in the TDC-containing crystal . The maps are shown as grey mesh ( contoured at 3 σ in A and 1 . 5 σ in B ) and carved around the ligand at a 1 . 6 Å radius . DOI: http://dx . doi . org/10 . 7554/eLife . 15718 . 01610 . 7554/eLife . 15718 . 017Figure 8—figure supplement 2 . Comparison between apo and TDC bound structure heterodimers . ( A ) Superposition of all the heterodimers in the asymmetric unit of the TDC-bound crystal . ( B–D ) Superposition of AB ( B ) , CD ( C ) and EF ( D ) heterodimers from the TDC-bound complex crystal structure and apo heterodimer . ( E ) Root-mean-square deviations ( RMSD ) for each superposition with the number of aligned α-carbon atoms in parethesis . Superpositions were made via the DaliLite server ( http://ekhidna . biocenter . helsinki . fi/dali_lite/start ) ( Holm and Park , 2000 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15718 . 017 We next purified the VtrA/VtrC heterodimer in the presence of the bile salt TDC and crystalized this complex and obtained its X-ray structure . The crystal structure was solved via molecular replacement by using our original VtrA/VtrC heterodimer as a search model , and refined to a resolution of 2 . 10 Å . The model reveals three VtrA/VtrC heterodimers in the asymmetric unit , each with clear electron density for one TDC molecule bound inside the β-barrel ( Figure 8B , Figure 8—figure supplement 1A ) . A fourth TDC molecule mediated lattice contacts between VtrA chain A and its symmetry mate ( Figure 8—figure supplement 1B ) . An analysis of the macromolecular interfaces found in this crystalline lattice by the web server PDBePISA ( Krissinel and Henrick , 2007 ) indicated that the heterodimer is the only stable quaternary structure in solution . Although the three heterodimers in the asymmetric unit are highly similar , average root mean squared deviations ( RMSD ) of 0 . 5 Å for VtrA and 0 . 6–1 . 5 Å for VtrC indicates that there are differences in the VtrC monomer ( Figure 8—figure supplement 2A , E ) . RSMD values of 0 . 4–0 . 5 Å for VtrA and 1 . 2–1 . 9 Å were obtained by superimposing each of the TDC-bound heterodimers to the apo structure ( Figure 8—figure supplement 2B-–E ) , indicating that the overall fold is maintained but there are considerable changes in the VtrC monomer . The largest differences between the VtrA/VtrC-TDC and the apo heterodimers are in the VtrC disordered loop and Fα that follows it . In the VtrA/VtrC-TDC heterodimer , the disordered loop has been displaced by the TDC molecule and is now observable in two out of the three heterodimers in the crystallographic asymmetric unit . Residues Thr122 and Ser123 , which were in the disordered loop in the apo structure , now form part of Fα in two of the TDC-bound heterodimers . Remarkably , one of these residues ( Ser123 ) now forms a hydrogen bond with the 12α-hydroxyl group of TDC ( Figure 8C ) . The main chain amide between F150 and Tyr151 ( Hβ-Iβ loop ) coordinates the 3α-hydroxyl group ( Figure 8C ) , which is present in all bile salts , suggesting that this interaction is important for specificity towards bile salts . Several VtrC residues in the binding pocket maintain hydrophobic interactions with TDC , including the side chains of Phe150 with steroid ring A and Tyr81 with the hydrophobic part of the bile acid ( Figure 8C ) . Based on our co-crystal structure of VtrA/VtrC with TDC , the role of several residues that contact the bile salt were investigated for their ability to affect the binding of bile salts and activation of the T3SS2 ( Figure 9 ) . The mutation of His50 , located at the top of the hydrophobic cleft , to an arginine residue is expected to produce steric hindrance by insertion of a large side chain into the bile salt binding pocket ( Figure 9A ) . When VtrC-H50R was expressed in the vtrC deletion strain and bile salts were added , the T3SS2 could not be induced as indicated by the lack of expression and secretion of the T3SS2 effector VopA ( Figure 9B , lanes 5 and 11 ) . We also changed Tyr81 to an alanine residue , which we predicted would disrupt bile salt binding by eliminating the interaction between the Tyr81 aromatic ring and the bile salt side chain ( Figure 9A ) . When VtrC-Y81A was expressed in the vtrC deletion strain and bile salts were added , the T3SS2 was not induced as indicated by the lack of expression and secretion of VopA ( Figure 9B , lanes 6 and 12 ) . As a control , we chose to mutate a surface exposed residue , Gln42 , to an alanine ( Figure 9A ) . As expected , this mutant rescued the vtrC deletion strain and upon addition of bile salts the T3SS2 was activated as indicated by the expression and secretion of VopA ( Figure 9B , lanes 4 and 10 ) . To assess whether these mutations had any effect on the VtrA/VtrC complex formation , we co-immunoprecitated the complex with the N-terminal FLAG-tagged VtrC . As seen in Figure 9C , the wild type and mutant VtrC proteins formed a stable complex with endogenous VtrA , as indicated by their co-immunopreciptation . 10 . 7554/eLife . 15718 . 018Figure 9 . Mutations in the hydrophobic chamber of VtrA/VtrC heterodimer disrupt signaling mediated by bile salts . ( A ) Structure of VtrA/VtrC heterodimer . Q42 , H50 and Y81 of VtrC are highlighted in yellow . ( B ) Restoration of T3SS2 activation in POR1ΔvtrC by VtrC mutant Q42A but not H50R or Y81A . Expression ( Cell ) and secretion ( Sup ) of V . parahaemolyticus T3SS2 effector VopA by POR1 derivative strains with the empty pBAD vector ( Empty ) , vtrC deletion ( ΔvtrC Empty ) , vtrC complementation by N-terminal FLAG-tagged wild type ( ΔvtrC p-FLAG-vtrC ) or mutant ( Q42A , H50R or Y81A ) VtrC that expresses pBAD vector induced protein under the arabinose inducible promoter . Protein-specific antibody was used to detect VopA . Anti-FLAG antibody was used to detect N-terminal FLAG-tagged VtrC . -/+ Bile , V . parahaemolyticus grown in LB without bile salts ( - ) or supplemented with 0 . 05% bile salts ( + ) . Loading control ( LC ) is shown for total protein lysate . Dashed lines indicate lane that was deleted from the gel . ( C ) Co-IP demonstrating that all three VtrC mutants form a stable complex with VtrA . DOI: http://dx . doi . org/10 . 7554/eLife . 15718 . 018
Bile sensing plays a significant role in human infection caused by various enteric bacterial pathogens ( Begley et al . , 2005 ) . Deciphering how bacteria use bile as a signal to regulate the expression of virulence genes is essential to our understanding of disease mechanisms and also can offer insights for designing novel therapeutic strategies . Bile salt-induced T3SSs are the primary virulence factor in many Vibrio species ( Broberg et al . , 2011 ) . In this study , we uncovered a previously unidentified component of a widespread , conserved signaling platform that is essential for bile salt sensing and activation of virulence factors in Vibrios . Our biochemical data and the structural analysis demonstrated that the periplasmic domains of the inner membrane proteins VtrA and VtrC form a stable complex that functions as a receptor for bile salts . The crystal structure of the VtrA/VtrC complex reveals that their periplasmic domains form a obligate heterodimer where VtrC folds into an open eight-stranded β-barrel and VtrA adopts a mixed α/β structure whereby one of its a β-strands contributes to the formation of the β-barrel . The VtrA/VtrC heterodimer fold is similar to members of the calycin superfamily , which contain eight to ten-stranded β-barrels that often bind hydrophobic molecules ( Flower et al . , 2000 ) . Calycins have low sequence similarity , but relatively high structural conservation . They can be subdivided into several families based on structural details characteristic of each group ( Figure 7 ) ( Flower et al . , 2000 ) . Calycin families are markedly different in their distribution among biological kingdoms , cellular localization , and function , e . g . : ( 1 ) FABPs are cytosolic proteins involved in lipid homeostasis in animals , ( 2 ) MPIs are bacterial proteins that are secreted into the periplasm to inhibit metalloproteases , ( 3 ) lipocalins are found in both eukaryotes and prokaryotes , and they are mostly extracellular or attached to the outer membrane and bind a variety of ligands ( Flower et al . , 2000 ) . In terms of structure , VtrC does not seem to cluster with any particular family and it forms its own clade ( Figure 7 ) . Although calycins have been found to form complexes with other proteins , our inspection of the available structures of complexes did not reveal any arrangement that resembled that of the VtrA/VtrC obligate heterodimer . This complex is unique for this family because it is formed from two proteins and it is a complex that transmits a signal upon binding its ligand . Interestingly , members of the FABP family have been found to bind bile salts . This is exemplified by human and porcine ileal lipid binding proteins ( ILBPs ) , and their NMR structures in complex with taurocholate ( Kurz et al . , 2003 ) and glycocholate ( Lücke et al . , 2000 ) , respectively . Analogous to VtrC's disordered loop and short α-helix , FABPs have a helix-turn-helix motif that forms a lid over one side of the barrel and is proposed to act as a portal for ligand access ( Storch and McDermott , 2009 ) . In the case of FABPs , it has been suggested that ligand binding induces subtle conformational changes that promote protein-membrane or protein-protein interactions ( Storch and McDermott , 2009 ) . We observe a similar scenario for the VtrA/VtrC obligate heterodimer bound to bile salts where , upon binding bile salt , it appears an unstructured loop is translocated from the inside of the hydrophobic chamber to the outside of the chamber , where it becomes structured . Future studies will focus on how a conformational change during bile salt binding may induce a switch in the VtrA/VtrC interaction that will result in activation of the VtrA cytoplasmic transcription activator domain ( Figure 9 ) . The closest example with regard to complex formation and function is the GrlR/GrlA complex , a regulator of the T3SS-encoding locus of enterocyte effacement ( LEE ) in enterohaemorrhagic and enteropathogenic E . coli strains ( EHEC and EPEC , respectively ) ( Padavannil et al . , 2013 ) . GrlR is a cytosolic lipocalin that represses the LEE repressor GrlA by binding to its helix-turn-helix ( HTH ) domain . Similar to vtrA and vtrC , grlR and grlA are transcribed from a bicistronic operon , although the order of the lipocalin and transcriptional regulator genes is reversed in the latter pair . Analytical ultracentrifugation and the structure of GrlR in complex with the GrlA HTH domain revealed a heterotrimeric complex where a GrlR dimer binds one GrlA HTH monomer ( PDB ID: 4KT5 ) ( Padavannil et al . , 2013 ) . It has been shown that the inner cavity of GrlR binds lipids ( Jobichen et al . , 2009 ) . However , the role of lipid binding in this system is unknown . Although GrlR and VtrC have analogous roles regulating T3SS through their interaction with a transcription factor , they do so from different cellular compartments at different stoichiometries , and their binding partners ( GrlA HTH domain and VtrA periplasmic domain , respectively ) have different folds . Several signaling features of VtrA/VtrC appear to parallel two other pairs of regulatory proteins found in V . cholerae , ToxR/ToxS and TcpP/TcpH , which control the expression of numerous virulence factors including CT and TCP . First , the genes encoding the proteins in each pair are arranged in bicistronic operons ( Miller et al . , 1989; Häse and Mekalanos , 1998 ) . Second , VtrA/ToxR/TcpP and VtrC/ToxS/TcpH have the same cellular localization and membrane topology ( Miller et al . , 1989; Häse and Mekalanos , 1998; Miller et al . , 1987 ) . Third , VtrC , ToxS and TcpH affect the activity of VtrA , ToxR and TcpP , respectively . On this last point , we showed that VtrA and VtrC form a complex in vivo both before and after T3SS2 activation and in vitro through their periplasmic domains . Our data supports a model in which VtrA and VtrC form a complex both in the absence and presence of bile salts . Upon binding bile salts to the hydrophobic chamber in the VtrA/VtrC complex , the cytoplasmic DNA binding domain of VtrA is activated and in turn induces VtrB to activate the T3SS2 ( Figure 10 ) . The specific changes caused by binding bile salts and their effects on the VtrA DNA-binding domain are yet to be determined . Although VtrA contains no cysteine residues in its periplasmic domain , it is still possible that bile salts increase VtrA activity by inducing its dimerization or oligomerization , similar to what has been proposed for the ToxR/ToxS and TcpP/TcpH systems ( Ottemann and Mekalanos , 1996; Yang et al . , 2013 ) . 10 . 7554/eLife . 15718 . 019Figure 10 . Model for bile salts sensing and T3SS2 regulation in V . parahaemolyticus . In the absence of bile salts , VtrA ( green ) and VtrC ( blue ) form a complex where the VtrA cytoplasmic DNA binding domain is kept inactive . Upon binding of bile salts , the VtrA/VtrC complex activates the VtrA DNA binding domain by a yet to be determined mechanism , which induces VtrB ( purple ) expression . VtrB , in turn , activates T3SS2 . DOI: http://dx . doi . org/10 . 7554/eLife . 15718 . 019 The VtrA/VtrC complex is highly conserved in a group of diverse Vibrionaceae family species , as well as in Moritella marina . While the bile salt-induced T3SS in V . cholerae non-O1/O139 strains is necessary for bacterial colonization and the primary cause of disease during infection ( Chaand et al . , 2015 ) , the role of T3SS in G . hollisae infection is yet to be characterized . For other bacteria encoding vtrA , vtrC and vtrB , such as Photobacterium marinum and V . caribbeanicus , only a few T3SS related genes or no T3SS related genes are observed in their genomes , respectively . The differences between T3SS gene representation in these bacteria demonstrate this system's diversity and imply that the T3SS may be evolving in response to different living environments . Based on surrounding genes , this bile salt receptor complex may have adapted in other species to bind other hydrophilic ligands . For example , in the ten strains that lack a vtrB homolog but possess a vtrA/vtrC pair , the gene downstream of the vtrA/vtrC operon encodes a putative sphingomyelin phosphodiesterase . This enzyme is a hydrolase involved in the metabolism of sphingolipids ( Goñi and Alonso , 2002 ) and is a putative virulence factor ( Oda et al . , 2012 ) . It is tempting to hypothesize that the homologues of VtrA/VtrC in these bacteria function as a sensor for sphingolipids and regulate the induction of the downstream gene that produces sphingomyelin phosphodiesterase . In conclusion , using microbial genetics and biophysics , we have identified and characterized a bile salt sensor that is widely distributed and used to induce virulence by many Vibrios . Additionally , we have found that a family of monomeric lipid binding calycin domain proteins has expanded to include an obligate heterodimer that binds bile salts and can be used to transmit a signal . To our knowledge , our study provides the first biochemical and structural analysis of a prokaryotic receptor involved in mediating a response to bile salts , a significant environmental cue during infection .
The V . parahaemolyticus clinical RimD2210633 derivative strain POR1 ( ΔtdhAS ) and its derivative strains were cultured in Marine LB ( MLB ) broth ( LB broth containing 3% NaCl ) or Marine minimal medium ( MMM ) at 30°C . E . coli DH5α , E . coli S17 ( λ pir ) , B834 ( DE3 ) , Rosetta 2 ( DE3 ) , and BL21 ( DE3 ) were cultured in LB broth at 37°C . The medium was supplemented with kanamycin ( 30 μg/ml for E . coli and 240 μg/ml for V . parahaemolyticus ) or chloramphenicol ( 25 μg/ml ) when necessary . For vector induced expression of VtrC that is regulated by its native promoter , the 1kb upstream of vtrA followed by the vtrC coding sequence was PCR amplified and cloned into the pBAD/Myc-His vector ( Invitrogen ) in which ampicillin resistance was replaced with kanamycin resistance . To generate pBAD-FLAG-vtrCΔN30 , vtrCΔN30 ( aa 31–161 ) was amplified by PCR ( a FLAG epitope was added via the forward primer ) and cloned into pBAD/Myc-His under the control of the arabinose inducible promoter . Amino acid residues 1 to 30 are predicted to encode the N-terminal transmembrane region . Since membrane localization is predicted to be important for the activity of VtrC , the VtrCΔN30 mutant protein should localize to the cytoplasm and , thus , be inactive . PhoA and LacZ fusions to VtrC in pBAD/Myc-His were generated by PCR amplifying phoA ( amino acids 22–471 , lacking the endogenous signal peptide ) and the full length lacZ ( amino acids 1–1024 ) and cloning directly upstream of VtrC aa 1 for N-terminal fusions or downstream of VtrC's last residue ( aa 161 ) for C-terminal fusions . To generate the vtrC deletion , the 1kb upstream and downstream of the nucleotide sequence for vtrC gene ( nucleotides 34–486 ) to be deleted were cloned into the pDM4 , a CmROriR6K suicide plasmid containing the sacB gene , which encodes an enzyme that metabolizes sucrose into a toxic product . C-terminal FLAG tag knock-in of vopD2 and vopS were generated by cloning the 1kb upstream of the knock-in site , followed by the nucleotide sequence encoding the FLAG tag and the 1kb downstream of the knock-in site into pDM4 . The resulting pDM4 plasmids were conjugated into the POR1 strain from E . coli S17 ( λ pir ) and the transconjugants were selected on medium containing 25 μg/ml chloramphenicol . Bacteria were then counterselected by growing on medium containing 15% sucrose to select for clones were the suicide plasmid had recombined out of the chromosome . PCR analysis was performed to confirm successful deletions and knock-ins and that the vtrA gene remained intact . FLAG antibodies were purchased from Sigma-Aldrich ( F3165 and F7425 ) . Polyclonal antibodies were produced in house with rabbits for the recombinant proteins for VopA and VopC and for the VtrA peptide aa 235–253 . In order to identify VtrA and VtrC homologues , we performed PSI-BLAST ( Altschul et al . , 1997 ) searches against a non-redundant ( NR ) database ( threshold E-value cutoff 0 . 02 ) with query sequences corresponding to the predicted periplasmic domains of VtrA ( VPA1332: gi|28901187 , aa 150–253 ) and VtrC ( VPA1333: gi|28901188 , aa 27–161 ) . Because our VtrC PSI-BLAST hits were from different species than our VtrA hits , we initiated a transitive PSI-BLAST search against the NR database ( threshold E-value cutoff 0 . 01 ) using the most distant identified VtrA sequence as a query ( gi|494726765 , aa 160–272 ) . The transitive search found the VtrA sequences from the remaining species containing VtrC . All identified VtrA and VtrC hits were found in tandem in their respective genomes , and representative sequences from non-redundant species were collected . VtrB sequences were identified by searching VtrA/VtrC-containing genomes for the VtrB sequence ( VPA1348: gi|28901203 ) . Due to widespread nature and strong conservation of the VtrB helix-turn-helix sequence in many transcription factors , we required VtrB orthologs to: 1 . be the top-scoring hit in the respective genomes , 2 . retain the predicted C-terminal transmembrane helix and 3 . be in close proximity to the VtrA/VtrC gene pair . Representatives from each sequence family were aligned using the MAFFT server ( http://mafft . cbrc . jp/alignment/server/ ) . To calculate the structure-based distance tree , representative structures were chosen from the lipocalins/streptavidin group in the Evolutionary Classification Of protein Domain structures ( ECOD ) database . We chose structures with bound ligands , when available , from the main branches of the calycin superfamily ( Flower et al . , 2000 ) : lipocalins ( 1kt7:1–175 , 2aco:10–177 ) , triabin ( 4n7c:3–176 ) , fatty acid binding proteins ( FABPs ) ( 1o1v: 1–126 , 3wvm:0–132 ) , streptavidin ( 2f01:14–134 ) , and metalloprotease inhibitor ( MPI ) ( 1jiw:1–105 , chain I ) . We also included the HRI1 N-terminal domain structure ( HRIN-like; 3rby:1–151 ) with a unique binding site that traverses the lipocalin-like barrel core , as well as the VtrC structure . The representative lipocalin-like domains were compared pairwise all-against-all using DaliLite ( Holm and Park , 2000 ) . Pairwise DaliLite Z-scores ( ZAB ) were transformed to distances by comparing to self DaliLite Z-scores ( ZAA , ZBB ) using the following equation: -ln ( ZAB/ ( 0 . 5*ZAA+0 . 5*ZBB ) ) . The structure-based tree was produced using the FITCH program ( with global optimization ) of the Phylip package ( Felsenstein , 1997 ) . Total RNA was extracted from V . parahaemolyticus strains using RNeasy Plus Mini Kit ( QIAGEN 74134 ) . Extracted RNA was then reverse transcribed into cDNA using ProtoScript First Strand cDNA Synthesis Kit ( NEB E6300S ) utilizing Random Primer Mix . The resulting cDNA served as the template for quantitative RT-PCR analysis using iTaq Universal SYBR Green Supermix ( Bio-Rad ) and the ViiA7 Real-Time PCR System ( Applied Biosystems ) . The 2-ΔΔCT method was used to determine the mRNA level of vtrA and vtrB in each sample relative to POR1 grown in LB without bile salts . The expression of vtrA and vtrB was normalized against the expression of fliA . Primers for specific genes were as follows: vtrA , 5’-TTGGAACCCACGAACATCTC-3’ and 5’-CAGTCACAAATTTTCCTGGCC-3’; vtrB , 5’-ATTATCAGCTTAGGTGGGCG-3’ and 5’-ACTTTACCCCACACTTTGTCG-3’; control gene fliA , 5’-AAGCGATAACCTATGACCAGC-3’ and 5’-TCCTCTACCTGAACACTCGG-3’ . For the PCR that tests if vtrA and vtrC are cotranscribed , the cDNA served as the template and the primers used were: 5’- AATTGTTCCAGAAAGGCTCTATGTCATGCTTAATG’-3 and 5’- GTTTCATAAAAATGAACTGGTTGAAAAAAATTG-3’ . V . parahaemolyticus strains were grown overnight in MLB at 30°C . For experiments that involved vector induced expression of VtrC under its endogenous promoter , overnight cultures were diluted to OD600 nm = 0 . 3 in LB supplemented with 0 . 05% bile salts or 0 . 5 mM individual bile acid taurodeoxycholate ( TDC ) , glycodeoxycholate ( GDC ) , chenodeoxycholate ( CDC ) or cholate ( CA ) and induced for 4 hr at 37°C . For experiments that involved vector induced expression of VtrC under the arabinose inducible promoter of pBAD , overnight cultures were diluted to OD600 nm = 0 . 6 in MLB supplemented with 0 . 1% arabinose and induced for 3 hr at 30°C . The cultures were then diluted to OD600nm = 0 . 3 in LB supplemented with 0 . 1% arabinose and 0 . 05% bile salts and induced for 4 hr at 37°C . For the expression fraction ( cell ) , the same OD600 nm of bacterial cultures were collected and cell pellets were resuspended in 2x protein sample buffer ( 100 mM Tris·HCl pH 6 . 8 , 20% glycerol , 2% sodium dodecyl sulfate ( SDS ) , 2% β-mercaptoethanol , 150 mM sodium hydroxide , bromophenol blue ) . For the secretion fraction ( sup ) , bacterial culture supernatants were filtered and precipitated with deoxycholate and trichloroacetic acid ( Kimata et al . , 2004 ) . Precipitated proteins were pelleted and washed with acetone and then resuspended in 2x protein sample buffer . Protein expression and secretion were detected by western blot analysis . V . parahaemolyticus strains were grown overnight in MLB at 30°C . Overnight cultures were diluted to OD600 nm = 0 . 3 in LB or Dulbecco's Modified Eagle Medium ( DMEM ) and induced for 4 hr at 37°C . For the expression fraction ( cell ) , the same OD600 nm of bacterial cultures were collected and cell pellets were resuspended in 2x protein sample buffer . For the secretion fraction ( sup ) , bacterial culture supernatants were filtered and precipitated with deoxycholate and trichloroacetic acid ( Kimata et al . , 2004 ) . Precipitated proteins were pelleted and washed with acetone and then resuspended in 2x protein sample buffer . Protein expression and secretion were detected by western blot analysis . V . parahaemolyticus strains containing the empty pBAD vector , pBAD-phoA-vtrC , or pBAD-vtrC-phoA were grown overnight in MLB at 30°C . Overnight cultures were diluted to OD600 nm = 0 . 6 in MLB supplemented with 0 . 1% arabinose and induced for 5 hr at 30°C . 1 ml of bacterial culture was collected , washed once , and resuspended with 1 ml Tris solution ( 1M Tris·HCl , pH 8 . 0 ) . 50 μl of resuspended cells were transfered to 1 ml Tris solution and permeabilized by adding 30 μL 0 . 1% SDS and vortexing for 10 s . The mixture was then incubated at 37°C for 5 min . 100 μl of p-nitrophenyl phosphate ( PNPP , Thermo Scientific Pierce 37621 ) was added to the samples to start the reaction and the time was recorded as T1 ( min ) . Samples were incubated at 37°C until a pale yellow color develops . The reaction was stopped by adding 100 μl of cold 1 M KH2PO4 and the time was recorded as T2 ( min ) . The supernatant was collected after centrifugation at 20000 x g for 5 min . Measure the absorbance of the supernatant with a spectrophotometer at wavelength 420nm . Alkaline phosphatase activity was calculated as below: Alkaline phosphatase activity ( Miller unit ) = 1000* OD420 nm/0 . 05* OD600 nm * ( T2-T1 ) V . parahaemolyticus strains containing the empty pBAD vector , pBAD-lacZ-vtrC or pBAD-vtrC-lacZ were grown overnight in MLB at 30°C . Overnight cultures were diluted to OD600 nm = 0 . 6 in MLB supplemented with 0 . 1% arabinose and induced for 5 hr at 30°C . Bacterial cells were permeablized by mixing 20 μl of bacterial culture with 80 μl permeabilization solution ( 100 mM Na2HPO4 , 20 mM KCl , 2 mM MgSO4 , 0 . 8 mg/ml hexadecyltrimethylammonium bromide , 0 . 4 mg/ml sodium deoxycholate , 5 . 4 μl/ml β-mercaptoethanol ) . 600 μl substrate solution ( 60 mM Na2HPO4 , 40 mM NaH2PO4 , 1 mg/ml 2-Nitrophenyl β-D-galactopyranoside ( ONPG , Sigma-Aldrich N1127 ) , 2 . 7 μl/ml β-mercaptoethanol ) was added to the samples to start the reaction and the time was recorded as T1 ( min ) . Samples were incubated at 30°C until a pale yellow color develops . The reaction was stopped by adding 100 μl 1 M Na2CO3 and the time was recorded as T2 ( min ) . The supernatant was collected after centrifugation at 20000 x g for 5 min . Measure the OD420nm of the supernatant . β-galactosidase activity was calculated as below: β-galactosidase activity ( Miller unit ) = 1000* OD420 nm/0 . 02* OD600 nm * ( T2-T1 ) The following strains were used for the Co-IP experiments on Figure 5C: V . parahaemolyticus strains POR1 + pBAD , POR1ΔvtrA + pBAD-FLAG-vtrC and POR1 + pBAD-FLAG-vtrC . The following strains were used for the Co-IP experiments on Figure 9C: V . parahaemolyticus strains POR1 + pBAD , POR1ΔvtrA + pBAD-FLAG-vtrC and POR1 ΔvtrC + pBAD-FLAG-vtrC wild type and mutants H50R , Y81A and Q42A . Strains were grown overnight in MLB at 30°C . Overnight cultures were diluted to OD600 nm = 0 . 6 in MLB supplemented with 0 . 1% arabinose and induced for 3 hr at 30°C . The cultures were then diluted to OD600 nm = 0 . 3 in LB supplemented with 0 . 1% arabinose and 0 . 05% bile salts and induced for 4 hr at 37°C . 200 ml bacterial culture were resuspended with lysis buffer ( 50 mM Tris·HCl pH 7 . 4 , 100 mM NaCl , 0 . 2% Triton X-100 , 5 mM EDTA , 1 mg/ml lysozyme and 1 mM PMSF ) . Resuspended samples were shaken at 22°C for 30 min and then subjected to three freeze/thaw cycles . The supernatant was collected after centrifugation at 25 , 000 x g , 4°C for 30 min . Immunoprecipitation was performed by incubating the supernatant with Anti-FLAG M2 beads at 4°C with gentle shaking for 4 hr . Beads were collected and washed with wash buffer ( 50 mM Tris·HCl pH 7 . 4 , 100 mM NaCl , 0 . 2% Triton X-100 ) 3 times . Proteins bound to beads were eluted with 2x protein sample buffer . VtrA and VtrC were detected by western blot analysis using Anti-VtrA and Anti-FLAG antibodies . The periplasmic domain of VtrA ( aa 161–253 ) was cloned as an N-terminally hexahistidine and maltose binding protein-tagged fusion protein into pET28b ( Novagen ) to produce pET28b-HisMBP-VtrA , which contains a Tobacco Etch Virus ( TEV ) protease cleavage site between the HisMBP-tag and VtrA . The pET28b-HisMBP-VtrA construct was expressed in E . coli Rosetta 2 ( DE3 ) cells ( Novagen ) . For VtrC/VtrA coexpression , the periplasmic domains of VtrC ( aa 31–161 ) and VtrA ( aa 161–253 ) were cloned into the first and second multiple cloning site , respectively , of pACYCDuet-1 ( Novagen ) to produce pACYCDuet-VtrC/VtrA , where VtrC is N-terminally hexahistidine-tagged . The pACYCDuet-VtrC/VtrA construct was expressed in E . coli BL21 ( DE3 ) cells . All cultures were grown in LB at 37°C until OD600nm 0 . 5–0 . 6 and induced with 0 . 4 mM isopropyl β-D-thiogalactopyranoside ( IPTG ) overnight at 22°C for pET28b-HisMBP-VtrA , and 17°C for pACYCDuet-VtrC/VtrA . Selenomethionyl-derivatized VtrA/VtrC complex was expressed in E . coli B834 ( DE3 ) cells ( Novagen ) in SelenoMet medium ( Molecular Dimensions ) . Cells were harvested by centrifugation , resuspended in buffer A ( 50 mM Tris pH 8 . 0 and 100 mM NaCl ) , and lysed by extrusion . Lysates were clarified by centrifugation and filtered ( 0 . 45-μm pore size ) . All proteins were purified by nickel-affinity purification using HisPur Ni-NTA resin ( ThermoFisher , Rockford , Illinois ) on a gravity flow column . Briefly , lysates were incubated with the resin for 30 min at 4°C with nutation . Lysate and beads were applied to the column and washed with 20 column volumes of buffer A supplemented with 15 mM imidazole . Proteins were eluted with 5 column volumes of buffer B ( 50 mM Tris pH 8 . 0 , 100 mM NaCl and 250 mM imidazole ) . For VtrA , the HisMBP fusion domain was removed by treating with TEV protease overnight at 4°C , followed by a second round of nickel-affinity chromatography . Proteins were further purified by size exclusion chromatography ( SEC ) on a Superdex 75 16/600 column ( Pharmacia Biotech ) with buffer A . All proteins were buffer exchanged to 10 mM Tris pH 8 and 10 mM NaCl for crystallographic studies . Selenomethionyl-derivatized VtrA/VtrC complex was purified in a similar manner with the addition of 1 mM DTT to all buffers . Gel filtration was performed on a Superdex S75 16/600 column ( Pharmacia Biotech ) with buffer containing 50 mM Tris pH 8 . 0 and 100 mM NaCl at a flow rate of 1 ml/min . The protein complex solution was injected into the column at a final concentration of 1 mg/ml in a total volume of 1 ml . Fractions were analyzed by SDS-PAGE and Coomassie blue staining . For protein molecular weight determination , the column was calibrated with a gel filtration LMW calibration kit ( GE Healthcare ) using the running buffer described above . Crystals of native VtrA/VtrC periplasmic domain heterodimer were grown using the sitting-drop vapor diffusion method from drops containing 0 . 2 μl protein ( 13 mg/ml ) and 0 . 2 μl reservoir solution ( 1 . 0 M lithium sulfate , 0 . 5 M ammonium sulfate , 0 . 1 M sodium citrate pH 5 . 6 ) and equilibrated over 50 μl reservoir solution at 20°C . Crystals of selenomethionyl-derivatized heterodimer were grown using the hanging-drop vapor diffusion method from drops containing 1 μl protein ( 10 mg/ml ) and 1 μl of reservoir solution ( 1 . 0 M lithium sulfate , 0 . 5 M ammonium sulfate , 0 . 1 M sodium citrate pH 5 . 9 ) and equilibrated over 500 μl of reservoir solution . Crystals appeared after 2 days at 20°C and grew to their maximal extent by 1 week . Native and selenomethionyl-derivatized crystals were cryoprotected by transferring to a final solution of 28 . 7% ethylene glycol , 1 . 1 M lithium sulfate , 0 . 6 M ammonium sulfate , and 0 . 1 M sodium citrate pH 5 . 6–5 . 8 , then the crystals were flash-cooled in liquid nitrogen . Crystals of the VtrA/VtrC heterodimer bound to the bile salt TDC were grown with protein that had been loaded with TDC by performing purification in the presence of 0 . 5 mM TDC in all buffers and buffer exchanging into 10 mM Tris pH 8 , 10 mM NaCl . Drops containing 1 μl protein ( 5 mg/ml ) and 1 μl of reservoir solution ( 2 . 0 M ammonium sulfate , 0 . 1 M sodium acetate pH 4 . 6 ) were set up by the hanging-drop vapor diffusion method , and equilibrated over 500 μl of reservoir solution . Crystals appeared after 6 days at 20°C and grew to their maximal extent by 2 weeks . Crystals were cryoprotected by transferring to a final solution of 22 . 5% ethylene glycol , 2 . 1 M ammonium sulfate , 0 . 1 M sodium acetate pH 4 . 6 , and 0 . 5 mM TDC , and flash-cooled in liquid nitrogen . Data were collected at APS beamline 19-ID at 100 K , and were indexed , integrated and scaled using the HKL-3000 program package ( Minor et al . , 2006 ) . Native and selenomethionyl-derivatized VtrA/VtrC heterodimer crystals exhibited the symmetry of space group F432 with unit cell parameters of a = 211 . 01 Å and contained one molecule each of VtrA and VtrC per asymmetric unit , with a solvent content of 65% . TDC-containing crystals belonged to space group P212121 with unit cell parameters of a = 55 . 39 Å , b = 71 . 28 Å and c = 203 . 73 Å and contained three molecules each of VtrA/VtrC heterodimer per asymmetric unit , with a solvent content of 50% . Native , selenomethionyl-derivatized and TDC-containing crystals diffracted isotropically to a dmin of 2 . 70 Å , 2 . 60 Å and 2 . 10 Å , respectively , when exposed to synchrotron radiation . Data collection statistics are provided in Table 1 . 10 . 7554/eLife . 15718 . 020Table 1 . Data collection and refinement statistics , VtrC/VtrA complex . DOI: http://dx . doi . org/10 . 7554/eLife . 15718 . 020Data collectionCrystalSeMetpeak*SeMet inflection point*NativeNative + bile saltSpace groupF432 F432 F432 P212121 Cell constants ( Å ) a = 211 . 01 a = 211 . 46 a = 211 . 39 a = 55 . 39 , b = 71 . 28 , c = 203 . 73 Wavelength ( Å ) 0 . 97927 0 . 97943 0 . 97935 0 . 97926 Resolution range ( Å ) 35 . 67–2 . 60 ( 2 . 64–2 . 60 ) 35 . 74–2 . 65 ( 2 . 70–2 . 65 ) 40 . 68–2 . 65 ( 2 . 70–2 . 65 ) 35 . 64–2 . 10 ( 2 . 14–2 . 10 ) Unique reflections12 , 839 ( 620 ) 12 , 267 ( 593 ) 12 , 322 ( 594 ) 45 , 762 ( 1868 ) Multiplicity22 . 6 ( 20 . 2 ) 22 . 7 ( 21 . 8 ) 36 . 1 ( 36 . 5 ) 3 . 9 ( 2 . 8 ) Data completeness ( % ) 99 . 9 ( 100 . 0 ) 99 . 9 ( 100 . 0 ) 99 . 9 ( 100 . 0 ) 95 . 6 ( 79 . 4 ) Rmerge ( % ) †7 . 2 ( 369 . 2 ) 7 . 7 ( 292 . 5 ) 7 . 6 ( 172 . 4 ) 7 . 9 ( 36 . 2 ) Rpim ( % ) ‡1 . 5 ( 83 . 7 ) 1 . 7 ( 63 . 5 ) 1 . 3 ( 28 . 8 ) 4 . 3 ( 24 . 9 ) CC1/2 ( last resolution shell ) 0 . 448 0 . 520 0 . 934 0 . 754 I/σ ( I ) 49 . 9 ( 0 . 8 ) 46 . 8 ( 1 . 1 ) 63 . 6 ( 4 . 2 ) 15 . 9 ( 2 . 5 ) Wilson B-value ( Å2 ) 79 . 0 78 . 6 77 . 6 31 . 1 Wilson B-value , sharpened ( Å2 ) §35 . 8 35 . 4 37 . 5 24 . 2 Phase determinationAnomalous scatterersselenium , 2 out of 5 possible sitesFigure of merit ( 121 . 8–2 . 60 Å ) 0 . 25 after Selenium MAD phasing; 0 . 88 after density modificationRefinement statisticsResolution range ( Å ) 40 . 68–2 . 70 ( 2 . 79–2 . 70 ) 35 . 64–2 . 10 ( 2 . 16–2 . 10 ) No . of reflections Rwork/Rfree 11 , 593/1 , 470 ( 899/129 ) 41 , 304/2 , 000 ( 1 , 807/91 ) Data completeness ( % ) 99 . 9 ( 100 . 0 ) 86 . 3 ( 57 . 0 ) Atoms ( non-H protein/ions/ligands/solvent ) 1 , 797/5/NA/NA 5 , 411/NA/375/204 Rwork ( % ) 26 . 0 ( 34 . 0 ) 20 . 1 ( 22 . 3 ) Rfree ( % ) 29 . 8 ( 39 . 2 ) 23 . 3 ( 28 . 3 ) R . m . s . d . bond length ( Å ) 0 . 002 0 . 002 R . m . s . d . bond angle ( ° ) 0 . 42 0 . 51 Mean B-value ( Å2 ) ( protein chain ID ) ( ligands/ions/solvent ) VtrA ( A ) : 77 . 2 VtrC ( B ) : 62 . 2 . ions: 70 . 2 VtrA ( A ) : 25 . 7 VtrC ( B ) : 29 . 8 VtrA ( C ) : 34 . 9 VtrC ( D ) : 37 . 8 VtrA ( E ) : 39 . 1 VtrC ( F ) : 36 . 2 ligands: 43 . 3 ions: 86 . 2 solvent: 37 . 0 Ramachandran plot ( % ) ( favored/additional/disallowed ) # 94 . 3/5 . 2/0 . 5 96 . 5/3 . 2/0 . 3 Maximum likelihood coordinate error 0 . 36 0 . 23 Missing residues , protein ( chain ID ) VtrA ( A ) : 161– 163 . VtrC ( B ) : −13 –0 , 113–117 . VtrA ( A ) : 161–164 . VtrC ( B ) : −13–0 . VtrA ( C ) : 161–163 . VtrC ( D ) : 13–0 , 119–124 . VtrA ( E ) : 161–165 . VtrC ( F ) : 13–0 . Data for the outermost shell are given in parentheses . *Bijvoet-pairs were kept separate for data processing . †Rmerge=100∑h∑i∣ Ih , i−⟨Ih⟩∣/∑h∑i⟨Ih , i⟩ , where the outer sum ( h ) is over the unique reflections and the inner sum ( ii ) is over the set of independent observations of each unique reflection . ‡Rpim is the precision indicating R-factor , i . e Rpim=100∑h∑i[1/ ( nh−1 ) ]1/2∣Ih , i−⟨Ih⟩∣/∑h∑i⟨Ih , i⟩ , where nh is the number of observations of reflections h ( Evans , 2011 ) . §B-factor sharpening was performed in the autocorrection mode of HKL3000 ( Borek et al . , 2013 ) . #Asdefined by the validation suite MolProbity ( Chen et al . , 2010 ) . Phases for the native VtrA/VtrC heterodimer were obtained from a two-wavelength anomalous dispersion experiment using selenomethionyl-derivatized heterodimer protein with data to a dmin of 2 . 60 Å . Two selenium sites were located using the program SHELXD ( Schneider and Sheldrick , 2002 ) , and phases were refined with the program Mlphare ( Otwinowski , 1991 ) , resulting in an overall figure-of-merit of 0 . 25 for data between 121 . 8 and 2 . 60 Å . Phases were further improved by density modification in the program Parrot ( Cowtan , 2010 ) resulting in a figure-of-merit of 0 . 88 . An initial model containing 84% of all VtrA/VtrC heterodimer residues was automatically generated in the program Buccaneer ( Cowtan , 2006 ) . As the selenomethionyl-derivatized and native crystals were isomorphous , all further calculations for the native structure were performed versus the native data . Additional residues for the VtrA/VtrC heterodimer were manually modeled in the program Coot ( Emsley et al . , 2010 ) . Positional and isotropic atomic displacement parameter ( ADP ) as well as TLS ADP refinement was performed to a resolution of 2 . 70 Å using the program Phenix ( Afonine et al . , 2010 ) with a random 10% of all data set aside for an Rfree calculation . The current model contains one VtrA/VtrC heterodimer; included are residues residues 164–253 of VtrA 31–112 and 118–161 of VtrC , and one sulfate ion . The Rwork is 0 . 260 , and the Rfree is 0 . 298 . A Ramachandran plot generated with Molprobity ( Chen et al . , 2010 ) indicated that 94 . 3% of all protein residues are in the most favored regions and 0 . 5% ( one residue ) in disallowed regions . Phases for the TDC-containing heterodimer were obtained by the molecular replacement method in the program Phaser ( McCoy et al . , 2007 ) using the coordinates for the native VtrA/VtrC heterodimer . Model building and refinement were performed to a resolution of 2 . 1 Å using a similar protocol to the native structure . Four TDC molecules were located in the asymmetric unit , one each bound to the VtrC monomer of a VtrA/VtrC heterodimer and one at a lattice contact between VtrA molecules . The current model contains three VtrA/VtrC heterodimers in the asymmetric unit , four TDC molecules , seven sulfate ions and 204 water molecules . A Molprobity ( Chen et al . , 2010 ) generated Ramachandran plot indicates that 96 . 5% of all protein residues are in the most favored regions and 0 . 3% ( two residues ) in disallowed regions . Phasing and model refinement statistics for all structures are provided in Table 1 . The VtrA/VtrC periplasmic domain complex was dialyzed at 4°C overnight against the assay buffer ( 50 mM Tris pH8 , 100 mM NaCl ) . Taurodeoxycholic acid ( 400 μM ) was prepared by dissolving the dry powder ( Sigma ) with the same dialysis buffer . ITC experiments were performed at 25°C on a MicroCal iTC200 system ( Malvern ) , with reference power at 5 μcal/s and stirring rate at 750 rpm . Measurements were performed as 19 injections of 400 μM taurodeoxycholic acid ( 1 μl for the first injection and 2 μl for injections 2–19 ) into approximately 200 μl of 36 μM VtrA/VtrC . ITC data were integrated and analyzed using NITPIC 1 . 1 . 5 ( Keller et al . , 2012; Scheuermann and Brautigam , 2015 ) and ITCsy version 1a ( Brautigam et al . , 2016 ) . ITC data plots were prepared with GUSSI 1 . 1 . 0 ( Brautigam , 2015 ) . | When we eat a meal , bile salts in the intestine help the body to absorb molecules of fat . Certain bacteria that cause food poisoning in humans , such as Vibrio parahaemolyticus , can also detect bile salts and use them to sense that they have reached the intestine of a suitable host . The bacteria then produce toxins that inflame the intestine; this can result in days of diarrhea . However , it was not clear how the bacteria detect the bile salt signal and how this triggers them to produce the toxins . Li , Rivera-Cancel et al . now discover that not one , but two , genes are required for this process . The genes encode two proteins , called VtrA and VtrC , that interact to form protein ‘complex’ on the surface of the membrane that surrounds the bacterial cell . The two proteins create a barrel-like structure that can bind to bile salts and trigger the cell to produce the toxins . Future experiments will aim to understand how the binding of bile salts to this protein complex causes an increase toxin production . A future challenge is to find out how other disease-causing bacteria sense environmental cues to produce toxins . With this knowledge , researchers might be able to design new drugs that could prevent the production of toxins to relieve symptoms of food poisoning and other illnesses . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"structural",
"biology",
"and",
"molecular",
"biophysics",
"microbiology",
"and",
"infectious",
"disease"
] | 2016 | Bile salt receptor complex activates a pathogenic type III secretion system |
Phytochromes are photoreceptors regulating growth and development in plants . Using the model plant Arabidopsis , we identified a novel signalling pathway downstream of the far-red light-sensing phytochrome , phyA , that depends on the highly conserved CCR4-NOT complex . CCR4-NOT is integral to RNA metabolism in yeast and animals , but its function in plants is largely unknown . NOT9B , an Arabidopsis homologue of human CNOT9 , is a component of the CCR4-NOT complex , and acts as negative regulator of phyA-specific light signalling when bound to NOT1 , the scaffold protein of the complex . Light-activated phyA interacts with and displaces NOT9B from NOT1 , suggesting a potential mechanism for light signalling through CCR4-NOT . ARGONAUTE 1 and proteins involved in splicing associate with NOT9B and we show that NOT9B is required for specific phyA-dependent alternative splicing events . Furthermore , association with nuclear localised ARGONAUTE 1 raises the possibility that NOT9B and CCR4-NOT are involved in phyA-modulated gene expression .
As sessile organisms , plants constantly monitor their surrounding and adapt growth and development to changes in the ambient environment . Light is a key factor for plants that promotes , for instance , seed germination , early seedling development , and transition to flowering ( Paik and Huq , 2019 ) . To sense the spectral composition and measure the light intensity in the environment , plants have evolved a set of photoreceptors sensitive to wavelengths in the UV-B , blue ( B ) , red ( R ) , and far-red ( FR ) light range of the spectrum . Phytochromes are receptors for R and FR light . They are synthesised in the inactive Pr state and convert to the physiologically active Pfr state upon absorbing light ( Legris et al . , 2019 ) . Pfr can then revert to the Pr state either by absorption of light or by thermal relaxation . This toggle-switch like behaviour enables phytochromes to sense the red:far-red light ratio in the environment , which is key to distinguish sunlight from canopy shade . PhyA and phyB are the primary phytochromes in seed plants . PhyB has a dominant function as R light receptor , while responses induced by FR light depend on phyA ( Legris et al . , 2019 ) . Phytochromes are located to the cytosol in the inactive state and translocate into the nucleus upon conversion to the active Pfr state ( Klose et al . , 2015 ) . Two comparably well-characterised signalling pathways link light-activation of phytochromes to regulation of gene expression . The PHYTOCHROME INTERACTING FACTORs ( PIFs ) are bHLH transcription factors that suppress photomorphogenesis , and in parallel the CONSTITUTIVELY PHOTOMORPHOGENIC 1/SUPPRESSOR OF PHYA-105 ( COP1/SPA ) E3 ubiquitin ligase complex targets positive factors of light signalling for degradation by the 26S proteasome ( Legris et al . , 2019 ) . Light-activated phytochromes inhibit PIF and COP1/SPA action , and thereby regulate gene expression in response to light . In addition to these well-established light signalling components , also factors involved in splicing or microRNA biogenesis modulate light responses . Mutants with reduced activity of DICER-LIKE 1 ( DCL1 ) , SERRATE ( SE ) , and ARGONAUTE 1 ( AGO1 ) , or lacking functional HYPONASTIC LEAVES 1 ( HYL1 ) or HUA ENHANCER 1 ( HEN1 ) are hypersensitive to light ( Achkar et al . , 2018; Sun et al . , 2018; Cho et al . , 2014; Tsai et al . , 2014; Sorin et al . , 2005 ) and functional interactions with the classical light signalling components COP1 , HY5 , and PIF4 have been described ( Sun et al . , 2018; Cho et al . , 2014; Tsai et al . , 2014 ) . Mutants with defects in miRNA biogenesis or action often have pleiotropic phenotypes . Several components of light signalling pathways and the circadian clock are subject to alternative splicing ( Tognacca et al . , 2019; Hartmann et al . , 2016; Mancini et al . , 2016; Shikata et al . , 2014 ) . SPLICING FACTOR FOR PHYTOCHROME SIGNALLING ( SFPS ) and REDUCED RED-LIGHT RESPONSES IN CRY1 CRY2 BACKGROUND ( RRC1 ) are related to the human splicing factor SR140 and SPF45 , respectively ( Xin et al . , 2017; Shikata et al . , 2012 ) . They interact with each other , directly bind phyB and also associate with pre-mRNAs in vivo ( Xin et al . , 2019; Xin et al . , 2017 ) . RRC1 regulates thousands of splicing events . Therefore , it is not surprising that rrc1 loss-of-function mutants are lethal; mutants lacking functional SFPS are viable but show altered splicing of hundreds of transcripts ( Xin et al . , 2019 ) . Both sfps and hypomorphic rrc1 mutants have general defects in light signalling ( Xin et al . , 2019; Xin et al . , 2017; Shikata et al . , 2012 ) . Light conditions that do not activate photosynthesis , such as monochromatic FR light or short light pulses , can modulate alternative splicing through photosensory photoreceptors , including phytochromes ( Mancini et al . , 2016 ) . However , many light-regulated splicing events are not affected in photoreceptor mutants and can be induced by exogenous sugar supply , indicating that they are connected to photosynthetic activity ( Hartmann et al . , 2016; Mancini et al . , 2016 ) . Recent findings describe the CARBON CATABOLITE REPRESSION 4-NEGATIVE ON TATA-LESS ( CCR4-NOT ) complex as a major player in mRNA metabolism in eukaryotes ( Collart , 2016; Ukleja et al . , 2016; Villanyi and Collart , 2015 ) . The complex consists of several evolutionary conserved subunits that assemble with the scaffold protein NOT1 . These subunits have deadenylase ( CCR4 ASSOCIATED FACTOR 1 [CAF1] and CCR4 ) or E3 ubiquitin ligase ( NOT4 ) activity , or are transcriptionally active ( NOT2 ) ( Collart and Panasenko , 2017; Ukleja et al . , 2016; Villanyi and Collart , 2016 ) . Another highly conserved subunit of the complex is NOT9 ( also called REQUIRED FOR CELL DIFFERENTIATION 1 [RQCD1] , CAF40 , CNOT9 ) , which in animals spans the bridge between the RNA-induced silencing complex ( RISC ) and CCR4-NOT ( Chen et al . , 2014; Mathys et al . , 2014 ) . The CCR4-NOT complex controls gene expression at all steps from transcription in the nucleus to translation and mRNA degradation in the cytosol ( Collart , 2016 ) . Although extensively studied in yeast and animals , the CCR4-NOT complex was only recently shown to exist in plants ( Zhou et al . , 2020; Arae et al . , 2019 ) . The full complex regulates RNA-directed DNA methylation in Arabidopsis ( Zhou et al . , 2020 ) , and the two NOT2 homologues , NOT2A and NOT2B , promote polymerase II-dependent transcription and contribute to miRNA biogenesis ( Wang et al . , 2013 ) . Similar to the CCR4-CAF1 module in yeast and animals , Arabidopsis homologues of CCR4-CAF1 have mRNA deadenlyase activity and play a role as integrators of environmental stresses ( Suzuki et al . , 2015; Walley et al . , 2010; Liang et al . , 2009 ) . However , the function of other components of the complex and their relevance for specific signalling pathways is largely unknown . Mutants lacking the scaffold protein NOT1 are lethal in plants , animals , and yeast ( Zhou et al . , 2020; Motomura et al . , 2020; Pereira et al . , 2020; Ito et al . , 2011; DeBella et al . , 2006; Maillet et al . , 2000 ) . There are two close homologues of CNOT9/CAF40/RQCD1 in Arabidopsis , NOT9A and NOT9B , and a more distantly related protein , NOT9C ( Figure 1—figure supplement 1 ) . In this study , we show that NOT9B binds to light activated phyA and acts as repressor of early seedling development during the dark-to-light transition . We describe an unexpected role of the Arabidopsis CCR4-NOT complex in light signalling and show that NOT9B – despite being a component of this evolutionary conserved complex – has a very specific role in phyA-regulated photomorphogenesis .
In a yeast-two-hybrid ( Y2H ) screen for novel phyA-interacting proteins , we identified NOT9B , an Arabidopsis homologue of CNOT9/CAF40 from yeast and animals . As a primary validation , we performed yeast-two-hybrid ( Y2H ) growth assays ( Figure 1A ) . Irrespective of whether phyA was in the active Pfr or inactive Pr state , we found interaction with NOT9B in yeast , while FAR-RED ELONGATED HYPOCOTYL 1 ( FHY1 ) , a control protein known to specifically interact with light-activated phyA in the Y2H system ( Hiltbrunner et al . , 2005 ) , only bound phyA Pfr but not Pr . Further Y2H experiments revealed that the N-terminal 406 amino acids of phyA are sufficient for binding NOT9B ( Figure 1—figure supplement 2 ) . NOT9A , the closest homologue of NOT9B in Arabidopsis , and the more distantly related NOT9C did not interact with phyA in the Y2H system under high stringency conditions and were therefore not further tested ( Figure 1—figure supplements 1 , 3A and B ) . Interaction of NOT9B with phyA was also observed in co-immunoprecipitation ( CoIP ) assays from HEK293T cells transfected with plasmids coding for FLAG-myc-mCherry-NOT9B and phyA-GFP ( Figure 1—figure supplement 4 ) . In order to verify the interaction in planta , we performed CoIP assays using extracts from transiently transformed tobacco plants . We could co-precipitate phyA-NLS-3×HA with HA-YFP-NOT9B using anti-GFP traps ( Figure 1B ) . In addition , YFP fluorescence lifetime was significantly reduced in FRET/FLIM experiments with transiently transformed leek cells co-expressing HA-YFP-NOT9B and phyA-NLS-tagRFP compared to cells expressing HA-YFP-NOT9B and tagRFP only ( Figure 1C ) . Finally , we used stable transgenic Arabidopsis lines expressing p35S:HA-YFP-NOT9B in Col-0 background for CoIP assays to verify the phyA/NOT9B interaction and test for Pfr dependence . Seedlings were exposed to FR light for 5 hr to facilitate nuclear import of phyA and subsequently treated for 5 min with either R light ( 660 nm ) or long-wavelength FR light ( 760 nm ) to convert phyA predominantly to the Pfr or Pr state , respectively . The amount of phyA that co-purified with NOT9B was substantially higher when it was in the active Pfr state compared to the inactive Pr state ( Figure 1D ) . For unknown reasons , the interaction in the Y2H system was not affected by the Pr/Pfr state of phyA , however we have previously observed that the Pfr-dependency of phyA interactions can be less pronounced in the Y2H system than in CoIP assays ( Enderle et al . , 2017; Sheerin et al . , 2015 ) . Light-activated phyA translocates from the cytosol into the nucleus and forms subnuclear structures termed photobodies ( Klose et al . , 2015 ) . Consistent with the interaction of phyA and NOT9B , HA-YFP-NOT9B colocalised with phyA-CFP in photobodies following exposure to FR light ( Figure 1E ) , but did not form photobodies in R light ( Figure 1—figure supplement 5 ) . Furthermore , recruitment of HA-YFP-NOT9B into photobodies was abolished in phyA-211 , a mutant lacking functional phyA ( Reed et al . , 1994 ) , and fhy1-3 fhy1-like-1 ( fhl-1 ) , which is impaired in phyA nuclear transport ( Figure 1F; Hiltbrunner et al . , 2006 ) . In contrast to NOT9B , NOT9A and NOT9C did not form detectable photobodies , consistent with the observations from interaction assays that only NOT9B binds to phyA ( Figure 1—figure supplement 3C ) . Overall , we conclude that NOT9B interacts with phyA and that interaction in planta is enhanced when phyA is in the active Pfr state . To investigate the potential role of NOT9B in light signalling , we isolated two independent not9b T-DNA insertion alleles . not9b-2 carries a T-DNA insertion in the third exon of NOT9B and lacks detectable expression , whereas the not9b-1 allele has an insertion in the 3’ UTR , which results in reduced transcript levels ( Figure 2A , B ) . Both not9b mutant alleles are hypersensitive to FR light with regard to inhibition of hypocotyl growth , while hypocotyl length of not9b mutants is indistinguishable from the wildtype when grown in the dark or either R or B light ( Figure 2C , D , Figure 2—figure supplement 1A , B ) . The not9b-2 phyB-9 double mutant is as hypersensitive to FR light as the not9b-2 single mutant but lacks a response to R light ( Figure 2—figure supplement 1C , D ) . In contrast , not9b-2 phyA-211 is fully insensitive to FR light , similar to phyA-211 , but behaves like the wildtype when grown in R light ( Figure 2—figure supplement 1C , D ) . Expression of LUC-NOT9B under the control of the native promoter complements the not9b-2 mutant phenotype , confirming that lack of functional NOT9B in not9b-2 is responsible for the increased sensitivity to FR light ( Figure 2E ) . To further investigate a function of NOT9B in phyA-dependent light signalling , we examined HA-YFP-NOT9B overexpression lines ( p35S:HA-YFP-NOT9B in Col-0 , NOT9Box ) . NOT9Box lines are hyposensitive to FR light in a dose-dependent manner ( Figure 2D , F , Figure 2—figure supplement 2 ) , whereas overexpression of NOT9A and NOT9C did not affect hypocotyl growth in FR light ( Figure 2—figure supplement 3A , B ) . Similar to not9b , NOT9Box lines are phenotypically indistinguishable from the wildtype when grown in the dark or R light ( Figure 2D , Figure 2—figure supplement 1E ) . NOT9Box does not further increase hypocotyl growth of phyA-211 in FR light , and the effect of NOT9Box is not affected by the absence of functional phyB ( Figure 2—figure supplement 1F ) . Furthermore , we found that NOT9Box seedlings expressing constitutively nuclear localised phyA are equally hyposensitive to FR light as the NOT9Box parental line , in which phyA nuclear transport depends on FHY1 and FHL ( Hiltbrunner et al . , 2006 ) , and that phyA-CFP still accumulates in the nucleus in NOT9Box background ( Figure 1E , Figure 2—figure supplement 1G ) . Therefore , overexpression of NOT9B does not interfere with phyA nuclear transport but possibly affects phyA downstream signalling events in the nucleus . In addition to hypocotyl growth , we investigated anthocyanin biosynthesis and seed germination in not9b mutants and NOT9Box lines . Compared to the wildtype , NOT9Box lines accumulate reduced amounts of anthocyanin specifically in response to FR but not B light , whereas anthocyanin levels are increased in not9b mutant alleles ( Figure 2G ) . HY5 and CHS are involved in anthocyanin biosynthesis ( Shin et al . , 2007; Tepperman et al . , 2006; Zhou et al . , 2005 ) , and we found that HY5 and CHS expression is downregulated in NOT9Box lines and upregulated in the not9b-1 mutant ( Figure 2H , I ) . Under conditions where phyA promotes seed germination , not9b mutant seeds had a slightly higher germination rate than wildtype seeds ( Figure 2—figure supplement 1H ) . In pNOT9B:GUS reporter lines , the activity of the NOT9B promoter was most prominent in the cotyledons and the upper part of the hypocotyl , which is consistent with a function of NOT9B in modulation of phyA-controlled hypocotyl growth ( Figure 2—figure supplement 4A ) . NOT9B transcript levels were not regulated by light at early time points but increased after 24 hr in W , FR , or B light; in FR light , regulation was dependent on phyA ( Figure 2B , Figure 2—figure supplement 4A , B , C ) . In seedlings expressing pNOT9B:LUC-NOT9B , LUC-NOT9B levels slightly increased after 24 hr exposure to W , R , FR , or B light , while there was no difference compared to dark-grown seedlings at early time points ( Figure 2—figure supplement 4D ) . Overall , we conclude that NOT9B is a negative regulator of early seedling development specifically acting downstream of phyA . NOT9B has high sequence similarity to yeast CAF40 and human CNOT9 , which are core components of the CCR4-NOT complex ( Figure 3A , Figure 1—figure supplement 1 ) . AP-MS experiments and Y2H assays with truncated NOT1 suggested that a similar complex also exists in plants ( Zhou et al . , 2020; Arae et al . , 2019 ) . We confirmed interaction of full-length NOT1 with NOT9B , NOT2B , CAF1A , and CAF1B in Y2H assays ( Figure 3B , C ) . CNOT9 in humans binds to the DUF3819 domain ( DUF ) of NOT1 ( Chen et al . , 2014 ) . In Y2H experiments , Arabidopsis NOT9B did interact with NOT1 M , a NOT1 fragment consisting of the DUF and the MIF4G domains , but it did not interact with NOT1 MIF4G fused to GFP , which is similar in size to NOT1 M but lacks the DUF domain ( Figure 3D , E ) . NOT1 is 269 kDa in size and tagged full-length NOT1 is difficult to express in plants . In contrast , fragments corresponding to the NOT1 MIF4G and NOT1 M domain express well . Binding of NOT9B to NOT1 M and the DUF domain of NOT1 ( NOT1 DUF ) was confirmed by CoIP from transiently transformed tobacco leaves . Expression of NOT1 DUF was very low and the protein was not detectable in the input fraction , but we could co-precipitate detectable amounts of NOT1 DUF with NOT9B ( Figure 3F ) . Similar to NOT9B , NOT9A did interact with NOT1 in the Y2H system , whereas we did not observe interaction between NOT9C and NOT1 ( Figure 1—figure supplement 3B ) . NOT9B has a dual localisation , both nuclear and also forming cytoplasmic structures that are reminiscent of processing bodies ( p-bodies ) ( Figure 1E , F , Figure 1—figure supplement 5; Collart , 2016; Maldonado-Bonilla , 2014; Nissan and Parker , 2008 ) . Since CCR4-NOT components in yeast and animals are associated with p-bodies , we tested for co-localisation of NOT9B and the established p-body markers DCP1 , XRN4 , and AGO1 in plants ( Maldonado-Bonilla , 2014; Pomeranz et al . , 2010; Weber et al . , 2008; Xu et al . , 2006 ) . CFP-tagged DCP1 , XRN4 , and AGO1 co-localised with HA-YFP-NOT9B in cytosolic foci in transiently transformed tobacco leaves , and also in stable transgenic Arabidopsis lines we observed colocalisation of DCP1-CFP and HA-YFP-NOT9B in p-bodies ( Figure 4A , Figure 4—figure supplement 1A , B , C ) . GW-repeat proteins bind to CNOT9 , the human homologue of NOT9B , and possibly contribute to recruitment of the CCR4-NOT complex into p-bodies ( Hicks et al . , 2017; Mathys et al . , 2014; Behm-Ansmant et al . , 2006; Yang et al . , 2004 ) . Structural analyses have shown that TNRC6 , a GW-repeat containing protein , binds to CNOT9 through a Trp-binding pocket ( Chen et al . , 2014; Mathys et al . , 2014 ) , which is conserved in NOT9B ( Figure 1—figure supplement 1 ) . The NOT9B ΔGWB mutant ( deficient in GW-binding; NOT9B R217D A220D R256E A260L ) contains amino acid substitutions at positions that are critical for the interaction of human CNOT9 and TNRC6 . This mutant is not affected in binding NOT1 and phyA in Y2H assays , which we confirmed by CoIPs from transiently transformed tobacco leaves and stable transgenic Arabidopsis lines ( Figure 4B–F ) . However , p-body localisation of HA-YFP-NOT9B ΔGWB in Arabidopsis is strongly reduced compared to NOT9Box , suggesting that proteins associating with NOT9B through the Trp-binding pocket promote assembly of NOT9B into p-bodies ( Figure 4G , H , Figure 4—figure supplement 2 ) . The dual localisation of NOT9B raises the question whether the nuclear or the cytoplasmic fraction is causing the effect on phyA signalling . Therefore , we generated an Arabidopsis line expressing exclusively nuclear localised NOT9B by adding the strong SV40 NLS ( Figure 4I ) . Using CoIP assays , we found that phyA co-precipitates with NLS-YFP-NOT9B in a Pfr dependent manner , indicating that the nuclear localised fraction of NOT9B interacts with light-activated phyA ( Figure 4F ) . Seedlings overexpressing constitutively nuclear localised NOT9B are equally hyposensitive to FR light with regard to inhibition of hypocotyl growth as seedlings overexpressing YFP-tagged wildtype NOT9B ( Figure 4J–L ) . Therefore , we conclude that the nuclear localised fraction of NOT9B is sufficient for modulation of phyA signalling . The interaction interface of human CNOT1 and CNOT9 has been characterised using co-crystallisation experiments ( Chen et al . , 2014; Mathys et al . , 2014 ) . Introducing four mutations in CNOT9 disrupted interaction with CNOT1 ( Chen et al . , 2014 ) . The CNOT1 interaction interface is conserved in Arabidopsis NOT9B ( Figure 1—figure supplement 1 ) , and Y2H growth assays show that NOT9B containing H70A V72A A77Y V85Y substitutions is unable to bind NOT1 M ( Figure 5A ) , which we validated by CoIP from transiently transformed tobacco leaves ( Figures 4D and 5B ) . Most interestingly , this NOT9B mutant version is also impaired in phyA binding , as shown by Y2H and CoIP ( Figures 4E and 5C–E ) , therefore we refer to it as NOT9B ΔPNB ( defective in phytochrome and NOT1 binding ) . To further investigate the function of the PNB site , we generated Arabidopsis NOT9B ΔPNBox lines ( p35S:HA-YFP-NOT9B ΔPNB in Col-0 ) . Confirming the previous finding , phyA did not co-precipitate with NOT9B ΔPNB ( Figure 5E ) and , in line with impaired phyA binding , NOT9B ΔPNB did not form phyA-dependent photobodies ( Figure 5F ) . However , NOT9B ΔPNB still formed p-bodies and colocalised with DCP1 , suggesting that the ΔPNB mutation does not lead to a generally misfolded protein ( Figure 5F , G , Figure 4—figure supplement 2 ) . We then compared hypocotyl growth of NOT9Box and NOT9B ΔPNBox lines expressing the transgene at similar levels ( Figure 5H–J ) . Wildtype and NOT9B ΔPNBox seedlings grown in FR light were indistinguishable , whereas NOT9Box seedlings had considerably longer hypocotyls than the wildtype and NOT9B ΔPNBox . This suggests that the PNB site , and therefore the interaction with NOT1 and/or phyA , is important for the impact of NOT9B on phyA-mediated light signalling . NOT1 and phyA both interact with NOT9B through the PNB site ( Figure 6A ) , therefore we used Y3H assays to test if they compete for binding to NOT9B . In yeast cells expressing AD-NOT9B and phyA-BD , we co-expressed either NOT1 M , NOT1 MIF4G-GFP , or NOT1 DUF . Co-expression of either NOT1 M or NOT1 DUF strongly reduced growth on medium selective for interaction of NOT9B and phyA , whereas NOT1 MIF4G-GFP , which lacks the NOT9B-binding site , had no effect ( Figure 6B ) . None of the NOT1 fragments competed with FHY1 for binding to phyA , suggesting that co-expression of NOT1 does not generally interfere with interaction of phyA and phyA-binding proteins ( Figure 6—figure supplement 1A ) . To confirm that expression of AD-NOT9B and phyA-BD is not reduced by co-expression of NOT1 fragments , we used HA-tagged AD-NOT9B and NOT1 M , and LUC-tagged phyA-BD for Y3H competition analysis . Co-expression of HA-NOT1 M clearly reduced growth of cells expressing the Y2H pair HA-AD-NOT9B/phyA-LUC-BD but had only a minor effect on HA-AD-NOT9B and phyA-LUC-BD protein levels ( Figure 6—figure supplement 1B , C ) . To test for competition in planta , we took advantage of the findings that recruitment of NOT9B into photobodies depends on interaction with phyA ( Figures 1E , F and 5F ) and that NOT1 does not form photobodies nor binds phyA ( Figure 6—figure supplement 2 ) . Thus , if there is competition for binding to NOT9B , we expect that high expression of NOT1 would hinder recruitment of NOT9B into photobodies ( Figure 6—figure supplement 3 ) . In tobacco leaves transiently expressing NLS-YFP-NOT9B and phyA-NLS-CFP , co-expression of a NOT1 fragment impaired in binding NOT9B ( FLAG-myc-mCherry-NLS-NOT1 MIF4G-mCherry ) had no effect on recruitment of NOT9B into photobodies ( Figure 6C ) . In contrast , co-expression of FLAG-myc-mCherry-NLS-NOT1 M , which includes the NOT9B binding site of NOT1 , almost fully abolished assembly of NOT9B into photobodies whilst not affecting phyA photobody formation . This finding is consistent with competition of phyA and NOT1 for binding NOT9B in planta . To further investigate this notion , we generated stable transgenic Arabidopsis lines expressing p35S:FLAG-myc-mCherry-NLS-NOT1 M either in Col-0 or NOT9Box backgrounds . FLAG-myc-mCherry-NLS-NOT1 M co-purified with HA-YFP-NOT9B in CoIPs ( Figure 6D ) , which is in agreement with data shown in Figures 3E , F and 4C , D . Interestingly , the amount of phyA Pfr co-precipitating with HA-YFP-NOT9B appeared to be lower in CoIPs from seedlings co-expressing FLAG-myc-mCherry-NLS-NOT1 M than from seedlings not expressing NOT1 M ( Figure 6D ) . Quantification of three independent CoIP experiments analysed on the same membrane showed that the amount of phyA bound to NOT9B is significantly reduced when NOT1 M is coexpressed ( Figure 6E , F ) . This finding supports the idea that binding of phyA and NOT1 to NOT9B is mutually exclusive and they compete for binding NOT9B in planta . Therefore , increasing the levels of one of the proteins binding to the PNB site of NOT9B would decrease binding of the other protein interacting with this site . Since phyA levels in the nucleus are likely much higher and more dynamic than levels of NOT1 , the impact of phyA on NOT1/NOT9B would be of greater importance in planta than the effect of NOT1 on the NOT9B/phyA interaction . Such competition-based regulation of NOT1/NOT9B interaction by phyA is favoured by phyA levels that are higher than NOT9B levels , and therefore we compared the luciferase activity in a phyA-211 pPHYA:PHYA-LUC complementation line and not9b-2 complemented with pNOT9B:LUC-NOT9B . Quantification of the luciferase signal showed that levels of phyA are several orders of magnitude higher than the levels of NOT9B ( Figure 6—figure supplement 4 ) . Consequently , based on the phenotype of not9b mutants and NOT9Box lines , we propose that CCR4-NOT complexes containing NOT9B ( CCR4-NOTNOT9B ) repress phyA signalling . In seedlings exposed to FR light , phyA is transported into the nucleus and competition for binding NOT9B could then displace NOT9B from CCR4-NOTNOT9B ( Figure 6A , Figure 6—figure supplement 3 ) , thereby relieving repression of light signalling . A potential explanation for the NOT9Box phenotype is that phyA levels are not high enough to fully prevent NOT9B from binding NOT1 when NOT9B is overexpressed . If so , blocking the PNB site in NOT9B by overexpression of NOT1 M should alleviate the NOT9Box phenotype . In line with this reasoning , the negative effect of NOT9Box on both inhibition of hypocotyl growth in FR light and expression of FR light-induced genes such as EARLY LIGHT INUDCIBLE PROTEIN 1 and 2 ( EPLI1/2 ) and CHS ( Harari-Steinberg et al . , 2001; Zhou et al . , 2005 ) is partially suppressed by co-expression of FLAG-myc-mCherry-NLS-NOT1 M without affecting NOT9B protein levels ( Figure 6G–I , Figure 6—figure supplement 5 ) . Recent AP-MS approaches to characterise the CCR4-NOT complex in plants ( Zhou et al . , 2020; Arae et al . , 2019 ) and data shown in Figure 3 support the notion that a complex similar to the CCR4-NOT complex from yeast and animals also exists in plants . We propose that phyA prevents the assembly of NOT9B into the CCR4-NOT complex , which offers a potential explanation for regulation of CCR4-NOT by FR light . A wide range of functions in control of RNA stability and metabolism have been described for yeast and animal CCR4-NOT but it is still unknown how CCR4-NOT modulates phyA mediated light responses in plants . Therefore , we investigated the NOT9B interactome in the NOT9Box line . Silver staining of GFP-trap purified proteins revealed several proteins co-precipitating with HA-YFP-NOT9B but not with YFP-HA ( Figure 7—figure supplement 1A ) . By MS-MS we identified 450 proteins that are only present in the HA-YFP-NOT9B eluate fraction ( Figure 7—figure supplement 1B , Supplementary file 1 ) . This includes the known NOT9B interactors NOT1 and phyA , and most components of the CCR4-NOT complex . We did not find NOT9A , the closest homologue of NOT9B , and the more distantly related NOT9C . Interestingly , AGO1 and proteins with a function in miRNA biogenesis and/or splicing were among NOT9B-associated proteins . Highly enriched GO terms among NOT9B-associated proteins were rRNA binding , mRNA binding , RNA binding , and translation elongation factor activity ( Figure 7—figure supplement 1C ) , supporting a conserved role of the CCR4-NOT complex in mRNA metabolism in eukaryotes , including plants . In the following , we investigated a potential function of AGO1 and alternative splicing in NOT9B-mediated modulation of phyA signalling . We did not find any evidence for direct interaction of NOT9B and AGO1 in Y2H assays ( Figure 7—figure supplement 2 ) but could co-precipitate endogenous AGO1 with HA-YFP-NOT9B , and vice versa found HA-YFP-NOT9B associated with endogenous AGO1 in CoIPs with αAGO1 antibodies ( Figure 7A , B ) . We also observed colocalisation of HA-YFP-NOT9B and AGO1-CFP in p-bodies of transiently transformed tobacco leaves ( Figure 4—figure supplement 1A ) . Overall , AP-MS and CoIP data show that AGO1 and NOT9B are in complex but possibly do not directly interact . AGO1 is most known for its function in miRNA-mediated cleavage of target mRNAs in the cytosol , but recent work has shown that Arabidopsis AGO1 also has a function independent of miRNAs , and binds to chromatin and regulates gene expression in the nucleus ( Bajczyk et al . , 2019; Bologna et al . , 2018; Liu et al . , 2018 ) . In CoIP assays , similar amounts of NOT9B co-purified with AGO1 from lines overexpressing wildtype and constitutively nuclear localised NOT9B , suggesting that NOT9B associates with AGO1 in the nucleus ( Figure 7B ) . The function of AGO1 in regulation of gene expression in the nucleus is independent of miRNAs but requires small RNAs that define AGO1 target sites in the genome ( Liu et al . , 2018 ) . Several genes involved in light signalling or light-regulated processes such as hypocotyl growth , anthocyanin biosynthesis , establishment of photosynthetic capacity , photoperiodic flowering , or circadian rhythms are among AGO1-bound genes , suggesting that the nuclear localised fraction of AGO1 might be involved in modulation of light responses ( Liu et al . , 2018 ) . Sorin et al . , 2005 have shown that hypocotyl growth at intermediate light intensities is reduced in weak , viable ago1 mutants ( e . g . ago1-33 , ago1-34 , and ago1-35 ) compared to the wildtype . To further investigate a potential function of AGO1 in light signalling , we measured detailed fluence rate response curves in FR light for the weak ago1-27 allele , and quantified anthocyanin levels and cotyledon opening . The ago1-27 mutant has only slightly shorter hypocotyls than the wildtype in the dark but is hypersensitive to FR light over a wide range of fluence rates ( Figure 7C , Figure 7—figure supplement 3 ) . In addition anthocyanin levels are increased in ago1-27 exposed to FR light , whereas there is no significant difference in the dark and in B light ( Figure 7D ) . Enhanced accumulation of anthocyanin in FR light is therefore wavelength-specific and not a general light or stress phenotype . Cotyledon unfolding is typically enhanced by light but we found that this response is partially impaired in the ago1-27 and the not9b-1 mutants ( Figure 7E , F ) . Thus , while generally being hypersensitive to FR light , ago1-27 and not9b-1 are hyposensitive regarding cotyledon unfolding , which distinguishes them from other light signalling mutants and may indicate that NOT9B and AGO1 are functionally linked . Interestingly , the NOT9Box line is also impaired in proper cotyledon unfolding in FR light ( Figure 7E , F ) , for which we present a potential explanation in the discussion . Since the NOT9B interactome includes proteins with a function in splicing ( Figure 7—figure supplement 1 ) , we searched the literature for reports on FR light regulated splicing events . MYBD is a MYB-related transcription factor that promotes anthocyanin biosynthesis downstream of HY5 ( Nguyen et al . , 2015 ) . MYBD is regulated by alternative splicing; splice variant MYBD . 1 encodes a functional protein , while MYBD . 2 has a premature stop codon due to intron retention ( Figure 8A; Hartmann et al . , 2016 ) . The ratio of MYBD . 2/ . 1 is strongly reduced upon R or FR light treatment , meaning that light shifts the MYBD . 2/ . 1 ratio toward the splice variant that codes for functional MYBD and thereby promotes anthocyanin biosynthesis ( Hartmann et al . , 2016; Nguyen et al . , 2015 ) . We found that phyA is required for this response to FR light ( Figure 8B ) and that the MYBD . 2/ . 1 ratio in dark-grown not9b-2 seedlings is similar to wildtype seedlings treated with FR light ( Figure 8C ) . Thus , NOT9B is essential to suppress the shift of the MYBD . 2/ . 1 ratio toward the physiologically active splice variant in etiolated seedlings and FR light perceived by phyA removes this negative regulation . Seedlings lacking functional NOT9B do not have an obvious constitutively photomorphogenic ( cop ) phenotype ( Figures 2D and 7E , Figure 2—figure supplement 3A ) , but the splicing pattern of MYBD in etiolated not9-2 resembles the pattern in wildtype seedlings exposed to light ( Figure 8C ) . Thus , as an additional molecular read out of light signalling , we quantified transcript levels of the early light response genes EARLY PHYTOCHROME RESPONSIVE 1 ( EPR1 ) /REVEILLE 7 ( RVE7 ) , ELIP1 , and ELIP2 ( Kuno , 2003; Harari-Steinberg et al . , 2001 ) . Expression of EPR1 , ELIP1 , and ELIP2 was increased in dark-grown not9b-2 seedlings compared to the wildtype , supporting the concept that not9b mutant seedlings have a partial cop phenotype at the molecular level ( Figure 9A ) . When exposed to FR light for 1 hr , EPR1 , ELIP1 , and ELIP2 transcript levels are similar in not9b-2 and wildtype seedlings . Expression of EPR1 and ELIP1 was also increased in dark-grown NOT9Box seedlings , although not to the same extent as in not9b-2 .
Seedlings with altered NOT9B levels have a FR light-specific phenotype and do not show any obvious developmental defects despite the evolutionary conservation and general function of the CCR4-NOT complex ( Figure 2 , Figure 2—figure supplement 1 ) . This is in stark contrast to many other mutants with defects in complexes or pathways of general function , such as splicing and miRNA biogenesis . Several of these mutants show altered responses to light but they are not wavelength-specific , and often these mutants have highly pleiotropic phenotypes with loss-of-function alleles being lethal ( Xin et al . , 2019; Xin et al . , 2017; Shikata et al . , 2012; Laubinger et al . , 2008; Sorin et al . , 2005 ) . While CAF40/CNOT9 are single copy genes in humans and yeast , plants generally contain two or more genes coding for proteins with similarity to CAF40/CNOT9 , including NOT9A , NOT9B , and NOT9C in Arabidopsis ( Zhou et al . , 2020; Arae et al . , 2019 ) . Residues required for recruitment into p-bodies are conserved in all NOT9 proteins , including the more distantly related NOT9C , and consistently , we found that all Arabidopsis NOT9 proteins form p-bodies ( Figure 1—figure supplements 1 and 3C ) . The region corresponding to the PNB site of NOT9B is more variable ( Figure 1—figure supplement 1 ) . NOT9C is most divergent and did not interact with phyA and NOT1 nor did it co-purify with CCR4-NOT in previous AP-MS approaches ( Figure 1—figure supplement 3A , B; Zhou et al . , 2020; Arae et al . , 2019 ) . In contrast , NOT9A and NOT9B both bind NOT1 but we only observed interaction of phyA and NOT9B ( Figure 1 , Figure 1—figure supplement 3A , B ) . We hypothesise that residues in NOT9A and NOT9B that correspond to the residues mutated in NOT9B ΔPNB possibly contribute to specify the NOT9A/B interaction profile . Y2H and CoIP from HEK293T cells show that no additional plant-specific proteins are required for NOT9B/phyA complex formation but we cannot formally rule out the possibility that components of the CCR4-NOT complex conserved in animals , yeast , and plants bridge between NOT9B and phyA in these assays . Yet , if phyA would bind indirectly to NOT9B through other components of the CCR4-NOT complex , we would also expect complex formation for phyA and both NOT9A and NOT1 in the Y2H assay and recruitment of NOT9A and NOT1 into phyA-dependent photobodies . However , we did not observe either ( Figure 1—figure supplement 3A , B , C , Figure 6—figure supplement 2 ) and therefore favour a model in which phyA directly binds NOT9B . NOT9A and NOT9B co-precipitated with NOT1 , NOT3 , and CCR4b in previous AP-MS approaches ( Zhou et al . , 2020; Arae et al . , 2019 ) but we did not find NOT9A among NOT9B associated proteins in our AP-MS experiment ( Figure 7—figure supplement 1 ) . Thus , two pools of CCR4-NOT complexes , CCR4-NOTNOT9A and CCR4-NOTNOT9B , may exist that contain either NOT9A or NOT9B . Whether they differ regarding functionality is still unknown , but we expect that CCR4-NOTNOT9A and CCR4-NOTNOT9B are different in terms of regulation by light , since phyA binds NOT9B but not NOT9A . PhyA and NOT1 bind to the PNB site of NOT9B and several approaches provide evidence that binding to NOT9B is mutually exclusive ( Figure 6 ) . Increasing the levels of one PNB binding protein will therefore reduce interaction of NOT9B with the other PNB binding protein . PhyA is several orders of magnitude more abundant than NOT9B ( Figure 6—figure supplement 4A ) , and it is also expected to be more abundant and more dynamic than NOT1 . Under such conditions , phyA likely has a greater impact on the regulation of the NOT9B-NOT1 interaction , than NOT1 on the phyA-NOT9B interaction . Nuclear localised NOT9B is sufficient for modulation of phyA mediated light signalling , which therefore does not depend on p-bodies ( Figure 4I–L ) . We propose that upon transport into the nucleus , phyA binds NOT9B and thereby displaces NOT9B from the CCR4-NOTNOT9B complex ( Figure 9B ) . Based on physiological and gene expression data , we propose a NOT1/phyA competition model and assume that CCR4-NOTNOT9B is the active unit that represses light signalling while the complex without NOT9B does not affect light responses . Both the light-dependence of phyA nuclear transport and the preferential interaction of NOT9B with phyA in the active Pfr state would contribute to disassembly of CCR4-NOTNOT9B specifically in FR light , which could explain the phyA- and FR light-specific phenotype of not9b mutant and NOT9Box seedlings . In an alternative phyA repression model , NOT9B could bind phyA and repress phyA activity . In contrast to the NOT1/phyA competition model , the effect of NOT9B on light signalling would be independent of the CCR4-NOT complex in the phyA repression model . However , we favour the NOT1/phyA competition model over the phyA repression model for three reasons . First of all , levels of endogenous NOT9B are much lower than levels of phyA ( Figure 6—figure supplement 4A ) . Under such conditions , efficient competition between NOT1 and phyA for binding NOT9B is possible for which we have experimental evidence ( Figure 6 ) . In contrast , it appears unlikely that NOT9B levels , that are orders of magnitude lower than phyA levels , can efficiently repress phyA activity . Further insight could be gained by affinity measurements using heterologously expressed proteins , which also could support the notion that the phyA-NOT9B interaction is direct . Secondly , dark-grown not9b mutant seedlings show a partial constitutively photomorphogenic ( cop ) phenotype and are unable to fully suppress photomorphogenesis in the dark ( Figures 8 and 9A ) . This finding is consistent with the NOT1/phyA competition model , whereas dark-grown not9b seedling should be fully etiolated in the phyA repression model . Lastly , if the phyA repression model were true , we would expect all phyA-regulated responses to be increased in the not9b mutant , which is not the case ( i . e . cotyledon unfolding in FR light is reduced in not9b ) . In general , not9b mutant and NOT9Box seedlings have opposite phenotypes . However , we observed that both not9b and NOT9Box seedlings are hyposensitive to light with regard to cotyledon opening and both show increased expression of a subset of light-induced genes in the dark ( Figures 7E , F and 9A ) . A possible explanation for this seemingly counterintuitive observation is that NOT9B is part of a multimeric complex and could act as a linker protein between CCR4-NOTNOT9B and other proteins bound to NOT9B . In contrast to a dimeric complex , where increasing the level of one component does not have a negative effect on complex formation , there is an optimal concentration of the bridge protein at which a ternary complex is most abundant ( Figure 9—figure supplement 1A , B; Douglass et al . , 2013 ) . For a hypothetical NOT1-NOT9B-protein X ternary complex , in which NOT9B is the bridge protein , increasing NOT9B levels above the optimal concentration would predominantly result in formation of NOT1-NOT9B and NOT9B-protein X dimeric complexes at the expense of the NOT1-NOT9B-protein X ternary complex ( Figure 9—figure supplement 1B–E ) . Thus , if only the ternary complex is functional , not9b and NOT9Box seedlings can have the same phenotype . Different protein Xs may exist ( protein XA and XB in Figure 9—figure supplement 1F ) and associate with NOT9B to regulate light responses in a context-specific manner . The optimal concentration of NOT9B leading to highest levels of NOT1-NOT9B-protein X complexes depends on the concentration of protein XA and XB and their affinity for NOT9B . Therefore , overexpression of NOT9B may increase NOT9B levels to concentrations that favour the formation of ternary complexes with one protein X ( protein XB in Figure 9—figure supplement 1F ) , while the same NOT9B levels may exceed this optimal concentration for the assembly of ternary complexes with another protein X ( protein XA in Figure 9—figure supplement 1F ) . As a consequence , for some responses , NOT9Box lines and not9b mutants may show the same phenotype ( responses depending on NOT1-NOT9B-protein XA in Figure 9—figure supplement 1F ) , while the phenotype of the mutant and the overexpression line may be opposite for other responses ( responses depending on NOT1-NOT9B-protein XB , Figure 9—figure supplement 1F ) . However , this hypothesis is difficult to test at the moment for three reasons: firstly , the concentrations of NOT1 , NOT9B , and protein Xs in planta are unknown and they may vary depending on tissue and developmental stage , secondly , binding affinities for NOT1/NOT9B and NOT9B/protein Xs are unknown , and lastly , protein Xs themselves are unknown . A potential approach to identify candidates for protein X is to use a line expressing exclusively nuclear localised NOT9B ΔPNB for AP-MS . In such an approach , we would expect to find nuclear localised proteins that associate with NOT9B through binding sites other than the PNB site . Potential candidates for X might directly or indirectly interact with NOT9B through the GWB motif but binding to yet unknown sites is also possible . We found that AGO1 associates , possibly indirectly , with NOT9B and therefore AGO1 could be part of a protein complex that comprises a protein X . The GWB site in NOT9B corresponds to the binding site of TNRC6/GW182 proteins in the human NOT9B homologue CNOT9 ( Chen et al . , 2014; Mathys et al . , 2014 ) , and AGO proteins bind to GW-repeats in TNRC6/GW182 ( Lazzaretti et al . , 2009 ) . Plants lack sequence homologues of TNRC6/GW182 but they contain proteins with GW-repeats ( Pfaff et al . , 2013; Karlowski et al . , 2010 ) . Therefore , we speculate that a still unknown GW-repeat containing protein could bind NOT9B at the GWB site and link it to AGO1 . A potential function of AGO1 in a CCR4-NOTNOT9B-AGO1 complex could be to recruit CCR4-NOTNOT9B to specific sites in the genome . Target sites of such complexes could be defined by small RNAs bound by AGO1 ( Liu et al . , 2018 ) , while activities linked to CCR4-NOT could affect the transcription at the target site . Potential small RNAs in complex with CCR4-NOTNOT9B-AGO1 could be identified by immunoprecipitation on an NLS-YFP-NOT9B expressing line followed by smallRNAseq . Interestingly , we found that phyA-dependent regulation of splice site selection is disturbed in the not9b mutant ( Figure 8 ) and several proteins involved in splicing co-purified with NOT9B in the AP-MS approach ( Figure 7—figure supplement 1 ) . Ago2 , an AGO protein in Drosophila was shown to regulate alternative splicing patterns ( Lee and Rio , 2015; Taliaferro et al . , 2013 ) . We speculate that nuclear localised AGO1 in complex with CCR4-NOTNOT9B could play a role in splice site selection in Arabidopsis and that NOT9B through its specific interaction with phyA could put this event under FR light control . In this study , we identified NOT9B as novel phyA interacting protein providing a potential link between light signalling and the evolutionary conserved CCR4-NOT complex . We show that NOT9B is involved in light-dependent development of plants , and suggest a competition-based mode of action that depends on displacement of NOT9B from the CCR4-NOT complex by light-activated phyA . These findings can provide a basis for understanding of how light signalling in plants could feed into central processes common to all eukaryotes , such as mRNA metabolism . Future research will need to reveal the molecular mechanisms that connect the evolutionary ancient CCR4-NOT complex to gene expression in plants .
All vectors , oligonucleotides , and gBlocks used in this study have either been described previously or are listed in Supplementary file 2 . All plasmid constructs have been verified by sequencing and analytical digest . One gram of indicated plant material ( infiltratred tobacco leaves or Arabidopsis seedlings ) was collected under indicated light conditions and ground for 5 min in liquid N2 . Working in safe green light conditions at 4°C , protein was extracted using 4 ml IP buffer ( pH 7 . 8; 134 mM Na2HPO4 , 1 . 56 mM NaH2PO4 , 450 mM NaCl , 1 mM KCl , 1 mM EDTA , 1% PEG 4000 , 0 . 5% Triton X-100 , 1 mM Na3VO4 , 2 mM Na4P2O7 , 10 mM NaF ) , one vial Protease Inhibitor Cocktail ( Sigma-Aldrich , Cat-No: I3911 ) per litre , 1× Protease cOmplete Inhibitor Cocktail ( Sigma-Aldrich , Cat-No: 04693159001 ) . Buffer was supplemented with 5 mM DTT if N . benthamiana was used as an expression system . Lysate was further homogenised using Potter-Elvehjem homogenisers . Lysate was cleared by centrifugation , the soluble fraction was separated , and 50 μl Anti-GFP MicroBeads ( Miltenyi , Cat-No: 130-091-125 ) or Protein A beads ( Miltenyi , Cat-No: 130-071-001 ) and indicated antibody were added and incubated under gentle shaking for 2 hr at 4°C . Columns were equilibrated and washed post pulldown according to manufacturers instructions . Elution was performed using preheated ( 95°C ) Elution Buffer ( 100 mM Tris/HCl , pH 6 . 8 , 4% SDS , 20% glycerol , 0 . 05% Bromphenol blue ) . Elution fractions were analysed by SDS-PAGE and immunoblotting . HEK293T cells were used as platform for heterologous expression of proteins for co-immunoprecipitation . Culture conditions and IP conditions have been described previously ( Enderle et al . , 2017 ) . The HEK293T cell line was provided by the Core Facility Signalling Factory , University of Freiburg; the cell line was tested negatively for mycoplasma contamination and its identity was confirmed by STR analysis . All plant lines used in this study and primers used for genotyping are listed in Supplementary file 2 . Seeds were surface sterilised by incubation in 1 ml of 70% ethanol ( v/v ) for 10 min under constant shaking followed by washing with 1 ml 100% ethanol for 10 min under constant shaking . Seeds were dried on filter paper under sterile condition . If not indicated otherwise , Arabidopsis thaliana seedlings were grown on four layers of filter paper ( Macherey-Nagel; Cat-No: MN 615 ) wetted with 4 . 5 ml sterile H2O . Seeds were stratified for 2–3 days in darkness at 4°C , germination was induced by incubation in white light ( 70 μmol m−2 s−1 ) for 8 hr at 22°C . Plates were afterwards kept for additional 16 hr in D at 22°C , prior to transfer to respective light conditions . For propagation and breeding purposes , plants were grown under long day ( LD ) conditions ( 16 hr W , 70 μmol m−2 s−1 , 22°C/8 hr D , 18°C ) on standard soil ( Einheitserde , Cat-No: 540203 ) . N . benthamiana plants were grown under LD conditions ( 16 hr W , 70 μmol m−2 s−1 , 22°C/8 hr D , 18°C ) on standard soil ( Einheitserde , Cat-No: 540203 ) . Plants were grown for 4 days under the respective light conditions . 20 individual seedlings of each genotype were measured using the ImageJ software . Either the relative hypocotyl length ( mean hypocotyl length in light divided by mean hypocotyl length in D ) or absolute hypocotyl length is shown . Transient expression of fluorescently tagged ( YFP and mCherry ) proteins in leek epidermal cells was performed by particle bombardment using a particle gun , PDS1000/-He Biolistic Particle Delivery system ( Bio-Rad , USA ) . 400 ng of each plasmid were mixed and added to 5 μl gold particles ( 1 . 0 μm diameter #1652263 , Bio-Rad ) , 10 μl 2 . 5 M CaCl2 , and 4 μl 0 . 1 M Spermidine ( S2626-1G , Sigma-Aldrich ) in microcentrifuge tubes . The mixtures were then incubated at room temperature for 15 min with occasional vortexing , pulse-centrifuged ( 10 , 000 rpm for 10 s ) , washed initially with 100 μl 70% ethanol , followed by 50 μl 100% ethanol . The supernatants between the washes were removed by pipetting and the pellets were resuspended finally in 12 μl 100% ethanol . Macrocarriers ( #165–2258 , Bio-Rad ) were assembled on the metal rings and the resuspended gold particles carrying the plasmids ( microcarriers ) were applied onto the macrocarriers . The assemblies were incubated at room temperature for 5–10 min to evaporate the ethanol , and to make the microcarriers stick to the macrocarriers . Leek pieces were prepared by separating the 3rd , 4th , and 5th layers from outside , cut into 3 × 1 inch pieces and placed in petri dishes moistened with wet paper towels . The metal rings with the macrocarriers were loaded into the particle gun and high-pressure helium gas was used to bombard the microcarriers through 900 psi rupture disks ( #165–2257 , Bio-Rad ) in vacuum ( 25 inHg ) through stopping screens onto the leek pieces . After 14–16 hr of incubation in darkness , the epidermal layers of bombarded leek pieces were subjected to confocal microscopy followed by FRET-FLIM ( Förster Resonance Energy Transfer by Fluorescence Lifetime Imaging ) . The leek cells were imaged with a Confocal Laser Scanning Microscope ( Leica CLSM SP8 , Leica Microsystems ) . For FRET-FLIM analyses , the lifetime of the donor fluorophore YFP was compared between the leek cells that express YFP-tagged protein alone or co-expressed with mCherry-tagged proteins . YFP was excited with a pulsed picosecond laser at 470 nm ( detection gate of 521–549 nm , and a frequency of 40 MHz ) and the emission signals were detected by HyD SMD hybrid detector ( Leica Microsystems ) . In the same cell the expression of mCherry- ( excited by a 561 nm PMT laser with detection gating of 620–730 nm ) tagged proteins ( if co-expressed ) was also detected . The fluorescence lifetime of YFP was determined by Time Correlated Single Photon Counting ( TCSPC ) using FLIM hardware from PicoQuant ( PicoQuant GmbH , Berlin , Germany ) . The detections were limited to 1000 photons/pixel . In cells , in which the tagged proteins formed nuclear bodies ( NBs ) , appropriate regions of interest ( ROIs ) were chosen for the calculation of time correlated histograms . The ROIs were expanded to nuclear peripheries in the cells that did not form any NBs . The time correlated histograms were then deconvoluted and fitted by n-exponential tailfit algorithm into mono- ( for the cells expressing YFP-tagged proteins only ) or bi-exponential decay ( for the cells expressing YFP and mCherry-tagged proteins ) curves using the FLIM package of SymphoTime 64 software ( PicoQuant GmbH , Berlin , Germany ) . Only lifetime values from cells with χ2<1 were considered for analysis . The experiments were conducted on different days with independent bombardments and the data from several experiments were combined for the final calculations . Seeds were sown on ½ MS/2% agar plates and incubated in D for 2 hr . Then , a 5 min FR pulse ( 740 nm , 40 μmol m−2 s−1 ) was applied followed by 48 hr incubation in D . In the following , the plates were either incubated for 3 min in FR ( 740 nm , 40 μmol m−2 s−1 ) or for 30 s in R ( 660 nm , 10 μmol m−2 s−1 ) followed by incubation in D for another 5 days . One set of plates was kept in W ( 70 μmol m−2 s−1 PAR ) to estimate seed viability , one set was not illuminated and kept in D . After 5 days seedlings were counted and germination frequency calculated . Seedlings were grown for 4 days under indicated conditions . Total RNA was extracted using the Plant concert reagent ( Invitrogen , Carlsbad , California ) followed by RNA clean up using the RNAII plant RNA kit ( Bioline , London , UK ) . RNA quantity and quality was assessed by UV-Vis spectroscopy . Total RNA ( 1 μg ) was reverse transcribed into cDNA using the High Capacity Reverse transcription kit ( Thermo Fisher Scientific , Waltham , Massachusetts ) . qRT-PCR was performed using target-specific primer and/or probe sets as described in Supplementary file 3 . Primers were designed exon-exon spanning , if possible . Primer and probe specificity was analysed using Primer Blast ( Ye et al . , 2012 ) . qPCR reactions were set up as indicated in the 2× qPCRBIO SyGreen Mix Separate-ROX manual ( Nippon Genetics Europe GmbH , Düren , Germany , Cat . : PB20 . 14–51 ) . 5 ng of reverse transcribed RNA was used as template , 400 nM of each primer ( and probe if applicable ) was used for transcript specific amplification . Reactions were set up in 384 FastGene plates ( Nippon Genetics Europe GmbH , Düren , Germany , Cat . : FG-300150 ) and measured on a CFX384 Touch real-Time PCR Detection System ( Bio-Rad Laboratories , Inc , Hercules , California ) . The cycling conditions were set to 2 min at 95°C , followed by 40 cycles of 5 s denaturation at 95°C and 30 s annealing/extension at 60°C . A melting curve analysis was performed for each qPCR run to identify secondary products . For each primer pair a dilution series was assayed to analyse amplification efficiency , which was used for calibration of the quantification of the starting quantity ( Livak , 1997 ) . Each biological replicate was measured in technical triplicates . For light treatment , plants were kept in FR ( 740 nm , 40 μmol m−2 s−1 ) , B ( 30 μmol m−2 s−1 PAR ) , R ( 660 nm , 20 μmol m−2 s−1 ) , or W light ( 70 μmol m−2 s−1 PAR ) for the indicated time . The initial Yeast-2-Hybrid screen , in which NOT9B was identified as phyA-interacting protein , was done as previously described ( Sheerin et al . , 2015 ) . For Yeast-2-Hybrid experiments performed in this study , yeast strain AH109 ( Takara Clontech , Kyoto , Japan ) was transformed with plasmids coding for AD and either BD or BD-Aux using the Frozen-EZ yeast Transformation Kit ( Zymo Research , Freiburg , Germany Cat-No: T2001 ) . Transformants were isolated on Leucin-Tryptophane dropout medium ( CSM LT- ) , resuspended in sterile ddH2O and diluted to an OD600 of 0 . 1 . 5 μl of yeast suspension was spotted onto Leucin-Tryptophane-Histidine dropout medium ( CSM LTH- ) and incubated for 3–5 days at 30°C . For Yeast-2-Hybrid experiments involving phytochromes , phycocyanobilin ( Frontier Scientific , Logan , Utah , Cat-No: FSIP14137 ) was supplemented to the media as indicated in the figure legends and incubation temperature was reduced to 26°C . Light treatment was performed using either 660 nm ( 1 μmol m−2 s−1 , R ) or 740 nm ( 10 μmol m−2 s−1 , FR ) LED panels . Filter lift assays and quantification of β-Gal activity using ONPG were performed as described ( Sheerin et al . , 2015; Clontech , 2009 ) . Total protein extraction for immunoblotting was performed as described ( Kushnirov , 2000 ) . For microscopic live imaging of A . thaliana seedlings and N . benthamiana mesophyll cells , plant material was transferred under green light conditions to microscope slides and mounted in sterile H2O . Microscopic images were acquired using a Zeiss Axioplan 2io mot ( Carl Zeiss , Göttingen , Germany ) equipped with a Photometrics CoolSnap-HQ 12-bit monochrome CCD camera ( Roper Scientific , Tucson , AZ ) , external filter wheels ( LUDL , Hawthorne , NY ) , and filter sets for mCherry ( F36-508 , excitation 562 nm , emission 641 nm; AHF Analysentechnik , Tübingen , Germany ) , YFP ( F31-028 , excitation 500 nm , emission 515 nm; AHF Analysentechnik , Tübingen , Germany ) , and CFP ( F31-044 , excitation 436 nm , emission 455 nm; AHF Analysentechnik , Tübingen , Germany ) . For FR light treatment , plants were kept for 4–6 hr in FR light ( 740 nm , 40 μmol m−2 s−1 ) . Seedlings were grown for four days under the light conditions indicated in the figure legends . Total protein was extracted by grinding 100 mg of seedlings in liquid N2 followed by the addition of 250 μl of extraction buffer ( 65 mM Tris/HCl pH 7 . 3 , 4 M Urea , 3% SDS , 10% Glycerol , 0 . 05% Bromphenol blue , 20 mM DTT , 1× Protease Inhibitor Cocktail ( Sigma-Aldrich , Cat-No: I3911 ) ) preheated to 95°C . Soluble protein was separated by centrifugation ( 15 min , 20 , 000× g ) and protein content measured using the Amido black method ( Popov et al . , 1975 ) . Equal amounts of proteins were separated by 10% SDS-PAGE and transferred to PVDF membrane . Membranes were blocked with 5% skim milk powder in PBS-T ( 137 mM NaCl , 2 . 7 mM KCl , 10 mM Na2HPO4 , 1 . 8 mM KH2PO4 , pH 7 . 4 , 0 . 5% Tween-20 ) . Membranes were probed with antibodies ( see Supplementary file 3 for details and dilutions ) . Immunodetection was performed using CDP-Star ( Sigma-Aldrich , Cat-No: 11759051001 ) according to manufacturers instructions . αACT antibodies or amidoblack staining served as loading control . Two grams of plant material was collected under respective light conditions and ground for 8 min in liquid N2 . Protein was extracted using 4 ml IP buffer ( pH 7 . 8; 134 mM Na2HPO4 , 1 . 56 mM NaH2PO4 , 150 mM NaCl , 1 mM KCl , 1 mM EDTA , 0 . 5% Triton X-100 , 1 mM Na3VO4 , 2 mM Na4P2O7 , 10 mM NaF , 1× Protease Inhibitor Cocktail ( Sigma-Aldrich , Cat-No: I3911 ) , 1× Protease cOmplete Inhibitor Cocktail ( Sigma-Aldrich , Cat-No: 04693159001 ) ) . The lysate was further homogenised using Potter-Elvehjem homogenisers and cleared by centrifugation ( 20 , 000× g , 4°C ) . A total of 100 μl Anti-GFP MicroBeads ( Miltenyi , Cat-No: 130-091-125 ) were added to the supernatant and incubated for 2 hr in darkness at 4°C . The pulldown was performed as indicated in the manufacturers protocol . Elution was done using 3× 50 μl 0 . 1 M TEA pH 13 . 1 and directly neutralised with 3 μl 1 M MES pH 2 . Eluate fractions 1 , 2 , and 3 were combined ( approx . 140–150 μl with pH 6–7 ) and 200 mM ammonium bicarbonate buffer pH 8 was added to obtain a total volume of 200 μl . The disulfide bonds in the proteins were reduced with TCEP and the free sulfhydrils derivatised with MMTS . Proteins were then precipitated with cold acetone ( 6× volume ) overnight at −20°C . The washed pellets were rehydrated in 25 mM ammonium bicarbonate buffer pH 8 at 37°C for 1 hr , then trypsin was added and the proteins digested overnight . An aliquot of the digests was analysed by LC-MS/MS on a Thermo Orbitrap Fusion Lumos mass spectrometer on-line coupled to a Waters nanoAcquity UPLC in data-dependent fashion using HCD fragmentation as follows . Five μl of the sample was loaded onto a Waters Symmetry trap column ( C18 , 5 μm , 180 μm × 20 mm ) in 99% solvent A ( 0 . 1% formic acid in water ) at a flow rate of 5 μl min−1 for 5 min . Peptides were separated by increasing solvent B ( 0 . 1% formic acid in acetonitrile ) from 5% to 35% in 90 min , then to 50% in 5 min and up to 90% in 4 min . MS survey scans ( m/z 380–1580; AGC 400 , 000; max inject time 50 ms , resolution 120 , 000 ) were followed by HCD scans ( m/z auto; AGC 50 , 000; max inject time 100 ms , resolution 15 , 000; isolation width 1 . 6 Da; cycle time 2 s , normalised collision energy 35% ) on the most abundant multiply charged ions with a minimum intensity threshold of 25 , 000 , then excluded for 10 s . Raw data were processed by Protein Discoverer ( v1 . 4 ) . The resulting peaklists were submitted to database search by ProteinProspector ( v5 . 22 . 0 ) first against the full Swissprot 2019 . 6 . 12 database ( 560 , 292 entries ) , then against Arabidopsis thaliana entries in the Uniprot 2019 . 6 . 12 database also considering the YFP-HA and HA-YFP-NOT9B sequences and contaminants identified from the Swissprot database in the initial database search ( 89 , 235 entries ) . A precursor mass tolerance of 5 ppm and a fragment mass tolerance of 20 ppm was used . Only fully tryptic peptides were considered with a maximum of two missed cleavages . Methylthio modification of Cys residues was used as fixed modification and Met oxidation , acetylation of protein N-termini and pyroglutamic acid formation of peptide N-terminal Gln residues as variable modifications . Acceptance criteria were as follows: protein score >22 , peptide score >15 , protein E < 0 . 01 , peptide E < 0 . 05 , protein FDR < 1% , peptide FDR < 1% , and a minimum of two unique peptide identifications per protein . Peptide search results are summarised in Supplementary file 1; raw data are available at MassIVE ( ftp://massive . ucsd . edu/MSV000086324/ ) . Plants were grown for 4 days under the respective light conditions . At least 17 individual seedlings of each line were measured using the ImageJ software as described elsewhere ( Kretsch , 2010 ) . For measurement of hypocotyl elongation , modified Prado 500 W universal projectors ( Leitz , Wetzlar , Germany ) were used as light sources with Xenophot longlife lamps ( Osram , Berlin ) . Light was passed through narrow-band filters ( 716 DAL for FR , KG65 for R ) . For all other experiments 740 nm LEDs have been used for FR light and 656 nm LEDs for R light . For plant cultivation fluorescent bulbs were used . For induction of germination plants were kept in a growth cabinet in W light ( Sanyo , Osaka , Japan ) . For cultivation plants were kept under fluorescent white light . Spectra of all light sources can be found in Figure 10 . Seedlings were grown for four days in either D , FR ( 740 nm , 40 μmol m−2 s−1 ) , or B ( 436 nm , 30 μmol m−2 s−1 PAR ) on ½ MS/1 . 2% agar supplemented with 1 . 5% sucrose . Anthocyanin was extracted by collecting 25 seedlings from each treatment and genotype into 500 μl extraction buffer ( 18% ( v/v ) 1-propanol , 0 . 37% ( v/v ) HCl ) . Samples were heated for 2 min to 95°C , chilled on ice for 5 min , and incubated overnight at 4°C under constant shaking in darkness . Plant material was separated by centrifugation for 10 min ( 13 , 000× g , room temperature ) and the supernatant was analysed . Anthocyanin was quantified by measuring A535 and A650 using a plate reader . The relative amount ( A535-A650 ) was calculated . Transient transformation of tobacco ( N . benthamiana ) was performed as described ( Chapman et al . , 2004 ) . For all experiments p35S:P19 was coinfiltrated . Transformation of A . thaliana using the floral dip method was performed as described ( Clough and Bent , 1998 ) . For in vivo detection of luciferase activity , seedlings were grown for 4 days under indicated light conditions . For detection seedlings were sprayed with luciferin solution ( 2 mM D-Luciferin ( Biosynth #L-8220 ) , 0 . 001% Triton-X , 100 mM Tris-Phosphate pH 8 ) . Pictures of the luminescence were taken using a cooled CCD-camera . False colour images were generated using the ImageJ software package . For in vivo detection of luciferase activity in yeast cells , plates were sprayed with luciferin solution ( 2 mM D-Luciferin ( Biosynth #L-8220 ) , 0 . 001% Triton-X , 100 mM Tris-Phosphate pH 8 ) . Pictures of the luminescence were taken using a cooled CCD-camera . False colour images were generated using the ImageJ software package . Staining of pNOT9B:GUS transgenic seedlings was done as described ( Enderle et al . , 2017 ) . Blots and microscopic images were brightness/contrast adjusted and cropped using ImageJ v1 . 52a and assembled using Inkscape 0 . 92 . 4 . In some figures , blots were cropped for spacing reasons; dotted lines show cropping . Statistical analysis was performed as indicated in the figure legends . Plots were created using the Matplotlib package in Python 3 . 7 . 6 using Spyder IDE v4 . 0 . 1 . Identifiers have been parsed to the PANTHER Classification system and analysed towards the GO aspect ‘molecular function’ . p-value cut-off was set to p<0 . 01 . Sequences were retrieved from UniProt and aligned in JalView using MAFFT with default settings ( UniProt Consortium , 2019; Katoh and Standley , 2013; Waterhouse et al . , 2009 ) . UniProt sequence identifiers are included in the sequence names . ACT1 , AT2G37620; AGO1 , AT1G48410; CAF1A , AT3G44260; CAF1B , AT5G22250; CCR4A , AT3G58560; CHS , AT5G13930; COP1 , AT2G32950; DCL1 , AT1G01040; DCP1 , AT3G13300; ELIP1 , AT3G22840; ELIP2 , AT4G14690; EPR1 , AT1G18330; FHL , AT5G02200; FHY1 , AT2G37678; HEN1 , AT4G20910; HY5 , AT5G11260; HYL1 , AT1G09700; MYBD , AT1G70000; NOT1 , AT1G02080; NOT2B , AT5G59710; NOT4A , AT5G60170; NOT4B , AT3G45630; NOT9A , AT3G20800; NOT9B , AT5G12980; NOT9C , AT2G32550; PHYA , AT1G09570; PHYB , AT2G18790; PIF4 , AT2G43010; RH8 , AT4G00600; RRC1 , AT5G25060; SE , AT2G27100; SFPS , AT1G30480; SPA1 , AT2G46340; XRN4 , AT1G54490 . | Place a seedling on a windowsill , and soon you will notice the fragile stem bending towards the glass to soak in the sun and optimize its growth . Plants can ‘sense’ light thanks to specialized photoreceptor molecules: for instance , the phytochrome A is responsible for detecting weak and ‘far-red’ light from the very edge of the visible spectrum . Once the phytochrome has been activated , this message is relayed to the rest of the plant through an intricate process that requires other molecules . The CCR4-NOT protein complex is vital for all plants , animals and fungi , suggesting that it was already present in early life forms . Here , Schwenk et al . examine whether CCR4-NOT could have acquired a new role in plants to help them respond to far-red light . Scanning the genetic information of the plant model Arabidopsis thaliana revealed that the gene encoding the NOT9 subunit of CCR4-NOT had been duplicated in plants during evolution . NOT9B , the protein that the new copy codes for , has a docking site that can attach to both phytochrome A and CCR4-NOT . When NOT9B binds phytochrome A , it is released from the CCR4-NOT complex: this could trigger a cascade of reactions that ultimately changes how A . thaliana responds to far-red light . Plants that had not enough or too much NOT9B were respectively more or less responsive to that type of light , showing that the duplication of the gene coding for this subunit had helped plants respond to certain types of light . The findings by Schwenk et al . illustrate how existing structures can be repurposed during evolution to carry new roles . They also provide a deeper understanding of how plants optimize their growth , a useful piece of information in a world where most people rely on crops as their main source of nutrients . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"plant",
"biology"
] | 2021 | Uncovering a novel function of the CCR4-NOT complex in phytochrome A-mediated light signalling in plants |
Diverse clustered protocadherins are thought to function in neurite morphogenesis and neuronal connectivity in the brain . Here , we report that the protocadherin alpha ( Pcdha ) gene cluster regulates neuronal migration during cortical development and cytoskeletal dynamics in primary cortical culture through the WAVE ( Wiskott-Aldrich syndrome family verprolin homologous protein , also known as Wasf ) complex . In addition , overexpression of proline-rich tyrosine kinase 2 ( Pyk2 , also known as Ptk2b , Cakβ , Raftk , Fak2 , and Cadtk ) , a non-receptor cell-adhesion kinase and scaffold protein downstream of Pcdhα , impairs cortical neuron migration via inactivation of the small GTPase Rac1 . Thus , we define a molecular Pcdhα/WAVE/Pyk2/Rac1 axis from protocadherin cell-surface receptors to actin cytoskeletal dynamics in cortical neuron migration and dendrite morphogenesis in mouse brain .
The human brain contains approximately 86 billion neurons , and each neuron engages in several thousand specific synaptic connections , resulting in complex neural networks with over 1015 specific connections . These complex neural circuits are required for normal brain function , and inappropriate assemblies of neural circuits underlie neurodevelopmental and neuropsychiatric disorders ( Hyman , 2008 ) . A remarkable feature of neurodevelopment is the long-distance neuronal migration from the site of origin to the final destination ( Angevine and Sidman , 1961; Ayala et al . , 2007 ) . For example , cortical immature neurons generated from the proliferative ventricular and subventricular zones ( VZ/SVZ ) migrate radially through specific phases to appropriate laminar positions in an ‘inside-out’ manner and then differentiate into distinct subtypes of cortical neurons ( Angevine and Sidman , 1961; LoTurco and Bai , 2006; Rakic , 1974 ) . The cortical migration phases include somal translocation , multipolar migration , and glial-guided locomotion ( Ayala et al . , 2007; Cooper , 2014; Noctor et al . , 2004 ) . Newly born bipolar neurons in SVZ assume multipolar or stellate morphology and migrate randomly in the intermediate zone ( IZ ) , moving tangentially , up , or down ( Ayala et al . , 2007; Cooper , 2014; Jossin and Cooper , 2011; Nadarajah et al . , 2003; Noctor et al . , 2004; Tabata and Nakajima , 2003 ) . They then transit into bipolar again near the border of IZ/CP ( cortical plate ) and resume final radial migration to settle in appropriate cortical layers ( Ayala et al . , 2007; Cooper , 2014; Jossin and Cooper , 2011; Nadarajah et al . , 2003; Noctor et al . , 2004; Tabata and Nakajima , 2003 ) . Abnormal neuronal migration results in various neurodevelopmental and psychiatric diseases ( Ayala et al . , 2007; LoTurco and Bai , 2006; Valiente and Marín , 2010 ) ; however , the underlying molecular mechanisms for the abnormal neuronal migration is largely unknown . Human genetics studies have implicated mutations of the clustered protocadherin ( Pcdh ) cell adhesion genes in the 5q31 region for various developmental and psychiatric disorders ( Anitha et al . , 2013; Iossifov et al . , 2012; Pedrosa et al . , 2008; Shimojima et al . , 2011 ) . Similar to Dscam1 in Drosophila ( Zipursky and Sanes , 2010 ) , diverse clustered Pcdh genes play an important role in establishing neuronal identity and connectivity in the vertebrate brain ( Garrett et al . , 2012; Lefebvre et al . , 2012; Molumby et al . , 2016; Nicoludis et al . , 2016; Rubinstein et al . , 2015; Schreiner and Weiner , 2010; Suo et al . , 2012; Thu et al . , 2014; Wu and Maniatis , 1999 ) . In mice , 58 clustered Pcdh genes are organized into three closely linked Pcdh α , β , and γ clusters ( Pcdha , Pcdhb , and Pcdhg ) ( Wu et al . , 2001 ) . The Pcdh α and γ clusters are each consisted of variable and constant regions , similar to that of the Ig , Tcr , and Ugt1 gene clusters ( Wu , 2005; Wu and Maniatis , 1999; Wu et al . , 2001; Zhang et al . , 2004 ) . In particular , the variable region of the mouse Pcdhα cluster contains 12 highly similar ‘alternate exons’ , α1-α12 , whose promoters are stochastically activated by distal enhancers , and two divergent c-type ‘ubiquitous exons’ , αc1 and αc2 , whose promoters are constitutively activated by distal enhancers ( Figure 1A ) ( Guo et al . , 2012 ) . Each variable exon is separately spliced to the common set of downstream constant exons , generating diverse mRNAs and proteins . CCCTC-binding factor ( CTCF ) /Cohesin-mediated topological chromatin-looping domains are crucial for proper expression of Pcdhα proteins ( Guo et al . , 2015; Huang and Wu , 2016 ) . Each variable exon encodes an extracellular domain ( ectodomain EC1-6 ) , a transmembrane segment , and a juxtamembrane variable cytoplasmic domain ( VCD ) ( Shonubi et al . , 2015; Wu and Maniatis , 1999 ) , whereas the three constant exons encode a common membrane-distal constant domain ( CD ) of all Pcdhα proteins ( Figure 1A ) . This suggests that diverse extracellular cues converge on a single intracellular signaling pathway . However , the functional significance of this intriguing arrangement remains obscure . A large family of cell-surface receptors , including Pcdhα6 ( Pcdha6 ) , recruit WAVE complex to the plasma membrane ( Chen et al . , 2014; Nakao et al . , 2008; Tai et al . , 2010 ) . The WAVE complex is a conserved two-partite pentameric complex consisting of a pseudosymmetric dimer of Sra1/Cyfip1 and Nap1/Hem2 , and a heteromeric trimer of HSPC300/Brick , Abi1/2/3 , and WAVE1/2/3/SCAR ( Chen et al . , 2010 ) . First , Abi2 interacts with Abelson tyrosine kinase ( Abl kinase ) and has been implicated in cortical radial migration ( Xie et al . , 2013 ) . Second , WAVEs/SCARs are members of the Wiskott-Aldrich syndrome protein ( WASP ) and WASP verprolin homologous protein family , defined by a conserved VCA domain ( verprolin homologous , cofilin homologous or central hydrophobic , and acidic regions ) ( Chen et al . , 2010 ) . Third , VCA is inhibited by intermolecular interaction with Sra1 and intramolecular interaction within WAVE ( Chen et al . , 2010; Padrick et al . , 2011; Rohatgi et al . , 1999 ) . Fourth , Rac1 binds to WAVE complex and induces a conformational change to release VCA from its inhibitory state and to activate actin filament nucleation and branching through the Arp2/3 complex ( Chen et al . , 2010; Lebensohn and Kirschner , 2009; Padrick et al . , 2011; Rohatgi et al . , 1999; Ti et al . , 2011 ) . Finally , Pyk2 , a calcium-dependent cell-adhesion kinase and scaffold protein highly expressed in the brain and inhibited by Pcdhα , also regulates neurodevelopment ( Chen et al . , 2009; Hsin et al . , 2010; Lev et al . , 1995; Suo et al . , 2012 ) . However , whether and how WAVE complex and Pyk2 kinase function in cortical neuron migration are not clear . Here , we report that Pcdhα proteins play a critical role in neuronal migration and cytoskeletal dynamics . Specifically , we define an actin cytoskeleton remodeling pathway by which Pcdhα regulates lamellipodial and filopodial dynamics and neuronal migration as well as dendrite morphogenesis through interaction with WAVE complex via the WIRS ( WAVE interacting receptor sequence ) motif of Pcdhα constant domain ( CD ) . In addition , Pyk2 regulates cortical neuron migration by inactivating the small GTPase Rac1 . Given that actin cytoskeletal dynamics are central for neurite morphogenesis and neuronal migration , our findings have interesting implications for mechanisms of Pcdhα functions in dendrite self-avoidance and neuronal self/nonself recognition in normal brain development as well as aberrant neuron migration and dendrite morphogenesis underlying complex neurodevelopmental diseases .
We mapped the embryonic expression pattern of Pcdhα by using a GFP knockin mouse line ( PcdhαGFP ) ( Wu et al . , 2007 ) and found that Pcdhα proteins are expressed throughout the developing forebrain ( Figure 1B ) . Immunostaining with an antibody against alpha constant domain ( αCD ) revealed that Pcdhα proteins are expressed in all cortical regions and most prominently in the intermediate zone and marginal zone ( IZ and MZ ) of the developing neocortex ( Figure 1C ) . RT-PCR with isoform-specific primers showed that , starting at E10 , every member of the Pcdhα cluster is expressed in the developing brain ( Figure 1—figure supplement 1A ) . Pcdhα knockdown ( αKD ) with two independent shRNAs , each targeting a distinct subdomain of the constant region by in utero electroporation ( IUE ) , revealed a significant decrease of migrating neurons in the cortical plate ( CP ) and a concomitant increase within the lower intermediate zone , suggesting defects in multipolar migration ( Figure 1D and Figure 1—figure supplement 1B ) . The αKD multipolar neurons in the intermediate zone display stunted processes , as shown by lucida drawings ( Figure 1E ) . Live cell imaging of brain organotypic slice culture demonstrated the slower velocity of multipolar migration of αKD neurons compared to controls ( Figure 1F–H and Video 1 ) . In addition , early born αKD neurons also have migration defects , suggesting that Pcdhα is also required for glia-independent somal translocation ( Figure 1I and J ) . This suggests that Pcdhα is required for migration of immature cortical neurons . To rule out the possibility of altered progenitor proliferation , we labeled αKD mouse brain with BrdU and analyzed cell proliferation . Compared with controls , αKD results in no significant difference of percentage of BrdU+ cells ( Figure 1—figure supplement 1C and D ) . In addition , αKD does not alter the percentage of Tbr2+intermediate progenitor cells ( IPCs ) ( Figure 1—figure supplement 1E and F ) , nor the morphology of brain lipid binding protein ( BLBP ) -labeled radial glia cells ( Figure 1—figure supplement 1G ) . Moreover , the defect is not due to increased apoptosis ( Figure 1—figure supplement 1H ) . Finally , there is no cortical migration defect ( Figure 1—figure supplement 1I ) in mice with deletion of the entire Pcdhα cluster ( αKO ) ( Wu et al . , 2007 ) . The phenotypic discrepancy may be due to known genetic compensation mechanisms induced by deletion but not knockdown ( Rossi et al . , 2015 ) . To rescue the migration defect , we constructed shRNA-resistant forms of α6 ( α6* ) , which represents members of the alternate α1-α12 , and of the two divergent c-types ( αc1* and αc2* ) ( Figure 2—figure supplement 1A ) . Indeed the single α6* isoform rescues the αKD migration defect . The Pcdh αc1* also rescues the migration defect; however , αc2* does not ( Figure 2A and Figure 2—figure supplement 1B ) . This suggests that αc2 has distinct functions other than cortical neuron migration , consistent with very recent findings that αc2 endows serotonergic neurons with a single cell-type identity and specifically mediates the axonal tiling and assembly of serotonergic neural circuitries ( Chen et al . , 2017 ) . To investigate whether the extracellular domain and transmembrane segment play a role in cortical neuron migration , we replaced them with a myristoylation signal to attach the shRNA-resistant intracellular domain ( ICD ) to the plasma membrane ( Myr-α6ICD* , Myr-αc1ICD* , Myr-αc2ICD* ) ( Figure 2—figure supplement 1A ) . We found that Myr-α6ICD* and Myr-αc1ICD* rescue the migration defect , and Myr-αc2ICD* does not ( Figure 2—figure supplement 1C and D ) . This suggests that the intracellular domain of Pcdhα plays an important role in cortical neuron migration . To investigate why Myr-αc2ICD* cannot rescue the migration defect , we constructed an αc2 VCD-deleted form , which is , by definition , a myristoylated α constant domain ( Myr-αCD* ) ( Figure 2—figure supplement 1A ) . Intriguingly , we found that Myr-αCD* rescues the migration defect ( Figure 2—figure supplement 1C and D ) . This demonstrated that αc2 variable cytoplasmic domain has an inhibitory function . Consistently , sequence analysis revealed that αc2 variable cytoplasmic domain is the longest and most divergent among those of αc1 as well as of α1-α12 ( Figure 2—figure supplement 1E ) . Together , these data suggest that members of the Pcdhα family except αc2 regulate cortical neuron migration through their common constant domain . Recent studies linked Pcdhα6 to the WAVE complex through the WIRS ( WAVE interacting receptor sequence ) motif within the Pcdhα constant domain ( Chen et al . , 2014 ) . We thus investigated whether Pcdhα regulates cortical neuron migration through WAVE . Remarkably , we found that overexpression of either WAVE2 ( Wasf2 ) or Abi2 in vivo rescues the cortical neuron migration defect of αKD neurons ( Figure 2B ) although they themselves have no apparent influence on cortical neuron migration ( Figure 2C ) . Consistently , endogenous Pcdhα and WAVE2 co-localize in primary cultured cortical neurons ( Figure 2D and E ) . In addition , mutating the WIRS motif ( from FITFGK to FIAAGK ) of α6* , αc1* , and Myr-αCD* ( α6*-AA , αc1*-AA , and Myr-αCD*-AA ) abolishes the rescue effect ( Figure 2F and G , in comparison to Figure 2A and Figure 2—figure supplement 1D ) . As controls , these WIRS-mutated isoforms as well as wild types appears to reach the plasma membrane ( Figure 2—figure supplement 1F ) . Thus , Pcdhα regulates cortical neuron migration through the WAVE complex . Pcdhα physically interacts with and negatively regulates the Pyk2 kinase ( Chen et al . , 2009 ) . In addition , we previously found that Pcdhα regulates dendritic and spine morphogenesis through inhibiting Pyk2 activity ( Suo et al . , 2012 ) . To this end , we investigated whether knockdown of Pyk2 could rescue cortical neuron migration defects of αKD . Although Pyk2 ( Ptk2b ) knockdown ( Pyk2KD ) per se or CRISPR knockout of Pyk2 ( Pyk2KO ) does not affect cortical neuron migration ( Figure 3A and Figure 3—figure supplement 1A ) , we found that Pyk2KD rescues the defect of cortical neuron migration in αKD ( Figure 3A and Figure 3—figure supplement 1B ) . This suggests that Pcdhα regulates cortical neuron migration , at least in part , through the inhibition of Pyk2 . We next asked whether overexpression of Pyk2 ( Pyk2OE ) could recapitulate αKD cortical neuron migration defects . We found that the majority of Pyk2OE cells are stalled in the middle intermediate zone ( mIZ ) ( Figure 3B ) , a stage little later than the stalling of αKD cells ( Figure 3A ) . In addition , these mIZ cells have aberrant multipolar morphology with supernumerary primary processes in comparison to single leading processes of control cells ( Figure 3C–E ) . For the very few Pyk2OE cells in the lower cortical plate ( CP ) , they harbor elaborated leading processes ( Figure 3F and G ) ; by contrast , control cells displayed typical bipolar morphology with a single or bifurcated thick leading process ( Figure 3F and G ) . Pyk2OE leads to the inhibition of Rac1 activity ( Suo et al . , 2012 ) . As Rac1 is thought to provide the spatial information for actin polymerization ( Tahirovic et al . , 2010 ) , loss of Rac1 activity leads to aberrant actin polymerization at many sites with no controlled spatial information , resulting in supernumerary primary processes ( Figure 3C–E ) and more branchy morphology ( Figure 3F and G ) . Finally , time-lapse imaging showed that there is a significant difference of velocity of cortical neuron migration between Pyk2OE and control cells ( Figure 3H and I , and Video 2 ) . These data suggest that Pyk2OE partially recapitulates cortical neuron migration defects . We next examined the orientation of the Golgi apparatus of cells in mIZ , which is essential for transporting vesicles for oriented motility ( Jossin and Cooper , 2011 ) , by immunostaining with a Golgi marker GM130 ( Figure 3J ) . Most Golgi complexes are normally localized in front of the cell nucleus and are oriented toward the cortical plate ( Jossin and Cooper , 2011 ) . However , the polarity of most Pyk2OE cells is disrupted , showing oblique or inverted orientation of the Golgi apparatus ( Figure 3J and K ) . Thus , Pyk2OE blocks multipolar-bipolar transition by disrupting proper localization of the Golgi apparatus . Finally , early-born Pyk2OE neurons are also stalled at the intermediate zone , suggesting that Pyk2 also plays a role in somal translocation ( Figure 3—figure supplement 1C and D ) . To rule out the potential nonspecific effect of the CAG promoter , which is active in both progenitors and postmitotic neurons , we ectopically overexpressed Pyk2 at E15 . 5 only in postmitotic neurons using the NeuroD promoter ( Jossin and Cooper , 2011 ) . We found that Pyk2OE under the NeuroD promoter also significantly impairs cortical neuron migration in postmitotic neurons ( Figure 3—figure supplement 1E-G ) . Taken together , this suggests that Pcdhα regulates cortical neuron migration , at least in part , through inhibiting Pyk2 kinase activity . We previously found that Rac1 is epistatic downstream of Pyk2 in dendrite development and spine morphogenesis ( Suo et al . , 2012 ) . To investigate whether Pyk2-Rac1 pathway also functions in cortical neuron migration , we overexpressed a constitutive active form Rac1 ( Rac1Q61L ) in Pyk2OE neurons . We found that Rac1Q61L rescues defects of multipolar migration and morphology of Pyk2OE neurons ( Figure 4A–C ) , although Rac1Q61L itself has no apparent effect on cortical neuron migration ( Figure 4D ) . However , overexpression of another constitutively active form of Rac1 ( Rac1G12V ) impairs cortical neuron migration ( Figure 4D ) ( Konno et al . , 2005 ) and cannot be used to rescue , likely because it has a lower affinity for GTP and thus lower constitutive activity than Rac1Q61L . Thus , the two constitutively active forms of Rac1 have distinct roles in cortical neuron migration ( Figure 4A and D ) . Together , we conclude that Pyk2OE inhibits multipolar-bipolar transition and leads to aberrant branchy morphology in the intermediate zone by inactivating the small GTPase Rac1 . Pyk2 functions as an enzyme through its middle kinase domain and as a molecular scaffold through its N-terminal FERM ( four-point-one , ezrin , radixin , moesin ) domain ( Figure 4—figure supplement 1A ) ( Chen et al . , 2009; Lev et al . , 1995; Suo et al . , 2012 ) . We systematically engineered Pyk2 by mutating a series of key residues of its enzymatic kinase cascade . We found that overexpression of Pyk2Y402F , an autophosphorylation mutant that still can be activated by endogenous Pyk2 , as well as Pyk2Y579F , Pyk2Y580F , and Pyk2Y881F , still recapitulate the migration defects of αKD ( Figure 4—figure supplement 1A and B ) . However , overexpression of Pyk2K457A , which has a mutation at the catalytic center and is completely kinase-dead ( Suo et al . , 2012 ) , cannot recapitulate the migration defects of αKD ( Figure 4—figure supplement 1A and B ) . This suggests that the catalytic activity of overexpressed Pyk2 is essential for recapitulating the migration defects of αKD . Remarkably , overexpression of the Pyk2 FERM domain alone recapitulates the blocking activity of Pyk2OE ( Figure 4—figure supplement 1A and C ) , whereas deletion of FERM domain abolishes the blocking ( Figure 4—figure supplement 1A and C ) . Consistently , the C-terminal FAT domain of Pyk2 is not required for the blocking effect and the kinase domain alone cannot block cortical neuron migration ( Figure 4—figure supplement 1A and C ) . This is consistent with that Pyk2 has important kinase-independent functions in contextual fear memory ( Suo et al . , 2017 ) . Together , we conclude that both Pyk2 kinase cascade and FERM scaffold are crucial for blocking cortical neuron migration . As stated above , constitutive active Rac1Q61L rescues the blocking effect of Pyk2OE ( Figure 4A ) . However , we found that Rac1Q61L cannot rescue the blocking activity of FERM domain ( Figure 4—figure supplement 1D ) . This suggests that constitutive active form of Rac1 only functions downstream of the kinase cascade but not the FERM scaffold of Pyk2 . We next investigated actin dynamics underlying neuronal migration in primary cultured cortical neurons . The early development of primary cultured neurons can be divided into two stages: stage 1 , in which the cell body is surrounded by flattened lamellipodia and stage 2 , in which the lamellipodia transform into definitive processes with growth cones ( Dotti et al . , 1988 ) . At stage 1 , we found that the size of lamellipodia around cell cortex in αKD neurons decreases significantly compared with controls ( Figure 5A and B ) . In addition , α6* , αc1* , or Myr-αCD* rescues the αKD lamellipodial defect . By contrast , αc2* does not rescue ( Figure 5C and D ) , which is consistent with that αc2* cannot rescue the defects of cortical neuron migration ( Figure 2A ) . Moreover , mutating the WIRS motif ( from FITFGK to FIAAGK ) in either α6* , αc1* , or Myr-αCD* abolishes their rescue effects ( Figure 5E and F ) , similar to the situation in cortical neuron migration ( Figure 2G ) . Finally , both WAVE2 and Abi2 rescue the lamellipodial defect ( Figure 5G and H ) . At stage 2 , αKD results in a significant decrease of percentage of primary neurites with lamellipodia-like protrusions ( Figure 5—figure supplement 1A and B ) . Consistent with the situation at stage 1 , α6* , αc1* , or Myr-αCD* rescues this αKD lamellipodial defect while αc2* does not ( Figure 5—figure supplement 1C and D ) , and mutating the WIRS motif ( from FITFGK to FIAAGK ) abolishes the rescue effects of either α6* , αc1* , or Myr-αCD* ( Figure 5—figure supplement 1E and F ) . In addition , consistent with stage 1 , both WAVE2 and Abi2 rescue the lamellipodial defect of stage 2 αKD neurons ( Figure 5—figure supplement 1G and H ) . Finally , αKD lamellipodial dynamics are significantly impaired in comparison with control neurons , whose veil-like lamellipodia are motile and are constantly extending and retracting in both stage 1 and stage 2 neurons ( Figure 5I , Figure 5—figure supplement 1I , Video 3 and Video 4 ) . These data demonstrated that Pcdhα is indispensable for lamellipodial dynamics . Because lamellipodial dynamics are essential for cell migration ( Krause and Gautreau , 2014 ) , this suggests that cortical neuron migration defects of αKD are a consequence of impairment of lamellipodial formation and cytoskeletal dynamics . Consistent with that Pyk2KD rescues cortical neuron migration defects of PcdhαKD ( Figure 3A ) , we found that knockdown of Pyk2 in αKD cells results in a significant increase of lamellipodial sizes of stage1 neurons as well as of the percentage of primary neurites with lamellipodia of stage2 neurons ( Figure 6A–C ) . In addition , Pyk2OE results in a significant decrease of lamellipodial sizes , consistent with that of αKD ( Figure 6D and E ) . Filopodia are thin membrane protrusion pushed by underlying actin bundles and filopodial formation is also dependent on Arp2/3 complex ( Mattila and Lappalainen , 2008 ) , we found that Pyk2OE results in a significant increase of filopodial number per stage 1 neuron as well as of primary neurite number per stage 2 neuron despite no alternation in αKD cells ( Figure 6D–G ) . Finally , similar to the rescue of cortical neuron migration defects of PykOE ( Figure 4A ) , we found Rac1Q61L rescues both lamellipodial and filopodial defects of Pyk2OE ( Figure 6D–G ) . In summary , although both αKD and Pyk2OE impact cytoskeletal dynamics , they have subtle differences on both lamellipodia and filopodia . To see whether growth cones with lamellipodia and filopodia are affected in vivo , we co-electroporated Lifeact , an actin marker , with either αKD or Pyk2OE plasmids into the developing mouse cortex . In the lower intermediate zone , αKD neurons exhibit abnormal enrichment of Lifeact-labeled actin structures in stunted processes and cell bodies , while the control neurons extend long processes with growth cones ( Figure 6—figure supplement 1A ) . In the upper intermediate zone , Pyk2OE neurons exhibit branchy morphology with multiple aberrant processes; however , the control neurons have normal bipolar morphology with single leading processes and growth cones ( Figure 6—figure supplement 1B ) .
Recent studies revealed that a zipper-like ribbon structure assembles from combinatorial cis- and trans-interactions between like-sets of the clustered Pcdhs located in apposed plasma membranes of neighboring cells ( Nicoludis et al . , 2016; Rubinstein et al . , 2015; Schreiner and Weiner , 2010; Thu et al . , 2014; Wu , 2005 ) . These protocadherin interactions could provide enormous diversity and exquisite specificity for neuronal connectivity and neurite self-avoidance required for mammalian brain development . While exquisite specificity is determined by strict homophilic trans-interactions of highly diversified EC2/3 ( Goodman et al . , 2017; Molumby et al . , 2016; Nicoludis et al . , 2016; Rubinstein et al . , 2015; Schreiner and Weiner , 2010; Thu et al . , 2014; Wu , 2005 ) ; enormous diversity is mainly generated by promiscuous cis-interactions of highly conserved EC5/6 ( Nicoludis et al . , 2016; Rubinstein et al . , 2015; Schreiner and Weiner , 2010; Thu et al . , 2014; Wu , 2005 ) . One intriguing genomic architecture of the Pcdhα cluster is multiple tandem variable exons followed by a single set of three constant exons , encoding a common cytoplasmic constant domain , which is shared by all members of the Pcdhα family ( Figure 1A ) ( Huang and Wu , 2016; Wu and Maniatis , 1999 ) . The extracellular domains of Pcdhα provide enormous diversity and exquisite specificity for cell recognition and adhesion ( Nicoludis et al . , 2016; Rubinstein et al . , 2015; Schreiner and Weiner , 2010; Thu et al . , 2014; Wu , 2005 ) . However , the intracellular Pcdhα signaling pathway is largely unknown . We propose a Pcdhα-based WAVE clustering model for cortical neuron migration ( Figure 7 ) . Distinct Pcdhα isoforms on the cell surface recruit WAVE complex to the cell cortex under the plasma membrane . This is strongly supported by ( 1 ) the specific interaction between members of the Pcdhα family and the WAVE complex through the WIRS motif in Pcdhα constant domain ( Chen et al . , 2014 ) ; ( 2 ) the rescue of migration and lamellipodial defects of αKD neurons by WAVE complex subunits WAVE2 and Abi2; and ( 3 ) the abolishment of the rescue effect by WIRS mutations . The WIRS motif of members of the Pcdhα family binds to a composite surface formed by Abi2 and Sra1 of the WAVE complex ( Chen et al . , 2014 ) . In addition , the Pcdhα proteins may also recruit WAVE complex through the direct binding of Abi2 C-terminal SH3 domain to the four protocadherin PXXP motifs , which are specific to the constant domain of the Pcdhα but not Pcdhγ family ( Wu and Maniatis , 1999 ) . Consistently , WAVE2 and Abi2 are required for growth cone activity during cortical neuron migration ( Xie et al . , 2013 ) . We recently found that N-WASP , a homolog of WAVE2 , also regulates cortical neuron migration ( Shen et al . , 2018 ) . In addition , Pcdhα binds to Pyk2 via the intracellular domain and inhibits Pyk2 phosphorylation and activation ( Chen et al . , 2009; Suo et al . , 2012 ) , resulting in disinhibition of small GTPase Rac1 ( Figure 7 ) . Moreover , our data suggest that Pyk2 also has kinase-independent scaffolding activity through its FERM ( four-point-one , ezrin , radixin , moesin ) domain , similar to the FERM domain of FAK , which binds numerous interacting partners and connects cell cortex to diverse downstream intracellular pathways ( Frame et al . , 2010 ) . Rac1 , in conjunction with Pcdhα , activate the WAVE complex ( Chen et al . , 2010; Lebensohn and Kirschner , 2009; Rohatgi et al . , 1999 ) . Two activated WAVE complexes , probably induced by protocadherin dimerization , in turn stimulate actin-nucleating activity of Arp2/3 through the two VCAs ( Padrick et al . , 2011; Ti et al . , 2011 ) . The Arp2/3-mediated actin branching nucleation is central for cytoskeletal dynamics and cell motility ( Krause and Gautreau , 2014; Lebensohn and Kirschner , 2009 ) . Our finding that αKD blocks lamellipodial and filopodial formation and cytoskeletal dynamics is also consistent with the WAVE clustering model . Taken together , we suggest that Pcdhα regulates the formation and dynamics of lamellipodial and filopodial protrusions underlying cortical neuron migration through the WAVE/Pyk2/Rac1 axis ( Figure 7 ) . We noted that αKD neurons stall in the lower intermediate zone and Pyk2OE neurons stall in the middle intermediate zone . In other words , αKD phenotype is more severe than that of Pyk2OE . In addition , αKD neurons display stunted processes while Pyk2OE neurons have branchy morphology . Consistently , the WAVE clustering model suggests that , in addition to disinhibition of Pyk2 and consequently inhibition of Rac1 , αKD also impairs the membrane recruiting of the WAVE complex directly ( Figure 7 ) . It is puzzling why Pcdhαc2 is different from other members of the Pcdhα family ( Figures 2A , 5C and D , and Figure 2—figure supplement 1D , Figure 5—figure supplement 1C and D ) . However , a recent study revealed an intriguing role of αc2 in serotonergic axonal local tiling and global assembly ( Chen et al . , 2017 ) . Given the known role of variable cytoplasmic domain of clustered Pcdh proteins in their cytoplasmic association ( Shonubi et al . , 2015 ) , the unique sequences of the αc2 variable cytoplasmic domain may restrict its role to axonal tiling of serotonergic neurons but not cortical neuron migration . Diverse roles of the clustered Pcdh genes in axonal targeting , dendritic tiling and self-avoidance , spine morphogenesis , synaptogenesis and connectivity have been reported ( Garrett et al . , 2012; Katori et al . , 2009; Lefebvre et al . , 2012; Molumby et al . , 2016; Rubinstein et al . , 2015; Suo et al . , 2012; Thu et al . , 2014; Zipursky and Sanes , 2010 ) . In particular , genetic studies demonstrated that Pcdhα functions in axonal projection of olfactory sensory and serotonergic neurons ( Chen et al . , 2017; Hasegawa et al . , 2008; Katori et al . , 2009; Mountoufaris et al . , 2017 ) . In addition , another WIRS-containing protocadherin , Celsr3 , is also central for interneuron tangential migration and Globus Pallidus axonal connectivity in the mouse forebrain ( Jia et al . , 2014; Ying et al . , 2009 ) . It will be interesting to see whether these diverse protocadherin functions , in addition to the crucial role in cortical neuron migration , also require the complex WAVE/Pyk2/Rac1 signaling cascade ( Figure 7 ) . Sholl analysis demonstrated that the WIRS domain point mutation rescues the Pcdhα dominant-negative effects on dendrite outgrowth and branching of primary cultured cortical neurons , suggesting that the Pcdhα/WAVE/Pyk2/Rac1 signaling axis indeed functions in dendrite morphogenesis ( Figure 7—figure supplement 1 ) . Thus , the regulation of neuronal migration and neurite development by the Pcdhα/WAVE/Pyk2/Rac1 axis through actin cytoskeletal dynamics may be a general mechanism for diverse roles of protocadherins in brain development and function .
The PcdhαGFP mice were previously described ( Suo et al . , 2012; Wu et al . , 2007 ) . Pyk2KO and Pyk2Y402F mice were generated by CRISPR/Cas9 . Animals were maintained at 23°C in a 12 hr ( 7:00–19:00 ) light and 12 hr ( 19:00–7:00 ) dark schedule . The day of vaginal plug was considered to be embryonic day 0 . 5 ( E0 . 5 ) . All animal experiments were approved by the Institutional Animal Care and Use Committee ( IACUC ) of the Shanghai Jiao Tong University . Mouse lines of Pyk2KO and Pyk2Y402F were generated by using CRISPR/Cas9 . Briefly , sgRNA scaffold sequences were constructed in the pLKO . 1 plasmid . The construct was then used as template for amplifying a PCR product containing T7 promoter and sgRNA target sequence . The PCR product was gel-purified and used as templates for in vitro transcription of sgRNA ( T7-Transcription Kit , Invitrogen ) . Cas9 mRNA was transcribed in vitro from linearized pcDNA3 . 1-Cas9 plasmid ( T7-ULTRA-Transcription Kit , Ambion ) . Both Cas9 mRNA and sgRNAs were purified ( Transcription Clean-Up Kit , Ambion ) , mixed in M2 ( Millipore ) at the concentration of 100 ng/μl , and then injected into the cytoplasm of fertilized eggs of C57BL/6 mice . For Pyk2Y402F mice , single-stranded oligo-donor nucleotides ( ssODN ) with mutation at Y402 residue and nonsense mutation at PAM sequence were co-injected together with the Cas9 mRNA and sgRNA . After equilibration for 30 min , 15–25 injected fertilized eggs were transferred into fallopian tube of pseudopregnant ICR mouse females . Offspring of these mice were genotyped by PCR , restriction endonuclease digestion , and Sanger sequencing . All oligos used are listed in Supplementary file 1 . The following antibodies were used for biochemistry experiments: mouse anti-β-actin ( 1:5000 , Proteintech ) , mouse anti-Myc ( 1:1000 , Millipore ) , rabbit anti-Pyk2 ( 1:500 , Abcam ) . The following antibodies were used for immunocytochemistry and immunohistochemistry: mouse anti-Tuj1 ( 1:300 , Covance ) , rabbit anti-αCD ( 1:500 , Synaptic Systems ) , rabbit anti-GFP ( 1:1000 , Invitrogen ) , rabbit anti-BLBP ( Brain lipid binding protein ) ( 1:500 , Chemicon ) , mouse anti-GM130 ( 1:1000 , BD Bioscience ) , rat anti-BrdU ( 1:1000 , Bio-Rad ) , rabbit anti-activated caspase 3 ( 1:500; Cell Signaling Technology ) , rabbit anti-Tbr2 ( 1:500 , Abcam ) , rabbit anti-WAVE2 ( 1:500 , Millipore ) , goat anti-rabbit Alexa Fluor 488 ( 1:300 , Molecular Probes ) , goat anti-rabbit Alexa Fluor 568 ( 1:300 , Molecular Probes ) , goat anti-mouse Alexa Fluor 568 ( 1:300 , Molecular Probes ) , goat anti-mouse Alexa Fluor 647 ( 1:300 , Molecular Probes ) . Full-length cDNAs of Pcdha6 , Pcdhac1 , Pcdhac2 , WAVE2 ( GenBank AY135643 . 1 ) , Abi2 ( GenBank NM_198127 . 2 ) were cloned from mouse brain total RNA preparations by reverse transcriptase PCR ( RT-PCR ) . The cDNAs of Myr-αCD , Rac1 and Rac1G12V , Pyk2 and Pyk2 mutations ( Pyk2Y402F , Pyk2K457A , Pyk2Y579F , Pyk2Y580F , Pyk2Y881F ) , Pyk2 fragments ( ΔFERM , ΔFAT , FERM domain , Kinase domain ) were cloned from previously published plasmids ( Suo et al . , 2012 ) . WIRS-mutated and αKD-resistant Pcdhα isoforms ( α6* , αc1* , αc2* , Myr-αCD* , α6*-AA , αc1*-AA , Myr-αCD*-AA , Myr-α6ICD* , Myr-αc1ICD* , Myr-αc2ICD* ) , Rac1Q61L , were constructed from the above plasmids . Constructs used in IUE for overexpression were cloned into the pCAG-Myc vector or pNeuroD-IRES-GFP vector ( kindly provided by Dr . Franck Polleux , Columbia University ) using restriction enzyme sites . For knockdown , short-hairpin RNA ( shRNA ) coding sequences were cloned into the pLKO . 1 vector . All oligo sequences with corresponding restriction enzyme sites are listed in Supplementary file 1 . Plasmids were validated by Sanger sequencing . IUE was performed as previously described with modifications ( Saito and Nakatsuji , 2001 ) . Briefly , dams were anesthetized with pentobarbital sodium . pLKO . 1-shRNAs ( 2 μg/μl ) for knockdown or pCAG-Myc ( 2 μg/μl ) constructs for overexpression were mixed with GFP-expressing plasmid pCAG-eGFP ( 0 . 5 μg/μl ) and 0 . 05% fast green . Laparotomy was performed to expose the uteri . The plasmid mixture was injected into the lateral ventricle of the embryonic brain . Five electrical pulses were applied at 40 Volts for a duration of 50 ms at 900 ms intervals using a tweezertrode ( 3 mm , BTX ) with an electroporator ( Gene Pulser System , Bio-Rad ) . The uterine horns were placed back into the abdominal cavity to allow the embryos to continue normal development . For cortical neuron primary culture , electroporated cortices were collected from E17 . 5 embryos in Hanks’ Balanced Salt Solution ( HBSS ) with 0 . 5% glucose , 10 mM Hepes , 100 μg/ml penicillin/streptomycin . The cortices were then digested with 0 . 25% trypsin for 10 min at 37°C . The reaction was terminated with 0 . 5 mg/ml trypsin inhibitor for 3 min at room temperature ( RT ) . The cortical tissues were gently triturated in the plating medium ( MEM with 10% FBS , 1 mM glutamine , 10 mM Hepes , 50 μg/ml penicillin/streptomycin ) until fully dissociated . Cell viability and density were determined using 0 . 4% trypan blue and a hemocytometer . The dissociated cells ( 1 × 105 ) were plated into four-well chamber or 35-mm glass-bottom Petri dish precoated with 100 μg/ml poly-L-lysine ( Sigma ) and 5 μg/ml laminin ( Invitrogen ) . The cells were incubated with 5% CO2 at 37°C for 4 hr . The plating medium was then replaced with a serum-free culture medium ( Neurobasal medium , 2% B27 , 0 . 5 mM glutamine , 50 μg/ml penicillin/streptomycin supplemented with 25 μM glutamate ) . For immunocytochemistry , cells were cultured for additional 20 hr in vitro ( hiv ) . For cortical organotypic slice culture , the head of E17 . 5 embryos were briefly placed in 70% ethanol and the brains were carefully dissected . The brains were embedded in 3% low-melting agarose and glued to the chuck of a water-cooled vibratome ( Leica ) . The 250-μm-thick whole-brain coronal sections were cut and collected in the sterile medium . The organotypic slices were carefully placed in a 0 . 4 μm membrane cell culture insert ( Millipore ) in a six-well plate . Slices were cultured in slice culture medium: 67% Basal Medium Eagle ( BME ) , 25% HBSS , 5% FBS , 1% N2 , 1% penicillin/streptomycin/glutamine ( Invitrogen ) and 0 . 66% glucose ( Sigma ) . Slices ( three per well ) were cultured in six-well plates at 37°C and 5% CO2 , incubated for 6–8 hr . The membrane insert with slices was then transferred on to a glass-bottom Petri dish ( MatTek ) . Images were taken at 3 μm steps with 10–15 optical sections and were captured every 15 min for up to 16 hr with the Nikon A1 confocal laser microscope system . For single-cell time-lapse imaging , cortical neurons were plated into a 35-mm glass-bottom Petri dish . Images were taken at 1 μm steps with 10–15 optical sections and were captured every 5 min for up to 10 hr with Nikon A1 confocal laser microscope system . Primary cultured cortical neurons were washed once with PBS , fixed in 4% PFA for 20 min at RT , washed and permeabilized with 0 . 2% Triton X-100 for 10 min . After blocking with 5% BSA , cells were incubated with primary antibodies at 4°C overnight followed by incubation of secondary antibodies for 1–2 hr at RT . F-actin was labeled by Alexa-546 phalloidin ( Sigma ) . For immunohistochemistry , the dams were sacrificed , and embryonic brains were fixed in 4% PFA overnight at 4°C . The brains were then sectioned at 50 μm with a vibratome ( Leica ) . Sections were washed three times in PBS , blocked in 3% BSA , 0 . 1% Triton X-100 in PBS for 1 hr at RT , and then incubated with primary antibodies at 4°C overnight and secondary antibodies at RT for 1–2 hr . Cell nuclei were visualized with DAPI . Images were collected with a confocal microscope ( Leica ) under a 10x objective for brain sections . High-resolution images were collected under a 60x oil objective with a 3x digital zooming factor for primary cultured neurons . HEK293T cells were maintained in DMEM with 10% FBS and 100 μg/ml penicillin/streptomycin . Cultured cells were transfected using Lipofectamine 2000 ( Invitrogen ) . Total protein of HEK293T cells was extracted by lysis buffer ( 50 mM Tris–HCl , pH 7 . 5 , 150 mM NaCl , 1% NP40 , 0 . 5% sodium deoxycholate , 0 . 1% SDS ) with protease inhibitors and then centrifuged at 12 , 000 × g at 4°C for 30 min . The lysates were subjected to Western blot analyses . Total RNA was extracted from embryonic mouse brain tissues with TRIzol ( Ambion ) . The reverse-transcription reaction was performed with 1 μg total RNA preparations . All oligos used are listed in Supplementary file 1 . For each group , the IUE experiments were performed using at least three pregnant female mice , by which we usually harvested at least six embryonic brains . We obtained 15 ~ 20 sections from each electroporated brain , and quantified one typical section per brain . Nearly identical areas in the presumptive somatosensory cortices of anatomically matched brain sections were chosen for imaging and quantification . For bin analysis , the cortices were divided into ten equal bins and all GFP+ neurons in each bin were counted . In total , about 150 ~ 300 cells were counted per section . Statistical significance was assessed using one-way ANOVA , followed by a post hoc Tukey’s multiple comparisons test . In primary culture experiments , the development stage of cultured neurons were defined as in Dotti’s paper: at stage 1 , the cell body was surrounded by flattened lamellipodia; at stage 2 , the lamellipodia transformed into neural processes with growth cones ( Dotti et al . , 1988 ) . We immunostained the cultured cells with Tuj1 ( Neuron-specific class III beta-tubulin ) antibody , a neuron-specific marker , to exclude differentiated glia or radial glia . For quantification , we selected neurons with typical stage 1 or stage 2 morphology based on GFP and phalloidin signals . For stage 1 neurons , we selected the lamellipodia region by the wand tool in the ImageJ software ( NIH ) and measured the area size . For stage 2 neurons , the neurite tips with F-actin-enriched protrusions two folds larger than its width were defined as ‘neurite with lamellipodia’ . Sholl analysis was performed as previously described ( Suo et al . , 2012 ) . The significance of differences between two groups was analyzed using unpaired Student’s t tests . One-way ANOVA was used for multiple comparisons by the GraphPad software . | There are hundreds of billions of neurons in a human brain , and each one can form several thousand connections with other neurons . This complex network determines our thoughts , memories , personality , and behavior , but how does it form ? During brain development , specific areas give rise to new neurons , which then migrate long distances to other parts of the brain . Upon arrival , they generate several structures , called dendrites , which connect with other neurons . To distribute themselves correctly , the migrating immature neurons must be able to travel long distances and steer clear of one another . The dendrites from a single mature neuron must also avoid each other , a phenomenon known as self-avoidance . Certain membrane-spanning proteins , called clustered protocadherins , may help neurons achieve this . The portion of the protocadherins that sits on the cell surface is highly variable , and acts as a zipcode that helps cells to recognize one another . However , the section of the protein inside the cell varies little and is shared by all members of a protocadherin family . When the clustered protocadherin is ‘switched on’ , this internal segment can trigger a cascade of reactions that create changes in the cell . Yet , little was known about the nature of this signaling cascade . Using gene editing in mice , Fan , Lu et al . focus on the signaling cascade of the clustered protocadherin alpha family . The experiments show that the internal portion of these proteins interacts with a protein complex called WAVE . It also inhibits an enzyme known as Pyk2 , which increases the activity of another enzyme called Rac1 GTPase , that then further activates WAVE . This results in the WAVE complex also interacting with the internal skeleton inside the neurons and dendrites , which regulates the ability of these cells to migrate and of the dendrites to avoid each other . Many brain conditions , such as autism spectrum disorders or depression , result from abnormal neuronal migration and connectivity . Mutations in the genes of clustered protocadherins increase the risk of these disorders . By showing how these proteins help to regulate the migration and connectivity of neurons , Fan , Lu et al . add to our understanding of brain development in health and disease . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2018 | Alpha protocadherins and Pyk2 kinase regulate cortical neuron migration and cytoskeletal dynamics via Rac1 GTPase and WAVE complex in mice |
Hippocampal place cells fire at different rates when a rodent runs through a given location on its way to different destinations . However , it is unclear whether such firing represents the animal’s intended destination or the execution of a specific trajectory . To distinguish between these possibilities , Lister Hooded rats ( n = 8 ) were trained to navigate from a start box to three goal locations via four partially overlapping routes . Two of these led to the same goal location . Of the cells that fired on these two routes , 95 . 8% showed route-dependent firing ( firing on only one route ) , whereas only two cells ( 4 . 2% ) showed goal-dependent firing ( firing similarly on both routes ) . In addition , route-dependent place cells over-represented the less discriminable routes , and place cells in general over-represented the start location . These results indicate that place cell firing on overlapping routes reflects the animal’s route , not its goals , and that this firing may aid spatial discrimination .
A long-standing view of the hippocampus is that it contains a neural representation of space , a ‘cognitive map’ ( Tolman , 1948 ) , that encodes locations via the spatial receptive fields of place cells ( O'Keefe and Nadel , 1978; O'Keefe , 1999 ) . However , when a rat repeatedly traverses the same location on its way to different destinations , the place fields of hippocampal place cells are strongly modulated by where the animal is going or where it has come from ( Wood et al . , 2000; Frank et al . , 2000; Ferbinteanu and Shapiro , 2003; Bower et al . , 2005; Ainge et al . , 2007; Ji and Wilson , 2008; Pastalkova et al . , 2008; Ferbinteanu et al . , 2011; Allen et al . , 2012; Catanese et al . , 2014; Ito et al . , 2015 ) . This suggests that place cells represent not just near instantaneous location ( Muller and Kubie , 1989 ) , but also aspects of the animal’s goal directed behaviour . In tasks where consistent differences in place cell firing are observed , rats are usually well-trained and execute rapid trajectories to goal locations . For example , in a study by Ainge et al . , 2007 , rats ran up the central stem of a double-Y maze to gather reward in one of four goal boxes . Place fields on the central stem and on adjacent portions of the maze often exhibited strong modulation depending on the goal box the animal was headed towards ( see Figure 3 in Ainge et al . , 2007 ) . One interpretation of this ‘splitter cell’ pattern of firing ( hereafter differential firing ) is that it represents the animal’s intended destination . However , another interpretation is possible . To reach each goal location , the rat traversed a partially overlapping , but distinct route . Each route was repeated multiple times until the reward was moved to a different goal box . It is possible that the prospective differential firing observed in the overlapping portions of the routes did not reflect the intended goal location per se , but rather the position along one of four separate trajectories . In this view , differential firing reflects routes , as opposed to goals . This distinction between executing a series of responses and learning a goal location has a long tradition in spatial learning ( Tolman et al . , 1946; McCutchan , 1947; Restle , 1957 ) . To distinguish between goal and route accounts of differential firing , we designed a new apparatus in which different overlapping routes led to the same goal ( Figure 1A and B ) . If differential firing on the overlapping sections of different trajectories reflects the animal’s intended destination , then firing on the overlapping sections of the two routes leading to the same goal should be similar ( but should differ from firing on the routes leading to other goals ) . In contrast , if differential firing reflects position along a specific route , the firing on these two routes should differ ( Figure 1D ) . Our results support the latter interpretation . Place cells with fields on the overlapping portions of different routes leading to a common goal showed strong differential firing , and failed to show similar firing on different routes leading to the same goal . 10 . 7554/eLife . 15986 . 003Figure 1 . Maze apparatus with two routes leading to the same goal . ( A ) Top down view of the maze apparatus showing its layout including the start box , the three goal boxes , and the alleyways and choice points linking them . ( B ) The four trained routes through the maze . ( C ) Maze areas analysed for differential place cell firing . ( D ) Predictions of goal and route accounts of differential place cell firing . If differential firing of a place cells in the maze stem reflects the animal’s intended goal ( Prediction 1 - left plots ) , then a given cell should fire when the animals takes either the left or right route to the same goal . If such firing reflects the animal’s route ( Prediction 2 - right plots ) , firing should be seen on one route , but not the other . ( E ) Schematic of a representative daily session . Trials were blocked such that the same goal box was correct for at least 11 trials . The reward was then moved to a different goal box , and once it had been encountered by the rat , 11 further trials were run . In each session , all four routes were reinforced , although the order of these changed across sessions . DOI: http://dx . doi . org/10 . 7554/eLife . 15986 . 003
Over the course of the first 10 sessions prior to surgery , the number of errors the animals made after the first correct trial in each block of trials decreased significantly over sessions ( F ( 9 , 99 ) = 5 . 26 , p<0 . 001 , ηp2 = 0 . 32; Figure 2A solid line ) . The time the rats took to complete each trial once they had located the rewarded goal in a block of trials also decreased significantly across sessions ( F ( 9 , 99 ) = 6 . 87 , p<0 . 001 , ηp2 = 0 . 38; Figure 2B solid line ) . In contrast , neither the number of errors per session ( F ( 9 , 99 ) = 1 . 80 , p>0 . 05 ) , nor the time per trial ( F ( 9 , 99 ) = 1 . 95 , p>0 . 05 ) on trials prior to finding the rewarded goal box in each block changed significantly over sessions ( Figure 2A and B respectively , dashed lines ) . Together , these results suggest that the rats could remember the last rewarded goal and learned to apply the win-stay , lose-shift rule , but that their efficiency in searching for the new goal location at the beginning of each block did not improve significantly across sessions . 10 . 7554/eLife . 15986 . 004Figure 2 . Acquisition of the win-stay task . ( A ) The mean number of errors preceding the identification of the reinforced goal box in each block of trials ( broken line ) did not change significantly across training sessions . However , the number of errors following the identification of the reinforced goal box in each block of trials ( solid line ) decreased significantly across training . ( B ) The mean time taken to complete each trial preceding the identification of the reinforced goal box ( broken line ) did not change significantly across training sessions . However , the time taken following the identification of the reinforced goal box ( solid line ) decreased significantly . Error bars depict SEM . ( C ) Mean total number of errors summed across 10 training sessions on trials in which the centre goal box was rewarded ( black ) and on trials in which the Left and Right Goal Boxes were rewarded ( white ) , after the rewarded goal box had been identified in a block of trials . Rats made significantly more errors on trials when the two routes to the Centre Goal Box were rewarded than on trials when the two routes leading to the Left and Right Goal Boxes were rewarded . ( D ) Mean total number of errors on each session for trials on which the Centre Goal Box ( black ) and the Left and Right Goal Boxes ( white ) were rewarded . ( E ) Mean number of errors summed across 10 training sessions broken down by the nature of the error . For example , the first bar shows the average number of times the rats incorrectly chose Route 1 when Route 2 was rewarded , plus the number of times they chose Route 2 when Route 1 was rewarded . The number of confusion errors between routes to different goal boxes ( hollow bars ) was similar , regardless of route combination . However , there were significantly more confusion errors for the two routes to Centre Goal Box ( filled bar ) . Error bars depict SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 15986 . 004 Consistent with this , the rats navigated from the start box to a goal box significantly faster once they were aware of the reward location within each block of trials ( F ( 1 , 11 ) = 25 . 83 , p<0 . 001 , ηp2 = 0 . 70 ) . Across all sessions , the rats averaged 9 . 08 s ( S . D . = 6 . 19 s ) to travel from the start box to the end of the maze on trials before they had identified the goal box which contained reward . Once the rewarded goal box had been visited , travel time on later trials in that block decreased to 5 . 38 s ( S . D . = 3 . 48 s ) . During training the animals made a greater number of errors on the trials when Routes 2 or 3 to the Centre Goal Box were rewarded than on trials when the outer routes ( Routes 1 or 4 ) to the Left and Right Goal Boxes , respectively , were rewarded ( t ( 9 ) = 4 . 53 , p<0 . 005 , paired t-test , see Figure 2C ) . This difference decreased across the training period ( inner/outer goal x session interaction: F ( 9 , 99 ) = 2 . 99 , p<0 . 005 , ηp2 = 0 . 21; Figure 2D ) . We sought to define the nature of the errors which rats made after finding the location of the food reward in each block . An error where the rat took Route 1 to the Left Goal Box when Route 2 to the Centre Goal Box was rewarded can be interpreted as a similar form of navigation error as taking Route 2 to the Centre Goal Box when Route 1 to the Left Goal Box was rewarded . Both results may reflect an inability to discriminate between those two reward locations or routes . Figure 2E shows the distribution of post-reward errors when grouped into the six possible pairs of these confusion errors . From this figure it is clear that the rats made more errors between the two routes to the same goal ( Routes 2 and 3 to the Centre Goal ) as opposed to any other route pairs ( F ( 5 , 55 ) = 11 . 75 , p<0 . 001 , ηp2 = 0 . 52 ) . Post-hoc multiple comparison tests confirmed Routes 2 and 3 were confused more than any other route pair ( p<0 . 05 in all cases , with Sidak correction ) . Electrode tracks confirmed placement of the electrodes in the CA1 region of the HPC ( Figure 9 ) . The final electrode placement for one of the eight animals was less obvious due to tissue damage near the implant site . For this animal , part of the electrode track was seen in the cortex above the HPC and appeared to have descended at the correct ML and AP coordinates , and complex spikes and theta oscillations were observed . The neurons recorded from this animal which passed our criteria for place cells in the maze apparatus did not differ from the other animals in terms of isolation distance , l-ratio , average firing rate , maximum firing rate ( found in the firing rate map ) , width of waveform or spatial information content ( p>0 . 05 in all cases , Exact Kolmogorov-Smirnov tests ) . Thus , this rat’s data were included in the analyses above . 10 . 7554/eLife . 15986 . 016Figure 9 . Histological confirmation of electrode placement . ( A ) Coronal section of hippocampus with electrode track ( inset: higher magnification view ) . ( B ) Schematic of individual electrode tracks towards the CA1 cell layer of the hippocampus . Arrows represent the angle and depth of implantation with the arrow tip showing the point at which the electrode passed through the CA1 cell layer . No electrodes contacted lower cell layers . Each arrow is labelled with a rat number and an estimated anterior-posterior ( AP ) coordinate ( the schematic shows a slice at an AP -3 . 48 mm from bregma - the intended coordinate ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15986 . 016
Previous studies have shown that when a well-trained rat runs through the common segment of a maze on its way to or from different destinations , clear modulation of hippocampal place cell firing rates is observed ( e . g . , Wood et al . , 2000; Frank et al . , 2000; Ferbinteanu and Shapiro , 2003; Bower et al . , 2005; Ji and Wilson , 2008; Pastalkova et al . , 2008; Ferbinteanu et al . , 2011; Catanese et al . , 2014; Ito et al . , 2015 ) . In many cases , however , it is unclear what this differential firing represents . In a continuous maze , differential firing appears to represent previous locations early in the common stem of the maze , and intended destinations later in the common stem ( Catanese et al . , 2014 ) . In discrete trial tasks , where the animal is picked up and returned to a start box on each trial , the finding that place cells are active on journeys to one goal box and not active for journeys to other goal boxes has been interpreted as an encoding of the animal’s intended destination ( Ainge et al . , 2007 , 2012 ) . However , an alternative explanation is also plausible . In the Ainge et al . ( 2007 ) experiment , as in other studies , animals were well-trained . The rats ran rapidly to the rewarded goal box with little hesitation along the route . Thus the modulation of place cell firing observed in this task might not represent the intended destination per se , but rather a read-out of a specific trajectory sequence ( as in Pastalkova et al . [2008] ) . To differentiate between these possibilities , we tested animals on a task where two different routes led to the same goal location . Differential place cell firing on the overlapping portions of these routes would suggest that such activity encodes specific routes . Similar firing on the two routes would be consistent with the encoding of a common intended destination . Our results demonstrate that differential firing is associated with the animal’s route and not with its final destination . Of the cells with differential firing in the start box and central stem which fired preferentially on trajectories to the Centre Goal Box , nearly 96% showed differential firing between the two potential routes . Furthermore , at an ensemble level , the firing rate of place cells in the start box or central stem was sufficient to decode the animal’s route at an above chance level , regardless of which route was taken . This indicates that the ensemble firing of place cells carries sufficient information to distinguish the two central routes to the shared goal location and is thus also route-dependent , and not goal-dependent . A potential caveat to these findings is that we may have biased the rats’ representation towards routes by blocking one of the possible routes to the Centre Goal Box . We attempted to address this by training a second , naïve group of rats on the same task but without the transparent goal barrier . However , the majority of these animals formed a strong bias for one of the two routes to the Centre Goal Box . Thus , given the choice , the rats appeared to minimize the number of trajectories to learn , suggesting that even in the absence of a barrier , rats choose to solve the task in terms of routes , not goals . Similar navigation has been suggested in baboons , who simplify a complex jungle environment to a limited number of favourite routes ( Byrne et al . , 2000 ) . The potential biasing of place cell representations is consistent with previous reports of differential place cell firing with respect to deprivation state ( i . e . , hunger or thirst ) ( Kennedy and Shapiro , 2009 ) , spatial strategy ( Ferbinteanu et al . , 2011 ) , type of reward ( Allen et al . , 2012 ) , or order of non-spatial ( olfactory ) stimuli ( Allen et al . , 2016 ) . Though route encoding is predominant in the current study , it is certainly possible that different task contingencies might yield a significant representation of individual goals . Recent work by Ito et al . ( 2015 ) has suggested that trajectory-dependent cells are found in the nucleus reuniens , an input to CA1 , and in the medial prefrontal cortex . They show that the nucleus reuniens appears necessary for the differential firing in CA1 place cells . Together with the current results , this finding implies that trajectory encoding may arise in the medial prefrontal cortex , and may be passed through the nucleus reuniens to the CA1 layer of the hippocampus . The current results , like those of Ito et al . ( 2015 ) , deal only with prospective place cell firing . Findings from ( Ferbinteanu and Shapiro , 2003 ) indicate that retrospective firing may not reflect trajectories . They trained rats on a plus maze task , and showed that even when the animal took indirect trajectories to a goal , differential retrospective firing was observed . That is , retrospective place cell firing depended on which maze arm the rat started from , regardless of its subsequent trajectory . A likely possibility is that the hippocampus represents both where the animal is going , and where it has been . Such an account is consistent with recent demonstrations of prospective firing ( e . g . , Pfeiffer and Foster , 2013 ) , and with earlier lesion studies using retrospective homing tasks ( Gorny et al . , 2002; Wallace and Whishaw , 2003 ) . A second main finding in the current experiment was that the representation of the maze environment by hippocampal place fields was non-uniform . This occurred in two domains: the distribution of place fields on the maze , and the distribution between routes . In the current experiment , we observed a linear decrease in the frequency of active place cells from the start box to the goal locations . In previous work , place cells appear to over-represent goal locations ( Markus et al . , 1995; Hollup et al . , 2001; Hölscher et al . , 2003; Kobayashi et al . , 2003; Hok et al . , 2007; Dupret et al . , 2010 ) . However , in previous studies with a double-Y or continuous-T maze , over-representation of the start areas of the maze has also been observed ( Ainge et al . , 2007 , 2012 ) . One possibility is that in mazes where rats run overlapping routes but represent these independently , over-representation is expected for the common segments of the maze . A second , intriguing possibility is that the dorsal hippocampus represents distance to a goal , and this is more apparent in structured tasks with constrained routes . Recent findings suggest that in humans , the posterior hippocampus - which corresponds to the rodent dorsal hippocampus - exhibits more activation the farther one is from a navigational goal ( Howard et al . , 2014 ) . The second type of over-representation observed was a larger number of route-dependent cells coding specifically for the two routes to the Centre Goal Box than for routes to the two outer goal boxes . In learning the task , rats made significantly more errors when they were navigating these routes and they confused these two routes more than any other route pair . Over-representation may be the result of the recruitment of additional neural resources in the face of a difficult discrimination . The hippocampus has been implicated in the discrimination of structurally similar spatial environments ( Sanderson et al . , 2006; Aggleton and Pearce , 2001 ) , and it is possible that the similarity of two central routes requires greater hippocampal resources to discriminate , yielding an increase in route specific firing . The current study makes three contributions . First , we show that the strong modulation of place fields when rats run through a common location on their way to different destinations reflects the encoding a specific route or trajectory , and not the encoding of an intended goal per se . Second , we show that routes leading to the same goal are more difficult to discriminate than routes leading to different goals . Finally , we found that place fields over-represent the early portions of the maze , and difficult-to-discriminate routes . These results suggest that although the hippocampus represents places , the firing of place cells can also represent well-learned routes . It is possible that this representation , coupled with increased activity further from a goal , allow the animal to determine the distance and the route to a goal .
For the behavioural portion of the experiment 12 male Lister hooded rats , with an average weight of 300 g , were used as subjects . Eight of these animals were subsequently used in the electrophysiological portion of the experiment at which point they weighed approximately 400–450 g . A further eight naïve animals , with an average weight of 300 g , were used to test an alternative training protocol . All animals were housed in groups of four in standard cages , but housed individually in custom designed cages after surgery . The animals were maintained under a 12 hr light/dark cycle and testing was performed during the light phase of this cycle . Throughout testing , rats were food restricted such that they maintained approximately 90% ( and not less than 80% ) of their free-feeding weight . This experiment complied with the national [Animals ( Scientific Procedures ) Act , 1986 , United Kingdom] and international [European Communities Council Directive of November 24 , 1986 ( 86/609/EEC ) ] legislation governing the maintenance of laboratory animals and their use in scientific experiments . Local ethical approval was granted by the University of Edinburgh Animal Welfare and Ethical Review Board . Microdrives were based on a modified tripod design described previously ( Kubie , 1984 ) . The drives were comprised of eight tetrodes , each of which was composed of four HML coated , 17 µm , 90% platinum 10% iridium wires ( California Fine Wire , Grover Beach , CA ) . Tetrodes were threaded through a thin-walled stainless steel cannula ( 23 Gauge Hypodermic Tube , Small Parts Inc , Miramar , FL ) . The day before surgery and again immediately before surgery the tip of every electrode was gold plated ( Non-Cyanide Gold Plating Solution , Neuralynx , MT ) in order to reduce the impedance of the wire from a resting impedance of 0 . 7–0 . 9 MΩ to a plated impedance in the range of 200–300 kΩ ( 250 kΩ being the target impedance ) . Electrodes were implanted using standard stereotaxic procedures under isoflurane anaesthesia . Hydration was maintained by subcutaneous administration of 2 . 5 ml 5% glucose and 1 ml 0 . 9% saline . Animals were also given an anti-inflammatory analgesia ( small animal Carprofen/Rimadyl , Pfizer Ltd . , UK ) subcutaneously . Electrodes were lowered to just above the CA1 cell layer of the hippocampus ( -3 . 5 mm AP from bregma , +2 . 4 mm ML from the midline , −1 . 7 mm DV from dura surface ) . The drive assembly was anchored to the skull screws and bone surface using dental cement . Animals were given at least two hours recovery in their home cage , heated to body temperature . Following this , at least one week of recovery time passed before animals were screened for cells . During this week , the animals’ food was tapered from free feeding to the pre-surgery level of restriction . Single unit activity was observed and recorded using a 32-channel Axona USB system ( Axona Ltd . , St . Albans , UK ) . Mill-Max connectors built into the rat’s microdrive were attached to the recording system via two unity gain buffer amplifiers and a light , flexible , elasticated recording cable . The recording cable passed signals through a ceiling mounted slip-ring commutator ( Dragonfly Research and Development Inc . , Ridgeley , West Virginia ) to a pre-amplifier where they were amplified 1000 times . The signal was then passed to a system unit; for single unit recording the signal was band-pass ( Butterworth ) filtered between 300 and 7000 Hz . Signals were digitized at 48 kHz and could be further amplified 10–40 times at the experimenter’s discretion . The position of the animal was recorded using infra-red LEDs fixed to the unity gain amplifiers attached to the rat’s microdrive . A ceiling mounted , infrared sensitive CCTV camera tracked the animal’s position . Rats were screened for single unit activity and for the presence of theta oscillations once or twice a day , five days a week . The maze environment was constructed from wood and consisted of seven octagonal enclosures ( 25 × 25 cm with 25 cm high walls ) , which formed the start box , the three choice points , and the three goal boxes ( Figure 1A ) . Seven wooden alleyways connected these enclosures; these were 20 cm long × 10 cm wide with 10 cm high walls . All alleyways and octagonal enclosures were painted blue and the maze was elevated 60 cm from the floor on wooden stools . A moveable wooden barrier ( 10 cm wide 25 cm tall ) could be placed at the exit of the start box to confine the rat to the start box . A moveable transparent Perspex barrier ( also 10 × 25 cm ) could be used at one or other entrance to the central goal box to prevent the animal entering the goal box via that entrance . The maze was curtained off from the remainder of the room on the left with a large white sheet . On the right wall there was a large window blackout shutter , and on the wall in front of the maze was an upwardly directed light source . To add to the distinctiveness of the goal boxes within the maze , each contained a different object: a small grey elephant statue , a small white opaque bottle with cork stopper , and a small black and red box with slanted lid . Each goal box also had a Latin alphabet character in a different reflective colour affixed to the wall . Heavy ceramic reward dishes were placed in each goal box , directly beneath these reflective letters . The start box did not contain any objects , but an orange fluorescent star was affixed to the block which kept the animal from entering the alleyways . The square open field environment ( 100 × 100 cm with 25 cm high walls ) was also constructed of wood and painted black . When in use ( for screening for place cells ) , this box was placed on top of the maze apparatus , and so was elevated 85 cm above the floor , and was surrounded by the same set of distal cues . The current task was similar to that employed in previous experiments using a double-Y maze ( Ainge et al . , 2007 ) . A trial started when the experimenter raised the wooden block holding the animal in the start box at the base of the maze . The animal then navigated through the maze to one of the three goal boxes . On any given trial , only one goal box contained reward . Rats were not permitted to return towards the start box during a trial . If the rat entered the rewarded goal box , a correct choice was scored , and it was allowed to eat the food reward ( CocoPops , Kelloggs , Warrington , UK ) for a minimum of three seconds . The rat was then lifted by the experimenter , placed back into the start box and allowed to finish consuming any carried food reward . If the rat entered an unrewarded goal box , an incorrect choice was scored , and the animal was returned to the start box and held there using the wooden barrier for a minimum of three seconds . For the Centre Goal Box , one of the two entrances was blocked with a piece of clear Perspex positioned in the doorway . Thus , on trials in which the Centre Goal Box was baited , only one of the two paths from the start box to this goal box ( Routes 2 or Route 3 ) allowed access to the reward . For trials where the Left or Right Goal Boxes were reinforced , the transparent barrier was present , but was placed on the entry to the Centre Goal Box furthest from the rewarded box ( i . e . , at the right entry to the Centre Box if the Left Goal Box was reinforced ) . This was to ensure that the choice at the final junction was between two open goal boxes . Rats completed trials as described above until they entered the correct goal box . They were then given a further 11 trials , and in these trials the same goal box was reinforced . At the end of this first block of trials , the food was moved to a new goal box and the process was repeated ( Figure 1E ) . We trained an additional eight naïve rats on the same task described above in the absence of the Perspex barrier to the middle goal box . In this case , either route to the Centre Goal Box allowed access to the reward . As before , reward was available in one goal box for a block of trials , and the Centre Goal Box was reinforced for two blocks . We hoped that the rats would sample both routes at a roughly equal frequency without needing to direct their behaviour within blocks of trials to the Centre Goal Box . These rats were trained for a total of 12 sessions . Unfortunately , all of the rats rapidly developed a preference for only one route to Centre Goal Box ( mean 71% of trials via preferred route ) , and thus this variant of the task was unsuitable for assessing differences in place cell activity between routes and goals . After recovery from surgery , rats were screened daily for place cells in the open field apparatus . Upon the identification of place cells , an uninterrupted recording session was conducted . After a 10–15 min long recording session in the open field , rats were placed in the start box of the maze and the open field environment was removed . Rats were allowed a 60 s rest period in the start box before starting the win-stay , lose-shift task in the maze . The behavioural protocol during the maze phase was comparable to that used during pre-surgery training described above , with the exception that rats were required to make at least six correct trials within a block of trials before the reward location was changed . This was necessary to ensure adequate sampling of each trajectory . Between each trial , rats were confined to the start box for a minimum of six seconds ( session mean = 9 . 34 s , SD = 0 . 62 s ) in order to capture any possible differential firing which occurred there . At the end of the recording session , the animals were removed from the maze apparatus and the electrodes were lowered in order to maximise the chance of recording from a different population of cells on the following day . No attempt was made to track cells across days , and thus a subset of the cells may have been recorded on more than one session . Rats were tested until cells were no longer observed ( range 3–17 sessions ) . Single unit activity was analysed offline using a Matlab script that allowed the data files to be processed by the Klustakwik spike sorting program ( Kadir et al . , 2013 ) . The dimensionality of the waveform information was reduced to first principal component , energy , peak amplitude , peak time , and width of waveform . The energy of a signal x was defined as the sum of squared moduli given by the formula:εx≜ ∑n=0N−1| xn |2 Based on these parameters , Klustakwik spike sorting algorithms were then used to distinguish and isolate separate clusters . The clusters were then further checked and refined manually using the manual cluster cutting GUI , Klusters ( Hazan et al . , 2006 ) . As well as the previously mentioned features , manual cluster cutting also made use of spike auto- and cross-correlograms . Cluster quality was operationalised by calculating isolation distance ( Iso-D ) , Lratio , signal to noise ratio ( S/N ) and peak waveform amplitude , taken as the highest amplitude reached by the four mean cluster waveforms . For cluster C , containing nc spikes , Iso-D is defined as the squared Mahalanobis distance of the nc-th closest non-c spike to the centre of C . The squared Mahalanobis distance was calculated as:Di , C2= ( xi−μC ) T∑C−1 ( xi−μC ) where xi is the vector containing features for spike i , and µc is the mean feature vector for cluster C . A higher value indicates better isolation from non-cluster spikes ( Schmitzer-Torbert et al . , 2005 ) . The L quantity was defined as:L ( c ) =∑i∉C1−CDFxdf2 ( Di , C2 ) where i∉C is the set of spikes which are not members of the cluster and CDFxdf2 is the cumulative distribution function of the distribution with 8 degrees of freedom . The cluster quality measure , Lratio was thus defined as L divided by the total number of spikes in the cluster ( Schmitzer-Torbert and Redish , 2004 ) . As the signal and noise are both measured across the same impedance signal to noise ratio ( S/N ) was defined as:Signal to Noise Ratio= ( ( √μsignal ) 2 ( √μnoise ) 2 ) 2 where µ is the mean of the waveform amplitude . For noise we used the noise cluster which accompanied unit spikes on that tetrode . All four quality measures were assessed for their potential impact on our analyses by comparing the values observed in differential cells to non-differential place cells and by assessing the relationship between these measures and rANCOVA F-statistic ( as suggested by Schmitzer-Torbert et al . , 2005 ) . A cluster was classified as a place cell on the maze if it satisfied the following criteria: i ) the width of the waveform was >250 μs , ii ) the mean firing rate on the maze was greater than 0 . 1 Hz but less than 5Hz ( see Figure 6A ) and iii ) the spatial information content was greater than 0 . 5 b/s . Spatial information content is given by the equation:Information content= ∑Pi ( Ri/R ) log2 ( Ri/R ) where i is the bin number , Pi is the probability for occupancy of bin i , Ri is the mean firing rate for bin i , and R is the overall average firing rate ( Skaggs et al . , 1993 ) . Firing rate maps were used to quantify the number of distinguishable place fields on the maze . The rate maps were generated using an algorithm described by the following equations . The Gaussian kernel used is given by:g ( x ) = exp ( −x22 ) The algorithm for calculating firing rate is then given by:λ ( x ) = ∑i=1ng ( si−xh ) /∫0Tg ( y ( t ) − xh ) dt where Si represents the positions of every recorded spike , x is the centre of the current bin , the period [0 T] is the recording session time period , y ( t ) is the position of the rat at time t , and h is a smoothing factor , which was set to 2 . 5 cm . Bins in which the rat did not explore within 5 cm of the centre were regarded as having never being visited . If rats were aware that both Routes 2 and 3 led to the same goal box we would expect place cells to represent this location similarly regardless of the route taken there . Furthermore , we would expect higher correlations between the place cell activity in the Centre Goal Box when accessed via Route 2 and Route 3 , than between activity in other pairs of boxes ( e . g . the Left Box accessed via Route 1 and the Centre Goal Box when accessed by Route 2 ) , and also that the activity in the Centre Goal Box when accessed by the two routes would more highly correlated than expected by chance . To test this we generated a population vector , consisting of the firing rates of all place cells from all animals and sessions in the Centre Goal Box when the animals navigated using Route 2 . We calculated the Spearman’s ranked correlation between this vector and the population vector for the Centre Goal Box when animals navigated using Route 3 . Lastly we calculated the ranked correlation between these two vectors and vectors for the Left and Right Goal Boxes; each of these boxes was represented by only one vector as there was only one possible route to each . This analysis provides six correlation values representing the comparison of the goal box activity at the end of each route to every other one . Next we calculated the probability that each of these values could occur by chance . To do this we calculated the ranked correlation between each of the population vectors described above with a novel vector composed of random firing rate values from each cell ( in effect shuffling route identity , whilst maintaining the order of the cells ) after removing those firing rates belonging to the original population vector . This process was repeated 10000 times and the probability of the original correlation value occurring by chance was then estimated as the percentile position of that value in the distribution of correlation values resulting from the shuffled correlations ( using a kernel smoothed cumulative density function ) . This analysis tells us the probability of any two population vectors being similar by chance or in other words the likelihood that all place cells would fire similarly in two goal boxes by chance . To assess whether place fields in areas of the maze that were traversed on more than one route were modulated by the route and/or by the goal , we focussed on activity occurring in four segments of the maze: the start box and the initial corridor or central stem ( which were each common to all four routes and all three goals ) and the right and left stems ( each common to two routes and two goals ) ( Figure 1C ) . The analysis for each segment was conducted only for neurons that had been identified as place cells in the maze environment , and that were active in that segment ( active being defined as a mean firing >1 Hz in the maze segment on at least one of the four ( or two ) possible routes when all of the individual trajectories along one route were combined ) . For each place cell that was active in a given segment of the maze , the firing rate in that segment was calculated for each trial ( total number of spikes in segment/time in segment ) , together with the average x- and y- coordinates occupied by the animal for each trial , and the average velocity of the animal in that segment ( total distance travelled in segment/total time spent there ) for each trial . We then assessed whether firing rate differed between trials on which the animal had taken the four different routes using three different methods . See Figure 3—figure supplement 1 for an example of the parameters used in the following analyses . Method one , reported in the main text , used a nonparametric ANCOVA described previously ( Quade , 1967 ) . Briefly , this test consists of replacing the dependent variable ( DV ) and covariates ( COV ) with ranked equivalents . A linear regression is then performed on these ranked covariates against the DV , ignoring the independent variable ( IV ) . A one-way ANOVA is then performed on the unstandardized residuals resulting from this regression against the original IV . If this was found to be significantly modulated by the animal’s route ( p<0 . 05 ) after controlling for the effects of the covariates , we conducted planned post-hoc tests . These consisted of six pairwise comparisons of the estimated marginal means between the four routes . This method was carried out in Matlab using the functions tiedrank for ranking , fitlm for the linear regression , anova1 for the one way ANOVA and multcompare for the post hoc tests . Method two employed a similarly nonparametric form of the ANCOVA test . We conducted an ANCOVA on the firing rate data using the same DV , IV and COV as above . We then compared the F-statistic obtained from this test to a distribution of F-statistics obtained after randomly shuffling the DV and repeating the ANCOVA . We then computed a p-value by the following method:p= #Fshuff≥ Fobsk where p is the probability that the observed F-value could have been obtained by chance , Fshuff is the distribution of F-values obtained by shuffling the DV , Fobs is the F-value obtained when testing the unshuffled DV and k is the number of shuffles conducted . For our tests we conducted 5000 shuffles . Method three employed a Generalized Linear Model ( GLM ) approach instead of an ANCOVA . We conducted a linear GLM on the DV , IV and COV described above with the underlying distribution assumed to be Poisson ( Di Lorenzo and Victor , 2013 ) and a log link function . If firing rate was found to be significantly modulated by the animal’s route ( p<0 . 05 ) after controlling for the effects of the covariates , we conducted planned post-hoc tests . These consisted of six Mann-Whitney U tests comparing firing rates for every possible combination of the four routes . This method was carried out in Matlab using the functions fitglm to build the GLM and ranksum for the post hoc tests . As an additional means of assessing the route- versus goal-related place cell firing , we tested whether the animal’s trajectory could be derived from the ensemble activity of recorded place cells . If place cells show route-dependent activity at the ensemble level , then we would expect an automated route determination method to be equally accurate for all four routes , as each would be discriminable using the ensemble . In contrast , if ensembles reflect goal anticipation , then route determination should be significantly less accurate for the two central routes , as both lead to the same goal . Ensemble analyses were conducted separately for two segments of the maze: the start box and central stem . The animals traversed each of these zones on all trials and the trajectories within them should have remained similar regardless of which route the animals were taking . For each session and for each of these maze segments we compared the firing of all place cells on an individual trajectory ( i . e . Route 1 to the Left Goal Box ) to the average firing of these cells across all trajectories to each goal . In this way , we compared population vectors for every route to four , average firing rate , population vectors . However , the single trajectory being compared was not included in the calculation of its average goal vector , removing the possibility that it influenced the outcome of the assessment . To assess the similarity of each route population vector to each of the four goal vectors we used a cosine distance or cosine similarity measure defined as:CosSim= ∑ixiyi∑ixi2∑iyi2= ⟨x , y⟩∥x∥∥y∥ This calculation gives a value bounded between 0 and 1 if x and y are non-negative ( such as firing rates ) . Cosine similarity can be interpreted as the cosine of the angle between two vectors , or as an alternative to the Pearson correlation that is sensitive to shifts in group values ( i . e . If x is shifted to x + 1 , the cosine similarity between x and y changes ) . We next calculated the proportion of these comparisons which resulted in a ‘correct’ match between route vector and its corresponding goal vector and the proportion of those which suggested a similarity to one of the other goal vectors ( a match was taken as the goal vector resulting in the highest similarity score ) . This was repeated for every session included in the analysis . In order to calculate the probability of correct matches being made by chance , we also repeated the above process 10000 times using four goal vectors where the identity of the route ( but not of the contributing neuron ) were shuffled , therefore disrupting any relationship between firing rate and the animal's trajectory . The probability that the proportion of correct matches made in our unshuffled analysis was the result of chance was estimated by calculating the percentile position of our observed proportion of matches in the distribution observed in the shuffled data . We did this using a kernel smoothed cumulative density function - Matlab function ksdensity . An Epanechnikov kernel was employed as it is one of the most widely used , optimal filters and we set bandwidth to the ‘default’ mode for all distributions . A brief outline of this process can be seen in Figure 7 . We reasoned that if the central routes were represented more similarly ( due to goal location dependent firing at an ensemble level ) then we would expect an above chance level of matches between trajectories along Route 2 and the goal vector calculated for Route 3 ( and vice versa ) or we may expect fewer correct matches for the central routes than for the outer ones . At the end of the experiment animals were given an overdose of pentobarbital intraperitoneally ( Euthatal , Merial Animal Health Ltd . , Essex , UK ) , and perfused with 0 . 9% saline solution followed by a 4% formalin solution . The brain was extracted and stored in 4% formalin for at least seven days prior to any histological analyses . The brains were sliced in 32 µm sections on a freezing microtome at −20° . These sections were stained with a 0 . 1% cresyl violet solution and the slice best representing the electrode track was then imaged using ImageJ software ( ImageJ , NIH , Bethesda ) . | How does the brain represent the outside world ? One way of answering this question is to study the brains of rats , because the basic plan of a rodent’s brain is similar to that of other mammals , such as humans . For example , the brains of rodents and humans both contain a structure called the hippocampus , which plays important roles in navigation and spatial memory . Cells within the hippocampus called place cells support these processes by firing electrical impulses whenever the animal occupies a specific location . When a rat runs along a corridor in a maze , its place cells often fire as it approaches a choice point . A given place cell will typically fire before the rat chooses a path leading towards one particular location , but not before choices that lead to other locations . The firing that occurs prior to the choice point is termed “prospective firing” . However , it is not known whether the prospective firing of place cells represents the rat’s final destination , or the specific route the animal takes to get there . To address this question , Grieves et al . designed a maze in which two different paths from a starting corridor led to the same goal location . If place cells represent the goal location , they should fire whichever route the rat chooses . However , if they represent the specific path the rat takes to the goal , they should fire on one or the other route , but not both . Grieves et al . found that almost all place cells with prospective activity in the starting corridor fired on a single route , as opposed to firing on both routes to the common goal . This suggests that the prospective firing in the hippocampus reflects the route the animal will take , rather than its intended destination . A future challenge will be to understand how the way the hippocampus codes routes interacts with brain circuits that code for intended goals , and how the activity of these circuits influences the animal’s ability to navigate . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2016 | Place cells on a maze encode routes rather than destinations |
Embryonal Rhabdomyosarcoma ( ERMS ) and Undifferentiated Pleomorphic Sarcoma ( UPS ) are distinct sarcoma subtypes . Here we investigate the relevance of the satellite cell ( SC ) niche in sarcoma development by using Hepatocyte Growth Factor ( HGF ) to perturb the niche microenvironment . In a Pax7 wild type background , HGF stimulation mainly causes ERMS that originate from satellite cells following a process of multistep progression . Conversely , in a Pax7 null genotype ERMS incidence drops , while UPS becomes the most frequent subtype . Murine EfRMS display genetic heterogeneity similar to their human counterpart . Altogether , our data demonstrate that selective perturbation of the SC niche results in distinct sarcoma subtypes in a Pax7 lineage-dependent manner , and define a critical role for the Met axis in sarcoma initiation . Finally , our results provide a rationale for the use of combination therapy , tailored on specific amplifications and activated signaling pathways , to minimize resistance emerging from sarcomas heterogeneity .
Rhabdomyosarcoma ( RMS ) , the most common soft tissue sarcoma of childhood , is a rare but aggressive malignancy ( Hawkins et al . , 2013; Saab et al . , 2011 ) . Rhabdomyoblasts are positive for markers of muscle stem cells ( satellite cells ) such as Pax7 ( Tiffin et al . , 2003 ) , myoblasts ( MyoD and Myogenin ) and differentiated skeletal muscle ( Desmin and Myosin Heavy Chain MHC ) ( Merlino and Helman , 1999; Morotti et al . , 2006 ) . Thus , the identification of the RMS cell of origin is still debated and remains elusive . On the other hand undifferentiated pleomorphic sarcoma ( UPS ) , which is one of the most common subtypes of adult soft tissue sarcoma , is characterized by a lack of tissue-specific differentiation markers . UPS may originate from cells of different lineages that converge into a common undifferentiated histological presentation , or may derive by a process of de-differentiation from more committed sarcomas ( Weiss and John , 2007 ) . Histopathological classification of RMS includes two major subgroups , the alveolar ( ARMS ) and the embryonal ( ERMS ) subtype . ARMS are 20% of all newly diagnosed RMS . Although rarer , they are often metastatic at diagnosis and have a poor prognosis . 80% of ARMS are characterized by the PAX3/7-FOXO1 chromosomal translocation ( Mercado and Barr , 2007 ) . The remaining ones , negative for the translocation , are indistinguishable from ERMS at the molecular level ( Williamson et al . , 2010 ) . In the last decades standard treatments including radiotherapy , chemotherapy and surgery , have not significantly improved RMS and UPS patient survival . Thus , novel effective precision-based therapeutic approaches are required . While the mutational status of UPS has been only sporadically analyzed ( Li et al . , 2015 ) , the comprehensive genomic and epigenetic landscape of RMS tumors was recently described ( Chen et al . , 2013; Seki et al . , 2015; Shern et al . , 2014 ) . These studies highlight the difference between ARMS and ERMS in terms of mutational load . While ARMS carry only a few genetic lesions in addition to the pathognomonic ones , the ERMS subtype is highly heterogeneous , with recurrent mutations/copy number variations in genes coding for tyrosine kinase receptors ( RTKs ) and their downstream effectors ( RAS and PIK3CA ) . The early onset of ERMS , concomitant with a period of intense muscle growth and their positivity for Pax7 , suggest that the muscle stem cells ( satellite cells , SCs ) could be at the origin of this subtype . Quiescent SCs express the Met receptor ( Allen et al . , 1995 ) and are found adherent to the muscle fibers in a specialized sublaminar microenvironment called SCs niche . The niche microenvironment controls their fate by orchestrating the homeostatic balance between stem cell quiescence and activation . Upon injury , the niche releases HGF , which is one of the extrinsic signals involved in SC activation and proliferation ( Allen et al . , 1995; Tatsumi et al . , 1998; Thomas et al . , 2015 ) . Although others have linked injury with sarcoma also through activation of Met signaling ( Sharp et al . , 2002; Tremblay et al . , 2014; Van Mater et al . , 2015 ) , we here describe a unique model aimed at investigating the effect of HGF-mediated SC niche perturbation in sarcoma development and maintenance . Specifically , HGF expression was confined to the SC niche and could be temporally regulated by Doxycycline ( Dox ) . In a wild type background , HGF production promoted only limited activation of satellite cells , without inducing an overt phenotype . Conversely , in a Cdkn2a null background all mice developed sarcoma , 92% of which classified as ERMS and only 8% as UPS . Genetic ablation of the muscle stem cells ( obtained by moving the system in a Pax7 null background ) strongly influenced the sarcoma subtype . In this different genetic setting the majority of tumors were classified as UPS , suggesting that in the absence of satellite cells , fibroblasts resident in the SC niche were the more susceptible population to HGF-mediated perturbation . Finally , we investigated the relevance of novel therapeutic approaches using our preclinical model of sarcoma . The vast majority of tumors grew in a HGF/Met-independent manner and were genetically heterogeneous . Tumor cells were sensitive to Met inhibitors only in the rare cases harboring Met amplification , but the continuous treatment with a single agent resulted in selection and expansion of resistant clones . However , the use of a combination of drugs hitting different targets was effective in bypassing resistance . Altogether , our data show that perturbation of the SC niche with HGF can promote distinct sarcoma subtypes in a Pax7 lineage-dependent manner , thus offering a possible explanation of why ERMS and UPS are part of a tumor continuum . Finally , the use of our model for the preclinical assessment of targeted therapy revealed that combination , rather than single agent treatment , could be more effective in treating genetically heterogeneous sarcomas .
At variance with other sarcoma subtypes , a satellite cell-like signature is considered a hallmark of ERMS ( Hatley et al . , 2012; Rubin et al . , 2011 ) while the complete absence of tissue-specific markers in UPS suggests an origin from early mesenchymal precursors . By taking advantage of previously validated Met and satellite signatures ( Fukada et al . , 2007; Bertotti et al . , 2009; Pallafacchina et al . , 2010 ) we performed unsupervised clustering analysis of two large panels of human primary RMS ( Davicioni et al . , 2009; Williamson et al . , 2010 ) . We selected 5 clusters subdivisions to segregate normal muscle apart from RMS samples . Interestingly , in both datasets the Met signature was more effective than the satellite signature in distinguishing ERMS from ARMS ( Figure 1A ) . We determined the Met score , which is essentially the sum contribution of the expression values of the signature genes and is a measure of the level of Met activity in each sample . The Met score was higher in RMS compared to muscle ( Figure 1B ) and in ERMS rather than ARMS ( Figure 1C ) . The satellite score , measured in an analogous way , was also able to discriminate RMS from muscle , but did not consistently distinguish ERMS from ARMS ( Figure 1—figure supplement 1A , B ) . We then analyzed the top 25% of patients with high Met and satellite scores . Interestingly , only by using the Met score the first top quartile was enriched in ERMS samples ( Figure 1D ) . However , in the top quartiles there was a significant overlap of ERMS samples with both high Met and satellite scores ( Figure 1E ) . Finally , we used the Met signature and a fibroblast signature ( ad hoc calculated ) to analyze a large human dataset covering different sarcoma subtypes ( Gibault et al . , 2011 ) . Interestingly , the Met score was significantly enriched in UPS compared to other sarcomas , while the fibroblast signature distinguished UPS from RMS ( Figure 1F , G ) . 10 . 7554/eLife . 12116 . 003Figure 1 . Met and satellite signatures are both preferentially associated with the ERMS subtype , while UPS show high Met and fibroblast scores . ( A ) Unsupervised hierarchical clustering of RMS samples according to the Met or satellite signature genes; colors highlight RMS subtype enrichment in single clusters ( ARMSp: translocation positive; ARMSn: translocation negative ) . ( B ) Box-plot of the Met signature scores for RMS and muscles in the indicated datasets . ****p<0 . 0001 ( t test ) . ( C ) Box-plot of the Met signature scores for ERMS and ARMSp in the indicated datasets . **p<0 . 01; ****p<0 . 0001 ( t test ) . ( D ) Ranked distribution of RMS subtypes according to the indicated signature scores . The boxed area includes the top 25% samples . NSp>0 . 05; *p<0 . 05; ***p<0 . 001; ****p<0 . 0001 ( hypergeometric test ) . ( E ) Venn diagrams of ERMS included in the box in D , showing a significant intersection between high Met and high satellite signatures . *p<0 . 05; ****p<0 . 0001 ( hypergeometric test ) . ( F ) Box-plot of the Met signature scores for UPS and other sarcomas in the indicated dataset . ***p<0 . 001 ( t test ) . ( G ) Box-plot of the fibroblast signature scores for UPS and RMS in the indicated datasets . ****p<0 . 0001 ( t test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12116 . 00310 . 7554/eLife . 12116 . 004Figure 1—figure supplement 1 . Satellite signature analysis in human RMS datasets . ( A ) Box-plot of the satellite signature scores for RMS and muscles in the indicated datasets . ****p<0 . 0001 ( t test ) . ( B ) Box-plot of the satellite signature scores for ERMS and ARMSp in the indicated datasets . NSp>0 . 05; *p<0 . 05; ( t test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12116 . 004 Overall our bioinformatic analyses provide a basis to investigate the functional relationship between the HGF/Met axis and the SC niche in ERMS/UPS initiation and maintenance . To generate a model of ERMS based on SC niche perturbation ( herein defined MH ) we conditionally modulated transgenic Hgf and eGFP expression using the Muscle Creatine Kinase-driven tTA ( Ckm-Tet-Off ) promoter ( Figure 2A ) . MH mice , born at the expected Mendelian frequency , showed normal postnatal muscle growth ( Figure 2—figure supplement 1A , B ) . Expression of Hgf and eGFP was restricted to skeletal muscle and was Dox-dependent ( Figure 2B , C ) . While HGF was measurable in transgenic muscle , it was undetectable in blood ( Figure 2D ) . Muscle sections showed a modest reduction of fiber size ( Figure 2E , F , Figure 2—figure supplement 1C ) . However , Pax7 immunofluorescence revealed a higher number of satellite cells compared to control littermates ( Figure 2G , H ) . More Pax7-positive cells were also detected on freshly isolated single myofibers ( Figure 2I , J ) . Surprisingly , at time 0 some satellite cells were MyoD-positive , indicating basal activation ( Figure 2K , L ) . These differences were Dox-dependent ( Figure 2I , K ) and lasted for three days in culture ( Figure 2—figure supplement 1D , E ) . Furthermore , myofibers isolated from MH mice contained more myonuclei than controls ( Figure 2—figure supplement 1F ) . MyoD-positive cells were also detectable in transgenic muscle sections ( Figure 2—figure supplement 1G ) , indicating the presence of activated satellite cells also in vivo . 10 . 7554/eLife . 12116 . 005Figure 2 . SC niche perturbation results in satellite cells activation . ( A ) Schematic representation of the system used to generate Ckm-Tet-Off Hgf ( MH ) mice . In the absence of Doxycycline ( -Dox ) , tTA binds to the Tetracycline Responsive Element ( TRE ) , inducing the expression of HGF and eGFP in skeletal muscle . ( B ) Upper panel: fluorescent image of P10 mice . Middle panel: MH hindlimb under visible ( Vis ) and fluorescent ( Fluo ) light . Lower panel: fluorescent image of a muscle cross-section of a MH mouse . ( C ) Upper panel: semi-quantitative RT PCR of the indicated genes on Tibialis anterior muscles . Lower panel: fluorescent images of hindlimbs from a MH mouse with or without Dox treatment . ( D ) HGF-ELISA quantification in muscle and serum extracts from Dox-treated or control mice . ( E ) Distribution of Tibialis anterior cross-sectional areas showing a leftward shift in MH mice relative to their respective M controls . The mean value for MH mice was 585 ± 37 µm2 ( n=8 ) ; the mean value for control M mice was 844 ± 63 µm2 ( n=10 ) . **p<0 . 01 ( t test ) . ( F ) Representative H&E staining of Tibialis anterior cross sections . ( Scale bar = 100 µm ) . ( G ) Quantification ( mean ± SEM ) of Pax7 positive cells/field in Tibialis anterior sections ( M mice n=7; MH mice n=6 ) . ***p<0 . 001 ( t test ) . ( H ) Representative immunofluorescence of Pax7 ( green ) , Laminin ( red ) and DAPI ( blue ) staining in G . Arrowheads indicate Pax7-positive cells . ( Scale bar = 50 µm ) . ( I ) Quantification ( mean ± SEM ) of Pax7 positive cells/fiber after single fiber isolation ( M mice n=13; MH mice n=12; MH +Dox n=2 ) . *p<0 . 05; ***p<0 . 001 ( t test ) . ( J ) Representative immunofluorescence of Pax7 ( red ) and DAPI ( blue ) staining in I . ( K ) Quantification ( mean ± SEM ) of MyoD-positive cells/fiber after single fiber isolation ( M mice n=3; MH mice n=3; MH +Dox n=4 ) . **p<0 . 01; ***p<0 . 001 ( t test ) . ( L ) Representative immunofluorescence of MyoD ( red ) and DAPI ( blue ) staining in K . Dashed areas are shown at threefold magnification . DOI: http://dx . doi . org/10 . 7554/eLife . 12116 . 00510 . 7554/eLife . 12116 . 006Figure 2—figure supplement 1 . SC niche perturbation results in satellite cells activation . ( A ) Body weight ( mean ± SEM ) of the indicated mice ( M mice n=6; MH mice n=9 ) . NSp>0 . 05 ( t test ) . ( B ) Representative image of Tibialis anterior cross sections in the indicated mice . ( C ) Fiber size distribution ( mean ± SEM ) of Tibialis anterior showing a significant increase in the number of fibers with cross-sectional area of 300–800 μm2 and a significant decrease in the number of fibers with cross-sectional area of 1000–1300 μm2 in MH mice ( n=8 ) respective to control M mice ( n=10 ) . *p<0 . 05; **p<0 . 01; ***p<0 . 001 ( t test ) . ( D ) Quantification ( mean ± SEM ) of Pax7-positive cells/fiber after three days from single fiber isolation ( M mice n=6; MH mice n=5 ) . *p<0 . 05 ( t test ) . ( E ) Quantification ( mean ± SEM ) of MyoD-positive cells/fiber after three days from single fiber isolation ( M mice n=2; MH mice n=2 ) . *p<0 . 05 ( t test ) . ( F ) Quantification ( mean ± SEM ) of myonuclei/fiber after single fiber isolation ( M mice n=5; MH mice n=7 ) . **p<0 . 01 ( t test ) . ( G ) Representative immunofluorescence of MyoD ( red ) , Laminin ( green ) and DAPI ( blue ) staining . Arrowheads indicate MyoD-positive cells . ( Scale bar = 50 µm ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12116 . 006 Homozygous deletions of the CDKN2A locus has been reported in the majority of human tumors including sarcomas ( Miller et al . , 1997; Roussel , 1999; Ruas and Peters , 1998 ) . Specifically , CGH analysis revealed that more than 17% of UPS exhibited the loss of the locus , while up to 39% of the UPS did not express the corresponding p14 protein ( Pérot et al . , 2010 ) . The frequent inactivation of the locus has been confirmed by an additional study ( Simons et al . , 2000 ) reporting the deletion in 32% of UPS cases . Furthermore , the genetic inactivation of the CDKN2A locus in RMS has been extensively reported ( Chen et al . , 2007; Obana et al . , 2003; Paulson et al . , 2011; Seki et al . , 2015 ) . Thus we moved our model into a Cdkn2a ( alias Ink4a/Arf ) null genetic background ( herein defined MHI-null , Figure 3—figure supplement 1A ) . Approximately 30% of all MHI-null mice displayed an apparently hypertrophic phenotype ( Figure 3A–C ) . All of these mice quickly developed tumors and the histology revealed a picture of multistage progression ( Figure 3D , stage 1 to 4 ) . In the periphery of the tumor ( Figure 3D , stage 1 ) proliferating ( Ki67-positive ) satellite cells , surrounding old fibers , were mixed with more mature MyoD- and Myogenin-positive cells . These cells were intermingled with areas of neomyogenesis , including small fibers with central nuclei , positive for embryonal MHC ( eMHC ) . Immunofluorescence confirmed the presence of Desmin-positive centro-nucleated fibers contiguous to old fibers ( Figure 3E , left panel ) . At this early stage , Pax7-positive cells were detectable beneath the basal lamina of the newly formed fibers , suggesting that satellite cells retained the ability to return to quiescence ( Figure 3E , right panel ) . Closer to the tumor mass the expanded population of proliferating satellite cells appeared to take over , while the number of newly formed fibers was reduced ( Figure 3D , stages 2 and 3 ) . The central area of the tumor consisted of tissue with high cellularity and no further evidence of terminal myogenic differentiation ( Figure 3D , stage 4 ) . Accordingly , the tumor bulk was eGFP-negative indicating that the cells were undifferentiated ( Figure 3B , Figure 3—figure supplement 1B ) . Morphologically , small round cells were associated with elongated polygonal cells or with cells displaying pleomorphic nuclei ( Figure 3D , stage 4 ) . The histological markers ( Pax7 , MyoD and Myogenin , Figure 3D , stage 4 ) and the anatomical localization ( trunk , neck and occasionally the orbit ) ( Figure 3—figure supplement 1B ) , were compatible with the ERMS subtype . MHI-null mice developed tumors with a short latency ( 3 . 95 months ) , while mice heterozygous for the Cdkn2a locus ( MHI-het ) showed delayed tumor formation ( 6 months ) ( Figure 3—figure supplement 1C ) . PCR analysis revealed that the majority of the tumors in MHI-het mice had lost the wild type allele ( Figure 3—figure supplement 1D ) . To exclude the involvement of fetal myoblasts in ERMS development , MHI-null mice were maintained under Dox treatment until P10 . Given the persistence of Dox in the tissues , transgenic Hgf and eGFP expression became detectable from day 40 onward ( Figure 3—figure supplement 2A , B ) , when only resident muscle stem cells are present . In this cohort of mice , tumor development occurred with the previously described latency and incidence ( Figure 3—figure supplement 2C ) . 10 . 7554/eLife . 12116 . 007Figure 3 . SC niche perturbation in Cdkn2a-null mice results in a multi-step model of ERMS development . ( A ) Representative image of a control ( M ) and a hyperplastic mouse ( MHI-null ) . ( B ) Representative fluorescent images of MHI-null trunk muscles at different stages of progression: normal muscle , hyperplasia and ERMS . ( C ) Distribution of MHI-null mice developing ERMS with or without hyperplasia . ( D ) H&E staining and representative immunohistochemical analysis of MHI-null specimens collected at different stages of tumor progression ( from hyperplasia to ERMS , stage 1–4 ) . ( Scale bar = 100 µm ) . ( E ) Representative immunofluorescence on MHI-null hyperplastic muscle sections . Pax7 ( red ) , Laminin ( green ) , Desmin ( red ) and DAPI ( blue ) . ( Scale bar = 100 µm ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12116 . 00710 . 7554/eLife . 12116 . 008Figure 3—figure supplement 1 . SC niche perturbation in Cdkn2a-null mice results in sarcoma development . ( A ) Schematic representation of the system used to generate Ckm Hgf Cdkn2a-null ( MHI-null ) mice . The effect of Cdkn2a locus loss on Rb and p53 oncosuppressors is also depicted . ( B ) Representative images of tumor-bearing MHI-null mice under visible ( Vis ) and fluorescent ( Fluo ) light . Neoplastic lesions are marked with yellow dashed lines . ( C ) Kaplan-Meier curve showing the relative sarcoma-free survival of MHI-het and MHI-null mice treated or not with Dox . ( D ) PCR for Cdkn2a locus on genomic DNA extracted from normal muscles and sarcomas . DOI: http://dx . doi . org/10 . 7554/eLife . 12116 . 00810 . 7554/eLife . 12116 . 009Figure 3—figure supplement 2 . Evaluation of Hgf expression and mortality in MHI-null mice treated with Dox from conception to P10 . ( A ) Schematic representation of the procedure used to evaluate RMS formation in mice treated with Dox ( Hgf Off ) from conception to P10 . Muscle samples were collected from different mice at the indicated time . Green background corresponds to detectable GFP expression . ( B ) Real-time PCR analysis of transgenic Hgf expression in MHI-null muscle samples indicated in panel A and in control muscles . ( C ) Kaplan-Meier curve showing the survival of MHI-null mice treated with Dox from conception to P10 ( Ex Dox ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12116 . 009 It has been shown that ERMS and UPS are part of a tumor continuum in terms of histological presentation and expression profiling ( Rubin et al . , 2011 ) . Intriguingly , the SC niche , in addition to muscle stem cells , also contains resident muscle fibroblasts that are responsible for extracellular matrix production within skeletal muscle fibers ( Thomas et al . , 2015 ) . To assess the contribution of these two cell types to sarcomagenesis we moved the MHI-null mice in a Pax7 null genetic background . Pax7-deficient C57/B6 mice are viable at birth , but progressively die in 2–3 weeks ( Mansouri et al . , 1996; Seale et al . , 2000 ) . In these mice the number of satellite cells is reduced at birth , and further declines during postnatal development . However , in a mixed genetic background , between 5 and 10% of mutant animals survive into adulthood ( Oustanina et al . , 2004 ) . Thereby , we generated a cohort of MHI-null Pax7-null mice ( herein referred as MHl-null P-null ) ( Figure 4A ) sufficient to determine the effect of the Pax7 null background on sarcoma development . The mixed background had no effect on sarcoma incidence and latency ( Figure 4B ) . However MHI-null P-null mice showed a drastic reduction in ERMS incidence ( Figure 4C ) . The MHI-null P-null tumors were characterized by cells displaying a variegate morphology , that occasionally included the presence of visible cross-striations . Notably , immunohistochemistry revealed the positivity for MyoD and Myogenin in less than 30% of cases , which were classified as ERMS ( Figure 4C , D ) . Interestingly , the majority of tumors ( more than 70% ) presented extensive cellular heterogeneity from round to spindle-shape cells , with nuclear appearance ranging from hyperchromatic to pleomorphic . The complete absence of tissue-specific markers of differentiation categorized these tumors as UPS ( Figure 4C , D ) . The switch in sarcoma subtype is in line with the expansion of myofibroblasts observed in Pax7-null muscles upon exposure to a regenerative microenvironment ( Maltzahn et al . , 2013 ) . Accordingly , while both UPS and ERMS tumors expressed Met , only UPS were positive for the pan-fibroblast marker PDGFRα ( Figure 4D , E ) , and FACS analysis on freshly isolated tumors revealed that they were Sca1-positive and Integrin-α7-negative ( Figure 4F ) , indicating a mesenchymal-like phenotype ( Driskell et al . , 2013; Guarnerio et al . , 2015; Joe et al . , 2010 ) . In contrast , the ERMS were Sca1-negative and Integrin-α7-positive , confirming the myogenic identity of this subtype . Altogether our data clearly indicate that perturbation of the SC niche microenvironment can give rise to distinct sarcoma subtypes in a Pax7 lineage-dependent manner , and provide evidence for a fibroblast cell origin of UPS . 10 . 7554/eLife . 12116 . 010Figure 4 . Genetic ablation of Pax7 results mainly in UPS development . ( A ) Schematic representation of the system used to generate Ckm Hgf Cdkn2a-null Pax7-null ( MHI-null P-null ) mice . ( B ) Kaplan-Meier curve showing the relative sarcoma-free survival in the indicated genotypes . ( C ) Penetrance ( pie charts ) and histological distribution ( bar graphs ) of tumors . ( D ) Representative immunohistochemical analysis of MHI-null tumors in Pax7 heterozygous or null genetic background . UPS: undifferentiated pleomorphic sarcoma . ( Scale bar = 250 µm ) . ( E ) Western blot analysis of murine UPS and ERMS probed for the indicate proteins . ( F ) Sca1 and Int-α7 expression analysis in primary murine UPS and ERMS ( left panel , single staining; center and right panels , double staining ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12116 . 010 Genetically engineered animal models represent a robust preclinical platform to investigate the efficacy of novel treatments , among which targeted therapies and combination therapies are the most attractive ( Day et al . , 2015 ) . In our model transgenic Hgf comes from differentiated muscle fibers . Thus the tumor mass is no longer exposed to transgenic HGF for lack of differentiated muscle cells , as shown by the absence of expression of the eGFP reporter in the tumors ( see Figure 3B , Figure 3—figure supplement 1B ) . Accordingly , Dox-mediated Hgf downmodulation did not impair tumor growth ( Figure 5—figure supplement 1A , B ) indicating that transgenic HGF production was not essential for tumor maintenance . Although all primary murine tumors retained basal Met expression , in a subset of ERMS ( which later turned out to be Met-amplified ) , Met was present at much higher level and overexpression was accompanied by receptor phosphorylation ( Figure 5A , B , and Figure 5—figure supplement 2B ) . On the other hand , at earlier stages , paracrine HGF stimulation promoted receptor downregulation ( Figure 5—figure supplement 2A ) and this resulted in a difficult detection of Met and phospho-Met ( Figure 5—figure supplement 2B ) . 10 . 7554/eLife . 12116 . 011Figure 5 . Only a minor fraction of ERMS harbors MET amplification . ( A ) Western blot analysis in a panel of primary murine MHI-null ERMS probed for the indicate proteins . ( B ) Immunohistochemical analysis of primary murine MHI-null ERMS shown in A , probed for Met and phosphorylated Met . ( Scale bar = 100 µm ) . ( C ) Waterfall plot showing the MET Z-score expression from a panel of previously characterized human RMS tumors ( Shern et al . , 2014 ) ( http://pob . abcc . ncifcrf . gov/cgi-bin/JK ) . The exceptionally high MET expression in Patient 2083 ( highlighted ) is associated with focal amplification of chromosome 7q31 . 2 ( Shern et al . , 2014 ) . ( D ) FISH for MET in a human ERMS sample ( INT291 ) characterized by small cluster signals ( with an average of 9 signals per cell; MET is labeled in green and centromeres in red ) . ( E ) CGH analysis of murine MHI-null ERMS . Red indicates copy number gain , green indicates copy number loss . Arrowheads mark the Met ( red ) and Kras ( blue ) loci . DOI: http://dx . doi . org/10 . 7554/eLife . 12116 . 01110 . 7554/eLife . 12116 . 012Figure 5—figure supplement 1 . Transgenic Hgf downmodulation does not impair tumor growth . ( A ) Growth curve ( mean ± SEM ) of tumors in MHI-null mice treated with Dox when tumors became palpable ( n=3 ) . ( B ) Representative images of a MHI-null mouse described in A . DOI: http://dx . doi . org/10 . 7554/eLife . 12116 . 01210 . 7554/eLife . 12116 . 013Figure 5—figure supplement 2 . Determination of the level of Met protein and phosphorylation in rhabdomyosarcomagenesis . ( A ) Western blot analysis in the indicated samples of control ( MI-null ) and MHI-null mice , probed for the indicate proteins . ( B ) Representative immunohistochemical analysis of Met and P-Met in ERMS carrying or not Met amplification . For the indicated stages of rhabdomyosarcomagenesis see Figure 3D . ( Scale bar = 100 µm ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12116 . 01310 . 7554/eLife . 12116 . 014Figure 5—figure supplement 3 . Only a minor fraction of murine ERMS displays Met activation and amplification . ( A ) Array-based detection of phosphorylated RTKs ( p-RTK ) in the indicated cell lysates . Arrows indicate duplicate signals for P-Met ( in red ) , P-Erbb2 ( in green ) and P-Axl ( in violet ) ; positive controls are shown on the top left , top right , and lower left corners . ( B ) Densitometric quantification of P-Met signal detected through p-RTK analysis in the indicated MHI-null ERMS-derived cell lysates . ( C ) Real-time PCR analysis of Met and Kras relative copy number variation ( CNV ) performed in the indicated MHI-null ERMS . Arrowheads mark the Met ( red ) and Kras ( blue ) loci . DOI: http://dx . doi . org/10 . 7554/eLife . 12116 . 01410 . 7554/eLife . 12116 . 015Figure 5—figure supplement 4 . Evaluation of Met expression in human RMS datasets . ( A ) Box plot of MET mRNA expression level in primary human RMS tumors and skeletal muscles in the indicated datasets . NSp>0 . 05; **p<0 . 01; ( t test ) . ( B ) Box plot of MET mRNA expression level in human ARMSp tumors and in ERMS+ARMSn tumors in the indicated datasets . *p<0 . 05; ****p<0 . 0001 ( t test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12116 . 015 Despite extensive attempts to stabilize cell lines from both sarcoma subtypes , we obtained a significant number of cell lines only from ERMS . To identify signaling pathways involved in tumor maintenance , we tested a panel of 18 ERMS lines with receptor tyrosine kinase ( RTK ) phosphoarrays . Met activation was detected in three cell lines ( #187R , #1640 and #1796 , 17% of the total ) ( Figure 5—figure supplement 3A , B ) that all turned out to be sensitive to Met inhibition ( see below ) . Notably , Met had already been found overexpressed and phosphorylated in the corresponding primary tumor specimens ( Figure 5A , B and not shown for tumor #187R ) ruling out the possibility that Met activation resulted from in vitro culture conditions . Comparative genomic hybridization ( CGH ) revealed amplification of the Met locus in three samples ( #1640 , #1671 and #187R ) , and gain of the entire chromosome 6 in another one ( #1796 ) ( Figure 5E , Figure 7H ) . Amplification of the Met locus was confirmed by real-time PCR on genomic DNA ( Figure 5—figure supplement 3C , Figure 7—figure supplement 1D ) . 10 . 7554/eLife . 12116 . 016Figure 6 . Murine ERMS are genetically heterogeneous , with subsets specifically sensitive to pharmacological inhibition of distinct drivers . ( A , B ) Proliferation analysis ( mean ± SD ) of the indicated MHI-null ERMS cells treated with Met ( PHA ) and PI3K ( LY ) inhibitors . The number of cells at day 0 was set at 100% , representative assay of at least 2 independent experiments . ( C ) Quantification and representative images of soft agar growth assays of cells treated with PHA . The number of colonies obtained from cells maintained in DMSO control was set at 100% ( 3 independent experiments , mean ± SEM ) . ( Scale bar = 500 µm ) . ( D ) MHC expression analysis and representative MHC immunostaining of cells treated with PHA for 3 days ( 3 independent experiments , mean ± SEM ) . Dashed areas are shown at threefold magnification . ( Scale bar = 250 µm ) . ( E , F ) Proliferation analysis ( mean ± SD ) of the indicated MHI-null ERMS cells treated with Met ( PHA ) and PI3K ( LY ) inhibitors . The number of cells at day 0 was set at 100% , representative assay of at least 2 independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 12116 . 01610 . 7554/eLife . 12116 . 017Figure 6—figure supplement 1 . Murine ERMS are genetically heterogeneous , with subsets specifically sensitive to pharmacological inhibition of distinct drivers . ( A ) Western blot analysis of #1640 and #1796 MHI-null ERMS cells , treated with PHA for 2 hr . ( B ) Cell cycle distribution of cells treated with PHA for 1 day ( 2 independent experiments , mean ± SEM ) . ( C ) Apoptosis assessment in cells treated with PHA for 1 and 3 days ( 2 independent experiments , mean ± SEM ) . ( D ) Tumor growth curve ( mean ± SEM ) of #1640 and #1796 ERMS cells subcutaneously injected in nude mice . Mice were treated with Crizotinib ( 100 mg/kg/day ) starting on the day indicated by the arrow . ( E ) Western blot analysis of #59 and #2320 MHI-null ERMS cells , treated with LY for 2 hr . ( F ) Cell cycle distribution of cells treated with LY for 1 day ( 2 independent experiments , mean ± SEM ) . ( G ) Apoptosis assessment in cells treated with LY for 1 and 3 days ( 2 independent experiments , mean ± SEM ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12116 . 01710 . 7554/eLife . 12116 . 018Figure 7 . Treatments based on combination therapy can be effective in preventing ERMS recurrence and clonal evolution . ( A ) Western blot analysis on the indicated proteins in L and R cells . ( B ) Western blot on total lysate and ALK-immunoprecipitated fraction of #187 ERMS tumor . ( C ) Western blot analysis of R cells treated with PHA for 2 hr . ( D ) Proliferation analysis ( mean ± SD ) of L and R cells treated with the indicated inhibitors . The number of cells at day 0 was set at 100% , representative assay of at least 2 independent experiments . ( E ) Western blot analysis on total cell lysate and ALK-immunoprecipitated fraction of L cells , treated with 500 nM Crizotinib for 2 hr . ( F ) Quantification and representative images of soft agar growth assays of L and R cells treated with the indicated inhibitors . The number of colonies obtained from cells in DMSO control was set at 100% ( 2 independent experiments , mean ± SEM ) . ( Scale bar = 500 µm ) . ( G ) MHC expression analysis and representative MHC immunostaining of L and R cells , treated with the indicated inhibitors for 3 days ( 2 independent experiments , mean ± SEM ) . Dashed areas are shown at threefold magnification . ( Scale bar = 250 µm ) . ( H ) CGH analysis of MHI-null ERMS cell lines . Red indicates copy number gain , green indicates copy number loss . Arrowheads mark Alk ( orange ) and Met ( red ) loci . ( I ) Western blot analysis of R and PHA-selected R cells . ( J ) Proliferation analysis ( mean ± SD ) of PHA-selected R cells treated with the indicated inhibitors . The number of cells at day 0 was set at 100% , representative assay of at least 2 independent experiments . ( K ) Proliferation analysis ( mean ± SD ) of PHA-selected #1640 cells treated with the indicated inhibitors . The number of cells at day 0 was set at 100% , representative assay of at least 2 independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 12116 . 01810 . 7554/eLife . 12116 . 019Figure 7—figure supplement 1 . Treatments based on combination therapy can be effective in preventing ERMS recurrence and clonal evolution . ( A ) Table of L and R cells showing oncogene addiction and relative pharmacological response . ( B ) Cell-cycle distribution of L and R cells , treated with the indicated inhibitors for 1 day ( 2 independent experiments , mean ± SEM ) . **p<0 . 01 ( t test ) . ( C ) Apoptosis assessment in L and R cells treated with the indicated inhibitors for 1 and 3 days ( 2 independent experiments , mean ± SEM ) . ( D ) Real-time PCR analysis of Met and Alk relative copy number variation ( CNV ) performed in the indicated MHI-null ERMS cell lines . Arrowheads mark Alk ( orange ) and Met ( red ) loci . ( E ) Real-time PCR analysis of Alk expression in a subset of ERMS cell lines . ( F ) Western blot analysis in TS and SU-DHL-1 cells , treated with the indicated inhibitors for 6 hr . ( G ) Proliferation analysis ( mean ± SD ) of cells described in F treated with the indicated inhibitors . The number of cells at day 0 was set at 100% , representative assay of at least 2 independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 12116 . 019 Analysis of the level of MET transcript in human RMS datasets ( Davicioni et al . , 2009; Williamson et al . , 2010 ) confirmed the variability in terms of Met expression ( Figure 5—figure supplement 4A , B ) . Similar results were obtained by the analysis of RNA sequencing data on a different large cohort of human patients ( Figure 5C ) . The only ERMS sample where Met expression rose above all others ( #2083 ) was found to harbor MET amplification ( Shern et al . , 2014 ) . Interestingly , another example of human ERMS carrying MET amplification emerged from an independent FISH analysis of a small set ( 8 cases ) of ERMS ( Figure 5D ) . Overall murine and human data indicate a considerable variability of Met expression and rare Met amplification . Treatment of #1640 and #1796 cell lines ( #1671 cells were unfortunately lost ) with the Met-specific inhibitor PHA-665752 ( herein referred as PHA ) resulted in the abrogation of Met signaling ( Figure 6—figure supplement 1A ) , in the arrest of cell proliferation ( Figure 6A , B ) , in the accumulation of cells in the G0/G1 phase of the cell cycle ( Figure 6—figure supplement 1B ) , in the impairment of anchorage-independent growth ( Figure 6C ) and in the induction of apoptosis ( Figure 6—figure supplement 1C ) . Interestingly , the rare surviving cells were fully differentiated ( Figure 6D ) . Finally , nude mice bearing ERMS tumors derived from both cell lines were treated with Crizotinib , a well-tolerated dual Met and ALK inhibitor , approved for the treatment of patients with metastatic ALK-positive NSCLC ( Rodig and Shapiro , 2010 ) . Crizotinib administration to mice with palpable tumors blocked ERMS growth and reduced the tumor volume , without symptoms of toxicity ( Figure 6—figure supplement 1D ) . RTK downstream effectors are often dysregulated in ERMS . Accordingly , some ERMS cases of mutated phosphatidylinositol 3 kinase A ( PIK3CA ) have been reported ( Chen et al . , 2013; Shern et al . , 2014 ) and 82 . 5% of RMS tumors have been found to display a strong activation of the pathway ( Renshaw et al . , 2013 ) . LY 294002 ( herein referred as LY ) , a PI3K pathway inhibitor ( Vlahos et al . , 1994 ) , was employed in our panel of murine ERMS cells . In two cell lines ( #59 and #2320 ) , LY dramatically reduced cell proliferation ( Figure 6E , F ) , strongly inhibited Akt phosphorylation , promoted accumulation in the G0/G1 phase of the cell cycle and induced apoptosis ( Figure 6—figure supplement 1E–G ) . To investigate whether the addiction to Met and PI3K were mutually exclusive , PHA-sensitive cells ( #1640 and #1796 ) were treated with LY , while LY-sensitive cells ( #59 and #2320 ) were grown in the presence of PHA . PHA and LY did not interfere , respectively , with the proliferation rate of #59/# 2320 and #1640/#1796 , confirming that these ERMS cells depend on the sustained activation of distinct signaling pathways ( Figure 6A , B , E , F ) . Unfortunately , almost all cancer patients treated with a drug aimed at a single target relapse and this is particularly evident in cases of intratumoral heterogeneity , in which such treatment inevitably results in the selection and expansion of resistant clones . In our effort to stabilize murine ERMS cell lines for pre-clinical studies , we routinely implanted the tumor in the left ( L ) and right ( R ) flanks of recipient nude mice . Intriguingly , in the case of tumor #187 we obtained two cell lines ( #187L and R ) , one of which ( R ) expressed high levels of phosphorylated Met while the other one ( L ) did not ( Figure 7A ) . R cells , as expected , were sensitive to PHA and Crizotinib ( Figure 7C , D , F , G , Figure 7—figure supplement 1A–C ) . Conversely , L cells were resistant to PHA but responded to Crizotinib ( Figure 7D , Figure 7—figure supplement 1A ) , suggesting amplification of Alk , a lesion frequently found in human RMS ( van Gaal et al . , 2012 ) . Indeed , while visualization of ALK protein by Western blot in the primary tumor extract required enrichment by immunoprecipitation ( Figure 7B ) , L cells expressed unusually high levels of ALK at baseline ( Figure 7A ) . In L cells Crizotinib caused a strong inhibition of ALK signaling , impairment of anchorage-independent growth and induction of myogenic differentiation ( Figure 7E–G ) , as well as G0/G1 accumulation and induction of apoptosis ( Figure 7—figure supplement 1B , C ) . Accordingly , CGH and CNV analysis revealed Met and Alk amplification , respectively , in R and L cells ( Figure 7H , Figure 7—figure supplement 1D ) . Interestingly , real-time PCR showed a low level of Alk expression also in R cells ( Figure 7—figure supplement 1E ) , suggesting that the original tumor contained a mixed population with a minor Alk-amplified component . We therefore treated R cells for one month with 250nM PHA , a dose that specifically inhibits Met signaling without interfering with the ALK pathway ( Figure 6—figure supplement 1A , Figure 7—figure supplement 1F , G ) . PHA-selected R cells were negative for phospho-Met ( Figure 7I ) , but were sensitive to Crizotinib ( Figure 7J ) . CGH , CNV and real-time PCR analysis revealed the expansion , upon PHA treatment , of the Alk-amplified population ( Figure 7H , Figure 7—figure supplement 1D , E ) . Conversely , prolonged PHA treatment of #1640 cells resulted in the emergence of a clone harboring also Kras amplification ( Figure 5E , Figure 5—figure supplement 3C ) . Notably , the combination of Met and MEK inhibitors resulted in complete arrest of proliferation in these cells ( Figure 7K ) . Overall these data suggest the presence of intratumoral heterogeneity in our model and that combination therapy can be a more effective strategy in preventing recurrence due to clonal evolution in ERMS .
ERMS and UPS are two distinct subtypes of sarcoma . While ERMS is characterized by features of myogenic cells , UPS lacks any tissue-specific marker of differentiation . Intriguingly , although phenotypically and genetically distinguishable , ERMS and UPS have emerged as parts of a tumor continuum , suggesting an intrinsic relationship between these two subtypes ( Blum et al . , 2013; Rubin et al . , 2011 ) . Our muscle-restricted inducible HGF model is a unique tool to explore the influence of SC niche perturbation ( an issue which has not been thoroughly investigated so far ) in the development of these two distinct but associated pathological entities . Interestingly , in a Pax7 wild type background , where satellite cells are present in normal number , HGF induction prevalently resulted in ERMS formation with only less than 10% of mice developing UPS . Satellite cells are indeed particularly sensitive to the activation of the HGF/Met axis . They respond in vitro to HGF stimulation with an increase in proliferation rate ( Allen et al . , 1995 ) , and injection of HGF in uninjured muscle directly stimulates satellite cells activation ( Tatsumi et al . , 1998 ) . Recently , it has been shown that quiescent satellite cells residing in the controlateral limb with respect to the site of an injury respond to distant tissue damage by transitioning into an 'alert' state , a process that is again dependent on the HGF/Met axis ( Rodgers et al . , 2014 ) . In the absence of satellite cells , muscle regeneration is dramatically impaired and this results in the aberrant deposition of fibrotic tissue ( Maltzahn et al . , 2013; Murphy et al . , 2011; Oustanina et al . , 2004 ) . This process depends on the extensive expansion of resident fibroblasts , the other major component of the niche . Our model of perturbation of the SC niche showed that in a Pax7 null genetic background ERMS incidence dramatically drops in favor of UPS development . In this background , while rare ERMS may still derive from the few remaining satellite cells ( Oustanina et al . , 2004 ) , UPS are likely to originate from the aberrant expansion of fibroblasts ( Maltzahn et al . , 2013; Murphy et al . , 2011 ) . Accordingly , while murine ERMS were positive for satellite cell markers , UPS were positive for mesenchymal markers , such as PDGFRα . Interestingly , PDGFRα expression has been observed in 62% of human UPS samples ( Rüping et al . , 2014 ) . Furthermore , bioinformatic analysis on human ERMS and UPS datasets revealed a satellite signature for ERMS and a fibroblast signature for UPS . Thus , changes in the microenvironment of the niche produce distinct subtypes of sarcoma depending on different susceptible cell types . Genetic dissection of the cell of origin of ERMS has been extensively investigated . The preferred line of action has been the introduction of the most frequent human genetic lesions in mouse models by using cell/lineage specific promoters . However , these efforts , rather than unequivocally identifying the cell of origin , have further highlighted the complexity of sarcomagenesis . Hatley et al . ( Hatley et al . , 2012 ) obtained an aggressive form of head and neck ERMS by activating the Sonic Hedgehog pathway in combination with Cdkn2a deletion in adipocyte precursors . According to Rubin et al . ( Rubin et al . , 2011 ) , 100% of ERMS incidence occurred when p53 was knocked out in late myogenic precursors ( Myf6-driven Cre ) , while its deletion in satellite cells ( Pax7-driven Cre ) resulted in UPS . Notably , activation of Sonic Hedgehog in satellite cells in the same context of p53 deficiency , triggered ERMS formation , while its dysregulation in Myf6-positive-cells , promoted UPS development . Conversely ERMS , rather than UPS , represented the major histological subtype when Pax7-driven KrasG12D was combined with loss of Trp53 ( Van Mater et al . , 2015 ) . In this genetic setting , tumorigenesis was accelerated upon muscle injury or local administration of HGF . Notwithstanding , the same group showed that MyoD-driven KrasG12D in association with loss of Trp53 principally resulted in UPS rather than ERMS . Furthermore , expression of the activated form of the Hippo effector YAP1 in satellite cells resulted in fully penetrant ERMS only upon injury ( Tremblay et al . , 2014 ) . These models suggest two important mechanistic concepts . First , a given sarcoma subtype likely arises from the combination of specific genetic lesions with a well-defined cell lineage/stage of maturation . Second , tissue damage , such as injury or trauma , accelerates or , as in the case of YAP , triggers sarcoma development when occurring in a genetically predisposed background . Instead of imposing the most frequent genetic lesions to a cell in a specific developmental stage , we aimed at understanding how the most relevant component of the stem niche microenvironment , HGF , can contribute to ERMS and UPS formation . With our model , where only the soil of the niche is specifically modified while cell-intrinsic genetic events are left to chance , we proved that the HGF/Met axis mediates satellite cells release from quiescence , promoting their activation . In a wild type background , HGF stimulation perturbed muscle stem cell homeostasis only slightly . Conversely , in a tumor prone background , HGF caused massive expansion of cycling satellite cells , triggering ERMS initiation . Although we could not determine at which stage of the myogenic lineage full blown transformation occurs , the histological presentation supports a model of multistep ERMS progression . In the first stages of the disease , HGF-mediated stem cell niche perturbation resulted in neomyogenesis . At this early stage , Pax7-positive cells surrounded old fibers and were still able to form novel small muscle fibers . However at later stages , upon several cycles of proliferation and most likely through the acquisition of additional genetic lesion ( s ) , stem cells lost the ability to differentiate and formed full blown tumors . Genetic validation of satellite cells as ERMS initiators was obtained by finding a drastic reduction of their incidence in the Pax7 null background . In our model , activation of the HGF/Met axis is involved in the initiation of all tumors . While variable Met expression was present in all murine ERMS , its phosphorylation was observed only in Met-amplified tumors that were sensitive to Met inhibition . Thus in our system its role in maintenance seems to be limited only to tumors where the Met locus is amplified , presumably as a consequence of an additional genetic hit . The importance of the HGF/Met axis in ERMS tumorigenesis had been first shown by Sharp et al . by combining ectopic expression of the Hgf transgene with loss of the Cdkn2a locus ( Sharp et al . , 2002 ) . In their model phosphorylated Met was detectable in most tumor samples , indicating widespread constitutive activation . This discrepancy between the two models is likely to be due to differences in the strength and tissue specificity of the promoters . In Sharp et al . transgenic Hgf was expressed by the tumors , while in our model this was not the case . In human cancer MET is rarely mutated or amplified while its overexpression is frequently observed in a variety of tumors including RMS and UPS ( Ferracini et al . , 1996; Lahat et al . , 2011; Taulli et al . , 2006 ) . Our bioinformatic analysis revealed an enriched Met score in both human RMS and UPS . New evidence indicates that Met expression in cancer cells is a paradigm of 'inherence' ( Boccaccio and Comoglio , 2013 ) . The innate presence of Met in tumor cells is attributed to cancer stem cells ( CSC ) that inherit Met expression from their physiological counterpart ( normal stem/progenitor cells ) as part of their normal phenotype . In the CSC context , Met is not only a marker of the stemness status of the tumor but supports the self-renewal program and the expansion of CSC . Indeed , the HGF/Met axis sustains the stem cell phenotype in glioblastoma and colon cancer . Colon-derived xenospheres express Met in stem cell medium , but its downregulation rapidly occurs in differentiating conditions ( Luraghi et al . , 2014 ) . In RMS as well , the Met pathway could sustain a stem cell phenotype . Our bioinformatic analysis on human datasets suggests that the Met and satellite signatures converge on the ERMS subtype , and our in vivo data clearly show that the HGF/Met axis acts prevalently on muscle stem cells to promote ERMS initiation . Furthermore , we previously showed that sustained expression of Met in RMS is partly due to the loss of muscle-specific microRNAs ( Taulli et al . , 2009 ) . Ectopic introduction of myomiRs in RMS cells re-activated the differentiation program , in a process that involved Met downregulation and epigenetic reprogramming ( Coda et al . , 2015; Taulli et al . , 2014 ) , again confirming the role of Met in maintaining an undifferentiated , stem cell-like phenotype . However , the functional significance of Met in cancer has also been attributed to another property of some tumor cells , Met-'addiction' . Tumor cells harboring MET amplification display exquisite sensitivity to Met inhibitors , providing a rationale for the use of targeted therapies in patients carrying this lesion ( Smolen et al . , 2006 ) . Accordingly , in our model ERMS tumors bearing Met amplification showed Met-addiction . In these cases , Met inhibition stopped tumor growth and promoted terminal differentiation of the rare surviving cells , suggesting that , while the tumor bulk mainly consisted of Met-addicted cells , some residual cells required Met activity to retain their stem cell phenotype . These results nicely illustrate the two distinct roles of HGF/Met signaling in tumorigenesis based , respectively , on Met activity and Met-addiction and altogether suggest that , in the rare cases of human ERMS harboring this lesion , the patient may benefit from the use of a Met inhibitor . Recent deep sequencing data of multiple regions of human tumors have shown that they often harbor a dominant genetic clone , plus one or more genetically and topologically distinct subclones ( Gerlinger et al . , 2012 ) . The latter represent a major roadblock for a targeted therapeutic approach in terms of their potential to foster both progression and resistance ( McGranahan and Swanton , 2015 ) . Intratumoral heterogeneity is poorly reproduced in experimental models . Importantly , our model also exemplifies this feature , as shown by the isolation , from the same tumor , of a major Met- and a minor emerging Alk- or Kras-driven population . Expansion of the latter occurred following treatment with a Met inhibitor , underscoring the risk of drug-induced clonal evolution . Overall , our model demonstrates the functional relevance of the SC niche in ERMS and UPS development , providing an alternative explanation for the existence of a tumor continuum between these two subtypes . Although the high level of genetic complexity of sarcomas cannot be fully resolved with our model , we provide a rationale for the use of combination therapy instead of a single agent for a robust precision-guided therapeutic approach of genetically heterogeneous sarcomas .
Ckm-tTA mice were donated by P . Plotz ( Nagaraju et al . , 2000 ) . The Hgf cDNA was inserted in the PvuII site of pBI-GFP vector ( Clontech , Mountain View , CA ) . The fragment obtained after AseI digestion was microinjected into FVB fertilized eggs at the San Raffaele-Telethon Core Facility for Conditional Mutagenesis ( Milano , Italy ) . Cdkn2a-null mice were from Jackson Laboratory ( Bar Harbor , ME ) . Pax7-null mice were provided by P . Gruss . Genotyping primers are specified in Supplementary file 1 . Doxycycline diluted in drinking water ( 1 mg/ml ) was changed every 3 days . Approval for all animal procedures was granted by the Ethical Committee of the University of Torino during the session held in Turin on Sept 26 , 2013 and later communicated to the Italian Ministry of Health on Oct 24 , 2013 . RNA was extracted using TRIzol ( Invitrogen , Carlsbad , CA ) and retrotranscribed to cDNA using the iScript cDNA Synthesis Kit ( Bio-Rad , Hercules , CA ) . Real-time PCR was performed with iQ SYBR Green ( Bio-Rad ) . Primers are specified in Supplementary file 1 . Disaggregated Tibialis anterior muscles and sera were lysed in lysis buffer [20 mmol/L Tris ( pH 7 . 5 ) , 150 mmol/L NaCl , 1 mmol/L EDTA , 1 mmol/L EGTA , 1% Triton X-100 , 1 mmol/L h-glycerolphosphate] with Protease Inhibitor Cocktail . 200 µg of proteins were used for HGF determination using ELISA Kit ( B-Bridge , Santa Clara , CA ) according to the manufacturer’s protocol . Immunohistochemistry was performed as previously described ( Taulli et al . , 2009 ) with the specified antibodies . Ki67 ( #NCL-Ki67p ) was from Leica Biosystems ( UK ) ; MyoD ( #M3512 ) was from Dako ( Denmark ) ; Myogenin ( DSHB ( Iowa City , IA ) Hybridoma Product F5D was deposited by Wright , Woodring E . ) ; Pax7 ( DSHB Hybridoma Product PAX7 was deposited by Kawakami , Atsushi ) ; eMHC ( DSHB Hybridoma Product F1 . 652 was deposited by Blau , H . M . ) ; P-Met ( #3126 ) and PDGFRα ( #3164 ) were from Cell Signaling Technology ( Danvers , MA ) ; Met ( #18–7366 ) was from Invitrogen . Fiber cross-sectional areas were measured using ImageJ software ( rsb . info . nih . gov/ij ) . Statistical analyses were performed using Microsoft Excel software , with a moving average ( period 4 ) trendline . Muscles were immersed into isopentane for 30 sec and frozen in liquid nitrogen . Cryosections were fixed with 4% paraformaldehyde , permeabilized with methanol at -20°C for 6 min and incubated with the specified antibodies . Laminin ( #L9393 ) and Desmin ( #D1033 ) were from Sigma-Aldrich; Pax7 ( DSHB Hybridoma Product PAX7 was deposited by Kawakami , Atsushi ) ; MyoD ( #sc-760 ) was from Santa Cruz Biotechnology ( Dallas , TX ) ; 488 , 555 and Cy3-conjugated secondary antibodies were from Invitrogen . For Pax7 and MyoD the signal was amplified by incubation with biotin-conjugated goat anti-mouse/rabbit IgG1 ( Jackson ImmunoResearch , West Grove , PA ) followed by incubation with 488-coniugated streptavidin ( Jackson ImmunoResearch ) or with Cy3-coniugated streptavidin ( Jackson ImmunoResearch ) . Nuclei were stained with DAPI . EDL single fiber isolation was performed as previously described ( Zammit et al . , 2002 ) . Single fibers were plated in matrigel-coated culture plates and maintained in DMEM supplemented with 0 . 5% Chicken Embryo extract , 10% Horse Serum . Fibers were fixed in 4% PAF and permeabilized in 0 . 5% Triton X-100 . Nonspecific antibody binding was blocked by incubation in 20% PBS-BSA . Fibers were incubated overnight at +4°C with the specified antibodies . Pax7 ( DSHB Hybridoma Product PAX7 was deposited to the DSHB by Kawakami , Atsushi ) ; MyoD ( #sc-760 ) was from Santa Cruz Biotechnology . Alexa 555-conjugated secondary antibodies were from Invitrogen . Nuclei were stained with DAPI . The number of cells/fiber was counted ( ± SEM ) . Murine ERMS cells and satellite cells were isolated as previously described ( Crepaldi et al . , 2007 ) . ERMS cells were maintained in DMEM supplemented with 10% FBS . Satellite cells were maintained as previously described ( Crepaldi et al . , 2007 ) . SU-DHL-1 and TS cells ( a subclone of SUP-M2 ) were kindly provided by R . Chiarle and maintained in RPMI 1640 supplemented with 10% FBS . Methods of characterization from the cell bank include karyotyping and DNA fingerprinting . All cell lines were regularly tested with MycoAlert ( Lonza , Walkersville , MD ) to ascertain that cells were not infected with mycoplasma . Met inhibitor PHA-665752 was used at 250 nmol/L . Crizotinib ( PF-02341066 ) ( Selleckchem , Houston , TX ) was used at 250 nmol/L . LY-294002 was used at 10 µmol/L . Selumetinib ( AZD6244 ) ( Selleckchem ) was used at 1 µmol/L . DMSO was used as a control . Western blot was performed using the specified antibodies . MHC ( #sc-32732 ) and Myogenin ( #sc-12732 ) were from Santa Cruz Biotechnology; Met ( #18–7366 ) was from Invitrogen; P-Met ( #3126 ) , P-Akt ( #9271 ) , GAPDH ( #5174 ) , P-TYR ( #9411 ) , hALK ( #3633 ) , P-hALK ( #3341 ) and PDGFRα ( #3164 ) were from Cell Signaling Technology; mALK ( #ab16670 ) was from Abcam ( UK ) ; MyoD ( #M3512 ) was from Dako; P-MAPK ( # M8159 ) , Tubulin ( #T5201 ) , Actin ( #A5060 ) were from Sigma-Aldrich . mALK immunoprecipitation: total cell extracts ( RIPA ) were 5x diluted in IP buffer ( 50 mM TrisHCl pH 7 . 9; 150 mM NaCl; 1 mM EDTA; 5 mM MgCl2; 0 . 1% NP-40; 20% glycerol ) with 1 mM phenylmethylsulfonyl fluoride , 10 mM NaF , 1 mM Na3VO4 and protease inhibitor cocktail . Samples were precleared with equilibrated Dynabeads protein G ( Invitrogen ) ; mALK antibody and normal-IgG ( Santa Cruz Biotechnology ) were incubated overnight at 4°C . Mouse Phospho-Receptor Tyrosine Kinase Array ( R&D Systems , Minneapolis , MN ) was performed following the manufacturer’s instructions . The mean pixel density was measured using Quantity One software and expressed as a percentage of the mean pixel density of positive controls . Cells were seeded in 12-well plates at a density of 1 × 104 cells/well . Proliferation was evaluated by CellTiter-Glo ( Promega , Madison , WI ) following the manufacturer’s instructions . Cell-cycle and apoptosis analyses were performed as previously described ( Taulli et al . , 2009 ) . Cells were washed in PBS and incubated for 30 min with primary antibodies ( anti Sca1: FITC Rat Anti-Mouse Ly-6A/E BD Biosciences , Franklin Lakes , NJ , #553335; anti mouse Integrin-α7 MBL International , Woburn , MA , #K0046-3 ) diluted in PBS-0 . 1% BSA . After washing , cells were incubated for 30 min with secondary antibody ( required only for Integrin-α7: APC goat anti-mouse IgG Invitrogen #A865 ) and resuspended in PBS-0 . 1% BSA . Cells were analyzed by FACS scan using CellQuest Software ( BD Biosciences ) . MHC immunofluorescence and MHC staining for FACS analysis was performed as previously described ( Coda et al . , 2015; Taulli et al . , 2009 ) . Cells were suspended in 0 . 45% type VII low-melting agarose in 10% FBS DMEM at 5 × 104/well and plated on a layer of 0 . 9% agarose in 10% FBS DMEM in 6-well plates and cultured for two weeks at 37°C with 5% CO2 . Cells were trypsinized and resuspended at 4 × 106 cells/ml in sterile PBS . Cells ( 100 μl ) were injected subcutaneously into the flank of female nu/nu mice ( Charles River Laboratories , Wilmington , MA ) . Tumor size was measured with Vernier calipers every 2 days and tumor volumes were calculated as a sphere volume . Mice were treated for 8 days with 100 mg/kg/day Crizotinib by oral gavage . Crizotnib was resuspended in 0 . 5% methylcellulose and 0 . 4% polysorbate 80 . Comparative genomic hybridization analysis was performed at ‘Fondazione Edo ed Elvo Tempia’ , Biella , Italy . Total DNA was extracted using the DNeasy Blood & Tissue Kit ( Qiagen , Germany ) . Comparative genomic hybridization using aCGH microarrays wad carried out using the enzymatic labeling method . Digestion , labeling , hybridization , washing and slide scanning were performed following the manufacturer’s protocols ( Agilent Technologies , Santa Clara , CA ) . Images were analyzed using Feature Extraction software version 10 . 7 ( Agilent Technologies ) . CNV primers are specified in Supplementary file 1 . Analysis was performed using the standard curve method . Actl6a was used as a control: for each sample , the copy number was calculated using the ratio of Met/Kras/Alk Vs Actl6a copy number . Relative copy number variation was determined for each sample in comparison with a non-amplified reference sample . RNA sequencing of primary tumors was performed as previously described ( Shern et al . , 2014 ) . Briefly , PolyA-selected RNA libraries were prepared for RNA sequencing on Illumina HiSeq2000 and Illumina NextSeq using TruSeq v3 chemistry according to the manufacturer’s protocol ( Illumina , San Diego , CA ) . One hundred base-long paired-end reads were assessed for quality and reads were mapped using CASAVA ( Illumina ) . The generated fastq files were used as input for TopHat2 ( Trapnell et al . , 2009 ) . Reads were mapped according to the hg19 human genome assembly . Cufflinks ( http://cufflinks . cbcb . umd . edu/ ) ( Trapnell et al . , 2010 ) was used to assemble and estimate the relative abundances of transcripts mapped with TopHat2 at the gene and transcript level ( FPKM ) . FPKM values were log2 transformed . Sample specific Z-scores of expression were calculated using a panel of normal tissue . The generated data is publically available via the National Cancer Institute Oncogenomics web site ( http://pob . abcc . ncifcrf . gov/cgi-bin/JK ) . Formalin-fixed paraffin-embedded tissue sample was collected at Fondazione IRCCS Istituto Tumori in Milan , Italy , according to the Internal Review and the Ethics Boards of the Istituto Nazionale dei Tumori of Milan . Patients gave their written consent for research activities . FISH was performed on 2–4 μm-thick paraffin embedded sections by counting at least 100 tumor cells . MET amplification was assessed using a commercial probe at 7q31 ( Zyto-Light SPEC MET/CEN7 Dual Color Probe , ZytoVision , Germany ) , used according to the manufacturer’s instructions . MET/CEN7 ratio , the percentage of tumor cells with >4 MET signals , and the average MET copy number per cell were calculated . FISH results were evaluated in concordance with the criteria described by Schildhaus et al . ( Schildhaus et al . , 2015 ) . We ran expression analysis on the following publically available sets of microarray data: for 'Davicioni' dataset ( Affymetrix Human Genome U133A Array platform ) we used 134 RMS ( Davicioni et al . , 2009 ) . For 'Williamson' dataset ( Affymetrix Human Genome U133 Plus 2 . 0 Array platform ) we used 101 RMS samples ( Williamson et al . , 2010 ) . Muscle datasets GSE3307 and GSE1462 were used in comparison with RMS dataset 'Davicioni' , while muscle datasets GSE3526 and GSE2328 were used in comparison with RMS dataset 'Williamson' . For 'Gibault' dataset ( Affymetrix Human Genome U133 Plus 2 . 0 Array platform ) we used 73 UPS samples and 79 other sarcoma samples ( 20 myxofibrosarcomas MFS , 10 liposarcomas LPS and 49 leiomyosarcomas LMS ) ( Gibault et al . , 2011 ) . All samples were normalized by the RMA algorithm as implemented in R free software environment , with custom CDF ( Lembo et al . , 2012 ) . Differential expression was evaluated by limma package . As satellite signature we used the overlap between the mouse signatures as in Fukada ( Fukada et al . , 2007 ) and in Pallafacchina ( Pallafacchina et al . , 2010 ) translated in human genes by the Homologene mapping , build 67 . Fukada's gene IDs have been mapped to mouse Entrez IDs by RefSeq database , release 57 , and Pallafacchina's gene IDs by affymetrix mouse430_2 annotation , release na31 . As Met signature , we used Bertotti signature ( Bertotti et al . , 2009 ) . Clustering classification was made on the respective signature genes found on the platform by Bioconductor hclust function . As fibroblast signature , we derived differentially expressed genes between H9F cardiac fibroblasts cell line ( n=3 samples ) and H9 cardiomyocytes cell line ( n=3 samples ) by using GEO2R ( http://www . ncbi . nlm . nih . gov/geo/geo2r/ ) , imposing a cut-off of 0 . 05 on the adjusted p-value on the selection of differentially expressed genes ( Fu et al . , 2013 ) . Signatures scores are essentially the algebraic sum of the signature gene expression level: the contribution is added if the gene is up in the signature , subtracted otherwise . 2-tailed unpaired Student’s t test was used to evaluate statistical significance: NSp>0 . 05; *p<0 . 05; **p<0 . 01; ***p<0 . 001; ****p<0 . 0001 . In the CGH analysis , raw data were processed using the Agilent Genomic Workbench version 7 . Aberrant regions were detected using ADM-2 algorithm with threshold set to 6 . To avoid false positive calls , the minimum number of consecutive probes for amplifications/deletions was set at 3 , together with a minimum average absolute Log Ratio for aberrations ≥0 . 25 . For human datasets analysis gene differential expression was based on the bioconductor limma package . Enrichments have been evalutated by hypergeometric test . Heteroscedastic one-tailed Student t-test has been used to evaluate the differential signature's scores . | Soft tissue sarcomas are rare cancers that originate in tissues such as muscles , tendons , cartilage and fat . These cancers are further classified into subtypes based on their appearance . For example , rhabdomyosarcoma cells resemble the cells that normally develop into muscle , while other soft tissue tumors that do not look like a distinct cell type are called undifferentiated pleomorphic sarcomas . Recent experiments have suggested that although these subtypes appear different , they may both arise from the cells that build muscles . However , this had not been confirmed . Morena et al . investigated whether changing the environment – also known as the “niche” – of muscle stem cells could influence what type of sarcoma developed in mice that were prone to cancer . Normally muscle stem cells in an adult only regenerate injured muscles , and need to receive the correct cues before they divide . Among these cues is a protein called Hepatocyte Growth Factor ( or HGF for short ) , which is produced by cells in the muscle stem cells’ niche . Morena et al . engineered mice so that the production of HGF in the muscles could be switched on or off at will . Mice that were already prone to cancer and produced a lot of HGF tended to develop rhabdomyosarcomas . However , when HGF was turned on in similar mice that also lacked normal muscle stem cells , the resulting sarcomas were predominantly undifferentiated pleomorphic sarcomas . These data indicate that rhabdomyosarcomas probably originate from muscle stem cells , whereas undifferentiated pleomorphic sarcomas develop from other cells in the niche . Lastly , Morena et al . studied the sarcomas in their mice in more detail and observed that , similar to what has been found in human rhabdomyosarcomas , individual tumors had different genetic mutations . These differences make it difficult to treat sarcomas with a single anti-cancer drug . However , the new results suggest that a combination of targeted drugs may prove effective in blocking tumor growth and in preventing resistance . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology",
"cell",
"biology"
] | 2016 | Hepatocyte Growth Factor-mediated satellite cells niche perturbation promotes development of distinct sarcoma subtypes |
Cell-cell interactions influence all aspects of development , homeostasis , and disease . In cancer , interactions between cancer cells and stromal cells play a major role in nearly every step of carcinogenesis . Thus , the ability to record cell-cell interactions would facilitate mechanistic delineation of the role of the cancer microenvironment . Here , we describe GFP-based Touching Nexus ( G-baToN ) which relies upon nanobody-directed fluorescent protein transfer to enable sensitive and specific labeling of cells after cell-cell interactions . G-baToN is a generalizable system that enables physical contact-based labeling between various human and mouse cell types , including endothelial cell-pericyte , neuron-astrocyte , and diverse cancer-stromal cell pairs . A suite of orthogonal baToN tools enables reciprocal cell-cell labeling , interaction-dependent cargo transfer , and the identification of higher order cell-cell interactions across a wide range of cell types . The ability to track physically interacting cells with these simple and sensitive systems will greatly accelerate our understanding of the outputs of cell-cell interactions in cancer as well as across many biological processes .
Cell-cell interactions contribute to almost all physiological and pathological states ( Deb , 2014; Komohara and Takeya , 2017; Konry et al . , 2016; Zhang and Liu , 2019 ) . Despite the explosion of interest in uncovering and understanding cellular heterogeneity in tissues and across disease states , the extent to which cell-cell interactions influence cell state , drive heterogeneity , and enable proper tissue function remains poorly understood ( Konry et al . , 2016; Tsioris et al . , 2014; Zhang and Liu , 2019 ) . Detailed analysis of the impact of defined cell-cell interactions has illuminated critical aspects of biology; however , these analyses have been limited to a small number of juxtacrine signaling axes that are tractable to study ( Dustin and Choudhuri , 2016; Meurette and Mehlen , 2018; Yaron and Sprinzak , 2012 ) . Interactions between cancer cells and stromal cells play central roles in cancer initiation , progression , and metastasis ( Kitadai , 2010; Orimo and Weinberg , 2006 ) . While secreted factors relaying pro- or anti-tumorigenic signals have been extensively investigated , the impact of direct physical interactions between cancer cells and stromal cells remains understudied ( Bendas and Borsig , 2012; Dittmer and Leyh , 2014; Nagarsheth et al . , 2017 ) . A greater understanding of the constellation of direct interactions that cancer cells undergo will not only deepen our understanding of tumor ecology but also has the potential to uncover novel therapeutic opportunities ( Nagarsheth et al . , 2017; Swartz et al . , 2012 ) . Furthermore , how diverse cell-cell interactions differentially impact cancer cells at different stages of carcinogenesis and within different organ environments remains largely uncharacterized . Molecular methods to profile cell state , including in situ approaches within intact tissues , largely fail to uncover the causal relationship between cell-cell interactions and the underlying biology ( Giladi et al . , 2020; Halpern et al . , 2018 ) . Computational and experimental methods to characterize cell-cell interactions yield additional layers of dimensionality; however , modalities to capture cell-cell interactions are limited ( Boisset et al . , 2018; Morsut et al . , 2016; Pasqual et al . , 2018 ) . Much as diverse systems to detect and quantify protein-protein interactions have revolutionized our biochemical understanding of molecular systems , the development of novel systems to detect and quantify cell-cell interactions will accelerate the mapping of the interaction networks of multicellular systems . Endogenous cell-cell interactions can result in transfer of surface proteins between cells , mainly through either trans-endocytosis or trogocytosis ( Langridge and Struhl , 2017; Li et al . , 2019; Ovcinnikovs et al . , 2019 ) . Thus , we sought to integrate this phenomenon with fluorescent protein tagging to label cells that have undergone direct interactions . We describe a surprisingly robust system ( which we term GFP-based Touching Nexus or G-baToN ) that enables sensitive and specific interaction-dependent labeling of cancer cells and various primary stromal cells , including endothelial cells , T cells and neurons . We extensively characterize this approach and describe several novel applications of this versatile system .
To create a system in which a fluorescent signal could be transferred between neighboring cells , we adapted a synthetic ligand-receptor system based on the expression of surface GFP ( sGFP ) on sender cells and a cell surface anti-GFP ( αGFP ) nanobody on receiver cells ( Fridy et al . , 2014; Lim et al . , 2013; Morsut et al . , 2016 ) . Co-culturing sGFP sender cells with αGFP receiver cells led to GFP transfer and labeling of the receiver cells ( Figure 1A , B and Figure 1—figure supplement 1A ) . Receiver cell labeling required direct cell-cell contact , active membrane dynamics , and pairing between sGFP and its cognate αGFP receptor ( Figure 1C , D and Figure 1—figure supplement 1B , C ) . Notably , sGFP transfer was accompanied by reduced GFP on the sender cells , downregulation of αGFP from the surface of the receiver cells and was partially blocked by chemical inhibitors of endocytosis – all consistent with active GFP transfer and internalization into receiver cells ( Figure 1—figure supplement 1D–F ) . To characterize the kinetics of G-baToN-mediated receiver cell labeling , we performed co-culture time course experiments with time-lapse imaging and flow cytometry readouts . Time-lapse imaging showed rapid transfer and internalization of GFP by receiver cells ( Figure 1E and Video 1 ) . GFP transfer could be detected within five minutes of co-culture and was half-maximal after 6 hr ( Figure 1F and Figure 1—figure supplement 1G-H ) . Importantly , GFP fluorescence in receiver cells decayed rapidly after isolation of touched receiver cells from sender cells , thus documenting the transient labeling of receiver cells ( Figure 1—figure supplement 1I ) . To determine the sensitivity of this system , we co-cultured receiver cells with different ratios of sender cells . The fraction of labeled receiver cells was proportional to the number of sender cells , and even the addition of very few sender cells ( representing less than one sender cell to 105 receiver cells ) was sufficient to label rare receiver cells ( Figure 1G , H ) . Thus , the transfer of GFP to αGFP-expressing cells is a rapid and sensitive method to mark cells that have physically interacted with a predefined sender population . To further characterize the interaction reporter system , we deconstructed the G-baToN design into three functional modules: ( 1 ) the transmembrane domain of αGFP on the receiver cells; ( 2 ) the pairing between GFP and αGFP; and ( 3 ) the transmembrane domain of sGFP on the sender cells . We initially used a published sGFP-αGFP pair in which the Notch1 transmembrane domain links the LaG17-αGFP nanobody onto the receiver cell surface and the PDGFR transmembrane domain links sGFP onto the sender cell surface ( Morsut et al . , 2016 ) . Replacement of the Notch1 transmembrane domain of αGFP with different transmembrane domains allowed us to quantify their impact on GFP transfer efficiency . The VEGFR2 transmembrane domain enabled the highest transfer efficiency , resulting in about a threefold increase relative to the original design ( Figure 2A–C ) . We next replaced the LaG17-αGFP nanobody with αGFP nanobodies with varying affinity for GFP ( Figure 2D , E ) . While nanobodies exhibiting the highest affinities performed similarly , we noted a minimal affinity required for GFP transfer ( Figure 2F ) . Overall , the efficiency of GFP transfer correlated with GFP affinity . Lastly , permutation of the transmembrane domain of sGFP on the sender cell revealed that the rate of retrograde transfer of αGFP-VEGFR2-BFP from receiver to sender cells was influenced by the sGFP transmembrane domain ( Figure 2G–I ) . The PDGFR transmembrane domain minimized bidirectional transfer and thus was the optimal design for minimizing retrograde transfer which could generate false-positive signals ( Figure 2G–I ) . Collectively , the permutation of the transmembrane domains anchoring sGFP and αGFP , as well as varying the αGFP nanobody affinity identified designs that maximized unidirectional receiver cell labeling . Cancer cells interact with a variety of stromal cells at both the primary and metastatic sites ( Kota et al . , 2017; Nielsen et al . , 2016 ) . Thus , we employed the G-baToN system to record various cancer-stroma interactions in conventional 2D and 3D microfluidic culture systems as well as in vivo . Co-culturing sGFP-expressing lung adenocarcinoma cells with primary human umbilical vein endothelial cells ( HUVECs ) in a 2D format led to robust endothelial cell labeling ( Figure 3A , B ) . Additionally , within 3D microfluidic chips , pre-seeded HUVECs expressing αGFP were robustly labeled following co-incubation with sGFP-expressing lung adenocarcinoma cells ( Figure 3E–G ) . Thus , the G-baToN system is able to efficiently record cancer cell-endothelial cell interactions across multiple culture conditions . Given the importance of interactions with adaptive immune cells during carcinogenesis ( Crespo et al . , 2013; Joyce and Fearon , 2015 ) , we assessed the ability of the G-baToN system to track the interaction of primary human CD4 and CD8 T cells with lung cancer cells . αGFP-expressing CD4 and CD8 T cells that interacted with sGFP-expressing lung cancer cells in culture were specifically labeled ( Figure 4A–C ) . To test the ability of the G-baToN system to capture cancer cell-T cell interactions in vivo , we established lung tumors from a sGFP-expressing lung adenocarcinoma cell line prior to intravenous transplantation of αGFP-expressing CD4 T cells . 24 hr after T cell transplantation , over 60% of αGFP-expressing CD4 T cells within the tumor-bearing lungs were labeled with GFP , while control CD4 T cells remained unlabeled ( Figure 4D , E ) . Thus , the G-baToN system is capable of recording cancer cell-T cell interactions both in vitro and in vivo . Recent studies have demonstrated a supportive role for neurons within the primary and metastatic niche in the context of brain ( Venkatesh et al . , 2019; Zeng et al . , 2019 ) . To determine whether G-baToN can record cancer cell-neuron interactions , we co-cultured sGFP-expressing lung adenocarcinoma cells with primary cortical neurons expressing αGFP . Physical contact between cancer cells and neuronal axons led to punctate-like GFP granule transport into receiver neurons ( Figure 5A–B ) . These results demonstrate the successful application of G-baToN system to record a variety of cancer cell-stromal cell interactions . To assess the generalizability of the G-baToN system across cell types , we expressed αGFP in a panel of cell lines and primary cells . Each receiver cell type was able to uptake GFP from sGFP-expressing lung cancer sender cells upon cell-cell contact ( Figure 5—figure supplement 1A ) . Furthermore , diverse cancer cell lines and primary cell types expressing sGFP were able to transfer GFP to αGFP-expressing HEK293 receiver cells ( Figure 5—figure supplement 1B–F ) . As anticipated , receiver cell labeling required sGFP-expression on the sender cell and αGFP expression on the receiver cells . Thus , G-baToN-based labeling extends beyond transformed cell types and can label diverse primary cell types in co-culture . To further test the generalizability of the system and determine whether primary cells can serve as both sender and receiver cells , we assessed GFP transfer between interacting primary cells in the context of two well-established heterotypic cell-cell interactions: endothelial cells interacting with smooth muscle cells and astrocytes interacting with neurons . Co-culturing sGFP-expressing HUVEC and αGFP-expressing primary human umbilical vein smooth muscle cells ( HUVSMC ) resulted in efficient receiver smooth muscle cell labeling ( Figure 3C , D ) . Furthermore , sGFP-expressing astrocytes were able to transfer GFP to αGFP-expressing cortical neurons ( Figure 5C , D ) . Collectively , these results document the efficiency of G-baToN-based cell labeling across diverse cell types . Given the high efficiency with which sGFP labels receiver cells upon interaction with cognate sender cells , we tested whether other surface antigen/antibody pairs could lead to protein transfer and labeling . Due to the cross reactivity of αGFP with BFP , co-culture of surface BFP ( sBFP ) sender cells with αGFP receiver cells generated BFP-labeled receiver cells ( Fridy et al . , 2014; Figure 6—figure supplement 1A , B ) . Orthogonal systems consisting of surface-mCherry/αmCherry ( LaM4 ) ( Fridy et al . , 2014 ) and surface-GCN4-GFP/αGCN4 ( single-chain variable fragment , scFV ) ( Tanenbaum et al . , 2014 ) also led to efficient and specific receiver cell labeling ( Figure 6—figure supplement 1C–F ) . Thus , the G-baToN labeling system can be extended to additional antigen/antibody pairs . We next integrated these orthogonal systems to enable reciprocal labeling and detection of higher order multi-cellular interactions . Engineering cells with these orthogonal systems in an anti-parallel fashion should enable reciprocal labeling of both interacting cells . Co-culture of cells expressing sGFP and αmCherry with cells expressing smCherry and αGFP resulted in reciprocal labeling of both interacting cell types ( Figure 6A , B , and Figure 6—figure supplement 2A ) . This reciprocal labeling system may be particularly useful when the interaction elicits changes in both interacting cell types . Using orthogonal ligand-receptor pairs , we also created an AND gate dual labeling strategy . Specifically , co-expression of αmCherry and αGFP on receiver cells enabled dual color labeling of receiver cells that had interacted with smCherry-expressing , sGFP-expressing , or both sender cell types ( Figure 6C , D , and Figure 6—figure supplement 2B ) . Analogously , we achieved dual-color labeling of receiver cells by leveraging the ability of αGFP to bind to both sGFP and sBFP ( Figure 6E , F ) . Thus , derivatives of the G-baToN system allow for additional degrees of resolution of complex cell-cell interactions . We next extended our labeling system further by generating sender cells expressing the HaloTag protein fused to sGFP ( sHalo-GFP; Figure 7A; Los et al . , 2008 ) . Covalent attachment of a synthetic fluorophore to sHalo-GFP enabled specific loading onto sender cells ( Figure 7B ) . Co-culture of Alexa Fluor 660 ( AF660 ) -loaded sHalo-GFP sender cells with αGFP receiver cells enabled co-transfer of both GFP and AF660 ( Figure 7C ) . Compared to GFP , transfer of the chemical fluorophore using sHalo-GFP-based labeling of receiver cells led to increased signal-to-noise ratio and higher sensitivity ( Figure 7C , D ) . Importantly , changing from a protein ( GFP ) to a chemical fluorophore also extended the half-life of labeling , thus enabling partially tunable persistence of labeling after touching ( Figure 7E ) . Next , we coupled the enhanced properties of chemical fluorophore-based labeling with the generalizability of the GCN4-baToN system to assemble a robust and versatile system to label receiver cells that have interacted with two or more different sender cell types ( Figure 7F ) . Co-culturing αGCN4 receiver cells with AF488- and AF660-loaded sGCN4-Halo sender cells generated a spectrum of receiver cells with varying degrees of AF488 and AF660 labeling ( Figure 7G ) . Importantly , the ratio of AF488 to AF660 transferred to the dually labeled receiver cells strongly correlated with the ratio of the two sGCN4-Halo sender populations within the co-culture , suggesting that this system can quantitively measure higher order cell-cell interactions ( Figure 7H ) . Given the high efficiency of protein transfer using the G-baToN system , we investigated whether cargo molecules could be co-transferred with GFP from sender cells to receiver cells . In addition to the co-transfer of Halo-Tag with sGFP , we also generated sender cells with surface expression of a GFP-tdTomato fusion protein ( sGFP-Tom ) and uncovered stoichiometric tdTomato and GFP transfer to αGFP receiver cells ( Figure 8A , B ) . Beyond fluorescent labels , we tested whether other cargo could be transferred to receiver cells . We generated sGFP-PuroR-expressing sender cells and found that co-culture of sGFP-PuroR sender cells with αGFP receiver cells led to moderate puromycin resistance of touched receiver cells ( Figure 8C , D ) . Finally , loading of sGCN4-HaloTag sender cells with HaloTag-conjugated , AF647-coupled ssDNA prior to co-culture with αGCN4 receiver cells revealed successful co-transfer of fluorescently labeled ssDNA to receiver cells ( Figure 8E–G ) . Thus , baToN systems enable contact-dependent transport of different macromolecules between cells .
Here , we developed and optimized a novel cell-cell interaction reporter system and showed that this G-baToN system can record diverse cell-cell interactions in a specific and sensitive manner . Our data document the ability of diverse primary cell types to serve as both sender and receiver cells , suggesting that the G-baToN system is not only simple , sensitive and rapid , but also generalizable . Multicolor derivatives of G-baToN enable qualitative and quantitative analyses of higher order interactions involving more than two cell types . Finally , the ability to co-transfer protein , DNA and chemical cargo suggests that this platform could be leveraged to manipulate target cell function . Cancer cell-stromal cell interactions can be relatively stable ( such as a cancer cell interacts with other cancer cells or stromal cells ) or transient ( such as cancer cell-immune cell interactions and circulating tumor cell ( CTC ) interactions with endothelial cells during metastasis ) . The G-baToN system labels receiver cells through transfer of cell surface GFP which , due to its lability , ensures only transient labeling . This is similar to other cell-cell interaction labeling systems ( Figure 1—figure supplement 2 ) . Transient labeling is sufficient to label stable cancer cell-stromal cell interactions and many other diverse cell-cell interactions when sender cells consistently express GFP ( Figure 1F ) . This transient labeling should allow dynamic interactions to be detected , ensuring that the labeled receiver cells either are in contact with , or have recently interacted with sender cells . Further optimization of the G-baToN systems could allow shorter or longer term labeling within different biological systems . For example , a G-baToN system where sGFP is inducible may allow physical interactions between cancer cells and stromal cells to be captured with even more precise temporal control . Conversely , we have shown that using chemical fluorophores significantly extends label persistence within receiver cells ( Figure 7E ) , which can be used for longer term labeling of receiver cells that undergo dynamic interactions . Finally , due to the intrinsic attribute of G-baToN as a cell-cell contact dependent cargo transfer system ( Figure 8 ) , it should be possible to develop systems that allow for stable receiver cell labeling , perhaps through the transfer of site-specific recombinases ( e . g . Cre or FLP ) or programmable nucleases ( e . g . Cas9 ) to genetically modify receiver cells upon contact . There will likely be some challenges when combing NLS-signal with sGFP . Nevertheless , additional G-baToN systems will greatly facilitate study of cell-cell interaction induced cell fate determination via contact-dependent lineage tracing . As with other cell-cell interaction reporter systems , the G-baToN system relies on cell surface ligand-receptor recognition ( Supplementary files 1–2 ) . While LIPSTIC-based labeling is driven by endogenous ligand-receptor interactions , SynNotch and G-baToN systems rely on exogenous ligand-receptor pairs ( Morsut et al . , 2016; Pasqual et al . , 2018 ) . Consequently , these systems could stabilize normal cell-cell interactions . It is possible that tuning the affinity or expression level of the nanobody could minimize this effect . Inducible G-baToN systems may circumvent these issues , thus ensuring the recording of only bona fide interactions . Given that G-baToN-based cell-cell interaction reporter systems can be used as a discovery approach , the consequences of these interactions can be validated by orthogonal methods in the absence of exogenous ligand-receptor pairs . A recent paper used secreted cell-permeable mCherry for proximity labeling of cells . This system circumvents the problem of abnormal cell-cell interactions caused by ligand-receptor binding; however , this system labels all cells in proximity and lacks true cell-cell interaction labeling ( Ombrato et al . , 2019 ) . An interesting advantage of the G-baToN system is its ability to mediate cargo transfer . We demonstrated the feasibility of transferring small molecules ( HaloTag ligand , Figure 7 ) , functional proteins ( puromycin resistant protein , Figure 8C–D ) , and non-protein macromolecules ( ssDNA , Figure 8E–G ) . Transferred cargo proteins may be able to modify receiver cell signaling or promote cell death . In the future , additional design features could allow cancer cell-stromal cell interaction dependent drug delivery , cell-cell interaction facilitated sgRNA transfer between interacting cells , and digital recording of cell-cell interaction via DNA-barcode transfer . Thus , we expect the G-baToN system to facilitate an even wider array of discoveries about cell-cell interactions in cancer , across other physiological or pathological processes , and within different model organisms . The simplicity of this two-component system , combined with its generalizability across cell types , excellent foreground to background ratio , and rapid labeling , should enable facile analysis of the dynamics of cellular interaction . These types of approaches have the potential to have a broad impact on our ability to understand the outputs of cell-cell interactions in cancer and various other biological systems .
HEK-293T , B16-F10 , A549 , H460 , and HUVEC cells were originally purchased from ATCC; HUASMC were purchased from PromoCell ( C-12500 ) ; H82 cells were kindly provided by Julien Sage ( Stanford School of Medicine ) ; KP ( 238N1 ) and KPT ( 2985T2 ) lung adenocarcinoma cells were generated previously from tumors in KrasLSL-G12D; p53f/f and KrasLSL-G12D; p53f/f; R26LSL-Tom mice . HEK-293T , 238N1 , 2985T2 and B16-F10 cells were cultured in DMEM containing 10% FBS , 100 units/mL penicillin and 100 μg/mL streptomycin . A549 , H460 and H82 cells were cultured in RPMI1640 media containing 10% FBS , 100 units/mL penicillin and 100 μg/mL streptomycin . HUVECs were cultured in Vascular Cell Basal Medium ( ATCC , PCS-100–030 ) with Endothelial Cell Growth Kit ( ATCC , PCS-100–041 ) ; HUASMC were cultured in Smooth Muscle Cell Growth Medium 2 ( PromoCell , C-22062 ) . All cell lines were confirmed to be mycoplasma negative ( MycoAlert Detection Kit , Lonza ) . Pitstop ( ab120687 ) and Dyngo 4a ( ab120689 ) were purchased from Abcam . All plasmids used in this study are listed in Supplementary file 1 and key plasmids for multiple G-baToN systems will be available on Addgene . Anti-GFP antibody was purchased from MyBioSource ( MBS560494 ) , anti-RFP antibody was purchased from Rockland ( 600-401-379 ) , anti-human mitochondria antibody was purchased from Abcam ( ab92824 ) , anti-GAPDH antibody was purchased from Cell Signaling Technology ( 5174S ) . Lentiviral vectors were produced using polyethylenimine ( PEI ) -based transfection of 293 T cells with the plasmids indicated in Supplementary file 1 , along with delta8 . 2 and VSV-G packaging plasmids in 150 mm cell culture plates . Sodium butyrate ( Sigma Aldrich , B5887 ) was added 8 hr after transfection to achieve a final concentration of 20 mM . Medium was refreshed 24 hr after transfection . 20 mL of virus-containing supernatant was collected 36 , 48 , and 60 hr after transfection . The three collections were then pooled and concentrated by ultracentrifugation ( 25 , 000 rpm for 1 . 5 hr ) , resuspended overnight in 100 µL PBS , then frozen at −80°C . Parental cells were seeded at 50% confluency in a six-well plate the day before transduction ( day 0 ) . The cell culture medium was replaced with 2 mL fresh medium containing 8 µg/mL hexadimethrine bromide ( Sigma Aldrich , H9268-5G ) , 20 µL ViralPlus Transduction Enhancer ( Applied Biological Materials Inc , G698 ) and 40 µL concentrated lentivirus and cultured overnight ( Day 1 ) . The medium was then replaced with complete medium and cultured for another 24 hr ( Day 2 ) . Cells were transferred into a 100 mm cell culture dish with appropriate amounts of puromycin ( Dose used: 293T: 2 µg/mL; 238N1: 3 µg/mL; 2985T2: 2 µg/mL ) and selected for 48 hr ( Day 3 ) . After selection , FACS analysis was performed using fluorescent markers indicated in Supplementary file 2 for validation of selection efficiency . The Corning Transwell polycarbonate membrane cell culture inserts were purchased from Corning Inc ( 3422: CS , Corning , NY ) . sGFP sender cells were seeded in the upper chamber inserts of the transwell ( 1 × 105/insert ) . The inserts were then placed back into the plate pre-seeded with 1 × 105/well αGFP receiver cells and cultured in a humidified incubator at 37°C , with 5% CO2 for 24 hr . sGFP sender and αGFP receiver cells co-cultured in the same plate under the same conditions were used as control . After 24 hr , the upper chamber inserts were removed , cells in the lower chamber were trypsinized and analyzed by flow cytometry . For live cell microscopy , 2 × 104 sGFP sender and 2 × 104 αGFP receiver cells were seeded into 35 mm FluoroDish Cell Culture Dishes ( World Precision Instruments , FD35-100 ) and immediately imaged under a DeltaVision OMX ( GE Healthcare ) microscope with a 60x oil objective lens ( Olympus ) in a humidified chamber at 37°C with 5% CO2 . One image was taken per minute for 3 hr . Images were collected with a cooled back-thinned EM-CCD camera ( Evolve; Photometrics ) . For fixed cell microscopy , sender and receiver cells were seeded at the ratios indicated in Supplementary file 2 with a total number of 1 × 105 cells onto Neuvitro-coated cover slips ( Thermo Fisher Scientific , NC0301187 ) in a 12-well cell culture plate . 24 hr after co-culture , cells were fixed in 4% paraformaldehyde ( PFA ) PBS solution at room temperature for 10 min and washed with PBS and distilled water three times each , before mounting onto slides using 50% glycerol . Images were captured using a Leica DMI6000B inverted microscope with an 40x oil objective lens . For quantification , GFP-containing receiver cells were counted . Multiple coverslips were analyzed across independent experiments ( n = 10 ) . 5 × 106 sGFP sender and 5 × 106 αGFP receiver cells were co-cultured in a 100 mm cell culture dish for 24 hr . Cells were trypsinized , resuspended in FACS buffer containing propidium iodide ( PI ) ( PBS , 2% FBS , 1 mM EDTA , and 1 . 5 µM PI ) . tdTomatonegPIneg cells were sorted and lysed in RIPA buffer ( 50 mM Tris-HCl ( pH 7 . 4 ) , 150 mM NaCl , 1% Nonidet P-40 , and 0 . 1% SDS ) and incubated at 4°C with continuous rotation for 30 min , followed by centrifugation at 12 , 000 × rcf for 10 min . The supernatant was collected , and the protein concentration was determined by BCA assay ( Thermo Fisher Scientific , 23250 ) . Protein extracts ( 20–50 μg ) were dissolved in 10% SDS-PAGE and transferred onto PVDF membranes . The membranes were blocked with 5% non-fat milk in TBS with 0 . 1% Tween 20 ( TBST ) at room temperature for 1 hr , followed by incubation with primary antibodies diluted in TBST ( 1:1000 for anti-GFP , anti-Tomato ( RFP ) and anti-human mitochondria ( hu-Mito ) , 1:5000 for anti-GAPDH ) at 4°C overnight . After three 10 min washes with TBST , the membranes were incubated with the appropriate secondary antibody conjugated to HRP diluted in TBST ( 1:10000 ) at room temperature for 1 hr . After three 10 min washes with TBST , Protein expression was quantified with enhanced chemiluminescence reagents ( Fisher Scientific , PI80196 ) . To assess the stability of GFP and AF660 in touched receiver cells , 1 × 107 sGFP or sHalo-GFP sender cells were co-cultured with 1 × 107 αGFP receiver cells in a 150 mm cell culture dish for 6 hr . Cells were then trypsinized and resuspended in FACS buffer containing propidium iodide . mCherrynegPInegBFPpos cells were sorted and 1 × 105 cells were re-plated in 12-well plate and cultured for 2 , 4 , 6 , 8 , 12 , 24 , and 48 hr in DMEM containing 10% FBS , 100 units/mL penicillin and 100 μg/mL streptomycin . GFP or AF660 intensity was assessed via FACS analysis of mCherrynegPInegBFPpos cells and shown as Mean +/- SD of GFP/AF660 MFI in triplicate cultures . To validate puromycin-resistant protein ( GCN5-Related N-Acetyltransferases , PuroR ) function in sender cells , 5 × 106 HEK-293T cells were transfected with sGFP-PuroR-PDGFR , sGFP-PDGFR or sGFP-PDGFR-IRES-PuroR in 100 mm cell culture dishes for 12 hr before re-plating into a 12-well plate ( 1 × 105 cells/well ) . 24 hr after transfection , cells were treated with 1 , 2 , or 5 µg/mL puromycin for 24 hr . To count the number of viable receiver cells co-cultured with sGFP-PuroR sender cells , 2 × 105 αGFP receiver cells were co-cultured with 8 × 105 sGFP or sGFP-PuroR sender cells in a six-well plate . 24 hr after co-culture , cells were treated with 0 , 1 , and 3 µg/mL puromycin for 48 hr . Viable tdTomatonegPInegBFPpos receiver cells were counted via FACS . Mouse ( C57BL/6J , The Jackson Laboratory ) lung , kidney , heart , hindlimb skeleton muscle , spleen , and liver tissue were dissected , cut into small pieces and digested in 5 mL tissue digest media ( 3 . 5 mL HBSS-Ca2+ free , 0 . 5 mL Trypsin-EDTA [0 . 25%] , 5 mg Collagenase IV [Worthington] , 25 U Dispase [Corning] ) for 30 min in hybridization chamber at 37°C with rotation . Digestion is then neutralized by adding 5 mL ice cold Quench Solution ( 4 . 5 mL L15 media , 0 . 5 mL FBS , 94 µg DNase ) . Single-cell suspensions were generated by filtering through a 40 µM cell strainer , spinning down at 500 rcf for 5 min and washed with PBS twice . For primary mouse lung epithelial cells , kidney epithelial cells and cardiomyocyte isolation and culture , the single-cell pellets were resuspended in 1 mL FACS buffer containing 1:300 dilution of anti-EpCam-AF467 ( Biolegend , 118211 ) ( for lung and kidney epithelial cell ) or anti-Sirpa-AF467 ( Biolegend , 144027 ) ( for cardiomyocyte ) antibody and incubated on ice for 20 min before FACS sorting . DAPInegEpCampos or DAPInegSirpapos cells were sorted and seeded onto a 100 mm culture dish precoated with 5 µg/cm2 Bovine Plasma Fibronectin ( ScienCell , 8248 ) . For primary skeleton muscle cell , splenocyte and hepatocyte culture , the single cell pellets were resuspended in DMEM containing 20% FBS , 200 units/mL penicillin and 200 μg/mL streptomycin , amphotericin and cultured in 100 mm culture dish at 37°C for 1 hr to remove fibroblast cells . The supernatant containing primary skeletal muscle cells , splenocytes and hepatocytes was then transferred into a new 100 mm culture dish precoated with 5 µg/cm2 Bovine Plasma Fibronectin ( ScienCell , 8248 ) . Oligonucleotides to be conjugated with the HaloTag ligand were synthesized with a 5’ C12-linked amine and a 3’ biotin group ( IDT ) . Oligonucleotides were initially ethanol-precipitated and subsequently resuspended to 1 mM in conjugation buffer ( 100 mM Na2HPO4 ( Sigma-Aldrich S9763 ) , 150 mM NaCl ( Thermo Fisher Scientific S271 ) , pH 8 . 5 ) . Resuspended oligos were combined with an equal volume of the HaloTag ligand succinimidyl ester ( O4 ) ( Promega P6751 ) resuspended in N , N- dimethylformamide ( Sigma-Aldrich D4551 ) with a 30-fold molar excess of the ligand . Conjugation reactions were conducted overnight at room temperature with constant agitation prior to final cleanup via ethanol precipitation . Prior to loading with HaloTag-conjugated elements , sender cells were washed once in cold PBS following detachment and subsequently resuspended in cold Cell Staining Buffer ( BioLegend 42021 ) . For loading of HaloTag-conjugated fluorophores , sender cells were stained at a density of 1 . 00E+07 cells/mL on ice for 5 min in the presence of either 1 µM HaloTag-Alexa Fluor 488 ( Promega G1001 ) or 3 . 5 µM HaloTag-Alexa Fluor 660 ( Promega G8471 ) . Stained sender cells were then washed twice in Cell Staining Buffer ( 500 rcf for 5 min at 4°C ) prior to resuspension in growth media in preparation for co-culture . For loading with HaloTag-conjugated oligonucleotides , sender cells were initially resuspended and incubated with 100 µg/mL salmon sperm DNA ( Thermo Fisher Scientific 15632011 ) . Sender cells were then incubated with 3 . 5 µM HaloTag-conjugated oligonucleotides ( 5AmMC12/TCTAGGCGCCCGGAATTAGAT/3Bio ) and subsequently washed once . Oligonucleotide-loaded sender cells were then stained with 5 µg/mL streptavidin-conjugated Alexa Fluor 647 ( Thermo Fisher Scientific S32357 ) for 30 min on ice . The loaded , stained sender cells were then washed twice and resuspended in growth media in preparation for co-culture . The master mold of microfluidic chips was fabricated using a 3D printer ( Titan HD , Kudo3D Inc Dublin , CA ) . The surface of the molds was spray-coated with silicone mold release ( CRC , cat . No . : 03300 ) and PDMS ( poly-dimethyl siloxane , Sylgard 182 , Dow Corning ) was poured on it . After heat curing at 65°C for approximately 5 hr , the solidified PDMS replica was peeled off from the mold . Holes were made at both ends of each channel in the PDMS replica using a biopsy punch . The PDMS replica was then bonded to precleaned microscope glass slides ( Fisher Scientific ) through plasma treatment ( Harrick Plasma PDC-32G , Ithaca , NY ) . Microfluidic chips were UV-treated overnight for sterilization before cell seeding . A basement membrane extract ( BME ) hydrogel ( Cultrex reduced growth factor basement membrane matrix type R1 , Trevigen , Cat #: 3433–001 R1 ) was injected into the middle hydrogel channel of the chips placed on a cold pack and then transferred to rectangular 4-well cell culture plates ( Thermo Scientific , Cat #: 267061 ) followed by incubation at 37°C in a cell culture incubator for 30 min for gelation . After gelation , 10 µL of human umbilical vein endothelial cells ( HUVECs ) resuspended at the density of ~1 × 106 cells/mL was injected to the blood channel of the chips and endothelial cell growth medium was added to the other side channel . After incubation for 3 hr for cells to adhere , medium in both side channels was replaced with fresh medium . The next day , samples were placed on a rocking see-saw shaker ( OrganoFlow L , Mimetas ) that generates a pulsatile bidirectional flow to mimic the dynamic native environment and cultured for 4 more days to form a complete endothelium . Cell culture medium was changed every other day . Then , medium in the blood channel of the chips was removed and 10 µL of sGFP-expressing lung adenocarcinoma cell at the density of ~1 × 105 cells/mL was injected and cultured for 24 hr before imaging . Images were captured using an EVOS fl auto imaging system ( Life Technologies ) . Primary cortical neurons were dissociated from mouse ( C57BL/6J , The Jackson Laboratory ) E16 . 5 embryonic cortices into single-cell suspensions with a papain dissociation system ( Worthington Biochemical Corporation ) . Tissue culture plates were coated with poly-L-lysine ( 0 . 1% w/v ) before seeding cells . Neurons were grown in Neurobasal media ( Gibco ) supplemented with B-27 serum-free supplement ( Gibco ) , GlutaMAX ( Gibco ) , and penicillin-streptomycin ( Gibco ) in a humidified incubator at 37°C , with 5% CO2 . Half media changes were performed every 4–5 days . Primary astrocytes were dissociated from P0-P1 mouse cortices using the same papain dissociation methods as neurons , except the single-cell suspensions were then plated onto tissue culture plates without poly-L-lysine in DMEM with 10% FBS and penicillin-streptomycin . Primary astrocyte cultures were passaged using Accutase ( Stemcell Technologies ) . Blood from healthy donors collected in leukoreduction system ( LRS ) chambers was separated by Ficoll-Paque density gradient to obtain peripheral blood mononuclear cells ( PBMCs ) . CD4pos and CD8pos T cells were isolated by negative selection using EasySep Human CD4+ T Cell Isolation Kit and EasySep Human CD8+ T Cell Isolation Kit ( STEMCELL Technologies ) , respectively , according to the manufacturer’s instructions . T cells were cultured for 3 days with CD3/CD28 Dynabeads ( ThermoFisher Scientific ) with 40 IU/mL IL-2 and spinoculated with lentivirus for 2 hr at 400 rcf in the presence of 8 µg/mL polybrene . T cells were expanded following transduction for two days in the presence of CD3/CD28 Dynabeads and 300 IU/mL IL-2 prior to use in assays . Transduced or untransduced CD4pos or CD8 pos T cells were co-cultured for 24 hr together with A549 cells . Following co-culture , cells were harvested , stained with antibodies against CD45 , CD4 , or CD8 ( BioLegend ) and analyzed on a LSRFortessa flow cytometer ( BD Biosciences ) . Each condition was run in triplicate , and two independent experiments were conducted using T cells from different donors . For microscopy , A549 cells were co-cultured with CD4 pos T cells in glass-bottom plates ( MatTek Corporation ) and imaged on an LSM 700 confocal microscope ( Zeiss ) . Sample or experiment sizes were estimated based on similar experiments previously performed in our laboratory , as well as in the literature . For the experiments in which two or more cell types are co-cultured , we used at least three samples per group for FACS analysis , at least 10 images were taken per group for image quantification . For the experiments in which cancer cells and T cells were transplanted into mice , we used at least three mice per group . In all the experiments reported in this study , no data points were excluded . No randomization was used in this study . There was no blinding method used to assign individuals to experimental groups . Each experiment was repeated at least three times . All values are presented as mean ± SEM , with individual data points shown in the figure . Comparisons of parameters between two groups were made by two-tailed Student’s t-tests . The differences among several groups was evaluated by one-way ANOVA with Tukey-Kramer post hoc evaluation . p-Values less than 0 . 05 and 0 . 01 were considered significant ( * ) or very significant ( ** ) , respectively . | It takes the coordinated effort of more than 40 trillion cells to build and maintain a human body . This intricate process relies on cells being able to communicate across long distances , but also with their immediate neighbors . Interactions between cells in close contact are key in both health and disease , yet tracing these connections efficiently and accurately remains challenging . The surface of a cell is studded with proteins that interact with the environment , including with the proteins on neighboring cells . Using genetic engineering , it is possible to construct surface proteins that carry a fluorescent tag called green fluorescent protein ( or GFP ) , which could help to track physical interactions between cells . Here , Tang et al . test this idea by developing a new technology named GFP-based Touching Nexus , or G-baToN for short . Sender cells carry a GFP protein tethered to their surface , while receiver cells present a synthetic element that recognizes that GFP . When the cells touch , the sender passes its GFP to the receiver , and these labelled receiver cells become ‘green’ . Using this system , Tang et al . recorded physical contacts between a variety of human and mouse cells . Interactions involving more than two cells could also be detected by using different colors of fluorescent tags . Furthermore , Tang et al . showed that , alongside GFP , G-baToN could pass molecular cargo such as proteins , DNA , and other chemicals to receiver cells . This new system could help to study interactions among many different cell types . Changes in cell-to-cell contacts are a feature of diverse human diseases , including cancer . Tracking these interactions therefore could unravel new information about how cancer cells interact with their environment . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology",
"tools",
"and",
"resources",
"cancer",
"biology"
] | 2020 | A versatile system to record cell-cell interactions |
Generating recombinant monoclonal antibodies ( R-mAbs ) from mAb-producing hybridomas offers numerous advantages that increase the effectiveness , reproducibility , and transparent reporting of research . We report here the generation of a novel resource in the form of a library of recombinant R-mAbs validated for neuroscience research . We cloned immunoglobulin G ( IgG ) variable domains from cryopreserved hybridoma cells and input them into an integrated pipeline for expression and validation of functional R-mAbs . To improve efficiency over standard protocols , we eliminated aberrant Sp2/0-Ag14 hybridoma-derived variable light transcripts using restriction enzyme treatment . Further , we engineered a plasmid backbone that allows for switching of the IgG subclasses without altering target binding specificity to generate R-mAbs useful in simultaneous multiplex labeling experiments not previously possible . The method was also employed to rescue IgG variable sequences and generate functional R-mAbs from a non-viable cryopreserved hybridoma . All R-mAb sequences and plasmids will be archived and disseminated from open source suppliers .
Antibodies ( Abs ) are the workhorses of biomedical research . Enhancing the research community’s access to extensively validated Abs remains an important goal of antibody developers in both academia and industry ( Taussig et al . , 2018 ) . This has been the topic of a number of recent conferences and commentaries , led to changes in journal practices as related to transparent reporting of antibody-based research including the use of Research Resource Identifiers or RRIDs , and resulted in large-scale NIH and EU Ab development initiatives ( e . g . , the NIH Common Fund Protein Capture Reagent Program , the EU Affinomics Program ) . It is widely recognized that production of Abs in a renewable form , such as monoclonal antibodies ( mAbs ) , represents a substantial advance over polyclonal Abs from antisera that are available in finite quantity , and that comprise a heterogeneous and less definable population of Abs ( Harlow and Lane , 1988; Greenfield , 2014; Busby et al . , 2016 ) . Having mAbs available in recombinant form as recombinant mAbs ( R-mAbs ) offers numerous additional advantages ( Bradbury and Plückthun , 2015 ) . Recombinant expression systems allow for the reliable production of an R-mAb not prone to loss by genetic instability of tetra- or hexa-ploid hybridoma cells or other factors ( Andreeff et al . , 1985; Barnes et al . , 2003 ) . Moreover , unlike hybridoma cell lines that can express multiple functional heavy and light immunoglobulin chains ( Zack et al . , 1995; Blatt et al . , 1998; Bradbury et al . , 2018 ) , recombinant expression ensures production of a single , molecularly defined R-mAb . Recombinant expression can also yield production levels hundreds or even thousands of times higher than is possible with endogenous expressions of mAbs from hybridoma cells ( e . g . , Backliwal et al . , 2008; Fischer et al . , 2015; Kunert and Reinhart , 2016 ) . Furthermore , the cloning of R-mAbs provides for permanent , dependable and inexpensive archiving of the R-mAb as plasmid DNA and nucleic acid sequences versus an archiving system relying on expensive cryopreservation of hybridoma cell lines in liquid nitrogen , and their subsequent recovery as viable cell cultures . The conversion of existing mAbs to R-mAbs also allows for more effective dissemination of R-mAbs as plasmids , bacterial stocks or as DNA sequences . However , in spite of these advantages , it remains that many mAbs used for research are produced by hybridomas in culture . Modern techniques employing Abs as immunolabels [e . g . , immunoblotting ( IB ) , immunocytochemistry ( ICC ) and immunohistochemistry ( IHC ) ] utilize multiplexing of numerous Abs to simultaneously detect multiple targets within a single cell or tissue sample . This allows for direct comparison of the relative amounts and respective characteristics of multiple target molecules within the same sample , while reducing the number of samples needed to accomplish a comprehensive analysis . Typically , the individual primary Abs bound to the sample are detected with secondary antibodies conjugated to distinct reporters , most commonly organic fluorescent dyes , although enzymes , gold particles , etc . are also routinely employed as detection modalities . Multiplex labeling is often accomplished using Abs raised in distinct species , with their subsequent individual detection accomplished using species-specific secondary antibodies . However , mouse mAbs offer an important advantage for multiplex labeling procedures . Each mouse mAb is a single immunoglobulin ( Ig ) isotype , generally of the IgG class and if so specifically of a single IgG subclass , most commonly IgG1 , IgG2a or IgG2b . Mouse mAbs of distinct IgG subclasses can be robustly , reliably and specifically detected with commercial subclass-specific secondary antibodies , and , as such , can be multiplexed in a manner analogous to Abs from different species ( e . g . , Bekele-Arcuri et al . , 1996; Rhodes et al . , 1997; Lim et al . , 2000; Rasband et al . , 2001; Manning et al . , 2012 ) . One limitation to greater adoption of this approach is that mouse mAb collections generally have an extremely high representation ( ≈70% ) of IgG1 mAbs ( Manning et al . , 2012 ) , which limits the flexibility of multiplex labeling . The conversion of mAbs into R-mAbs allows for their subsequent engineering into forms with properties distinct from their parent mAb , as is routinely done to impact diverse aspects , including target binding affinity , of therapeutic R-mAbs ( Kennedy et al . , 2018 ) . Such engineering could also include switching the heavy chain constant region to impact subclass-specific secondary Ab binding specificity , an approach similar to that used to successfully modify subclass-specific in vivo effector functions ( Wang et al . , 2018 ) . There are numerous routes to obtaining validated R-mAbs , including their de novo generation from high complexity immune repertoire libraries produced from naïve or immunized animals , combined with selection of target-specific R-mAbs by in vitro display ( Bradbury et al . , 2011 ) . Alternatively , R-mAbs can be generated from existing hybridoma cell lines , which express well-characterized mAbs ( e . g . , Crosnier et al . , 2010 ) . Here we undertook conversion of a widely used collection of hybridoma-generated mAbs extensively validated for neuroscience research applications ( Bekele-Arcuri et al . , 1996; Rhodes and Trimmer , 2006; Gong et al . , 2016 ) into recombinant form . We developed a coherent pipeline of protocols for effective cloning of intact R-mAbs from cryopreserved hybridoma cells and their subsequent validation compared to their parent mAbs . Further , we developed a process to engineer these R-mAbs to IgG subclass-switched forms that provide additional utility for multiplex labeling employing mouse IgG subclass-specific secondary Abs . This approach is feasible and relatively inexpensive for any laboratory that uses standard molecular biology and mammalian cell culture techniques . This approach also represents a reliable method to convert valuable cryopreserved hybridoma collections to the immortalized form of a DNA sequence archive , including hybridomas that are no longer viable in cell culture .
We previously generated a large library of mouse mAbs that have been extensively validated for efficacy and specificity for immunolabeling endogenous target proteins in mammalian brain samples in immunoblotting ( IB ) and immunohistochemistry ( IHC ) applications ( Bekele-Arcuri et al . , 1996; Rhodes and Trimmer , 2006; Gong et al . , 2016 ) . Here , we undertook the systematic conversion of a sizable subset of this existing mAb collection to R-mAbs . We developed an innovative pipeline for R-mAb cloning , expression and validation . For the cloning steps , we built upon a previously described method ( Crosnier et al . , 2010; Müller-Sienerth et al . , 2014 ) to clone IgG VH and VL region sequences , but with the modification that we cloned directly from cryopreserved hybridomas , without the need for their labor-intensive recovery into cell culture . Our overall cloning strategy ( Figure 1A ) employs PCR-mediated amplification to generate IgG VH and VL region sequences from hybridoma-derived cDNA , followed by PCR-based fusion of these VH and VL regions with a joining fragment that contains a subset of the elements needed for high level expression of R-mAbs from transfected mammalian cell lines . The product of this fusion PCR reaction is then inserted into a plasmid containing the remainder of the elements for propagation in bacteria and expression of intact heavy and light chains in , and secretion of R-mAbs from transfected mammalian cells ( Figure 1B ) . As described in detail in the Materials and methods section , we extracted RNA from cryopreserved hybridoma cell vials , followed by first strand cDNA synthesis and RT-PCR of the IgG VH and VL region sequences ( Figure 1A ) . We amplified the IgG VL kappa ( κ ) and VH domain sequences from the hybridoma cDNA templates using a degenerate mouse Ig variable region primer set developed by Gavin Wright and colleagues ( Crosnier et al . , 2010 ) . This expanded on a previously used set ( Krebber et al . , 1997 ) that at the time of their design recognized 97% and 98% of known functional heavy and kappa ( κ ) light chain sequences , respectively . Lambda light chains were not targeted for amplification because they constitute only a small percentage of light chains used in mouse immunoglobulins ( Haughton et al . , 1978; Woloschak and Krco , 1987 ) . This PCR amplification reliably gave products of the expected 360 base pairs ( bp ) for both the VL and VH domain sequences , examples of which are shown in Figure 2A for representative mAbs N59/36 ( ‘N59’ , anti-NR2B/GRIN2B glutamate receptor ) and K39/25 ( ‘K39’ , anti-Kv2 . 1/KCNB1 potassium channel ) . To permit cloning of both the VH and VL regions into a single expression vector , fusion PCR ( F-PCR ) was performed using as templates the VH and VL PCR products , as well as a joining fragment amplified from the P1316 expression plasmid ( Crosnier et al . , 2010; Müller-Sienerth et al . , 2014 ) to produce a 2 . 4 kbp amplicon ( Figure 2B ) . The joining fragment ( Figure 1B ) contains kappa light chain constant region sequences and an associated polyadenylation signal , followed by a CMV promoter to drive VH expression , and a VH leader sequence ( Crosnier et al . , 2010; Müller-Sienerth et al . , 2014 ) . The F-PCR reaction products were then treated with NotI and AscI restriction enzymes and purified in preparation for cloning into a NotI/AscI restriction enzyme-digested fragment of the P1316 expression plasmid ( Figure 1B ) . Upstream of the NotI cloning site , the digested P1316 fragment contains a CMV promoter to drive light chain expression and a VL leader sequence , and downstream of the AscI cloning site a mouse IgG1 CH sequence , and a polyadenylation signal ( Figure 1B ) . Clones expressing the full-length IgG expression cassette were identified by colony PCR ( Figure 2C ) . Following NotI/AscI restriction digestion to verify the correct insert size ( Figure 2D ) , these plasmid clones were subjected to further analysis ( Figure 1A ) , including expression in mammalian cells , and R-mAb validation and sequencing as detailed in the subsequent sections . While a number of bona fide R-mAbs were isolated using this approach , we found a high degree of variability in the number of colony PCR- and restriction enzyme digest- verified positive plasmids that yielded functional expression . A major obstacle in cloning functionally rearranged IgG sequences from many mouse hybridomas is the presence of an aberrant kappa IgG light chain transcript expressed by the Sp2/0-Ag14 ( Sp2/0 ) hybridoma ( Carroll et al . , 1988 ) that frequently serves as a ‘myeloma’ partner for fusion with mouse splenocytes to generate mAb-producing hybridomas ( Shulman et al . , 1978 ) , and that was used as the fusion partner in all of our mAb generation efforts ( Bekele-Arcuri et al . , 1996; Gong et al . , 2016 ) . The source of this non-productive IgG light chain is the MOPC-21 myeloma cell line used to generate the Sp2/0 hybridoma ( Shulman et al . , 1978 ) . As we experienced , and as previously reported by others ( Carroll et al . , 1988 ) , aberrant chain mRNA expression varies greatly among distinct hybridoma cell lines but in certain cases can exceed the levels of functional light chain transcripts . For certain of our projects , this resulted in >90% of the colony PCR positive clones failing to produce detectable levels of functional R-mAbs , thus necessitating a high volume of screening . As such , we sought to eliminate this aberrant light chain during the cloning process . We treated the VL PCR products with the restriction enzyme BciVI . The restriction site for this enzyme is present in the VL region of the aberrant Sp2/0-derived transcript , but is predicted to occur at a low frequency in functional mouse VL kappa sequences ( Juste et al . , 2006 ) . We used VL PCR products derived from the Sp2/0 cell line and from pooled BALB/c mouse splenocytes as positive and negative controls , respectively , for sensitivity to BciVI digestion . Due to the exclusive presence of aberrant light chain in Sp2/0 cells , VL PCR products from these cells were completely digested , as shown by the decreased size of the VL PCR products from ≈360 bp typical of VL PCR products ( see Figure 2A for examples ) to the ≈180 bp fragment that results from BciVI digestion ( Figure 2E ) . In contrast , the sample prepared from the pooled mouse splenocytes was not detectably affected by BciVI digestion ( Figure 2E ) . Treatment of VL PCR products from various Sp2/0-derived hybridomas with BciVI resulted in varying degrees of digestion , yielding different proportions of the bands representing the intact VL PCR product of ≈360 bp and the cleaved aberrant SP2/0-derived VL fragment of ≈180 bp ( Figure 2E ) . After BciVI digestion was incorporated into the protocol , DNA sequencing of 149 colony PCR-positive clones from 26 different hybridomas revealed that only 12 ( 8% ) still contained the aberrant Vk light chain ( Table 1 ) . In certain cases , digestion of hybridoma VL PCR products resulted in fragments of unexpected sizes ( see the K58/35 lane in Figure 2E ) , indicating , as predicted by an earlier bioinformatics analysis ( Juste et al . , 2006 ) , that in rare cases ( in our hands , 3/248 clones pursued to this step ) the BciVI restriction site was also present in these functionally rearranged splenocyte-derived VL genes . As such , we attempted to clone these VL PCR products without BciVI treatment . As one example , the splenocyte-derived Vk light chain PCR products from the K58/35 hybridoma were sensitive to BciVI digestion , which necessitated their cloning in the absence of the BciVI digestion step . This project yielded somewhat lower frequency of clones that produced functional R-mAbs able to detect target antigen ( ≈37% ) than , on average , those containing splenocyte-derived VL PCR products refractory to BciVI digestion ( ≈48%; Supplementary file 1 ) . For the bulk of mAbs encoded by splenocyte-derived VL PCR products resistant to BciVI digestion , ligations were performed following BciVI digestion , and 10–14 candidate clones that were colony PCR-positive were selected . Plasmid DNA was digested with AscI/NotI restriction enzymes to confirm the correct insert ( Figure 2D ) . On average , ≈93% of all clones subjected to restriction analysis passed this screening step . We next transfected the colony PCR and restriction digest-validated plasmids into COS-1 cells and after a 3 to 6 day incubation , tested the conditioned culture media for production of target-specific R-mAbs . While the parent mAbs had previously been validated for efficacy and specificity in a variety of applications ( IF-ICC , IB and IHC on brain samples ) , we selected the IF-ICC assay in transiently transfected heterologous cells for R-mAb validation . This method was chosen because it is high-throughput , employing 96 well microtiter plates , it requires only a small amount of R-mAb sample , and the robust difference between target-expressing and non-expressing cells in the same sample allows for sensitivity and clarity of results . Importantly , each of the parent mAbs had been previously validated in this procedure . We took advantage of the fact that in most cases we switched the IgG subclass during conversion of the hybridoma-derived mAb into the corresponding R-mAb , allowing for their separate detection by subclass-specific secondary Abs ( Manning et al . , 2012 ) . This assay involved expressing the full-length target protein in transiently transfected COS-1 cells cultured in individual wells of a black polystyrene 96 well clear bottom plate that allows for microscopic visual analysis and imaging using indirect immunofluorescence . For each of 1 to 15 R-mAb candidate clones to be assayed from a given project , a set of replicate wells were prepared expressing the given target protein . After fixation and permeabilization , the individual wells were then immunolabeled with either the hybridoma-generated mAb alone , the R-mAb alone , or the mAb and R-mAb together . Each well was subsequently incubated with a cocktail of the two distinct subclass-specific secondary Abs , one specific for the respective mAb and one specific for the subclass-switched R-mAb mouse IgG subclasses , and both conjugated to spectrally distinct Alexa Fluors . The ‘mAb only’ well was used to demonstrate that the target protein was expressed in a subset of the transiently transfected cells , and that the only detectable secondary Ab labeling was for the IgG subclass of the parent hybridoma-generated mAb . Similarly , the 'R-mAb only' wells were used to show that the R-mAb labeled a comparable number of cells , and that the only detectable secondary Ab labeling was for the IgG subclass of the subclass-switched R-mAb . The wells containing both the parent mAb and candidate R-mAb were used to show that the mAb and R-mAb gave indistinguishable labeling patterns at both the cellular and subcellular level and could be detected separately using subclass-specific secondary Abs . Numerous examples of this assay are shown in the following sections focusing on specific R-mAb projects . We note that only in rare cases was the labeling for one or both secondaries noticeably depressed in the well containing both mAb and R-mAb relative to the wells with these primary antibodies alone , as would occur due to competitive binding to the same epitope . However , this was generally not apparent , suggesting that in this assay system , in which the target protein was overexpressed , we were typically operating under conditions of antigen excess . In comparing the overall results from 180 recent projects that had BciVI-resistant splenocyte-derived VL PCR products , and that were taken through this entire pipeline one time ( Supplementary file 1 ) , we found that a range of 1–15 colony PCR- and restriction digest-positive candidate R-mAbs per project were evaluated in the COS-ICC assay ( 919 total , mean/project = 5 . 11 ± 0 . 23 S . E . M . ) . For these 180 projects , cloning was performed after BciVI digestion , and all restriction digest-validated candidates were tested for functional mAb production . Of these 180 projects , 72% ( 129 projects ) yielded at least one positive R-mAb on their first pass through the pipeline . A retrospective analysis of these 129 successful projects revealed a range of 1 to 13 restriction digest-validated candidates were evaluated ( mean/project = 5 . 74 ± 0 . 26 S . E . M . ) , with an overall success rate of 48 . 4% for all 741 restriction digest-validated candidate R-mAbs tested . A parallel analysis of the 51 projects that did not yield a positive R-mAb on their first pass through the pipeline revealed a similar range of 1 to 15 restriction digest-validated candidates per project evaluated . However , an overall lower number of colony PCR- and restriction digest-positive candidates were evaluated in the COS-ICC assay ( 178 total , mean/project = 3 . 5 ± 0 . 37 S . E . M . ) in these projects than for the successful projects . A statistically significant difference ( two-tailed P value = 2 . 48×10−6 ) existed between the number of candidates evaluated in successful versus unsuccessful projects . The per-project COS-ICC rate success was impacted by number of clones tested . For 1–3 clones tested ( 66 projects ) , the project success rate was 50% , for 4–6 ( 62 projects ) it was ≈ 77% , and for ≥7 ( 52 projects ) it was ≈92% . However , the per clone success rate was similar between the three bins ( ≈ 45% vs . ≈39% vs . ≈ 45% ) . Following validation in the COS-ICC assay , a subset of positive clones for each project was subjected to DNA sequencing . We employed a set of sequencing primers that allowed for determination of sense and antisense strands of the VH and VL domain-encoding cDNA inserts that were unique to each R-mAb . The sequences were searched against the NCBI database , and against a custom database that contained the VH and VL domain sequences of the parent P1316 plasmid , the VL domain of the aberrant Sp2/0 cell line , and the VH and VL domain sequences of all of the R-mAbs we had cloned to date . The set of COS-ICC validated R-mAb plasmids derived from a single hybridoma that had matching sequences unique from any sequences in the custom database were subsequently archived as frozen plasmid DNA and bacterial glycerol stocks , and their sequence used as the archival sequence of that particular R-mAb . We initiated our R-mAb cloning efforts with the K28/43 mAb , a mouse mAb specific for the neural scaffolding protein PSD-95 . This mAb is widely used as a marker of excitatory synapses ( e . g . , as of 1/1/19 ≈520 publications have cited the use of the K28/43 mAb as obtained from the UC Davis/NIH NeuroMab Facility alone ) . While the K28/43 mAb is already of a less common IgG subclass ( IgG2a ) , generating a subclass switched version with an alternate IgG subclass would provide greater flexibility in its use in multiplex labeling experiments , and also provide proof of concept that we could use our process to effectively generate subclass switched R-mAbs with efficacy and specificity comparable to the parent mAb . We generated VL and VH domain cDNA fragments from the cryopreserved K28/43 hybridoma and cloned them into the original P1316 plasmid that contains a mouse IgG1 CH domain ( Crosnier et al . , 2010 ) . COS-1 cells were transiently transfected with the resultant plasmids . We evaluated the conditioned medium from the transfected cells for the presence of the subclass-switched K28/43R IgG1 R-mAb by ICC against fixed and permeabilized transiently transfected COS-1 cells expressing full-length PSD-95 ( Figure 3A ) . We note that we have designated the recombinant R-mAb versions of each of our mAbs by a capital ‘R’ after the clone designation , in this case the R-mAb cloned from the K28/43 hybridoma is designated K28/43R . We screened the COS-1 cell-generated R-mAbs for functional R-mAb immunoreactivity in the 96-well IF-ICC assay detailed in the previous section . After primary antibody labeling , all wells received both IgG1- and IgG2a-subclass-specific , fluorescently labeled secondary antibodies . As shown in Figure 3A , as expected , the sample receiving only the native K28/43 hybridoma-generated mAb exhibited a signal corresponding to the IgG2a subclass-specific secondary Ab ( red ) with no detectable signal for the IgG1 subclass-specific secondary Ab ( green ) . Conversely , labeling with the K28/43R R-mAb alone produced only an IgG1 subclass-specific green signal demonstrating a successful IgG subclass switch for the R-mAb ( Figure 3A ) . Simultaneous multiplex labeling with the hybridoma-generated mAb and a positive R-mAb resulted in an identical pattern of immunolabeling at the cellular level in the specific cells recognized , and at the subcellular level as to the pattern of immunolabeling within the labeled cells , as shown by the uniform hue of the signal in the merged panel , indicating that both the mAb and R-mAb were recognizing the same target ( Figure 3A ) . We next performed multiplex immunofluorescent IHC on adult mouse brain sections with the K28/43 mAb and the K28/43R R-mAb . As shown in Figure 3B , as detected with secondary Abs specific for their respective mouse IgG subclasses , the signal from these two primary antibodies was indistinguishable in its laminar and subcellular pattern in cerebellar cortex , consistent with previous studies of PSD-95 ( e . g . , Kistner et al . , 1993 ) . Both signals were especially intense in the terminal pinceau of basket cells located adjacent to the Purkinje cell layer , and both signals were also present in the molecular layer , and for the most part lacking in the granule cell layer ( Figure 3B ) . This demonstrates that consistent with the validation in heterologous COS-1 cells expressing exogenous PSD-95 , the K28/43 R-mAb can be used reliably for multiplex immunolabeling of endogenous PSD-95 in brain sections . The specificity of recombinant K28/43R was also assessed on immunoblots of samples from COS-1 cells exogenously expressing various representatives of the MAGUK superfamily of scaffolding proteins , of which PSD-95 is one member ( Figure 3C ) . Immunoblots were probed with a rabbit polyclonal antibody raised against PSD-95 and that also cross-reacts with SAP97 as a positive control , and recombinant K28/43R in either the IgG1 or IgG2a mouse IgG subclass form ( Figure 3C ) . Generation of the IgG2a expression plasmid is described below . Both the IgG1 and IgG2a subclass isoforms of the K28/43R R-mAb gave identical immunolabeling patterns against samples from rat brain and COS-1 cells overexpressing PSD-95 . Moreover , the pattern of R-mAb immunolabeling was similar to that obtained with the rabbit polyclonal antibody , and absent against samples of COS-1 cells exogenously expressing other MAGUK superfamily members ( Figure 3C ) . To confirm expression of these MAGUK proteins , immunoblots were probed with rabbit polyclonal anti-PSD-95 and with the hybridoma-generated K28/43 mAb ( IgG2a ) or a mAb that recognizes all mammalian MAGUK proteins mAb ( K28/86; IgG1 ) ( Rasband et al . , 2002 ) ( Figure 3C ) . These initial results demonstrated that we could use this cloning and expression process to generate and validate subclass-switched R-mAbs that recapitulate the immunolabeling characteristics of the native mAbs . Most ( ≈70% ) mouse IgG mAbs are of the IgG1 subclass ( Manning et al . , 2012 ) . Generating R-mAbs employing the efficient VH and VL cloning approach developed by Gavin Wright and colleagues includes their subsequent insertion into an expression plasmid that yields R-mAbs of the mouse IgG1 subclass due to the presence of the mouse γ1 CH domain in the plasmid backbone ( Crosnier et al . , 2010 ) . To generate R-mAbs of less common mouse IgG subclasses , we modified this plasmid by replacing the γ1 CH domain with a mouse γ2a CH domain . We amplified the γ2a CH domain sequence from cDNA generated from the hybridomas producing the K28/43 IgG2a mAb and then replaced the γ1 CH domain present in the K28/43R plasmid with this γ2a CH domain using standard cloning . This plasmid was sequenced verified and validated for expression of a K28/43R R-mAb of the IgG2a subclass ( Figure 3C ) . We subsequently used this plasmid as the target cloning vector for generation and expression of numerous R-mAb plasmids encoding functional IgG2a R-mAbs . Our strategy entailed cloning all native IgG1 and IgG2b mAbs into this plasmid to generate forms distinct from the parent native mAbs . Toward this end , we have successfully cloned and validated as functional R-mAbs a total of 178 mAbs ( Supplementary file 2 ) . This includes 148 mAbs of distinct subclasses converted to IgG2a R-mAbs ( 125 IgG1 , 21 IgG2b , and 3 IgG3 ) , and 29 IgG2a mAbs that were retained in their native IgG subclass ( Supplementary file 2 ) . We also converted the IgG2a mAb K28/43 to an IgG1 R-mAb . Each of these R-mAbs have been assigned unique RRID numbers in the Antibody Registry ( Supplementary file 2 ) and all will be deposited in plasmid form at Addgene , a subset of which are already available ( https://www . addgene . org/James_Trimmer/ ) . One benefit of subclass switching R-mAbs is the ability to perform multiplex immunolabeling not previously possible due to IgG subclass conflicts . Examples of such enrichment in cellular protein localization are shown in Figure 4 . Figure 4A shows labeling with the subclass switched IgG2a R-mAb derived from the widely used pan-voltage-gated sodium channel or ‘pan-Nav channel’ IgG1 mAb K58/35 ( Rasband et al . , 1999 ) . Like the corresponding mAb , K58/35 R-mAb gives robust labeling of Nav channels concentrated on the axon initial segment ( AIS , arrows in Figure 4A main panel ) , and at nodes of Ranvier ( arrows in Figure 4A insets ) . Importantly , subclass switching allowed Nav channel labeling at nodes to be verified by co-labeling with K65/35 an IgG1 subclass antibody directed against CASPR , a protein concentrated at paranodes ( Menegoz et al . , 1997; Peles et al . , 1997 ) . Similarly , simultaneous labeling for the highly-related GABA-A receptor β1 and β3 subunits ( Zhang et al . , 1991 ) with their widely used respective mAbs N96/55 and N87/25 could not be performed as both are IgG1 subclass . Switching the N96/55 mAb to the IgG2a N96/55R R-mAb allowed simultaneous detection of these two highly-related but distinct GABA-A receptor subunits . In Figure 4B localization of GABA-A receptor β1 and β3 subunits appeared completely non-overlapping in separate layers of cerebellum . Figure 4C illustrates localization of protein Kv2 . 1 and AnkyrinG in separate subcellular neuronal compartments . Labeling for the K89/34R R-mAb ( IgG2a , red ) , specific for the Kv2 . 1 channel , highly expressed in the plasma membrane of the cell body and proximal dendrites ( arrows in panel C1 ) is shown together with labeling for N106/65 ( green ) , an IgG1 mAb specific for AnkyrinG , a scaffolding protein highly expressed in the AIS ( arrows in panel C2 ) and at nodes of Ranvier . Subclass switching N229A/32 ( IgG1 , GABA-AR α6 ) to IgG2a , allows comparison with Kv4 . 2 potassium channel ( K57/1 , IgG1 ) in the cerebellum where both are highly expressed in the granule cell layer ( Figure 4D ) . While both are prominently found in the glomerular synapses present on the dendrites of these cells , simultaneous labeling reveals that some cells express both ( Figure 4D , magenta ) while others appear to predominantly express Kv4 . 2 ( Figure 4D , blue ) . Labeling for both proteins is in contrast to that for mAb N147/6 ( IgG2b ) which recognizes all isoforms of the QKI transcription factor and labels oligodendrocytes within the granule cell layer and throughout the Purkinje cell layer ( PCL , green ) . In Figure 4E , localization of pan-QKI ( N147/6 , IgG2b , blue ) is compared with GFAP ( N206A/8 , IgG1 , green ) predominantly thought to be in astrocytes . Surprisingly many ( but not all ) cells co-label both proteins . We also labeled cortical neurons with these two mAbs ( pan-QKI in blue , GFAP in green ) . Multiplex labeling for the neuron-specific Kv2 . 1 channel , using subclass-switched K89/34R ( IgG2a , red ) confirms non-neuronal localization of both proteins . Lastly , we labeled for the postsynaptic scaffold protein PSD-93 using R-mAb N18/30R ( IgG2a , red ) , which in the cerebellum is prominently localized to Purkinje cell somata and dendrites ( Brenman et al . , 1998 ) . As shown in Figure 4F , the R-mAb labeling is consistent with the established localization of PSD-93 . Because of subclass switching the N18/30R R-mAb , this labeling can now be contrasted with labeling for the excitatory presynaptic terminal marker VGluT1 , labeled with mAb N28/9 ( IgG1 , blue ) , which exhibits robust labeling of parallel fiber synapses in the molecular layer , and glomerular synapses in the granule cell layer . Together these results demonstrate the utility of employing subclass-switched R-mAbs to obtain labeling combinations not possible with native mAbs . While not a common event , mAb-producing B cell hybridomas can lose their mAb production due to genetic instability ( Morrison and Scharff , 1981; Frame and Hu , 1990; Castillo et al . , 1994 ) , mycoplasma contamination leading to amino acid depletion and cytopathic effects ( Drexler and Uphoff , 2002 ) , or other factors . Non-optimal cryopreservation due to problems during the freezing process itself or inadequate storage conditions can result in non-recoverable frozen seeds . As the method described here for cloning of VH and VL domain sequences from cryopreserved hybridomas does not require expansion of cells in culture prior to RNA extraction , we speculated that it may be possible to use this approach to generate functional R-mAbs from even those hybridoma cell lines that are no longer viable in cell culture . We had in our cryopreserved hybridoma archive a hybridoma cell line that produced the D3/71 mAb . This hybridoma cell line had been generated in 1994 among a series of projects targeting the Kv2 . 1 voltage-gated potassium channel α subunit , projects that also yielded the D4/11 mAb ( Bekele-Arcuri et al . , 1996 ) . The D3/71 mAb has particular value in binding at a site distinct from other anti-Kv2 . 1 mAbs , which unlike most other available anti-Kv2 . 1 mAbs ( and pAbs ) , remains intact in truncated isoforms of Kv2 . 1 found in patients with severe neurodevelopmental disorders linked to de novo frameshift or nonsense mutations in the KCNB1 gene ( de Kovel et al . , 2016; Marini et al . , 2017 ) . Unlike our experience with other cryopreserved hybridomas in our collection , when we attempted to resuscitate this cryopreserved hybridoma cell line over 20 years after it was cryopreserved , the D3/71 hybridoma cells were no longer viable . We extracted RNA from a cryopreserved vial of D3/71 hybridoma cells in an attempt to generate a functional D3/71R R-mAb . Cloning was performed as described above including amplification of D3/71 VH and VL domain cDNAs ( Figure 5A–B ) and their insertion into the mouse IgG2a expression plasmid . Most of the bacterial colonies tested ( 20/24 ) were positive for the insert by PCR ( Figure 5C ) , and subsequent restriction analysis of 12 of these positive clones confirmed that they each contained the full-length VL-joining fragment-VH cassette ( Figure 5D ) . We next tested whether these plasmids encoded functional D3/71R R-mAb in the COS-1-ICC assay , using a distinct native anti-Kv2 . 1 IgG1 mAb K89/34 ( Misonou et al . , 2004 ) and the subclass-switched K89/34R R-mAb ( Figure 6B ) as controls ( Mandikian et al . , 2014; Bishop et al . , 2015 ) . A subset of the recovered and subclass-switched D3/71R R-mAbs were positive in this assay ( Figure 6A ) . Moreover , the D3/71R R-mAb exhibits immunolabeling of endogenous Kv2 . 1 in neurons in mouse neocortex that recapitulates the previously established pattern of Kv2 . 1 immunolabeling in these neurons ( e . g . , Bishop et al . , 2015; Trimmer , 1991; Rhodes et al . , 1995 ) and that overlaps precisely with simultaneous immunolabeling obtained with the native K89/34 mAb ( Figure 6C ) . The D3/71R R-mAb also yields an immunoreactive band on immunoblots similar to that obtained with K89/34 and K89/34R ( Figure 6D ) . These results suggest that the mAb cloning approach described here can be used to effectively recover functional R-mAbs from non-viable hybridomas and could be applied by researchers that have such hybridomas in their collections .
In this study , we developed a novel R-mAb generation and validation procedure that we used to generate a valuable R-mAb resource by converting a library of widely used mouse mAbs into recombinant reagents . Our approach was adapted from that of Gavin Wright and colleagues ( Crosnier et al . , 2010 ) with modifications to facilitate higher throughput cloning and validation steps . The efficiency of the process was further enhanced by restriction digest of VL PCR products to eliminate the MOPC-21 derived aberrant light chain . R-mAbs were produced in the culture medium from transiently transfected cells under standard mammalian cell culture conditions in sufficient quantities that they did not require purification for effective use in IB , ICC , and IHC assays . We were also able to recover functional R-mAbs from a cryopreserved hybridoma cell line that was no longer viable in cell culture via RNA extraction , cloning and expression . Moreover , in the process we generated subclass-switched R-mAbs that are more amenable to multiplex labeling in IB , ICC , and IHC applications . We then employed these mAbs in multiplex IHC experiments in combinations that were not possible prior to their cloning and expression as subclass-switched R-mAbs . Using the method described here , most researchers with standard molecular biology expertise can generate functional R-mAb expression plasmids with the advantages associated with recombinant reagents from their own cryopreserved hybridomas , whether viable or non-viable . Moreover , producing R-mAbs from cells transfected with any of the large collection of expression plasmids whose generation is described here is within the reach of anyone with standard mammalian cell culture capabilities . We previously undertook a systematic analysis of anti-mouse IgG subclass-specific secondary antibodies and demonstrated that the use of mouse mAbs of different IgG subclasses in combination with anti-mouse IgG subclass-specific secondary antibodies for simultaneous multiplex immunolabeling allows for robust and specific labeling in several immunolabeling applications , and also provides enhanced signal-to-noise ( background ) in immunolabeling of brain samples when compared to generic anti-mouse IgG ( H + L ) secondary Abs ( Bekele-Arcuri et al . , 1996; Rhodes et al . , 1997; Lim et al . , 2000; Rasband et al . , 2001; Manning et al . , 2012 ) . Simultaneous multiplex labeling of multiple targets within a single sample allows for better comparison of their relative levels , co-localization , and overall tissue architecture . Multiplex labeling also conserves valuable samples , as multiple targets can be interrogated in the same sample instead of across multiple samples , which can be a particular concern in studies using human tissue samples that may be available in limited quantity . However , multiplex labeling using the classical approach of employing primary Abs raised in different species and their subsequent detection with species-specific secondary Abs is highly limited by the lack of available Abs from numerous species . As one example , interrogating the Abs registered with the Antibody Registry ( 2 , 381 , 068 as of 9/9/18 ) shows a preponderance ( 84% in total ) of Abs raised in only three species: rabbits ( 44 . 6% ) , mice ( 31 . 2% ) and goats ( 8 . 2% ) , with all other species together accounting for the remaining 16% . Among the predominant species , goat Abs are exclusively polyclonal and as such cannot be used with one another in simultaneous multiplex labeling . Moreover , the bulk of secondary Abs are raised in goats , constraining facile detection of primary Abs in multiplex labeling experiments employing goat primary Abs . Rabbit Abs ( whether poly- or mono-clonal ) have limited utility for multiplex labeling , as unlike most other mammals , rabbits do not make IgGs of distinct subclasses but only a single generic IgG ( Knight et al . , 1985 ) . As such employing multiple goat or rabbit primary Abs for multiplex immunolabeling involves costly and time-consuming procedures such as direct labeling with different fluorophores or sequential multiplexing technologies such as Opal ( Stack et al . , 2014 ) . While the utility of mouse mAbs ( and less common rat mAbs ) for simultaneous multiple immunolabeling is enhanced by the availability of subclass-specific secondary antibodies , it remains that mouse mAbs against any given target may be available in only a single IgG subclass . In general , mouse mAb collections reflect the representation of these subclasses in the circulating serum IgGs in immunized BALB/c mice , which is ≈70% IgG1 , ≈20% IgG2a and ≈10% IgG2b ( Natsuume-Sakai et al . , 1977 ) . As a prominent example , the large mouse mAb catalog of ThermoFisher Scientific ( 10 , 992 independent IgG mouse mAbs as of 6/28/18 ) follows this remarkably closely , comprising 69% IgG1 , 18% IgG2a and 12% IgG2b ( Matt Baker , ThermoFisher Scientific , personal communication ) . As such , the flexibility of using mouse mAbs in multiplex labeling is restricted by the prevalence of those of the IgG1 subclass . It is possible to selectively screen for mAbs of the rarer IgG2a and IgG2b subclasses of mouse IgGs in the process of their development and screening ( Gong et al . , 2016; Liu et al . , 2015 ) . It is also possible , although labor-intensive , to manipulate hybridomas in culture and select and/or screen for subclass switched hybridoma-generated mAbs ( Faguet and Agee , 1993 ) . However , it remains that the optimal utility of mouse mAbs in research and diagnostics is limited by the preponderance of IgG1 mAbs . Here , we have enhanced the flexibility of a substantial fraction of an existing library of widely used mouse mAbs by converting them to subclass-switched R-mAbs without altering target binding specificity . We show using mixtures of mAbs and R-mAb that we can obtain effective simultaneous multiplex labeling for combinations that were previously unattainable due to conflicting IgG subclass . Antibody reformatting in our system is easily accomplished by the choice of plasmid backbone used for the original cloning , or by subcloning the VL-joining fragment-VH cassette ( Figure 1B ) between plasmids with distinct CH domains . Ultimately , one can envision using this approach to convert any mouse mAb into the corresponding set of mouse IgG1 , IgG2a , and IgG2b subclass R-mAb expression plasmids . Multiplex labeling could be further expanded by constructing IgG expression vectors containing CH regions of other species that have IgG subclasses such as rat and human ( but not rabbit , in which all IgG Abs are of a generic IgG class ) , for which subclass-specific secondary antibodies are also widely available . For example , we interrogated the ThermoFisher secondary Ab catalog , which contains 52 different IgG subclass-specific secondary antibodies for mouse IgG subclasses , 47 for rat IgG subclasses , and 43 for human IgG subclasses . Together , the wide availability of such a broad range of reliable secondary antibodies with distinct conjugates provides ample flexibility for specific detection of multiple R-mAbs with distinct CH regions engineered for separate detection in multiplex labeling experiments . Previous attempts by others to generate R-mAbs have been complicated by the MOPC-21 derived aberrant kappa light chain that is present in many widely used myeloma fusion partners ( Crosnier et al . , 2010; Cochet et al . , 1999; Duan and Pomerantz , 1994; Yuan et al . , 2004 ) . For example , Crosnier et al . , ( Crosnier et al . , 2010 ) showed that depending on the hybridoma , 26% to 70% of R-mAb plasmids they generated contained this aberrant sequence . We have incorporated a simple restriction enzyme digest into our process to eliminate the aberrant kappa light chain ( Juste et al . , 2006 ) . For most of our hybridomas , the majority of the VL PCR product was cleaved by BciVI , indicating that the MOPC-21 aberrant light chain comprised the greatest fraction of total light chain amplicons . By eliminating this transcript , the overall screening and cloning burden for each mAb was reduced substantially . However , depending on the particular hybridoma , the efficiency of generating clones that produced functional R-mAbs even after BciVI digestion of the VL domain PCR products was still highly variable . This was evidenced by the overall efficiency of slightly below 50% for the successful projects tabulated in Supplementary file 1 , and also by the projects that were unsuccessful on the first pass in which clones passed the colony PCR and restriction digest screening steps but failed in the COS-ICC screening assay for functional R-mAbs . Moreover , a lack of allelic exclusion at heavy and light chain IgG loci has been documented in hybridomas , resulting in expression of multiple IgG heavy and/or light chains at the mRNA and protein levels ( Zack et al . , 1995; Blatt et al . , 1998; Bradbury et al . , 2018; Ruberti et al . , 1994 ) . In these previous studies , while multiple combinations of heavy and light IgG chains were produced by monoclonal hybridoma cells , only one combination of heavy and light IgG chains present in the hybridoma produced an R-mAb capable of recognizing the target antigen . Varying degrees of allelic exclusion in the hybridomas that served as the source of our R-mAbs could also contribute to the variable efficiency of whether colony PCR and restriction enzyme assay-positive clones would produce an R-mAb that recapitulated immunolabeling obtained with the native mAb . Technical issues , such as PCR-induced mutations may have also contributed to the generation of R-mAb clones that failed in the COS-IF-ICC assay . We did not systematically analyze these negative clones , so we do not know whether these plasmids contain residual aberrant light chains from the BciVI digestion step , plasmids containing other heavy and/or light chains , or mutated forms of bona fide R-mAb chains with impaired/lost functionality . That the vast majority ( 84% ) of the 114 projects for which ≥ 4 clones were evaluated yielded at least one positive R-mAb suggests that the presence of a small and variable number of negative clones does not impact the overall success of the approach described here , given the simple but effective nature of microplate screening assays , such as the ICC assay employed here , to rapidly and easily screen candidate R-mAbs . There is a growing impetus to enhance research reproducibility by both improving the quality of Abs used in basic research and in diagnostics , and the transparency of their reporting as related to the exact molecularly-defined Ab that was used ( Taussig et al . , 2018; Bradbury and Plückthun , 2015; Uhlen et al . , 2016 ) . Using mAbs in their recombinant form and defining them unambiguously at the molecular level by publishing their VL and VH domain sequences is a step towards achieving this goal and elevate reproducibility world-wide . Among the different forms of research and diagnostic Abs , polyclonal Ab preparations have unique benefits ( Ascoli and Aggeler , 2018 ) , due to the presence of Abs recognizing distinct epitopes on the target protein . However , in any of the diverse forms in which they are available ( antiserum , IgG fractions , affinity-purified ) polyclonal Abs are inherently limited by their intrinsic lack of precise molecular definition . The inability to rigorously define the clonal composition of specific preparations can contribute to batch-to-batch variation in the efficacy and specificity of the polyclonal collection for any specific application , negatively impacting the reproducibility of research performed with such Abs . Moreover , unlike mAbs , polyclonal Abs are a finite resource whose depletion can lead to batch-to-batch variability or complete lack of availability , stymieing independent reproduction of research using the same reagent . Conventional mAbs are advantageous in lacking the molecular complexity of polyclonal preparations , and in being renewable reagents . However , the conversion of conventional hybridoma-generated mAbs into R-mAbs offers numerous additional advantages . This includes ensuring their broad availability by generating a publicly accessible DNA sequence archive that can be used to synthesize the functional VL and VH domain sequences . R-mAbs are also more easily distributed as plasmids or DNA sequence , a feature that should enhance dissemination of this resource , relative to the more labile cryopreserved and/or living hybridomas that are less amenable to long-distance transport . Cloning also enhances research reproducibility by the unambiguous definition of mAbs at the level of their cDNA sequence , as well as the enhanced control of molecular composition afforded by expressing a single light and heavy chain combination from transfected cells and their high-level animal-free production in cell culture . Cloning of mAbs also allows for increased utility as offered by their subsequent engineering into alternate forms , such as the subclass-switched R-mAbs generated here . The method described here represents a straightforward approach to producing functional R-mAbs from existing murine hybridomas that could be applied to other valuable mAb collections to ensure their permanent archiving , which cannot be assured when they exist solely in the form of hybridomas cryopreserved in liquid nitrogen . In particular , there are many cases of important hybridomas , including entire collections , that have been discarded upon the closure of the laboratory that formerly housed and financially supported the maintenance of the cryopreserved collection . The enhanced utility , permanence , and cost-effectiveness of R-mAbs as generated by the approach described here is one route to circumvent the negative impact to research effectiveness and reproducibility that such losses represent . Simple conversion of mAbs to R-mAbs , as described here , also paves the way for higher throughput approaches , for example those employing high throughput ‘next generation’ sequencing to simultaneously obtain light and heavy chain sequences from large pools of hybridomas ( Chen et al . , 2018 ) . In theory this should allow for larger scale ‘en masse’ ( as opposed to the mAb-by-mAb approach used here ) conversion efforts more suitable to large hybridoma collections . For example , our larger collection of ≈40 , 000 target-specific mAbs archived in our cryopreserved hybridoma collection , and other widely used collections , such as that housed at the Developmental Studies Hybridoma Bank ( http://dshb . biology . uiowa . edu/; comprising 3 , 672 mAbs as of 9/9/18 ) might represent appropriate targets for such next generation sequencing approaches . Future efforts to generate R-mAbs from cryopreserved hybridoma collections will further enhance the rigor , reproducibility and overall effectiveness of Ab-based research .
Primer sets for mouse Ig V region amplification and fusion PCR ( F-PCR ) were used as described previously ( Crosnier et al . , 2010; Müller-Sienerth et al . , 2014 ) . The following primers were used for other PCR steps . Amplification of the Joining Fragment ( Crosnier et al . , 2010 ) Primer 21: 5’- GGGCTGATGCTGCACCAACTGTA-3’ Primer 26: 5’-ACTGCTTGAGGCTGGACTCGTGAACAATAGCAGC-3’ Colony PCR: UpNotI: 5’-TTTCAGACCCAGGTACTCAT-3’ DownAscI: 5’-GGGCAGCAGATCCAGGGGCC-3’ ( reverse primer for IgG1 vector ) Rev IgG2a: 5’- ACCCTTGACCAGGCATCCTAGAGT- 3’ ( reverse primer for IgG2a vector ) Mouse γ2a CH domain amplification ( restriction sites are underlined ) : IgG2a-F-AscI: 5’-ATATCACGGCGCGCCCAACAGCCCCATCGGTCTATCCA-3’ IgG2a-R-XbaI: 5’GACTGATCTAGATCATTTACCCGGAGTCCGGGAGAA-3’ R-mAb Sequencing: Forward strand of VL region: UpNotI = 5’-TTTCAGACCCAGGTACTCAT-3’ Reverse strand of VL region ( IgG2a plasmids ) : Seq_VL_Rev_IgG2a = 5' - CCAACTGTTCAGGACGCCATT −3' Forward strand of VH region: VH_seq_Forward = 5'- TCCCAGGCCACCATGAA −3' Reverse strand of VH region ( IgG1 plasmids ) : DownAscI = 5’-GGGCAGCAGATCCAGGGGCC-3’ Reverse strand of VH region ( IgG2a plasmids ) : Rev IgG2a = 5’- ACCCTTGACCAGGCATCCTAGAGT- 3’ The Ambion RNAqueous 96 Extraction kit ( Thermo Fisher Cat# AM1920 ) was used for high throughput RNA extraction . Frozen vials containing 0 . 5–1 × 107 hybridoma cells per vial were thawed in a 37°C water bath for 5 min , in batches of 20 vials for high through-put purposes . Cells were spun down in a table top centrifuge at 2000 rpm for 5 min , the supernatant was removed , and cells were washed with 1 mL cold PBS . A 250 µL aliquot of this cell suspension , representing 1–3 × 106 cells , was used for RNA extraction according to the manufacturer’s instructions . We replica plated using 4 . 0 µL of RNA ( 1/16 of the extract volume ) to a second 96 well plate and used the Superscript III Reverse Transcriptase First Strand Synthesis System ( Thermo Fisher Cat# 18080051 ) for high throughput cDNA synthesis using oligo ( dT ) primers . Following the cDNA synthesis reaction , we replica plated ≈5% of the volume ( 1 . 0 µL of 10-fold dilution ) of the cDNA synthesis product to a third 96 well plate to serve as the template for PCR amplification of the IgG VH and VL domain sequences . As such , the templates for the VH and VL domain PCR amplification represented the cDNA yield from ≈3 , 000–7 , 500 hybridoma cells . Amplification of Ig V region sequences and fusion PCR ( F-PCR ) to join VL and VH PCR products were performed as described ( Crosnier et al . , 2010; Müller-Sienerth et al . , 2014 ) with the noted modifications . Briefly , degenerate primer sets were used to amplify mouse Ig kappa VL and VH sequences using PFU Ultra II Fusion HS DNA polymerase ( Agilent Technologies Cat# 600670 ) and Advantage 2 Polymerase Mix ( ClonTech Cat# 639201 ) for VL and VH amplification , respectively . PCR conditions for VL amplification were: 95°C for 5 min; 5 cycles of 95°C for 20 s , 60°C for 20 s , 72°C for 30 s; 19 cycles of 95°C for 20 s , 60 . 5°C for 20 s with 0 . 5°C decrement per cycle , 72°C for 30 s; 10 cycles of 95°C for 20 s , 55°C for 20 s , 72°C for 30 s; 72°C for 15 min . PCR conditions for VH amplification were: 95°C for 5 min; 5 cycles of 95°C for 45 s , 62°C for 30 s , 72°C for 1 min; 19 cycles of 95°C for 45 s , 64 . 5°C for 30 s with 0 . 5°C decrement per cycle , 72°C for 1 min; 10 cycles of 95°C for 45 s , 55°C for 30 s , 72°C for 1 min; 72°C for 15 min . VL PCR products ( 7 . 0 µL per reaction ) were digested with 5 units of the restriction enzyme BciVI ( BfuI ) ( Thermo Fisher Cat# ER1501 ) in a 20 µL reaction at 37°C for 2 hr . The enzyme was inactivated by heating to 80°C for 20 min . In preparation for fusion of the VL and VH PCR products , a joining fragment was produced using the mouse Ig expression plasmid P1316 ( a gift of Dr . Gavin Wright , Sanger Institute , Cambridge , UK , now available from Addgene as plasmid #28217 ) . The joining fragment contains the following sequences ( 5’ to 3’ ) : mouse kappa light chain constant region , a polyadenylation signal , a CMV promoter , and the leader sequence for the Ig heavy chain . P1316 was used as a template in a 50 µL PCR containing 0 . 2 µM of primers 21 and 26 , 0 . 2 mM dNTPs , and 1 . 0 µL PFU Ultra II Fusion HS DNA polymerase . PCR conditions were: 95°C for 5 min; 5 cycles of 95°C for 20 s , 60°C for 20 s , 72°C for 45 s; 19 cycles of 95°C for 20 s , 60 . 5°C for 20 s with 0 . 5°C decrement per cycle , 72°C for 45 s; 10 cycles of 95°C for 20 s , 55°C for 20 s , 72°C for 45 s; 72°C for 15 min . The 1 . 7 kb joining fragment was purified ( Qiagen/QiaQuick PCR Purification Cat# 28106 ) in preparation for F-PCR . VL ( BciVI restriction enzyme digested ) , the joining fragment , and VH PCR products were joined via F-PCR in a 96-well format . We observed that purification of VL and VH PCR products was not necessary . Each 50 µL reaction consisted of the following: 0 . 2 µM of primers 51 and 52 ( Crosnier et al . , 2010; Müller-Sienerth et al . , 2014 ) , 0 . 2 mM dNTPs , 1 . 5 µL VL ( BciVI digested ) , 0 . 5 µL VH , 0 . 5 µL purified joining fragment ( 50 ng ) , and 1 . 0 µL PFU Ultra II Fusion HS DNA polymerase . PCR conditions were: 95°C for 2 min; 11 cycles of 95°C for 45 s , 63°C for 30 s , 72°C for 5 min; 7 cycles of 95°C for 45 s , 62°C for 30 s with 1°C decrement per cycle , 72°C for 5 min , 95°C for 45 s; 26 cycles of 56°C for 30 s , 72°C for 5 min; 72°C for 15 min . F-PCR products were digested with FastDigest NotI and AscI ( Thermo Fisher Cat# ER0595 and ER1891 , respectively ) at 37°C for 20 min , followed by inactivation at 80°C for 5 min and column purification ( Qiagen/QiaQuick PCR Purification Cat# 28106 ) . The P1316 plasmid was also NotI/AscI digested and gel purified ( Qiagen/QiaQuick Gel Extraction Cat# 28706 ) . P1316 is a derivative of the pTT3 expression vector ( Durocher et al . , 2002 ) and consists of ( 5’ to 3’ ) : a CMV promoter , the mouse V kappa leader sequence , a NotI restriction site , an insert consisting of VL/joining fragment/VH , and the mouse IgG1 CH sequence amplified from mouse genomic DNA , flanked by AscI and XbaI restriction sites ( Crosnier et al . , 2010; Müller-Sienerth et al . , 2014 ) . Ligation was performed overnight at 16°C with T4 DNA ligase ( Thermo Fisher Cat# 15224017 ) using 20 ng insert and 20 ng vector , a 3:1 molar ratio . Half of each ligation reaction was used to transform 25 µL of Mach I chemically competent E . coli . ( Thermo Fisher Cat# C862003 ) . Cells and DNA were incubated on ice for 30 min , heat shocked at 42°C for 30 s , incubated on ice for 2 min , and allowed to recover for 1 . 0 hr in 250 µL SOC medium in a 37°C shaking incubator . The cells were spun in a centrifuge at 3000 rpm ( ≈950 x g ) for 2 min and the supernatant was removed until 150 µL remained . Cells were resuspended , and the entire volume was plated on LB plates containing 100 µg/mL ampicillin and incubated overnight at 37°C . Colony PCR was performed to identify E . coli colonies that contained the full-length , 2 . 4 kb Ig cassette . Colonies were diluted in 96-well plates containing 50 µL water and patch plates were made for later recovery of positive clones . 2 µL of diluted colony was used in each PCR . Conditions were 94°C for 5 min; 23 cycles of 94°C for 20 s , 58°C for 30 s , 72°C for 2 . 5 min; 72°C for 10 min . For additional confirmation of the presence of the full-length Ig cassette , plasmid DNA was isolated from PCR positive clones and subjected to restriction enzyme digestion with NotI/AscI at 37°C for 20 min followed by agarose gel electrophoresis . The VL and VH regions of functional R-mAbs were subjected to sequencing in both orientations to generate a permanent archive . The primers ‘UpNotI’ and ‘Seq_VL_Rev_IgG2a’ were used for VL domain sequencing , and the ‘VH seq forward’ , and either ‘DownAscI’ or ‘Rev IgG2a’ for sequencing of VH regions in the IgG1 or IgG2a expression plasmids , respectively . The mouse γ2a CH domain was amplified from the cDNA preparation that was obtained from the K28/43 ( RRID: AB_2292909 ) hybridoma ( Rasband et al . , 2002; Shibata et al . , 2003 ) that was used for cloning of the K28/43 VL and VH domains . PCR conditions were: 94°C for 5 min; 29 cycles of 94°C for 30 s , 65°C for 30 s , 72°C for 30 s . The forward primer included an AscI restriction site and the reverse primer included an XbaI restriction site to facilitate cloning into the K28/43 IgG1 recombinant R-mAb plasmid . The K28/43 IgG1 recombinant R-mAb plasmid was derived from the P1316 plasmid ( Crosnier et al . , 2010 ) by restriction enzyme-based cloning of K28/43 variable region sequences as described above . The IgG2a CH PCR product and the K28/43 IgG1 recombinant plasmid were both digested with AscI and XbaI restriction enzymes ( New England BioLabs Cat# R0558 and R0145 , respectively ) and column purified ( Qiagen/QiaQuick PCR Purification Cat# 28106 ) or agarose gel purified , respectively ( Qiagen/QiaQuick Gel Extraction Cat# 28706 ) . Because XbaI is methylation sensitive , the K28/43 IgG1 plasmid was sourced from dam-/dcm- E . coli ( New England Biolabs Cat# C2925 ) . T4 DNA ligase ( New England Biolabs Cat# M0202 ) was used to insert the IgG2a CH fragment into the digested K28/43R plasmid to generate a K28/43 IgG2a R-mAb plasmid , which was confirmed by DNA sequencing . COS-1 cells ( ATCC Cat No CRL-1650; RRID:CVCL_0223 ) were used for R-mAb expression . To rule out inter-species contamination , cells were authenticated at ATCC as being from African Green Monkey by a PCR based method to detect species-specific variants of the cytochrome C oxidase I gene ( COI analysis ) . Cells were tested in house for mycoplasma contamination using the MycoAlert Mycoplasma Detection Kit ( Lonza Catalog#: LT07-318 ) . For production of recombinant mAbs in mammalian cell culture , 3 × 105 COS-1 cells were plated on 35 mm tissue culture dishes and cultured overnight in DMEM ( high glucose/pyruvate , Thermo Fisher Cat# 11995065 ) with 10% Fetal Clone III ( HyClone Cat# SH30109 . 03 ) and 100 µg/ml penicillin/streptomycin ( Thermo Fisher Cat# 15140122 ) . Cells were then transfected with a 1:1 ratio of plasmid ( 1 µg ) :Lipofectamine 2000 ( 1 µL ) ( Thermo Fisher Cat# 11668019 ) diluted in Opti-MEM reduced serum medium ( Thermo Fisher Cat# 31985070 ) . Following an overnight incubation , the transfection solution was replaced with culture medium , and the cells incubated for an additional 3–6 days , at which time the conditioned medium was collected as R-mAb tissue culture supernatant ( TC supe ) for analysis . In certain cases , R-mAbs were produced from COS-1 cells cultured in 12 well plates , using proportionally reduced amounts of cells ( 1 . 5 × 105 ) plasmid ( 0 . 5 µg ) :Lipofectamine 2000 ( 0 . 5 µL ) and Opti-MEM . R-mAb TC supes were screened for immunoreactivity in an immunofluorescence assay against transiently transfected COS-1 cells cultured in 96-well plates . COS-1 cells were plated in black , clear bottom 96-well plates ( Greiner Cat# 655090 ) at a density of 4 , 700 cells/well . After overnight incubation , each well received 50 ng plasmid DNA encoding the R-mAb target protein plus Lipofectamine 2000 at a 1:1 ratio as described above . On day three post-transfection , cells were washed three times with DPBS ( 138 mM NaCl , 2 . 67 mM KCl , 1 . 47 mM KH2PO4 , 8 . 1 mM Na2HPO4 , 1 mM CaCl2 and 1 mM MgCl2 ) , pH 7 . 4 and then fixed using 3 . 0% formaldehyde ( prepared fresh from paraformaldehyde ) in in DPBS plus 0 . 1% Triton X-100 on ice for 20 min . Cells were washed three times with DPBS/0 . 1% Triton X-100 , blocked with Blotto/0 . 1% Triton X-100 for 1 hr , and stored in DPBS/0 . 02% sodium azide . For primary antibody labeling , R-mAb TC supes were used without dilution and hybridoma-generated mAb TC supe controls ( see Table 2 for details of non-R-mAb Abs used in this study ) were diluted 1:10 in COS-1 cell culture medium . Each R-mAb was tested alone and in combination with the corresponding hybridoma-generated mAb TC supe . Primary antibodies were incubated at room temperature for 1 hr and cells were washed 3 × 10 min with Blotto/0 . 1% Triton X-100 . Secondary labeling was performed at room temperature for 30 min using subclass-specific , anti-mouse secondary antibodies conjugated to Alexa Fluors ( Thermo Fisher , Cat#/IgG subclass/Alexa Fluor dye conjugates: ( A-21121/IgG1/488 and A-21241/IgG2a/647 ) and diluted to 1 . 3 µg/mL in Blotto/0 . 1% Triton X-100 . Hoechst 33342 ( Thermo Fisher Cat# H3570 ) was used at 0 . 1 µg/mL in the secondary antibody cocktail to stain nuclear DNA . Cells were washed 3 × 10 min with DPBS/0 . 1% Triton X-100 . Imaging was performed using a Zeiss M2 AxioImager microscope . Images were processed using Axiovision ( Carl Zeiss Microimaging , RRID:SCR_002677 and Fiji ( NIH , RRID:SCR_002285 ) software . For higher resolution imaging , COS-1 cells were plated on poly-L-lysine coated #1 . 5 glass cover slips and cultured overnight followed by transfection with plasmids encoding the target protein . Cells were fixed and immunolabeled as described in the previous section . Images were acquired on a Zeiss AxioImager M2 microscope using a 40x/0 . 8 NA plan-Apochromat oil-immersion objective and an AxioCam MRm digital camera . Optical sections were acquired using an ApoTome two structured illumination system ( Carl Zeiss MicroImaging ) . Imaging and post processing were performed in Axiovision and Photoshop ( Adobe Systems; RRID:SCR_014199 ) . Multiplex immunofluorescence labeling of immunoblots using mouse IgG subclass-specific secondary antibodies was performed as described previously ( Manning et al . , 2012 ) . In brief , samples were generated from COS-1 cells transiently transfected to express individual target proteins essentially as described above for the immunofluorescence experiments except that the cells were cultured in 35 mm tissue cultures dishes . Transfected COS-1 cells were washed once with ice-cold PBS and lysed with 150 µL of ice-cold lysis buffer containing 1% v/v Triton X-100 , 150 mM NaCl , 1 mM EDTA , 50 mM Tris-HCl ( pH 7 . 4 ) , 1 mM sodium orthovanadate , 5 mM NaF , 1 mM PMSF and a protease inhibitor cocktail for 10 min at 4°C ( Shi et al . , 1994 ) . The cell lysates were centrifuged at 12 , 000 x g at 4 ˚C for 10 min . The cell lysate supernatants were mixed with 150 µL of 2X RSB and size-fractionated by 7 . 5% SDS–PAGE . Following SDS-PAGE , proteins were transferred to nitrocellulose membranes ( Bio-Rad Cat# 1620115 ) , which were blocked for 1 hr with Blotto ( 3% w/v nonfat milk in Tris-buffered saline ( TBS: 50 mM Tris , pH 7 . 5 , 150 mM NaCl ) plus 0 . 1% v/v Tween-20 followed by 2 hr or overnight incubation with primary antibodies . Primary antibodies were mAb TC supes diluted 1:10 , non-diluted R-mAb TC supes , and an in-house anti-PSD-95 rabbit polyclonal antibody raised against a GST fusion protein , GSTKAP1 . 13 , containing amino acids 77–299 of human PSD-95 [clone 2 , ( Kim et al . , 1995 ) ] and that crossreacts with SAP97 ( see Table 2 for details of non-R-mAb Abs used in this study ) . After three washes with Blotto , the membranes were incubated with the appropriate subclass-specific Alexa Fluor conjugated secondary antibodies ( Manning et al . , 2012 ) for 1 hr . After three washes with TBS containing 0 . 1% v/v Tween-20 , the immunoblots were visualized directly on a FluorChem Q imager ( Cell Biosciences Cat# DE500-FCQ ) . Alternatively , crude rat brain membranes ( RBM ) ( Shi et al . , 1994 ) were subjected to immunoblotting as described above except that a single RBM sample ( 3 mg protein ) were size fractionated on a curtain gel , and after transfer to nitrocellulose the membrane was cut into 30 strips , each containing 100 μg of RBM protein . Immunolabeling was detected on autoradiography film after treatment of strip immunoblots with HRP-conjugated anti-mouse IgG-specific secondary antibody and enhanced chemiluminescence ( ECL ) . Multiplex immunofluorescence labeling of rat brain sections was performed essentially as described previously ( Manning et al . , 2012; Bishop et al . , 2015 ) . All experimental procedures were approved by the UC Davis Institutional Animal Care and Use Committee and conform to guidelines established by the National Institutes of Health ( NIH ) . Rats were anesthetized with sodium pentobarbital ( Fatal-Plus solution , 100 mg/kg sodium pentobarbital ) and perfused transcardially with 100 mL of phosphate buffered saline ( PBS ) , containing 10 units/mL heparin , pH 7 . 4 , followed by 400 mL of 4% formaldehyde ( prepared fresh from paraformaldehyde ) in 0 . 1 M sodium phosphate buffer or PB ( pH 7 . 4 ) . The brains were removed , cryoprotected for 48 hr in 30% sucrose , frozen in a bed of pulverized dry ice , and then cut into 30 μm sections on a freezing-stage sliding microtome . Sections were collected in 0 . 1 M PB and processed immediately for immunohistochemistry . Free-floating brain sections were blocked with 10% goat serum in 0 . 1 M PB containing 0 . 3% Triton X-100 ( vehicle ) for 1 hr at RT and then incubated overnight at 4°C in vehicle containing different combinations of primary antibodies ( see Table 2 for details of non-R-mAb Abs used in this study ) . The following day sections were washed 4 × 5 min each with vehicle , and then incubated for 1 hr at RT in mouse IgG subclass-specific or anti-rabbit Alexa-conjugated secondary antibodies as described previously ( Manning et al . , 2012; Strassle et al . , 2005 ) . Sections were then washed 2 × 5 min each with 0 . 1 M PB , 2 × 5 min each with 0 . 05 M PB and mounted on gelatin-coated microscope slides and air dried . Sections were cover slipped after adding ProLong Gold Antifade Mountant ( Thermo Fisher Scientific catalog # P36930 ) . Images were obtained on a Zeiss Axiovert 200 microscope with Apotome . Imaging and post-imaging processing was performed in Zeiss Axiovision and Adobe Photoshop software , taking care to maintain any linear differences in signal intensities present in the original samples . | The immune system fights off disease-causing microbes using antibodies: Y-shaped proteins that each bind to a specific foreign molecule . Indeed , these proteins bind so tightly and so specifically that they can pick out a single target in a complex mixture of different molecules . This property also makes them useful in research . For example , neurobiologists can use antibodies to mark target proteins in thin sections of brain tissue . This reveals their position inside brain cells , helping to link the structure of the brain to the roles the different parts of this structure perform . To use antibodies in this way , scientists need to be able to produce them in large quantities without losing their target specificity . The most common way to do this is with cells called hybridomas . A hybridoma is a hybrid of an antibody-producing immune cell and a cancer cell , and it has properties of both . From the immune cell , it inherits the genes to make a specific type of antibody . From the cancer cell , it inherits the ability to go on dividing forever . In theory , hybridomas should be immortal antibody factories , but they have some limitations . They are expensive to keep alive , hard to transport between labs , and their genes can be unstable . Problems can creep into their genetic code , halting their growth or changing the targets their antibodies recognize . When this happens , scientists can lose vital research tools . Instead of keeping the immune cells alive , an alternative approach is to make recombinant antibodies . Rather than store the whole cell , this approach just stores the parts of the genes that encode antibody target-specificity . Andrews et al . set out to convert a valuable toolbox of neuroscience antibodies into recombinant form . This involved copying the antibody genes from a large library of preserved hybridoma cells . However , many hybridomas also carry genes that produce non-functional antibodies . A step in the process removed these DNA sequences , ensuring that only working antibodies made it into the final library . Using frozen cells made it possible to recover antibody genes from hybridoma cells that could no longer grow . The recombinant DNA sequences provide a permanent record of useful antibodies . Not only does this prevent the loss of research tools , it is also much more shareable than living cells . Modifications to the DNA sequences in the library allow for the use of many antibodies at once . This could help when studying the interactions between different molecules in the brain . Toolkits like these could also make it easier to collaborate , and to reproduce data gathered by different researchers around the world . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"tools",
"and",
"resources",
"neuroscience"
] | 2019 | A toolbox of IgG subclass-switched recombinant monoclonal antibodies for enhanced multiplex immunolabeling of brain |
Chemosensory neurons extract information about chemical cues from the environment . How is the activity in these sensory neurons transformed into behavior ? Using Caenorhabditis elegans , we map a novel sensory neuron circuit motif that encodes odor concentration . Primary neurons , AWCON and AWA , directly detect the food odor benzaldehyde ( BZ ) and release insulin-like peptides and acetylcholine , respectively , which are required for odor-evoked responses in secondary neurons , ASEL and AWB . Consistently , both primary and secondary neurons are required for BZ attraction . Unexpectedly , this combinatorial code is altered in aged animals: odor-evoked activity in secondary , but not primary , olfactory neurons is reduced . Moreover , experimental manipulations increasing neurotransmission from primary neurons rescues aging-associated neuronal deficits . Finally , we correlate the odor responsiveness of aged animals with their lifespan . Together , these results show how odors are encoded by primary and secondary neurons and suggest reduced neurotransmission as a novel mechanism driving aging-associated sensory neural activity and behavioral declines .
Animals have evolved specialized sensory systems to detect relevant information in their environment . This sensory information is relayed to downstream neural circuitry , generating appropriate food-seeking and toxin-avoiding behaviors , which enhance animal fitness . In particular , olfactory sensory neurons have an additional challenge in detecting a large set of volatile cues ( Buck , 2005 ) . In mammals , odors are detected by G-protein coupled odorant receptors that are expressed on olfactory sensory neurons . Moreover , while the mammalian genome encodes approximately 1000 receptors ( Buck and Axel , 1991 ) , each olfactory sensory neuron is known to express only one type of receptor ( Vassar et al . , 1993; Chess et al . , 1994 ) . Since mammals can detect far more than 1000 odors ( Duchamp-Viret et al . , 1999; Rubin and Katz , 1999 ) , this suggests that olfactory information at this level is encoded by a combinatorial code ( Malnic et al . , 1999 ) . Calcium imaging and electrophysiological studies have confirmed that individual odorants bind multiple odorant receptors and activate the corresponding olfactory sensory neurons ( Malnic et al . , 1999; Abaffy et al . , 2006 ) . Moreover , activity in an individual olfactory sensory neuron represents not only the molecular receptive field of its odor receptors ( Araneda et al . , 2000 ) , but also gating by feedback circuits ( Gomez et al . , 2005; Wachowiak et al . , 2009 ) and modulation by sniffing behavior in mammals ( Wesson et al . , 2009 ) . Information from these sensory neurons is then further processed and relayed to other brain regions ( Ghosh et al . , 2011; Miyamichi et al . , 2011; Sosulski et al . , 2011 ) . Despite this understanding , little is known about how specific activity patterns in the olfactory sensory neurons are correlated with behavioral outputs . One solution to this problem is to analyze numerically simpler invertebrate olfactory circuits where information flow can be traced at the resolution of individual neurons and correlated with animal behavior . The nematode Caenorhabditis elegans , with its small nervous system consisting of just 302 neurons , is ideally suited for a circuit-level analysis of chemosensory processing and behavior . Chemosensory stimuli are detected by twelve sensory neuron pairs located in the amphid ganglia ( Figure 1A ) ( White et al . , 1986; Bargmann , 2006 ) . All 24 of these neurons send their dendrites to the nose of the animal where they detect environmental changes and relay that information through their axons to the downstream circuitry ( White et al . , 1986 ) . C . elegans uses small numbers of sensory neurons to drive locomotion towards or away from particular sensory stimuli ( Bargmann , 2006 ) . For example , single cell ablation experiments showed that the bilaterally asymmetric pair of AWC sensory neurons is necessary for attraction to benzaldehyde ( BZ ) odor , while the AWA sensory neuron pair is required for diacetyl odor attraction ( Bargmann and Horvitz , 1991; Bargmann , 2006 ) . Functional imaging experiments revealed that AWC neurons are activated by the removal of odor stimuli ( Chalasani et al . , 2007 ) , while AWA neurons respond to the addition of odors ( Zaslaver et al . , 2015 ) . However , these sensory neuron activity patterns are not sufficient to explain how animals behave when they encounter diverse olfactory stimuli in the environment . We hypothesized that multiple amphid ganglia neurons could encode odor information and drive plastic olfactory behaviors; therefore , we performed the first comprehensive analysis of odor-evoked neural activity in all amphid neurons . We identified a novel circuit motif consisting of primary and secondary olfactory neurons that collectively encode odor and drive behavioral plasticity . We then analyzed the reliability of this combinatorial code and found that it degrades during aging . Our experiments suggest that a selective vulnerability of neurotransmitter release pathways in aged animals is the underlying mechanism that leads to a specific decay in secondary olfactory neuron activity and associated behavioral decline . Furthermore , we find that olfactory circuit function is correlated with an animal's longevity . 10 . 7554/eLife . 10181 . 003Figure 1 . Multiple sensory neurons detect the odor benzaldehyde ( BZ ) . ( A ) Image of a young adult C . elegans and schematic depicting the twelve pairs of sensory neurons in the anterior amphid ganglia whose dendrites project to the nose of the animal where they detect sensory stimuli . ( B ) Average GCaMP fluorescence change in young adult ( day 1 ) , wild-type sensory neurons in response to medium concentration ( 0 . 005% vol/vol ) BZ stimulation . Shaded box indicates two minute BZ odor stimulation beginning at t = 10 s . The light color shading around curves indicates s . e . m . and numbers in parentheses indicate number of neurons imaged . ( C ) Summary chart of the calcium responses of all amphid sensory neurons to low ( 0 . 0001% vol/vol ) , medium ( 0 . 005% vol/vol ) and high ( 0 . 1% vol/vol ) concentrations of BZ odor . This chart shows the composition of the C . elegans olfactory neural circuit and depicts a combinatorial sensory neuron code for odor concentration . The calcium signal in some neurons ( as indicated ) is suppressed by the addition of odor ( see methods and materials section ) . ( D ) Chemotaxis assay schematic depicting C . elegans attraction to a point source of BZ . Animals are placed at the origin ( O ) and allow to chemotax towards a point of BZ or control ( Ctrl ) . The putative BZ gradient is shown in shades of green with darker colors representing higher BZ concentrations . ( E ) Young adult ( day 1 ) chemotaxis performance of wild-type , AWC or AWB or ASH neuron-specific genetic ablation , AWA neuron-specific tetanus toxin expression worms or che-1 mutants missing ASE neurons to a medium concentration point source of BZ odor ( Uchida et al . , 2003 ) . See Figure 1—source data 1 for raw chemotaxis data . Numbers on bars indicate number of assay plates and error bars indicate s . e . m . *p < 0 . 05 , two-tailed t-test with Bonferroni correction , compared to wild-type . DOI: http://dx . doi . org/10 . 7554/eLife . 10181 . 00310 . 7554/eLife . 10181 . 004Figure 1—source data 1 . Young adult chemotaxis performance data . DOI: http://dx . doi . org/10 . 7554/eLife . 10181 . 00410 . 7554/eLife . 10181 . 005Figure 1—source data 2 . Odor-evoked responses in wild-type young adult data . DOI: http://dx . doi . org/10 . 7554/eLife . 10181 . 00510 . 7554/eLife . 10181 . 006Figure 1—figure supplement 1 . Combinatorial olfactory coding in C . elegans . ( A ) Maximum ΔF/F of each individual wild-type animal's AWCON , AWA , ASEL or AWB neuron response to medium BZ . ( B ) Quantification of the time to maximum ΔF/F following stimulus change for each wild-type AWCON , AWA , ASEL or AWB neuron response to medium BZ in seconds . ( A , B ) These graphs show additional quantification of the young adult odor response data presented in Figure 1B . Horizontal lines show mean and error bars represent s . e . m . ( C ) Average calcium responses of young adult , wild-type amphid sensory neurons to medium concentration BZ stimulus . ( D , E ) Average GCaMP fluorescence change in young adult , wild-type ( D ) AWCOFF or ( E ) ASE right ( ASER ) sensory neurons in response to low , medium or high concentration BZ stimulus . ( F–I ) Average calcium responses of wild-type amphid sensory neurons to ( F , G ) low and ( H , I ) high concentration BZ stimulus . ( C–I ) Shaded box represents two minute BZ stimulation ( low 0 . 0001% vol/vol , medium 0 . 005% vol/vol and high 0 . 1% vol/vol ) beginning at t = 10 s . Light shading around curves indicates s . e . m . and numbers in parentheses indicate number of neurons imaged . See Figure 1—source data 2 for raw data . DOI: http://dx . doi . org/10 . 7554/eLife . 10181 . 006
We used functional imaging to identify the amphid sensory neurons that detect the food odor Benzaldehyde ( BZ ) ( Figure 1A ) . We trapped young adult animals expressing GCaMP family of genetically encoded calcium indicators ( Tian et al . , 2009 ) , under cell selective promoters , in individual amphid sensory neurons in our custom-designed microfluidic device ( Chalasani et al . , 2007 ) and recorded their responses to BZ . Consistent with previous studies , we observed a large calcium transient indicating increased AWC activity upon removal of a medium concentration BZ stimulus ( Figure 1B , C , Figure 1—figure supplement 1A , B ) ( Chalasani et al . , 2007 ) . Unexpectedly , we found additional BZ responsive neurons: the diacetyl sensing AWA neurons ( Bargmann et al . , 1993 ) were activated by the addition of BZ , while ASE and AWB neurons ( that were previously shown to sense salts [Bargmann and Horvitz , 1991] and volatile repellents [Troemel et al . , 1997; Bargmann , 2006] , respectively ) also responded to the removal of this stimulus in young adults ( Figure 1B , C , Figure 1—figure supplement 1A , B ) . Furthermore , none of the other amphid neurons responded to this medium concentration BZ stimulus ( Figure 1B , C Figure 1—figure supplement 1C ) . While the two AWC and ASE neurons can be genetically and functionally separated ( Wes and Bargmann , 2001; Suzuki et al . , 2008 ) , each one in the pair showed similar responses to the removal of the BZ stimulus; therefore , we chose to focus our subsequent analysis on the AWCON and ASEL ( left ) neurons ( Figure 1B , C , Figure 1—figure supplement 1D , E ) . We also noted that the ASEL responses to BZ were slower to reach the maximum response ( average of 5 . 46 s after stimulus change ) than the other odor responsive neurons ( average 1–2 s after stimulus change ) , indicating that the kinetics of odor-evoked activity are different in different cells ( Figure 1—figure supplement 1B ) . Moreover , different neural activity patterns distributed across AWC , ASE , AWA , AWB and ASH sensory neurons defined active neural circuits for different concentrations ( medium as well as high or low ) of BZ ( Figure 1C , Figure 1—figure supplement 1F–I ) . We focused on responses to the attractive medium concentration of BZ for the remainder of this study . Our data suggests that four pairs of sensory neurons ( AWC , AWA , ASE and AWB ) signal the presence of this BZ stimulus . Next , we tested whether all four of these sensory neuron pairs were also required to drive behavioral attraction to BZ . We used a chemotaxis assay ( Figure 1D ) and analyzed the behavior of animals with non-functional sensory neurons . We found that genetic ablation ( Beverly et al . , 2011; Yoshida et al . , 2012 ) or blocking synaptic transmission ( with tetanus toxin [Schiavo et al . , 1992] ) in any of the four AWC , ASE , AWA or AWB neurons impaired animals' chemotaxis to a point source of medium BZ ( Figure 1E ) . This data is consistent with our imaging experiments and confirms a role for multiple sensory neurons in driving attraction to the BZ odor . In particular , our results showing important roles for ASE , AWA and AWB neurons in BZ attraction are novel . Together , these results show that a combinatorial code of activity across multiple neurons is essential to drive plasticity in an animal's behavior to BZ odor . Previously , we defined two classes of sensory neurons: primary neurons , which directly detect stimuli , and secondary neurons , which respond to neurotransmission from primary neurons ( Leinwand and Chalasani , 2013 ) . To classify the BZ-responsive neurons , we combined laser cell ablation with functional imaging . We predicted that BZ responses in primary neurons would be preserved when other odor responsive sensory neurons were ablated , while secondary neuron responses would require functional signaling from intact primary neurons . We found that AWCON responses to BZ were not affected in animals with any of the other BZ responsive neuron pairs ( AWA , ASE or AWB neurons ) ablated , suggesting that AWCON neurons directly detect the odor stimulus ( Figure 2A ) . Similarly , AWA responses to BZ were not affected in animals with ablated AWC , ASE or AWB neurons ( Figure 2B ) . These experiments suggest that AWCON and AWA neurons directly detect BZ and function as primary sensory neurons . In contrast , ASEL responses to BZ were greatly reduced in animals with ablated AWC neurons , but unaffected by AWA or AWB neuron ablation ( Figure 2C ) . This suggests that ASEL neurons may respond to signals from AWCON primary sensory neurons ( Figure 2C ) . Similarly , AWB responses to BZ required signaling from AWA neurons as these responses were significantly reduced specifically in the AWA ablation condition ( Figure 2D ) . Interestingly , while AWA neurons responded to the addition of odor stimulus with an increase in the calcium signal , the AWB neuron calcium signal increased upon odor removal ( Figure 2B , D ) . We suggest that AWB neurons may be inhibited by AWA and , when odor is removed , AWA is no longer active , leading to a release from inhibition and an increase in AWB activity . Additionally , direct olfactory sensory inputs or signaling from other neurons may also contribute to AWB activity , accounting for the residual AWB responses to odor in the AWA neuron ablated animals ( Figure 2D ) . Collectively , these data show a novel sensory circuit configuration in which the odor responsive neurons are not equal: the olfactory circuit for BZ odor is composed of two primary sensory neurons ( AWCON and AWA ) and two secondary neurons ( ASEL and AWB ) ( Figure 2E ) . 10 . 7554/eLife . 10181 . 007Figure 2 . Cell ablation reveals primary and secondary BZ sensory neurons . ( A ) Average young adult AWCON neuron responses to medium BZ in control ( Ctrl ) mock-ablated animals compared to animals with the AWA , ASE or AWB sensory neurons ablated ( neurons ablated at an early larval stage ) . ( B ) Average young adult AWA neuron responses to BZ in Ctrl mock-ablated animals compared to animals with AWC , ASE or AWB sensory neurons ablated . ( C ) Average young adult ASEL neuron responses to BZ in Ctrl mock-ablated animals compared to animals with AWC , AWA or AWB sensory neurons ablated . ( D ) Average young adult AWB neuron responses to BZ in Ctrl mock-ablated animals compared to animals with AWC , AWA or ASE sensory neurons ablated . ( A–D ) Shaded box represents two minute medium BZ ( 0 . 005% vol/vol ) stimulation beginning at t = 10 s . Yellow box indicates the time period after stimulus change for which the fluorescence change was averaged in the bar graphs ( See Figure 2—source data 1 for raw data . ) . Light shading around curves and bar graph error bars indicate s . e . m . Numbers on bars indicate number of neurons imaged . *p < 0 . 05 , two-tailed t-test with Bonferroni correction , compared to mock-ablation . ( E ) Schematic of the BZ circuit depicting the primary , direct BZ sensory neurons and the secondary , indirect BZ sensory neurons whose odor responses are reduced by cell ablation . DOI: http://dx . doi . org/10 . 7554/eLife . 10181 . 00710 . 7554/eLife . 10181 . 008Figure 2—source data 1 . Odor responses in cell ablated animal data . DOI: http://dx . doi . org/10 . 7554/eLife . 10181 . 008 Based on the C . elegans wiring diagram ( White et al . , 1986 ) , we hypothesized that the primary olfactory neurons use chemical neurotransmission to signal the presence of odor to the secondary neurons . To identify the relevant primary neuron released neurotransmitters that activate the secondary neurons , we analyzed the neural activity patterns in various mutants . We first examined genetic mutants that primarily block the release of ( 1 ) small , clear synaptic vesicles containing classical neurotransmitters such as glutamate , gamma-aminobutyric acid ( GABA ) and acetylcholine [Munc13 or unc-13 in C . elegans ( Richmond et al . , 1999 ) ] or ( 2 ) neuropeptide-containing dense core vesicles [CAPS , calcium-dependent activator protein for secretion , or unc-31 in C . elegans ( Speese et al . , 2007 ) ] . We found that AWCON and AWA neurons retained their odor responsiveness in the absence of classical or peptidergic neurotransmission ( Figure 3A , B ) . This data confirms our cell ablation results , indicating that these neurons directly detect BZ and are primary olfactory sensory neurons . Interestingly , we found that AWCOFF responses to BZ were significantly reduced in unc-13 mutants , suggesting that classical neurotransmission might be required to potentiate odor-evoked activity in this neuron ( Figure 3—figure supplement 1A ) . We suggest that AWCOFF responses to BZ might be potentiated by classical neurotransmission from AWCON neuron . Together , these results confirm that AWCON and AWA are primary sensory neurons and can directly detect BZ in the environment . 10 . 7554/eLife . 10181 . 009Figure 3 . Primary olfactory neurons release neuropeptides and classical neurotransmitters to recruit secondary neurons into the BZ circuit . ( A , B ) Average young adult ( A ) AWCON and ( B ) AWA neuron calcium responses to BZ in wild-type , unc-13 mutants with impaired synaptic vesicle release , and unc-31 mutants with impaired dense core vesicle release . ( C ) ASEL responses to BZ in unc-31 mutants and unc-31; AWC-specific unc-31 rescue . ( D ) AWB responses to BZ in unc-13 mutants and animals with AWA- or AWC-specific expression of tetanus toxin . ( A–D ) Shaded box indicates two-minute medium BZ ( 0 . 005% vol/vol ) odor stimulation . Yellow box indicates the time period after stimulus change for which the fluorescence change was averaged in the bar graphs ( See Figure 3—source data 1 for raw data ) . The light color shading around curves and bar graph error bars indicate s . e . m . Numbers on bars indicate number of neurons imaged . *p < 0 . 05 , two-tailed t-test with Bonferroni correction , compared to wild-type or mutant as indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 10181 . 00910 . 7554/eLife . 10181 . 010Figure 3—source data 1 . Odor responses in neurotransmitter release pathway genetic mutant data . DOI: http://dx . doi . org/10 . 7554/eLife . 10181 . 01010 . 7554/eLife . 10181 . 011Figure 3—source data 2 . Odor responses in genetic mutant data . DOI: http://dx . doi . org/10 . 7554/eLife . 10181 . 01110 . 7554/eLife . 10181 . 012Figure 3—figure supplement 1 . Primary and secondary olfactory neurons respond to BZ . ( A-D ) Average calcium responses of young adult ( A ) AWCOFF , ( B ) ASER , ( C ) ASEL and ( D ) AWB neurons in wild-type , unc-13 mutants with impaired synaptic vesicle release , and unc-31 mutants with impaired dense core vesicle release to BZ stimulation . ( A–D ) Shaded box indicates two-minute medium BZ ( 0 . 005% vol/vol ) odor stimulation . Yellow box indicates the time period after stimulus change for which the fluorescence change was averaged in the bar graphs ( see Figure 3—source data 2 for raw data ) . Numbers on bar graphs indicate number of neurons imaged . Light color shading around curves and bar graph error bars indicate s . e . m . *p < 0 . 05 , two-tailed t-test with Bonferroni correction , compared to wild-type . DOI: http://dx . doi . org/10 . 7554/eLife . 10181 . 012 In contrast , we found that mutations impairing neurotransmission affected the odor responses of the ASEL and AWB secondary neurons , which we identified by cell ablation experiments . Specifically , odor-evoked ASEL activity required unc-31-dependent neuropeptide signaling ( Figure 3C ) . Restoring neuropeptide release function specifically to the AWC neurons rescued ASEL BZ responses in unc-31 mutants , suggesting that AWC neurons release peptides to recruit ASEL neurons ( Figure 3C ) . This gene mutant analysis suggests that the longer time required for ASEL neurons to reach their maximum response to odor may reflect the additional requirement of AWC-dependent peptidergic transmission ( Figure 3C and Figure 1—figure supplement 1B ) . Similarly , ASER responses to BZ also require neuropeptide signaling ( Figure 3—figure supplement 1B ) . While we have not identified the source of these neuropeptides , we suggest that AWC released peptides may also activate ASER neurons . Moreover , unc-13-dependent classical neurotransmission was not required for either ASEL or ASER responses to BZ ( Figure 3—figure supplement 1B , C ) . We then examined AWB responses to BZ in neurotransmission mutants . AWB responses were significantly and specifically reduced in unc-13 mutants , suggesting that these neurons are recruited to this olfactory circuit by classical neurotransmitter ( s ) ( Figure 3D , Figure 3—figure supplement 1D ) . To confirm that AWA was the source of these classical neurotransmitters ( Figure 2D ) , we used tetanus toxin to manipulate the neurotransmitter pathways . Tetanus toxin has been previously shown to cleave synaptobrevin and block neurotransmission ( Schiavo et al . , 1992 ) . We found that expressing tetanus toxin specifically in the AWA , but not AWC , sensory neurons significantly reduced AWB responses to BZ removal ( Figure 3D ) . This confirms that AWA signals to AWB and recruits it into the odor circuit . Nevertheless , the residual odor-evoked AWB responses observed in unc-13 mutants and transgenic animals with reduced AWA neurotransmission ( AWA::tetanus toxin ) confirm that direct sensory inputs or signaling from other neurons may also contribute to AWB activity ( Figure 3D ) . These data show that ASE and AWB neurons can function as secondary neurons because their responses to BZ require neuropeptide and classical neurotransmitter signaling respectively . Collectively , this defines a BZ odor-encoding circuit motif consisting predominantly of two primary and two secondary neurons wired as two parallel channels of olfactory information . We then mapped the identities of the neuropeptide and neurotransmitter pathways transferring information from primary to secondary olfactory neurons . The C . elegans genome includes at least 122 neuropeptide genes and pathways to generate several classical neurotransmitters including glutamate , GABA and acetylcholine ( Hobert , 2013 ) . To identify the cognate neuropeptide ( s ) activating ASEL neurons , we used ASEL activity as readout to screen a number of neuropeptide gene mutants . We found that the insulin-like peptide ins-1 ( Pierce et al . , 2001 ) was required for BZ-evoked ASEL responses ( Figure 4A ) . Moreover , restoring INS-1 function specifically to AWC neurons , but not to AWA neurons , rescued mutant ASEL activity deficits ( Figure 4A ) . This suggests that AWC neurons release INS-1 peptides to recruit ASEL neurons into the odor circuit . To confirm AWC as the source of the INS-1 peptides , we used an AWC neuron-specific RNAi approach to knockdown the ins-1 gene . Previous studies have shown that expressing the sense and anti-sense transcript under a cell-specific promoter can efficiently knockdown the gene of the interest in that cell ( Esposito et al . , 2007; Leinwand and Chalasani , 2013 ) . We found that knocking down ins-1 in AWC neurons significantly reduced the ASEL responses to BZ , confirming that AWC-released INS-1 is required for ASEL activity in the odor circuit ( Figure 4B ) . We suggest that the same insulin neuropeptide may be multifunctional . For example , INS-1 released from AIA interneurons inhibits AWC and ASER activity ( Tomioka et al . , 2006; Chalasani et al . , 2010 ) , while we show that INS-1 released from AWC recruits ASEL into the BZ circuit . Ultimately , this signaling can regulate odor circuit dynamics , salt chemotaxis plasticity and integrative thermotactic behavior ( Kodama et al . , 2006; Tomioka et al . , 2006; Chalasani et al . , 2010 ) . Collectively , these results suggest that the site of release and likely also signaling in the downstream neurons play key roles in determining the functionality of INS-1 peptides . 10 . 7554/eLife . 10181 . 013Figure 4 . Insulin peptidergic and cholinergic transmission from the two primary olfactory sensory neurons recruits two secondary olfactory neurons . ( A ) BZ-evoked activity in young adult ASEL neurons in wild-type , ins-1 insulin-like peptide mutants , ins-1; AWC-specific ins-1 rescue and ins-1; AWA-specific ins-1 rescue . ( B ) Average ASEL responses to BZ in young adult wild-type and AWC neuron-specific ins-1 RNAi knockdown animals . ( C , D ) BZ-evoked activity in young adult ASEL neurons in ( C ) daf-2 insulin receptor mutants and daf-2; ASEL-specific daf-2 rescue , and ( D ) age-1 PI3-Kinase mutants and age-1; ASEL-specific age-1 rescue compared to wild-type . ( E ) AWB neuronal activity in response to BZ in young adult wild-type , unc-17 vesicular acetylcholine transporter mutants and unc-17; AWA-specific unc-17 rescue . ( F ) AWB neuronal activity in response to BZ in young adult wild-type , AWA neuron-specific cha-1 choline acetyltransferase RNAi and AWA-specific cho-1 choline transporter RNAi knockdown transgenic animals . ( G , H ) Young adult chemotaxis performance of wild-type and ( G ) AWC neuron-specific ins-1 RNAi knockdown or ( H ) AWA neuron-specific cha-1 RNAi knockdown animals to a medium concentration point source of BZ odor . Numbers on bars indicate number of assay plates and error bars indicate s . e . m . *p < 0 . 05 , two-tailed t-test . ( I ) Proposed young adult BZ circuit model . ( A-F ) Shaded box indicates medium BZ ( 0 . 005% vol/vol ) odor stimulation . Yellow box indicates the time period after stimulus change for which the fluorescence change was averaged in the bar graphs . Numbers on bar graphs indicate number of neurons imaged . Light color shading around curves and bar graph error bars indicate s . e . m . *p < 0 . 05 , two-tailed t-test with Bonferroni correction , compared to wild-type or mutant as indicated . See also Figure 4—source data 1 for raw data . DOI: http://dx . doi . org/10 . 7554/eLife . 10181 . 01310 . 7554/eLife . 10181 . 014Figure 4—source data 1 . Odor responses and chemotaxis performance in insulin and acetycholine pathway mutant and transgenic data . DOI: http://dx . doi . org/10 . 7554/eLife . 10181 . 01410 . 7554/eLife . 10181 . 015Figure 4—source data 2 . Additional odor responses in insulin and acetycholine pathway mutant and transgenic data . DOI: http://dx . doi . org/10 . 7554/eLife . 10181 . 01510 . 7554/eLife . 10181 . 016Figure 4—figure supplement 1 . Odor-evoked calcium dynamics in genetic mutants . ( A–C ) Young adult AWCON neuron average responses to BZ stimulation in wild-type animals compared to ( A ) insulin-like peptide ins-1 mutants , ( B ) daf-2 insulin receptor mutants and ( C ) age-1 PI3-Kinase mutants . ( D ) Average AWB calcium responses to 2-nonanone in wild-type and unc-17 vesicular acetylcholine transporter mutants . ( E ) Average young adult AWA neuron responses to BZ in wild-type and unc-17 vesicular acetylcholine transporter mutants . ( F ) Average AWA neuron responses to BZ in wild-type and AWA neuron-specific cha-1 choline acetyltransferase RNAi knockdown animals . ( A–F ) Data presented as described for Figure 4 . NS , p > 0 . 05 , two-tailed t-test . See Figure 4—source data 2 for raw data . DOI: http://dx . doi . org/10 . 7554/eLife . 10181 . 016 Next , we investigated the receptor and downstream signaling components in ASEL neurons that transduce the AWC-released INS-1 signal . We found that odor-evoked ASEL activity required the canonical insulin receptor ( daf-2 in C . elegans [Pierce et al . , 2001] ) and PI3-Kinase ( age-1 in C . elegans [Morris et al . , 1996] ) signaling in ASEL neurons ( Figure 4C , D ) . We suggest that the increase in calcium in ASEL may result from rapid signaling downstream of PI3-Kinase acting directly on calcium channels ( Blair and Marshall , 1997 ) , and that this may represent an alternate pathway to the canonical , long term effects of DAF-2 signaling to regulate gene expression ( Murphy et al . , 2003 ) . Furthermore , we found that the insulin receptor mutant ( daf-2 ) had a stronger reduction in ASEL activity compared to the insulin ligand mutant ( ins-1 ) or the insulin ligand knockdown animals ( Figure 4A–C ) . These results suggest that AWC neurons may co-release additional insulin peptides along with INS-1 to bind the insulin receptor on ASEL neurons . Importantly , we also confirmed that AWCON primary olfactory neuron dynamics were normal in all of the insulin pathway mutants analyzed , indicating that insulin signaling functions downstream of primary olfactory sensory transduction ( Figure 4—figure supplement 1A–C ) . Furthermore , we have previously shown that ASEL responses to a different , directly detected stimulus , salt , are not affected in the daf-2 or age-1 mutants , suggesting that these primary ASEL responses do not depend on insulin signaling ( Leinwand and Chalasani , 2013 ) . Together , these results indicate that AWC-released insulin peptides signal to ASEL secondary neurons via the insulin receptor and PI3-Kinase to encode the BZ stimulus . We also mapped the classical neurotransmitter pathway recruiting AWB neurons into the circuit . We found that mutations in the vesicular acetylcholine transporter ( VAChT ) , unc-17 , which packs acetylcholine into synaptic vesicles ( Alfonso et al . , 1994 ) , reduced AWB odor responses ( Figure 4E ) . Restoring cholinergic function specifically in AWA primary neurons was sufficient to elicit wild-type-like activity in AWB secondary neurons ( Figure 4E ) . We also examined additional components of the cholinergic synthesis and release pathway through a cell-specific RNAi knockdown approach . We found that knocking down the C . elegans choline acetyltransferase ( ChaT ) , cha-1 , which is required for the biosynthesis of acetylcholine ( Rand and Russell , 1984; Alfonso et al . , 1994 ) , specifically in the AWA neurons significantly reduced AWB neuron responses to BZ ( Figure 4F ) . Together , these results suggest an essential role for cholinergic signaling from the AWA neurons to recruit AWB neurons to the olfactory circuit . Interestingly , AWA neuron-specific knockdown of the choline transporter cho-1 , which is required for high affinity choline reuptake at presynaptic terminals ( Okuda et al . , 2000 ) , had no effect on AWB responses to BZ ( Figure 4F ) . Therefore , we suggest that AWA requires the choline acetyltransferase , but may not require the high affinity choline transporter to release acetylcholine . While we cannot rule out the possibility that our attempts to knockdown the choline transporter were ineffective , our results are consistent with prior observations that loss of cho-1 has only mild effects on cholinergic neurotransmission and suggest that de novo choline synthesis and low affinity choline uptake may be sufficient for cholinergic signaling in the olfactory circuit ( Mullen et al . , 2007 ) . We considered whether acetylcholine modulates AWB activity by acting on muscarinic receptors . We found that odor-evoked AWB activity was not affected in mutants of any of the three identified C . elegans muscarinic receptors ( gar-1 , gar-2 and gar-3 ) ( data not shown ) , suggesting that acetylcholine might bind other receptors on AWB neurons . The C . elegans genome encodes 8 acetylcholine-gated chloride channels ( Hobert , 2013 ) and we suggest that AWA-released acetylcholine binds one of these receptors to inhibit AWB neuronal activity when odor is added , leading to a rebound from this inhibition when odor is removed . Moreover , we found that AWB responses to the directly detected repulsive odorant 2-nonanone ( Troemel et al . , 1997 ) were normal in unc-17 mutants ( Figure 4—figure supplement 1D ) . We suggest that AWB secondary ( to BZ ) , but not primary ( to 2-nonanone ) responses require cholinergic signaling . Importantly , we also confirmed that AWA primary olfactory neuron dynamics were normal in the genetic mutants and knockdown animals analyzed ( Figure 4—figure supplement 1E , F ) . These experiments support the conclusion that changes in the secondary neuron activity observed in these mutants and knockdown transgenic animals are downstream of sensory transduction in the primary neurons and related to transmitter release from primary neurons . Next , we tested whether insulin peptidergic and cholinergic signaling were required for chemotaxis behavior . Consistent with our imaging results , we found that knocking down the insulin-like peptide ins-1 in AWC neurons significantly reduced attraction to BZ ( Figure 4G ) . In addition , animals with the choline acetyltransferase cha-1 knocked down specifically in AWA neurons also displayed significantly reduced BZ chemotaxis behavior ( Figure 4H ) . Together , these data show that BZ stimulus is encoded by AWCON and AWA primary sensory neurons , which use insulin peptidergic and cholinergic neurotransmission to elicit activity in ASEL and AWB secondary neurons and to shape chemotaxis behavior ( Figure 4I ) . Thus , multiple neuropeptide and neurotransmitter pathways are integrated to shape odor encoding and behavior . We have shown that a combinatorial neural activity code comprising primary and secondary neurons encodes odors and drives behavior . Is this combinatorial olfactory code persistent and reliable throughout life ? Interestingly , olfactory behavioral performance has been previously shown to degrade with age , which in turn affects quality of life and overall safety and survival across species ( Doty and Kamath , 2014 ) . We used the detailed characterization of the combinatorial BZ olfactory circuit described above to investigate systems levels changes in olfactory function with age . We first tested whether aging affects BZ-evoked behavior . While young adults were strongly attracted to BZ odor , we found that older animals ( day 4–6 ) showed a significant impairment in their attraction ( Figure 5A ) . The behavioral deficit was largest for day 6 adults; however , we found that these animals had more variability in their size ( Figure 5—figure supplement 1A ) making it difficult to design an effective trap to image animals beyond day 5 of adulthood and analyze their odor-evoked neuronal activity . Therefore , for the remainder , we compared young adults ( day 1 , the age characterized above ) and animals at a post-reproductive , early stage of aging ( day 5 ) , which we refer to as ‘aged’ adults . Importantly , we found that the aging-associated chemotaxis behavioral deficit is unlikely to be caused by changes in locomotory ability since the speed of chemotaxing aged animals did not differ from that of young adults ( Figure 5B ) . This data establishes BZ chemotaxis as a model of aging-associated olfactory sensory behavioral decline . 10 . 7554/eLife . 10181 . 017Figure 5 . BZ-evoked secondary neuron activity and behavior specifically degrade with age . ( A ) Chemotaxis performance of wild-type worms from young adulthood ( day 1 ) through early stage aging ( day 6 ) towards a point source of medium BZ . ( B ) Speed of wild-type young ( day 1 ) and aged ( day 5 ) adult animals chemotaxing towards a point source of BZ odor . ( C–F ) Heat maps of ratio change in fluorescence to total fluorescence for wild-type young adult ( day 1 ) and aged adult ( day 5 ) sensory neuron responses to the addition ( at t = 10 s ) or removal ( at t = 130 s ) of a two-minute medium BZ stimulus ( 0 . 005% vol/vol ) , as indicated by shaded box and arrows . One row represents activity from one neuron . ( G ) Maximum ΔF/F for each individual young ( black dots ) or aged ( blue dots ) wild-type animal shown in C–F . ( H ) Averaged ΔF/F after odor addition ( for AWA ) or odor removal ( for all other neurons ) for each individual young ( black dots ) or aged ( blue dots ) wild-type animal shown in C–F . The red line represents a ΔF/F of 10% , the cutoff used to classify neurons as odor responsive or non-responsive . *p < 0 . 05 , two-tailed t-test comparing young and aged responses; statistical analysis performed only on odor responsive subset of data . ( I ) Quantification of the percent of odor responsive neurons shown in H . ( J ) Aged ( day 5 ) adult BZ chemotaxis performance of wild-type , AWC or AWB or ASH neuron-specific genetic ablation , AWA neuron-specific tetanus toxin expression worms or che-1 mutants missing ASE neurons . ( K , L ) The percent of wild-type young ( day 1 ) and aged ( day 5 ) adult ( K ) ASEL neurons responsive to sodium chloride and ( L ) AWB neurons responsive to 2-nonanone odor . ( I , K , L ) Odor or salt responsive defined as having a ΔF/F to stimulus greater than 10% . Numbers on bars indicate number of neurons imaged . *p < 0 . 05 , two-tailed Chi Square test . ( A , B , J ) Numbers on bars indicate number of assay plates and error bars indicate s . e . m . *p < 0 . 05 , two-tailed t-test with Bonferroni correction , compared to young adults or wild-type as indicated . See Figure 5—source data 1 for raw data . DOI: http://dx . doi . org/10 . 7554/eLife . 10181 . 01710 . 7554/eLife . 10181 . 018Figure 5—source data 1 . Age-related decay in odor responses and chemotaxis behavior data . DOI: http://dx . doi . org/10 . 7554/eLife . 10181 . 01810 . 7554/eLife . 10181 . 019Figure 5—source data 2 . Primary and secondary neuron activity in young and aged animal data . DOI: http://dx . doi . org/10 . 7554/eLife . 10181 . 01910 . 7554/eLife . 10181 . 020Figure 5—source data 3 . Correlated behavior and functional imaging in aged animal data . DOI: http://dx . doi . org/10 . 7554/eLife . 10181 . 02010 . 7554/eLife . 10181 . 021Figure 5—source data 4 . Dose-dependent odor response data . DOI: http://dx . doi . org/10 . 7554/eLife . 10181 . 02110 . 7554/eLife . 10181 . 022Figure 5—source data 5 . Salt and 2-nonanone responses in young and aged animal data . DOI: http://dx . doi . org/10 . 7554/eLife . 10181 . 02210 . 7554/eLife . 10181 . 023Figure 5—source data 6 . Longevity mutant odor response data . DOI: http://dx . doi . org/10 . 7554/eLife . 10181 . 02310 . 7554/eLife . 10181 . 024Figure 5—figure supplement 1 . Quantification of BZ-evoked primary and secondary neuron activity in young and aged animals . ( A ) Measurement of the perimeter of day 5 aged worms and the more variable day 6 aged worms ( see ‘Materials and methods’ section ) . Thick red line shows mean and error bars represent standard deviation ( day 5: 2656 . 3 μm ± 112 . 07 , day 6: 2950 . 5 μm ± 291 . 21 , n = 55 animals for each age ) . ( B ) Quantification of the time to maximum ΔF/F following stimulus change for individual wild-type young ( black dots ) or aged ( blue dots ) neuron responses to medium BZ , in seconds , for the subset of odor responsive recordings only . Horizontal red lines show mean and error bars represent s . e . m . NS , p > 0 . 05 , two-tailed t-test comparing young and aged response times . Graph shows additional quantification of the data presented in Figure 5C–F; see Figure 5—source data 2 for raw data . DOI: http://dx . doi . org/10 . 7554/eLife . 10181 . 02410 . 7554/eLife . 10181 . 025Figure 5—figure supplement 2 . Olfactory behavior in aged animals is correlated with reliability of odor-evoked neuronal activity . ( A ) Schematic of animals from a chemotaxis assay washed and sorted into two populations , based on success or failure in navigating up the BZ odor gradient , for calcium imaging . ( B , C ) Heat maps of ratio change in fluorescence to total fluorescence for BZ-evoked activity in wild-type ( day 5 ) aged ( B ) ASEL and ( C ) AWB neurons in animals that did or did not successfully chemotax towards the BZ point source . Two-minute medium BZ ( 0 . 005% vol/vol ) odor stimulation indicated by shaded box and arrows . One row represents activity from one neuron . ( D ) Quantification of the percent of BZ responsive neurons shown in B and C ( see Figure 5—source data 3 for raw data ) . Numbers on bars represent number of neurons imaged and odor responsive is defined as having a ΔF/F to odor greater than 10% . *p < 0 . 05 , two-tailed Chi Square test . DOI: http://dx . doi . org/10 . 7554/eLife . 10181 . 02510 . 7554/eLife . 10181 . 026Figure 5—figure supplement 3 . Dose-dependent odor-evoked calcium dynamics in young and aged adults . ( A ) Chemotaxis performance of wild-type worms of different ages towards a point source of high concentration BZ . Numbers on bars represent number of assay plates and error bars indicate s . e . m . NS , p > 0 . 05 , two-tailed t-test with Bonferroni correction , compared to young adults . ( B , C ) Heat maps of ratio change in fluorescence to total fluorescence for wild-type young adult ( day 1 ) and aged adult ( day 5 ) ( B ) ASEL and ( C ) AWB sensory neuron responses to high concentration BZ ( 0 . 1% vol/vol ) stimulation . One row represents activity from one neuron . ( D ) Quantification of the percent of high concentration BZ responsive neurons . Numbers on bars represent number of neurons imaged and odor responsive is defined as having a ΔF/F to odor greater than 10% . NS , p > 0 . 05 , two-tailed Chi Square test . See Figure 5—source data 4 for raw data . DOI: http://dx . doi . org/10 . 7554/eLife . 10181 . 02610 . 7554/eLife . 10181 . 027Figure 5—figure supplement 4 . ASE and AWB primary responses to salt and 2-nonanone , respectively , remain reliable with aging . ( A ) Chemotaxis performance of wild-type young ( day 1 ) and aged ( day 5 ) adults towards a point source of 500 mM NaCl . NS p > 0 . 05 , two-tailed t-test . ( B ) Heat maps of ratio change in fluorescence to total fluorescence for wild-type young adult ( day 1 ) and aged adult ( day 5 ) ASEL neurons to +50 mM NaCl stimulation . ( C ) Chemotaxis performance of wild-type young ( day 1 ) and aged ( day 5 ) adults towards a point source of repulsive 2-nonanone odor . ( D ) Heat maps of ratio change in fluorescence to total fluorescence for wild-type young and aged adult AWB neurons to 2-nonanone odor stimulation . ( E ) Heat maps of ratio change in fluorescence to total fluorescence for wild-type young adult ( day 1 ) and aged adult ( day 5 ) AWCON neurons to +50 mM NaCl stimulation . ( F ) Quantification of the percent of salt responsive neurons shown in E , with salt responsive defined as having a ΔF/F to +50 mM NaCl greater than 10% . ( A–F ) Data presentation and statistics are as in Figure 5—figure supplement 3; see Figure 5—source data 5 for raw data . DOI: http://dx . doi . org/10 . 7554/eLife . 10181 . 02710 . 7554/eLife . 10181 . 028Figure 5—figure supplement 5 . Long and short-lived mutants do not influence the aging-associated declines in neuronal function . ( A , B ) Heat maps of ratio change in fluorescence to total fluorescence for aged adult ( day 5 ) ( A ) ASEL and ( B ) AWB neurons stimulated with medium BZ ( 0 . 005% vol/vol ) in wild-type , glp-1 mutants and aak-2 gain of function ( gf ) mutants . ( C , D ) Heat maps of ratio change in fluorescence to total fluorescence for aged adult ( day 5 ) ( C ) ASEL and ( D ) AWB neurons stimulated with medium BZ ( 0 . 005% vol/vol ) in Ctrl , rab-10 , and hsf-1 RNAi treated animals . ( E–H ) Quantification of the percent of odor responsive neurons shown in A-D . NS , p > 0 . 05 , two-tailed Chi Square test . ( I ) Chemotaxis performance of young and aged wild-type and glp-1 mutant animals towards a point source of medium BZ . Data presentation and statistics are as in Figure 5—figure supplement 3; see Figure 5—source data 6 for raw data . DOI: http://dx . doi . org/10 . 7554/eLife . 10181 . 028 To determine the mechanism underlying this aging-associated decline in BZ-directed behavior , we probed neuronal activity in the combinatorial , BZ-encoding sensory neural circuit described above . We analyzed the responses of the primary ( AWCON and AWA ) and secondary ( ASEL and AWB ) neurons to BZ in both young ( day 1 ) and aged ( day 5 ) adult animals . Overall , aging did not affect the reliability , duration or magnitude of odor-evoked activity in AWCON and AWA primary neurons ( Figure 5C , D , G–I , Figure 5—figure supplement 1B ) . In contrast , odor-evoked ASEL and AWB secondary neuron activity was highly variable with aging , with many neurons failing to show any responses to odor , revealing a possible mechanism for behavioral decline ( Figure 5E–I , Figure 5—figure supplement 1B ) . Interestingly , the AWB neurons that did respond to odor in aged animals had calcium transients that were indistinguishable from responses in younger animals ( Figure 5F–H , Figure 5—figure supplement 1B ) . Additionally , considering only the animals with odor responsive ASEL neurons , the BZ responses of the aged animals were in fact significantly larger than that of the young animals ( Figure 5E , G , H ) . These results suggest that odor-evoked activity in ASEL and AWB secondary neurons selectively decays in some animals . Consistent with these results , we found that the weak chemotaxis performance of aged animals towards BZ only required the primary AWC and AWA neurons , and not the unreliable secondary ASE and AWB neurons ( Figure 5J ) . To further examine this , we tested whether performance in the chemotaxis assay is correlated with the odor responsiveness of the ASEL and AWB secondary neurons ( Figure 5—figure supplement 2A ) . We found that aged animals that failed to chemotax towards BZ were significantly more likely to have odor non-responsive ASEL and AWB neurons than aged animals that successfully found the odor source ( Figure 5—figure supplement 2B–D ) . Taken together , these data reveal a distributed neural circuit that detects attractive odors and suggest that BZ behavioral declines arise from unreliable activity of aged secondary ASEL and AWB neurons in this circuit . We then tested whether this aging-associated decline was dependent on odor concentration . We showed that a distinct , but overlapping set of sensory neurons encodes high concentration BZ ( Figure 1C , Figure 1—figure supplement 1H , I ) . Behaviorally , we found that high BZ was similarly repulsive in young and aged animals ( through day 5 ) ( Figure 5—figure supplement 3A ) . Consistently , high BZ-evoked neural activity did not significantly decline between day 1 and day 5 adults ( Figure 5—figure supplement 3B–D ) . These data suggest that the aging-associated decline in neuronal function is dependent on odor concentration; consistent with previous studies showing relatively preserved behavioral detection of strong sensory stimuli with age ( Hummel et al . , 2007 ) . Next , we investigated whether aging impairs all or only selective functions of ASEL and AWB neurons . To test this , we analyzed responses to salt ( sodium chloride ) and the repulsive odorant 2-nonanone , which are directly transduced by ASEL ( Bargmann , 2006; Suzuki et al . , 2008 ) and AWB neurons ( Troemel et al . , 1997 ) , respectively . We found that neuronal activity and behavior in response to these stimuli remained reliable and robust in aged animals ( Figure 5K , L , Figure 5—figure supplement 4A–D ) . These data indicate that functionality of both ASEL and AWB neurons in aged animals is sensory context dependent . Specifically , their primary responses to salt ( ASEL ) and 2-nonanone ( AWB ) are preserved , while their function as secondary neurons in encoding attractive BZ stimuli is impaired during aging . We have previously shown AWC sensory neurons act as secondary neurons in the salt sensory circuit and respond to salt stimuli in an ASE-dependent manner ( Leinwand and Chalasani , 2013 ) . Therefore , we tested whether AWC secondary responses salt were also degraded during aging . However , we found that AWC responses to salt were not reduced in aged animals ( Figure 5—figure supplement 4E , F ) . These data suggest that these early aging-associated deficits are specific to the BZ circuit , leaving the salt circuit fully functional . Together , these results show that there is a sensory context dependent decline in ASEL and AWB responses to BZ with age , disrupting the combinatorial code for attractive olfactory information specifically . C . elegans is short lifespan model and has proven to be useful in identifying conserved organismal-level longevity pathways , such as insulin and energy and stress sensing pathways ( Wolff and Dillin , 2006 ) . We hypothesized that long-lived mutants might alter the dynamics of the age-associated decline in the combinatorial neural code for odor . We tested several distinct pathways shown to mediate lifespan extension . Gain of function ( gf ) mutants in the energy sensing alpha subunit of the AMP-activated protein kinase ( AMPK , aak-2 in C . elegans [Apfeld et al . , 2004] ) have increased lifespan . Similarly , animals without a germline due to ablation ( Hsin and Kenyon , 1999 ) or mutations in the notch signaling pathway ( glp-1 in C . elegans [Berman and Kenyon , 2006] ) have increased lifespan . Furthermore , whole animal RNAi treatment to knockdown the Rab-like GTPase rab-10 also extends lifespan ( Hansen et al . , 2005 ) . We recorded ASEL and AWB secondary neuron responses to BZ in aged day 5 adults in wild-type , long-lived aak-2 ( gf ) and glp-1 mutants and rab-10 knockdown animals . Similar to wild-type , the ASEL and AWB responses to medium BZ in aak-2 ( gf ) and glp-1 mutants and rab-10 knockdown animals were unreliable in day 5 adults ( Figure 5—figure supplement 5 ) . We also tested whether the aging-associated declines in olfactory behavior were altered in long-lived mutants . We found that glp-1 mutants displayed a similar aging-associated decline in attraction to BZ compared to wild-type animals ( Figure 5—figure supplement 5I ) . These data show that signaling from the longevity-modulating germline , AMP kinase energy sensing and Rab GTPase pathways do not attenuate secondary neuronal activity and behavior declines . We also tested whether mutations that shorten lifespan could influence the aging-associated declines in neuronal function . A whole animal knockdown of the stress-induced heat shock factor 1 ( hsf-1 ) was shown to be short-lived ( Hsu et al . , 2003 ) . We found that animals with hsf-1 knocked down had similarly unreliable day 5 aged ASEL and AWB secondary neuron responses to BZ compared to wild-type ( Figure 5—figure-supplement 5 ) . Taken together , these data suggest that the aging-associated declines in olfactory neuronal functions are independent of many known longevity pathways ( glp-1 , aak-2 , rab-10 and hsf-1 ) . Our results show that the ASEL and AWB secondary neurons have unreliable odor-evoked activity in aged animals . This suggests that the neurotransmission that recruits these neurons to the odor circuit may break down with age . In particular , impaired ASEL neuronal activity may indicate a breakdown in the peptidergic neurotransmission that recruits this neuron into the BZ circuit . In order to identify the mechanisms for this aging-associated decline , we manipulated the primary to secondary neurotransmission pathway . First , we hypothesized that aging might downregulate the levels of the peptide receptors on ASEL neurons , thus reducing signaling in aged ASEL neurons . We tested this hypothesis by overexpressing the DAF-2 insulin receptor specifically in the ASEL neurons ( Figure 6A , left panel ) . However , we found no change in the reliability of these aged animals' odor-evoked ASEL activity compared to wild-type ( Figure 6B , C , F , Figure 6—figure supplement 1B ) . This result suggests that receptor expression is not limiting in these aged animals . We confirmed that our ASEL-specific DAF-2 overexpression ( OE ) was functional by analyzing ASEL responses in young day 1 adults . We found that the ASEL BZ responses were significantly larger in young adult animals overexpressing DAF-2 in ASEL ( Figure 6—figure supplement 1A , B ) confirming the efficacy of the transgene . Taken together , these results suggest that DAF-2 receptors in ASEL are not reduced during the aging process and signaling via these receptors does not limit olfactory circuit activity in aged animals . 10 . 7554/eLife . 10181 . 029Figure 6 . Increased neurotransmitter release from AWC neurons rescues aging-associated ASEL activity and behavioral deficits . ( A ) Schematic representation of genetic manipulations to overcome aging-associated decay of neurotransmission . ( B–E ) Heat maps of ratio change in fluorescence to total fluorescence for aged adult ( day 5 ) ASEL sensory neuron responses to the removal ( at t = 130 s ) of a two-minute medium BZ stimulus ( 0 . 005% vol/vol ) in ( B ) wild-type , ( C ) ASEL-specific daf-2 overexpression ( OE ) , ( D ) AWC-specific ins-1 OE and ( E ) AWC-specific tom-1 RNAi . ( F ) Quantification of the percent medium BZ responsive aged ASEL neurons in B–E . Odor responsive defined as having a ΔF/F to stimulus greater than 10% . Numbers on bars indicate number of neurons imaged . *p < 0 . 05 , two-tailed Chi Square test . ( G ) BZ chemotaxis in young and aged wild-type , che-1 mutants lacking ASE neurons , AWC-specific tom-1 RNAi and AWC-specific tom-1 RNAi in the che-1 background . *p < 0 . 05 , two-tailed t-test with Bonferroni correction . See Figure 6—source data 1 for raw data . DOI: http://dx . doi . org/10 . 7554/eLife . 10181 . 02910 . 7554/eLife . 10181 . 030Figure 6—source data 1 . Odor responses in AWC-released neurotransmitter manipulation animal data . DOI: http://dx . doi . org/10 . 7554/eLife . 10181 . 03010 . 7554/eLife . 10181 . 031Figure 6—source data 2 . Additional odor responses in AWC-released neurotransmitter manipulation animal data . DOI: http://dx . doi . org/10 . 7554/eLife . 10181 . 03110 . 7554/eLife . 10181 . 032Figure 6—figure supplement 1 . AWC-released neurotransmitters modify aging-associated neuronal activity and behavioral deficits . ( A ) Heat maps of ratio change in fluorescence to total fluorescence for young adult ( day 1 ) ASEL neuron responses to medium BZ ( 0 . 005% vol/vol ) in wild-type animals and in transgenic animals with AWC-specific tom-1 RNAi , AWC-specific ins-1 peptide OE and ASEL-specific daf-2 insulin receptor OE . ( B ) Plot of the averaged ΔF/F in the 15 s following odor removal for each individual young ( black dots ) or aged ( blue dots ) wild-type , AWC-specific tom-1 RNAi , AWC-specific ins-1 OE or ASEL-specific daf-2 OE transgenic animal . The red line represents the 10% ΔF/F cutoff used to classify neurons as odor responsive or non-responsive . *p < 0 . 05 , two-tailed t-test with Bonferroni correction comparing age-matched wild-type and transgenic animals; statistical analysis performed only on odor responsive subset of data . ( C ) Quantification of the percent of odor responsive neurons shown in A . ( D ) Chemotaxis behavior in young and aged wild-type and AWC-specific ins-1 OE animals , showing a trend towards dampened behavioral responses to medium BZ point sources in these transgenic animals . Numbers on bars indicate number of assay plates and error bars indicate s . e . m . NS , p > 0 . 05 , two-tailed t-test . See Figure 6—source data 2 for raw data . DOI: http://dx . doi . org/10 . 7554/eLife . 10181 . 032 We then tested whether primary AWC sensory neurons synthesize less neuropeptide as the animal ages , causing a breakdown in signaling to recruit ASEL neurons . To test this , we over-expressed the insulin-like neuropeptide INS-1 in the AWC neurons ( Figure 6A , right panel ) . This manipulation succeeded in improving the reliability of odor-evoked activity in aged ASEL neurons , suggesting that increased neuropeptide production , and consequently release , can rescue aging-associated deficits ( Figure 6D , F , Figure 6—figure supplement 1B ) . We also tested whether INS-1 ( OE ) could rescue aging-induced behavioral decline . We found that overexpressing INS-1 did not have a significant effect on the behavior of aged ( or young ) adults to BZ ( Figure 6—figure supplement 1D ) . Together , these results show that while INS-1 ( OE ) can rescue the age-induced neural activity deficits , this is not sufficient to rescue aging-induced behavioral deficits . We suggest that the temporal properties of neuropeptide signaling are likely to be complex and that the INS-1 ( OE ) might have predicted effects on short timescales ( a few seconds ) , but variable effects on longer timescales ( hours to days ) . To confirm a role for increased AWC neurotransmission in recruiting ASEL neurons , we also generated an AWC-specific RNAi knockdown of Tomosyn ( tom-1 in C . elegans [Gracheva et al . , 2007; Leinwand and Chalasani , 2013] ) , a syntaxin-interacting protein that normally acts as a brake on all neurotransmission , to increase neuropeptide and neurotransmitter release from AWC neurons ( Figure 6A , right panel ) . This manipulation to increase release from AWC neurons resulted in significantly more reliable odor-evoked ASEL activity ( Figure 6E , F ) . These manipulations did not significantly affect ASEL responses in young day 1 adults ( Figure 6—figure supplement 1A–C ) , suggesting that increased neurotransmission from the primary olfactory neurons specifically rescues the aging-associated ASEL defects . We also tested whether increased neurotransmission from AWC could rescue the aging-associated decline in chemotaxis behavior . We found that in aged day 5 adults , AWC-specific tom-1 knockdown animals showed a significant improvement over wild-type in their attraction to BZ ( Figure 6G ) . Moreover , this improvement required the presence of functional ASE neurons ( che-1 mutants do not have functional ASE neurons [Uchida et al . , 2003] ) ( Figure 6G ) . Taken together , these results show that experimental manipulations to increase neurotransmission from AWC neurons rescue aging-induced decline in ASEL secondary neuron activity and animal behavior . We have shown that AWA neurons release acetylcholine , which is required for AWB neuronal activity in the young adult odor circuit . We hypothesized that this process could be reduced during aging; therefore , we tested whether manipulations to increase neurotransmission from AWA neurons could rescue the decline in aged AWB neural activity . We over-expressed the vesicular acetylcholine transporter , UNC-17 , specifically in AWA neurons ( Figure 7A ) . OE of the vesicular acetylcholine transporter was previously shown to increase the quantity of acetylcholine packed into and released from synaptic vesicles ( Song et al . , 1997 ) . We found that this manipulation significantly increased the reliability of aged AWB odor responses ( Figure 7B , C , E , Figure 7—figure supplement 1B ) , further suggesting that increased signaling from the primary neurons can overcome aging-associated declines . We also confirmed a role for acetylcholine by using a pharmacological agent , arecoline . Arecoline is a cholinergic agonist known to act presynaptically to stimulate synaptic vesicle fusion ( Liu et al . , 2013 ) ( Figure 7A ) . Acute arecoline treatment in aged animals significantly increased the probability of AWB odor responses ( Figure 7D , E , Figure 7—figure supplement 1B ) , suggesting that a pharmacological approach to increase neurotransmission in aged animals can rejuvenate neuronal functions . Moreover , neither the UNC-17 OE nor acute arecoline had significant effects on AWB responses in day 1 adults ( Figure 7—figure supplement 1A–C ) , confirming a specific role for increased neurotransmission in rescuing aged-associated AWB defects . 10 . 7554/eLife . 10181 . 033Figure 7 . Increased release from AWA primary neurons rescues aging-associated AWB activity and behavioral deficits . ( A ) Schematic representation of genetic and pharmacologic manipulations to overcome aging-associated decay of neurotransmission . ( B–D ) Heat maps of ratio change in fluorescence to total fluorescence for aged adult ( day 5 ) AWB sensory neuron responses to the removal ( at t = 130 s ) of a two-minute medium BZ stimulus ( 0 . 005% vol/vol ) in ( B ) wild-type , ( C ) AWA-specific unc-17 OE , and ( D ) animals treated acutely with the cholinergic agonist arecoline . ( E ) Quantification of the percent BZ responsive aged AWB neurons in B–D . Odor responsive defined as having a ΔF/F to stimulus greater than 10% . *p < 0 . 05 , two-tailed Chi Square test . ( F ) Medium BZ chemotaxis in young and aged wild-type , AWB neuron ablated , AWA-specific unc-17 OE and AWB ablated in the AWA-specific unc-17 OE background . *p < 0 . 05 , two-tailed t-test with Bonferroni correction . See Figure 7—source data 1 for raw data . DOI: http://dx . doi . org/10 . 7554/eLife . 10181 . 03310 . 7554/eLife . 10181 . 034Figure 7—source data 1 . Odor responses in AWA-released neurotransmitter manipulation animal data . DOI: http://dx . doi . org/10 . 7554/eLife . 10181 . 03410 . 7554/eLife . 10181 . 035Figure 7—source data 2 . Additional odor responses in AWA-released neurotransmitter manipulation animal data . DOI: http://dx . doi . org/10 . 7554/eLife . 10181 . 03510 . 7554/eLife . 10181 . 036Figure 7—figure supplement 1 . AWA neurotransmission modifies aging-associated neuronal activity and behavioral deficits . ( A ) Heat maps of ratio change in fluorescence to total fluorescence for young ( day 1 ) adult AWB neuron responses to medium BZ ( 0 . 005% vol/vol ) in wild-type animals , transgenic animals with AWA-specific unc-17 OE and wild-type animals that received acute treatment with the cholinergic agonist arecoline . ( B ) Plot of the averaged ΔF/F in the 10 s after odor removal for each individual young ( black dots ) or aged ( blue dots ) wild-type , AWA-specific unc-17 OE or arecoline treated animal . The red line represents the 10% ΔF/F cutoff used to classify neurons as odor responsive or non-responsive . *p < 0 . 05 , two-tailed t-test with Bonferroni correction comparing wild-type and age-matched transgenic or drug treated animals; statistical analysis performed only on odor responsive subset of data . ( C ) Quantification of the percent of odor responsive neurons shown in A . ( D ) Chemotaxis behavior in young and aged wild-type animals that did or did not receive acute arecoline treatment . Numbers on bars indicate number of assay plates and error bars indicate s . e . m . *p < 0 . 05 , two-tailed t-test with Bonferroni correction . See Figure 7—source data 2 for raw data . DOI: http://dx . doi . org/10 . 7554/eLife . 10181 . 036 We also tested whether increased cholinergic transmission from the AWA neurons could rescue the aging-associated defects in behavioral attraction to BZ . We found that aged animals overexpressing the UNC-17 vesicular acetylcholine transporter in AWA neurons were significantly more attracted to BZ compared to aged wild-type animals ( Figure 7F ) . Moreover , this increased attraction required the secondary AWB neurons ( Figure 7F ) . These data confirm a role for AWA-AWB neurotransmission in rescuing aging-associated decline in BZ attraction . We note that while arecoline pharmacology rescued aged AWB neuronal activity , this treatment impaired BZ chemotaxis in both young and aged animals ( Figure 7—figure supplement 1D ) . We suggest that the known effect of arecoline to increase spontaneous locomotion may be counterproductive to the directed locomotion required to chemotax up an odor gradient ( Glenn et al . , 2004; Liu et al . , 2013 ) . Finally , we investigated the consequences of individual variation in aged olfactory abilities at the whole animal level by testing whether the olfactory abilities we analyzed could be correlated with longevity . We performed chemotaxis assays and separated the animals into two populations that did or did not navigate up an attractive BZ gradient ( Figure 8A ) . We then assayed the lifespan of these two populations of animals . Notably , we observed a significant extension ( average of 16 . 2% in three separate trials , p < 0 . 001 , Mantel–Cox test ) in the lifespan of animals that successfully chemotaxed to the odor as aged adults , compared to animals that failed to do so ( Figure 8B , Figure 8—figure supplement 1 ) . However , we found no difference in the lifespan of animals that were sorted on the basis of their chemotaxis performance as young adults ( Figure 8C , Figure 8—figure supplement 1 ) . These results suggest that the olfactory abilities of aged , but not young , animals may be correlated with their overall health , leading to lifespan differences . In contrast , we found that sorting aged animals based on their attraction to salt did not result in any significant differences in lifespan ( Figure 8—figure supplements 1 , 2A ) . These data show that the increase in lifespan is likely to be specific to BZ and not the salt associated neural circuit , consistent with the specific declines in BZ , not salt , evoked activity and behavior . Furthermore , these results indicate that the functionality of some , but not all , sensory neuronal circuits in early stage aged animals may predict animals' longevity . These data are also consistent with cell ablation experiments where loss of some chemosensory neurons affects C . elegans lifespan , while loss of other chemosensory neurons has no effect ( Alcedo and Kenyon , 2004 ) . Together , these results suggest that the olfactory prowess of aged animals is indicative of whole animal physiology , health and lifespan . 10 . 7554/eLife . 10181 . 037Figure 8 . Aged animal olfactory abilities and neurotransmission from primary neurons are correlated with lifespan . ( A ) Schematic of animals from a chemotaxis assay washed and sorted into two populations based on successful or failed navigation up the BZ odor gradient , for lifespan analysis . ( B ) Animals that chemotaxed to the BZ odor side of the chemotaxis plate as aged adults ( day 5 ) have a 16 . 2% average extension in their lifespan compared to animals from the opposite , control ( Ctrl ) side ( p < 0 . 01 by Mantel–Cox test , see Figure 8—figure supplement 1 and Figure 8—source data 1 for quantification ) . ( C ) Animals sorted by their young adult chemotaxis do not have significantly different lifespans ( see Figure 8—figure supplement 1 and Figure 8—source data 1 ) . ( D ) AWA-neuron specific unc-17 OE transgenic animals have a 26 . 6% average extension in lifespan compared to wild-type animals ( p < 0 . 0001 by Mantel–Cox test , see Figure 8—figure supplement 3 and Figure 8—source data 1 for quantification ) . ( E ) Survival of wild-type , AWC-neuron specific tom-1 RNAi , and AWC-specific ins-1 OE transgenic animals ( see Figure 8—figure supplement 3 and Figure 8—source data 1 for quantification ) . ( B–E ) Mean survival is reported in days of adulthood . BZ , benzaldehyde; OE , overexpression . DOI: http://dx . doi . org/10 . 7554/eLife . 10181 . 03710 . 7554/eLife . 10181 . 038Figure 8—source data 1 . Lifespan of animals sorted by their chemotaxis performance and lifespan of neurotransmitter manipulation transgenic animal data . DOI: http://dx . doi . org/10 . 7554/eLife . 10181 . 03810 . 7554/eLife . 10181 . 039Figure 8—source data 2 . Additional lifespan of neurotransmitter manipulation transgenic animal data . DOI: http://dx . doi . org/10 . 7554/eLife . 10181 . 03910 . 7554/eLife . 10181 . 040Figure 8—figure supplement 1 . Sorting animals based on their performance on odor chemotaxis affects lifespan . Wild-type ( N2 ) worms were separated into a population that successfully reached the BZ odor or salt side of the chemotaxis plate and a population that failed to do so ( Ctrl side ) as young adults ( day 1 ) or aged adults ( day 5 ) and then their survival was analyzed . Data from three separate trials are shown for odor experiments and two separate trials for salt experiments . Animals were censored if they bagged , exploded or desiccated on the side of the plate . Mean survival , s . e . m . of survival , median survival and percent change in mean survival are reported in days of adulthood . *p < 0 . 05 by the Mantel–Cox test ( the chi-square statistic value is reported in parentheses ) . The percent change in mean survival was calculated as the mean survival of animals from the BZ odor ( or salt ) side minus the mean survival of animals from the Ctrl side divided by the mean survival of the BZ odor ( or salt ) side . BZ , benzaldehyde . DOI: http://dx . doi . org/10 . 7554/eLife . 10181 . 04010 . 7554/eLife . 10181 . 041Figure 8—figure supplement 2 . Sorting animals based on their performance on salt chemotaxis and silencing primary neurons modifies lifespan . ( A ) Animals sorted by their aged ( day 5 ) adult chemotaxis to sodium chloride do not have significantly different lifespans ( see Figure 8—figure supplement 1 and Figure 8—source data 2 for quantification ) . ( B ) AWC or AWA-neuron specific tetanus toxin expression to silence these neurons significantly extends lifespan compared to wild-type ( +27 . 6% and +36 . 0% , respectively , *p < 0 . 01 by Mantel–Cox test , see Figure 8—figure supplement 3 and Figure 8—source data 2 for quantification ) . ( A , B ) Mean survival is reported in days of adulthood . DOI: http://dx . doi . org/10 . 7554/eLife . 10181 . 04110 . 7554/eLife . 10181 . 042Figure 8—figure supplement 3 . Manipulating neurotransmission from primary olfactory neurons modifies lifespan . The survival of wild-type , AWC-neuron specific tom-1 RNAi , AWC-specific ins-1 OE , AWA-specific unc-17 OE , AWA-specific tetanus toxin expression and AWC-specific tetanus toxin expression transgenic animals was analyzed . Data from two or three separate trials ( as indicated ) is shown . Animals were censored if they bagged , exploded or desiccated on the side of the plate . Mean survival , s . e . m . of survival , median survival and percent change in mean survival are reported in days of adulthood . *p < 0 . 05 by the Mantel–Cox test ( the chi-square statistic value is reported in parentheses ) . The percent change in mean survival was calculated as the mean survival of the transgenics minus mean survival of wild-type , divided by mean survival of wild-type . DOI: http://dx . doi . org/10 . 7554/eLife . 10181 . 042 We next investigated the mechanisms linking aged olfactory abilities and longevity . We tested whether more reliable olfactory circuit functioning resulting from increased neurotransmission from the primary AWA or AWCON neurons affected animal lifespan . We found that animals overexpressing the UNC-17 vesicular acetylcholine transporter in AWA neurons lived an average of 26 . 6% longer than their wild-type counterparts ( over three separate trials , p < 0 . 001 , Mantel–Cox test ) ( Figure 8D , Figure 8—figure supplement 3 ) . Moreover , increasing neurotransmission from AWC either by overexpressing the INS-1 peptide or by knocking down tom-1 , the C . elegans homolog of Tomosyn , resulted in a trend towards a small extension in lifespan ( Figure 8E , Figure 8—figure supplement 3 ) . These data suggest that both classical neurotransmission ( from AWA neurons ) and neuropeptide signaling ( from AWC neurons ) , which are key components of the combinatorial code for BZ , may have a longevity promoting effect . These data are in contrast with previously published results showing that animals with AWC or AWA neurons ablated live longer ( Alcedo and Kenyon , 2004 ) ; therefore , we probed the role of these neurons in lifespan further . We analyzed the lifespan of animals in which AWA or AWC neurons were silenced rather than ablated . Animals expressing tetanus toxin in either AWA or AWC neurons to block their neurotransmission lived significantly longer than wild-type ( average of 36 . 0% and 27 . 6% longer , respectively , over two independent trials , p < 0 . 01 , Mantel–Cox test , Figure 8—figure supplement 2B , 3 ) . Together , these results suggest that a balance in neurotransmission from the primary olfactory neurons is crucial to an animal's longevity; both higher than normal and lower than normal levels of neurotransmission extend lifespan . We suggest that signaling from these primary olfactory neurons is integrated by the downstream circuitry to mediate effects on the animal's lifespan .
Our results define a novel neural circuit mechanism for encoding sensory information to drive behavior and demonstrate age-related functional declines in this circuit . These data provide the first indication that C . elegans employ a combinatorial olfactory coding strategy as in flies and mice , suggesting that this strategy is essential for behavioral plasticity ( Wang et al . , 2003; Oka et al . , 2006 ) . Moreover , we suggest that primary olfactory neurons directly detect odors and use neurotransmission to recruit additional secondary neurons . However , activity in the secondary neurons declines with aging , leading to behavioral deficits . We propose that the combination of primary and secondary neurons may be a common motif in sensory neural circuits from worms to mammals . A distinct , but similarly distributed neural circuit ( which does not include the ASE neurons ) encodes a different attractive odorant , isoamyl alcohol ( data not shown , [Yoshida et al . , 2012] ) . Furthermore , we have previously shown that the C . elegans salt neural circuit is composed of a primary salt sensory neuron , ASEL , which releases INS-6 insulin neuropeptides to recruit a secondary sensory neuron , AWCON , into the circuit in particular sensory contexts ( Leinwand and Chalasani , 2013 ) . This combined primary and secondary neuron coding strategy is likely to increase the signal-to-noise ratios , thus preventing failures in encoding sensory information . Combinatorial coding of this sort may also be broadly useful for distinguishing different concentrations of the same stimulus , as they will be encoded by overlapping but distinct subsets of neurons . This approach may also enhance the ability of young adults to successfully find food , perhaps to enhance reproductive success , while the aging-associated declines occur in post-reproductive animals that may have reduced nutritional demands . Furthermore , the insulin peptidergic and cholinergic signaling from primary to secondary olfactory neurons could add salience to volatile food signals in a complex , multisensory environment . Previous studies have shown that insulin ( Lacroix et al . , 2008 ) and cholinergic receptors ( Ogura et al . , 2011 ) are expressed in mammalian olfactory processing centers , suggesting that these signaling pathways might also be used to encode odors in mammals . Detailed analyses of the architecture of sensory circuits , including the neurotransmission between sensory neurons , in other species are needed to determine whether the circuit motif described here is broadly conserved . We find that primary sensory transduction remains robust as animals age . However , the combinatorial code for attractive volatile cues degrades with age because the activity of cells functioning as secondary neurons decays with age . Our results show that aging-induced decline in neuronal function is dependent on the interplay between sensory context and neuronal identity . For example , we find that primary ASEL responses to salt and AWB responses to 2-nonanone are preserved while their secondary responses to BZ are reduced in aged animals . This is in contrast with studies showing an early stage age-induced decline in the primary ASH neuron responses to hyperosmotic stimuli ( Chokshi et al . , 2010 ) . Taken together , these results indicate that aging differentially affects sensory circuits , perhaps reflecting differences in physiological demand and the importance of diverse sensory contexts as the animal ages . Furthermore , these aging-associated sensory declines occur independently of many known longevity pathways . Insulin signaling has been shown to promote longevity in a number of model systems including C . elegans , Drosophila melanogaster and Mus musculus ( Broughton and Partridge , 2009; Kenyon , 2010 ) . We find that insulin signaling is required for the primary AWCON to secondary ASEL neurotransmission and so are unable to separate its longevity promoting effect from its role in encoding sensory information . We speculate that the insulin signaling pathway might affect both the quality of an animal's life by encoding odors based on sensory context and also its lifespan . Our experiments show that experimental manipulations targeting neurotransmission pathways improve the aging-associated neuronal activity and olfactory behavioral declines . Several different mechanisms could underlie the impairments observed in aged animals and overcome by our manipulations . A decline in peptidergic and cholinergic gene expression with age could contribute; however , quantitative RT-PCR experiments suggest that there is no aging-associated reduction in the expression of these genes at the whole animal level [data not shown and ( Jin et al . , 2011 ) ] . Changes in gene expression specifically in the primary olfactory neurons cannot be ruled out . Nevertheless , we speculate that the early aging-associated sensory impairments are driven at least in part by reduced neurotransmitter release from primary neurons , a mechanism likely applicable across species . We find that both increasing neurotransmitter production and release capacity rescue the aging-associated deficits . Therefore , it is likely that aging affects multiple steps in the neurotransmitter release pathway , emphasizing the key role played by this machinery in regulating animal behavior and physiology . These results are consistent with reports of reduced synapse number in the aged mammalian olfactory bulb , which should disrupt olfactory circuits ( Richard et al . , 2010 ) . We speculate that these differences in synaptic transmission also explain some of the inter-individual variability in aging phenotypes ( Pinto et al . , 2014; Vijg , 2014 ) . Subsequently , these circuit-level changes could produce hyposmia or anosmia , which may be among the earliest predictors of lifespan and mortality across species ( Toth et al . , 2012; Liu et al . , 2013; Pinto et al . , 2014 ) . More generally , we suggest that alterations in transmitter release , which disrupt neuronal communication throughout the brain ( Dickstein et al . , 2007 ) are likely to underlie variability in individual animal behavior and age-related cognitive and behavioral decline .
cDNA corresponding to the entire coding sequences of unc-31 ( isoform a ) , daf-2 ( isoform a ) , age-1 ( isoform a ) , tom-1 ( isoform a ) , and the ins-1 genomic region were amplified by PCR and expressed under cell-selective promoters . unc-17 cDNA was synthesized ( GenScript ) and expressed under a cell-selective promoter . For cha-1 and cho-1 knockdown experiments , 1 kb fragments corresponding to exons 3–7 and the 3′ end of the gene , respectively , in the sense and antisense orientation were synthesized ( GenScript ) . Neuron-selective RNAi transgenes were created as previously described by co-injection of equal concentrations of sense and antisense oriented gene fragments driven by cell-specific promoters ( Esposito et al . , 2007 ) . Cell-specific expression was achieved using the following promoters: ceh-36deletion or odr-3 for both AWC , str-2 for AWCON , srsx-3 for AWCOFF , gpa-4 for AWA and ASI , gpa-4deletion for AWA , gcy-7 for ASEL , gcy-5 for ASER , str-1 for AWB , sre-1 for ADL , srh-142 for ADF , gcy-8 for AFD , ops-1 for ASG , sra-6 for ASH , trx-1 for ASJ and sra-9 for ASK . For all experiments , a splice leader ( SL2 ) fused to a mCherry or gfp transgene was used to confirm cell-specific expression of the gene of interest . Germline transformations were performed by microinjection of plasmids ( Mello and Fire , 1995 ) at concentrations between 25 and 200 ng/μl with 10 ng/μl of unc-122::rfp , unc-122::gfp or elt-2::gfp as co-injection markers . For rescue and OE experiments , DNA was injected into mutant or wild-type C . elegans carrying GCaMP arrays . Transgenic worms expressing GCaMP calcium indicators under a cell-selective promoter were grown to day 1 or day 5 of adulthood and trapped in a custom designed PDMS microfluidic device and exposed to odor stimuli ( Chalasani et al . , 2007; Chronis et al . , 2007 ) . For aging experiments , a new PDMS device with larger channels was designed to trap and stimulate day 5 adult worms ( Chokshi et al . , 2010 ) . Older , day 6 adult worms exhibit much larger variation in whole animal size than day 5 adults ( see Figure 5—figure supplement 1A ) and could not be trapped consistently without introducing bias into the experiment . For aging experiments , animals were transferred to new OP50 bacteria plates every other day to track the aging animals and to avoid contamination by their progeny . Additionally , for whole animal RNAi experiments to knockdown rab-10 and hsf-1 , animals were fed either control ( Ctrl ) empty pL4440 , rab-10 RNAi or hsf-1 RNAi expressing bacteria beginning at day 1 of adulthood as previously described ( Hansen et al . , 2005 ) . Fluorescence from the neuronal cell body was captured using a Zeiss inverted compound microscope for 3 min . We first captured 10 s of baseline activity ( t = 0–10 s ) in chemotaxis assay buffer ( 5 mM K3PO4 ( pH 6 ) , 1 mM CaCl2 , 1 mM MgSO4 , and 50 mM NaCl ) , then 2 min ( t = 10–130 s ) of exposure to an odor ( or salt ) stimulus dissolved in chemotaxis buffer , and lastly 50 s ( t = 130–180 s ) of buffer only . BZ refers to a 0 . 005% vol/vol dilution in chemotaxis assay buffer , except where low BZ ( 0 . 0001% vol/vol ) or high BZ ( 0 . 1% vol/vol ) is specifically mentioned . Additionally , a 0 . 1% vol/vol dilution of 2-nonanone and 50 mM sodium chloride stimulus were used as indicated . For arecoline experiments , worms were pre-treated with 0 . 15 mM arecoline in chemotaxis buffer for approximately 20 min and immediately imaged in the presence of the drug . Laser ablations of the paired AWC , AWA , AWB or ASE sensory neurons , along with mock ablations , were performed as previously described ( Bargmann and Avery , 1995 ) in transgenic animals expressing GCaMP . In all experiments , a single neuron was imaged in each animal , and each animal was imaged only once . Wild-type Ctrls , mutants , and transgenic or drug treated strains for each figure were imaged in alternation , in the same session . We used Metamorph and an EMCCD camera ( Photometrics ) to capture images at a rate of 10 frames per second . A MATLAB script was used to analyze the average fluorescence for the cell body region of interest and to plot the percent change in fluorescence for the region of interest relative to F0 , as previously described ( Chalasani et al . , 2007 ) . Specifically , data was plotted and statistical analysis was performed as follows: ( 1 ) for line graphs of ΔF/F over time ( Figures 1–4 and corresponding figure supplements ) , the average fluorescence in a 8 s window ( t = 1–9 s ) was set as F0 . Average and standard error at each time point were generated and plotted using MATLAB scripts , as previously described ( Leinwand and Chalasani , 2013 ) . ( 2 ) For heat maps ( Figures 5–7 and corresponding figure supplements ) , the average fluorescence in a 8 s window ( t = 1–9 s ) was set as F0 . To quantify calcium responses , F0 was consistently set to the average fluorescence signal from 1 s to 9 s prior to the relevant change ( addition or removal ) of stimulus . For statistical analysis , the average fluorescence and standard error were calculated for each animal over a short period corresponding to the duration of a response . Specifically , to analyze on responses to the addition of stimulus , the average fluorescence and standard error were calculated in the 10 s period following the addition of odor or salt ( t = 10–20 s ) . For AWA neurons , the response duration was very brief; therefore , a 4 s time period was used instead ( t = 10–14 s ) so that small , fast responses could be appropriately quantified . To analyze off responses to the removal of stimulus , the average fluorescence and standard error were calculated in the period following the removal of odor ( t = 130–140 s for all cells except ASE , and t = 130–145 for the slower , longer duration ASE responses ) . Traces in which an averaged ΔF/F of greater than 600% was recorded were excluded as they are likely to be artifacts of the neurons moving out of the focal plane and these usually account for less than 1% of the traces collected . To determine whether there was an odor-evoked increase or suppression of the calcium signal ( see Figure 1C ) , the average fluorescence in these time windows in buffer only trials was compared ( by a two-tailed unpaired t-test ) to the average fluorescence in odor stimulation trials , for each neuron . The maximum ΔF/F in these time periods following odor addition or removal and the time to reach this maximum ΔF/F ( from the stimulus change , in seconds ) were also quantified ( see Figure 5G , Figure 1—figure supplement 1A , B and Figure 5—figure supplement 1B ) . More specifically: ( 1 ) For bar graphs of averaged ΔF/F after odor addition or removal ( Figures 2–4 , Figure 3—figure supplement 1 and Figure 4—figure supplement 1 ) : ( a ) F0 was set to the average fluorescence from 1–9 s for quantification of AWA neuron responses to the addition of BZ stimulus and ( b ) F0 was set to the average fluorescence from 121–129 s for quantification of AWC , ASE and AWB responses to the removal of BZ or 2-nonanone . Two-tailed unpaired t-tests were used to compare the responses of different genotypes or cell ablation conditions , and the Bonferroni correction was used to adjust for multiple comparisons . ( 2 ) For scatter plots of maximum ΔF/F ( Figure 1—figure supplement 1A and Figure 5G ) and scatter plots of averaged ΔF/F after stimulus change ( Figure 5H , Figure 6—figure supplement 1B and Figure 7—figure supplement 1B ) : ( a ) for AWA neurons' response to the addition of odor stimulus F0 was set to the average fluorescence from 1–9 s and ( b ) for AWCON , ASEL and AWB responses to odor stimulus removal F0 was set to the average fluorescence from 121–129 s . For the subset of odor-responsive neurons ( exceeding the 10% ΔF/F cut-off ) , the averaged ΔF/F after the stimulus change and the time to the maximum ΔF/F were also analyzed using two-tailed unpaired t-tests to compare different ages or genotypes ( Figure 5H , Figure 5—figure supplement 1B , Figure 6—figure supplement 1B and Figure 7—figure supplement 1B ) . Furthermore , considering only the odor responsive neurons , no significant differences were observed in the magnitude of the odor-evoked suppression of young and aged animals ( comparing the average fluorescence in ten second windows tiling the period of odor stimulation , by two-tailed t-test ) , indicating that our subsequent analyzes of the odor removal time period are not biased by the choice of the F0 . ( 3 ) For bar graph quantifications of the % odor or salt responsive neurons in the aging experiments ( Figure 5I , K , L , 6F , and 7E and the corresponding figure supplements ) : ( a ) F0 was set to the average fluorescence from 1–9 s for quantification of the percent of AWA and ASH neurons responsive to the addition of BZ stimulus and for the percent of ASEL and AWC neurons responsive to the addition of NaCl salt stimulus . ( b ) F0 was set to the average fluorescence from 121–129 s for AWCON , ASEL and AWB responses to BZ or 2-nonanone odor stimulus removal . The percent of odor responsive neurons was calculated by determining the proportion of cells displaying an average fluorescence ( ΔF/F ) greater than 10% after odor addition ( for AWA and ASH ) or odor removal ( all other neurons ) . 10% ΔF/F was used as the cut-off for odor responsiveness because , for all neurons imaged , changing buffer around the nose of the animal elicited a response smaller than this cut-off . Similarly , neurons displaying an average fluorescence ( ΔF/F ) greater than 10% after salt addition were considered to be salt responsive . A two-tailed Chi–Square test was used to compare the percent of odor or salt responsive neurons in different conditions . Odor chemotaxis assays were performed as previously described ( Ward , 1973 ) . For aging assays , worms were synchronized by hatch offs in which 8 young adult worms were given 150 min to lay eggs on a large plate before being picked off . These eggs were grown at 20° until the appropriate day of adulthood , except for glp-1 mutants , which were raised at the restrictive temperature , 25° . Aging animals were transferred to new bacteria plates every other day to track the aging animals and to avoid contamination by their progeny . Chemotaxis assays were performed on 2% agar plates ( 10 cm diameter ) containing 5 mM potassium phosphate ( pH 6 ) , 1 mM CaCl2 and 1 mM MgSO4 . Animals were washed once in M9 and three times in chemotaxis buffer ( 5 mM K3PO4 ( pH 6 ) , 1 mM CaCl2 and 1 mM MgSO4 ) . For arecoline chemotaxis experiments , 0 . 15 mM arecoline was added to the M9 and chemotaxis buffer washes , yielding a 16–20 min drug treatment immediately prior to the behavioral experiment . Odor concentration gradients were established by spotting diluted BZ ( 0 . 2% vol/vol , in ethanol ) near the edge of the plate , with a Ctrl 1 μl of ethanol spotted at the opposite end of the plate . Where noted , 1 μl of neat BZ was used for high concentration point source assays . For 2-nonanone experiments , a 50% vol/vol dilution of 2-nonanone in ethanol was used . For salt chemotaxis experiments , salt gradients were established by placing a Ctrl or a high salt ( 500 mM NaCl ) agar plug on the assay plate and allowing 16–20 hr for the salt to diffuse and form a gradient ( Leinwand and Chalasani , 2013 ) . 1 μl of sodium azide was added to the odor ( or salt ) and the Ctrl spots to anesthetize animals reaching the end points . Washed worms were placed on the plate and allowed to move freely for one hour . The chemotaxis index was computed as the number of worms in the region near the odor ( or salt ) minus the worms in the region near the Ctrl divided by the total number of worms that moved beyond the origin . Nine or more assays were performed , over at least three different days . Two-tailed unpaired t-tests were used to compare the responses of different genotypes or ages , and the Bonferroni correction was used to adjust for multiple comparisons . Transgenic worms bearing GCaMP arrays , synchronized by a hatch off as described above , were grown until day 5 of adulthood at 20° . Aging animals were transferred to new bacteria plates every other day to track the aging animals and to avoid contamination by their progeny . Animals were tested in a ( 0 . 2% vol/vol ) BZ odor chemotaxis assay as above , with two modifications . First , no sodium azide was used to paralyze the animals . Second , animals were given only 30 min to move freely on the chemotaxis plate . The chemotaxis assay plate was then cut into three regions corresponding to the BZ odor side , the middle , and the ethanol Ctrl region immediately after 30 min and worms were washed off each section separately and allowed to recover on OP50 bacteria plates for at least 90 min . Worms from the odor and the Ctrl sections of the chemotaxis assay were imaged in alternation as described above . Worms , synchronized by a hatch off as described above , were grown until day 1 or 5 of adulthood at 20° . To sort animals on the basis of their chemotaxis performance , wild-type animals were tested in a ( 0 . 2% vol/vol ) BZ odor or ( 500 mM NaCl ) salt chemotaxis assay as above , but without sodium azide and with only 30 min for the animals to move freely on the chemotaxis plate . The chemotaxis assay plate was then cut into a BZ odor ( or salt ) , middle , and Ctrl region and worms were washed off each section separately . 100 adults from the odor ( or salt ) region or the Ctrl region were transferred onto 10 small OP50 plates ( 10 adults per plate ) and grown at 20° . For experiments with transgenic animals , day 1 animals bearing the appropriate transgene were picked from the hatch off plate directly onto 10 small OP50 plates ( 10 adults per plate ) and grown at 20° . In all experiments , aging animals were transferred to new bacteria plates every other day to track the aging animals and to avoid contamination by their progeny . Survival was analyzed every other day and worms were scored alive or dead based on their response to a gentle head touch ( or lack thereof ) as previously described ( Kenyon et al . , 1993 ) . Worms were censored if they bagged , exploded or desiccated on the side of the plate . The chemotaxis assay followed by lifespan analysis or lifespan assays with transgenic animals were repeated two or three times per condition as indicated , beginning on separate days . The percent change in mean survival was calculated as the mean survival of animals from the odor side minus the mean survival of animals from the Ctrl side divided by the mean odor side survival or the mean transgenic animal survival minus the mean wild-type survival divided by the mean wild-type survival . Statistical analysis of lifespan was performed by the Mantel–Cox Log–Rank test , using GraphPad Prism and OASIS ( Yang et al . , 2011 ) . Chemotaxis assays to BZ were set up as described above , but with modifications to enable automated analysis of animal speed . 200 mM Cu ( II ) SO4-soaked filter paper was placed on a standard chemotaxis assay plate to contain the worms in a reduced chemotaxis arena ( 1 . 25 by 1 . 25 inch square ) . 1 μl of BZ ( 0 . 2% vol/vol dilution in ethanol ) and a Ctrl 1 μl of ethanol were spotted at opposite corners of the square arena , without any paralytic . After washing , only 5 worms were placed on the chemotaxis plate; this number minimized collisions and enabled more accurate tracking . The movement of the animals was tracked over 60 min using a Pixelink camera and speed was analyzed using previously published MATLAB scripts to track the centroid of the animal ( Ramot et al . , 2008 ) . The results from eleven chemotaxis plates were averaged for each age . NS indicates p > 0 . 05 , two-tailed t-test . Day 5 and day 6 adult worms from hatch offs performed on three separate days were immobilized with tetramisole and imaged on 2% agarose pads . Images were captured on a Zeiss Observer D1 microscope using a 10× objective with DIC . The perimeters of 55 worms were measured using MetaMorph software . | A sense of smell can help animals to find food and detect danger . Odor molecules activate so-called olfactory neurons that relay signals to the brain in the form of nerve impulses . This information is then processed , and the appropriate response is triggered; for example , an animal might move towards the smell of food , or away from the scent of a predator . But how can the activity of olfactory neurons trigger the right behavioral response ? Leinwand et al . have now explored the activity of olfactory neurons in a roundworm called C . elegans . The experiments revealed that a food odor activated two olfactory neurons directly , and that each of these ‘primary’ neurons then in turn activated another ‘secondary’ olfactory neuron . This communication between primary and secondary olfactory neurons was essential for worms to respond to the food odor . Further experiments revealed that the primary olfactory neurons send chemical signals , called neurotransmitters and neuropeptides , to communicate with the secondary neurons . Importantly , mutations that blocked this chemical signaling prevented the worms from responding appropriately to the smell of food . Aging animals , including people , often have impaired senses and can therefore find it difficult to identify and respond to odors . Leinwand et al . found that aged worms were no different . Further experiments suggested that aging worms' responses to odor decline because the communication between the primary and secondary olfactory neurons may be impaired with age . When Leinwand et al . strengthened this communication it reversed the effects of aging on the worms' sense of smell . Moreover , the experiments also showed that an animal's performance on the odor task was correlated with its longevity , such that the better performers also lived longer . A challenge for the future is to understand the precise changes that occur at early stages of aging to impair the sense of smell . Future studies could also test if similar combinations of olfactory neurons are needed to trigger certain behavioral responses to odors in young and old mammals . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2015 | Circuit mechanisms encoding odors and driving aging-associated behavioral declines in Caenorhabditis elegans |
Genetic variants regulating RNA splicing and transcript usage have been implicated in both common and rare diseases . Although transcript usage quantitative trait loci ( tuQTLs ) have been mapped across multiple cell types and contexts , it is challenging to distinguish between the main molecular mechanisms controlling transcript usage: promoter choice , splicing and 3ʹ end choice . Here , we analysed RNA-seq data from human macrophages exposed to three inflammatory and one metabolic stimulus . In addition to conventional gene-level and transcript-level analyses , we also directly quantified promoter usage , splicing and 3ʹ end usage . We found that promoters , splicing and 3ʹ ends were predominantly controlled by independent genetic variants enriched in distinct genomic features . Promoter usage QTLs were also 50% more likely to be context-specific than other tuQTLs and constituted 25% of the transcript-level colocalisations with complex traits . Thus , promoter usage might be an underappreciated molecular mechanism mediating complex trait associations in a context-specific manner .
Genome-wide association studies ( GWAS ) have discovered thousands of genetic loci associated with complex traits and diseases . However , identifying candidate causal genes and molecular mechanisms at these loci remains challenging . Complex trait-associated variants are enriched in regulatory elements and are therefore thought to act via regulation of gene expression levels , often in a cell type- and context-specific manner ( Alasoo et al . , 2018; Fairfax et al . , 2014; Kim-Hellmuth et al . , 2017 ) . However , such variants are equally enriched among splicing quantitative trait loci ( QTLs ) ( Fraser and Xie , 2009; Li et al . , 2016 ) and incorporating splicing QTLs in a transcriptome-wide association study increased the number of disease-associated genes by twofold ( Li et al . , 2018 ) . In addition to splicing , genetic variants can also alter transcript sequence by regulating promoter and 3ʹ end usage , which we refer to collectively hereafter as transcript usage QTLs ( tuQTLs ) . Alternative transcript start and end sites underlie most transcript differences between tissues ( Pal et al . , 2011; Reyes and Huber , 2018 ) , they are dynamically regulated in response to cellular stimuli ( Alasoo et al . , 2015; Richards et al . , 2017 ) and they are also frequently dysregulated in cancer ( Demircioğlu et al . , 2017; Lee et al . , 2018 ) . Moreover , experimental procedures designed to capture either 5ʹ or 3ʹ ends of transcripts have identified disease-relevant genetic variants that regulate promoter or 3ʹ end usage ( Garieri et al . , 2017; Zhernakova et al . , 2013 ) . However , well-powered RNA-seq-based tuQTL studies performed across cell types ( Battle et al . , 2014; Chen et al . , 2016; Lappalainen et al . , 2013; Li et al . , 2016; Ongen and Dermitzakis , 2015 ) and conditions ( Nédélec et al . , 2016; Ye et al . , 2018 ) have thus far not distinguished between promoter usage , splicing and 3ʹ end usage . Thus , how these distinct transcriptional mechanisms contribute to complex traits and how context-specific these genetic effects are is currently unclear . In addition to splicing analysis , RNA-seq data can also be used to quantify promoter and 3ʹ end usage . The simplest approach would be to first quantify the expression of all annotated transcripts using one of the many quantification algorithms ( benchmarked in Teng et al . , 2016 ) . Linear regression can then be used to identify genetic variants that are associated with the usage of each transcript of a gene ( Li et al . , 2018; Ongen and Dermitzakis , 2015 ) . Comparing the associated transcripts to each other can reveal which transcriptional changes take place ( Figure 1A ) . A key assumption here is that all expressed transcripts are also part of the annotation catalog . If some of the expressed transcripts are missing , then reads originating from the missing transcripts might be erroneously assigned to other transcripts that are not expressed at all ( Figure 1B ) ( Soneson et al . , 2018 ) . This can lead to individual genetic variants being spuriously associated with multiple transcriptional changes . For example , a genetic variant regulating promoter usage might also appear to be associated with the inclusion of an internal exon ( Figure 1B ) , although there are no reads originating from that exon . Importantly , this is not just a theoretical concern , because 25–35% of the exon-exon junctions observed in RNA-seq data are not present in transcript databases ( Ongen and Dermitzakis , 2015 ) , and up to 60% of the transcripts annotated by Ensembl ( Zerbino et al . , 2018 ) are truncated at the 5ʹ or 3ʹ end ( Figure 1—figure supplement 1 , Figure 1—figure supplement 2 ) . To overcome the issue of missing transcript annotations , recent tuQTL studies have focussed on quantifying transcription at the level of individual exons ( Fadista et al . , 2014; Lappalainen et al . , 2013; Odhams et al . , 2017 ) , introns ( Odhams et al . , 2017 ) or exon-exon junctions ( Figure 1C ) ( Li et al . , 2018; Odhams et al . , 2017; Ongen and Dermitzakis , 2015 ) . While these approaches often discover complementary genetic associations ( Odhams et al . , 2017; Ongen and Dermitzakis , 2015 ) , they do not explicitly reveal the transcriptional mechanism ( promoter usage , alternative splicing or 3ʹ end usage ) underlying the genetic associations . The most successful approach to differentiate between distinct transcriptional mechanisms has been ‘event-level’ analysis where reference transcripts are split into independent events ( e . g . promoters , splicing events and 3ʹ ends ) whose expressions is then quantified using standard transcript quantification methods ( Figure 1C ) . This approach was pioneered by MISO ( Katz et al . , 2010 ) and was recently used to identify promoter usage QTLs in the GEUVADIS dataset ( Richards et al . , 2017 ) . Despite its success , MISO covers only a subset of promoter events ( alternative first exons ) and its event annotations have not been updated since it was first published . Thus , there is a need for a method that is able to detect a comprehensive set of promoter , splicing and 3ʹ end usage QTLs in an uniform manner . In this study , we re-analysed RNA-seq data from human induced pluripotent stem cell-derived macrophages ( IPSDMs ) exposed to three inflammatory stimuli ( 18 hr IFNɣ stimulation , 5 hr Salmonella infection and IFNɣ stimulation followed by Salmonella infection ) ( Alasoo et al . , 2018 ) . We also collected a new dataset of IPSDMs from 70 individuals stimulated with acetylated LDL ( acLDL ) for 24 hr . We mapped genetic associations at the level of total gene expression , full-length transcript usage and exon-exon junction usage in each experimental condition . In addition to existing quantification methods , we also developed a complementary approach ( txrevise ) that stratifies reference transcript annotations into independent promoter , splicing and 3ʹ end events . Using txrevise , we found that promoter and 3ʹ end usage QTLs constituted 55% of detected tuQTLs , exhibited distinct genetic architectures from canonical expression or splicing QTLs , and often colocalised with complex trait associations . Promoter usage QTLs were also 50% more likely to be context-specific than canonical splicing QTLs . Thus , context-specific regulation of promoter usage might be a previously underappreciated molecular mechanism underlying complex trait associations .
We analysed RNA-seq data from human induced pluripotent stem cell ( iPSC ) -derived macrophages exposed to three inflammatory stimuli ( 18 hr IFNɣ stimulation , 5 hr Salmonella infection , and IFNɣ stimulation followed by Salmonella infection ) and one metabolic stimulus ( 24 hr acLDL stimulation ) . While the gene expression analysis of the IFNɣ+Salmonella dataset from 84 individuals has previously been described ( Alasoo et al . , 2018 ) , the acLDL data from 70 individuals was newly generated for the current study . The acLDL dataset allowed us to assess how our results generalise to weaker , non-inflammatory stimuli . Both datasets included independent unstimulated control samples ( denoted as ‘naive’ and ‘Ctrl’ ) . In each condition , we quantified gene expression and transcript usage using the following established quantification approaches: ( i ) gene-level read count quantified with featureCounts ( Liao et al . , 2014 ) , ( ii ) full-length transcript usage quantified with Salmon ( Patro et al . , 2017 ) ( Figure 1C ) , and ( iii ) exon-exon junction usage quantified with Leafcutter ( Li et al . , 2018 ) ( Figure 1C ) . Inspired by event level analysis proposed by MISO ( Katz et al . , 2010; Richards et al . , 2017 ) , we also developed a complementary approach ( txrevise ) to stratify reference transcript annotations into independent promoter , splicing and 3ʹ end events . To achieve this , txrevise identifies constitutive exons shared between all transcripts of a gene and uses those to assign non-constitutive exons to promoter , internal exon or 3ʹ end events ( Figure 1C ) . Since up to 60% of the transcripts annotated by Ensembl ( Zerbino et al . , 2018 ) are truncated at the 5ʹ or 3ʹ end ( Figure 1—figure supplement 1 ) , txrevise extends truncated transcripts by copying over exons from the longest transcript of the gene ( Figure 1—figure supplement 2 ) . This step eliminates a large number of implausible alternative promoter and 3ʹ end events that lack experimental evidence . To make the approach suitable for genes with non-overlapping transcripts , we also select a subset of transcripts that share the largest number of exons between them ( Figure 1—figure supplement 3 ) . Finally , to ensure that the new alternative promoter and 3ʹ end events do not capture splicing changes , txrevise masks alternative exons in promoter and 3ʹ end events that are not the first or last exons ( Figure 1—figure supplement 4 ) . Although this means that some splicing events near the promoters and 3ʹ ends of the genes may remain undetected by txrevise , it is a trade-off that improves the overall interpretability of txrevise tuQTLs . The R package as well as custom transcriptional events constructed by txrevise are available from GitHub ( https://github . com/kauralasoo/txrevise; Alasoo , 2018a ) . Depending on the experimental condition and quantification method , we detected between 1500 and 3500 QTLs at a 10% false discovery rate ( FDR ) ( Figure 2A ) . Leafcutter consistently detected the lowest number of QTLs per condition , while txrevise detected approximately 30% more associations than other methods ( Figure 2A ) , 55% of which affected promoter or 3ʹ end usage instead of internal exons ( Figure 2—figure supplement 1 ) . However , this increase in QTLs can be partially explained by the fact that txrevise detected multiple associations for ~24% of the genes while the full-length tuQTL analysis was limited to single lead association per gene ( Figure 2—figure supplement 2 , Figure 2—figure supplement 3 ) . Some of these additional QTLs are likely to represent independent causal variants , such as the three independent tuQTLs detected for the IRF5 gene ( Figure 2—figure supplement 4 ) while others could be explained by technical biases such as large gene expression QTL ( eQTL ) effects ( Figure 2—figure supplement 5 ) or positional biases in the RNA-seq data ( Figure 2—figure supplement 6 ) . Alternatively , additional associations could also be caused by transcriptional coupling where promoter or 3ʹ end choice directly influences the splicing of an internal exon or vice versa ( Anvar et al . , 2018; Bentley , 2014 ) . Different quantification methods may be biased towards discovering events with specific genomic properties , which is not captured by the number of QTLs detected . To address this , we quantified how often the lead QTL variants ( FDR < 0 . 01 ) from different methods were in high linkage disequilibrium ( LD ) ( r2 >0 . 8 ) with each other ( see Materials and methods ) . Consistent with previous reports that tuQTLs are largely independent from eQTLs ( Li et al . , 2016 ) , we found that only 11–25% of the lead variants detected at the read count level replicated at the transcript level ( r2 >0 . 8 , irrespective of the replication p-value ) , independent of which quantification method was used ( Figure 2B ) . In contrast , ~50% of the Leafcutter QTLs were also detected by txrevise or full-length transcript usage approaches . Similarly , the tuQTLs detected by txrevise and full-length transcript usage quantification were in high LD more than 60% of the time ( Figure 2B ) . Finally , we found that while 44% of the txrevise internal exon QTLs were in high LD with Leafcutter QTLs , this decreased to ~20% for promoter and 3ʹ end QTLs ( Figure 2—figure supplement 1 ) , suggesting that Leafcutter is less suited to capture those events . Thus , different quantification approaches appear to capture complementary sets of genetic associations . To characterise the genetic associations detected by different quantification methods , we compared the relative enrichments of the identified QTLs across multiple genomic annotations . We constructed genomic tracks for eight annotations: open chromatin measured by ATAC-seq ( Alasoo et al . , 2018 ) , promoter flanking regions ( −2000 bp to +200 bp ) , 5ʹ UTRs , coding sequence ( CDS ) , introns , 3ʹ UTRs , polyadenylation sites ( Gruber et al . , 2016 ) , and eCLIP-binding sites for RNA-binding proteins involved in splicing regulation ( splicing factors ) ( Van Nostrand et al . , 2017 ) . We then used the hierarchical model implemented in fgwas ( Pickrell , 2014 ) to estimate the enrichment of each genomic annotation among the QTLs detected by each quantification method . Consistent with the limited overlap between eQTLs and tuQTLs ( Figure 2B ) , we found that eQTLs were strongly enriched in sites of open chromatin ( Figure 2C; log enrichment of 3 . 31 , 95% CI [3 . 15 , 3 . 47] ) , whereas all transcript-level QTLs were enriched at the binding sites of splicing factors detected by eCLIP ( Figure 2C , mean log enrichment of 2 . 29 ) . Importantly , when all txrevise tuQTLs were pooled , the enrichment patterns were broadly similar to tuQTLs detected by full-length Ensembl transcripts ( Figure 2C ) . This suggests that txrevise events and full-length transcripts capture similar genetic associations but txrevise facilitates more accurate identification of the underlying transcriptional event ( i . e . promoter , internal exon or 3ʹ end usage ) ( Figure 2B ) . Finally , compared to Leafcutter , full-length transcript usage and txrevise QTLs were both more strongly enriched at 3ʹ UTRs ( Figure 2C , mean log enrichment of 1 . 85 ) , suggesting that they capture changes in 3ʹ UTR length that do not manifest at the level of junction reads and are thus missed by Leafcutter . To compare different types of transcriptional events , we repeated the fgwas analysis on the promoter , internal exon and 3ʹ end QTLs detected by txrevise as well as Leafcutter splicing QTLs . We found that Leafcutter and internal exon QTLs showed broadly similar enrichment patterns , with a strong enrichment at the binding sites of splicing factors ( Figure 2D , mean log enrichment of 2 . 53 ) . In contrast , promoter and 3ʹ end usage QTLs were specifically enriched at promoters ( Figure 2D; log enrichment of 2 . 76 , 95% CI [2 . 59 , 2 . 95] ) and 3ʹ UTRs ( Figure 2D; log enrichment of 3 . 60 , 95% CI [3 . 43 , 3 . 76] ) , respectively ( Figure 2D ) , and showed only a modest enrichment at the binding sites of splicing factors ( Figure 2D; mean log enrichment of 1 . 17 ) . Compared to other events , promoter usage QTLs were relatively more enriched in open chromatin regions ( log enrichment of 1 . 58 , 95% CI [1 . 42 , 1 . 74] ) . Thus , promoter usage , splicing and 3ʹ end usage appear to be regulated by largely independent sets of genetic variants enriched in distinct genomic regions . Motivated by the enrichment of promoter usage QTLs in open chromatin regions ( Figure 2D ) , we analysed chromatin accessibility QTLs that we previously identified in a subset of 41 individuals of the same study ( Alasoo et al . , 2018 ) . We wanted to determine how often changes in promoter usage also manifest at the level of promoter accessibility . We found that 124/786 ( 15 . 8% ) of the promoter usage QTLs were in high LD with at least one chromatin accessibility QTL ( r2 >0 . 9 ) compared to 10 . 2% of the internal exon and 3ʹ end usage QTLs ( Fisher’s exact test p-value=3 . 87×10−5 ) . These overlaps could correspond to both distal regulatory elements affecting promoter usage or direct changes in local promoter accessibility . To focus on local promoter accessibility , we further required the center of the accessible region to be no farther than 1000 bp from the closest promoter of the gene , leaving 46/786 ( 5 . 8% ) promoter usage QTLs with a putative coordinated effect on promoter accessibility . One such example affecting promoter usage and promoter accessibility of the HDLBP gene is highlighted in Figure 2—figure supplement 7 . However , larger studies with increased statistical power are needed to characterise the true extent of coordination between promoter accessibility and promoter usage . To assess the relevance of different QTLs for interpreting complex trait associations , we performed statistical colocalisation analysis with GWAS summary statistics for 33 immune-mediated and metabolic traits and diseases ( see Materials and methods ) . We found that 47 of 138 colocalised QTLs influenced total gene expression level ( Figure 3A ) ( PP3+PP4 >0 . 8 , PP4/PP3 >9; PP3 , posterior probability of a model with two distinct causal variants; PP4 , posterior probability of a model with one common causal variant ) . In contrast , the remaining 91 colocalised QTLs were associated with at least one of the transcript-level phenotypes ( full-length transcript usage , txrevise or Leafcutter ) but not with total gene expression ( Figure 3A ) . Similarly , 44 of 91 transcript-level colocalisations were detected only by a single transcript quantification approach ( Figure 3A ) . An important caveat of this analysis is that it does not directly test if the colocalisations are specific to one quantification method or simply missed by others because of limited power . Thus , our estimates of method-specificity are likely to be inflated . Finally , to quantify the relative contribution of promoter usage , splicing and 3ʹ end usage to complex traits , we stratified the txrevise colocalisations by transcriptional event type . We found that 44 of 77 colocalised QTLs influenced internal exons and the rest regulated promoters and 3ʹ ends ( Figure 3B ) . We were able to replicate known associations between splicing of exon two in CD33 and Alzheimer’s disease ( Figure 3—figure supplement 1 ) ( Malik et al . , 2013 ) and splicing of exon 13 in HMGCR and LDL cholesterol ( Figure 3—figure supplement 2 ) ( Burkhardt et al . , 2008 ) . Importantly , while half of the promoter and internal exon colocalisations were also detected by Leafcutter , only 1/10 3ʹ end events were captured by Leafcutter , probably because these are less likely to manifest at the level of junction reads ( Figure 3B ) . Next , we explored how the genetic effects of eQTLs and tuQTLs varied in response to stimuli . To define response QTLs , we started with QTLs detected ( FDR < 10% ) in each of the four simulated conditions ( I , S , I + S and acLDL ) and used an interaction test to identify cases where the QTL effect size was significantly different between the simulated and corresponding naive condition ( FDR < 10% ) . To exclude small but significant differences in effect size , we used a linear mixed model to identify QTLs where the interaction term explained more than 50% of the total genetic variance in the data ( see Materials and methods ) . Although the fraction of QTLs that were response QTLs varied greatly between conditions ( Figure 4A ) and correlated with the number of differentially expressed genes ( Figure 4—figure supplement 1 ) as previously reported ( Kim-Hellmuth et al . , 2017 ) , we found that the fraction of response tuQTLs was relatively consistent between the four quantification methods ( Figure 4A ) . While previous reports have highlighted that eQTLs are more condition-specific than tuQTLs ( Nédélec et al . , 2016 ) , we found no clear pattern in our data with stronger stimuli ( S and I + S ) showing larger fraction of condition-specific eQTLs , and weaker stimuli ( I , acLDL ) showing smaller fraction of response eQTLs ( Figure 4A ) compared to tuQTLs . However , when we focussed on the transcriptional events detected by txrevise , we found that promoter usage QTLs were 50% more likely to be response QTLs than tuQTLs regulating either internal exons or 3ʹ ends ( Figure 4B ) ( Fisher’s exact test combined p-value=2 . 79×10−6 ) . Finally , we assessed the condition-specificity of QTLs that colocalised with complex trait loci . We found that , on average , 12% of the GWAS colocalisations corresponded to response QTLs ( Figure 4C ) . One example is an IFNɣ-specific promoter usage QTL for the CD40 gene that colocalises with a GWAS signal for rheumatoid arthritis ( Okada et al . , 2014 ) . The alternative C allele of the rs4239702 variant is associated with increased usage of the transcript with the short 5ʹ UTR ( Figure 4E , F ) . This tuQTL was also visible at the absolute expression level of the two alternative promoters ( Figure 4—figure supplement 2 ) , but was missed by Leafcutter , because there is no change in junction reads . Although the variant was not significantly associated with total gene expression level ( Figure 4—figure supplement 2 ) , the two promoters contain the same start codon . As a result , the likely functional consequence of the CD40 tuQTL is modulation of protein abundance . Although the same tuQTL was also detected at the full-length transcript usage level , the affected transcripts also differ from each other by alternatively spliced exon 6 , making it challenging to interpret the result ( Figure 4E ) . The preferential upregulation of the transcript with the short 5ʹ UTR after exposure to an inflammatory stimulus is also supported by FANTOM5 capped analysis of gene expression ( CAGE ) data from primary macrophages ( Figure 4—figure supplement 3 ) ( Baillie et al . , 2017 ) .
We have performed a comprehensive analysis of the genetic determinants of transcript usage in human iPSC-derived macrophages exposed to four different stimuli . Our approach to stratify transcripts into individual events greatly improved the interpretability of molecular mechanisms underlying tuQTLs . Consequently , we were able to discover that 55% of the transcript-level associations affected promoter or 3ʹ end usage and these variants were enriched in markedly different genomic features relative to canonical splicing QTLs . We also found that promoter usage QTLs were 50% more likely to be condition-specific than other transcriptional events and often colocalised with GWAS hits for complex traits . Thus , event-level analysis might be preferable over transcript-level analysis when the aim is to identify specific transcriptional changes underlying genetic associations . We were able to link 6% of the promoter usage QTLs to coordinated changes in promoter accessibility . A likely reason for such a small overlap is limited statistical power in our chromatin accessibility dataset that contained only 41 individuals , leading us to miss many true effects on promoter accessibility . Alternatively , as other studies have suggested , promoter accessibility might not be an accurate proxy of activity and may merely be a prerequisite for transcription to take place ( Pliner et al . , 2018 ) , but demonstrating this would require better powered datasets to confidently demonstrate lack of effect on promoter accessibility . There is a great potential to study this further in larger datasets that have profiled gene expression , chromatin accessibility or histone modifications in hundreds of individuals ( Chen et al . , 2016; Kumasaka et al . , 2019 ) . Choosing the optimal quantification method for RNA-seq data is a challenging problem . The field of detecting and quantifying individual transcriptional changes from RNA-seq data has been developing rapidly . One of the most successful approaches has been the use of reads spanning exon-exon junctions to detect differential usage of individual exons within genes . In our study , we used Leafcutter to perform junction-level analysis , but other options are available such as JUM ( Wang and Rio , 2018 ) or MAJIQ ( Vaquero-Garcia et al . , 2016 ) . A key advantage of junction-level analysis is that it can discover novel exon-exon junctions and is thus well-suited for characterising rare or unannotated splicing events . On the other hand , changes in 5ʹ and 3ʹ UTR length are not captured by junction-level methods , because these events do not overlap exon-exon junctions . Changes in UTR length can only be detected by methods that consider all reads originating from alternative transcript ends such as MISO ( Katz et al . , 2010 ) or txrevise proposed here . MISO provides more fine-grained events that can differentiate between various types of splicing events . Txrevise , on the other hand , provides a more comprehensive catalog of promoter and 3ʹ end events that can be continuously updated as reference annotations improve . A promising alternative to both of these methods is Whippet , which quantifies transcriptional events by aligning reads directly to the splice graph of the gene ( Sterne-Weiler et al . , 2017 ) . Thus , no single approach is consistently superior to others and characterizing the full spectrum of transcriptional consequences of genetic variation requires a combination of analytical strategies ( Odhams et al . , 2017; Ongen and Dermitzakis , 2015 ) . An important limitation of txrevise is that it is only able to quantify splicing events present in reference transcript databases . However , our approach can easily be extended by incorporating additional annotations such experimentally determined promoters from the FANTOM5 ( Forrest et al . , 2014 ) projects or alternative polyadenylation sites from the PolyAsite database ( Gruber et al . , 2016 ) , as is done by QAPA ( Ha et al . , 2018 ) . Another option might be to incorporate novel transcripts identified by transcript assembly methods such as StringTie ( Pertea et al . , 2015 ) into existing annotation databases . Nevertheless , since txrevise relies on Salmon for event-level quantification , it is still susceptible to some of the same limitations as full-length transcript quantification . Even though event-level analysis reduces the problem slightly , a positive transcript expression estimate does not guarantee that any specific exon is actually present in the transcript , especially if the transcript annotations are incomplete ( Figure 1B ) ( Soneson et al . , 2018 ) . Secondly , large eQTL effects and positional biases in the RNA-seq data can occasionally lead to spurious changes in transcript usage ( Figure 2—figure supplements 5 and 6 ) . Therefore , it is important to visually confirm candidate transcriptional events using either base-level read coverage plots ( Alasoo , 2017 ) or Sashimi plots ( Katz et al . , 2015 ) before embarking on follow-up experiments . A key aim of QTL mapping studies is to elucidate the molecular mechanisms underlying complex trait associations . In our analysis , we found that over 50% of the genetic effects that colocalise with complex traits regulated transcript usage and did not manifest at the total gene expression level . Moreover , 42% of the transcript-level colocalisations affected promoter or 3ʹ end usage instead of splicing of internal exons . Importantly , no single quantification method was able to capture the full range of genetic effects , confirming that different quantification approaches often identify complementary sets of QTLs ( Odhams et al . , 2017; Ongen and Dermitzakis , 2015 ) . Thus , there is great potential to discover additional disease associations by re-analysing large published RNA-seq datasets such as GTEx ( Battle et al . , 2017 ) with state-of-the-art quantification methods .
All RNA-seq libraries from the acLDL study were constructed manually using poly-A selection and the Illumina TruSeq stranded library preparation kit . The TruSeq libraries were quantified using Bioanalyzer and manually pooled for sequencing . The samples were sequenced on Illumina HiSeq 2000 using V4 chemistry and multiplexed at six samples/lane . The control and acLDL stimulated RNA samples from a single donor were always sequenced in the same experimental batch . Sample metadata is presented in Supplementary file 2 . RNA-seq reads from both studies were aligned to the GRCh38 reference genome and Ensembl 87 transcript annotations using STAR v2 . 4 . 0j ( Dobin et al . , 2013 ) . Subsequently , VerifyBamID v1 . 1 . 2 ( Jun et al . , 2012 ) was used to detect and correct any sample swaps between donors . Two samples from one donor ( HPSI0513i-xegx_2 ) were excluded from downstream analysis , because they appeared to be outliers on the principal component analysis ( PCA ) plot of the samples . We used four alternative strategies to quantify transcription from RNA-seq data: ( i ) gene-level read count quantified with featureCounts ( Liao et al . , 2014 ) , ( ii ) full-length transcript usage quantified with Salmon ( Patro et al . , 2017 ) ( Figure 1C ) , ( iii ) promoter , internal exon and 3ʹ end usage quantified with txrevise , and ( iv ) exon-exon junction usage quantified with Leafcutter ( Li et al . , 2018 ) . To compare the QTLs detected by different quantification methods , we estimated the fraction of QTL lead variants detected by each method that were replicated by the other methods . Since read count and full-length transcript usage analysis were performed at the gene level , we decided to perform the replication analysis at the gene level as well . Because txrevise and Leafcutter quantified multiple events per gene and sometimes detected multiple independent QTLs ( Figure 2—figure supplement 2 ) , we picked the lead variant with the smallest p-value across all of the events quantified for a given gene as the gene-level lead variant . For each pairwise comparison of quantification methods , we first identified all lead variant-gene pairs with FDR < 0 . 01 detected by the query method . Subsequently , we extracted the lead variants for the same genes detected by the replication method and estimated the fraction of those that were in high LD ( r2 >0 . 8 ) with each other . We then repeated this analysis for all pairs of quantification methods . Note that this measure is not necessarily symmetric between the quantification methods and also depends on the statistical power of each method . Since Leafcutter had lower statistical power than other methods on our dataset , it also replicated smaller fraction of QTLs detected by the other methods . In contrast , ~50% of the Leafcutter QTLs were replicated by txrevise and full-length transcript usage ( Figure 2B ) . We acknowledge that our definition of replication ignores the direction of the effect of the genetic variant on gene expression or transcript usage . For example , if Leafcutter detects a genetic variant that is associated with increased inclusion of an exon in a gene and txrevise detects that the same variant is associated with decreased inclusion of the same exon in the same gene , we would still consider it to be a ‘successful’ replication . However , in practice it is difficult to map Leafcutter events to specific Ensembl transcripts or txrevise events , especially if Leafcutter includes novel exon-exon junctions not present in the Ensembl database . Furthermore , comparing the effect size direction between eQTLs and tuQTLs is not possible , because any variant that is associated with increased usage of one transcript is by definition also associated with decreased usage of some other transcripts of the same gene . To identify response QTLs , we started with QTLs detected ( FDR < 10% ) in each of the four stimulated conditions ( I , S , I + S and acLDL ) and used an interaction test to identify cases where the QTL effect size was significantly different between one of the stimulated and corresponding naive condition ( FDR < 10% ) . We performed this test separately for each of the four stimulated condition ( I , S , I + S and AcLDL ) . Furthermore , to take advantage of our profiling of gene expression in overlapping set of donors in the stimulated and naive conditions , we also included the cell line as a random effect and fitted a linear mixed model using the lme4 ( Bates et al . , 2015 ) package . Specifically , for each phenotype and lead variant pair , we used the anova function to compare the following two models: H0: phenotype ~genotype + condition + ( 1|donor ) H1: phenotype ~genotype + condition+condition:genotype + ( 1|donor ) where ( 1|donor ) denotes the donor-specific random effect . We obtained the p-value of rejecting the null hypothesis and used the p . adjust function to identify phenotype and lead variant pairs that were significant at 10% Benjamini-Hochberg FDR . For some QTLs , we noticed that although the interaction test p-value was significant , the difference in the effect size between the two conditions was very small . To identify response QTLs with large effect size differences between naive and stimulated conditions , we turned to variance component analysis . Specifically , for the same phenotype and lead variant pairs tested above , we also fitted the following linear mixed model: phenotype ~ ( 1|genotype ) + ( 1|condition ) + ( 1|condition:genotype ) where genotype , condition and the interaction between the two were all fitted as random effects . We then quantified the variance explained by each of the three components using the VarCorr function form the lme4 package . Finally , we calculated the variance explained by the interactions term relative to the total genetic variance: σ2relative = σ2interaction / ( σ2interaction+σ2genotype ) We defined response QTLs as those with FDR < 10% from the interaction test and σ2relative > 0 . 5 from the variance component analysis . Although fitting genotype as a random effect in this way is suboptimal because it ignores the expected linear relationship between the alternative allele dosage and phenotype , we empirically found that filtering both on the p-value of the interaction test as well as σ2relative was effective at identifying QTLs with large effect size differences between conditions . The Snakemake ( Köster and Rahmann , 2012 ) files used for gene and transcript expression quantification , QTL mapping and colocalisaton are available from the project’s GitHub repository ( https://github . com/kauralasoo/macrophage-tuQTLs; copy archived at https://github . com/elifesciences-publications/macrophage-tuQTLs; Alasoo , 2018b ) . The same repository also contains R scripts that were used for all data analysis and figures . The txrevise R package is available from GitHub ( https://github . com/kauralasoo/txrevise; copy archived at https://github . com/elifesciences-publications/txrevise; Alasoo , 2018a ) and wiggleplotr R package that was used to make transcript read coverage plots is available from Bioconductor ( http://bioconductor . org/packages/wiggleplotr/ ) . RNA-seq data from the acLDL stimulation study is available from ENA ( PRJEB20734 ) and EGA ( EGAS00001000876 ) . RNA-seq data from the IFNɣ+Salmonella study is available from ENA ( PRJEB18997 ) and EGA ( EGAS00001002236 ) . The imputed genotype data for HipSci cell lines is available from ENA ( PRJEB11749 ) and EGA ( EGAD00010000773 ) . Processed data and QTL summary statistics are available from Zenodo: https://zenodo . org/communities/macrophage-tuqtls/ . | Genes contain all instructions needed to build an organism in form of DNA . Humans share around 99 . 5% of DNA , but it is the remaining 0 . 5% that contain the small genetic variations that make us unique . Subtle differences in genes can , for example , influence the color of our hair or eyes . To build gene products , such as proteins , DNA first needs to be transcribed into RNA . Some genetic variants can affect how a gene is transcribed into an RNA molecule , for example by making it be transcribed too much or too little , which can lead to diseases . These variants can also influence where the transcription begins through a process called promoter usage . This can lead to shorter or longer RNAs , which can have different biological impacts . With current research methods it is difficult to detect changes in the latter kind of alteration . As a result , it is harder to distinguish these from other types of changes . Now , Alasoo et al . wanted to find out what proportion of genetic variants that alter traits influence promoter usage , compared to other changes . To do so , a new computational method was developed to directly measure how genetic variants influence different parts of the RNA , such as promoters , middle sections and ends . The method was then applied to datasets of human immune cells . The experiments revealed that genetic variants often influence promoter usage . Many of the effects could only be found when cells are exposed to external stimuli , such as bacteria . The results highlight that to discover genes responsible for human traits and disease we need to consider all the possible ways genetic differences between individuals could alter the gene products . Large published datasets could be reanalyzed using this method to identify new genes that could be implicated in human health and disease , potentially leading to new treatment options in future . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"computational",
"and",
"systems",
"biology",
"genetics",
"and",
"genomics"
] | 2019 | Genetic effects on promoter usage are highly context-specific and contribute to complex traits |
BST2/tetherin , an antiviral restriction factor , inhibits the release of enveloped viruses from the cell surface . Human immunodeficiency virus-1 ( HIV-1 ) antagonizes BST2 through viral protein u ( Vpu ) , which downregulates BST2 from the cell surface . We report the crystal structure of a protein complex containing Vpu and BST2 cytoplasmic domains and the core of the clathrin adaptor protein complex 1 ( AP1 ) . This , together with our biochemical and functional validations , reveals how Vpu hijacks the AP1-dependent membrane trafficking pathways to mistraffick BST2 . Vpu mimics a canonical acidic dileucine-sorting motif to bind AP1 in the cytosol , while simultaneously interacting with BST2 in the membrane . These interactions enable Vpu to build on an intrinsic interaction between BST2 and AP1 , presumably causing the observed retention of BST2 in juxtanuclear endosomes and stimulating its degradation in lysosomes . The ability of Vpu to hijack AP-dependent trafficking pathways suggests a potential common theme for Vpu-mediated downregulation of host proteins .
The interferon-inducible restriction factor BST2 ( also named tetherin , CD317 and HM1 . 24 ) presents a potent innate immune restriction to many enveloped viruses ( Neil et al . , 2008; Van Damme et al . , 2008; Jouvenet et al . , 2009; Evans et al . , 2010; Arias et al . , 2012 ) . BST2 has a short cytoplasmic tail at the N-terminus followed by a single transmembrane ( TM ) helix , a long coiled-coil extracellular domain , and a C-terminal glycosyl phosphatidylinositol ( GPI ) anchor ( Kupzig et al . , 2003 ) . The availability of two membrane anchors connected by a long coiled-coil allows BST2 to efficiently inhibit viral transmission by tethering newly formed virions at the infected cell surface and preventing their release to the surrounding environment ( Hinz et al . , 2010; Schubert et al . , 2010; Yang et al . , 2010 ) . The short intracellular , N-terminal tail of BST2 has been implicated in the natural trafficking of this antiviral protein ( Rollason et al . , 2007; Masuyama et al . , 2009 ) . HIV-1 overcomes restriction by BST2 using the viral protein Vpu ( Neil et al . , 2008; Van Damme et al . , 2008 ) . The mechanism by which Vpu antagonizes BST2 appears to be multifaceted , involving both degradation and mistrafficking within the endosomal system . Vpu ( about 81 amino acids in most viral isolates ) is a transmembrane protein consisting of an N-terminal transmembrane α-helix , followed by a cytoplasmic domain that is likely to be flexible ( Cohen et al . , 1988; Strebel et al . , 1988 ) . Vpu associates with BST2 through an anti-parallel interaction between the transmembrane domains of each protein ( Mangeat et al . , 2009; Skasko et al . , 2012; McNatt et al . , 2009; Vigan and Neil , 2011; Kobayashi et al . , 2011; Vigan and Neil , 2010 ) . This interaction is species-specific and essential for the antagonism of BST2 by Vpu ( McNatt et al . , 2009; Skasko et al . , 2012 ) . Part of Vpu's activity against BST2 can be explained by the viral hijacking of the β-TrCP-associated ubiquitin–proteasome degradation pathway ( Van Damme et al . , 2008; Douglas et al . , 2009; Goffinet et al . , 2009; Iwabu et al . , 2009; Mangeat et al . , 2009; Mitchell et al . , 2009 ) . A component of the ESCRT-0 machinery , HRS , has also been suggested to recognize ubiquitinated BST2 and target it for lysosomal degradation ( Janvier et al . , 2011 ) . However , these degradation pathways are only partially responsible for the antagonism of BST2 by Vpu ( Van Damme et al . , 2008; Douglas et al . , 2009; Iwabu et al . , 2009; Mangeat et al . , 2009; Mitchell et al . , 2009 ) . Efficient BST2 downregulation from the cell surface can occur in the absence of BST2 degradation ( Dube et al . , 2010; Goffinet et al . , 2010; Tervo et al . , 2011 ) . Moreover , Vpu induces the mistrafficking of BST2 ( Douglas et al . , 2009; Dube et al . , 2010; Hauser et al . , 2010; Lau et al . , 2011; Schmidt et al . , 2011 ) , causing the accumulation of BST2 at the trans-Golgi network ( TGN ) . Both recycled and newly synthesized BST2 are retained at the TGN , blocking the resupply of BST2 to the plasma membrane and eventually leading to its depletion at the cell surface ( Dube et al . , 2011; Lau et al . , 2011; Schmidt et al . , 2011 ) . Moreover , mutations in the juxta-membrane hinge region of Vpu that interfere with the localization of Vpu to the TGN impair the antagonism of BST2 ( Dube et al . , 2009; Vigan and Neil , 2011 ) . Clathrin-dependent trafficking pathways have been suggested to be involved in the Vpu-mediated mistrafficking of BST2 ( Kueck and Neil , 2012; Lau et al . , 2011; Mitchell et al . , 2009; Ruiz et al . , 2008 ) . Such pathways regulate the trafficking of cellular membrane proteins by selectively packaging these membrane cargos into clathrin-coated vesicles ( CCV ) ( Bonifacino and Traub , 2003; Canagarajah et al . , 2013; Traub , 2009 ) . The clathrin adaptor protein ( AP ) complexes mediate this cargo selection . Two canonical sorting motifs in the cytoplasmic domains of the membrane cargo proteins are recognized by the AP complexes: a tyrosine-based Yxxϕ motif ( ϕ represents a large hydrophobic residue; x for any amino acid ) and an acidic dileucine motif , [E/D]xxxL[L/I] . Five AP complexes exist in the cell and each is responsible for trafficking by distinct routes . For example , AP1 traffics cargo between the TGN and sorting endosomes , while AP2 selects cargo for transport between the plasma membrane and early endosomes ( Canagarajah et al . , 2013 ) . In the absence of Vpu , the natural trafficking of endogenous BST2 depends on the clathrin-associated pathways , and the involvement of both AP1 and AP2 has been suggested ( Rollason et al . , 2007; Masuyama et al . , 2009 ) . An unusual double-tyrosine motif , YxY , in the BST2 cytoplasmic domain ( BST2CD ) is critical for this natural trafficking ( Rollason et al . , 2007; Masuyama et al . , 2009 ) . In Vpu , a putative clathrin-sorting motif , ExxxLV ( ELV ) , located in the membrane-distal half of the protein's cytoplasmic domain ( VpuCD ) was shown to be important for BST2 antagonism ( Kueck and Neil , 2012 , McNatt et al . , 2013 ) . Furthermore , Vpu-induced virion release and removal of BST2 from the cell surface are inhibited by a dominant negative mutant of AP180 , a protein required for the assembly of the CCV at the lipid membrane ( Kueck and Neil , 2012; Lau et al . , 2011 ) . However , the critical question remains as to whether any AP complexes , and if so which , are hijacked by Vpu for the downregulation of BST2 . To understand the mechanisms of Vpu-mediated mistrafficking of BST2 , we examined the interaction of the cytoplasmic domains of these proteins with recombinant AP complexes and their subunits . Moreover , we determined the crystal structure of a three-component complex containing AP1 , the cytoplasmic domain of BST2 ( BST2CD ) and the cytoplasmic domain of Vpu ( VpuCD ) . The structure shows that Vpu mimics a membrane cargo by occupying the acidic dileucine-binding site of AP1 , while BST2 is bound at the tyrosine-binding site of AP1 . This , together with biochemical and functional evidence , suggests that HIV-1 Vpu is a virally encoded modulator of clathrin-dependent trafficking pathways and supports the involvement of AP1 in the Vpu-mediated mistrafficking of BST2 .
To identify the AP complexes involved in the trafficking of BST2 , we investigated their ability to bind BST2 . BST2CD contains a putative tyrosine motif that is believed to bind to the μ subunits of AP complexes , and intracellular interactions between BST2 and the μ subunits of either AP1 or AP2 have been reported ( Rollason et al . , 2007; Masuyama et al . , 2009 ) . We tested interactions between BST2CD and the C-terminal domains ( CTD ) of the μ subunits of three AP complexes ( AP1 , AP2 , and AP3 ) . The purified μ1-CTD of AP1 bound BST2CD as the two co-migrated as a higher molecular weight complex on a size exclusion column ( Figure 1A ) . In contrast , no such binding was observed for either the μ2-CTD ( Figure 1B ) or μ3-CTD ( Figure 1C ) . These observations were further confirmed by a yeast two-hybrid ( Y2H ) assay ( Figure 1D ) . Consistent with the in vitro binding assay , BST2CD exhibited binding to μ1 , but not μ2 or μ3 , in the Y2H assay . As controls , the TGN38 peptide containing a canonical YxxΦ motif displayed binding to all three μ subunits , and a mutant μ1 subunit ( μ1 D174A ) with a disrupted binding pocket for tyrosine-based motifs bound neither TGN38 nor the BST2CD . Furthermore , this binding was abolished by the double alanine mutation of the YxY motif ( Y6/8A ) in BST2CD . These results indicate that BST2CD not only contains a tyrosine-based motif to allow for AP binding , but also a specificity determinant to select specifically for μ1 . 10 . 7554/eLife . 02362 . 003Figure 1 . BST2 interacts with µ1 but not µ2 or µ3 . ( A–C ) Size exclusion chromatograms and SDS PAGE analysis of purified MBP-BST2CD ( purple curve ) , MBP-µ CTD ( green curve ) , and their mixture ( red curve ) . A MBP-µ1 CTD and MBP-BST2CD complex is formed in ( A ) , as indicated by the shift ( P1 ) of its elution volume from those of the individual components ( P2 and P3 ) . No complex formed between BST2 and µ2 ( B ) or µ3 ( C ) . ( D ) Yeast 2-hybrid assays showing binding of BST2CD to µ1 , but not µ2 or µ3 . Growth on -Leu/-Trp/-His + 3 mM 3AT plates indicates interaction . Growth on -Leu/-Trp serves as a growth control . ‘0 . 5x’ indicates plating of one-half the amount of yeast cells relative to other spots . The BST2CD-µ1 interaction is abolished by the tyrosine motif mutation Y6/8A in BST2CD or the tyrosine-binding pocket mutation D174A in µ1 . DOI: http://dx . doi . org/10 . 7554/eLife . 02362 . 00310 . 7554/eLife . 02362 . 004Figure 1—figure supplement 1 . BST2 does not interact with the α appendage domain of AP2 . Size exclusion chromatograms and SDS PAGE analysis of purified MBP-BST2CD ( purple curve ) , MBP-α appendage ( green curve ) , and their mixture ( red curve ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02362 . 004 Our results support the involvement of AP1 in trafficking of BST2 and are consistent with a previous observation that BST2 binds to the μ1 subunit of AP1 , but not μ2 of AP2 ( Masuyama et al . , 2009 ) . In the absence of Vpu , BST2 has been suggested to localize to the TGN through clathrin-dependent trafficking ( Kupzig et al . , 2003; Rollason et al . , 2007; Masuyama et al . , 2009 ) , and the BST2–AP1 interaction may be responsible for this localization . In contrast , an involvement of AP2 in BST2 trafficking has been suggested to occur through its α appendage domain . However , this interaction was not detected in our in vitro binding test using size exclusion chromatography ( Figure 1—figure supplement 1 ) . A specific site on AP2 for BST2 binding remains to be clearly elucidated . Given the native affinity of BST2 for AP1 and the retention of BST2 in the TGN by Vpu , we sought to define the potential interactions between Vpu and AP1 . As the functionally active ELV motif in VpuCD is a putative acidic dileucine sorting signal , we hypothesized that Vpu binds AP1 at the acidic dileucine-binding site located at the γ and σ1 subunits . To test this , we created a truncated AP1 core with the μ1-CTD removed . Such a truncation construct ( AP1t ) mimics the open conformation of AP1 , exposing the dileucine binding site on AP1 for interaction with cargo ( Jackson et al . , 2010 ) . We co-expressed VpuCD with casein kinase II ( CKII ) to phosphorylate VpuCD at S52 and S56 to mimic its cellular state ( Figure 2—figure supplement 1 ) . Unless mentioned otherwise , doubly phosphorylated Vpu constructs were used in all in vitro studies . Indeed , MBP-VpuCD co-migrated with the μ1-truncated AP1 as a complex on a size exclusion column ( Figure 2A ) . Importantly , the binding was abolished by the alanine-mutation of the ELV motif , signifying the crucial role of the ELV motif in the interaction between Vpu and AP1 ( Figure 2B ) . We further tested the binding between VpuCD and μ1-CTD and observed complex formation using size exclusion chromatography ( Figure 2C ) . These interactions demonstrate that Vpu has evolved the ability to associate with multiple subunits of AP1 , potentially allowing it to modulate the cellular trafficking machinery to target host proteins such as BST2 . 10 . 7554/eLife . 02362 . 005Figure 2 . Vpu interacts with multiple subunits of AP1 and AP2 , but not μ3 of AP3 . ( A and B ) Size exclusion chromatograms and SDS PAGE analysis of purified MBP-VpuCD constructs ( cyan curve and bottom SDS gel ) and its mixture with μ1-truncated AP1 ( AP1t ) ( red curve and top SDS gel ) . ( A ) The elution profile of MBP-VpuCD on the size exclusion column is altered in the presence of AP1t ( marked by the red arrow ) , indicating complex formation . ( B ) The alanine mutation of the ELV motif in VpuCD ( MBP-VpuCD ELV/AAA ) abolishes the interaction with AP1t , as the elution profile of MBP-VpuCD ELV/AAA does not change with or without AP1t . ( C ) MBP-VpuCD binds to MBP-µ1 CTD . A shift occurs in the elution volume of the MBP-µ1 CTD and MBP-VpuCD mixture ( P1 ) relative to the individual components ( P2 and P3 ) . ( D ) Size exclusion chromatogram and SDS PAGE analysis showing the co-migration of MBP-VpuCD and α398-σ to a higher molecular weight region relative to the individual components . ( E ) MBP-VpuCD and MBP-μ2-CTD also form a co-migrating interaction complex . ( F ) No complex is formed between MBP-VpuCD and MBP-μ3-CTD as the mixture migrates in the same manner as the individual components . DOI: http://dx . doi . org/10 . 7554/eLife . 02362 . 00510 . 7554/eLife . 02362 . 006Figure 2—figure supplement 1 . Efficient and complete double phosphorylation of recombinant VpuCD . ( A ) SDS-PAGE gel , size-exclusion chromatogram and ( B ) mass spectrometry of purified phosphorylated VpuCD . The molecular weight measured by mass spectrometry confirms that VpuCD is doubly phosphorylated . DOI: http://dx . doi . org/10 . 7554/eLife . 02362 . 006 We next investigated the interaction between Vpu and subunits of AP2 and AP3 . We used size exclusion chromatography to test the binding between VpuCD and a truncated hemicomplex of AP2 , α ( 1–398 ) -σ , which contains the binding pocket for the acidic dileucine sorting motif; an interaction complex was observed ( Figure 2D ) . Furthermore , binding between VpuCD and the μ2-CTD of AP2 was also observed ( Figure 2E ) , while no such binding was observed between VpuCD and the μ3-CTD of AP3 ( Figure 2F ) . Altogether , these results suggest specific interactions between Vpu and AP1 and AP2 , which may allow the viral protein to hijack the associated trafficking pathways . However , because BST2CD specifically binds only to μ1 , but not to μ2 or μ3 ( Figure 1 ) , AP1 may play a more significant role than AP2 in the Vpu-mediated antagonism of BST2 . This notion is supported by multiple observations , with one exception ( Iwabu et al . , 2010 ) , that Vpu does not increase the rate of BST2 internalization from the cell surface ( Mitchell et al . , 2009; Dube et al . , 2010; Andrew et al . , 2011; Schmidt et al . , 2011 ) . The ability of Vpu to interact simultaneously with BST2 and AP1 suggests that Vpu may enhance a native but weak affinity between BST2 and AP1 to increase their binding and consequently retain BST2 in endosomes including the TGN and/or target it to lysosomes . We used a fusion of BST2CD and VpuCD to mimic in vitro the membrane-assisted binding that occurs in cells . BST2 and Vpu each have a transmembrane ( TM ) helix through which the two proteins associate . The C-terminus of BST2CD and the N-terminus of VpuCD are placed close to each other by the interacting TM helices ( Skasko et al . , 2012 ) . This facilitates the convenient design of a 10-amino acid fusion linker that mimics the restraints exerted by the TM helices and links the cytoplasmic domains in an appropriate spatial arrangement ( Figure 3A ) . The fusion protein exhibited strong binding to the GST-tagged AP1 in the pulldown assay ( Figure 3B ) . In contrast , although both VpuCD and BST2CD interact with AP1 subunits in our size exclusion chromatography assays ( Figure 1 , Figure 2 ) , their individual interactions with the full AP1 core complex were not observed under stringent pulldown conditions ( Figure 3B ) . This result suggests that an additive or cooperative effect occurs when BST2CD and VpuCD bind to AP1 . 10 . 7554/eLife . 02362 . 007Figure 3 . BST2CD-VpuCD fusion binds tightly to AP1 . ( A ) Schematic of the BST2–Vpu interaction at the lipid membrane ( top ) and the fusion protein used for in vitro studies ( bottom ) showing the location of important residues in both BST2 ( YxY ) and Vpu ( ELV ) . S52 and S56 are phosphorylation sites required for the binding of Vpu to β-TrCP . The regions of Vpu and BST2 used in the fusion are boxed in green . ( B ) GST pulldown assay using an AP1 with a GST-tagged γ subunit . GST bead-bound AP-1 captured MBP-BST2CD-VpuCD , but not MBP-VpuCD or MBP-BST2CD alone . SDS PAGE analysis shows the amount of protein in the load and eluted from the GST beads . DOI: http://dx . doi . org/10 . 7554/eLife . 02362 . 007 We tested the requirement of the important motifs in BST2 and Vpu for the binding of AP1 . We first used a GST-tagged AP1 core to pull down mutants of the BST2CD-VpuCD fusion ( Figure 4 ) . As expected , alanine mutation of the YxY motif ( Y6/8A ) greatly reduced the binding to AP1 . Additional mutation of the γ subunit R15E on AP1 that disrupts the acidic dileucine-motif binding site affected the binding further . To validate the in vitro observations made with the TM-free BST2CD-VpuCD fusion , we designed a TM-containing chimera and performed immunoprecipitation experiments using human cells ( Figure 4B , C ) . This chimera contains the N-terminal 66 residues of BST2 ( including the cytoplasmic domain , TM , and two of the three cysteines in the ectodomain involved in dimerization ) followed by a flexible linker ( GGGSx3 ) , a FLAG epitope tag , and the entirety of Vpu; we expected it to faithfully recapitulate the geometry of the two interacting molecules in cells ( Figure 4B ) . Consistent with the in vitro observations ( Figure 3B ) , binding between endogenous AP1 ( the µ1subunit ) and this chimera was detected , while no such binding was detected for either BST2 or Vpu alone ( Figure 4C ) . Although the alanine mutations of either Y6/8 or ELV did not substantially reduce the binding , the mutant combining both sets of mutations was unable to bind AP1 ( Figure 4C ) . 10 . 7554/eLife . 02362 . 008Figure 4 . BST2/Vpu/AP1 interaction involves the BST2 YxY sequence and the dileucine motif-binding pocket on γ/σ1 and is independent of Vpu phosphorylation . ( A ) GST affinity pulldown assay using an AP1 core complex with a GST-tagged γ subunit . Mutation of the acidic dileucine pocket on γ ( ‘AP1 R15E’ ) and the YxY motif on BST2 substantially reduced binding to AP1 . Phosphorylation or S52/56N mutation of Vpu has no effect on AP1 binding . ( B ) Schematic of the TM-containing BST2 and Vpu fusion chimera . The broken line indicates the first 20 residues of the native BST2 ectodomain followed by the fusion linker , the FLAG epitope tag , and the Vpu ectodomain . ( C ) HEK293T cells were transfected to express FLAG-tagged Vpu , BST2 , or the fusion chimera or related mutants . ‘no FLAG’ indicates mock transfection without any FLAG-tagged construct . Notably , Vpu is much more abundantly expressed than either BST2 or the BST2-Vpu chimeras , and the chimeras appear to be partly proteolyzed . Nonetheless , immunoprecipitation ( IP ) using anti-FLAG antibody shows endogenous µ1 only in the presence of the BST2-Vpu chimera . Mutation of both the BST2 YxY and Vpu ELV motifs is necessary to prevent co-IP of µ1 . DOI: http://dx . doi . org/10 . 7554/eLife . 02362 . 008 Our experiments show that the binding of Vpu to AP1 is independent of serine-phosphorylation ( Figure 4A ) . Phosphorylation of S52 and S56 in Vpu is critical for recruiting β-TrCP and its associated E3 ubiquitin ligase complex . The unphosphorylated fusion protein bound AP1 as tightly as the phosphorylated construct ( Figure 4A ) . Furthermore , the binding of AP1 was not affected by the double mutation , S52/56N , that destroys the phosphorylation sites ( Figure 4A ) . These results are consistent with the notion that the β-TrCP-dependent BST2-degradation by Vpu and the mistrafficking of BST2 by Vpu are governed by different determinants in the VpuCD . To understand the Vpu-enhanced binding of BST2 to AP1 , we determined the crystal structure of the BST2CD-VpuCD fusion in complex with the AP1 core at 3 . 0 Å resolution ( Figure 5A ) . The AP1 core adopts an activated , open conformation , with both of its cargo binding sites exposed for interaction with BST2CD-VpuCD . The AP1 in the current structure adopts an open conformation distinct from that observed previously for AP2 and AP1 ( Jackson et al . , 2010; Ren et al . , 2013 ) . BST2CD binds to the tyrosine motif-binding site on AP1 through critical interactions involving the YxY motif , while VpuCD occupies the acidic dileucine motif-binding site of AP1 through the ELV motif . Only a short region of VpuCD flanking the ELV motif is well ordered and successfully built in the structure . There is no direct interaction between BST2CD and VpuCD . Overall , the structure reveals how Vpu enhances the native interaction between BST2 and AP1 . By combining the viral protein's affinity for AP1 via the ELV motif and the tight transmembrane interaction with the host protein , Vpu appears to act as an adaptor to increase the affinity of AP1 for BST2 . 10 . 7554/eLife . 02362 . 009Figure 5 . Crystal structure of the BST2CD-VpuCD/AP1 complex . ( A ) The crystal structure of the BST2CD-VpuCD/AP1 complex . AP1 is colored by subunit ( β1 in gray , γ in orange , μ1 in green and σ1 in yellow ) . VpuCD ( cyan ) binds to the acidic dileucine binding pocket of γ/σ1 , and BST2CD ( magenta ) binds to the tyrosine-binding pocket in μ1 . ( B ) The difference Fourier map ( mFo-DFc at 3σ level , blue mesh ) of BST2CD ( magenta sticks ) binding to μ1 ( green surface ) . Important residues in BST2CD are labeled . ( C ) V11 partly fills the canonical Φ residue binding-site on µ1 . ( D ) Y6 and Y8 make extensive interactions to µ1 residues . Hydrogen bonds are indicated with dashed lines . ( E ) Yeast 2-hybrid assays showing binding of BST2 to µ1 . Growth on -Leu/-Trp/-His + 3 mM 3AT plates indicates interaction . Growth on -Leu/-Trp serves as a growth control . ‘0 . 2x’ indicates plating of one-fifth the amount of yeast cells relative to other spots . The BST2-µ1 interaction is abolished by the alanine mutations of BST2 Y6/8A , Y6A , Y8A , or V11A . ( F ) The difference Fourier map ( mFo-DFc at 3σ level , blue mesh ) of VpuCD ( cyan sticks ) binding to σ1 ( yellow surface ) and γ ( orange surface ) subunits of AP1 . Important VpuCD and γ residues are labeled . ( G ) Overlay of α/σ2 of AP-2 ( PDB ID 2JKT ) ( Kelly et al . , 2008 ) and γ/σ1 of AP1 shows that the Vpu ELV motif ( cyan ) binds in the same way as the canonical peptide ( gray ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02362 . 00910 . 7554/eLife . 02362 . 010Figure 5—figure supplement 1 . Structural incompatibility prevents the binding of BST2 to either μ2 of AP2 or μ3 of AP3 . ( A ) µ2 ( gray surface ) and ( B ) µ3 ( gray surface ) do not support binding to the YxY motif of BST2 . Note the steric hindrance for the Y6 side chain and the lack of stacking interaction as provided by Y384 of µ1 . The models were made by superimposing the coordinates of μ1 in the current structure with those of μ2 ( PDB ID 1BXX ) ( Owen and Evans , 1998 ) and μ3 ( PDB ID 4IKN ) ( Mardones et al . , 2013 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02362 . 010 BST2CD occupies the conserved tyrosine motif-binding site on the μ1 subunit of AP1 ( Figure 5B–D ) . However , the observed interface differs from the canonical tyrosine peptide ( YxxΦ ) binding in that the interaction is achieved with an unusual double-tyrosine motif ( YxYxxΦ ) ( Figure 5B , D ) . Y8 of BST2CD forms the canonical interactions with μ1 residues and inserts into the conserved tyrosine-binding pocket on μ1 . Unlike the canonical cases in which a leucine or isoleucine residue is found at the Y+3 position , V11 in BST2 only partly fills the corresponding hydrophobic pocket on μ1 ( Figure 5C ) . This pocket would be better satisfied by a larger hydrophobic residue . This observation is consistent with results from a previous combinatorial screen of tyrosine-based µ-binding sequences: for µ1 , valine was disfavored at the Y+3 position , whereas leucine was favored ( Ohno et al . , 1998 ) . The relatively modest interaction mediated by V11 is presumably compensated by Y6 , which stabilizes the binding through hydrogen-bonding interactions with N308 and E381 of μ1 and through stacking interactions with μ1 P383 and Y384 . Consistent with the notion that Y6 , Y8 , and V11 each contribute to the overall binding , substitution of any of these single residues with alanine was sufficient to disrupt the interaction between BST2CD and µ1 as detected using the Y2H assay ( Figure 5E ) . The structure also explains why BST2 , specifically the YxYxxV sequence , has a preference for μ1 , but not μ2 or μ3 ( Figure 1D ) . An overlay of the BST2CD-bound μ1 structure to the μ2 structure shows that μ2 lacks a tyrosine residue , corresponding to μ1 Y384 , which provides a stabilizing stacking interaction with BST2 Y6 ( Figure 5—figure supplement 1A ) . In addition , the presence of μ2 Q318 , in place of the smaller N308 of μ1 , disrupts a hydrogen-bonding interaction with the side chain of BST2 Y6 and potentially causes steric hindrance . In the case of μ3 , the structural difference is much more pronounced: severe steric clashes appear to prevent the binding of BST2 Y6 ( Figure 5—figure supplement 1B ) . Overall , these results not only explain a critical feature of Vpu-mediated hijacking of the AP1-dependent CCV pathway for BST2 antagonism , but also serve as the first structural example of a cellular YxY-based sorting signal bound to the μ1 subunit of AP1 . Opposite to the BST2-binding site on the activated AP1 core , VpuCD associates with the γ and σ1 subunits of AP1 by mimicking the canonical acidic dileucine-sorting motif ( Figure 5 ) . E62 forms a salt bridge with R15 of AP1 γ subunit , fulfilling the role of the ‘acidic residue’ within the sorting motif , while L66 and V67 embed into the hydrophobic pocket on AP1 σ1 that accommodates the canonical dileucine residues ( Figure 5F ) . The Vpu ELV motif overlays closely with a canonical acidic dileucine-sorting motif when bound to AP2 ( Figure 5G ) . Our results reveal that the ELV motif of Vpu acts as a sorting motif mimic for hijacking AP1 and thus the CCV pathway for mistrafficking of BST2 . While the previously observed Arf1-bound AP1 exhibits the same level of opening as the activated AP2 ( Jackson et al . , 2010; Ren et al . , 2013 ) , the AP1 in the BST2/Vpu-bound structure adopts a conformation that is much more open than the previously observed structures . When the BST2/Vpu-activated AP1 structure was overlaid with the Arf1-activated AP1 structure using the β1 subunits , a twisting of the γ and σ1 subunits was observed ( Figure 6A ) , which further exposes the dileucine-binding pocket at the γ/σ1 interface in AP1 . The conformational change involves a ∼20° rotation of γ/σ1 around an axis at the base of γ where it contacts β1 , with the largest Cα movement of ∼35 Å at the tip of γ ( Figure 6A ) . Of note , however , the relative positions either between β1 and μ1 or between γ and σ1 are well maintained . As a result of the γ/σ1 movement , new interactions occur between the μ1-CTD and the N-terminal portion of γ . This new γ–μ1 interface has a buried surface area of 685 Å2 , and consists of extensive , hydrogen-bonding and salt bridge interactions between the two subunits ( Figure 6B , C ) . We further created a double mutation , γ Q28R and μ1 D319R , to disrupt this new interface and tested its role in the binding of AP1 to BST2/Vpu in the GST pulldown assay . The binding was decreased , indicating that this newly observed γ–μ1 interface is important for the Vpu-mediated manipulation of AP1 ( Figure 6D ) . This new AP1 conformation supports the recent hypothesis that the AP complexes might be able to access a wider conformational space beyond what has been previously observed in the locked , unlatched , and open states ( Canagarajah et al . , 2013 ) . 10 . 7554/eLife . 02362 . 011Figure 6 . AP1 adopts a novel open conformation when bound to BST2CD and VpuCD . ( A ) Overlay of the Arf1-bound AP1 structure ( PDB ID 4HMY , grey surface ) ( Ren et al . , 2013 ) and the BST2CD-VpuCD-bound AP1 structure ( ribbon representation with β1 in blue , γ in orange , μ1 in green and σ1 in yellow ) . The broken arrows point to the large relative movement of the γ/σ1 subunits of our structure with respect to Arf-1/AP1 . Note that the β1/μ1 subunits overlay well in the two structures . ( B and C ) Close-up views of the new γ-μ1 interface observed in the present study , boxed in ( A ) , with hydrogen bonds represented by dashes . ( D ) GST affinity pulldown assay using an AP1 core complex with GST-tagged γ subunit . Mutations at the γ-μ1 interface reduced MBP-BST2CD-VpuCD binding to AP1 . DOI: http://dx . doi . org/10 . 7554/eLife . 02362 . 011 The newly formed γ–μ1 interface leads to local structural changes . Specifically , the μ1 loop from P363 to G372 , which is disordered in the previous AP1 structures ( Heldwein et al . , 2004; Jia et al . , 2012; Ren et al . , 2013 ) , became well ordered in the current structure . Although all of the nearby residues from both γ and μ1 are well ordered , considerable additional electron density remains near the γ–μ1 interface , suggesting the possible presence of Vpu residues ( Figure 7A ) . Although the quality of the additional electron density does not permit model building , we suspect it may belong to a portion of VpuCD that is N-terminal to the ELV , judging by the location of the electron density relative to Vpu ELV ( Figure 7A ) . Specifically , Vpu residues R44 and L45 ( NL4-3 sequence , corresponding to I45 of Vpu HV1S1 used in the structural study ) upstream of ELV may participate in the interaction with AP1 , as these conserved residues have been implicated in Vpu-mediated antagonism of BST2 ( Lucas and Janaka , 2012; Pickering et al . , 2014 ) . 10 . 7554/eLife . 02362 . 012Figure 7 . Vpu R44 , L/I45 residues may interact with AP1 at the γ-μ1 interface and contribute to the optimal Vpu activity . ( A ) The difference Fourier map ( mFo-DFc at 3σ level , blue mesh ) near γ ( orange ) -μ1 ( green ) interface of AP1 . VpuCD is shown in cyan sticks and AP1 σ1 is in yellow . ( B and C ) Affinity pulldown assays using an AP1 with GST-tagged γ subunit ( B ) or using a His-tagged MBP-μ1-CTD ( C ) . SDS PAGE analysis shows the amount of protein in the load and elution . Mutation of these potentially important Vpu residues at the γ-μ1 interface , R44A:I45A , significantly affected the binding interactions . ( D ) Downregulation of surface BST2 by Vpu is impaired by the R44A:L45A ( RL/AA ) mutation as well as by the S52N:S56N ( S52/56N ) mutation . ( E ) HEK293T cells were co-transfected to express provirus lacking vpu , and the indicated codon-optimized Vpu proteins . The Vpu R44A:L45A and S52N:S56N mutations each impair Vpu mediated-virion release , and their effects are additive . ( F ) Western blot of the experiment shown in E , indicating the expression levels of BST2 and Vpu; gp120 is the viral envelope glycoprotein . ( G ) Co-IP showing that β-TrCP binding was abolished by the S52/56N ( ‘2/6’ ) mutation , but not the RL/AA mutation . DOI: http://dx . doi . org/10 . 7554/eLife . 02362 . 01210 . 7554/eLife . 02362 . 013Figure 7—figure supplement 1 . VpuCD exhibits large conformational flexibility . Comparison of VpuCD conformations observed in previous NMR studies ( A: PDB ID 1VPU and B: PDB ID 2K7Y ) ( Willbold et al . , 1997; Wittlich et al . , 2009 ) and the current work ( C ) . The molecules in ( A ) and ( B ) are colored in a blue-to-red color ramp from N-terminus to C-terminus . Residues R44 and L/I45 and the ELV motif are shown in sticks . DOI: http://dx . doi . org/10 . 7554/eLife . 02362 . 013 We tested the Vpu R44A and I45A double mutation for its effect on the binding to AP1 in vitro . The binding was impaired by this mutation ( Figure 7B ) . Specifically , the mutation substantially reduced the binding between the BST2CD-VpuCD fusion and μ1-CTD of AP1 in our pulldown assay ( Figure 7C ) . This interaction likely explains the observed affinity between Vpu and μ1 ( Figure 2C ) . It also suggests that the extra electron density in our structure ( Figure 7A ) may come from Vpu residues including R44 and L/I45 . These Vpu residues may interact with γ/μ1 subunits of AP1 and stabilize γ–μ1 contacts and the novel AP1 conformation . We characterized the effect of the R44A:L45A mutation on the activity of Vpu as an antagonist of BST2 in human cells . The R44A:L45A mutation significantly impaired the ability of Vpu to reduce the amount of BST2 at the cell surface ( Figure 7D ) . It also greatly impaired the ability of Vpu to enhance virion release ( Figure 7E ) . Interestingly , the effect of the R44A:L45A mutation on virion release was additive with the S52/56N mutation , which ablates the binding of Vpu to β-TrCP ( Figure 7E , G ) , despite that both mutations impair the apparent degradation of BST2 ( Figure 7F ) . These data are consistent with the proposed role of R44 and I45 in the binding of Vpu to AP1 rather than to β-TrCP . Indeed , the R44A:L45A mutation did not affect the interaction with β-TrCP as measured by immunoprecipitation ( Figure 7G ) . Together , the impaired abilities of Vpu R44A:L45A to bind AP1 and to decrease the steady-state expression of BST2 suggest that AP1 is important for the Vpu-mediated endo-lysosomal degradation of BST2 . We note that to achieve the interaction of R44 , L/I45 with µ1 and the interaction of the ELV sequence with γ-σ1 , VpuCD must adopt an extended conformation . Although observed with considerable helical content in previous NMR studies , VpuCD was believed to be flexible ( Figure 7—figure supplement 1 ) ( Willbold et al . , 1997; Wittlich et al . , 2009 ) . Furthermore , these secondary structures were observed under conditions that induce helix formation . Specifically , the helical feature of the first half of VpuCD was shown to be relatively more pronounced , whereas the latter helix , harboring the ELV sequence , exhibited low helical content and was more likely to be unstructured ( Wittlich et al . , 2009 ) . Our study provides further evidence for the flexible nature of VpuCD . We performed virion release assays to confirm the functional requirements of the BST2 YxY and Vpu ELV motifs in human cells ( Figure 8 ) . The mutation Y6/8A in BST2CD markedly impaired the ability of Vpu to promote virion release , supporting a critical role for this trafficking motif in antagonism of BST2 by Vpu . Since the BST2 YxY motif is not likely required for the β-TrCP-dependent degradation pathway , which functions via ubiquitination , its importance likely comes from its affinity to AP1 and clathrin-dependent pathways . Of note , the Y6/8A mutant was better expressed than wild-type BST2 and restricted virion release more effectively both in the presence and absence of Vpu . Alanine mutation of Vpu ELV also impaired virion release , to a degree similar to that of the S52/56N mutation of the β-TrCP binding site . Notably , mutation of the ELV sequence in Vpu and the YxY sequence in BST2 increased the steady-state expression of BST2 , presumably by inhibiting endo-lysosomal degradation either as it occurs natively or as stimulated by Vpu . This scenario is consistent with our structural and biochemical observations that both BST2 YxY and Vpu ELV motifs interact with AP1 , and it further supports the hypothesis that this interaction is part of the endo-lysosomal degradation mechanism that supports Vpu-activity . 10 . 7554/eLife . 02362 . 014Figure 8 . The BST2 YxY as well as the Vpu ELV and di-serine motifs are each important for Vpu-mediated reduction in BST2-expression and for optimal virion release . ( A ) HEK293T cells were co-transfected to express the indicated BST2 variants , provirus lacking vpu , and the indicated codon-optimized Vpu proteins . Mutations of the BST2 YxY , Vpu ELV and S52/S56 ( ‘2/6’ ) sequences have additive effects in decreasing the efficiency of Vpu mediated- virion release . ( B ) Western blot showing the expression levels of BST2 ( WT or Y6/8A ) and Vpu ( WT or mutants ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02362 . 014 We used immunofluorescence microscopy to verify the co-localization of BST2 , Vpu and AP1 at the juxtanuclear region near the cell center ( Figure 9; Figure 9—figure supplement 1 ) . In HeLa cells that express BST2 constitutively , BST2 and AP1 co-localized even in the absence of Vpu , and all three proteins co-localized when Vpu was expressed ( Figure 9—figure supplement 1 ) . As expected , this co-localization was not affected by the Vpu S52/56N mutation , consistent with our in vitro observation that the β-TrCP binding motif of Vpu is not required for interaction with AP1 . We further performed the immunofluorescence experiments after stably transfecting HT1080 cells to express BST2 ( allowing the analysis of mutants since these cells do not naturally express BST2 ) followed by transient transfection to express Vpu . As was seen using the HeLa cells , BST2 and AP1 co-localized in the absence of Vpu , and all three proteins co-localized at the cell center when Vpu was expressed ( Figure 9A ) . Mutation of the YxY sequence of BST2 , and to a lesser extent the ELV sequence of Vpu , caused mislocalization of the proteins from the cell center region into more peripheral puncta , although some overlap at the cell center with AP1 persisted . Quantitative image analyses indicated that the YxY sequence of BST2 contributes to that protein's co-localization with AP1 , as does the ELV sequence of Vpu ( Figure 9B ) . 10 . 7554/eLife . 02362 . 015Figure 9 . YxY and ELV motifs contribute to BST2/Vpu/AP1 co-localization and the trafficking of BST2/Vpu complexes . ( A ) HT1080 cells stably expressing BST2 WT or the Y6/8A mutant were transfected to express Vpu or the ELV/AAA mutant . The cells were stained for Vpu , BST2 , and AP1 ( γ subunit ) . ( B ) Pearson co-efficient of correlation between AP1 and Vpu or between AP1 and BST2 vs the intensity of each protein . Each datum point represents a single cell . The correlation between Vpu and AP1 was analyzed in cells expressing wild type BST2 , whereas that between BST2 and AP1 was analyzed in cells that were microscopically negative for Vpu expression ( although transfected with the wild type Vpu-expression construct ) . Mutation of the YxY sequence in BST2 and the ELV sequence in Vpu decrease each protein's overlap with AP1 . Statistical significance of these data is shown on the bottom . ( C ) HT1080 cell were transfected to express the BST2-Vpu chimera . The wild type chimera co-localized with AP1 at the perinuclear region . Alanine mutations of the YxY and ELV motifs decreased the co-localization of the BST2-Vpu chimera and AP1 and caused displacement of the chimera to peripheral endosomes and the plasma membrane . ( D ) The surface expression of the BST2-Vpu chimera after expression in HEK293T or HT1080 cells by transient transfection was measured using anti-FLAG antibody and flow cytometry . GFP was expressed from a separate plasmid as a transfection marker . The histograms show the cell number vs the intensity of surface stain for the chimera in the GFP-positive ( transfected ) cells . The Y6/8A mutation caused increased surface expression , as did the ELV/AAA mutation , although to a lesser extent . The highest expression at the cell surface was observed in case of the mutant containing both Y6/8A and ELV/AAA mutations . DOI: http://dx . doi . org/10 . 7554/eLife . 02362 . 01510 . 7554/eLife . 02362 . 016Figure 9—figure supplement 1 . Co-localization of BST2 , Vpu , and AP1 in HeLa cells transfected to express either the wild type or the S52N/S56N mutant of Vpu . DOI: http://dx . doi . org/10 . 7554/eLife . 02362 . 016 To clarify the role of these sequences in the trafficking of BST2–Vpu complexes , we analyzed the subcellular localization of our BST2–Vpu chimera that includes each protein's transmembrane domain ( Figure 4B ) . This chimera has the advantage of ‘forcing’ the interaction between the two proteins , allowing the experiment to follow the fate of the chimera as a surrogate for BST2/Vpu complexes , without a potentially large background of the individual , uncomplexed proteins . The wild-type chimera localized to the cell center with AP1 as expected ( Figure 9C ) . In contrast , the Y6/8A-ELV/AAA mutant colocalized less well with AP1; moreover , it was dispersed into peripheral endosomes and highlighted the plasma membrane . Correspondingly , enhanced surface expression was detected for the mutant chimera by flow cytometry ( Figure 9D ) , indicating the loss of Vpu's ability to downregulate BST2 from the cell surface . Overall , these data support the proposed roles of the BST2 YxY and Vpu ELV sequences in the hijacking of AP1 by Vpu for the mistrafficking of BST2 .
Mistrafficking of host immune molecules via clathrin-dependent pathways is a strategy frequently employed by primate lentiviruses ( Tokarev and Guatelli , 2011 ) . For example , hijacking of AP-mediated trafficking pathways is well documented for the viral protein Nef ( Piguet et al . , 1999; Roeth and Collins , 2006; Arhel and Kirchhoff , 2009 ) . Nef binds AP1 to prevent MHC-I from reaching the cell surface , providing evasion of immune surveillance by cytotoxic T cells ( Collins et al . , 1998; Le Gall et al . , 1998; Roeth et al . , 2004; Noviello et al . , 2008 ) . Nef also engages AP2 to remove the primary viral receptor , CD4 , from cellular membranes to avoid super-infection , interference with virion release and infectivity , and the exposure on the cell surface of CD4-induced epitopes within the viral Env glycoprotein ( Garcia and Miller , 1991; Foti et al . , 1996; Craig et al . , 1998; Chaudhuri et al . , 2007; Veillette et al . , 2013; Pham et al . , 2014 ) . In SIV , Nef uses AP2 to remove BST2 from the cell surface to allow for the efficient release of progeny virions ( Zhang et al . , 2011; Serra-Moreno et al . , 2013 ) . In addition , the HIV-2 Env protein antagonizes BST2 through an AP2-dependent mistrafficking mechanism ( Hauser et al . , 2010 ) . The work presented herein reveals that HIV-1 Vpu is another viral modulator of host membrane trafficking pathways , specifically , for the AP1-mediated mistrafficking of BST2 . This is conceptually similar to the HIV-1 Nef-mediated MHC-I downregulation but occurs through different interaction mechanisms with AP1 ( Figure 10 ) . In each case , the HIV-1 protein , Nef or Vpu , causes the retention of its cellular target , MHC-I or BST2 , at the TGN and eventually leads to lysosomal degradation through late endosomal pathways . Nef builds upon an incomplete tyrosine-based sorting motif in the MHC-I cytoplasmic domain ( CD ) and promotes a cooperative three-protein interaction involving MHC-I CD , Nef and μ1 of AP1 ( Figure 10B ) ( Noviello et al . , 2008; Wonderlich et al . , 2008; Singh et al . , 2009; Jia et al . , 2012 ) . The binding mode for the three-component complex involving BST2 , Vpu and AP1 , is different , with no three-protein interface . The complex is instead stabilized by pair-wise binary interactions: BST2 and Vpu bind through the transmembrane regions , BST2CD binds the AP1 μ1 subunit , and VpuCD binds the AP1 σ1/γ/μ1 subunits ( Figure 10A ) . This model is consistent with the previous findings that the transmembrane interaction between BST2 and Vpu is of pivotal importance in determining the activity of Vpu against BST2 . The BST2/Vpu/AP1 interaction also causes an open conformation of AP1 that has not been observed before . This new AP1 conformation might only be induced by Vpu or alternatively it might exist as a physiologically functional state that Vpu selectively uses to its advantage . 10 . 7554/eLife . 02362 . 017Figure 10 . Schematics of hijacking of AP1 by HIV-1 Vpu to target BST2 ( left ) and by HIV-1 Nef to target MHC-I ( right ) ( Jia et al . , 2012 ) . Transmembrane helices are represented by cylinders and the Nef myristoyl anchor is represented by a cyan sphere . DOI: http://dx . doi . org/10 . 7554/eLife . 02362 . 017 The Vpu-mediated antagonism of BST2 appears to be a multifaceted process . Besides AP1-mediated mistrafficking and degradation , Vpu can also induce the degradation of BST2 through β-TrCP-mediated ubiquitination and the ESCRT machinery . Abolishing either the ubiquitination pathway by the S52/56N mutation in Vpu or the mistrafficking pathway by the ELV/AAA mutation each led to defects in the enhancement of virion release ( Figure 8 ) ( Kueck and Neil , 2012; McNatt et al . , 2013 ) . Combining the two sets of mutations , however , appeared to cause the most substantial loss of Vpu function , suggesting a potentially parallel nature for these pathways . How these pathways , ubiquitination and AP1-mediated mistrafficking , might work together to enable the antagonism of BST2 by Vpu remains to be further elucidated in finer temporal and spatial details . The complexity of these trafficking and degradation mechanisms may have prevented a clear identification of the AP complex ( es ) responsible for the Vpu-activity in previous studies ( Kueck and Neil , 2012 ) . Despite the evidence supporting the involvement of clathrin-associated pathways , neither AP1 , 2 or 3 knockdown in 293T cells and HeLa cells nor AP1 γ knockout in mouse fibroblasts had any apparent effects on the ability of Vpu to antagonize restriction by BST2 ( Kueck and Neil , 2012 ) , although we previously reported a role for AP2 in the Vpu-mediated surface-downregulation of BST2 ( Mitchell et al . , 2009 ) . Conceivably , the individual roles of a given pathway are obscured by compensation from overlapping parallel pathways , such as the involvement of multiple AP complexes or the monomeric clathrin adaptor HRS . Another possibility is redundancy in the composition of an individual AP complex . A specific subunit of AP could be knocked down or out only to be replaced by a different isoform ( Boehm and Bonifacino , 2001 ) . Substitution might also happen between two related subunits from different AP complexes as they have highly conserved structures ( Keyel et al . , 2008; Li et al . , 2010 ) . Another level of complexity might come from the potential for the mechanism of Vpu activity to be at least partly cell-type dependent . For example , BST2 in some cell types appears to be equally vulnerable to β-TrCP-dependent degradation and to mistrafficking , whereas in others the role of β-TrCP-dependent degradation is less important , possibly due to lower levels of BST2 expression ( Schindler et al . , 2010 ) . The intrinsic , structural flexibility of Vpu may enable its functional versatility and underlie the complexity of BST2-antagonism . A shared feature of retroviral accessory proteins including Vpu , Nef , Vif , Vpr and Vpx is the presence of long and flexible loops that provide the structural basis for their multifunctional nature ( Xue et al . , 2012 ) . These unstructured regions can adopt different conformations to facilitate binding to different host target proteins . These regions often harbor motifs mimicking functional host sequences , enabling the virus to hijack the desired host machineries ( Kadaveru et al . , 2008 ) . The binding of the Vpu ELV motif to AP1 revealed in our structure , as well as that of the di-phosphoserine motif of Vpu to β-TrCP shown before , serve as two excellent examples of this viral strategy . The ExxxLV motif is conserved in the Vpu of subtype B of group M HIV-1 , to which both NL4-3 and HV1S1 strains of HIV-1 used in this study belong . However , in the Vpu sequence from the subtype C of the group M HIV-1 , the motif is replaced by the consensus sequence ExxxMV , which might or might not be functionally equivalent to the ExxxLV sequence . Interestingly , while subtype B Vpu localizes predominantly within internal membrane compartments , subtype C Vpu is substantially expressed at the cell surface ( Pacyniak et al . , 2005 ) . In addition to the sequence variation in the ELV region , subtype C Vpu contains another putative acidic dileucine motif in the membrane-proximal region of its CD , mutation of which affected the trafficking of Vpu ( Ruiz et al . , 2008 ) . Conceivably , subtype C Vpu hijacks AP complexes using residues distinct or partly overlapping with those used by subtype B Vpu . The intrinsic affinity of Vpu to AP1 and AP2 suggests that Vpu has the ability to hijack clathrin-dependent trafficking pathways to target other cellular proteins with which it interacts simultaneously . Besides BST2 and CD4 , HIV-1 Vpu also modulates the surface expression of two other important immune factors , NTB-A and CD1d ( Sandberg et al . , 2012 ) . NTB-A is a co-activator for natural killer ( NK ) cells , and the Vpu induced downregulation of NTB-A enables infected cells to evade lysis by NK cells ( Shah et al . , 2010 ) . This involves the mistrafficking of NTB-A , which phenotypically resembles Vpu-mediated BST2 mistrafficking ( Bolduan et al . , 2013 ) . Although degradation is not induced , Vpu causes retention of NTB-A in a perinuclear compartment . Like BST2 , the downregulation depends on an interaction between the transmembrane domains of Vpu and NTB-A ( Shah et al . , 2010 ) . In addition , residues other than the conserved phosphorylated serines in VpuCD are critical ( Bolduan et al . , 2013 ) . In the case of CD1d , an antigen-presenting molecule in dendritic cells , downregulation by Vpu occurs through interference with the recycling of CD1d in the endosomal compartments ( Moll et al . , 2010 ) . A critical residue of CD1d , Tyr331 in a canonical sorting motif , is key for both the natural trafficking and the Vpu-mediated CD1d downregulation . Whether CD1d also interacts with Vpu in the transmembrane or other regions remains to be determined . Given the similarities between these systems , we speculate that the modulation of NTB-A and/or CD1d by Vpu may also involve clathrin-dependent trafficking pathways . Finally , we consider the question of why BST2 interacts on its own with AP1 , since this must relate in some way to the protein's natural function . One possibility is that the interaction with AP1 allows BST2 that is internalized from the cell surface to recycle to the plasma membrane , where it acts to trap nascent virions . Another possibility derives from the ability of BST2 to induce the activity of the NF-κB family of transcription factors ( Cocka and Bates , 2012; Galao et al . , 2012; Tokarev et al . , 2013 ) ; this signaling function could occur from an endosomal compartment that BST2 reaches via AP1-dependent trafficking . Yet another possibility is that BST2 uses AP1 to divert tethered virions to late endosomal compartments in antigen presenting cells , thus facilitating the presentation of viral antigens in the context of class II or class I MHC ( Neil et al . , 2006; Miyakawa et al . , 2009; Moffat et al . , 2013 ) . Whether these or other processes underlie the physiologically relevant function of the interaction between BST2 and AP1 remains to be determined . In conclusion , our biochemical and biological data support , and our crystal structure elucidates in atomic detail , how HIV-1 Vpu hijacks the AP1-dependent membrane trafficking pathway to antagonize the host restriction factor BST2 . These results not only help unravel the perplexing mechanism of Vpu-mediated antagonism of BST2 but also suggest a potentially evolutionarily conserved function of Vpu in antagonizing other host immune targets . This function might be critical in understanding the full capacity of Vpu in promoting the infectivity and pathogenesis of the primate lentiviruses .
The BST2CD-VpuCD fusion consisted of BST2 ( 1-21 ) and Vpu ( HV1S1 , 28-80 ) , connected by a linker peptide of 10 amino acids: GSDEASEGSG . The encoding gene was cloned into the pMAT9s expression vector containing a N-terminal 6xHis tag followed by the maltose binding protein ( MBP ) and a SARS-CoV Mpro cleavage site ( Xue et al . , 2007 , 2008 ) . To introduce phosphorylation on Vpu S52 and S56 , this plasmid and a pCDFDuet vector encoding both the α and β subunits of the casein kinase II ( CK2 ) were co-transformed into the E . coli NiCo21 ( DE3 ) competent cells ( New England Biolabs ) for expression in the Terrific broth . Cells were induced with 0 . 1 mM isopropyl β-d-thiogalactopyranoside ( IPTG ) at OD600 of 0 . 8 and grown at 16°C overnight . The protein was first purified with the Ni-NTA affinity column . For crystallization , the 6xHis-MBP tag was cleaved off by the SARS-CoV Mpro protease . Protein was subsequently purified on a HiTrap Q anion exchange column and a Superdex 75 size exclusion column , which yielded homogenous monomeric protein . Double phosphorylation of the protein was confirmed by using MALDI mass spectrometry . For GST pull-down experiments , the 6xHis-MBP-tagged proteins , either wild-type or mutants , were purified through Ni-NTA , HiTrap Q and Superdex 200 size exclusion columns . The unphosphorylated form or the S52/56N ( Vpu ) mutant of the fusion was produced similarly except that the pCDFDuet-CK2 plasmid was not included during expression . 6xHis-MBP-VpuCD ( phosphorylated ) and 6xHis-MBP-BST2CD were expressed and purified similarly . For the AP1 core used in the crystallization , the genes encoding residues 1-613 of mouse γ and 1-158 of human σ1 were subcloned into the pCDFDuet vector , while genes encoding residues 1-584 of human β1 and 1-423 of mouse μ1 were subcloned into the pETDuet vector . Non-cleavable 6xHis tags were included at the N-termini of both the γ and β1 subunits . The heterotetrameric AP1 core was expressed overnight at 22°C in the NiCo21 ( DE3 ) cells in Terrific broth after induction by 0 . 1 mM IPTG . The complex was purified sequentially through Ni gravity , HiTrap Q , and Superdex 200 size exclusion columns . The μ1-CTD-truncated AP1 core was created by introducing a stop codon after residue 145 of the μ1 subunit in the pCDFDuet vector carrying both the γ and μ1 subunits . The truncated AP1 core was expressed and purified similarly . The AP1 core used in the GST pull-down experiments further included a GST tag at the C-terminus of the γ subunit . The AP1-GST complex was expressed similarly as above . The complex was purified by Ni gravity and GST columns , followed by buffer-exchange to remove the glutathione for its subsequent use in the GST pull-down assays . MBP-μ1-CTD and MBP-μ2-CTD were created previously ( Jia et al . , 2012 ) . MBP-μ3-CTD was created by subcloning the gene encoding residues 166-418 of rat μ3 into the pMAT9s expression vector . These proteins were overexpressed and purified as described previously ( Jia et al . , 2012 ) . The hemicomplex of AP2 , α398-σ , was created by subcloning the rat α ( 1-398 , with three surface mutations I370A:I374S:L393A ) and human σ into the pCDFDuet vector . The heterodimer was expressed similarly as the BST2CD-VpuCD fusion above and was purified sequentially on a Ni-NTA affinity column , a HiTrap S cation exchange column , and a Superdex 200 size exclusion column . Crystallization was carried out using the microbatch under-oil method . The purified AP1 core and the BST2CD-VpuCD fusion were mixed at 1:3 molar ratio to a final concentration of 4 . 5 mg/ml ( 25 mM Tris , pH 8 . 0 , 100 mM NaCl , 0 . 1 mM TCEP , 0 . 1 mM PMSF , 0 . 2 mM EDTA ) . Equal volumes of the protein solution and the precipitant solution ( 100 mM Tris , pH 7 . 0 , 150 mM NaCl , 8% PEG6000 ) were mixed . The drop was sealed using a mixture of paraffin and silicon oil at a 2:1 ratio . Crystals appeared within 24 hr at room temperature and grew to full size in about a week . Crystals were cryo-protected using the precipitant solution containing 20% glycerol and then frozen in liquid nitrogen . Datasets were collected at NE-CAT beamline 24ID-C at the Advanced Photon Source , Argonne National Laboratory , and beamline X29 at the National Synchrotron Light Source , Brookhaven National Laboratory . The crystals were in the P43 space group and diffracted to a highest resolution of 3 . 0 Å . The data collection statistics are summarized in Table 1 . 10 . 7554/eLife . 02362 . 018Table 1 . Crystallographic data collection and refinement statisticsDOI: http://dx . doi . org/10 . 7554/eLife . 02362 . 018NativeData collection Space groupP43 Cell dimensions a , b , c ( Å ) 160 . 5 , 160 . 5 , 118 . 4 Wavelength ( Å ) 0 . 9792 Resolution ( Å ) 48 . 8–3 . 0 ( 3 . 11–3 . 0 ) Rmerge0 . 077 I/σI13 . 6 ( 0 . 9 ) Completeness ( % ) 99 . 4 ( 99 . 5 ) Redundancy3 . 8 ( 3 . 7 ) Refinement Unique reflections59937 Rwork/Rfree0 . 186/0 . 229 No . atoms Protein13 , 926 Water14 B-factors Protein107 . 1 Water74 . 2 R . m . s deviations Bond lengths ( Å ) 0 . 013 Bond angles ( ° ) 1 . 27 Ramachandran Favored95% Outliers0 . 23%Values in parenthesis are for highest-resolution shell . The structural solution was obtained by molecular replacement using PHASER ( McCoy et al . , 2007 ) implemented in PHENIX ( Adams et al . , 2010 ) . Only one molecule exists in the asymmetric unit . The structure of the closed AP1 core ( PDB ID: 1W63 ) ( Heldwein et al . , 2004 ) was divided into three search models: the γ and σ1 hemicomplex , the β1 and μ1-NTD hemicomplex , and the μ1-CTD . These models were used in sequential searches to successfully obtain the solution . Iterative rounds of model building in COOT ( Emsley et al . , 2010 ) and refinement with PHENIX ( Adams et al . , 2010 ) were carried out . Data sharpening was performed to enhance the details in the electron density map ( Liu and Xiong , 2013 ) . BST2CD and VpuCD were built into the model based on the prominent difference densities at their corresponding locations . The final model has an Rwork/Rfree of 0 . 18/0 . 22 . The refinement statistics are summarized in Table 1 . The purified proteins , MBP-μ1/2/3-CTD ( 0 . 86 mg ) or α398-σ ( 0 . 73 mg ) and MBP-BST2CD/MBP-VpuCD ( 0 . 73 mg ) or their mutants , were mixed to a final volume of 500 µl . Similarly , the μ1-CTD-truncated AP1 ( 3 . 4 mg ) and MBP-VpuCD ( 1 . 2 mg , wt or the ELV/AAA mutant ) were mixed to a final volume of 500 µl . All samples were incubated overnight at 4°C and then applied to the Superdex 200 10/300 GL size exclusion column pre-equilibrated with the elution buffer ( 25 mM Tris , pH 8 , 100 mM NaCl , 0 . 1 mM TCEP ) . The elution fractions were analyzed by using SDS PAGE . The purified proteins , AP1-GST ( 0 . 24 mg ) and MBP-BST2CD-VpuCD/MBP-BST2CD/MBP-VpuCD ( 0 . 18 mg ) or their mutants , were mixed in a final volume of 100 µl and incubated at 4°C overnight . The protein solution was then loaded onto a small gravity flow column containing 0 . 2 ml GST resin . Flow through was collected and the resin was extensively washed with 5 × 0 . 9 ml GST binding buffer ( 50 mM Tris , pH 8 , 100 mM NaCl , 0 . 1 mM TCEP ) . The bound proteins were then eluted with 5 × 0 . 1 ml GST elution buffer containing 10 mM reduced glutathione . The eluted proteins were analyzed on SDS PAGE stained with Coomassie blue . 6xHis-MBP-μ1 ( 0 . 25 mg ) and the BST2CD-VpuCD fusion ( 0 . 1 mg ) or its mutants were mixed in a final volume of 100 µl and incubated at 4°C for 2 hr . The proteins were loaded on a small gravity column containing 0 . 2 ml Ni-NTA resin . Flow through was collected and the resin was extensively washed with 5 × 0 . 9 ml binding buffer containing 20 mM imidazole . The bound proteins were subsequently eluted with 5 × 0 . 1 ml Ni elution buffer containing 400 mM imidazole . The eluted proteins were analyzed on SDS PAGE stained with Coomassie blue . AH109 yeast cells ( Clontech Laboratories , Inc , Palo Alto , CA ) were co-transformed to express both hybrid proteins . The GAL4-Activation domain in pACT2 fused with the μ1 subunit was provided by Juan Bonifacino . The GAL4-DNA binding domain ( DBD ) in pGBT9 was fused to the BST2 cytoplasmic domain ( CD ) with the linker GGGSGGGSGGGS inserted between the BST2CD start codon and the DBD of the GAL4 protein . BST2CD was inserted between the EcoRI and SalI restriction sites of the pGBT9 multiple cloning site . Yeast cells were transformed by the lithium-acetate method . Transformants were picked; colonies were pooled and grown in -Leu/-Trp liquid media before plating as spots on -Leu/-Trp solid media as a growth control or on -Trp/-Leu/-His solid media containing 3-aminotriazole ( 3AT ) to test for interaction . For the virion release assays , the following plasmids were used: the proviral Vpu-mutant pNL43/Udel ( Klimkait et al . , 1990 ) ; pcDNA3 . 1-BST-2 from Autumn Ruiz and Edward Stephens; pcDNA3 . 1-BST2-Y6/8A constructed by site-directed mutagenesis of pcDNA3 . 1-BST2 using the QuikChange II Site-Directed Mutagenesis Kit ( Stratagene , La Jolla , CA ) ; pVpHu-FLAG constructed by cloning the codon-optimized Vpu sequence from pVpHu ( from Klaus Strebel ) with primers introducing a C-terminal FLAG tag ( DYKDDDDK ) into the NheI and XhoI sites of the pcDNA3 . 1 ( − ) vector ( Life Technologies , Carlsbad , CA ) ; and the FLAG-tagged Vpu-mutant expression plasmids pVpHu-FLAG-S52/56N , pVpHu-FLAG-ELV59 , 63 , 64AAA and pVpHu-FLAG-S52/56N + ELV/AAA constructed by site-directed mutagenesis of pVpHu-FLAG using QuikChange ( Stratagene ) . HEK293T cells in six-well plates were transfected using Lipofectamine 2000 ( Life Technologies , Carlsbad , CA ) with 50 ng of the indicated BST2 expression plasmid , 250 ng of the indicated VpHu-FLAG expression plasmid and 3600 ng of the Vpu-mutant proviral plasmid pNL43/Udel . Supernates were collected 24 hr after transfection as previously described ( Van Damme et al . , 2008 ) . Virion-associated p24 was pelleted through a 20% sucrose cushion before measurement by p24 ELISA ( Advanced Bioscience Laboratories , Rockville , MD ) . Cells were harvested and lysed 24 hr post-transfection and immunoblots for BST2 , gp120 , actin , and FLAG-epitope were performed as previously described ( Day et al . , 2004; Tokarev and Guatelli , 2011; Tokarev et al . , 2013 ) . 293T cells were transfected with the plasmids expressing the constructs indicated in the figure legends , using Lipofectamine 2000 and the manufacturer's protocol . The next day , cells were lysed in lysis buffer ( 50 mM Tris HCl pH7 . 4 , 150 mM NaCl , 1 mM EDTA , 1% Triton X-100 and 5% glycerol ) supplemented with protease inhibitors cocktail ( Roche Diagnostics ) . Lysates were cleared by centrifugation for 10 min at 16 , 000×g and incubated with magnetic anti-FLAG-coated beads for BST2 , Vpu or the BST2-Vpu chimera; or with anti-HA-coated beads for β-TrCP , which were pre-blocked with 3% BSA in PBS , for 2 hr at 4°C with continuous rotation . Beads were washed three times with lysis buffer containing 250 mM NaCl . The precipitated material was eluted with boiling in 2x Laemmli buffer and subjected to Western blotting . Endogenous µ1 was detected using a rabbit antiserum provided by Linton Traub . HeLa P4R5 or HT1080 cells were transfected with Vpu or BST2-Vpu chimera constructs , as indicated in the figure legends , using Lipofectamine 2000 . The next day , cells were fixed with 4% paraformaldehyde , permeabilized with 0 . 2% NP-40 , blocked with PBS containing 5% donkey serum , 5% goat serum ( Jackson Immunoresearch ) and 3% BSA . For Figure S6 , cells were first stained with mouse anti-γ adaptin ( clone 100/3 , Sigma-Aldrich ) and anti-Vpu rabbit serum , washed three times with PBS and stained with anti-rabbit antibody conjugated to rhodamine red-X and anti-mouse antibody conjugated to FITC ( Jackson Immunoresearch ) . Cells were then blocked with 5% mouse serum and stained with mouse anti-BST2 conjugated to Alexa Fluor 647 ( Biolegend , San Diego CA ) . For Figure 9A , the staining was performed as just described , except that the secondary antibody to detect Vpu was anti-rabbit conjugated to FITC and the secondary antibody to detect AP1 was anti-mouse conjugated to rhodamine red-X . For Figure 9C , the chimeras were visualized using anti-Vpu rabbit serum , and AP1 was visualized using mouse anti-γ adaptin . The secondary antibodies used were anti-mouse conjugated to FITC and anti-rabbit conjugated to rhodamine red-X . After staining , cells were washed , fixed again , and mounted in solid media ( Fluka ) . Images were acquired as Z-series using an Olympus microscope in wide-field mode . Images were deconvolved using a nearest neighbors method ( SlideBook software , Intelligent Imaging Innovations , Denver , CO ) and the Z-series were collapsed into single projection images . Co-localization was quantified using the Pearson correlation coefficient function in SlideBook . T-values were calculated using Microsoft Excel . Composite images were generated using Adobe Photoshop . Two-color flow cytometry was performed on unpermeabilized cells using an Accuri C6 flow cytometer to measure the intensity of GFP and the FLAG-epitope , which was detected indirectly using murine anti-FLAG ( Sigma-Aldrich ) followed by allophycocyanin ( APC ) -conjugated anti-mouse antibody ( Biolegend , San Diego , CA ) . | HIV is a retrovirus that attacks the immune system , making the body increasingly susceptible to opportunistic infections and disease and eventually leading to AIDS . While antiretroviral drugs have allowed people with AIDS to live longer , there is no cure or vaccine for HIV . Two types of HIV exist , with HIV-1 being much more common and pathogenic than HIV-2 . Like other ‘complex’ retroviruses , the HIV-1 genome contains genes that encode various proteins that allow the virus to disrupt the immune response of the host it is attacking . Viral protein u is a protein encoded by HIV-1 ( but not HIV-2 ) that counteracts an antiviral protein called BST2 in the host . BST2 , which is part of the host's innate immune response , prevents newly formed viruses from leaving the surface of infected cells . By counteracting BST2 , viral protein u allows the virus to spread in the host more efficiently . Like many proteins , newly produced BST2 is packaged inside structures called vesicles in a part of the cell called the trans-Golgi network , and then sent to its destination . Complexes formed by various proteins make sure that the vesicles take their cargo to their correct destinations within the cell . Two adaptor protein complexes—known as AP1 and AP2—are thought to be involved the transport of BST2 . However , it is not known how viral protein u stops BST2 from reaching the cell surface , or how it decreases the amount of BST2 in the cell as a whole . Jia et al . show how viral protein u and BST2 jointly interact with AP1 . This interaction leads to the mistrafficking and degradation of BST2 and the counteraction of its antiviral activity . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"structural",
"biology",
"and",
"molecular",
"biophysics",
"microbiology",
"and",
"infectious",
"disease"
] | 2014 | Structural basis of HIV-1 Vpu-mediated BST2 antagonism via hijacking of the clathrin adaptor protein complex 1 |
The molecular machinery responsible for DNA expression , recombination , and compaction has been difficult to visualize as functionally complete entities due to their combinatorial and structural complexity . We report here the structure of the intact functional assembly responsible for regulating and executing a site-specific DNA recombination reaction . The assembly is a 240-bp Holliday junction ( HJ ) bound specifically by 11 protein subunits . This higher-order complex is a key intermediate in the tightly regulated pathway for the excision of bacteriophage λ viral DNA out of the E . coli host chromosome , an extensively studied paradigmatic model system for the regulated rearrangement of DNA . Our results provide a structural basis for pre-existing data describing the excisive and integrative recombination pathways , and they help explain their regulation .
The rearrangement of DNA , either by homologous or site-specific recombination and transposition , is a fundamental feature of evolution , genetic variation , and gene regulation; among such pathways , the integration and excision of bacteriophage λ into and out of the E . coli chromosome is one of the most thoroughly characterized . Nevertheless , even for this pathway , it has not been possible until now to visualize the entire functional structure of the recombinogenic complex , despite the wealth of genetic , biochemical , functional and structural information accumulated by many laboratories over the past 50 years . The site-specific recombinase ( Int ) encoded by bacteriophage λ is the archetypical member of the tyrosine recombinase family , whose members carry out such diverse functions as chromosome segregation , chromosome copy number control , gene expression , conjugative transposition , gene dissemination , and viral integration and excision ( Craig et al . , 2015; Jayaram et al . , 2015; Landy , 2015 ) . All tyrosine recombinases use the same isoenergetic phosphoryl transfer chemistry and sequential strand exchange mechanism to execute DNA rearrangements via the formation and resolution of a transient four-way DNA junction , or Holliday junction ( HJ ) , recombination intermediate ( Hsu and Landy , 1984; Kitts and Nash , 1988a ) . Although these junctions are common intermediates in many different pathways responsible for rearranging genetic material in evolution , heredity , and gene expression , they have been particularly well-studied in the tyrosine recombinase family of reactions ( Gopaul et al . , 1998; Jayaram et al . , 2015; Van Duyne , 2015 ) . In contrast to two other well-studied and highly exploited family members , Cre and FLP , where the entire bidirectional recombination reaction is executed by a recombinase tetramer ( Jayaram et al . , 2015; Van Duyne , 2015 ) , Int carries out two opposing reactions that are both highly directional and tightly regulated . Each reaction requires the assembly of a unique 400 kDa multi-protein complex using accessory DNA-bending proteins and an overlapping ensemble of accessory binding sites . In response to a variety of physiological and environmental signals one complex is assembled to carry out unidirectional integrative recombination between “att” sites on the viral and bacterial chromosomes , while the other complex is formed to carry out unidirectional excisive recombination between att sites flanking the integrated viral chromosome ( [Seah et al . , 2014; Tong et al . , 2014]; see Figures 1 and 2 ) . 10 . 7554/eLife . 14313 . 003Figure 1 . Integration and excision of the λ viral chromosome into and out of the bacterial host chromosome is highly directional and tightly regulated . ( A ) Formation of the integrated prophage . In those infected cells where the decision has been made not to replicate the viral DNA , the circularized supercoiled viral DNA ( black lines ) is inserted into the bacterial chromosome ( curly red lines ) at a specific site ( called attB or BOB’ ) located between the gal and bio genes ( Campbell , 1963 ) . This insertion , integrative recombination , involves cutting the viral DNA at a specific site ( called attP or P’C’OCP ) and joining these cut ends to the cut ends of the bacterial chromosome in attB . The cutting , recombining , and resealing of viral and bacterial DNA generates new DNA sequences forming the junctions between bacterial and integrated viral DNA . These junction sequences , called attL ( BOC’P’ ) and attR ( PCOB’ ) on the left and right respectively , are themselves substrates for the cutting , recombining , and resealing reactions that will , sometime in the future , remove ( excise ) the viral DNA from the host chromosome and thereby regenerate the viral attP and bacterial attB sequences . The integrated provirus chromosome is stably inherited , with almost all of its genes repressed , for many bacterial generations . Upon instigation by the appropriate physiological signals , the viral chromosome is excised , replicated , and inserted into viral particles which are released into the environment . The integrative reaction requires the virally-encoded integrase protein ( Int ) and the bacterial accessory protein Integration Host Factor ( IHF ) . The excisive reaction additionally requires the virally encoded accessory protein Xis ( which also inhibits the integrative reactions ) . Both reactions are stimulated by the bacterial accessory protein Fis . ( See Panel C for DNA binding sites of all the proteins and their respective roles in each reaction ) . Both reactions proceed through a four-way DNA junction called a Holliday junction ( HJ ) ( see Panel B for details of these DNA strand exchanges ) ( reviewed in [Landy , 2015] ) . ( B ) The Holliday junction intermediate . Cutting , recombining , and resealing DNA during recombination proceeds by two pairs of sequential single-strand DNA exchanges that are staggered by seven base pairs that are identical in all four att sites; they are referred to as the 'overlap' region ( O ) . The molecular details of this recombination are common to all reactions catalyzed by the large family of tyrosine recombinases ( except for the size of the O regions and the order of strand exchanges ) , as first characterized for λ Int , Cre , Flp , and XerC/D ( reviewed in [Van Duyne , 2015] ) . It proceeds in the absence of exogenous energy via the formation of high-energy covalent 3’phospho-tyrosine intermediates in the active site of each Int protein . Illustrated here is the pathway for integrative recombination; it would be identical for the excisive reaction but the substrates ( left panel ) would be attL and attR , leading to attP and attB products ( right panel ) . Viral and bacterial DNAs are denoted as in panel A . ( i ) The attP ( C’OC ) and attB ( BOB’ ) sites are aligned anti-parallel with respect to their identical overlap sequences . ( ii ) The first pair of exchanges , always at the C ( green ) and B ( blue ) core sites , is initiated by formation of 3’ phospho-tyrosine linkages ( with Tyr342 ) and 5’ OH termini . ( iii ) The 5’ OH terminus generated by the cleavage at C attacks the phospho-tyrosine linkage at B to regenerate a new B–C’ strand ( orange arrow head ) . Concomitantly , the 5’ OH from the cleavage at C attacks the phospho-tyrosine linkage at B to regenerate a new B'–C strand ( orange arrow head ) . Together , this pair of single strand exchanges , at one boundary of the overlap region , forms the four-way DNA ( HJ ) intermediate . ( iv ) A similar pair of single strand cleavage , exchange , and resealing reactions is executed by the Ints at C’ ( brown ) and B’ ( purple ) on the other side of the overlap region . ( v ) As a result of the two sequential pairs of single-strand exchanges , all four DNA strands have new junction sequences and the HJ is resolved to recombinant products attL ( C’OB ) and attR ( COB’ ) . C ) Additional complexity , in the P’ and P arms , confers regulation and directionality to the λ Int reaction . In contrast to the 'simple' Cre and Flp tyrosine recombinases , λ Int ( and more than a thousand viral cousins in the public data bases ) has two DNA binding domains , as shown in Figure 2 . The carboxy-terminal domain of Int ( CTD ) binds at the four sites of DNA cleavage ( called core-type sites; blue boxes ) and catalyzes the chemistry of DNA cleavage and rejoining; this domain and the core-type sites ( B , B’ , C’ and C ) are analogous , and very similar , to the Cre and Flp enzymes and their respective DNA targets sites ( see panel B ) . In λ Int an additional small DNA binding domain at the amino terminus ( NTD ) binds with high affinity to a different family of DNA sequences ( arm-type sites ) ( green boxes ) distant from the sites of DNA cleavage and located in the P’ and P arms of viral DNA ( adjacent to C’ and C , respectively ) . To enable the ( 'hetero-bivalent' ) Int to bind simultaneously to both of its DNA targets the core- and arm-type DNA sites are interposed by binding sites for the accessory DNA bending proteins , IHF ( yellow boxes , H1 , H2 , and H’ ) , Xis ( gold boxes , X1 , X1 . 5 , X2 ) , and Fis ( magenta ) . The DNA bending proteins bring the core- and arm-type sites into close proximity and also serve as essential elements in forming the large multi-protein recombination complexes . Two distinct but overlapping ensembles of binding sites are employed ( solid boxes ) to generate either the integrative or excisive recombinogenic complexes ( reviewed in [Landy , 2015] ) . ( For the patterns of Int bridging between core- and arm-type sites in each of the complexes see Figure 2 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14313 . 00310 . 7554/eLife . 14313 . 004Figure 2 . Patterns of specific Int bridges between core- and arm-type DNA binding sites . ( A ) The amino-terminal domain ( NTD ) of Int ( yellow ) binds to one of the five arm-type DNA sites ( green ) via a helix-turn-helix recognition motif . It is joined via a flexible 10 residue linker region to the carboxy-terminal domain ( CTD ) ( light blue ) , which is composed of two smaller domains ( the core-binding ( CB ) and catalytic domains ) joined by a 12 residue linker . The CTD binds to core-type DNA sites ( blue ) via a C-clamp motif; the distal domain of the C-clamp contains the nucleophile Tyr-342 ( red ) and those residues comprising the catalytic site for DNA cleavage and ligation . The CTD is analogous , and very similar in structure and function , to the well-studied monovalent recombinases , Cre , Flp , and XerC/D . ( B ) Integrative recombination depends upon four Int bridges , as determined previously ( Tong et al . , 2014 ) . In the attP and attB substrates ( left-most panel ) only two of the four bridges are formed prior to synapsis; the accessory DNA bending proteins have been omitted for clarity ( see Figure 1C ) . The coloring scheme for the Ints bound to different arm-type sites is the same as that used in all of the following figures . ( C ) Excisive recombination involves a different pattern of Int bridges ( Tong et al . , 2014 ) and , as shown in Figure 1C , a different ensemble of DNA bending proteins . The HJ intermediate of this reaction ( brackets ) was purified and used for single particle cryo-electron microscopy , as described in the text . DOI: http://dx . doi . org/10 . 7554/eLife . 14313 . 004 Int is a heterobivalent DNA binding protein that binds to high affinity 'arm-type' DNA sites via its small amino-terminal domain ( NTD ) and more weakly to 'core-type' DNA sites , where DNA cleavage and ligation take place . Int binds to core sites via a central core binding domain ( CB ) and a C-terminal catalytic domain ( CAT ) ; the latter two domains are referred to here as the CTD ( see Figure 2; Moitoso et al . , 1989 ) . The Int subunits within a recombinogenic complex bind and bridge arm- and core-type DNA sites in patterns determined by the accessory DNA bending proteins IHF , Xis , and Fis . Differential occupancy of the 16 protein binding sites on the 240 bp att site generates two overlapping ensembles that differentiate the integrative and excisive recombination pathways ( [Seah et al . , 2014; Tong et al . , 2014]; ( see Figure 1 ) . Using techniques described previously ( Figure 3 ) , we trapped and purified the multi-protein HJ complex of excisive recombination and determined its structure at 11 Å resolution using single particle electron cryo-microscopy ( cryo-EM ) . We then used the EM density , known protein-DNA crystal and NMR structures , and the known Int-mediated arm-core bridges to build an atomic model of the intact functional complex . The experimental structure for the excisive complex then allowed us to construct a model for the integrative complex . The results provide a structural basis for understanding how these complex DNA recombination machines function and how they are so tightly regulated . 10 . 7554/eLife . 14313 . 005Figure 3 . Isolation and single particle electron cryo-microscopy of the Holliday junction recombination intermediate . ( A ) Native polyacrylamide gel electrophorograms of the recombination reactions prior to and after sucrose gradient purification . Recombination reactions between attL and attR partners designed to trap the HJ intermediate ( see Figure 2 ) were treated with 0 . 0035% glutaraldehyde for 10 min , quenched with 0 . 3 M glycine and concentrated by ultrafiltration . The concentrated 100 μL samples were loaded onto a 2 mL sucrose gradient ( 22%–40% sucrose , 10 mM Tris , pH 8 , 1 M Betaine ) and centrifuged for 16 hr at 4°C . After examining each of the 100 µl fractions by native PAGE the purest fractions were pooled , concentrated , and subjected to a second round of centrifugation . The samples were either immediately frozen in liquid nitrogen for storage or used for plunging grids ( see Materials and methods for details ) . ( B ) Electron micrograph of ice-embedded Holliday junction complexes . Some complexes used for further processing are encircled . Scale bar = 300 Å . ( C ) Some 2D class averages obtained from 66 , 033 particles that were selected from 1359 micrographs . Different views of the complex are apparent . Some of the views clearly show loops , other crosses . ( D ) Different views of the 3D reconstruction of the Holliday junction complex at 11 Å resolution calculated from 10 , 956 particles and sharpened with a B-factor of -2500 Å2 ( see Materials and methods for details ) . Scale bar = 100 Å . DOI: http://dx . doi . org/10 . 7554/eLife . 14313 . 005
Stable HJ complexes were trapped by in vitro recombination between attL and attR partners whose respective seven bp overlap sequences were designed to generate a fully paired HJ overlap sequence that would create mismatched base pairs upon reversal or progression of the recombination reaction ( Matovina et al . , 2010; Tong et al . , 2014 ) . Following brief crosslinking with 0 . 0035% glutaraldehyde , the complexes were purified by two cycles of sucrose gradient centrifugation and either frozen in liquid nitrogen or immediately prepared for cryo-EM . The complexes were analyzed for purity and homogeneity by native gel electrophoresis ( Figure 3A ) and negative staining electron microscopy . For details of the preparation and characterization of the complexes , see ( Tong et al . , 2014 ) . We selected 66 , 033 particles from 1359 images ( for example , Figure 3B ) and obtained two-dimensional ( 2D ) class averages using the ISAC procedure ( Yang et al . , 2012 ) after pixel binning ( effective pixel size 5 . 6 Å ) . Following visual inspection of the class averages , we selected 52 classes that showed clear features , such as DNA loops and crosses ( Figure 3C ) , and calculated initial maps using EMAN2 ( Tang et al . , 2007 ) . The highest-scoring map was used to initialize particle alignment and three-dimensional ( 3D ) classification in FREALIGN ( Lyumkis et al . , 2013 ) . All particles were aligned against the EMAN2 map using data limited to 60 Å resolution . Starting with reconstructions calculated from six randomly selected subsets of the data , particles were 3D-classified into six classes and their alignments refined in 40 iterations in which the resolution limit was increased in regular steps to a final limit of 20 Å . The reconstruction with the highest average particle score was then used to initialize a second round of alignment and 3D classification , starting again with a resolution limit of 60 Å and increasing the limit in steps to a final 20 Å . Due to the substantial heterogeneity of the sample – only 17% of the particles ended up in the final reconstruction ( Figure 3D ) – we performed six such refinement/classification rounds , always initializing the next round with the reconstruction of the previous round that had the highest average particle score . The highest scoring reconstructions from each round , as well as the initial and final maps are shown in Figure 4A . Lower-scoring reconstructions showed some of the features of the highest-scoring reconstruction but were missing density in the DNA loop and/or the Int tetramer regions . This suggests that the heterogeneity is the result of particle distortions and instabilities that often arise during the grid preparation procedure , for example when particles interact with the air/water interface ( Cheng et al . , 2015 ) . For the final map , we performed an additional 50 iterations of refinement and 3D classification using a resolution limit of 18 Å . The last three iterations used data with an effective pixel size of 2 . 8 Å . The best reconstruction contained 10 , 956 particles and reached a resolution of about 11 Å ( Figure 4B and C ) . 10 . 7554/eLife . 14313 . 006Figure 4 . Electron cryo-microscopy refinement and 3D classification . ( A ) Different stages of refinement and classification using FREALIGN ( Lyumkis et al . , 2013 ) . Reconstructions at each stage are shown . The initial map was generated from 2D class averages ( Figure 3C ) using EMAN2’s initial model procedure ( Tang et al . , 2007 ) . Subsequent maps resulted from several rounds of refinement and 3D classification in FREALIGN . The best-scoring reconstruction from each round was used to initialize the next round . The final reconstruction contained 10 , 956 particles . ( B ) Resolution estimation of the final reconstruction using the Fourier Shell Correlation ( FSC ) method . The red curve ( labeled FSC ) was calculated using the two reconstructions obtained from half the data each and masked using a tight mask that left a margin around the reconstructed density of about 15 Å . The blue curve ( FSC randomized ) was calculated using a second set of reconstructions obtained from data phase-randomized beyond a resolution of 18 Å . This curve was used to correct the first curve for any masking effects ( Chen et al . , 2013 ) , yielding the final green curve ( FSC true ) . The black curve ( FSC against model ) was calculated between the final map and a map derived from the atomic model presented in this study . The vertical line indicates 18 Å resolution , the limit beyond which no data was used for refinement and classification . The horizontal lines indicate values of 0 . 143 , the FSC threshold used to assess the resolution of the map ( Rosenthal and Henderson , 2003 ) , and 0 . 5 , the corresponding FSC threshold when comparing an experimental map against a noise-free model . Both ‘FSC true’ and ‘FSC against model’ suggest a resolution of about 11 Å . The FSC data are shown in Figure 4—source data 1 and 2 . ( C ) Local resolution map calculated using ResMap ( Kucukelbir et al . , 2014 ) . The resolution is fairly uniformly indicated as 11 Å , except for small regions within the Int tetramer and at one of the IHF heterodimer sites , indicating lower resolution and higher structural variability in these regions . DOI: http://dx . doi . org/10 . 7554/eLife . 14313 . 00610 . 7554/eLife . 14313 . 007Figure 4—source data 1 . FSC data in Excel format . FSC , randomized FSC , and true FSC are tabulated as a function of resolution . DOI: http://dx . doi . org/10 . 7554/eLife . 14313 . 00710 . 7554/eLife . 14313 . 008Figure 4—source data 2 . FSC data in csv format . FSC , randomized FSC , and true FSC are tabulated as a function of resolution . DOI: http://dx . doi . org/10 . 7554/eLife . 14313 . 008 We began by placing a complex containing the core-binding and catalytic domains of an integrase tetramer bound to core site HJ DNA into the EM envelope . This core complex was derived from a crystal structure of full-length integrase bound to HJ DNA and short oligonucleotides containing arm site sequences ( Biswas et al . , 2005 ) . The assembly fit well as a rigid body , indicating that the EM and crystal structures are similar in this region . Structural models for the protein-bound P and P' arms were assembled as previously described ( Seah et al . , 2014 ) , except that the Fis dimer ( which is not present in the purified HJ complex used here ) was omitted from the P arm model . Details of the arm constructions are provided in Materials and methods . The pre-assembled P and P' arms were spliced onto the core Int-HJ complex while maintaining standard B-DNA twist and fit into the EM envelope by applying small DNA roll angles ( ± 1° ) in the regions outside of the IHF , Xis , and Int-NTD footprints . Flexibility in the IHF-induced DNA bends at H' and H2 was also allowed by testing small in-plane and out of plane deviations from the IHF/H'-DNA crystal structure ( Rice et al . , 1996 ) . Two regions in the resulting structure could not be confidently modeled based on existing crystal structures . The linkers connecting the NTD and CB domains of Int adopt distinct and partially extended conformations , a situation quite different from the compact and symmetric arrangement of linkers observed in the crystal structures of Int bound to short oligonucleotide DNAs ( Biswas et al . , 2005 ) . In addition , the central base-pairs of the HJ DNA were not well-defined in the Int-HJ crystal structure . To obtain low-energy conformations for these regions within the EM envelope and to regularize the stereochemistry of the DNA splice junctions and sites of bending , we performed molecular dynamics refinement using dynamic elastic network ( DEN ) restraints ( Schröder et al . , 2007 ) . The protein•DNA sub-complexes derived from crystal structures were tightly restrained to adopt their experimental conformations , whereas the Int NTD-CB linkers and central HJ DNA base-pairs were allowed flexibility . To restrain flexible regions of the model to remain within the EM envelope , we included an energy term for agreement with the Fourier coefficients of the EM density . As shown in Figure 5 , the final structure agrees well with the EM envelope . There are no major features lacking density and only one small region of density that was not fit by the model ( discussed below ) . The P and P' arm trajectories are particularly well defined by the EM envelope ( Figure 5A ) , as is the core Int-HJ tetramer ( Figure 5B ) . 10 . 7554/eLife . 14313 . 009Figure 5 . Model and reconstructed density for the λ excision complex . ( A ) View from the 'top' of the complex , illustrating the IHF-mediated bends at H2 and H' and the close , parallel trajectory of the P and P' arms . Unmodeled density that could be occupied by Int-C NTD is indicated by an arrow . ( B ) View from the bottom of the complex , highlighting the catalytic domains of the Int tetramer . The Int subunits are labeled according to the core half-sites where they are bound ( B , B' , C , and C' ) . IHF heterodimers and Xis subunits are indicated . In the schematic representations shown below the structural models , the DNA strands near the junction centers are omitted for clarity . DOI: http://dx . doi . org/10 . 7554/eLife . 14313 . 009 The limited resolution of the cryo-EM map does not permit construction of an atomic model with accurate detail at the level of side chains and nucleotides . Consequently , the model coordinates deposited for the excisive HJ complex specify only the positions of protein domains and the paths of the DNA duplex arms . Inter-domain linkers for the Int-C' , Int-B , and Int-B' subunits are provided in arbitrary conformations in the final model to improve visualization of the structure and to facilitate modeling in future experiments; no attempt was made to predict the structures of these linkers . Three views of the excisive complex structure are shown in Figure 6 and a video illustrating several views of the complex is available as Video 1 . At the center of the excisive complex an integrase tetramer is bound to a core-site HJ , where strand exchange has taken place between the B half-site of attL and the C half-site of attR ( see Figures 1B and 2C ) . We refer to the integrase subunits by the core half-sites to which they are bound . Int-C , for example , is the integrase subunit bound to the C half-site . The most striking features of the complex are the tight bends of the P and P' arms as they emerge from the C and C' core half-sites . IHF heterodimers bend the P arm at the H2 site and the P' arm at the H' site , in each case re-directing the DNA back towards the Int tetramer . For P' , the IHF bend is sufficient for the P'1 and P'2 arm-binding sites to be engaged by the Int subunits bound at the C' and B core sites , respectively ( Figure 6B ) . These are two of the three Int bridging interactions identified using biochemical and genetic approaches ( Tong et al . , 2014 ) . 10 . 7554/eLife . 14313 . 010Figure 6 . The λ excision complex structure . ( A ) View towards the attR-derived half of the complex , where the path of the P arm can be seen . A sharp IHF ( yellow ) directed bend at the H2 site , combined with the cooperative binding and bending by three Xis molecules ( gold ) and the Int subunit bridging between the B' and P2 sites ( magenta ) , redirects the P arm over the top of the Int tetramer . The Int-C NTD is not present in the model because it could not be unambiguously fit into EM density ( approximate location is indicated in the corresponding cartoon ) . B ) View towards the attL-derived half of the complex , where IHF bound at the H' site sharply bends the P' arm and enables Int subunits to bridge between the C' and P'1 sites ( brown ) and between the B and P'2 sites ( blue ) . ( C ) View from the top of the excision complex , where the P'-arm ( from attL ) and P-arm ( from attR ) run nearly parallel and close to one another . An intimate interface is formed between the Xis subunits and the Int-C' and Int-B NTDs . The unoccupied , but accessible binding site for Fis is indicated . In the schematic views of the excision complex shown below the structural models , the DNA strands near the junction center are omitted for clarity . For reference , the brackets labeled attP/attB indicate the relative positions of the excision products . See Video 1 for multiple views of the excisive complex structure and isolation of individual subunits . DOI: http://dx . doi . org/10 . 7554/eLife . 14313 . 01010 . 7554/eLife . 14313 . 011Video 1 . The bacteriophage λ excisive HJ intermediate . Multiple views of the excisive complex are shown . The complex is stripped of all proteins to illustrate the path of the DNA arms as they loop around the 4-way junction . IHF , Xis , and Int subunits are added individually to the DNA to help visualize their structural and functional roles . The assembly sequence shown is for visualization only; nothing is implied regarding the actual order of addition or assembly during λ excision . DOI: http://dx . doi . org/10 . 7554/eLife . 14313 . 011 For the P arm , the IHF-mediated bend at H2 is phased differently with respect to the core tetramer compared to the bend at H' . Rather than being directed towards the NTDs of Int , the H2 bend causes the P arm to run alongside the Int tetramer . Three Xis subunits provide a second change in direction of the P arm , directing it across the top of the Int tetramer ( Figure 6A ) . As a result of the IHF and Xis-induced bends , the P2 site is positioned where it can be engaged by the Int subunit bound to the B' core site ( Figure 6A and C ) . The B'-P2 bridge is the third bivalent interaction required for excisive recombination ( Figure 2C ) ( Tong et al . , 2014 ) . Remarkably , the DNA bend observed in the Xis•DNA crystal structure ( Abbani et al . , 2007 ) fits well into the EM envelope , which has strong density for both the DNA and the Xis subunits . The P arm trajectory is also aided by an intrinsic A-tract sequence found in the Fis binding site . Fis was not present in the complexes analyzed by EM and as expected , there is no density corresponding to Fis at this site . We note , however that Fis can be readily accommodated by the structure ( Figure 6C ) , which fits well with the proposed role for Fis as a non-essential enhancer of recombination ( Ball and Johnson , 1991a; 1991b; Esposito and Gerard , 2003; Thompson et al . , 1987a ) . The only component of the excisive HJ complex that is not accounted for by EM density is the NTD of Int-C ( Figures 2C and 6A ) . This domain has no known function in excisive recombination , since only three arm-binding sites are used in the reaction ( Numrych et al . , 1990; Thompson et al . , 1987b; Tong et al . , 2014 ) . One plausible location is near the Int-NTD bound at P2 , where there is a small lobe of additional density present ( Figure 5A ) . It is clear , however , that the Int-C NTD is not positioned directly above its connected CB domain , since the Xis molecule bound at the X1 site occupies this space . We did not fit the Int-C NTD into the P2-associated density because the domain's orientation and position could not be clearly established . The EM structure described here provides an opportunity to re-examine the model and data recently reported in a FRET-based study of the λ excision HJ intermediate ( Seah et al . , 2014 ) . In that work , 12 DNA sites were labeled with fluorophores and 28 inter-site distances were estimated using in-gel fluorescence measurements . These distances , combined with knowledge of the Int bridging interactions and crystal structures of the component sub-complexes , were used to construct a model of the excisive HJ complex . The overall architecture of our EM structure is in broad agreement with the FRET-based model . As might be expected from the small number of structural restraints available , the FRET model is largely schematic in nature . For example , the tight , parallel paths of the P and P' arms were not predicted in the FRET model , nor was the extensive Xis-Int interface that links the two arms together ( see below ) . Indeed , the details that emerge from the EM structure provide important new insights that could not have been predicted based on distance measurements alone . The EM structure also provides an opportunity to ask if the dye positions chosen for the FRET study were appropriate and to determine how well the EM structure agrees with the estimated distances ( Table 1 ) . Six of the locations appear to have been good choices , since they are largely free from steric interference that would compromise free rotation and/or the volumes accessible to the fluorophores . The EM structure is in good agreement with all of the distance estimates involving these sites . Two of the sites , attP- ( -118 ) and attP- ( +79 ) , have serious steric conflicts based on the EM structure . The attP- ( -118 ) dye on the P arm is directed into the path of the P' arm , possibly explaining why the arms diverge in the FRET model . The dye at attP- ( +79 ) is on the P' arm and is directed into the Int-NTD bound at P2 . Thus , the distances involving these sites are expected to be compromised and indeed , we see much poorer agreement with the EM structure for the distances involving these locations . The remaining four sites used for the FRET study are located on extensions of the P and P' arms that were not included in the constructs used for EM . Thus , there is excellent agreement between the EM structure and the FRET distance estimates for those dye positions that were correctly predicted to be free from steric interference . 10 . 7554/eLife . 14313 . 012Table 1 . FRET vs . EM model distances in the λ excisive recombination complex . R ( FRET ) values were determined by Seah et al . ( 2014 ) using in-gel fluorescence measurements for dye pairs whose positions were designed in the absence of a model for the λ excision complex architecture . Dyes positioned at sites 4 , 5 , 6 , 7 , 11 , and 12 are predicted to be largely free from steric interference based on the EM structure . The FRET pairs involving these positions are in the first group below . Dyes positioned at site 3 ( second group ) and site 8 ( third group ) are predicted to have steric conflicts between the dye and the excision complex . Four additional sites ( 1 , 2 , 9 , & 10 ) used in the study are on extended P and P' arms not present in the construct used for EM studies . See Seah et al . ( 2014 ) for descriptions of the site position nomenclature . DOI: http://dx . doi . org/10 . 7554/eLife . 14313 . 012Position iPosition jR ( FRET ) ( Å ) R ( FRET ) Model ) ( Å ) R ( EM Model ) ( Å ) 7 [P+50T]6 [P+17B]565750 . 04 [P-58B]5 [P-15T]707270 . 65 [P-15T]12 [B+17B]767778 . 84 [P-58B]6 [P+17B]767067 . 95 [P-15T]6 [P+17B]898081 . 16 [P+17B]11 [B-15T]938185 . 37 [P+50T]4 [P-58B]99125108 . 911 [B-15T]12 [B+17B]1038682 . 53 [P-118T]12 [B+17B]806964 . 13 [P-118T]5 [P-15T]827795 . 33 [P-118T]11 [B-15T]99108100 . 08 [P+79B]3 [P-118T]565641 . 48 [P+79B]11 [B-15T]746575 . 28 [P+79B]5 [P-15T]9610981 . 38 [P+79B]6 [P+17B]9610785 . 48 [P+79B]4 [P-58B]108158124 . 88 [P+79B]12 [B+17B]1167368 . 2 In crystal structures of Int complexes containing simplified oligonucleotide arm-binding sites , the DNA bound by Int NTDs lies parallel to the Int tetramer and the Int NTD-CB linkers form a compact network of interactions ( Biswas et al . , 2005 ) . In contrast , the P' arm in the EM structure follows a relatively straight path after the H' bend , resulting in an upwards trajectory as it passes over the Int tetramer ( Figure 6B ) . Consequently , Int-C' and Int-B linkers must adopt extended and distinct conformations in order to bridge the P' arm and the Int CB domains . Int bridge formation at P'1 and P'2 of attL therefore does not require bending of the P' arm after the IHF-directed U-turn at H' . Instead , Int uses the inherent flexibility present in the NTD-CB linker regions to reach the P'1 and P'2 binding sites . A primary function of Xis is to bend the P arm to stabilize a synapsis-competent attR complex . The EM structure reveals that Xis also mediates a substantial protein-protein interface between the P and P' arms . The Xis molecules bound at the X1 and X1 . 5 sites of the P arm are packed against the Int-B and Int-C' NTDs bound at the P'2 and P'1 sites , respectively ( Figure 6C and Video 1 ) . The Xis1 . 5-Int-C' interface appears to be particularly extensive , although at the resolution of the EM density , we cannot visualize the conformations or roles of individual side chains . These Xis-Int interfaces are presumably important , given the close approach and parallel trajectory of the P and P' arms in the excisive complex . The identification of a Xis-mediated interface between the P and P' arms provides an explanation for how synapsis is facilitated in the excisive pathway . Int binds only weakly to the core sites of attL and attR and the Int-bound core sites do not form stable synaptic complexes that undergo recombination ( Bushman et al . , 1985 ) . What provides the interactions that stabilize the attL x attR synaptic complex to initiate excisive recombination ? Crystal structures of Int bound to core and a symmetrized mimic of arm DNA fragments suggested that the Int NTDs , along with their associated linkers , could play this role ( Biswas et al . , 2005 ) . However , earlier experiments with a Cre-Int chimeric recombinase ( Warren et al . , 2008 ) and the present EM structure indicate that this is unlikely . The three Int NTDs in the excisive complex are arranged in a linear manner , where the NTD bound at P2 makes no contact with the NTDs bound at P'1 or P'2 ( Figure 6C ) . Instead , Xis-NTD , and possibly Xis-linker interactions provide the additional glue required to hold attL and attR together and allow the initial strand exchange to take place . There are 18 residues at the C-terminus of Xis that were missing in the Xis•DNA crystal structures ( Abbani et al . , 2007; Sam et al . , 2004 ) and are not included in our excisive complex model . The C-termini of the Xis domains bound at the X1 . 5 and X2 sites extend into the gap between the P and P' arms ( i . e . , towards the viewer in Figure 6C ) . Six of the eighteen missing residues are arginine or lysine , suggesting that the basic C-terminal tails of Xis could interact with P and/or P' arm DNA , further stabilizing a synaptic interface . The C-terminus of Xis bound at X1 is instead directed at the Int-B NTD bound at P2 , consistent with the observed cooperativity of Xis and Int binding to attR arm DNA ( Bushman et al . , 1985; Numrych et al . , 1992; Sarkar et al . , 2002 ) . Thus , the EM structure reveals and/or explains three functional roles for Xis in excisive recombination: i ) promoting Int NTD binding at the weak P2 site; ii ) bending the P arm to place P2 in position for cooperative Int binding and formation of a synapsis-competent attR; iii ) mediating an Xis•Int interface between the P and P' arms that facilitates attL x attR synapsis . Although Fis is not required for excisive recombination , it does stimulate the reaction when Xis is limiting ( Thompson et al . , 1987a ) . The views in Figure 6A and C provide an explanation for this stimulation . The A-tract present in the Fis site , Xis binding at X1 , X1 . 5 , and X2 , and Int binding at P2 all contribute to the formation of a properly assembled P arm , and therefore a properly assembled attR at the start of the excision pathway . Fis binding stabilizes the A-tract bend in attR ( Stella et al . , 2010 ) and facilitates Xis binding ( Papagiannis et al . , 2007 ) , providing additional sources of cooperativity for P arm assembly that would enhance the activity of Xis at lower concentrations . Structures representing reaction intermediates in the Cre and λ systems have led to a model for understanding how strand exchange is coordinated ( Aihara et al . , 2003; Biswas et al . , 2005; Gopaul et al . , 1998; Guo et al . , 1997 ) . The essential geometric features of this model for λ integrase are shown in Figure 7 . The scissile phosphodiesters , which flank the 7-bp crossover sequence located at the center of the core sites , form a parallelogram in the HJ intermediate . The sites that are active for cleavage form the closer diagonal pair; those that are inactive are farther apart . Isomerization of the HJ intermediate between a top-strand cleavage configuration and a bottom-strand cleavage configuration involves exchanging angles within the scissile phosphate parallelogram ( Figure 7A and B ) . In the Cre system , isomerization is coupled to changes in the angles between the core site duplexes that converge at the center of the junction . Changes in the interfaces between catalytic domain subunits that result from isomerization are responsible for activation/deactivation of active site pairs ( Van Duyne , 2001 ) . 10 . 7554/eLife . 14313 . 013Figure 7 . Model for isomerization of HJ intermediate . The core HJ complex is fit to the reconstructed density , based on structure 1Z1G ( Protein Data Bank ) ( Biswas et al . , 2005 ) . ( A ) The HJ isomer corresponding to a top-strand cleavage configuration , where the Int subunits bound at the B and C half-sites are activated for cleavage and strand exchange . B ) Alternative isomer , where Int-B' and Int-C' are activated for cleavage of the bottom strands . Scissile phosphates ( red spheres ) are closer together in the active subunits compared to the inactive ones . Arrows indicate the activated phosphodiesters . The interconversion between ( A ) and ( B ) involves migration of the branch point by one base-pair , with only minimal changes at the ends of the duplex arms . An animation that illustrates the HJ isomerization is provided as Video 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 14313 . 013 The core HJ isomer that we fit in the EM structure corresponds to a bottom strand cleavage configuration , where the complex is poised to carry out strand exchange to form attP and attB products ( Figure 7B ) . We considered how the structure of the alternative isomer , corresponding to a configuration where the top strands have just been exchanged to generate the HJ from the synapsed attL and attR partners , might differ in this complex ( Figure 7A ) . Although there are significant changes near the center of the core complex where the HJ branch migrates by one bp , the ends of the duplexes remain essentially in the same position , with only small torsional changes ( Figure 7 and Video 2 ) . This finding is consistent with the pseudo-fourfold symmetry of the core HJ complex ( Biswas et al . , 2005 ) and the local four-fold symmetry of the EM density corresponding to this region ( Figure 5B ) . Thus , at the resolution of our EM data , we cannot distinguish between HJ isomers and the complexes analyzed may be a mixture of the two isomeric forms . 10 . 7554/eLife . 14313 . 014Video 2 . Isomerization model for the λ HJ intermediate . The core HJ is shown switching between the isomers described in Figure 7 . This type of isomerization could occur within the core of the recombination complex , without large changes in positions of the attached arms . DOI: http://dx . doi . org/10 . 7554/eLife . 14313 . 014 The nearly isosteric model for HJ isomerization shown in Figure 7 has important implications for considering how the P and P' arms of the excisive complex might change during recombination . In the absence of structural models , it has been difficult to rationalize how the arms could accommodate , or perhaps even direct the changes in quaternary structure expected from a Cre-like isomerization . In contrast to Cre , however , the core HJ structure observed in the EM and crystal structures of the λ intermediate indicates that isomerization could occur with relatively little impact on the flanking DNA segments . We suggest that small changes in position and torsion of the C and C' core sites can be accommodated by the in-plane and out of plane bending modes of IHF , with little impact on the Xis•Int interface assembled between the P and P' arms as they cross the Int tetramer . In other words , the EM structure implies that isomerization of the HJ intermediate can occur in the absence of large changes in quaternary structure and therefore does not require coupling to large changes in the P and P' arm positions . This type of overall isosteric isomerization may be facilitated by the odd-numbered 7-bp crossover region that is found in essentially all tyrosine integrases , e . g . ( Hakimi and Scocca , 1996; Kolot and Yagil , 1994; Peña et al . , 1996; Williams , 2002 ) . A stereochemical explanation for how a one-bp branch migration accomplishes this will be an interesting topic for future studies . The integrative HJ intermediate is formed when a complex containing attP DNA , an Int tetramer , and three IHF heterodimers captures the attB site in the host genome and carries out top strand exchange between the core DNA sequences ( Figure 1 ) . We previously proposed a model for this complex based on core-arm site bridge data and insights derived from the FRET-based excisive complex model discussed above ( Seah et al . , 2014 ) . The EM structure of the excisive HJ complex provides an opportunity to revise the integrative model in the context of a well-defined molecular architecture and correctly positioned P and P' arms for the excision complex . The new integrative HJ model , constructed using a similar procedure to that previously described , is shown in Figure 8 . In both the integrative and excisive complexes , IHF bending at the H' site directs the P' arm over the CB domains of the Int tetramer . In the integrative complex , the B'-NTD binds at P'3 , since Xis is not present to position the P2 site to bind to this domain . A striking aspect of the new integrative complex model is that the P' arm trajectory requires no positional changes in order to readily accommodate binding of the C' , C , and B' NTDs of Int ( Figure 8A ) . Thus , the P' arm position and trajectory appear to be invariant in excisive and integrative recombination . 10 . 7554/eLife . 14313 . 015Figure 8 . Model of the λ integration complex . ( A ) View towards the B'–C face of the complex , where the trajectory of the P arm can be seen . IHF bending at the H2 site , combined with an A-tract bend at the Fis site and additional bending provided by negative supercoiling , directs the P arm over the top of the Int tetramer and over the P' arm . IHF-bending at the H1 site re-directs the P arm down along the opposite face of the tetramer . ( B ) View towards the C'–B face of the complex , showing Int-B bridging between the B core site and the P1 site . The catalytic and CB domains of Int-B wrap around the opposite face of the attB helix ( with respect to the Int bound at the B' site ) . This arrangement , which is facilitated by the flexible P1 tether , must be accommodated during synapsis , as described in the text and Figure 9 . The nearly linear arrangement of NTDs bound at P'1 , P'2 , and P'3 of the P' arm can also be seen . ( C ) View from the top of the integration complex . Schematic drawings of the integrative complex are shown below the structural models , where the P arm is black and the P' arm gray . The DNA connectivity at the HJ center is omitted for clarity; consequently the junction between the P arm ( black ) and P' arm ( gray ) is not visible . The bend directions of the core sites are opposite to those shown in Figure 6 for the excision complex ( e . g . , compare the C–C' bend in ( C ) vs . Figure 6C ) , underscoring that integration and excision are not the mechanistic reverses of one another . DOI: http://dx . doi . org/10 . 7554/eLife . 14313 . 015 In the absence of Xis , the P arm is not directed towards the Int NTDs after the H2 bend . Instead , it follows an upward trajectory beyond the P' arm , where IHF binding at the H1 site reverses its course and directs it back towards the Int tetramer ( Figure 8B and C ) . This P arm trajectory places the P1 site within reach of the Int B-NTD , a high-affinity interaction that is crucial for integrative recombination ( Tong et al . , 2014 ) . Thus , the P arm crosses the P' arm to position Int-B during integration , but runs parallel to the P' arm and promotes Int-B' binding at P2 during excision . The P-P' arm crossing that occurs during integration generates a negative DNA-crossing node , a feature that provides a structural explanation for the long-standing puzzle of why the topologies of integrative and excisive recombination are different and why supercoiling is required for efficient integration ( Nash , 1975; Richet et al . , 1988 ) . Indeed , the integrative model shown in Figure 8 cannot be constructed based on IHF-phased bends at H2 and H1 alone; additional DNA-bending is required , which we suggest would occur naturally in a negatively supercoiled attP , but would be energetically unfavorable in linear substrates . The P-P' crossing also involves close DNA-DNA contacts between the arms , explaining in part the important role of polyamines such as spermidine and/or divalent cations in the integrative reaction ( Kikuchi and Nash , 1978; Nash , 1975 ) . The most remarkable feature of bacteriophage λ excisive and integrative recombination is the precisely controlled unidirectionality of these two isoenergetic reactions which , at the DNA level , are the simple reverse of each other . While many features of the distinct but overlapping protein ensembles responsible for each reaction ( and its regulation ) have been worked out , the present study finally allows us to consolidate and incorporate these features into the structure of a fully functional reaction intermediate . As discussed above , the structure reveals an important new role for Xis , clarifies the role of the Int linker regions , defines the paths of the arms as they embrace the enzymatically active core region , and suggests that HJ isomerization occurs with only minor changes in quaternary structure . Equally important , the good fit between the present structure and the large body of genetic , biochemical , and biophysical data that exists ( reviewed in [Landy , 2015] ) makes it relatively straightforward to extend the present structure to informative models for the other intermediates involved in excisive and integrative recombination . Based on the EM structure of the excisive HJ complex and the model of the integrative HJ complex described here , our current views of the mechanisms of directionality and regulation in λ site-specific recombination are summarized below and illustrated schematically in Figure 9 . 10 . 7554/eLife . 14313 . 016Figure 9 . Schematic of the λ integration ( Ia–Id ) and excision ( Ea–Ec ) pathways , based on the structural models presented here . ( Ia ) In the presence of Int and IHF ( but not Xis ) , an integration complex is assembled on supercoiled attP . We suggest that the Int-B CTD ( blue ) transiently associates with the Int-C' and Int-B' CTDs ( brown & magenta ) to form a tetrameric complex . ( Ib ) The P arm of the integration complex can swing open to test and engage candidate attB sequences and thereby deliver Int-B to the B site on the opposite face of the attB duplex , consistent with the implications of biochemical and genetic results ( Seah et al . , 2014; Tong et al . , 2014 ) and the especially high affinity of Int NTD for P1 ( Sarkar et al . , 2002 ) . ( Ic ) Stable attB binding leads to HJ formation and resolution . ( Id ) Disassembly results in unstable attR ( no bridges ) and attL ( one bridge ) complexes . ( Ea ) AttL forms a synapsis-competent complex with the two Int bridges , P’1-C’ and P’2-B . In the presence of Xis , attR forms a synapsis-competent complex , due to bending of the P-arm and formation of an Int bridge at P2 . Note that the C'-B and C-B' core sites are bent in the opposite directions compared to the integration products shown in ( Id ) . ( Eb ) Xis mediates formation of a synaptic complex , where recombination can occur . ( Ec ) Disassembly results in an unstable attP complex containing only one Int bridge . Xis competes with formation of the integration complex shown in ( Ia ) . In panels ( Ic ) and ( Eb ) , the DNA strands near the center of the complex are omitted for clarity . DOI: http://dx . doi . org/10 . 7554/eLife . 14313 . 016 The biochemical feature defining excisive and integrative recombination as two different pathways ( and not the forward and reverse of a single pathway ) is the fact that both reactions are initiated by exchange of the same ( 'top' ) strands to form their respective HJ intermediates ( Kitts and Nash , 1988b; Nunes-Düby et al . , 1987 ) . Examination of the models shows that the order of strand exchange in both pathways is determined prior to synapsis by the patterns of specific Int bridges between arm and core sites and by the IHF- and Xis-induced bends that enable those bridges . For example , the C' and B bridges to the P'1 and P'2 sites of attL require a specific bend of the B-C' core site , which in turn commits attL to top strand exchange in the synaptic complex with attR ( Figure 9Ea ) . Similarly , formation of an integration-competent attP assembly requires a specific bend direction of the C-C' core site in order to form the correct bridging interactions ( Figure 9Ia ) . We found that a stereochemically plausible attP model can be assembled if the C-C' core site is bent in the opposite direction to that shown in Figure 9Ia ( i . e . , a bottom-strand cleavage configuration ) , but the resulting complex has a ( + ) crossing node and would therefore be strongly disfavored in a negatively supercoiled substrate . We conclude that only a top strand exchange integrative complex configuration is consistent with negative supercoiling . The excisive complex structure delineates the mechanism by which Xis acts as a directionality switch between the excisive and integrative recombination pathways . During formation of a synapsis-competent attR complex , Xis enhances binding of the Int NTD at P2 through protein-protein interactions ( Sarkar et al . , 2002 ) and it enhances formation of the Int bridge between P2 and B’ by pronounced bending of the P arm . Xis interactions with the Ints bound to attL also likely play an important role in facilitating synapsis . Thus , the Xis switch functions at multiple levels involving both intramolecular ( attR ) and intermolecular ( attL-attR ) stabilization of intermediates in excisive recombination ( Figure 9Ea and Eb ) . The structural models also account for why the prophage doesn't rapidly excise upon completion of integration and why the excision reaction is not run efficiently in reverse . The attR and attL products of integration have zero and one intramolecular Int bridges , respectively , and are therefore unlikely to be stable ( Figure 9Id ) . Similarly , the attP complex formed at the end of excision has only a single Int bridge ( Figure 9Ec ) . The models also explain why the integrative reaction is inhibited in the presence of Xis ( Nash , 1975 ) : Xis-induced bending of the P arm ( Abbani et al . , 2007 ) prevents its upward trajectory in attP and thus defeats the proper positioning of the P1 site required for attB synapsis ( Figure 9Ec vs Figure 9ia ) . One of the especially interesting features suggested recently ( Seah et al . , 2014 ) and strengthened considerably by this work addresses the long-standing conundrum of how the large attP complex captures a naked attB site , whose core-type half-sites are arranged as imperfect inverted repeats ( Richet et al . , 1988 ) . Because the openings of the two attP-bound Ints , positioned to bind attB , face in opposite directions they cannot dock attB by simple collision , i . e . , the Int destined to bind the B core site must have the flexibility to wrap around the host chromosome from the opposite face . The integrative complex model shown in Figure 8 has an inherently flexible P arm which we propose can transiently swing out along with Int bound at P1 ( Figure 9Ib ) . This would provide the dynamic binding required to embrace the bacterial chromosome and clasp onto a synapsed attB ( Figure 9Ic ) , a structural feature that is consistent with genetic and biochemical results ( Tong et al . , 2014 ) and the especially high affinity of Int NTD for the P1 site ( Sarkar et al . , 2002 ) . The structure and models are also consistent with the changes in topology that occur when integrative and excisive recombination reactions are carried out within the same circular DNA molecule ( Crisona et al . , 1999 ) and with the requirement for negative supercoiling for efficient integration ( Richet et al . , 1988 ) . The highly asymmetric arrangement of the NTDs in the structure and models reported here contrasts sharply with the tightly packed complexes seen in crystal structures of HJ-Int tetramers bound to truncated and symmetrized arm-type 'consensus' oligonucleotides ( Biswas et al . , 2005 ) . Indeed , the asymmetry and inherent flexibility described here are critical elements in understanding the structural basis of the recombination mechanisms; they are both consistent with , and explanatory for , all of the extensive genetic , biochemical , and topological analyses of this recombination pathway . We suggest that , in addition to their significance for understanding regulated directionality in recombination , the results described here will also serve as exemplars for thinking about the complex regulatory machines so abundant in prokaryotic and eukaryotic biology .
The HJ complex was obtained through excisive recombination of attR and attL bubble substrates using the following conditions: 55 nM attR , 50 nM attL , 30 mM KCl , 60 mM NaCl , 0 . 005 mg/mL BSA , 25 mM HEPES at pH 7 . 5 , 150 nM IHF , 350 nM Xis and 350 nM Int were incubated at room temperature for 2 hr . The reaction mixture was then crosslinked with 0 . 0035% glutaraldehyde at room temperature for 10 min followed by quenching with 0 . 3 M glycine . The mixture was concentrated using Amicon ultracell 50K filters ( Merck Millipore , Billerica , MA ) to a volume of 100 μL and loaded onto a 2 mL sucrose gradient ( 22%–40% sucrose , 10 mM Tris , pH 8 , 1 M Betaine ) at 4°C . The gradient was run at 47 , 000 rpm using a TLS-55 rotor ( Beckman Coulter , Inc . , Brea CA ) in a TL-100 tabletop ultracentrifuge ( Beckman Coulter ) for 16 hr at 4°C . The gradient was fractionated into 100 μL slices which were examined on a 5% native PAGE to analyze the purity of the fractions . Pure fractions that were devoid of any secondary bands or aggregations were selected . They were further concentrated using Amicon ultracell 50K filters ( Merck Millipore ) to a volume of 30–40 μL and were buffer-exchanged into 10 mM Tris , 50 mM NaCl buffer using prewashed Micro Biospin P-30 columns ( Bio-Rad , Hercules , CA ) . The concentration of the buffer-exchanged sample was determined to be about 2 mg/mL by UV absorption using a Nanodrop spectrophotometer ( Thermo Scientific , Wilmington , DE ) . The sample was either immediately frozen in liquid nitrogen for storage or used for plunging grids . Details and data for characterizing the HJ complex have been described previously ( Tong et al . 2014 ) . The isolated protein-DNA HJ complex ( without crosslinking ) is stable in polyacrylamide gels and in solution for more than 12 hr at room temperature or 4C . Low levels of crosslinking ( 0 . 0035% glutaraldehyde ) are required to stabilize the complex upon dilution . The purified Holliday junction complex was checked for homogeneity by screening negatively-stained samples on a Morgagni electron microscope ( FEI , Hillsboro , Oregon ) . For cryo preparation , we applied 3 . 0 μL of complex solution to a C-flat 1 . 2/1 . 3 Cu grid ( 400 mesh ) ( Protochips , Raleigh , NC ) , which had been glow discharged at 20 mA for 30 s . Grids were plunge-frozen with a Vitrobot Mark II ( FEI ) at 85% humidity , offset -2 , blot time 7 s . Images were recorded on a Tecnai F30 electron microscope ( FEI ) operated at 300 kV and using a liquid-nitrogen cooled 626 cryo-specimen holder ( Gatan Inc . , Pleasanton , CA ) . We used the semi-automated acquisition program SerialEM ( Mastronarde , 2005 ) to record 1359 movies with a Falcon II direct detector ( FEI ) at an underfocus set between 2 . 7 – 4 . 2 μm . Each movie consisted of 25 frames , collected in a 2 s exposure of 35 . 5 electrons/Å2 . The nominal magnification was 78 , 000x , corresponding to a magnification of 100 , 000x on the detector and 1 . 4 Å pixel size on the specimen . Movie frames were aligned and summed using motioncorr ( Li et al . , 2013 ) . From 1359 images we picked a total of 66 , 033 particles using e2boxer ( Tang et al . , 2007 ) from 4x binned images ( pixel size on the specimen: 5 . 6 Å ) . We carried out 2D classification using the ISAC procedure ( Yang et al . , 2012 ) implemented in the image processing package SPARX ( Hohn et al . , 2007 ) . Image defocus was determined with CTFFIND3 ( Mindell and Grigorieff , 2003 ) . We selected 52 good class averages to calculate initial maps with EMAN2 ( e2initialmodel . py , [Tang et al . , 2007] ) . The highest scoring map was used to initialize refinement and 3D classification in FREALIGN ( Lyumkis et al . , 2013 ) , as described in the main text and Figure 4A . The resolution was limited to 18 Å throughout this refinement . The final reconstruction was calculated using 2x binned data ( pixel size on the specimen: 2 . 8 Å ) and had a resolution of about 11 Å ( Figure 4B ) , as indicated by the FSC = 0 . 143 threshold criterion ( Rosenthal and Henderson , 2003 ) . For this estimate , the two 3D reconstructions obtained from half the data and used to calculate the FSC curve were masked with a tight mask that added a margin of about 15 Å around the reconstructed density of the Holliday junction complex . To assess how the mask affected the FSC curve , the 3D reconstructions were recalculated with particle images with randomized phases beyond 18 Å resolution and used to correct the FSC curve obtained from the masked reconstructions ( Chen et al . , 2013 ) . The resolution of about 11 Å was corroborated by calculating the FSC between the map and atomic model presented in this study ( Figure 4B ) , and by the local resolution map ( Kucukelbir et al . , 2014 ) ; Figure 4C . The final map was filtered at 11 Å resolution and sharpened with a B-factor of -2500 Å2 . Source data for the FSC curves shown in Figure 4B are given in Figure 4—source data 1 and Figure 4—source data 2 . Construction of an excisive HJ complex model began with a core complex derived from the 3 . 8 Å crystal structure of full-length Int bound to unlinked HJ and arm DNA fragments ( Biswas et al . , 2005 ) . The core complex consists of the catalytic and CB domains of Int ( residues 75–356 ) and a 4-way DNA junction containing the C , C' , B , and B' core half-sites . A P' arm complex containing IHF bound at H' and Int NTDs ( residues 1–55 ) bound at P'1 and P'2 was constructed essentially as described ( Seah et al . , 2014 ) , except that the NTD•DNA segments were derived from the NMR structure of an NTD•P'2 DNA complex ( Fadeev et al . , 2009 ) . A P arm complex was also constructed as previously described ( Seah et al . , 2014 ) , except for the use of the NMR-based model for the NTD•P2 segment . The assembled P' arm was joined to the C' half-site and the P arm was joined to the C half-site to generate the initial excisive complex model . During all construction steps , DNA segments were joined by superimposing the flanking duplexes on an idealized B-DNA splint fragment , ensuring smooth continuity of DNA twist . DNA segments were extended by superposition of idealized B-DNA fragments of the appropriate length , using a similar overlap . The initial complex was fit manually into EM density as a rigid body . Excellent agreement with the core complex was readily obtained , but the P and P' arms were not well-positioned in the envelope . The P and P' arms were fit into density by iterative testing of small ( ± 1° ) roll angles in the protein-free regions and by varying the IHF-mediated DNA bends . Changes in bending and torsion of IHF•DNA complexes ( <10° ) were made by rotating the DNA duplexes about the existing kinks ( in-plane ) or rotating the DNA arms about an axis connecting the two kink positions ( torsion ) . The phosphate backbones of spliced , extended , and bent DNA segments were regularized to correct small deformations by highly restrained refinement in CNS ( Brunger , 2007 ) . To complete the model , the NTD-CB linker segments ( residues 56–74 ) for the C' , B , and B' subunits were fit into the EM envelope with essentially arbitrary conformations . The final model was refined with CNS using dynamic elastic network ( DEN ) restraints and torsion angle dynamics to obtain reasonable linker geometries and good overall stereochemistry ( Schröder et al . , 2007; 2010 ) . To restrain flexible regions of the model to remain within the EM envelope and to optimize the final model fit to density , we included an energy term for agreement with the Fourier coefficients of the EM density . An integrative HJ model was built from the excisive HJ complex based on the different bridging interactions that form during integrative recombination ( Figure 2B; Tong et al . , 2014 ) . Binding of the C' , C , and B' NTDs to the P' arm could be readily accommodated by changing the NTD-CB linker conformations; no change in the P' arm was required . A new P arm was assembled in the absence of Xis as previously described ( Seah et al . , 2014 ) , using the NMR-based model for Int B-NTD binding to the P1 site . The initial model , lacking the phased bend induced by Xis , clearly indicated that the P arm must pass over P' , with IHF bending at the H1 site responsible for re-directing the P1 site towards the Int tetramer . Small roll angle changes and bending at H1 and H2 ( < 10° ) were used to obtain a P arm geometry that avoids steric clashes and positions the Int NTD bound at P1 close to the Int B CB domain . A small amount of bending of the P' site was found to better accommodate the P arm crossing . After insertion of CB-NTD linkers for each of the Int subunits ( arbitrary conformations ) , the model was optimized using CNS to obtain reasonable linker geometries and good stereochemistry elsewhere . Figures 3–5 were generated with UCSF Chimera ( Pettersen et al . , 2004 ) . Figures 6–8 were generated with Pymol ( Delano , 2002 ) . | Some viruses can remain dormant inside an infected cell and only become active when conditions are right to multiply and infect other cells . Bacteriophage λ is a much-studied model virus that adopts this lifecycle by inserting its genetic information into the chromosome of a bacterium called Escherichia coli . Certain signals can later trigger the viral DNA to be removed from the bacterial chromosome , often after many generations , so that it can replicate and make new copies of the virus . Specific sites on the viral and bacterial DNA earmark where the virus’s genetic information will insert and how it will be removed . Remarkably , each of these two site-specific reactions ( i . e . insertion and removal ) cannot be reversed once started , and their onset is precisely controlled . These reactions involve a molecular machine or complex that consists of four enzymes that cut and reconnect the DNA strands and seven DNA-bending proteins that bring distant sites closer together . Despite decades of work by many laboratories , no one had provided a three-dimensional image of this complete molecular machine together with the DNA it acts upon . Now , Laxmikanthan et al . reveal a three-dimensional structure of this machine with all its components by trapping and purifying the complex at the halfway point in the removal process , when the DNA forms a structure known as a “Holliday junction” . The structure was obtained using electron microscopy of complexes frozen in ice . The structure answers many of the long-standing questions about the removal and insertion reactions . For example , it shows how the DNA-bending proteins and enzymes assemble into a large complex to carry out the removal reaction , which is different from the complex that carries out the insertion reaction . It also shows that the removal and insertion reactions are each prevented from acting in the opposite direction because the two complexes have different requirements . These new findings improve our understanding of how the insertion and removal reactions are precisely regulated . Laxmikanthan et al . ’s results also serve as examples for thinking about the complicated regulatory machines that are widespread in biology . | [
"Abstract",
"Introduction",
"Results",
"and",
"discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression",
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2016 | Structure of a Holliday junction complex reveals mechanisms governing a highly regulated DNA transaction |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.